Sie sind auf Seite 1von 1078

atomic charges of atoms i and j; e0 is the electric

constant with a value of 8.854 10-12 C2/(N m2); and rij is


the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevie
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandaku
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakuma
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxf L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995)ord, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53i, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53mar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.o
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevie
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandaku
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakuma
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxf L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995)ord, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53i, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53mar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53r, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevie
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandaku
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakuma
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxf L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995)ord, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53i, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53mar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53r, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevie
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandaku
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxford, 2005).
9C. Huang, K. Nandakuma
atomic charges of atoms i and j; e0 is the electric
constant with a value of 8.854 10-12 C2/(N m2); and rij is
the distance between atoms i and j. The Lorentz-Berelot combining
rules19 and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http:
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to are used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;an help provide and improve the u
nderstanding of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6//scitation.aip.org/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) typ
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulare used to calculate the interaction terms for
the Lennard-Jones parameters between two different species
of atoms i and j; the definition functions of Rmin(ij) and e(ij) are
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53tute of Physics. [http://dx.doi.org/10.1063/1.47946
85]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater;lar and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
124701-3 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
shown in Eqs. (3) and (4), respectively,
Rmin(ij ) = Rmin(ii) + Rmin(jj)
2
, (3)
e(ij ) =
v
e(ii) e(jj) (4)
indicating that the Rmin(ij) is an arithmetic mean and the e(ij) is
a geometric mean.
Two series of modified CNT membranes are produced
first by varying the carbon atom Lennard-Jones well-depth
values (e(CC)), and second by setting the atomic charges on
CNT carbon atoms (patterns of balanced positive and negative
charges are placed on the CNT). These modifications
affect the interactions between water molecules and the CNT
membrane. NEMD simulations are carried out for unmodified
and modified CNT membranes to analyze the NF transport
phenomenon of pressure-driven water flow in terms of water
flow rate and density and velocity (in z direction) distributions
along both radiNonequilibrium molecular dynamics simulation of pressure-driven w
ater transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American(Received 21 December 2012; accepted 13 February 2013; published online
22 March 2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Instis several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6e CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6

,
(1)
Eelect =

nonbonded atom pairs


qiqj
4pe0rij
, (2)
where e(ij) is the Lennard-Jones well-depth between atoms i
and j and Rmin(ij) is the distance at the Lennard-Jones minimum
interaction energy between atoms i and j; qi and qj are
the partialNonequilibrium molecular dynamics simulation of pressure-driven water
transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
Nonequilibrium molecular dynamics simulation of pressure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 TNonequilibrium molecular dynamics simulation of press
ure-driven water transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
he use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includeNonequilibrium molecular dynamics simulation of pressure-driven wat
er transport
through modified CNT membranes
Luying Wang, Randall S. Dumont, and James M. Dickson
Citation: The Journal of Chemical Physics 138, 124701 (2013); doi: 10.1063/1.479
4685
View online: http://dx.doi.org/10.1063/1.4794685
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/138/12?
ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
How fast does water flow in carbon nanotubes?
J. Chem. Phys. 138, 094701 (2013); 10.1063/1.4793396
Nonequilibrium molecular dynamics simulation of water transport through carbon n
anotube membranes at low
pressurea)
J. Chem. Phys. 137, 044102 (2012); 10.1063/1.4734484
Molecular simulation of pressure-driven fluid flow in nanoporous membranes
J. Chem. Phys. 127, 054703 (2007); 10.1063/1.2749236
Molecular dynamics simulations of transport and separation of carbon dioxidealkan
e mixtures in carbon
nanopores
J. Chem. Phys. 120, 8172 (2004); 10.1063/1.1688313
Kinetic theory and molecular dynamics simulations of microscopic flows
Phys. Fluids 9, 3915 (1997); 10.1063/1.869490
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
THE JOURNAL OF CHEMICAL PHYSICS 138, 124701 (2013)
Nonequilibrium molecular dynamics simulation of pressure-driven water
transport through modified CNT membranes
Luying Wang,1 Randall S. Dumont,2 and James M. Dickson1,a)
1Department of Chemical Engineering, McMaster University, Hamilton, Ontario L8S
4L7, Canada
2Department of Chemistry and Chemical Biology, McMaster University, Hamilton, On
tario L8S 4L8, Canada
(Received 21 December 2012; accepted 13 February 2013; published online 22 March
2013)
Nonequilibrium molecular dynamics (NEMD) simulations are presented to investigat
e the effect
of water-membrane interactions on the transport properties of pressure-driven wa
ter flow passing
through carbon nanotube (CNT) membranes. The CNT membrane is modified with diffe
rent physical
properties to alter the van der Waals interactions or the electrostatic interact
ions between water
molecules and the CNT membranes. The unmodified and modified CNT membranes are m
odels of
simplified nanofiltration (NF) membranes at operating conditions consistent with
real NF systems.
All NEMD simulations are run with constant pressure difference (8.0 MPa) tempera
ture (300 K),
constant pore size (0.643 nm radius for CNT (12, 12)), and membrane thickness (6
.0 nm). The water
flow rate, density, and velocity (in flow direction) distributions are obtained
by analyzing the NEMD
simulation results to compare transport through the modified and unmodified CNT
membranes. The
pressure-driven water flow through CNT membranes is from 11 to 21 times faster t
han predicted by
the Navier-Stokes equations. For water passing through the modified membrane wit
h stronger van
der Waals or electrostatic interactions, the fast flow is reduced giving lower f
low rates and velocities.
