Journal of Molecular Liquids 134 (2007) 71 89

www.elsevier.com/locate/molliq

Review
Molecular simulation of the thermophysical properties of fluids:
From understanding toward quantitative predictions
Philippe Ungerer a,, Carlos Nieto-Draghi a , Bernard Rousseau b ,
Gktug Ahunbay b,c , Vronique Lachet a
a
Institut Franais du Ptrole, 1-4 avenue de Bois Prau, 92852 Rueil-Malmaison, France
b
Laboratoire de Chimie Physique, Universit de Paris Sud CNRS, 91405 Orsay, France
c
Chemical Engineering Department, Istanbul Technical University, 34469 Maslak, Istanbul, Turkey
Available online 7 March 2007

Abstract

The purpose of the present paper is to review what kind of thermophysical properties can be predicted, either qualitatively or quantitatively
with molecular simulation. In a first part, the main types of molecular simulation methods are introduced. Molecular dynamics (MD) can be used
to address equilibrium properties and dynamic behaviour as well. Monte Carlo simulation (MC) is particularly adapted to phase equilibria or
physisorption. Both methods require to represent the potential energy, which is classically decomposed into intramolecular (bending, torsion, etc.)
and intermolecular (dispersion, repulsion, electrostatic, polarization) contributions.
In a second part, the prediction of fluid properties (PVT relationships, enthalpy, heat capacity, JouleThomson coefficient, etc.) is reviewed.
Either MC or MD can be used to relate these properties and molecular structure, as shown by examples like high pressure hydrocarbon gases,
CFCs, acid gases, and natural gases.
Fluid phase equilibria are discussed in the third part. Examples are given in which MC is used to provide pure component properties when pure
chemicals are not commercially available, such as heavy hydrocarbons of complex structure. MC is also capable of predicting phase behaviour for
mixtures with little (or no) calibration on binary system data. This aspect is illustrated by the prediction of Henry constants of gases in polar liquids
and by the prediction of phase diagrams of acid gases (H2S, CO2) with water, methanol or hydrocarbons. The self-association of polar molecules
and the critical scaling behaviour are also described.
In the fourth part, transport properties (viscosity, diffusion coefficients, and thermal conductivity) are discussed. For many systems like
hydrocarbons, carbon dioxide or hydrogen sulphide, very good predictions are obtained, and simulation is shown to predict detailed features such
as the differences in viscosity between isomers.
In Conclusion and perspectives, the current limitations and possible improvements of molecular simulation methods are mentioned.
2007 Published by Elsevier B.V.

Keywords: Molecular simulation; Molecular dynamics; Monte Carlo simulation; Fluid phase equilibria; Transport properties; Viscosity

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2. Molecular simulation methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.1. Statistical ensembles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.2. Energy of molecular systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.3. Monte Carlo methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.4. Molecular dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.4.1. Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.4.2. Computing transport properties. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Corresponding author.
E-mail address: presse@ifp.fr (P. Ungerer).

0167-7322/$ - see front matter 2007 Published by Elsevier B.V.

doi:10.1016/j.molliq.2006.12.019
72 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189

3. Equilibrium properties of fluids in the monophasic state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4. Fluid phase equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.1. Pure compounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2. Mixtures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5. Transport properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.1. Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2. Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3. Thermal conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6. Conclusion and perspectives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

1. Introduction simulation is a way to gather the prediction of all thermo-

Molecular simulation considers small size systems, at a
typical scale of a few nanometers, and determines their be-
haviour from a careful computation of the interactions between
their components. The position of most atoms is explicitly
computed, in order to account for the difference in geometry
between individual molecules (Fig. 1). These computations take
much more computer time than classical thermodynamic
models: from a few hours to several weeks of computing
time, depending on system size. Yet, these methods are in-
creasingly used in chemical engineering to compute thermo-
physical properties of poorly known systems, and they allow
also scientists to get a deeper understanding of property dif-
ferences versus molecular structure. Interestingly, molecular
physical properties in a single theoretical framework, which is
statistical thermodynamics [16]. Provided the potential energy
of the system is a derivable function of the molecular coor-
dinates through an adequate model, its thermophysical pro-
perties can be derived without additional hypothesis. This holds
for equilibrium properties like vapor pressure, as well as for
transport properties like shear viscosity. Although it is delicate
to find an adequate potential energy model, this field is suf-
ficiently mature that it can provide reliable predictions in many
instances, and it shows an unprecedented ability to extrapolate
prediction from small to complex molecules, from low to high
pressures, and from low temperatures to high temperatures. As
such, molecular simulation is now considered in the chemical
industry as capable of filling the gap between experimental data

