Sie sind auf Seite 1von 7

Measurement Science and Technology

Meas. Sci. Technol. 28 (2017) 014003 (7pp) doi:10.1088/1361-6501/28/1/014003

Can molecular dynamics help in


understanding dielectric phenomena?
Roberto Olmi1 and Marco Bittelli1,2
1
  Institute of Applied Physics, National Research Council Via Madonna del Piano, 10, 50019 Sesto
Fiorentino (Firenze), Italy
2
  Department of Agricultural Science, University of Bologna Viale Fanin, 44, 40127 Bologna, Italy

E-mail: r.olmi@ifac.cnr.it and marco.bittelli@unibo.it

Received 29 June 2016, revised 28 October 2016


Accepted for publication 1 November 2016
Published 25 November 2016

Abstract
Molecular dynamics (MD) is a modeling technique widely used in material science as well as
in chemical physics, biochemistry and biophysics. MD is based on ‘first principles’, allowing
one to compute the physical characteristics of a material, such as density, heat capacity,
isothermal compressibility and also the dielectric constant and relaxation, mixing a classical
physics approach and statistical mechanics. Although a number of papers exist in the literature
concerning the study of the dielectric properties of liquid and solid materials, the MD
approach appears to be almost ignored in the electromagnetic aquametry community. We use a
rather simple example, a mixture of ethanol and water at various concentrations, to introduce
MD as a theoretical tool for investigating the dielectric behavior of more complex moist
substances. We show that MD simulations suggest a time-domain model for alcohol–water
solutions, consisting in a mixture of a KWW stretched-exponential and a simple exponential,
whose validity could be subjected to an experimental verification.

Keywords: molecular dynamics, molecular modeling, dielectrics


(Some figures may appear in colour only in the online journal)

1. Introduction the University of Groeningen in 1991 and is nowadays very


widely employed as it is totally free and it runs faster than
Theoretical models and measurement science are tightly many other MD simulation programs.
bound. The ‘plain’ measurement process involves a measuring The idea behind MD is that of computing how a system of
instrument, an object to be measured and an environment. But particles (atoms, molecules, coarse-grained structures) evolves
the real purpose of measurements is the confirmation or the under the laws of classical mechanics. Although a rigorous
rejection of a model of the real world. In such a context, simu- solution of a many-body problem at the atomic level would
lation methods based on ‘first principles’, i.e. subordinated require quantum mechanics, a semi-empirical approach can be
to the weakest possible physical constraints, play an impor- used that is based on classical mechanics and statistical physics.
tant role in the interpretation of the measurement results. The Every particle is assigned a proper potential function (usually
experimental study of the physical properties of molecular called a Force Field in the MD community) that is used to
systems (e.g. the dielectric properties) can benefit from the compute the interactions among particles as a function of their
application of broad class of computer-simulation techniques coordinates. Force Fields usually subdivide the potential func-
known as ‘molecular modeling’. tions in two broad classes: bonded interactions, describing the
Among those techniques, molecular dynamics (MD) was forces among atoms in a molecule (more precisely, describing
first introduced in the late 50s [1], but the application to large- the interaction of up to four atoms, i.e. of a dihedral), and non-
scale problems had to wait for the explosive improvement bonded interactions between atoms in different molecules.
of the computational power in the last 20 years. Indeed, the The former consist in potential terms due to bond stretching,
GROMACS (GROningen MAchine for Chemical Simulations) to the bending of bond angles and to the torsion due to the
package [2], that we used in this paper, was first released by rotation around bonds. The latter include short-range, van der