These investigations show the effect of water-CNT membrane interactions on water
transport under
NF operating conditions. This work can help provide and improve the understandin
g of how these
membrane characteristics affect membrane performance for real NF processes. 2013
American
Institute of Physics. [http://dx.doi.org/10.1063/1.4794685]
I. INTRODUCTION
Molecular dynamics (MD) simulation provides a dynamic
view of microscopic systems. Water flow through
nanoscale channels, driven by external fields, is critical to
many phenomenafor example, biological channels, drug
delivery, membrane separations, fuel cells, and novel nanofluidic
applications (nanopumps, nanosyringes, nanosensors,
etc.). In recent years, MD simulations have been used to
study water flow driven by external fields such as pressure
difference,13 osmotic pressure difference,4, 5 and electric
field.6
Nanofiltration (NF) is a common membrane separation
process driven by a pressure difference between the two sides
of the NF membrane, where a concentrated stream on the high
pressure side passes through the NF membrane and becomes
a purified stream on the low pressure side. Applications of
NF worldwide have increased primarily in the water treatment
industry,7 such as water softening, organics removal,
radium and heavy metal removal from wastewater; and sulfate
removal from seawater. In addition, NF has been applied
in other industries: pulp and paper effluent treatment in the
papermaking industry, removal of dyes and other coloring
agents in the textile industry, concentration of intermediates
and antibiotics in the pharmaceutical industry.8 In view of the
a)Author to whom correspondence should be addressed. Electronic mail:
dickson@mcmaster.ca.
growing industrial applications of NF, the transport mechanism
of NF is examined using MD simulation with the goal
of improving the understanding of the types of membranes
best suited for NF processes.
To understand the microscopic dynamics properties of
the NF process with MD simulation, nonequilibrium molecular
dynamic (NEMD) simulations must be used because
pressure-driven fluid flow corresponds to a nonequilibrium
conditions. NEMD is a valuable tool to study fluid
flow through nanoscale channels, induced by a pressure
difference.1, 2, 911 The key to a NEMD simulation of a NF
process is a mean of imposing different constant pressures
on the two sides of the NF membrane. In our previous study,
we reported a NEMD simulation system derived by Huang9
and Takaba10 to study the transport phenomena of pressuredriven
water flow through CNT membranes under NF operating
conditions.12, 13
Real NF membranes are generally made from synthetic
polymers. The membrane structure (pore size, number
of pores, and membrane thickness) and the polymer characteristics
(functional groups, electric charge, hydrophilicity/
hydrophobicity, etc.) can both affect membrane performance.
The interaction between solution and polymeric NF
membrane is one of the most important factors determining
membrane separation and transport performance. In our
previous work, the effect of membrane structure on water
transport through CNT membranes was investigated.12 This
work investigates the effect of water-CNT membrane interactions
on modified CNT membrane performance with different
0021-9606/2013/138(12)/124701/9/$30.00 138, 124701-1 2013 American Institute of
Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-2 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
membrane characteristics. Although the solution-membrane
interactions in a real NF process are more complicated than
the water-CNT interactions, this work provides fundamental
information how membrane characteristics affect membrane
performance.
The modified CNT membranes have the same membrane
structure (pore size and membrane thickness) as the unmodified
CNT membrane; only the van der Waals interactions
and electrostatic interactions between water molecules and
CNT membranes are altered. Transport through the unmodified
CNT membrane and modified CNT membranes is compared
to establish the role of the water-CNT interactions in determining
observed transport properties. This work highlights
a unique advantage of MD simulationthe water-membrane
interaction can be adjusted to reveal the effect on the NF transport.
The unmodified and modified CNT membranes in this
paper can be considered as functionalized CNTs or models of
simplified NF membranes at operating conditions consistent
with real NF systems.
II. SIMULATION METHODS
The system model in Cartesian coordinates is shown
in Fig. 1. The system consists of two graphene sheets acting
as moveable walls, two water reservoirs, and the CNT
membrane model. The moveable walls and the membrane
model are 4 nm 4 nm in the x, y plane. The thickness and
the pore size of the CNT membrane is determined by the
(12, 12) type CNT, which has a 6.0 nm length and 0.643
nm internal radius.1, 14 The top and bottom water reservoirs,
connected by the CNT membrane model, provide the source
and sink of the water flow passing through the membrane,
along the z direction. External force is applied on each carbon
atom of the moveable walls to produce 8.1 MPa pressure
FIG. 1. Perspective snapshots of the simulation systems produced by VMD
package at the beginning status of MD simulations (adapted from the figure
in Ref. 12). Shown is the beginning empty CNT membrane model connecting
two liquid filled reservoirs, and the two graphene sheets acted as the movable
walls where the force ft or fb is applied on each carbon atom of the top or
bottom wall. Carbon atoms in green, hydrogen atoms in white, and oxygen
atoms in red.
on the top water reservoir and 0.1 MPa on the bottom water
reservoir to give a pressure difference of 8.0 MPa. We use
a simulated pressure difference across the CNT membrane
that is somewhat higher than typical NF systems (usually
less than 3 MPa pressure difference) as this higher pressure
difference reduces the (still large) CPU time needed for the
simulations.