Fig. 1. The two ways of building a statistical ensemble: Molecular dynamics (MD) and Monte Carlo (MC) simulation. Average properties can be determined from time
averages in MD and from ensemble averages in MC. Both averages are equivalent, as a consequence of the ergodicity theorem.
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Table 1
Statistical ensembles
Statistical ensemble Imposed variables
Canonical ensemble N, V, T
Grand Canonical Ensemble i, V, T
Isothermalisobaric ensemble N, P, T
Gibbs Ensemble at imposed global volume (m phases) N N1 N Nm ;
V V1 N Vm ;
T
Gibbs Ensemble at imposed pressure (m phases) N = N1 + Nm, P, T
The probability density is given in phase space, i.e. the space of all positions and momenta. E is total energy (kinetic + potential), N is number of molecules, V is system
volume, T is temperature, P is pressure and i is chemical potential of molecular type i.
and engineering models in various circumstances: unknown
chemicals, extreme conditions of temperature or pressure, toxic
compounds, among others.
The first part of this article provides a brief review of the
main simulation methods applicable to determine the thermo-
physical properties of fluids. These comprise molecular
dynamics (MD), which consists in integrating Newton's
equations of motion for all particles with time, and Monte
Carlo simulation (MC), in which a statistical method is used to
generate representative system configurations on the basis of
the probability distribution of potential energy. The main
features of the most common potential energy models are also
given.
In the second part, we will review the application of mole-
cular simulation to the equilibrium properties of fluids (PVT
relationships, enthalpy, heat capacity, JouleThomson coeffi-
cient). Either MC or MD can be used to relate these properties
and molecular structure, as shown by examples like high pres-
sure hydrocarbon gases, CFCs, acid gases, and natural gases.
In the third part, we will consider fluid phase equilibria, for
which MC is the most efficient technique. It can be applied to
the prediction of the vaporliquid equilibrium properties of
pure compounds on the basis of their molecular structure, if the
potential energy model has been previously parametrized.
Concerning mixtures, we will see that the distinction between
the major forms of energy (repulsion, dispersion, electrostatic)
may result in excellent predictions of phase diagrams for
mixtures involving polar fluids like water or methanol. We will
show also how these techniques may be used to model the near-
critical behaviour of fluid mixtures and to characterize
association phenomena.
Transport properties can be addressed with MD, the results
of which can be processed to determine transport coefficients:
viscosity, diffusion coefficients, and thermal conductivity. We
will show representative examples, and discuss which type of
potential energy model might be capable of providing general
answers in this field which has been less systematically ex-
plored than equilibrium properties.
Finally, we will briefly quote other applications of molecular
simulation in the field of interfacial phenomena and adsorption
73
Probability density Applications
exp( E) Phase properties (P, H, Cv, , )
! Adsorption isotherms, selectivities
X
exp bE b li Ni
i
exp( E PV) Phase properties (H, C(p, , , )
exp( E) Phase equilibrium of pure components and mixtures
exp( E PV) Phase equilibrium of mixtures
in microporous adsorbents. This technique is also expected to
provide key interpretations and predictions, because the
relevant length scales (interface thickness or pore diameter)
are comparable to molecular size.
2. Molecular simulation methods
2.1. Statistical ensembles
Depending on the application desired, different statistical
ensembles can be used (Table 1), each ensemble being
characterized by the constrained variables and by its probabil-
ity density, i.e. the probability of occurrence of each system state
in the ensemble [1]. For a given problem, the selection of the
type of ensemble is generally made in such a way that the
constraints correspond to variables that are controlled in the
experimental set-up. The variables that are not constrained are
fluctuating, and their statistical averages provide predictions
that may be compared with experimental results.
The canonical ensemble, in which the temperature, number
of molecules and volume are imposed, is used for monophasic
fluids when density is known. The predicted average properties
are then average energy, pressure, and chemical potential.
In the isothermalisobaric or NPT and ensemble, pressure is
imposed instead of volume, and the average volume of the
system is used to predict fluid density. This ensemble is used,
for instance, when the properties of a fluid are to be determined
at known pressure and temperature.
The ensemble which is the most widely used for phase
equilibrium calculations is the Gibbs ensemble [7] in which two
phases are introduced without explicit interface. This may be
done either by imposing the global volume of the two phases, or
the pressure [3]. The temperature and the total number of
molecules are also imposed in the Gibbs ensemble. When
applied to pure compounds, this ensemble is applied as constant
volume to respect the phase rule. Its outputs are then the
equilibrium properties: saturated vapor pressure, vaporization
enthalpy, and the coexisting densities of the liquid and vapor. In
the case of mixtures, the Gibbs ensemble may be applied either
at imposed volume or at imposed pressure. In both cases, it
74 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Fig. 2. Intermolecular and intramolecular interaction potentials for flexible molecules.
provides the equilibrium compositions and densities of coex-
isting phases.
The ensemble that is most adapted to adsorption in
microporous solids is the Grand Canonical ensemble, where
the temperature, the volume, and the chemical potential of each
species are the imposed variables. For such applications, the
computation must account for the interaction energy between
each molecule and the microporous adsorbent. The most desired
output is the average number of molecules of every species in
the system, i.e. the amount adsorbed.
Beside common average properties like density, pressure or
energy, the analysis of fluctuations allows to determine
thermodynamic properties like the heat capacity, the compress-
ibility, the thermal expansion coefficient or the JouleThomson
coefficient [8,9].
Practically, there are two main methods to simulate statistical
ensembles (Fig. 1). The first is through molecular dynamics,
which consists in solving the equations of motion, and the
second is Monte Carlo simulation, in which a statistical method
is used.
Once a large number of configurations has been generated,
the statistical average of property X (such as volume, energy, )
may be computed by an arithmetic average over all configura-
tions:
1X n
hX i Xi
n i1
Similarly, thermodynamic derivative properties can be
obtained from statistical fluctuations. For instance, heat ca-
pacity at constant pressure, Cp, is obtained from the fluctuations
of energy in the isothermalisobaric ensemble
1   ibj
Cp 2
hE PV 2 ihE PV i2 2
kB T
2.2. Energy of molecular systems
Although the total energy E appearing in the probability
densities of Table 1 is the sum of the kinetic energy and potential
energy, only the latter needs to be provided as an explicit func-
tion of molecular coordinates through a suitable intermolecular
potential energy model. In MD, this is sufficient to compute the
forces acting between molecules or parts of molecules, and
kinetic energy is an outcome of the simulation. In MC, an
analytical integration of the contribution of the kinetic energy is
performed, so that it does not need to be explicitly considered.
As summarized in Fig. 2, the potential energy is decomposed
classically into intermolecular (or external) energy and intra-
molecular energy:
U Uint Uext 3
The intermolecular energy arises from the interaction
between different molecules. In the applications considered
here, it is decomposed as: Uext = Udr + Uel + Upol where Udr
arises from dispersion and repulsion interactions, and Uel is the
electrostatic (or coulombic) interaction energy and Upol is the
polarization energy [10].
The dispersionrepulsion energy, which is dominant in low
polarity systems like alkanes, is obtained by a summation over
all the pairs of force centers. Although the Buckingham exp-6
1
model has been shown to be more accurate on several important
systems [1113] the most popular model is still the Lennard
Jones 612
X  12  6 !
rij rij
Udr 4eij 4
rij rij
i; j
where rij is the separation distance between force centers.
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Potential energy models considering individual atoms as
force centers are referred to as All Atoms [1416]. A specific
version of the All Atoms model has been proposed in which the
hydrogen force centers are located at the mid-point of the
carbonhydrogen bond [17]. If the force center represents a
group of atoms and is located on the main atom of the group, the
model is said United Atoms [1820]. Alternatively, the force
center may be located at an intermediate position between the
atoms belonging to the group (Fig. 2), and the model is then
known as Anisotropic United Atoms or AUA [2123]. The
parametrization of the AUA intermolecular potential for
hydrocarbons, which is used for several examples discussed
in the present article, is outlined in Fig. 3.
The electrostatic energy is computed directly from the
Coulomb law, assuming that the molecules bear electrostatic
point charges
X 1 qi qj
Uel
4peo rij
i; j
ibj
where the sum runs over all possible pairs i, j of partial charges
qi, qj, rij is the separation distance and o = 8.8541910 12 C2
N 1 m 2. Point charges may be determined from dipole
moment or quadrupole moment for small molecules like N2 or
CO2 but they are more frequently regressed from quantum
chemistry results [15,2426].
The polarization energy may be computed on the basis of the
molecular polarizabilities, which are known from experimental
measurements.
The intramolecular energy is assumed to be the sum of four
terms (Fig. 2): 1 the stretching energy associated with bond
length variations 2 the bending energy arising from the var-
Fig. 3. Parametrization of the Anisotropic United Atoms intermolecular potential for hydrocarbons. The LennardJones parameters for each group have been
determined from the saturation pressure, vaporization enthalpy and liquid density of 14 hydrocarbons comprising three n-alkanes [22], isobutane [75], two cyclic
alkanes [30], two aromatics [7678] and six olefins [79].
75
iation of the angle between two successive chemical bonds, 3
the torsion energy arising from the variation of the dihedral
angle defined by three successive chemical bonds, 4 the
dispersionrepulsion between distant atoms which prevents
overlaps within the same molecule. When simulating the
thermophysical properties of common fluids, the bond lengths
are often considered constant (they are taken from X-ray dif-
fraction results), and the stretching energy is accordingly
neglected. This approximation allows to use longer time steps in
molecular dynamics, and thus shorter computing time. However
it should be noted that vibrational temperatures are often much
higher than investigated temperatures (more than 1000 K) so
that the corresponding degrees of freedom are not fully ac-
tivated from a quantum mechanical point of view. A classical
description of bond stretching using a harmonic potential is thus
an approximation that may lead to wrong vibrational energies.
Bending and torsional parameters are determined to match
5 equilibrium conformation angles, infrared spectroscopy data
and quantum chemistry results [15]. In flexible molecules, these
contributions are summed over all bending and torsion angles in
the molecule. If a molecule (or a part of molecule) is considered
rigid, these contributions are not taken into account.
2.3. Monte Carlo methods
The basis of Monte Carlo (MC) methods is to generate
successive configurations of the simulated system by a statis-
tical method to respect the probability distribution of the desired
statistical ensemble. This method describes the configuration
space, in which only positions are considered, momenta (or
velocities) being treated implicitly. Each configuration is gen-
erated by applying an elementary change (or MC move) to the
previous configuration of the chain. In order to sample correctly
76 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
the desired equilibrium distribution, it is sufficient to respect the
microreversibility criterion, which states that the flux of con-
figurations from state i to state j is equivalent of the inverse flux
from j to i:
qi pij qj pji 6 The determination of energetic properties must take into
where i is the probability of configuration i, which is pro-
portional to the Boltzmann weight exp( U), and pij is the
transition probability from configuration i to j. More precisely,
the probability density in the configuration space is
V N expbU
qi ~ 7
KN N !
where is the de Broglie wavelength, which includes the
kinetic contribution to the probability distribution, and U is the
potential energy, including both inter- and intramolecular terms.
In a Monte Carlo move, the microscopic reversibility crite-
rion is used to determine appropriate transition probability from
i to j. The Metropolis algorithm provides the more commonly
way to satisfy this requirement
 