1361-6501/17/014003+7$33.00 1 © 2016 IOP Publishing Ltd  Printed in the UK


Meas. Sci. Technol. 28 (2017) 014003 R Olmi and M Bittelli

Waals interactions, and long-range, electrostatic ones. van der qiqj


Waals pair interactions are usually modeled by the empirical Vcoulomb = ∑
i, j 4πε0rij
Lennard-Jones function describing a ‘potential well’:
⎡⎛ δ ⎞12 ⎛ δ ⎞6⎤
The equations  of motion are numerically solved at discrete
VLJ(r ) = 4α ⎢⎜ ⎟ − ⎜ ⎟ ⎥ time steps, using various integrators like the Verlet and the
⎣⎝ r ⎠ ⎝r ⎠ ⎦
(1)
leap-frog algorithms [3]. Given the coordinates of an initial
‘meaningful’ geometric configuration of particles, and given
for a spherical particle of diameter δ, where α is the depth of their velocity distribution, the position and velocity of each
the potential well. particle is computed at every time step as a consequence of the
Practically, an MD simulation consists in solving the potentials (i.e. of the force) on each atom. In other words, the
Hamiltonian’s equations of motion or, in other terms in inte- particle system is evolved in time, updating the coordinates at
grating the Newton’s equations for each particle: every time step. The choice of the time step is critically impor-
→ ∂V (r→1, …, r→N ) tant, in particular it should be as large as possible to reduce
Fi = − the computation time (most of which is spent in computing
∂r→i
forces) but it must be small enough to properly reproduce
∂ 2r→i → the motion of the fastest particles. When hydrogen atoms are
mi = Fi involved, the time step is as low as 1–2 fs.
∂t 2
To simulate an infinite medium, at least in the directions
where the total potential V is a sum of terms such as: in which the interaction with the container is not of interest,
V (r ) = VB(r ) + VNB(r ) = (Vbonds + Vangles + Vtorsion )B periodic boundary conditions are applied. In order to avoid
 computations over infinite distances, when long range interac-
+ (VLJ + Vcoulomb )NB (2) tions are involved, several techniques are employed for poten-
The shape of the different terms in (2) is described in all MD tial truncation: from cut-off approaches, involving an abrupt
texts, for example in the introductory paper [3], and it is briefly decay of the interaction (working well on LJ short range inter-
summarized in the following. Bond stretching and bond angle actions, but not on Coulomb ones), to techniques manipulating
potentials are usually assumed as harmonic: the interactions by splitting them into slow and fast parts, as
the particle–mesh Ewald (PME) summation method [8].
1 MD simulations are performed at constant temperature,
Vbonds =
2
∑ kij (rij − req )2
in a so-called NVT (number of particles, volume, temper­
bonds
ature) ensemble (known as canonical ensemble in statistical
1 mechanics), or in a NPT (number, pressure, temperature)
Vangles =
2
∑ hijk (θijk − θeq )2
ensemble, i.e. in isothermal conditions at constant volume or
bond angles
pressure. Statistical mechanics come into play when, in order
respectively involving pairs and triads of atoms. The torsion to estimate a physical property (in our case, the dielectric con-
potential depends on a torsion angle φ, and it is generally stant), an average is performed on the computed geometrical
described as a cosine expansion: configurations. The key point in the average calculation is the
1 ergodic hypothesis, one of the most fundamental axioms of sta-
Vtorsion =
2
∑ An [1 + cos(nφ − γ )] tistical physics stating that the ‘ensemble’ average of a quantity
torsion angles
A can be computed as the ‘time’ average of the same quantity:
An and γ being constant depending on the molecule geometry.
The pair correlation potential represents the ‘essential’ physics, A ensemble = A time

but it is clearly a simplification as an n-body problem would meaning that during the time evolution the system will even-
require also higher-order terms. Such an approx­imation, how- tually pass through all possible states. The temperature and
ever, has been demonstrated to be sufficient [4] for describing pressure control are accomplished by means of a numerical
the physical properties of real liquids if the average three-body thermostat and barostat, respectively. In practice, temperature
effects are included in an effective pair potential. is controlled by rescaling the particle velocities, while pres­
A thorough discussion of the huge number of different sure requires rescaling the coordinates at each time step. The
Force Fields used in the literature is beyond the scope of this above described thermostat clearly violates energy conserva-
paper. Potentials used for large molecules in condensed phase tion, therefore it must be used only as a control step, i.e. data
include AMBER (assisted model building with energy refine- must not be collected at this stage. After bringing a system to
ment) [5], CHARMM (chemistry at Harvard macromolecular a desired state (for example at a given temperature) it must be
mechanics) [6], OPLS (optimized potentials for liquid simu- left free to evolve without such kind of constraints.
lations) [7] and many others. The functional form of all those As stated at the beginning, MD is an approximate method.
potentials is described by the above three interaction terms, In particular, it does not take into account electrons, as
while their specific parametres are typically determined by quantum effects are not modeled, and as such MD is not able
quantum chemical computations and by experimental data. at all in simulating the creation of rupture of chemical bonds
The last non-bonded interaction term is the straightforward and, in principle, it could not be used for rigorously inves-
Coulomb potential between the charges qi and qj a distance tigating hydrogen bonds (that have a quantum mechanical
rij apart: nature). However, investigations of H-bond distributions are
2
Meas. Sci. Technol. 28 (2017) 014003 R Olmi and M Bittelli

