You are on page 1of 23

Chem Soc Rev

View Article Online


REVIEW ARTICLE View Journal

Computer modelling of the surface tension of the


gas–liquid and liquid–liquid interface
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

Cite this: DOI: 10.1039/c5cs00736d


Aziz Ghoufi,a Patrice Malfreyt*b and Dominic J. Tildesleyc

This review presents the state of the art in molecular simulations of interfacial systems and of the
calculation of the surface tension from the underlying intermolecular potential. We provide a short
account of different methodological factors (size-effects, truncation procedures, long-range corrections
and potential models) that can affect the results of the simulations. Accurate calculations are presented
for the calculation of the surface tension as a function of the temperature, pressure and composition by
considering the planar gas–liquid interface of a range of molecular fluids. In particular, we consider the
challenging problems of reproducing the interfacial tension of salt solutions as a function of the salt
Received 25th September 2015 molality; the simulations of spherical interfaces including the calculation of the sign and size of the
DOI: 10.1039/c5cs00736d Tolman length for a spherical droplet; the use of coarse-grained models in the calculation of the
interfacial tension of liquid–liquid surfaces and the mesoscopic simulations of oil–water–surfactant
www.rsc.org/chemsocrev interfacial systems.

1 Introduction where A is the Helmholtz free energy and A is the surface area.
The corresponding expression at constant pressure involves the
1.1 The surface tension Gibbs free energy G
The surface tension arises from the unbalanced attractive forces  
at an interface, where the density of the system is changing @G
g¼ : (2)
rapidly. The surface appears to be covered with a stretched @A N;P;T
elastic membrane. For the vapour–liquid interface of water, the
surface tension is high (72.8 mN m1 at 20 1C), and small The surface tension can be measured experimentally by a
objects that are denser than water are able to float or move along variety of techniques including the Du Noy ring method, the
the surface. An understanding and control of the surface tension pendant drop method and the Wilhelmy plate method.5
is at the heart of many important industrial and practical However, one of the goals of modern liquid-state theory
processes, for example: the control of detergency through the has been the accurate prediction of g from a knowledge
addition of surface active molecules (surfactants);1 enhanced oil of the underlying intermolecular potential functions for
recovery from reservoirs through the injection of dilute solutions mixtures over a broad range of temperatures and pressures.
to reduce the surface tension;2 the phenomena of wetting and The powerful technique of computer simulation can be
spreading in the application of coatings;3 and the separation of brought to bear on this problem and in this review, we will
hydrophilic and hydrophobic materials using froth flotation.4 explore the progress that has been made towards this goal in
Formally the surface tension of a liquid is the force per unit the last 40 years.
length that opposes the expansion of the surface area. For a
1.2 Statistical mechanics of the vapour–liquid interface
system at constant volume, temperature and composition
  The connection between the underlying intermolecular potential
@A and the surface tension of an interface is made through the
g¼ (1)
@A N;V;T statistical mechanics of inhomogeneous systems.6–8
For an inhomogeneous fluid in the canonical ensemble, the
a
Institut de Physique de Rennes, UMR CNRS 6251, Université Rennes 1, structure at the interface is essentially described by the singlet
263 avenue du Général Leclerc, 35042 Rennes, France and pair distribution functions. For a fluid of N atoms at a
b
Institut de Chimie de Clermont-Ferrand, ICCF, CNRS, UMR 6296, Université Clermont temperature T
Auvergne, Université Blaise Pascal, BP 10448, F-63000 Clermont-Ferrand, France.
E-mail: Patrice.Malfreyt@univ-bpclermont.fr; Tel: +33 473407204 Ð Ð
ð1Þ    expðbVÞdrðN1Þ
c
CECAM École Polytechnique Fédérale de Lausanne, Batochime BCH 3101, r ðr1 Þ ¼ N Ð Ð (3)
CH 1015 Lausanne, Switzerland    expðbVÞdrðNÞ

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

and where V is the total potential energy of the fluid, dr(N) = dr1  drN,
Ð Ð where dr1 is the volume element for atom 1, and b = 1/kBT, where kB
   expðbVÞdrðN2Þ
rð2Þ ðr1 ; r2 Þ ¼ NðN  1Þ Ð Ð (4) is the Boltzmann constant. r(1) and r(2) are related through the first
   expðbVÞdrðNÞ
equation in the YBG hierarchy.9
For the planar vapour–liquid interface the cylindrical symmetry
allows us to express these distribution functions in terms of the
height of the atoms zi, and zj, and the distance sij parallel to the
planar interface (or equivalently through the variables {zi, zj, rij}
Aziz Ghoufi received his PhD degree or {zi, cos yij, rij}). The geometry of the planar interface is shown
in physical chemistry from the in Fig. 1.
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

University Blaise Pascal, Clermont- In an inhomogeneous system the pressure is a second-rank


Ferrand, France, in 2006. After two tensor. At equilibrium
Postdoctoral stages in Institut
Francais du Pétrole, Paris, France rp = 0. (5)
and in Institut Charles Gerhardt,
Montpellier, France, Aziz Ghoufi
joined in 2008 the physics institute
of Rennes (Institut de Physique de
Rennes, IPR) at the University of
Rennes1 (UR1). Aziz Ghoufi
Aziz Ghoufi currently works in ‘‘Nanosciences
Materials’’ department at IPR. His
research aims at understanding how confinement at the nanoscale
deeply modifies the phase diagram, the molecular dynamics, the flow
or self-assembly conditions of fluids. He aims at gaining a better
understanding of the microscopic mechanisms that control the
nanostructuration of confined fluids. His research interests include the
development and applications of advanced molecular and mesoscale
simulation techniques to model heterogeneous systems and to
understand the confinement effects of liquids and gases in the
nanoporous medium. Fig. 1 The geometry of the planar interface.