qj
pij min 1; 8
qi
If we consider the example of a translation move in the
canonical ensemble, the test configuration is accepted with the
following acceptance probability
pij min1; expbDU 9 where =U + PV is the configurational enthalpy and Na =
where U is the variation in potential energy when changing
from configuration i to j. The Metropolis algorithm may be also
adapted to bias the generation of the new test configuration, i.e.
to generate more frequently favorable configurations that have
an increased probability to be accepted. These statistical bias
are now standard in Monte Carlo simulation [3] to handle either
systems with preferred orientation [27], flexible linear chains
[18,28], flexible branched molecules [29] and flexible cyclic
molecules [12,30]. Indeed, it is a common problem when
complex molecules or dense phases are considered that the
acceptance probability of some Monte Carlo moves is very low.
Numerous different types of Monte Carlo moves are con-
sidered in our applications, so that complex molecules can be
considered: translation or rotation of a molecule, volume change
of the simulation box, transfer of a molecule from one box to the
other with two types of statistical bias (Gibbs ensemble),
destruction of an existing molecule or insertion of a new one
(Grand Canonical ensemble), regrowth of a part of a molecule
by configurational bias (flexible chain), rotation of an atom
around the axis formed by its immediate neighbours (flexible
molecules), reptation (destruction of one end of a chain and
growth of the other end), displacement (destruction of an
existing molecule and insertion of a new one in the same box),
pivot, i.e. rotation of the part of a molecule located beyond some
atom, around this atom (flexible chains).
In addition to these moves, it is also possible to perform test
insertions of a molecule in the simulation box with two types of
statistical bias. The analysis of the interaction energy of the test
particle with the rest of the system allows to determine the
chemical potential of the species [31]. This is useful because
chemical potential cannot be obtained by simple thermody-
namic averages like volume or energy.
account the fact that the kinetic energy is not computed in MC,
while it contributes to properties like enthalpy and heat capacity.
Thus the enthalpy must be obtained by summing the residual
enthalpy, which can be obtained from MC, and the ideal gas
enthalpy which may be taken from experimental data or group
contribution methods [8,32]. However the calculation of the
vaporization enthalpy does not require the knowledge of the
ideal contribution, as the latter is assumed identical in the liquid
and vapor phases [33].
Similarly, the residual heat capacity obtained by MC from
fluctuations, cpres, must be combined with the ideal heat capacity
cpid (i.e. the heat capacity in absence of intermolecular inter-
actions) to obtain the total heat capacity at constant pressure, cp
[8]
cp T ; P cres
p T ; P cp T
id
10
Na D E D E
p
cres U ext H hUext i H
kB NT 2 D E D E
Na P
2
V H hV i H Na kB 11
kB NT
6.023 10 23 mol 1 is Avogadro's number.
Combined with a similar evaluation of the thermal expansivity,
this allows to compute also the JouleThomson coefficient.
Finally, an interesting Monte Carlo technique to investigate a
system at several state points (for instance m different temper-
atures) is parallel tempering, also known as replica exchange
[34,35]. In this technique, m systems are simulated simulta-
neously, and they can exchange their temperatures with an
appropriate acceptance probability. As a result, one configura-
tion visits successively all the temperatures. The increased
sampling efficiency is useful to simulate fluids at low temper-
atures, because a single simulation would then be trapped in a
restricted, poorly representative region of configuration space.
For instance, it may be used in conjunction with thermodynamic
integration [36] to extend vapor pressure calculations well
below the normal boiling point, a region where Gibbs ensemble
calculations are hampered by the very low acceptance proba-
bility of transfer moves because of the high density of the liquid
phase.
2.4. Molecular dynamics
2.4.1. Principles
In contrast to the Monte Carlo procedure (MC), Molecular
Dynamics (MD) follow the time evolution of a molecular
system by numerically integrating Newton's equations of
motion for a set of N particles [2,6]. The result is a trajectory
that specifies how the positions and velocities of the particles
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
vary with time. In many respects, MC and MD are equivalent
techniques. Identical results should be obtained for structural
and thermodynamic properties. However, dynamic properties
can be studied only with MD, either at equilibrium or far from
equilibrium.
In MD, the dynamics of molecules is usually assumed to
obey classical equations of motion, which is a satisfactory
approximation for translational motion as the de Broglie
wavelength of a typical molecule is much less than the average
distance between molecules, i.e.
Kt V =N 1=3 12
h
where Kt p is the translational de Broglie wavelength. SHAKE [40] or Gauss constraints [41] which maintains bond
2kmkB T
In this limit, there is a continuous distribution of accessible
energy states. Rotational motion can also be treated classically
provided the rotational energy spacing is much less than kBT.
These criteria are true for the vast majority of molecules under
ordinary conditions and the classical approximation is routinely
used in MD. As indicated above, vibrational degrees of freedom
may be affected by quantum effects. However, they are either
removed (for instance by considering constant bond lengths)
or treated classically with a harmonic potential. The description
of the dynamics follows from the Hamiltonian of the system H
(rN, pN)
X p2
HrN ; pN i
U rN 13
i
2m i
where rN and pN represent the set of particle positions and
momenta, mi the particle mass and U(rN) the potential energy
function as defined in Section 2.2. The Hamiltonian equations
of motion are given by
AH p dp AH jr U rN
rd i i 14
Api mi i
Ari i
The equations of motion are integrated numerically using a
finite-difference method. The Verlet based or the Gear
predictorcorrector algorithms are the most commonly used
integrators because they can be easily implemented in NVE and
other statistical ensembles [2]. Repeatedly integrating for
several thousand times produce individual trajectories from
which time averages A can be computed for macroscopic
properties from instantaneous values of any mechanical
property A(t)
Z where vi(t), xy(t) and Jq(t) are respectively particle i velocity,
t0 t
1 pressure tensor component and heat flux. In the case of heat
hAi lim Asds 15
t Yl t t0
At equilibrium, this average cannot depend on the initial time
t 0.
The solution of the above Hamiltonian equations of motion
samples the microcanonical NVE ensemble. In order to com-
pare MD results with experimental data, it is often desirable to
work at conditions where temperature and/or pressure are held
constant. Different techniques have been developed to enable
77
simulations in the desired statistical ensembles, namely NVT
and NPT ensembles. The most widely used method consists in
adding external degrees of freedom (called extended Hamilto-
nian methods) to the system in order to describe the coupling
between the set of particles and a heat or pressure bath [37,38].
Other methods based on the Gauss constraint principle [39] or
simple velocity and/or particle rescaling are also used.
Another important feature of molecular dynamics concerns
the description of the internal degrees of freedom of flexible
molecules, like alkanes. Stretching, bending and torsion
motions can be described using intramolecular potential func-
tions (see Section 2.2). However, in most applications the fastest
modes can be freezed using a constraint algorithm like
distances to a constant value. Equations of motion can therefore
be integrated using a larger time step, hence enabling simu-
lations on a larger time scale.
2.4.2. Computing transport properties
Different methods exist to compute transport coefficients in
MD. From a practical point of view, the simplest approach
makes use of equilibrium MD simulations and of the Green
Kubo (GK) relationships [42]. The GK relationships relate
equilibrium fluctuations of the fluxes to the corresponding
phenomenological (or Onsager) coefficients Lij
Z l
V
Lij hJi tJj 0ieq dt 16
3kB 0
Given a microscopic and instantaneous expression for the
fluxes Ji(t), one can compute the Lij's from a single equilibrium
run. Within the framework of irreversible processes, the Lij's
can therefore be connected with the transport coefficients [43].
For example, self-diffusion D, shear viscosity and thermal
conductivity can be computed at equilibrium from the fol-
lowing expressions
Z lX
N
1
D hvi tdvi 0idt 17
3N 0 i1
Z l
V
g hrxy trxy 0idt 18
kB T 0
Z l
V
k hJq tdJq 0idt 19
3kB T 2 0
transfer, it should be noticed that there is no microscopic ex-
pression for heat flux. The thermal conductivity in mixtures is
given from the integral of the autocorrelation of the heat flux,
given as
X
Jq J e hi Ji 20
i
where Ji is the mass flux of particles i, hi is the specific enthalpy
of species i and Je, is the internal energy flux whose
78 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
microscopic expression was given by Irving and Kirkwood
[44]. Because of the presence of the enthalpic contribution in
the heat flux, there cannot be a microscopic expression for Jq
in mixtures. An approximation was proposed for computing Jq,
in ideal mixtures [45] but it does not apply to non-ideal systems.
This precludes the use of the GreenKubo method for the
prediction of thermal conductivity in non-ideal mixtures, unless
it is associated with a suitable procedure for evaluating partial
enthalpies from simulation results [46,47].
Using linear response theory, Evans and Moriss [48]
proposed another way of computation of the transport
coefficients called synthetic non-equilibrium molecular dynam-
ics. In this method, one introduces an external perturbation in
the (generally) non-Hamiltonian equations of motion. This
perturbation generates a non-equilibrium flux which can be
related to the equilibrium correlation function of fluxes and
therefore to the Lij. Because a (strong) perturbation is applied to
the system, the response is large and is obtained with a good
Fig. 4. Properties of methane according to the LennardJones intermolecular
potential of Mller et al. [53], compared with reference data gathered by IUPAC
[159] or with the reference equation of Setzmann and Wagner [160] (a)
Compressibility factor Z = PV / RT at 294.25 K and 377.55 K. (b) Residual heat
capacity at 294.25 K.
Fig. 5. JouleThomson coefficient of difluoromethane along isenthalps [54].
signal to noise ratio. However, the Lij must be determined from
the limit at zero perturbation and several simulations are
necessary in order to obtain the desired property. Just like
GreenKubo simulations, one also has to convert from Lij to
transport coefficient if comparison is desired with experimental
measurements, and this can be a problem for e.g. thermal
conductivity.
A third way to compute transport coefficient is to mimic the
physical phenomenon and drive the system out of equilibrium
by modifying the boundary conditions of the simulation box
[49,50]. The transport coefficients are then obtained from the
phenomenological relations (i.e. Fourier's law for thermal
conductivity, etc.). For example, a heat flux can be created in a
simulation box by exchanging kinetic energy between a hot and
a cold layer. At the stationary state, a thermal gradient is
established within the simulation box and the heat flux can be
computed from the rate of exchange of the kinetic energy per
surface area per unit time. The thermal conductivity is then
given by the ratio of the heat flux to the thermal gradient [49]. It
should be emphasized here that this method gives direct access
to the transport coefficients and not to the Lij. Viscosity is also a
property that can be conveniently determined by non-equilib-
rium molecular dynamics [50,51].
Generally speaking, non-equilibrium methods are more
efficient than equilibrium methods but they are problem-
specific, i.e. each transport coefficient requires a separate
simulation. If one is interested in several transport coefficients
(say D, and ) for a given fluid, then equilibrium MD might
be more efficient.
3. Equilibrium properties of fluids in the monophasic state
In the monophasic state, Monte Carlo methods in the NPT
ensemble allow first to determine the pressurevolume rela-
tionship, as illustrated in Fig. 4a for methane [8]. For low
pressures, the compressibility factor Z is decreasing as a result
of attractive interactions. This is followed by an increase of Z
which is the consequence of the hard core volume of the
methane molecules. The residual heat capacity of methane Cpres
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
(i.e. the difference between the real and the ideal heat capacity at
constant pressure) shown in Fig. 4b is an example of
thermodynamic derivative property that can be computed
from the analysis of energy fluctuations in NPT calculations
[8]. The maximum of the residual heat capacity is a
phenomenon which is linked with the strong variations of
molar volume in the supercritical region. Simulation has been
also shown to describe qualitatively the maximum of Cv, the
heat capacity at constant volume for a constant temperature
(220 K) when density increases [52]. This maximum, which
announces the non-classical anomaly of Cv in the critical
region, is not reproduced with standard equations of state. It is
remarkable that these results have been obtained with a very
simple model, as it involves only the two parameters of the
LennardJones potential [53].
Such calculations are not restricted to pure fluids of simple
structure like methane. For instance, the JouleThomson
Fig. 6. Top: Molecular modelling of a natural gas with 500 molecules of 19
different types. Bottom: JouleThomson coefficient at 463 K following IUPAC
recommendations for methane (dots), and from Monte Carlo simulations
(dashed line: methane; dotted line: ethane; full line: natural gas) [9].
79
coefficient of a refrigerant, difluoromethane, has been comput-
ed along several isenthalps, using an original Monte Carlo
procedure to simulate a statistical ensemble at imposed pressure
and enthalpy [54]. An intermolecular potential involving elec-
trostatic point charges was used, in which the dispersion and
repulsion parameters were calibrated against equilibrium pro-
perties. As shown in Fig. 5, the agreement between simulation
results and reference data is excellent over a large range of
conditions. Other simulations conducted on carbon dioxide
[55,56] have shown that the whole JouleThomson inversion
curve i.e. the locus of the points where the sign of the Joule
Thomson coefficient changes in the P,T diagram is accurately
predicted. The use of statistical fluctuations for computing the
JouleThomson coefficient is also applicable to mixtures [8]. In
Fig. 6, it is applied to a mixture of 19 components representing a
natural gas. This natural gas contains dominantly methane
(65 mol%) and a complex distribution of hydrocarbons up to 30
carbon atoms [57]. At the reservoir temperature of 463 K, the
inversion of the JouleThomson effect is predicted to occur at a
pressure of 42 MPa [9]. This prediction was confirmed by
volumetric measurements. These results, associated with previ-
ous work [58] confirm the feasibility of selecting a represen-
tative component for each fraction of a complex fluid, so that
most of the fluid behaviour is captured and yet the simulation is
tractable.
4. Fluid phase equilibria
4.1. Pure compounds
For chemical engineering applications, a good prediction of
the liquidvapor equilibrium (VLE) properties of pure com-
pounds is important for the design and optimization of sepa-
ration processes, among others. With adapted Monte Carlo
techniques such as the Gibbs ensemble, these properties can be
determined now for flexible or rigid molecules, either non-polar
or strongly polar. However, the accuracy of the equilibrium
properties largely depends on the intermolecular potential.
Giving a complete account of all the intermolecular poten-
tials that have been developed for the VLE of small molecules is
beyond the scope of this paper. However we can cite methane
[53], carbon dioxide [59], hydrogen sulphide [60], or methanol
[61] as examples of molecules of various polarities for which
VLE is well described with simple models involving Lennard
Jones potential, in which electrostatic charges are included to
represent polar or quadrupolar interactions. Potential parameters
have been also determined in a systematic way to describe the
VLE of many small molecules, either quadrupolar [62] or
dipolar [63]. In the specific case of water, several models
provide reasonable predictions [6466], but no simple model is
able to reproduce accurately the VLE properties over the whole
temperature range up to the critical point.
Intermolecular potentials have been also developed for com-
plex organic molecules involving alkyl chains, cycles, func-
tiona1 groups, etc. It has been often recognized that the force
fields which had been developed primarily to describe the liquid
phase properties at ambient temperature, such as the OPLS
80 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Fig. 7. Saturation pressures computed with the AUA potential for various
hydrocarbons, compared with experiment-based correlation from the DIPPR
data bank.
potential [15] do not always represent the whole coexistence
curve because the critical temperature is poorly reproduced
[25,67]. This deficiency, coupled with attempts to decrease the
computing time, has led to the development of force fields
based on United Atoms in which the LennardJones parameters
of each group are regressed to describe the coexisting densities
of the compounds considered. The first of these was the SKS
potential for n-alkanes [18]. It was followed by the TRAPPE
potential for n-alkanes [19], branched alkanes [68], aromatics
and olefins [69], alcohols [70], other oxygen- and nitrogen-
containing functions [71,72]. Another potential based on United
Atoms with the LennardJones functional form, NERD, was
developed for n-alkanes [20], branched alkanes [73], and alpha-
Fig. 8. Saturated liquid density computed with the AUA potential for various
hydrocarbons, compared with experiment-based correlation from the DIPPR
data bank.
olefins [74]. A United Atoms model, using the Buckingham
exp-6 functional form, was also developed for n-alkanes,
benzene and cyclohexane [12,13]. In our group, we have
concentrated on the Anisotropic United Atoms potential
(AUA), for which the LennardJones parameters have been
regressed for n-alkanes [22], branched alkanes [75], cyclic
alkanes [30], aromatics [7678], olefins [79] and sulphur-,
oxygen- or nitrogen-bearing functions [25,26,80]. In Figs. 79,
a comparison of experimental data with predictions using the
AUA-potential is provided for a series of hydrocarbons of
various families which were not part of the reference molecules
on which the AUA parameters were calibrated. Thus the com-
puted vapor pressures (Fig. 7), saturated liquid densities (Fig. 8)
and vaporization enthalpies (Fig. 9) are predictions from the
sole input of molecular structure, in the same way as we would
do for new compounds.
An important property of intermolecular potentials, transfer-
ability, is the ability to predict the contribution of each group
independently from the other groups present in the molecules.
Transferability can be tested by considering molecules contain-
ing several different groups. For instance, the behaviour of
propylbenzene is well described with the benzene-derived
parameters for the aromatic CH group and the alkane-derived
parameters for the CH2 and CH3 groups of the alkyl chain. It is
interesting to note that the differences in properties between the
three heptane isomers (n-heptane, 2-methylhexane and 2,4-
dimethylhexane) are qualitatively reproduced. It is also
remarkable that propylbenzene is predicted to display a lower
liquid density than styrene (Fig. 8), while the contrary would
be expected from the general tendency of density to increase
with increasing molecular weight. The influence of alkyl sub-
stituents to reduce the density of alkylaromatics may be
Fig. 9. Vaporization enthalpy computed with the AUA potential for various
hydrocarbons, compared with experiment-based correlation from the DIPPR
data bank.
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189 81

