Molecular dynamics simmulation studies

Hyo-Dong Pan
Department of Physics and Astronomy, Seoul National University, Seoul 151-747, Korea

In this letter shear viscosity of argon fulid was calculated by Molecular Dynamics simulation(MD) via python
2.7.10. The molecular system was dominated thoroughly by Lennard Jones potential and in liquid phase with
changing temperature

Introd
Introduction
Transport coefficients such as shear viscosity and
thermal conductivity, etc can be defined in terms of
the system response to a perturbation. For shear
viscosity, which is a measure of the shear stress
induced by an applied velocity gradient can be
calculate by the time integration of correlation
function of stress tensor by introducing perturbations
into the Hamiltonian, or directly into the equations of
motion[1],[2]. Getting the linear terms in the
perturbation, and comparing the equation for the
response with a macroscopic transport equation, we
can identify the transport coefficient. I.e. transport
characteristics be described by the infinite time
integral of a correlation function by the Green-Kubo
formula, as follows.[3]
(1)
where is the transport coefficient and A is a
variable appearing in the perturbation term in the
Hamiltonian. The problem in the calculation of
transport properties using the equilibrium molecular
dynamics (EMD) simulation is the non-decaying,
long-time tail of the time correlation functions in Eq.
(1). The velocity auto-correlation (VAC) function of
liquid argon in the Green-Kubo formula decays
quickly within 5 ps to give a well-defined diffusion
coefficient because the velocity is the property of
each particle. On the other hand, the stress (SAC)
and the heat-flux auto-correlation (HFAC) functions
for shear viscosity and thermal conductivity have
non-decaying longtime tails because the stress and
the heat-flux appear as the system properties. In this
letter I will calculate the shear viscosity by molecular
dynamics, using green kubo's fomula
Theory
By the green kubo fomula Shear viscosity is given
by The Green-Kubo formula for shear
viscosity is given by

(2)
where = xy, xz, yx, yz, zx, or zy and Pis
the component of the molecular stress tensor P
of the system, V is the volume of the system, K is
boltzmann constant, T is the absolute temperature of
the system. in Ref.[1] stress tensor P is give by
(3)
and an alternative equation is
(4)
but as was seen it Ref.[1] any result deriven from
eqn (3) and (4) is not converging which means the
integral of ensemble average goes to infinite and
have no meaning on the viscosity of the system. I
calculated the viscosity of argon liquid using eqn (3)
and shear viscosity tensor component of xy, yz, zx,
yx, zy, xz has all different values which is not
favorable results at all.
so I used the third equation of Ref.[1] which is
given
(5)
(6)
the result calculated by equation above gave the
same shear viscosity independent of component
which is favorable.
I used python 2.7.10 for coding. To get numerical
solution I used a library named numpy, which help to
read and calculate the matrix by component. And
also it reduced the simulation time
The first one I made has about ten row length and
cosinsts of only one step, which calculate the shear
viscosity all at once and make debugging extremely
difficult. There was another problem. Although the
code was simple, it took almost half and an hour to
calculate the viscosity.
So I separated it into three step, which reduces the
calculation time considerably. First, calculate the i-th
particle's stress tensor Pi which is give by eqn (5).

Second, calculate the i-th particle's correlation

function. Third, calculate the shear viscosity of the
system.
At first I calculated all for non diagonal component
of stress tensor. but I got the exactly the same result
and then calculated it only on the xy component.
Method
The global md algorithm is as follows. First of all,
conditions Potential interaction V as a function of
atom positions, Positions r of all atoms in the system,
Velocities v of all atoms in the system is needed as a
input.
Then repeat as follows for the required number of
steps. There is two kind of force, coulomb force and
van der waals force. First one contributes to the
bonding force while the other contributes to the non
bonding force.
First, Compute forces The force on any atom,

(7)
F =


which can be computed by calculating the force

between non-bonded atom pairs.
F = F (8)
Then plus the forces due to bonded interactions
(which may depend on 1, 2, 3, or 4 atoms), plus
restraining and/or external forces. Now the potential
and kinetic energies and the pressure tensor are
computed.
Second, Update configuration, that the movement of
the atoms is simulated by numerically solving
Newtons equations of motion

 



In this letter model liquid has no local charge or

ionic bond. so I dont need to consider the coulomb
force. So adding force term by bonded atom pairs is
not needed.
Our system consists of 512 atoms in liquid phase
and in canonical ensemble which has constant
pressure and temperature. This system is solely
dominated by Lennard Jones potential:


U = 4 [   # ]
r
r
 and  are the interaction strength and range
between particles i and j. Preceding research on the
same model has been done in ref [5]. For liquid
phase  is 0.34nm and  is 0.8KBT where KB is
Boltzmann constant.
Volume of the system is differ from each other and
can be determined by the Gromacs.
I simulated for 108 time steps of time constant.
Where time step is 1/400 of characteristic time
constant(~2ps).
Result
Shear viscosity was calculated in five different
temperatures from 220K to380K with 40K.interval.
Stress auto correlation by equation (3) does not decay
in the long term. So viscosity is not well defined.
And that made shear viscosity not to be isotropic. So
I simulated Stress tensor is using equation (5). As
mentioned before system is isotropic, so calculation
on only xy component is needed. Actually, all result
of each shear viscosity was identical.