often found in the literature, based on the ‘crude’ approx­ Table 1.  Number of ethanol and water molecules.
imation of describing them as Coulomb interactions of atomic cETH (%) NE NW
point charges. As a frequently cited primer on MD points out
[9], ‘the realism of the simulation depends on the ability of 0 0 216
the potential chosen to reproduce the behavior of the material 20 100 1022
under the conditions at which the simulation is run’. 40 100 383
60 100 170
80 100 64
2.  Simulation of ethanol–water mixtures 100 216 0

The interest in ethanol–water mixtures is motivated by the


wide use of ethanol in industrial processes and because the distances in a liquid mixture. In this step, the system
most part of alcoholic beverages are based on such solutions. is evolved assuming flexible bonds, treating long range
Apart from that, alcohol/water mixtures are interesting in interactions by the PME summation method. Integration
themselves, because they can constitute a simple model for of the equations  of motion is performed by a steepest
studying the hydrophobic effect. Indeed, alcohols have an descent algorithm, with 10 fs step size, for a total time
amphiphilic character, i.e. they possess both hydrophobic and duration of 500 ps. The suitability of the final configura-
hydrophilic parts. tion is tested by looking at the behavior of the system
From a scientific point of view, alcohols have been studied potential energy, that should converge to a negative
for a long time, with various analytical methods and for dif- value.
ferent objectives. The dielectric relaxation properties of pure 2. Equilibration. Once a reasonable geometry has been
alcohols and in water solutions are still an actual research obtained, solute and solvent should be equilibrated at
topic, particularly in relation with hydrogen bonding and the constant temperature (NVT ensemble) and at constant
fluctuation of hydrogen networks [10, 11]. MD simulations pressure (NPT ensemble). For that purpose, 100 ps of
have been conducted related to the study of alcohol hydration NVT simulation at 300 K are followed by 100 ps of NPT
[12] and specifically on the analysis of hydrogen-bonded pairs simulation at 300 K and 1 Atm, with PME long range
[13]. electrostatics and reducing the time step to 2 fs. In both
The dynamic properties of ethanol–water mixtures have cases, the molecules are taken rigid (i.e. not flexible)
been numerically investigated for the purpose of studying using LINCS bond constraints, to reduce the computa-
the relation between mobility and viscosity [14]. Dielectric tional time. Integration is performed using the leap-frog
properties of pure ethanol have been studied by means of MD algorithm. At the end of the first half-step (isothermal run)
[15]. In this paper, the dielectric relaxation of ethanol–water the constancy of temperature during time is checked as a
mixtures are investigated by MD as a function of the ethanol control variable. After the isobaric step, the constancy of
mass fraction. pressure is checked, and the density of the alcohol–water
mixture can also be computed.
3. Production. The physical quantities of interest are com-
2.1.  GROMACS implementation
puted during the last step. As in step 2, a time step of
The MD simulations have been conducted using GROMACS 2 fs is used, again with leap-frog integration. This time,
[16, 32], version 5.0.7, employing two force fields: the however, the system must evolve for a time sufficiently
simple point charge (SPC/E) [17] for the water molecules, long to allow the relevant physics to reach steady state
and the all-atoms optimized potential for liquid simulations conditions. A total time od 20 ns is chosen, also on the
(OPLS-AA) [18] for ethanol. Needless to say, several other basis of the considerations reported in [20] concerning
force fields have been developed, for example the largely used the minimum time required for dielectric simulations.
TIP3P potential whose characteristics are compared to SPC/E For the case of pure ethanol, the total simulation time has
in [19]. been increased to 40 ns, in consideration of the expected
The size of the simulation box and the number of ethanol greater relaxation time.
and water molecules, respectively NE and NW, have been
Table 1 summarizes the number of molecules used in the
computed such to give the correct mass fraction of ethanol in
MD unit cell. The simulations employed a cutoff of 0.9 nm for
water, i.e. in terms of the molar masses of ethanol and water:
Lennard-Jones interactions. All simulations refer to a temper­
MENE ature of 300 K and a pressure of 1 Atm.
c ETH(%) =
(3) × 100
MENE + MWNW The proper number of ethanol molecules is inserted in a
cubic box, by means of specific GROMACS commands. The
GROMACS is a command line code running under Linux.
size of the box is chosen such to contain the total number of
A number of scripts have been developed to perform the fol-
molecules, with reference to table 1. For example, referring to
lowing steps:
the second row (0.2 ethanol mass fraction) the box is gener-
1. Energy minimization. The initial configuration, con- ated by means of:
sisting in NE molecules of ethanol and NW molecules of gmx insert-molecules -ci 1-ethanol.pdb -nmol 100 -o ethanol.
water is a random one, not respecting the intermolecular pdb -box 3.5 3.5 3.5