Table 2
Comparison of critical parameters and boiling temperatures determined by simulation with the AUA potential and experimental measurements
Compound Critical temperature Critical density Normal boiling temperature
(K) (kg/m3) (K)
Exp. Simulation Exp. Simulation Exp. Simulation
Ethane 305.4 311 202 210 184 nd
n-pentane 469.7 469 237 218 307.7 307
n-heptane 540.3 547 232 231 372 376
n-decane 617.7 616 236 225 447 445
n-dodecane 658.2 652 239 217 489 481
n-pentadecane (708) nd (250) nd 543.1 541
n-eicosane (768) nd (251) nd 617 609
Isobutane 407.8 nd 229 nd 261.4 268
2-methylhexane 530.4 537 227 238 362 362
2,4-dimethylpentane 519.8 529 240 247 354 357
Cyclopentane 511.8 507 271 275 322.4 311
Cyclohexane 553.6 559 273 271 353.9 352
Cyclooctane 647.2 647 283 303 424.3 nd
Ethylene 282.3 280 214 212 169.2 166
Propene 365 373 223 226 225.1 224
1-butene 419.5 420 233 236 267.1 262
1-pentene 464.7 462 234 232 303.3 298
1-hexene 504 499 237 241 336.6 328
1-octene 567.1 566 241 240 394.4 388
cis-2-butene 435.6 448 240 238 276.8 279
trans-2-butene 428.6 439 236 234 274 265
isobutene 417.9 426 234 232 266.2 260
Butadiene 425 422 245 237 268.6 264
trans-2-pentene 475 480 234 243 309.5 316
Benzene 562.2 558 302 302 353.2 350
Toluene (methylbenzene) 591.8 580 292 307 383.8 376
p-xylene (1,4 dimethylbenzene) 616.2 598 280 303 411.5 403
o-xylene (1,2 dimethylbenzene) 630.3 610 289.1 311 417.2 401
1,3,5 trimethylbenzene 637.4 622 280.2 285 437.8 415
n-propylbenzene 638.2 627 273 281 432.4 421
Naphthalene (748) 716 (315) 323 487.8 491
Reference compounds.
Values in parentheses are likely results of extrapolations rather than true measurements [99].