An equilibrium MD simulation, in an isothermal-isobaric
(NPT) ensemble, is carried out first to energy-minimize and
equilibrate the system; and then a NEMD simulation is carried
out to simulate the pressure-driven water flow through
the CNT membrane. Each simulation is at 300 K, and takes
about 14 CPU days for 1.0 ns of simulation (the length of
each NEMD simulation is from over 100 ns to about 250 ns).
The detailed simulation methodology (force field parameters,
integration algorithm, periodic boundary conditions, etc.)
and simulation procedures are the same as in our previous
work.12
All MD simulations are performed using NAMD
(NAnoscale Molecular Dynamics) package15 and VMD
(Visual Molecular Dynamics) package16 in the CHARMM
(Chemistry at HARvard Molecular Mechanics)17 force field.
The water molecule is represented by the modified flexible
TIP3P (transferable intermolecular potential three-point) water
model. Unlike the original rigid TIP3P model in which
the van der Waals force is only set on the oxygen atom, the
TIP3P water model in NAMD package is a flexible model
and is modified additionally to include Lennard-Jones parameters
for the hydrogen atoms.18 The use of a flexible water
model here requires more calculations during a MD simulation
but produces a more accurate representation of the
water molecule needed in this work. The CHARMM force
field includes several terms representing intermolecular and
intramolecular interactions. The intermolecular potential energy
(also called long-range potential energy) includes the
Lennard-Jones 612 potential and the Coulomb potential to
describe van der Waals and electrostatic interactions, respectively.
The mathematical expressions of the Lennard-Jones
612 type potential (EVW) and Coulomb potential (Eelect) are
shown in Eqs. (1) and (2):
EVW =

nonbonded atom pairs


e(ij )

Rmin(ij )
rij
12
-2

Rmin(ij )
rij
6
al and flow directions.
III. RESULTS AND DISCUSSION
The theoretical flux and velocity in the z direction are
calculated using the Navier-Stokes equations for a cylindrical
channel (Eqs. (5) and (6)). Equation (5) is also called the
Hagen-Poiseuille equation, and we will simply refer to Eq. (6)
as the Navier-Stokes equation:
JW = LP P =
r2
p
8? (Lz/AK)
P,
AK =
p r2
p
Amem
=
p r2
p
Lx Ly
,
(5)
vz = P
4?Lz

r2
p
- r2
, (6
water reservoir. Average density and velocity values for
each section are obtained from effective collected trajectory
data of the NEMD simulation. The stream velocity of water
flow in the water reservoirs is too lowdue to the low
pressure differenceto determine with reasonable accuracy
through the NEMD simulation. Instead, the water velocity in
each reservoir is determined from the velocity of each movable
wall, which is calculated by fitting the moved distance of
the wall to a linear dependence on simulation time. The number
of water molecules passing through the membrane model
are counted every 50 ps and plotted as a function of the simulation
time. The slope of the linear profile (through the origin)
is the number flow rate. For each flow rate value, the estimated
error of the flow rate is represented by the standard error on
the slope with the assumption of uncorrelated data (a correlation
time shorter than 50 ps). An example of the error analysis
for the simulation flow rate is provided in our previous
work.12 The error analyses for density or velocity values are
performed by considering the integrated autocorrelation time,
which is used to calculate the effective sample number of time
correlated dat)
where LP is the permeability coefficient of the water fluid, rp
is the pore radius, ? is the viscosity of water at the operating
temperature, Lz/AK is the ratio of the membrane effective
thickness to the membrane porosity, P is the pressure difference,
and r is the radial position. The density and viscosity
of water at 300 K are 1000 kg/m3 and 8.54 10-4 Pa s,
respectively,20 and theoretical flux is converted to theoretical
flow rate in molecules/ns to compare with the simulation
flow rate.
The simulation water flow rate is defined as the number
flow rate (molecules/ns), which is calculated by the slope of
the linear trend between the number of water molecules in the
bottom reservoir and simulation time. To obtain the density
and velocity distributions along the radial direction, the pore
is divided into ten annular sections evenly spaced in the radial
direction. The distributions along the flow direction are
studied by dividing the entire system evenly into ten sections
with equal distance along the flow direction: four sections in
the top water reservoir, three in the CNT, and three in the bottoma (the number
of uncorrelated samples).21 The
statistical error of each density or velocity data point is calculated
directly from the standard deviation of the sampled
mean using the number of uncorrelated samples as discussed
previously.12
A. The modified Lennard-Jones parameters
of the CNT membranes
To study the effect of van derWaals interactions between
water and the membrane with respect to the transport phenomenon,
the Lennard-Jones parameters of carbon atoms in
the CNT model are modified. As listed in Table I, the welldepth
parameter of the carbon atom is set as the well-depth parameter
value of a hydrogen (Case I) or oxygen (Case II) atom
in the modified CNT models (corresponding to smaller and
larger well depths, respectively). The distance of the Lennard-
Jones minimum interaction energy is kept at the same value
i.e., that of a carbon atom. The value of the well-depth parameter
reflects the strength of the van der Waals interaction between
two atoms: a larger value means a stronger interaction.
The flow rate of eachns)
of unmodified and modified CNT membranes. The results are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. H-P
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I.
frictionless CNT surface, (2) the quite narrow pore size, and
(3) the relatively weak interactions between water molecules
and the CNT surface. However, water transport does depend
on the interactions between water and the CNT: strong interactions
impede flow, while weak interactions permit more
rapid flow. As the interactions increase (from Case I, to unmodified,
to Case II) the water flow rate decreases due to these
interactions.