3
Meas. Sci. Technol. 28 (2017) 014003 R Olmi and M Bittelli

using the geometry (ethanol.pdb) of a single ethanol mole-


cule. In the above case, the initial cubic box side is 3.5 nm.
The choice of a small number of molecules in the MD cell
is assumed to do not significantly alter the physical ‘picture’.
Indeed, simulations with an increasing number of molecules
in the case of pure water show a moderate influence on the
relaxation dynamics (as briefly discussed in section 3.1).

2.2.  Dielectric relaxation model

It is well known that the static dielectric constant of a polar


liquid can be computed from the fluctuations of the total dipole
moment [21]. A thorough analysis of the computer simulation
of polar systems has been developed in [22]. Although flexible
(polarizable) models are available for computing the dielectric
properties of polar liquids, here a rigid molecular model is
adopted, which greatly reduces the computational time and
has been proved to give meaningful results. The rigid model Figure 1.  Density versus ethanol mass fraction: comparison among
assumes a limit high-frequency permittivity equal to 1, and real-solution data (black line with markers) and average densities
the low-frequency value is given by: computed by MD (red line).

M2 − M 2 ⎡ ⎛ ⎡ dϕ ⎤⎞⎤
εs = 1 +
(4) ε′(ω ) = C ⎢1 +(εs − 1)R⎜L ⎢− ⎥⎟⎥
3ε0VkBT ⎣ ⎝ ⎣ dt ⎦⎠⎦

where M is the total dipole moment, 〈M2〉 and 〈M〉2 are ⎛ ⎡ dϕ ⎤⎞
ε″(ω ) = −C (εs − 1)I⎜L ⎢− ⎥⎟ (7)
⎝ ⎣ dt ⎦⎠
averages of the squared M and the square of the average
M, ε0 is the permittivity of vacuum, V the volume of the
MD cell, kB the Boltzmann constant and T the temperature. where the constant C is such to give a low-frequency limit
Equation (4) assumes a periodic arrangement of molecules, value equal to εs.
i.e. the simulation volume is infinitely repeated in all the
spatial directions. 3. Results
The relaxation dynamics depends on the auto-correlation
function (ACF) of the total dipole moment M. Effectively, the Six molecular assemblies have been considered, corresponding
frequency-dependent electric permittivity is expressed as the to the ethanol mass concentrations reported in table 1. At the
Laplace transform of the pulse-response function [23]. end of the simulations, various physical characteristics can be
ε (ω ) − 1 ⎡ dφ ⎤ computed in addition to the dielectric properties. In particular,
= L ⎢− ⎥ = 1 − L [ φ ]
⎣ dt ⎦
(5) as the dielectric constant is somehow related to the density,
εs − 1
this last parameter has been calculated and compared to the
In the linear response approximation [24], the thermal fluc- expected value for a real ethanol–water mixture. Figure  1
tuations of the polarization are equivalent to the macroscopic compares the MD-densities with the values taken from the
re-alignment caused by an external electric field, therefore the literature [26].
relaxation function coincides with the dipole ACF: The excellent agreement between computed and true den-
→ → sity makes us confident about the correctness of the simula-
M (t ) ⋅ M (0)
φ (t ) = tion, together with other control parameters not shown here
(6) → 2
M (0) (e.g. the total potential energy of the system, the average
temper­ature and pressure).
and the frequency dependent permittivity is obtained [25] The dielectric properties and the relaxation dynamics of
combining equations (4)–(6). ethanol–water mixtures have been thoroughly studied starting
GROMACS allows computing the static dielectric constant from experimental data. Usually, in the frequency domain a
εs of a system of molecules, by using the equivalence among mixture of relaxation functions is adopted to represent the
time averages and ensemble averages or, in other words, behavior of the alcoholic solution [27]. The asymmetric shape
assuming the ergodic hypothesis as true. The code also allows of the dielectric losses suggests that the dielectric function in
to obtain the normalized ACF Φ of the system, using the time- the time domain can consist in a stretched exponential, also
domain results obtained during the production phase (see sec- known as Kohlrausch–Williams–Watts (KWW) function [28],
tion 2.1). Once the above characteristics have been computed, or in a sum of such a function and a simple exponential, as we
the permittivity is obtained by a Laplace (or Fourier) trans- propose here for the ACF. Indeed, the relation among the time-
form of the time-derivative of the ACF or, in practice, of a domain KWW function and the frequency-domain Havriliak–
function ϕ fitting it: Negami (HN) distribution has been established [29], although