explained by their larger CC bond length, which is 1.535 as
compared to 1.40 in aromatic rings. More generally, the
good agreement between simulation and experimental results in
Figs. 68 illustrates the good reliability of the proposed
potential for a large range of molecules, as it comprises olefins,
naphthenoaromatics, alkylaromatics, etc.
A first application of these intermolecular potentials is to
compute the properties of organic compounds that are not
commercially available in the pure form. This is the case for
instance of most branched alkanes beyond ten carbon atoms,
and this has motivated some systematic studies by molecular
simulation with the NERD potential [81]. The TRAPPE
potential has been also used for predicting the properties of
heavy isoalkanes [82,83]. The Anisotropic United Atoms po-
tential has been applied in several industrial projects to obtain
the properties of poorly known compounds including organic
sulphides and thiols [25], ketones and aldehydes [80], heavy
isoalkanes and alkyl-substituted polyaromatics [84].
A characteristic feature of molecular simulation is that it
cannot be used in the close vicinity of the critical point, because
it would be necessary to simulate systems of excessively large
size that are incompatible with reasonable computing times.
Indeed, the characteristic size of density or energy fluctuations
increases when the critical point is approached, and larger
systems should be considered [85]. In the case of a pure com-
pound, the critical coordinates may be obtained by assuming the
following scaling law
qV qL AT Tc b 21
where 0.325 is a universal critical exponent, V and L are
the densities of the coexisting vapor and liquid phases, and Tc is
the critical temperature. This relationship may be used to regress
the critical temperature and critical density from Gibbs ensem-
ble calculations performed at lower temperatures [3]. An alter-
native, more rigorous method consists in using finite size
scaling [85,86]. Table 2 provides the critical coordinates of
various hydrocarbons according to AUA potential.
4.2. Mixtures
The second major application of well calibrated force fields
is of course the determination of phase diagrams for real mix-
tures, and we will mention only a few among the numerous
82 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Fig. 10. Gibbs ensemble simulations of the H2Scyclohexane system at 423 K,
compared with experimental data [161]. The full lines indicate the near-critical
correlation used to close the phase diagram [99]. (a) Pressurecomposition
diagram (b) coexisting densities (c) density difference (l v) versus (Pc
P)0.325.
contributions to this field. Mixtures of noble gases have been
investigated among others, by Panagiotopouplos [87] and by
Delhommelle [88], the latter questioning the usual rules that are
used in evaluating the cross interaction parameters of the
LennardJones potential. The mixtures of hydrocarbons of
various molecular weights have been also investigated by
several authors, generally with good results [8994]. While
hydrocarbons and noble gases can be treated with non polar
force fields limited to dispersionrepulsion forces, electrostatic
interactions are generally introduced to investigate the mixtures
of alkanes with polar gases like hydrogen sulphide or carbon
dioxide [92,9598]. As an example of Gibbs ensemble
simulations of such mixtures, we indicate in Fig. 10 the phase
diagram of the binary system H2Scyclohexane, using a stan-
dard potential with point charges for H2S [60] and the AUA
potential for cyclohexane [30]. The knowledge of this kind of
diagram is necessary to design the production of H2S-rich oil
and gas fields. It is clear from Fig. 10 that the pressurecom-
position phase diagram is very well described. Since H2S is
supercritical at the temperature considered (423 K) the phase
diagram displays a liquidvapor critical point. In the case of
binary mixtures, the determination of critical coordinates is
more complex than with pure fluids as the critical composition
must be determined. This can be made by finite size scaling [86]
or by near-critical correlations similar to the above equation [99]
as we have done in Fig. 10. Any analytical model, such an
equation of state would have predicted a density difference
between liquid and vapor varying like (Pc P)1/2 near the
critical point [100]. Simulation results, in agreement with ex-
perimental observations on other mixtures [101103], indicate
that the density difference is varying rather like (Pc P). The
critical point cannot be approached closely by simulation, even
using large systems as we did (up to 1200 molecules at high
pressure). Nevertheless, the (Pc P) dependence can be used to
provide a reliable estimation of the critical locus and to close the
phase diagram with appropriate functional forms.
When more polar compounds like alkanols or water are
considered, the electrostatic contribution plays a major role and
allows often simulation to predict phase diagrams with good
success, while classical thermodynamic models would require a
calibration of specific interaction parameters from mixture VLE
data. This is illustrated by the mixtures involving isobutene,
methanol and methyl-tert-butyl ether [104], alkanols and
alkanes [92] alkanols and CO2 [92,105], methanol and H2S
[99], alkanes and water [66,106], water and CO2 [107,108],
water and H2S [96,109], alkanols and water [105]. A major
outcome of the second AIChE Fluid Simulation Challenge was
that several independent groups were able to provide solubilities
in ethanol (or more exactly Henry constants) for nitrogen,
oxygen, methane and CO2 with fair accuracy [26,110,111].
However, standard force fields are sometimes unable to predict
the phase diagrams or excess enthalpies of mixtures involving
hydrogen bonding between different molecules unless partial
electrostatic charges are specifically determined for the corre-
sponding hydrogen-bonded dimer [26,112,113]. As an example
of prediction where both species are significantly polar, we
show in Fig. 11 the phase diagram of the methanolH2S system
[99]. Here, the force fields for methanol [61] and for H2S [60]
have been taken from the literature. In this system, the two-
phase region shows a smaller extension than in the alkane
methanol systems [99,105] or in the H2Swater system [109],
the latter showing a liquidliquid phase split. It is encouraging
that simple force fields allow to quantify the differences in
phase behaviour between these systems [99]. These differences
appear to be the consequence of molecular polarity and hydro-
gen bonding. Moreover, simulation results can be used to
quantify methanol or water clustering, the cluster size distri-
bution being obtained as a result of microscopic interactions.
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
5. Transport properties
Prediction of transport properties of fluids with industrial
interest is one of the main motivations to perform MD
simulations. The comparisons of simulation results with the
available experimental data become the key point to determine
which intermolecular potential model should be employed to
analyze a particular system. In this section we intend to present
a series of cases where MD simulations have been employed to
compute transport properties such as shear viscosity, self-
diffusion and thermal conductivity of different systems. Our
goal is to show illustrative examples to the reader where
different models have been applied to reproduce transport
properties at different thermodynamic conditions and to a wide
range of systems.
5.1. Viscosity
Simple linear hydrocarbons are the first type of systems of
our list of examples. In Fig. 12 we present our simulation results
for the effect of temperature variations on the shear viscosity of
ethane, n-pentane and n-dodecane computed with the AUA4
model [22] at low-moderate pressures (0.1 to 5 MPa). A
remarkable agreement is observed between the simulations and
the experimental data [114,115] despite the wide range of
temperatures employed, particularly for the case of ethane and
n-pentane. For the case of n-dodecane, it is worth to mention
the ability of the AUA4 to reproduce the strong change in the
viscosity experimentally observed [114] between 298 K and
473 K. Here, the average absolute deviation between the
simulation and the experimental data at this condition is of the
order of 20%.
Though the effect of temperature changes is well reproduced
in simulations, the effect of changes in the pressure of the
system is more subtle. In Fig. 13 we can observe the pressure
variation of the shear viscosity of n-octane at 348 K using the
UA-TraPPE [19,68] model using the non-equilibrium MD
method from the work of Kioupis et al. [116] and our simulation
data for n-hexane at 333 K using the AUA4 model computed
with equilibrium MD in the NPT ensemble. Despite the
Fig. 11. Phase diagram of the H2Smethanol system mixture at 348.15 K from Gibbs ensemble simulations, compared with experimental data [162].
83
differences in models and methodology, and the fact that both
examples underestimate the experimental value, the trend in
viscosity with pressure is captured reasonably well for both
simulations. Deviations become more important when the
pressure of the system is increased, the difference being more
pronounced for the AUA model. In this case the average
absolute deviation between simulations and experiments is of
the order of 40%. Dysthe et al. [117] have found that the mean
deviations on diffusion and viscosity of different UAAUA
models analyzed over different states were of the order of 24%
for AUA2, 26% for SKS2, 30% for AUA3, 48% for OPLS1 and
54% for SMMK [118]. In general, all models tend to under-
estimate the viscosity and to overestimate the self-diffusion.
The deviations increase when the density of the system is
increased [45,119] as it is observed at high pressure in the
examples of n-octane and n-hexane in Fig. 13. However, there
are certain cases where simple empiric intermolecular potentials
are robust enough to predict well the behaviour or fluids in a
wide range of conditions. This aspect is of particular interest
when the fluid is toxic and corrosive, or when the domain of
application requires extreme conditions of pressure and
temperature or when the experimental determination of
physicochemical data requires expensive equipment. This is
the case of hydrogen sulphide that is encountered in a
significant amount in certain oil reservoirs. Nieto-Draghi et al.
[120] have investigated the transport properties of H2S along the
liquid vapor coexisting line using equilibrium MD simulations
at constant pressure (NPT). In Fig. 14 we can observe the
comparison of the simulation results (using the model of Kristof
et al. [60]) and the experimental data [121] of the shear viscosity
of H2S at the LV equilibrium. The agreement between the
experimental data and the simulation results is very good. In this
case the maximum deviation is lower than 20%, the difference
being more important at the low temperature and high density
liquid phase. This example shows the prediction capacity of
molecular simulation method in direct industrial applications
such as mixtures of H2S, CO2 and hydrocarbons in oil and gas
production, for instance.
Molecular simulation techniques can also be used as sub-
stitute of experiments, for instance, in the understanding of the
84 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Fig. 12. Temperature variation of the shear viscosity of ethane (experiments
[115], MD simulations (), at 98.15 K; 173.15 K; 248.15; 298.15 K and
5 MPa), n-pentane (experiments [115], MD simulations (), at 148.15 K;
248.15 K; 298.15 K; 373.15 K; 473.15 K and 5 MPa) and n-dodecane
(experiments -- [114], MD simulations (), at 298.15 K, 373.15 K, 473.15 K
and 0.1 MPa).
relationship between molecular architecture and dynamic
properties. The case of poly--olefins (PAO) is illustrative
since these compounds are commonly used in synthetic motor
oil base sticks [122] in order to control the reduction of the
viscosity of lubricants as the temperature is increased [123]. In
particular the number and location of branch points in PAO
isomers have a large effect on the temperature dependence of
diffusivity and viscosity. Kioupis et al. [116] have investigated
the variation of viscosity of different branched iso-PAO mole-