According to the Eqs. (1) and (3), the Lennard-Jones potential
energy between the water molecules and the membrane
surfaces depends on the square root of the well-depth parameter
of the CNT membrane. Figure 3 illustrates the relationship
between the simulation flow rate and the water-membrane interaction,
which is represented by the square root of the welldepth
parameter of the unmodified and modified CNT membranes.
Here, we see the inverse relationship between flow
FIG. 3. Simulation flow rate (molecules/ns) as a function of the square root
of the well-depth parameter. The data points are based on membrane simulations
using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The
statistical error for each data point is less than 0.5%. The solid line is the
linear trend based on the simulation results.
FIG. 4. Effect of van der Waals interactions between water and membrane,
represented by the w CNT membrane and the theoretical
flow rate are shown in Fig. 2. As previously seen,12 CNT
membranes exhibit much higher flow rate than theoretical values.
Here, we see that the simulation flow rate decreases with
increasing CC well-depth. A continuum flow dominated by
the bulk properties can be described by the Hagen-Poiseuille
equation. However, water passing through a nanoscale CNT
membrane forms a non-continuum nanofluid flow. Transport
of water through the CNT membrane is not governed by the
Hagen-Poiseuille equation due to (1) a smooth and nearly
TABLE I. The Lennard-Jones parameters of the unmodified and modified
CNT membranes.
Lennard-Jones Modified Modified
parametersa Unmodified Case I Case II
emem (kcal/mol) 0.07 = e(CC) 0.046 = e(HH)
b 0.1521 = e(OO)
c
Rmin/2 (nm) 0.19924 0.19924 0.19924
aThe values of the parameters are based on CHARMM force field.17
bSet to the well-depth parameter for hydrogen, e(HH); a smaller interaction.
cSet to the well-depth parameter for oxygen, e(OO); a larger interaction.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-4 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 2. Bar chart of simulation and theoretical flow rates (molecules/ell-depth
parameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the density distributions
along the radial direction: density values are averaged over thin
annular sections in the pore at the indicated r values over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 1%. The values of the welldep
th
parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The solid
lines are the trend curves based on the simulation results, and the dotted lines
are the pore surface represented by the effective pore radius (0.643 nm).
rate and well depth. With only three points, the details of this
relationship are not clearonly that flow rate decreases with
increasing well depth. Clearly, stronger attraction of water to
the pore surface reduces the flow as expected.
Figures 4 and 5 sho from left
to right: the water reservoir on the high pressure side, the pore of the CNT
membrane, and the water reservoir on the low pressure side.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-5 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
For the axial direction, as shown in Fig. 5, the density
remains constant throughout the length of the pore for all
three well depths. However, the constant value is sensitive
to the well depth. If the carbon well depth is increased from
that of hydrogen (Case I), to carbon (unmodified) and then
to oxygen (Case II), the density in the pore increases while
the bulk values in the reservoirs remain unchanged. Overall
the increase in density (in the pore) from hydrogen to carbon
and hydrogen to oxygen well-depth parameters is about
2 molecules/nm3 (6%) and 4 molecules/nm3 (13%), respectively.
The density decrease on entering the pore for the unmodified
case (well depth of carbon) is due to the ordered
stacking of water molecules in the pore compared to bulk solution
and is consistent with results found previously.12 For
Case I, the attraction of water to the membrane wall is less and
the water is more free to align with itself and obtain a density
slightly higher (about 3%) than bulk water. On the other hand,
for Case II, the interaction of water with the wall is stronger
and this apparently forces the water into closer aligned rings
in the pore (as evidenced in Fig. 4) and an even higher
density than for the unmodified case (about 10% higher) is
obtained.
The radial velocity distribution for continuum flow
through a cylindrical pore should have a parabolic profile
determined by the Navier-Stokes equations. Figure 6 shows
the oscillatory velocity distributions along the radial direction
observed for pressure-driven water flow through these CNT
membranes. The velocity profiles obtained from the NEMD
simulations are higher than the theoretical values predicted
by the Navier-Stokes equations. The non-parabolic distribution
of water velocity in the CNT membrane is explained in
our previous work.12, 13 Figure 6 shows that Case I, with the
least interaction, has the highest velocity curves and Case II,
with the highest interaction, has the lowest velocity curves.
Overall, the velocity profiles increase in the order of decreas-
FIG. 6. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the velocity
distributions (in z direction) as a function of radial position: velocity
(in z direction) values are averaged over thin annular sections in the pore
at the indicated r values over the NEMD simulation at steady state water
flow. The data points are based on membrane simulations using the (12, 12)
CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical errors for most data
points are less than 10%. N-S equation refers to the Navier-Stokes equation
(Eq. (6)), and the values of the well-depth parameters (e(CC), e(HH), and e(OO))
are shown in Table I. The solid lines are the trend curves based on the simulati
on
results, the dashed lines are the trends based on the Navier-Stokes
equation, and the dotted lines are the pore surface represented by the effective
pore radius (0.643 nm).