(9)

There is two way to solve the equation (9): leap frog

algorithm and Verlet algorithm. In this letter I used
Verlet algorithm. In Verlet algorithm positions r and
velocities v at time t are used. In velocity Verlet,
positions r and velocities v at time t are used to
integrate the equations of motion; velocities at the
previous half step are not required.
The algorithm is as follows


 t + t = t +




t(10)


t + t = t + tt + (11)



t + t =  t +  + tt(12)


With no temperature or pressure coupling, and with
corresponding starting points, leap-frog and velocity
Verlet will generate identical trajectories. Given a
single starting file with the same starting point x(0)
and v(0), leap-frog and velocity Verlet will not give
identical trajectories, as leap-frog will interpret the

velocities as corresponding to t = , while

velocity Verlet will interpret them as corresponding
to the time point at t = 0.
Third, save the outputs if its needed, such as
positions, velocities, energies, temperature, pressure,
etc.[4]

Figure. 1. Correlation plot on time. This was done

from 220K to380K with the interval of 40K.
Figure 1 is the xy stress correlation function versus
time graph 220K to380K with the interval of 40K. Xaxis is time and the interval is 1ps. Y is the xy stress
correlation and is not normalized because only
converging property is important. If simulation result
of correlation function shows slowly decaying
property, molecular dynamics (MD) simulation is not
effective. In this case all the results have quickly

decaying tale, which converge to zero within 7ps

making viscosity well defined. At low temperature
there is shoulder (over shooting) around 3ps, while at
high temperature(=380K) there is no shoulder. I.e. in
higher temperature, it takes less time to converge into
zero.

Figure. 2. Simulation result of argon shear

viscosity by molecular dynamics simulation. This
was done from 220K to380K with the interval of
40K.
As shown in the figure 2, shear viscosity of argon
is disproportionate to temperature. Higher
temperature means faster relative molecular
movement that results in short relaxation time, which
means higher viscosity. Shear viscosity was 0.67cp at
220K, 0.63cp at 260K, 0.58cp at 300K, 0.55cp at
340K, and 0.46cp at 380K, which is in consensus
with former research. Fitting was done

Figure. 1. Radial distribution function of argon.

This was done from 220K to380K with the interval
of 40K.
Figure 3 shows radial distribution of argon. As
shown in this figure the fluctuation of number of
molecule along the radial distance is smaller for
higher temperature which means more homogeneous.

This is in consensus with low viscosity at high

temperature because spatial homogeneity is
originated from active molecular relative motion.
Conclusion
I simulated shear viscosity model liquid using
numpy, python 2.7.10 through molecular dynamic
simulation. Stress correlation shows quickly(~7ps)
decaying property which makes viscosity well
defined. Shear viscosity is disproportionate to
temperature, as predicted by more active relative
molecular motion. And also radial distribution
function has quickly converging property at higher
temperature because of the same reason.
Reference
[1] Bongsu Kim, et al, Shear-stress function
approach of hydration layer based on the GreenKubo formula, Phys. Rev. E 91, 032307 Published
16 March 2015
[2] Bongsu Kim, et al, Unified Stress Tensor of the
Hydration Water Layer, Phys. Rev. Lett. 111, 246102
Published 13 December 2013
[3] R. Kubo, J. Phys. Soc. Jpn. 12, 570 (1957)
[4] Emile Apol, et al, GROMACS USER
MANUAL Version 4.5.6
[5] Weijs, J. H., Snoeijer, J. H. & Lohse, D.
Formation of Surface Nanobubbles and the
Universality of Their Contact Angles: A Molecular
Dynamics Approach. Phys. Rev. Lett. 108, 104501
(2012).
[6] Song Hi Lee, Molecular Dynamics Simulation
Study of the Transport Properties of Liquid Argon:
The Green-Kubo Formula Revisited, Bull. Korean
Chem. Soc. 2007, Vol. 28, No. 8Received May 19,
2007
[7] W. T. Ashurst and W. G. Hoover, Argon Shear
Viscosity via a Lennard-Jones Potential with
Equilibrium
and
Nonequilibrium
Molecular
Dynamics Phys. Rev. Lett. 31, 206(1973)