cules with pressure. In Fig. 15 we can observe the comparison
of MD simulation results of the pressure dependence of shear
viscosity of three C18 isomers (highly branched 4,5,6,7-
tetraethyldecane (HB), star-like 7-butyltetradecane (S) and
linear n-octadecane (L)) at 473 K. The Newtonian viscosity
has been computed using the synthetic SLLOD NPT non-
equilibrium MD scheme (which impose Couette shear flow
conditions) and using the UA-TraPPE force field. The
viscosities of the three isomers increase with pressure. The
interesting result is that the L and S isomers present similar
pressure dependence, while the HB fluid presents greater
Fig. 13. Pressure variation of the shear viscosity of n-octane at 348 K
(simulations (), experimental data (---)[116]) and n-hexane at 333 K
(simulations (), experimental data () [115]).
increase in viscosity with pressure. It is important to understand
why HB isomers present a higher viscosity than the other two
isomers when its average radius of gyration (Rg2 = 10.15 2) is
almost a half of the S (Rg2 = 18.49 2) isomer and a third of the L
(Rg2 = 27.92 2) isomer. In fact, molecules with more stiff
backbone structure and with high cohesive energy (more dense
at the same thermodynamic conditions) present higher viscosity,
as it is the case of the HB PAO isomer. Then the motion of
molecules in the liquid relies on the ability to make jumps
between voids in the fluid or by displacing neighbouring
molecules to these voids. Species with low intramolecular
flexibility will find more difficult to do these jumps or
accommodate the movement of adjacent molecules than flexible
ones. Following the work of Kioupis, it is clear that the internal
dynamics or intramolecular mobility plays a more important
role than the size of the molecule in the value of the viscosity at
the same conditions. This case is a good example of how MD
simulations provide a useful tool to enhance the understanding
of the physical properties that are certainly inaccessible
experimentally.
Although many studies have been devoted to the transport
properties of linear or branched alkanes, or alkane mixtures,
much less has been done on aromatics. Rousseau et al. [124]
have investigated the transport properties of xylene isomers. As
m-xylene and p-xylene display very close shear viscosities and
o-xylene is significantly more viscous, the study of these
isomers represents a severe test of the predictive power of
molecular dynamics, mostly of the force fields. Therefore, the
effects of pressure and temperature on viscosity of the three
isomers of xylene have been investigated using equilibrium MD
and the GreenKubo formalism in the temperature and pressure
range of 298348 K and 0.1100 MPa, respectively. The
xylene isomers are described as multisite rigid molecules. Two
different force fields have been used: one derived from the
OPLS set for pure liquid substituted benzene, proposed by
Jorgensen et al. [125,126] the OPLS model and an AUA model
by Contreras et al. [77]. The OPLS model has been obtained by
merging the LennardJones parameters and partial charges of
an all-atom model of benzene and LennardJones parameters
Fig. 14. Comparison of the experimental shear viscosity () [115] of H2S along
the liquidvapor coexistence curve with MD simulations () using the NPT
ensemble [120].
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
Fig. 15. Molecular dynamic simulations results of the pressure dependence of
the shear viscosity of poly--olefin hydrocarbon C18 isomers at 200 C [116].
Three isomers have been analyzed: highly-branched 4,5,6,7-tetraethyldecane
(), star 7-butyltetradecane () and n-octadecane ().
for the substituent from some other OPLS set. The charge on the
substituent, the methyl groups, is equal to the partial charge on
the substituted hydrogen. The AUA model is much simpler and
considers only neutral united atom sites.
Viscosity data are presented on Fig. 16 along with
experimental data from Et Tahir [127] the accuracy of which
is better than 2%. The relative order of viscosities is reproduced
with AUA and OPLS potentials. However, the AUA model
underestimates the viscosity, presumably due to the lack of
electrostatic interactions. This potential reproduces also the
influence of temperature (not shown). A similar conclusion was
reached by Rowley and Ely about benzene [128]. The OPLS
force field is able to quantitatively reproduce the viscosity data
over the full range of pressure (Fig. 16) and temperature [124],
with a maximum deviation with experimental data smaller than
6%.
5.2. Diffusion
Diffusion of molecules is another interesting property of
fluids that has its direct application in the framework of
Fig. 16. Influence of pressure at T = 323 K on the viscosity of xylene isomers
using the OPLS model [124] (open symbols) and the AUA model (this work,
full symbols) compared with experimental data [127] (full lines).
85
chemical reaction kinetics and process design, among others.
Hydrocarbons represent a good example to test the predictive
capabilities of methods, intermolecular potentials and algo-
rithms. Recently, Krishna et al. [129] have compared MD
simulation results of self-diffusion coefficients of pure n-
alkanes (from C1 to C10) with experimental data [130] at 333 K
and 30 MPa. In this work a UA potential [131] has been
employed together with NVT ensemble and the density was
fixed to reproduce the pressure of each system in the state
considered. The agreement between the experimental data and
simulation results is very good as can be seen in Fig. 17,
particularly for small hydrocarbons. The self-diffusion of n-
alkanes decreases with the chain length. Simulation data slightly
overestimate the experimental self-diffusion coefficient for the
heaviest molecules but, in general, the global trend is well
reproduced for the model employed. In Fig. 18 we can observe
the variation of the self-diffusion Di and the MaxwellStefan
(MS) mutual diffusion coefficient ij of the system C1nC6
(methanen-hexane mixture) as a function of C1 molar fraction
at 333 K and 30 MPa. In this figure, simulation results (empty
symbols) are compared with the experimental data [130] (full
symbols) and the Darken approximation (line, ij = x2D1 +
x1D2), showing a good general agreement. For the case of Di,
the values obtained in the simulations for C1 are closer to the
experiments than those computed for n-C6 but, in any case the
deviation observed is less than 5%. Unfortunately, the MS
coefficient is not experimentally accessible and no direct
comparison between simulation and experimental data is
possible. However, the values of ij computed in the
simulations follow the same trend as those obtained using the
Darken relationship [132]. As mentioned before, the knowledge
of the diffusion of different species in a mixture is essential to
understand diffusion controlled reaction mechanism, particu-
larly in cases where the size difference between the solvent and
the solute is considerable, as in the case for the termination step
on free radical polymerization reactions. Harmandaris et al.
[133] have compared MD simulations of very asymmetric
mixtures (large differences in size between solute and solvent)
to test the free volume theory of Vrentas and Duda [134,135]
(VD) and the chain free volume theory proposed by Bueche and
Fig. 17. Comparison of the self-diffusion of pure alkanes of 110 carbons
obtained with MD simulations () and the experimental data () of Helbaek
[130] at 333 K and 30 MPa.
86 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
von Meerwall [136,137] (BM). Using the NERD full flexible
model [20] and Verlet multiple time step XI-RESPA [138]
algorithm with a NosHoover thermostat, Harmandaris et al.
computed the self-diffusion coefficient of a mixture of nC12
nC60 [133]. The variation of Di for each compound as a function
of the mass fraction (w1) of C12 at 403.15 K can be observed in
Fig. 19. The agreement between the regime of intermediate
values of w1. This fact is not surprising, since the ratio D1/D2 is
observed to remain constant over the whole concentration
range, exhibiting the ideal solution behaviour predicted by the
BM theory.
5.3. Thermal conductivity
Among the different transport coefficients, thermal conduc-
tivity is one of the most difficult to obtain using molecular
dynamics and many publications have been devoted to
methodological efforts to compute in dense fluids and fluid
mixtures [140148]. Thermal conductivity is sensitive to the
internal energy of the molecules. The relaxation of the internal
conformations of molecules should therefore be taken into
account as well. For long alkanes, one can consider that the
longest relaxation time is given by the end-to-end vector
correlation function decay. This quantity reaches up to 200 ps
for n-decane close to the triple point. Moreover, this relaxation
time scales with the square of the number of carbon atoms per
molecule (Rouse regime). In order to correctly sample the phase
space of the system, it is important to run simulations for several
relaxation times. Although the standard GreenKubo approach
is the safe way, it suffers from rather long convergence times for
the flux autocorrelation function and many authors have
proposed non-equilibrium methods to ensure faster
convergence.
Dysthe et a1. [117,149] have computed the thermal conduc-
tivity of several alkanes, from n-butane to n-hexadecane at
different state points and using different force fields. Some of
their data are summarized in Table 3. The simulation data
obtained with the AUA2 [149] and OPLS [14] force fields are
compared with the Assael correlations for n-alkanes [150]. The
Fig. 18. Comparison of MD simulations [129] of self-diffusion D1,self (C1, ),
D2,self (nC6, ) and MaxwellStefan mutual diffusion 12 () with the Darken
relation and experimental data (C1 , C6 ) of Helbaek [130] in function of the
molar fraction of C1 at 333 K and 30 MPa.
Fig. 19. Comparison of the self-diffusion coefficient of C12 and C60 molecules in
a C12C60 blend at 403.15 K at a function of the mass composition W1 [133].
Data for C12, simulations () and experiments (); and for C60, simulations ()
and experimental data ( ). Lines are the prediction of the Buechevon
Meerwall theory [136,139].
statistical uncertainties for thermal conductivity are 5%10%.
Assael correlation is known to represent all experimental data
on viscosity, thermal conductivity and diffusion to experimental
accuracy as long as one does not extrapolate outside the region
of validity. As can be seen in Table 3, simulation results
compare rather well with Assael results. For the AUA model,
average deviation from the Assael correlation is roughly 6%
(not taking into account n-hexadecane). The predictions are
worse for the OPLS model which systematically underpredicts
the thermal conductivity and presents a 14% deviation from
Assael correlation. Thermal conductivity predictions are less
sensitive to the interaction potential models and to thermody-
namic conditions than viscosity and self-diffusion. Therefore,
optimizing interaction potential may be more relevant based on
a comparison with viscosity and self-diffusion rather than
thermal conductivity.
6. Conclusion and perspectives
As illustrated by the examples showed in this brief overview,
molecular simulation has reached a stage where it can be
reliably used for a large array of equilibrium and transport
properties of molecular liquids.
Thanks to the development of efficient algorithms to allow
an efficient exploration of configuration space (statistical bias,
parallel tempering), MC methods are now applicable to a large
range of equilibrium properties. However there are still cases
that are difficult to investigate with standard methods. For
instance, the investigation of liquidliquid phase equilibria at
temperatures much lower than the normal boiling point of either
component would require more powerful MC moves such as
staged transfer moves [151]. This occurs systematically with
high molecular weight compounds. The promising extension of
phase equilibrium simulations from liquids to amorphous
polymeric materials is another example where the difficulty of
exploring correctly the configuration space requires specific
moves [152]. Although it did not fit in the scope of this review,
the application of MC simulation to the adsorption of molecular
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189 87