FIG. 7. Effect of van der Waals interactions between water and membrane,
represented by tw the density distributions along the
radial and axial directions, respectively, for the three CNT
membrane models. The radial distributions of density all exhibit
similar oscillatory waves, with low density in the pore
center and two pairs of symmetrical peaksindicating that
there are two cylindrical rings of water molecules within the
pore. The effect of increasing the CNT well-depth parameter
is to shift the outer ring towards the pore surface and both
the inner and outer water rings become more well defined
(sharper peaks).
FIG. 5. Effect of van der Waals interactions between water and membrane,
represented by the well-depth parameter of the Lennard-Jones potential (interact
ion
increasing from red , to black , to green ), on the density distributions
along the flow direction: density values are averaged over thin sections
about the indicated z values (cylindrical sections in the pore and cuboid
sections in the water reservoirs) over the NEMD simulation at steady state
water flow. The data points are based on membrane simulations using the
(12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T: 300 K. The statistical error for
each data point is less than 1%. The values of the well-depth parameters
(e(CC), e(HH), and e(OO)) are shown in Table I. The solid lines are the trend
curves based on the simulation results. The dashed lines are the membrane
boundaries along z direction separating the system in three partshe well-depth p
arameter of the Lennard-Jones potential (interaction
increasing from red , to black , to green ), on the velocity
distributions (in z direction) along the flow direction: velocity (in z directio
n)
values are averaged over thin sections about the indicated z values (cylindrical
sections in the pore and cuboid sections in the water reservoirs) over the
NEMD simulation at steady state water flow. The data points are based on
membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa; T:
300 K. The statistical error for each data point is less than 10%. The values of
the well-depth parameters (e(CC), e(HH), and e(OO)) are shown in Table I. The
solid lines are the trend curves based on the simulation results. The dashed
lines are the membrane boundaries along the z direction separating the system
in three parts from left to right: the water reservoir at high pressure side,
the pore of the CNT membrane, and the water reservoir at low pressure side.
ing attraction to the membrane wall as would be expected
since the lower water-membrane attraction allows the water
to move faster. The oscillatory shape of the velocity distribution
is approximately consistent with the two rings of water
observed in the density distribution (Fig. 4): a higher velocity
section near
whole length of the tube to get each average radial result; averaging
the values over the whole radius for each axial section to
get each flow-directional result). The almost constant results
along the tube in Figs. 5 and 7 show that the entrance/exit
effects are approximately insignificant here (also shown in
Sec. III B). Note that Nicholls et al.22 have reported that
the entrance/exit effects can be significant in their studies of
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-6 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
water transport through (7,7) CNTs at a 200 MPa pressure
difference.
B. The polarized CNT membranes
It has been shown that the delocalized p-electrons of the
CNT carbon atoms can result in polarized CNTs,2326 with
positive and negative partial charges of the polarized CNT
forming a pattern.27 Simplified charge distributions (as shown
in Fig. 8) are proposed here: positive and negative partial
FIG. 8. Charge patterns of the unmodified and modified CNTs: (a) unmodified
CNT, (b) unwrapped CNT of the ring polarized model, and (c) unwrapped
CNT of the band polarized model. This figure illustrates the relative
locations of the neutral atoms in green, the positive atoms in orange, and
the negative atoms in blue. Transport is from left to right as indicated by the
arrows on the z-axes.
charges arranged in distinct patterns are set on the unmodified
CNT to model a fictitious polarized CNT. Unlike the unmodified
CNT membrane, the polarized CNT membrane can affect
the water through electrostatic interactions. The water transport
properties of the unmodified and polarized CNT membranes
are investigated to show the effect of electrostatic interactions
on the transport properties.
Two modified CNTs with different charge patterns
(Fig. 8) are used to model the polarized CNT membranes.
Each CNT has a total net charge of zero, with positive and
negative atomic charges on selected carbon atoms along the
length or circumference of the CNT. Figure 8(b) shows the
ring polarized CNT with ten alternating charged rings with
equal magnitude along the length. Figure 8(c) shows the band
polarized CNT with six alternating charged bands with equal
magnitude along the circumference. The atomic charges of
each charged ring or band have the same magnitude, and
the adjacent positive and negative rings or bands are placed
evenly to form orderly charge patterns. Each charged ring
or band and neighbouring uncharged atoms form a charged
pore surface having a surface charge density of 1.0 C/m2 or
-1.0 C/m2, which corresponds to 0.040765 e or 0.031958
e on charged carbon atoms of the ring and band polarized
CNTs, respectively. Here we study artificially polarized
CNTs; the CNTs represent, for instance, end-functionalized
CNTs or CNTs modified due to the highly polarizable p-
electrons in the C the pore center and a lower velocity section near
the pore surface. The velocity distribution of the nanoscale
water flow is more complicated than the density distribution.
The velocity distribution along the radial direction may be affected
by the two-ring configuration of water in the pore. The
velocity profiles along the axial (flow) direction are shown in
Fig. 7, where the velocity dramatically increases/decreases
when water molecules enter/exit any of the CNT membranes.
Like the density function, the velocities are constant along the
pore length and are larger than the velocities in the reservoirs
due to the much smaller cross-sectional area in the pore compared
to in the reservoirs. As discussed above for Fig. 6, again
it can be seen from Fig. 7 that increasing the strength of the interaction
decreases the velocity of the water flow in the CNT.
The weaker interactions between water and the CNT membrane
result in the CNT having a higher flow rate, while the
stronger interactions increase the attraction to the pore wall
and decrease the flow rate.