4
Meas. Sci. Technol. 28 (2017) 014003 R Olmi and M Bittelli

Figure 3.  The computed relaxation times as a function of the


ethanol mass fraction (see text).
Figure 4 shows the dielectric spectra (real and imaginary
part) of the analyzed molecular systems. The computation of
the spectrum requires to compute the Laplace transform (actu-
ally, the FFT (fast Fourier transform) is computed instead) of
the ACF, using (7). The plots concerning the dielectric con-
stant (on the left) refer to an ethanol concentration increasing
from 0 (pure water) to 100% (pure ethanol). All dielectric
constants are slightly underestimated. This is a well known
situation with rigid-molecule assemblies, but here is not of
great concern, if the aim is that of understanding if MD is
Figure 2.  The normalized ACF for an ethanol–water mixture with
potentially useful for studying water in dielectric models. The
40% ethanol: (a) computed ACF (cross symbols) compared to
simple exponential fitting (dashed line) and exponential mixture (8) dielectric losses behavior with frequency and ethanol mass
(solid line); (b) log(ACF) versus time in the first 60 ps. fraction (plots on the right of figure 3) is also rather encour-
aging, correctly showing the expected shift to the right of the
a closed-form expression is not available. Therefore, a sum of loss maximum with decreasing ethanol.
a KWW, bringing to a HN in the frequency domain and thus
well representing the ethanol relaxation, and a simple expo-
nential, representing water relaxation, appears very reason- 3.1.  Relevant statistical functions
able. As it is well known, the Debye relaxation is the Laplace
The radial distribution function (RDF) for oxygen–oxygen
transform of an exponential function. Equation (8) shows such
pairs has been computed in the two extreme cases (pure sol-
a fitting function.
vents), for comparing the numerical results with experimental
⎡⎛ t ⎞ α ⎤ ⎛ t ⎞ data. The RDF is a statistical function describing the structure
φ(t ) = AKWW exp ⎢⎜ ⎟ ⎥ + AD exp⎜ ⎟
⎣⎝ τKWW ⎠ ⎦ ⎝ τD ⎠
(8) of a substance (actually, describing how the density depends
on the distance from a reference particle) that can be exper­
Figure 2(a) shows, as an example, the ACF obtained by the imentally determined by x-ray or Neutron diffraction methods.
MD computation for an ethanol mass fraction of 0.4 (cross Figure 5 shows the RDF (denoted by gOO) for pure water
symbols). Superimposed, the best fitting obtained with a (left) and pure ethanol (right), computed for three different
simple exponential (red dashed line) and with function (8) number of molecules (see figure caption). The experimental
(blue solid line). It is clear that a single exponential is not RDF for O–O pairs has been reported in [30] for water at
able to accurately represent the relaxation dynamics in the ambient temperature, and for ethanol [31] at 298 K. The
first 100 ps. At times greater than 200 ps the computed ACF agreement among the computed RDF’s and the experimental
fluctuates around zero, making the choice of the fitting func- data is very good. The gOO of water experimentally exhibits a
tion quite arbitrary in that time region. first peak at about 0.34 nm, followed by two smaller maxima
Figure 3 shows the relaxation times obtained on all mat­ between 0.4 and 0.5 nm and at about 0.7 nm, exactly as the
erials. It compares the KWW relaxation times (the Debye computed RDF shows. The experimental gOO of ethanol at
terms are not reported there) for pure ethanol and for solutions 298 K has a sharp maximum at about 0.2 nm, as found by
at decreasing ethanol fraction, with the Debye relaxation time the MD computation, followed by a small peak between
computed for pure water. 0.3 and 0.4 nm. The dpendence on the size of the MD cell