Table 3 Onsager coefficients to transport coefficients, which is ne-

Thermal conductivity for different n-alkanes as computed by equilibrium cessary with Einstein or GreenKubo methods. However, long
molecular dynamics at different state points
relaxation times always require long simulation times to explore
(W/m/K) T(K) Molecular simulations Assael sufficiently the phase space, and this puts practical limits to the
correlation
(kg/m )
3
AUA OPLS range that can be addressed by simulation. More precisely,
[P (MPa)] quantitative determination of transport coefficients is increas-
n-butane 510.9 0.083 0.001 0.074 0.002 0.072 ingly difficult for high molecular weights, because larger sys-
500.0 [63.8] tem size is associated with longer relaxation time.
150.0 0.186 0.006 0.173 0.005 0.181
Another difficulty in computing transport coefficients is the
732.3 [24.1]
n-decane 559.9 0.078 0.002 0.057 0.002 0.071 availability of adequate intermolecular potentials. As illustrated
510.9 [5.5] by the viscosity of xylene isomers and of n-alkanes, viscosity
300 0.137 0.002 0.103 0.004 0.134 predictions in dense liquids seem more sensitive to detailed
724.7 [0.1] features of the potential than equilibrium properties. For
293.2 0.166 0.005 0.138 0.004 0.168
instance, significant changes of viscosity are found when
783.6 [100.0]
n-hexadecane 563/574.8 0.080 0.003 NA 0.064 changing from All Atoms to United Atoms potentials. On one
hand, it appears that relaxation times are systematically
Results for the AUA2 [149] and OPLS [14] force fields are compared with
Assael correlations [150].
underestimated with United Atoms potentials, as if suppressing
some roughness of the true molecule was reducing the friction
with its neighbours. On the other hand, relaxation times are
fluids in microporous materials is a very active field (see [99] sometimes significantly overestimated by All Atoms potentials.
and references therein). Another growing field of application of Is it possible that versatile potentials can be parametrized to
MC methods is the direct simulation of fluid interfaces [153 provide quantitative predictions of equilibrium and transport
157] which allows to provide a detailed understanding of properties? This is of course highly desirable, but it is still an
interface structure as well as a microscopically based prediction open challenge.
method for the interfacial tension.
Although difficulties remain in the exploration of configu- Acknowledgements
ration space, the most stringent limitation of MC simulations is
rather the availability of well-tested potentials with a large range The authors express their thanks to Total for its support to
of application. Small molecules have been thoroughly investi- several research projects in which molecular simulation results
gated but unless specific calibration is performed on mixture were obtained.
data, the accuracy of predictions decreases when changing from
pure compounds to mixtures with standard combining rules. References
With more complex organic molecules, the force fields cali-
brated on phase equilibrium data like TRAPPE, NERD or AUA [1] D.A. McQuarrie, Statistical Mechanics, Harper and Collins, New York,
1976.
allow fair predictions, and they can be used efficiently to [2] M.P. Allen, D.J. Tildesley, Computer Simulation of Liquids, Oxford
supplement lacking data for poorly known systems. However, Science Publications, Oxford, 1987.
improvements are still possible in many respects. As far as [3] D. Frenkel, B. Smit, Understanding Molecular Simulation, Academic
hydrocarbons and low polarity compounds are considered, the Press, San Diego, 1996.
[4] R.L. Rowley, Statistical Mechanics for Thermophysical Property
transferability to large molecules combining several types of
Prediction, Prentice Hall, 1994.
structures (such as polycyclic hydrocarbons with attached alkyl [5] C.G. Gray, K.E. Gubbins, Theory of Molecular Fluids, Oxford Science,
chains, for instance) has been little investigated. With respect to Oxford, 1984.
polar compounds, we have mentioned that the use of fixed [6] J.M. Hayle, Molecular Dynamics Simulation Elementary Methods,
electrostatic point charges and standard combining rules may John Wiley & Sons, 1992.
not reproduce the behaviour of mixtures as well as it does with [7] A.Z. Panagiotopoulos, Mol. Phys. 61 (1987) 813.
[8] M. Lagache, P. Ungerer, A. Boutin, A.H. Fuchs, Phys. Chem. Chem.
pure compounds. It is likely that in the future, the mixtures of Phys. 3 (2001) 4333.
polar fluids, and particularly hydrogen-bonding fluids, will [9] M. Lagache, P. Ungerer, A. Boutin, Fluid Phase Equilib. 220 (2004) 211.
require an explicit account of polarization energy, as performed [10] J.-L. Rivail, Elments de Chimie Quantique l'usage des chimistes,
in recent models [158]. However this will require substantial InterEditions/CNRS Editions, Paris, 1994.
reparametrization. [11] J.R. Errington, A.Z. Panagiotopoulos, J. Chem. Phys. 109 (1998) 1093.
[12] J.R. Errington, A.Z. Panagiotopoulos, J. Chem. Phys. 111 (1999) 9731.
In molecular dynamics, the range of systems for which [13] J.R. Errington, A.Z. Panagiotopoulos, J. Phys. Chem., B 103 (1999)
transport properties can be reliably computed is also very 6314.
significant. Equilibrium MD, associated with GreenKubo or [14] W.L. Jorgensen, J.D. Madura, J. Am. Chem. Soc. 106 (1984) 6638.
Einstein procedures, is a versatile tool that allows several [15] W.L. Jorgensen, D.S. Maxwell, J. Tirado-Rives, J. Am. Chem. Soc. 118
transport coefficients to be computed with the same tool, if not (1996) 11225.
[16] H. Sun, J. Phys. Chem., B 102 (1998) 7338.
with the same simulation. Property-specific non-equilibrium [17] B. Chen, I.J. Siepmann, J. Phys. Chem. 103 (1999) 5370.
MD may be more efficient for properties like thermal con- [18] B. Smit, S. Karaborni, I.J. Siepmann, J. Chem. Phys. 102 (1995) 2126.
ductivity because it avoids the delicate task of converting from [19] M.G. Martin, J.I. Siepmann, J. Phys. Chem., B 102 (1998) 2569.
88 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189
[20] S.A. Nath, F.A. Escobedo, J.J. dePablo, J. Chem. Phys. 108 (1998) 9905.
[21] S. Toxvaerd, J. Chem. Phys. 93 (1990) 4290.
[22] P. Ungerer, C. Beauvais, J. Delhommelle, A. Boutin, B. Rousseau, A.H.
Fuchs, J. Chem. Phys. 112 (2000) 5499.
[23] S. Toxvaerd, J. Chem. Phys. 107 (1997) 5197.
[24] B. Lvy, M. Enescu, J. Mol. Struct. 432 (1998) 235.
[25] J. Delhommelle, C. Tschirwitz, P. Ungerer, G. Granucci, P. Milli, D.
Pattou, A.H. Fuchs, J. Phys. Chem., B 104 (2000) 4745.
[26] Y. Boutard, P. Ungerer, J.M. Teuler, M.G. Ahunbay, S.F. Sabater, J.
Perez-Pellitero, A.D. Mackie, E. Bourasseau, Fluid Phase Equilib. 236
(2005) 25.
[27] R.F. Cracknell, D. Nicholson, N.G. Parsonage, Mol. Phys. 71 (1990) 931.
[28] J.J. de Pablo, M. Laso, U.W. Suter, J. Chem. Phys. 96 (1992) 6157.
[29] M.D. Macedonia, E.J. Maginn, Mol. Phys. 96 (1999) 1375.
[30] E. Bourasseau, P. Un'gerer, A. Boutin, J. Phys. Chem., B 106 (2002)
5483.
[31] B. Widom, J. Chem. Phys. 39 (1963) 2808.
[32] F.A. Escobedo, Z. Chen, Mol. Simul. 26 (2001) 395.
[33] T. Spyriouni, I.G. Economou, D.N. Theodorou, J. Am. Chem. Soc. 121
(1999) 3407.
[34] Q. Yan, J.J. dePablo, J. Chem. Phys. 111 (1999) 9509.
[35] C. Beauvais, X. Guerrault, F.-X. Coudert, A. Boutin, A.H. Fuchs, J. Phys.
Chem., B 108 (2004) 399.
[36] D.A. Kofke, J. Chem. Phys. 98 (1993) 4149.
[37] S. Nos, J. Chem. Phys. 81 (1984) 511.
[38] W.G. Hoover, Phys. Rev., A 31 (1985) 1695.
[39] D.J. Evans, G.P. Morris, Chem. Phys. 77 (1983) 63.
[40] J.P. Ryckaert, G. Ciccotti, H.J.C. Berendsen, J. Comput. Phys. 23 (1977)
327.
[41] R. Edberg, D.J. Evans, G.P. Morriss, J. Chem. Phys. 84 (1986) 6933.
[42] R. Kubo, J. Phys. Soc. Jpn. 12 (1957) 570.
[43] S.R. deGroot, P. Mazur, Non-Equilibrium Thermodynamics, Dover
Publications, New York, 1984.
[44] J.H. Irwing, J.G. Kirkwood, J. Chem. Phys. 18 (1950) 817.
[45] D. Dysthe, A.H. Fuchs, B. Rousseau, J. Chem. Phys. 110 (1999) 4060.
[46] P. Sindzingre, G. Ciccotti, C. Massobrio, D. Frenkel, Chem. Phys. Lett.
136 (1987) 35.
[47] B. Hafskjold, T. Ikeshoji, Fluid Phase Equilib. 104 (1995) 173.
[48] D.J. Evans, G.P. Morriss, Statistical Mechanics of Nonequilibrium
Liquids, Academic Press, 1990.
[49] A. Tenenbaum, G. Ciccotti, R. Gallico, Phys. Rev., A 25 (1982) 2778.
[50] D.R. Wheeler, N.G. Fuller, R.L. Rowley, Mol. Phys. 92 (1997) 55.
[51] G. Galliero, C. Boned, A. Baylaucq, F. Montel, Fluid Phase Equilib. 234
(2005) 56.
[52] P. Ungerer, Oil Gas Sci. Technol. 58 (2003) 271.
[53] D. Mller, J. Oprzynski, A. Mller, J. Fischer, Mol. Phys. 75 (1992) 363.
[54] M. Lisal, W.R. Smith, K. Aim, Mol. Phys. 101 (2003) 2875.
[55] C.M. Colina, M. Lisal, F.R. Siperstein, K.E. Gubbins, Fluid Phase
Equilib. 202 (2002) 253.
[56] C.M. Colina, C.G. Oliveira-Fuentes, F.R. Siperstein, M. Lisal, K.E.
Gubbins, Mol. Simul. 29 (2003) 405.
[57] P. Ungerer, B. Faissat, C. Leibovici, H. Zhou, E. Behar, G. Moracchini, J.P.
Courcy, Fluid Phase Equilib. 111 (1995) 287.
[58] B. Neubauer, B. Tavitian, A. Boutin, P. Ungerer, Fluid Phase Equilib. 161
(1999) 45.
[59] J.G. Harris, K.H. Yung, J. Phys. Chem. 99 (1995) 12021.
[60] T. Kristof, J. Liszi, J. Phys. Chem., B 101 (1997) 5480.
[61] M.E. van Leeuwen, B. Smit, J. Phys. Chem. 99 (1995) 1831.
[62] J. Vrabec, J. Stoll, H. Hasse, J. Phys. Chem., B 105 (2001) 12126.
[63] J. Stoll, J. Vrabec, H. Hasse, Fluid Phase Equilib. 209 (2003) 29.
[64] W.L. Jorgensen, J. Chandrasekhar, J.D. Madura, J. Chem. Phys. 79
(1983) 926.
[65] G.C. Boulougouris, I.G. Economou, D.N. Theodorou, J. Phys. Chem., B
102 (1998) 1029.
[66] J.R. Errington, G.C. Boulougouris, I.G. Economou, A.Z. Panagiotopou-
los, D.N. Theodorou, J. Phys. Chem., B 102 (1998) 8865.
[67] B. Chen, M.G. Martin, J.I. Siepmann, J. Phys. Chem., B 102 (1998) 2578.
[68] M.G. Martin, J.I. Siepmann, J. Phys. Chem., B 103 (1999) 4508.
[69] C.D. Wick, M.G. Martin, J.I. Siepmann, J. Phys. Chem., B 104 (2000)
8008.
[70] B. Chen, J.J. Potoff, J.I. Siepmann, J. Phys. Chem., B 105 (2001) 3093.
[71] J.M. Stubbs, J.J. Potoff, J.I. Siepmann, J. Phys. Chem., B 108 (2004)
17596.
[72] C.D. Wick, J.M. Stubbs, N. Rai, J.I. Siepmann, J. Phys. Chem., B 109
(2005) 18974.
[73] S.K. Nath, J.J. dePablo, Mol. Phys. 98 (2000) 231.
[74] S.K. Nath, B.J. Banaszak, J.J. dePablo, J. Chem. Phys. 114 (2001) 3612.
[75] E. Bourasseau, P. Ungerer, A. Boutin, A.H. Fuchs, Mol. Simul. 28 (2002)
317.
[76] O. Contreras-Camacho, P. Ungerer, A. Boutin, A.D. Mackie, J. Phys.
Chem., B 108 (2004) 14109.
[77] O. Contreras-Camacho, V. Lachet, M.G. Ahunbay, J. Perez, P. Ungerer,
A. Boutin, A.D. Mackie, J. Phys. Chem., B 108 (2004) 14115.
[78] M.G. Ahunbay, J. Perez-Pellitero, R.O. Contreras-Camacho, J.M. Teuler,
P. Ungerer, A.D. Mackie, J. Phys. Chem., B 109 (2005) 2970.
[79] E. Bourasseau, M. Haboudou, A. Boutin, A.H. Fuchs, P. Ungerer,
J. Chem. Phys. 118 (2003) 3020.
[80] S. Kranias, D. Pattou, B. Levy, A. Boutin, Phys. Chem. Chem. Phys. 5
(2003) 4175.
[81] J.T. Wescott, P. Kung, S.K. Nath, Fluid Phase Equilib. 208 (2003) 123.
[82] N.D. Zhuravlev, J.I. Siepmann, Fluid Phase Equilib. 134 (1997) 55.
[83] N.D. Zhuravlev, M.G. Martin, J.I. Siepmann, Fluid Phase Equilib. 202
(2002) 307.
[84] M.G. Ahunbay, S. Kranias, V. Lachet, P. Ungerer, Fluid Phase Equilib.
224 (2004) 73.
[85] N. Wilding, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip.
Topics 53 (1995) 926.
[86] J.J. Potoff, A.Z. Panagiotopoulos, J. Chem. Phys. 109 (1998) 10914.
[87] A.Z. Panagiotopoulos, Int. J. Thermophys. 10 (1989) 447.
[88] J. Delhommelle, P. Milli, Mol. Phys. 99 (2001) 619.
[89] T. Spyriouni, I. Economou, D.N. Theodorou, Phys. Rev. Lett. 80 (1998)
4466.
[90] J. Delhommelle, A. Boutin, A.H. Fuchs, Mol. Phys. 96 (1999) 1517.
[91] P. Ungerer, A. Boutin, A.H. Fuchs, Mol. Phys. 99 (2001) 1423.
[92] J.J. Potoff, J.R. Errington, A.Z. Panagiotopoulos, Mol. Phys. 97 (1999)
1073.
[93] F. Escobedo, J. Chem. Phys. 110 (1999) 11999.
[94] S.K. Nath, B.J. Banaszak, J.J. dePablo, Macromolecules 34 (2001) 7841.
[95] J. Delhommelle, A. Boutin, A.H. Fuchs, Mol. Simul. 22 (1999) 351.
[96] P. Ungerer, A. Wender, G. Demoulin, E. Bourasseau, P. Mougin, Mol.
Simul. 30 (2004) 631.
[97] S. Nath, J. Phys. Chem., B 107 (2003) 9498.
[98] J. Vrabec, J. Fischer, AIChE J. 43 (1997) 212.
[99] P. Ungerer, B. Tavitian, A. Boutin, Applications of Molecular Simulation
in the Oil and Gas Industry Monte Carlo Methods, Editions Technip,
Paris, 2005.
[100] J.M.H. Levelt-Sengers, Ind. Eng. Chem. Fundam. 9 (1970) 470.
[101] J.C. Rainwater, F.R. Williamson, Int. J. Thermophys. 7 (1986) 65.
[102] M.R. Moldover, J.C. Rainwater, J. Chem. Phys. 88 (1988) 7772.
[103] J.J.-C. Hsu, N. Nagarajan, R.L. Robinson, J. Chem. Eng. Data 30 (1985)
485.
[104] M. Lisal, W.R. Smith, I. Nezbeda, J. Phys. Chem., B 103 (1999) 10496.
[105] M. Lisal, W.R. Smith, I. Nezbeda, Fluid Phase Equilib. 181 (2001) 127.
[106] I. Economou, Fluid Phase Equilib. 183184 (2001) 259.
[107] M. Lisal, W.R. Smith, K. Aim, Fluid Phase Equilib. 226 (2004) 161.
[108] J. Vorholz, V.I. Harismiadis, B. Rumpf, A.Z. Panagiotopoulos, G.
Maurer, Fluid Phase Equilib. 170 (2000) 203.
[109] J. Vorholz, B. Rumpf, G. Maurer, Phys. Chem. Chem. Phys. 4 (2002)
4449.
[110] T. Schnabel, J. Vrabec, H. Hasse, Fluid Phase Equilib. 233 (2005) 134.
[111] E.C. Cichkowski, T.R. Schmidt, J.R. Errington, Fluid Phase Equilib. 236
(2005) 58.
[112] G. Kamath, G. Georgiev, J.J. Potoff, J. Phys. Chem., B 109 (2005) 19463.
[113] R.C. Rizzo, W.L. Jorgensen, J. Am. Chem. Soc. 121 (1999) 4827.
[114] D. Cadwell, J.P.M. Trusler, V. Vesovic, W. Wakeham, Int. J. Thermophys.
25 (2004) 1339.
 P. Ungerer et al. / Journal of Molecular Liquids 134 (2007) 7189 89