Entrance/exit values are automatically included in the
NEMD calculation of the detailed axial and radial values of
density and velocities. As mentioned above, the entrance/exit
values are both included in the analysis of the radial and the
flow-directional distributions (averaging the values over theNT aromatic rings.
In the case of the endfunctionalized
CNT, Goldsmith and Hinds28 studied a watermethanol
solution through a CNT membrane with a charged
Pt monolayer attached at one end of the CNT by using NEMD
simulations.
The water flow rates of different CNT membranes are
compared with the theoretical flow rate in Fig. 9. The flow
rates of the two polarized CNT membranes are both smaller
than the unmodified CNT membrane and larger than the theoretical
flow rate calculated by the Hagen-Poiseuille equation.
In addition, the ring and band polarized CNT membranes exhibit
similar flow rates, which are about 88% and 84% of the
flow rate for the unmodified CNT membrane, respectively.
While adding a charge to the membrane does reduce the flux
FIG. 9. Bar chart of simulation and theoretical flow rates (molecules/ns) of
unmodified and modified polarized CNT membranes. The results are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical error for each flow rate data is less than 0.5%. HP
equation refers to the Hagen-Poiseuille equation (Eq. (5)), and the charge
patterns of the modified polarized CNTs are shown in Figs. 8(b) and 8(c).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-7 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 10. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the radial direction: density values are averaged over thin annular sectio
ns
in the pore at the indicated r values over the NEMD simulation at steady
state water flow. The data points are based on membrane simulations using
the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for
each data point is less than 1%. The charge patterns of the modified polarized
CNTs are shown in Figs. 8(b) and 8(c). The solid lines are the trend curves
based on the simulation results, and the dotted lines are the pore surface repre
sented
by the effective pore radius (0.643 nm).
the charge pattern apparently has little influence on the transport
properties.
The positive and negative charges of the polarized membranes
give rise to electrostatic interactions between water
and the pore surface, which attracts water molecules near the
pore surface. Since water has an electrical dipole consisting
of a partial negative charge on the O atom and partial positive
charges on the H atoms (net neutral, of course) then the water
molecules can interact electrostatically with polarized groups
on the membrane. This attraction results in a reduced water
flow through the CNT membrane. Although the electrostatic
interaction makes the total interactions between water and the
CNT membrane stronge
shown in Sec. III A), except not as pronounced. The similar
density distributions for the two polarized CNT membranes
tell us that the pattern of polarization is largely immaterial.
Radial distributions of velocity for unmodified and polarized
CNT membranes are shown in Fig. 12 along with
the parabolic profile predicted by the Navier-Stokes equation.
Again, observed velocity profiles are non-parabolic and much
FIG. 11. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns within the CNT, on the density distributions
along the flow direction: density values are averaged over thin sections about
the indicated z values (cylindrical sections in the pore and cuboid sections in
the water reservoirs) over the NEMD simulation at steady state water flow.
The data points are based on membrane simulations using the (12, 12) CNT,
Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The statistical error for each data point is
less than 1%. The charge patterns of the modified polarized CNTs are shown
in Figs. 8(b) and 8(c). The solid lines are the trend curves based on the simula
tion
results. The dashed lines are the membrane boundaries along the z
direction separating the system in three parts from left to right: the water
reservoir at high pressure side, the pore of the CNT membrane, and the water
reservoir at low pressure side.
higher than predicted by the Navier-Stokes equation. The
electrostatic interactions between the water molecules and the
polarized CNTs lead to lower velocities near the pore surface.
Otherwise, the radial distributions of velocity are all very similar.
The polarization increases the hydrophilicity of the pore,
as seen in Fig. 10, so that density increases near the pore surface.
Goldsmith and Martens29 studied the effect of modifications
of surface interactions on pressure-induced flow through
rigid ice-like membranes, which include a hydrophilic model
and a hydrophobic model. For the hydrophilic case, the membrane
is modeled by standard TIP3P water; while, for the hydrophobic
case, the partial charges of water in the membrane
model is set to zero in order to eliminate the water-membrane
FIG. 12. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the radial direction: velocity (in z direction) values are a
veraged
over thin annular sections in the pore at the indicated r values over
the NEMD simulation at steady state water flow. The data points are based
on membrane simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0 MPa;
T: 300 K. The statistical errors for most data points are less than 10%. N-S
equation refers to the Navier-Stokes equation (Eq. (6)), and the charge patterns
of the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The
solid lines are the trend curves based on the simulation results, the dashed
lines are the trends based on the Navier-Stokes equation, and the dotted lines
are the pore surfar, the water-CNT interactions are not
strong enough to reduce the flow to that predicted by continuum
mechanics (given by the Hagen-Poiseuille equation).