5
Meas. Sci. Technol. 28 (2017) 014003 R Olmi and M Bittelli

Figure 4.  Dielectric constant (left) and dielectric losses (right) of the ethanol–water mixtures (see text).

Figure 5.  The RDF of the O–O pairs for pure water (left) and pure ethanol (right) as a function of the number N of molecules.
Symbols:  +  refers to N  =  216, o to N  =  502, X to N  =  1050, solid line to N  =  2258.

Figure 6.  The ACF of pure water (left) and pure ethanol (right) as a function of the number N of molecules. Dotted lines correspond to
N  =  216, dashed line to N  =  502, dash-dot to N  =  1050, solid line to N  =  2258.

is clearly negligible, as the RDF’s of both water and ethanol is rather moderate, therefore the dielectric relaxation proper-
do not change when the number of molecules increases of an ties of the smaller model are meaningful.
order of magnitude, from 216 to 2258 (and the initial box size
increases from 1.8 to 4.1 nm for water, and from 2.9 to 7 nm 4. Conclusions
for ethanol).
Figure 6 shows the dependence of the normalized ACF of MD simulations, also employing small-size systems, show sev-
pure water (left) and pure ethanol (right) from the box size. eral interesting features of ethanol–water mixtures. In particular,
At least for the limited range of molecule numbers investi- they suggest that the relaxation dynamics of such materials in the
gated (respectively, 216, 502, 1050 and 2258 as specified in time domain can be described by a mixture involving a stretched
the figure caption), the dependence of the ACF on the cell size exponential (or KWW function) and a simple exponential decay,