[115] NIST, Thermophysical properties of pure fluids, http://webbook.nist.gov. [138] G.J. Martyna, M.E. Tuckerman, D.J. Tobias, M.L. Klein, Mol. Phys. 87
[116] L.I. Kioupis, E.J. Maginn, J. Phys. Chem., B 104 (2000) 7774. (1996) 1117.
[117] D. Dysthe, A.H. Fuchs, B. Rousseau, J. Chem. Phys. 112 (2000) 7581. [139] E. van Meerwall, J. Feick, R. Ozisik, W.L. Mattice, J. Chem. Phys. 111
[118] J.I. Siepmann, M.G. Martin, C.J. Mundy, M.L. Klein, Mol. Phys. 90 (1999) 750.
(1997). [140] D. McGowan, D.J. Evans, Phys. Lett., A 117 (1986) 414.
[119] D. Dysthe, A.H. Fuchs, B. Rousseau, J. Chem. Phys. 110 (1999) 4047. [141] D.J. Evans, Phys. Lett., A 91A (1982) 457.
[120] C. Nieto-Draghi, A.D. Mackie, J. Bonet-Avalos, J. Chem. Phys. 123 [142] P.J. Daivis, D.J. Evans, Mol. Phys. 68 (1989) 1219.
(2005) 014505. [143] P.J. Daivis, D.J. Evans, Mol. Phys. 81 (1994) 1289.
[121] C. Ho, P. Liley, T. Makita, Y. Tanaka, Properties of Inorganic and Organic [144] P.J. Daivis, D.J. Evans, Chem. Phys. 198 (1995) 25.
Fluids, Cindas Series on Material Properties, Hemisphere, New York, [145] F. Mller-Plathe, D. Reith, Comput. Theor. Polymer Sci. 9 (1999) 203.
1988. [146] D. Bedrov, G.D. Smith, J. Chem. Phys. 113 (2000) 8080.
[122] High viscosity index synthetic lubricant compositions, USA, 4 827 064. [147] C. Nieto-Draghi, J.B. Avalos, Mol. Phys. 101 (2004) 2303.
[123] B.J. Hamrock, Fundamentals of Fluid Film Lubrication, McGraw Hill, [148] T. Terao, F. Mller-Plathe, J. Chem. Phys. 122 (2005) 081103.
New York, 1994. [149] P. Padilla, S. Toxvaerd, J. Chem. Phys. 94 (1991) 5650.
[124] B. Rousseau, J. Petravic, J. Phys. Chem., B 106 (2002) 13010. [150] M.J. Assael, J.H. Dymond, M. Papadaki, P.M. Patterson, Int. J. Thermophys.
[125] W.L. Jorgensen, E.R. Laird, T.B. Nguyen, J. Tirado-Rives, J. Comput. 13 (1992) 269.
Chem. 14 (1993) 206. [151] D. Kofke, Mol. Phys. 102 (2004) 405.
[126] W.L. Jorgensen, T.B. Nguyen, J. Comput. Chem. 14 (1993) 195. [152] D.N. Theodorou, Mol. Phys. 102 (2004) 147.
[127] A. Et-Tahir, Dtermination des variations de la viscosit de divers [153] F. Goujon, P. Malfreyt, A. Boutin, A.H. Fuchs, Mol. Simul. 27 (2001) 99.
hydrocarbures en fonction de la pression et de la temprature. Etude [154] F. Goujon, P. Malfreyt, J. Chem. Phys. 116 (2002) 8106.
critique de modles reprsentatifs., PhD thesis, Universit de Pau et des [155] B. Chen, J.I. Siepmann, M.L. Klein, J. Am. Chem. Soc. 124 (2002)
Pays de l'Adour, Pau, 1993. 12232.
[128] R.L. Rowley, J.F. Ely, Mol. Phys. 75 (1992) 713. [156] L. Partay, P. Jedlovsky, G. Horvai, J. Phys. Chem., B 109 (2005) 12014.
[129] R. Krishna, J.M. van Baten, Ind. Eng. Chem. Res. 44 (2005) 6939. [157] L. Partay, P. Jedlovszky, A. Vincze, G. Horvai, J. Phys. Chem., B 109
[130] M. Helbaek, B. Hafskjold, D.K. Dysthe, G.H. Sorland, J. Chem. Eng. (2005) 20493.
Data 41 (1996) 598. [158] E. Bourasseau, J.B. Maillet, L. Mondelain, P.M. Anglade, Mol. Simul. 31
[131] D. Dubbeldam, S. Calero, T.J.H. Vlugt, R. Krishna, T.L.M. Maesen, B. (2005) 705.
Smit, J. Phys. Chem., B 108 (2004) 12301. [159] W. Wagner, K.M. de Reuck, Methane, International Thermodynamic
[132] L.S. Darken, Trans. Inst. Min. Metall. Eng. 175 (1948) 184. Tables of the Fluid State, International Union of Pure and Applied
[133] V.A. Harmandaris, D. Angelopoulou, V.G. Mavrantzas, D.N. Theodorou, Chemistry (IUPAC), Blackwell Science, Oxford, 1996.
J. Chem. Phys. 116 (2002) 7656. [160] U. Setzmann, W. Wagner, J. Phys. Chem. Ref. Data 20 (1991) 1061.
[134] J.S. Vrentas, J.L. Duda, J. Polym. Sci. 15 (1977) 403. [161] S. Laugier, D. Richon, J. Chem. Eng. Data 40 (1995) 153.
[135] J.S. Vrentas, J.L. Duda, J. Polym. Sci. 15 (1977) 417. [162] A.D. Leu, J.J. Carroll, D.B. Robinson, Fluid Phase Equilib. 72 (1992)
[136] F. Bueche, Physical Properties of Polymers, Interscience, New York, 163.
1962.
[137] E. van Meerwall, S. Beckman, J. Jang, W.L. Mattice, J. Chem. Phys. 108
(1998) 4299.