The density distributions resulting from polarized CNT
membranes are shown in Figs. 10 and 11. The radial distributions
of density in all three CNTs are similar. The only effect
of the CNT polarization is to shift the outer peaks lower and
nearer to the pore surface. The higher density values of water
in the polarized CNTs are seen from the density profile along
the axial (flow) direction, shown in Fig. 11. Polarization of
the CNT increases the hydrophilicity of the pore, contributing
to higher density of water molecules in the CNT. Through
analyzing the average number of water molecules in the CNT
for each membrane, the results show that there are about four
more water molecules in the polarized CNT than in the unmodified
CNT. Figure 11 also shows water density in the CNT
is uniformly lower than in the bulk reservoirs for all the CNT
membranes. The effect of polarizing the CNT membrane is
similar to that of increasing the van der Waals attraction (asce represented by
the effective pore radius (0.643 nm).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-8 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
FIG. 13. Effect of electrostatic interactions between water and membrane,
caused by the charge patterns of the CNT, on the velocity (in z direction)
distributions along the flow direction: velocity (in z direction) values are ave
raged
over thin sections about the indicated z values (cylindrical sections in
the pore and cuboid sections in the water reservoirs) over the NEMD simulation
at steady state water flow. The data points are based on membrane
simulations using the (12, 12) CNT, Lz: 6.0 nm; P: 8.0MPa; T: 300 K. The
statistical error for each data point is less than 10%. The charge patterns of
the modified polarized CNTs are shown in Figs. 8(b) and 8(c). The solid lines
are the trend curves based on the simulation results. The dashed lines are the
membrane boundaries along the z direction separating the system in three
parts from left to right: the water reservoir at high pressure side, the pore of
the CNT membrane, and the water reservoir at low pressure side.
electrostatic interactions. In that work, the hydrophilic case
showed a faster water flow through the nanopore than the hydrophobic
case. A non-parabolic water velocity distribution
was observed in each case. These results are consistent with
the results presented here.
The net effect of polarization is lower average velocity
and associated lower flow rate. The lower velocities of the polarized
membranes can be observed from the velocity profile
along the direction of flow, shown in Fig. 13. In this figure, we
see that polarization of the CNT reduces velocity uniformly
within the pore, though the velocity values in the CNT are still
higher than bT
(Fig. 14(c)), the +ve charged band regions (orange) attract the
partial -ve charged oxygen molecules while the -ve charged
band regions (blue) attract the partial +ve charged hydrogen
molecules of the water. For both the ring and the band polarized
CNTs the neutral regions (green) on the CNT wall,
on average, do not prefer the oxygen or hydrogen molecules.
The preferential orientation of water primarily occurs near the
pore surface and there is no noticeable difference observed for
the water in the pore center, which agrees with the radial density
distributions (Fig. 10) showing different densities near the
FIG. 14. Orthographic snapshots of water in the (a) unmodified CNT,
(b) ring polarized CNT, and (c) band polarized CNT. Each snapshot is
a cross-sectional view in the center section (facing z = -1.0 nm, from
z=-1.0 nm to 1.0 nm; see Fig. 8 for the coordinates). Neutral carbon atoms
in green, positively charged carbon atoms in orange, negatively charged carbon
atoms in blue, hydrogen atoms in white, and oxygen atoms in red. For
the unmodified (neutral) wall in (a) the water orientation near the pore surface
is random. However, for the polarized cases (b) and (c) there is preferential
orientation of water in the CNTs.
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
124701-9 Wang, Dumont, and Dickson J. Chem. Phys. 138, 124701 (2013)
pore surface but almost the same densities in the pore center.
These orthographic snapshots help explain and are consistent
with the discussion above on ring and band oriented CNTs.
IV. CONCLUSIONS
The effect of interactions between water and CNT
on water transport has been investigated by simulating
pressure-driven water flow through modified CNT membranes.
In one set of membranes, the interaction is modified
by varying the water-CNT van der Waals interactions. In
another set of membranes, electrostatic interactions are
introduced by putting charges on the CNT. The transport
properties investigatedsimulation flow rate, radial and axial
distributions of density and velocityshow that increasing
the water-CNT interactions increases density within the pore
while reducing velocity and overall flow rate through the
pore. Increasing interactions between water and the CNT
draws water molecules to the pore surface and reduces the
pressure-driven water flow as the pore interacts more strongly
with the water molecules. Still the water transport is much
faster than that predicted by estimation, in Quantum Simulations of Complex Many-
Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.estimatio
n, in Quantum Simulations of Complex Many-Body Systems:
From Theory to Algorithms, edited by G. Johannes, D. Marx, and A.
Muramatsu (John von Neumann Institute for Computing, Jlich, 2002 Nakao, J. Chem.
Phys. 127, 054703 (2007).
11I. Hanasaki and A. Nakatani, Modell. Simul. Mater. Sci. Eng. 14, S9
(2006).
12L. Wang, R. S. Dumont, and J. M. Dickson, J. Chem. Phys. 137, 044102
(2012).
13L. Wang, R. S. Dumont, and J. M. Dickson, Comparison of two NEMD
simulation systems for the pressure-driven water permeation through carbon
nanotube membranes, Modell. Simul. Mater. Sci. Eng. (submitted).
14A. Noy, H. G. Park, F. Fornasiero, J. K. Holt, C. P. Grigoropoulos, and O.
Bakajin, Nanotoday 2, 22 (2007).
15J. Ph).
22W. D. Nicholls, M. K. Borg, D. A. Lockerby, and J. M. Reese, Microfluid.
Nanofluid. 12, 257 (2012).
23L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 52, 8541
(1995).
24T. Dumitrica, C. M. Landis, and B. I. Yakobson, Chem. Phys. Lett. 360,
182 (2002).
25G. Y. Guo, K. C. Chu, D. S. Wang, and C. G. Duan, Comput. Mater. Sci.the Navie
r-Stokes equations.
Similarly introducing electrostatic interactions has the
same effect as increasing the van der Waals interactions.