6
Meas. Sci. Technol. 28 (2017) 014003 R Olmi and M Bittelli

that is strictly related to a mixture of a Debye function and of an mixtures studied by computer simulation J. Chem. Phys.
HN distribution in the frequency domain. 119 7308–17
[15] Saiz L, Guardia E and Padro J 2000 Dielectric properties of
The results obtained on ethanol–water solutions demon-
liquid ethanol. A computer simulation study J. Chem. Phys.
strate that MD could contribute to the understanding of the 113 2814–22
dielectric behavior of complex moist substances. For example, [16] Abraham M J, van der Spoel D, Lindahl E, Hess B and the
it could hopefully provide hints to advance in the comprehen- GROMACS development team 2015 GROMACS User
sion of the controversial problem of the water bound to a Manual version 5.0.7 www.gromacs.org
[17] Berendsen H J C, Grigera J R and Straatsma T P 1987 The
porous surface.
missing term in effective pair potentials J. Phys. Chem.
91 6269–71
[18] Jorgensen W L, Maxwell D S and Tirado-Rives J 1996
References
Development and testing of the OPLS all-atom force field
on conformational energetics and properties of organic
[1] Alder B J and Wainwright T E 1959 Studies in molecular liquids J. Am. Chem. Soc. 118 11225–36
dynamics. I. General method J. Chem. Phys. 31 459 [19] Mark P and Nilsson L 2001 Structure and dynamics of the
[2] Lindahl E, Hess B and van der Spoel D 2001 GROMACS 3.0: TIP3P, SPC, and SPC/E water models at 298 K J. Phys.
a package for molecular simulation and trajectory analysis Chem. A 105 9954–60
J. Mol. Mod. 7 306–17 [20] Gereben O and Pusztai L 2011 On the accurate calculation of
[3] Allen M P 2004 Introduction to molecular dynamics the dielectric constant and the diffusion coefficient from
simulation Computational Soft Matter: From Synthetic molecular dynamics simulations: the case of SPC/E water
Polymers to Proteins, Lecture Notes (NIC Series vol 23) Chem. Phys. Lett. 507 80–3
ed N Attig et al (Julich: John von Neumann Institute for [21] Froehlich H 1958 Theory of Dielectrics (New York: Oxford
Computing (NIC)) University Press)
[4] Allen M P and Tildesley D J 1991 Computer Simulation of [22] Neumann M 1983 Dipole moment fluctuation formulas
Liquids (New York: Oxford University Press) in computer simulations of polar systems Mol. Phys.
[5] Weiner S J, Kollman P A, Case D A, Singh U C, Ghio C, 50 841–58
Alagona G, Profeta S and Weiner P 1984 A new force-field [23] Boettcher C J F and Bordewijk P 1978 Theory of Electric
for molecular mechanical simulatio of nucleic acids and Polarization (Amsterdam: Elsevier)
proteins J. Am. Chem. Soc. 106 765–84 [24] Feldman Y, Puzenko A and Ryabov Y 2006 Dielectric
[6] Brooks B R, Bruccoleri R E, Olafson B D, States D J, relaxation phenomena in complex materials Fractals,
Swaminathan S and Karplus M 1983 CHARMM—a Diffusion, and Relaxation in Disordered Complex Systems:
program for macromolecular energy, minimization and a Special Volume of Advances in Chemical Physics vol
dynamics calculations J. Comput. Chem. 4 187–217 133, ed W T Coffey and Y P Kalmykov (Hoboken, NJ:
[7] Jorgensen W L and Tirado-Rives J 1988 The OPLS force field Wiley)
for proteins. Energy minimizations for crystals of cyclic [25] Neumann M and Steinhauser O 1983 On the calculation of
peptides and crambin J. Am. Chem. Soc. 110 1657–66 the frequency-dependent dielectric constant in computer
[8] Rapaport D C 2004 The Art of Molecular Dynamics simulations Chem. Phys. Lett. 102 508–13
Simulation (Cambridge: Cambridge University Press) [26] Perry R H and Green D W 2007 Perry’s Chemical Engineers’
[9] Ercolessi F 1997 A Molecular Dynamics Primer (Trieste: Handbook (New York: McGraw-Hill)
International School of Advanced Studies (SISSA-ISAS)) [27] Sato T and Buchner R 2004 Dielectric relaxation processes in
[10] Kaatze U, Behrends R and Pottel R 2002 Hydrogen network ethanol/water mixtures J. Phys. Chem. A 108 5007–15
fluctuations and dielectric spectrometry of liquids [28] Sudo S, Shinyashiki N, Kitsuki Y and Yaghihara S 2002
J. Non-Cryst. Solids 305 19–28 Dielectric relaxation time and relaxation time distribution
[11] Dzida M and Kaatze U 2015 Compressibility and dielectric of alcohol–water mixtures J. Phys. Chem. A 106 458–64
relaxation of mixtures of water with monohydroxy alcohols [29] Alvarez F, Alegria A and Colmenero J 1991 Relationship
J. Phys. Chem. B 119 12480–9 between the time-domain Kohlrausch–Williams–Watts and
[12] Noskov S Y, Lamoureux G and Roux B 2005 Molecular frequency-domain Havriliak–Negami relaxation functions
dynamics study of hydration in ethanol–water mixtures Phys. Rev. B 44 7306–12
using a polarizable force field J. Phys. Chem. B [30] Clark G N I, Cappa C D, Smith J D, SayKally R J and Head-
109 6705–13 Gordon T 2010 The structure of ambient water Mol. Phys.
[13] Gereben O and Pusztai L 2015 Investigation of the structure of 108 1415–33
ethanol–water mixtures by molecular dynamics simulation [31] Saiz L, Padro J A and Guardia A 1997 Struncture and
I: analyses concerning the hydrogen-bonded pairs J. Phys. dynamcis of liquid ethanol J. Phys. Chem. 101 78–86
Chem. B 199 3070–84 [32] Berendsen H J C, van der Spoel D and van Drunen R 1995
[14] Wensink E J W, Hoffmann A C, Van Maaren P J and van GROMACS: a message-passing parallel molecular
der Spoel D 2003 Dynamic properties of water/alcohol dynamics implementation Comp. Phys. Commun. 91 43–56

Das könnte Ihnen auch gefallen