The transport of water through the modified membranes is
reduced but is still much faster than that predicted by the
Navier-Stokes equation. The modified membranes still produce
non-uniform nanofluid flow with lower friction than that
consistent with the Navier-Stokes equations. Overall, the increases
in water-CNT attractions considered here are not sufficient
to eliminate the flow enhancement characteristic of a
CNT membrane system. The nanoscale interior of the CNT
and the smooth surface, features retained by the modified
CNT membranes, are apparently the key to the observed flow
enhancement.
Polarized CNT membranes with zero net charge are constructed
with two different patterns of polarization; a ring
pattern and a band pattern. Both show similarly reduced
flow rates. While the pattern of polarization on the polarized
CNTs makes little difference on the water flow rate,
there is a noticeable difference on the water orientation in the
CNT. Orthographic snapshots of these ring and band polarized
membranes are consistent with the observed simulation
data.
This work reveals that the weak water-CNT interactions
are important in determining the degree of flow enhancement
seen in CNT membrane systems; the modified CNT membrane
systems with incrulk values. The pattern of polarization is seen to
have little impact, though band polarization does appear to reduce
flow a little more than ring polarization. The difference
is too small to prove a charge pattern effect, however.
Finally, we examine the detailed arrangement of water
molecules in the unmodified CNT and the ring and band polarized
CNTs as shown in Fig. 14. Water configuration in
the unmodified CNT (Fig. 14(a)) is similar to reported previously
with the water molecules in two rings12 and with no
discernable orientation towards the graphene CNT wall. However,
patterns are noticeable in Figs. 14(b) and 14(c). For
water in the ring polarized CNT, the facing section shows
that partial +ve H atoms (white) of the water molecules near
the pore surface are oriented towards -ve charged (blue)
ring region on the wall. Similar for the band polarized CNeased water-CNT intera
ctions show
enhanced flow, but less so. While these modifications of the
CNT membranes all act to reduce water flow, the effect on
the transport of solute-water mixtures could be even larger.
Membrane performance (water transport) can be controlled
through manufacturing/modifying CNT membranes having
weak or strong solution-membrane interactions, which could
have a more significant effect on organic solutes/liquids than
the pure water considered here. Real polymeric NF membranes
are more complex than these CNT membrane models
and water flow through the NF membrane would not be
as fast as through CNT membranes due to the much stronger
water-NF membrane interactions. The expected significance
of this work, in the future, is to help to understand the transport
mechanisms of CNT membranes and polymeric membranes,
and then to improve applications of NF process.
ACKNOWLEDGMENTS
The Natural Sciences and Engineering Research
Council Canada (NSERC) and McMaster University are
thanked for their financial support. The Shared Hierarchical
Academic Research Computing Network (SHARCNET,
http://www.sharcnet.ca) is gratefully acknowledged for
providing high performance computing technology for this
work.
1J. Goldsmith and C. C. Martens, J. Phys. Chem. Lett. 1, 528 (2010).
2K. Leung and S. B. Rempe, J. Comput. Theor. Nanosci. 6, 1948 (2009).
3F. Zhu, E. Tajkhorshid, and K. Schulten, Biophys. J. 83, 154 (2002).
4A. Kalra, S. Garde, and G. Hummer, Proc. Natl. Acad. Sci. U.S.A. 100,
10175 (2003).
5A. N. Raghunathan and N. R. Aluru, Phys. Rev. Lett. 97, 024501 (2006).
6J. Su and H. Guo, ACS Nano 5, 351 (2011).
7X. Lu, X. Bian, and L. Shi, J. Membr. Sci. 210, 3 (2002).
8A. I. Schfer, A. G. Fane, and T. D. Waite, Nanofiltration: Principles and
Applications (Elsevier, Oxf L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995)ord, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53i, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53mar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53r, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53rg/termsconditions. Downloaded to IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53r, Oxford, 2005).
9C. Huang, K. Nandakumar, P. Choi, and L.W. Kostiuk, J. Chem. Phys. 124,
234701 (2006).
10H. Takaba, Y. Onumata, and S.illips, R. Braun, W. Wang, J. Gumbart, E. Tajkhor
shid, E. Villa, C.
Chipot, R. D. Skeel, L. Kal, and K. Schulten, J. Comput. Chem. 26, 1781
(2005).
16W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics 14, 33 (1996).
17A. D. MacKerell, N. Banavali, and N. Foloppe, Biopolymers 56, 257
(2000).
18W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 (1983).
19M. P. Allen and D. J. Tildesley, Computer Simulation in Chemical Physics
(Kluwer Academic, The Netherlands, 1993).
20U. Kaatze, Radiat. Phys. Chem. 45, 539 (1995).
21W. Janke, Statistical analysis of simulations: Data correlations and error
30, 269 (2004).
26S. M. Hou, Z. Y. Shen, X. Y. Zhao, and Z. Q. Xue, Chem. Phys. Lett. 373,
308 (2003).
27D. Lu, Y. Li, U. Ravaioli, and K. Schulten, J. Phys. Chem. B 109, 11461
(2005).
28J. Goldsmith and B. J. Hinds, J. Phys. Chem. C 115, 19158 (2011).
29J. Goldsmith and C. Martens, Phys. Chem. Chem. Phys. 11, 528 (2009).
This article is copyrighted as indicated in the article. Reuse of AIP content is
subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded t
o IP: 27.48.137.3
On: Tue, 22 Apr 2014 06:16:53

Das könnte Ihnen auch gefallen