Patrice Malfreyt obtained his PhD in Dominic Tildesley received his BSc
Physical Chemistry on computer in Chemistry from Southampton
modelling in 1995 from the University in 1973 and his DPhil
University Blaise Pascal, Clermont- from Oxford University in 1977.
Ferrand, France. After a postdoctoral After postdoctoral positions at
stage at Imperial College (London), Pennsylvania State University,
he joined the group of Cornell and Oxford, he joined the
Thermodynamics of Clermont- Chemistry Department at
Ferrand where he is professor Southampton University in 1981.
since 2004. He is the head of the He was awarded the Chair of
group of ‘‘Molecular architectures Theoretical Chemistry at
at interfaces’’ at the Institute of Southampton in 1990. In 1998
Patrice Malfreyt Chemistry of Clermont-Ferrand Dominic J. Tildesley Dominic joined Unilever Research
(ICCF) and he is vice-president Port Sunlight as the Head of the
of the Physical Chemistry Division of the French Chemical Society Physical Sciences Group. In 2003 he became Chief Scientist for
(SCF). His research is dedicated to the development of Unilever’s Home and Personal Care Division. He retired from
methodologies for simulations of interfacial systems and of multi- Unilever in June 2012. He moved to Switzerland in January 2013, to
scale simulations for the study of thermodynamic and mechanical become Director of CECAM (Centre Européen de Calcul Atomique et
properties of polymers at interfaces. Moléculaire) based in Lausanne in Switzerland. He is currently
Professor Titulaire at the École Polytechnique Fédérale de Lausanne.
His research interests include the statistical mechanics of liquids,
computer modelling of complex fluid, simulation of adsorption and
application of high performance computing to industrial problems.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

The pressure tensor p is diagonal and for a planar interface


it is only a function of z. The independent components are
1 
pN = pzz(z), the normal component, and pT ¼ pxx þ pyy , the
2
tangential component.
Extending the well known virial equation for the pressure of a
homogeneous fluid is not straightforward. There is no unique way
of deciding whether the force between two atoms 1 and 2 con-
tributes to the stress across a microscopic element of the fluid
at a particular z. For pairwise additive potentials Schofield and
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

Henderson10 have shown that the pressure tensor can be written as


Fig. 2 The affine transformations required in the test-area method for
*  ð +
N X
X N ra the calculation of g: (a) A decrease in the surface area at constant V to
ij dv rij
pab ðrÞ ¼ kB TrðrÞdab  dl b dðr  lÞ (6) the perturbed system (blue); (b) an increase in the surface area at
r drij Cij
i j 4 i ij constant V, to the perturbed system (red). Note in both cases the atom
positions are scaled between the reference system and the final system,
where Cij is any contour joining ri to rj, dab is the Kronecker delta, v(rij) see eqn (15).
is the potential between atoms i and j and d is the Dirac delta
function. There are infinitely many ways of choosing the contour
from i to j. Irving and Kirkwood11 consider a straight line between the where DV ¼ V 1  V 0 , the potential energy difference between
two atoms. An alternative contour due to Harasima12 extends from i the perturbed and the reference system, is calculated over the
parallel to the plane of the surface and then directly down to j. These states of the reference system. Note that V 1 is calculated by
different contours all lead to the same value of the normal pressure, scaling all of the atom positions with the box. Small scalings of
but can give rise to different values of the tangential pressure. DA=A0  5  104 produce accurate values for g.18 The test-
The surface tension, g can be calculated from the pressure tensor. area method avoids the calculation of the pressure tensor and
ð þ1 is not limited to pairwise additive potentials. This method
g¼ dz½ pN ðzÞ  pT ðzÞ (7) represents then an interesting alternative for the calculation
1 of the surface tension with many body interactions.
where pN and pT are the normal and tangential components of the These formulae for the surface tension can be readily
pressure. The limits of this integral can be truncated at values of z extended to molecular and ionic fluids, where the forces are
within the vapour and liquid where the integrand is zero. g is long-range, and to interfaces of different shapes, such as the
independent of the choice of contour chosen to define p. spherical droplet or the solid–liquid interface. We will discuss
The surface tension can also be obtained directly by calculating the extensions in the later sections of this review or direct the
the change in the free energy or thermodynamic potential13,14 with reader to the original references for the appropriate equations.
surface area at constant volume. Since the planar interface is two dimensional, it is char
ð ð   acterized by a strong asymmetry in the intermolecular forces. In
1 þ1 3z122 0 the bulk phases, there is a force balance between the attractive
g¼ dz1 dr12 r12  v ðr12 Þrð2Þ ðr1 ; r2 Þ (8)
4 1 r12 and repulsive forces due to the symmetry around a given
where v0 (r) is the first derivative of the intermolecular potential molecule. In the liquid–vapour interfacial region, for example,
with respect to r. Eqn (8) is formally equivalent to eqn (7). the surface tension is explained by the pressure anisotropy of
The expressions given in eqn (6) and (8) make the assumption the attractive forces since the repulsive forces are isotropic and
of pairwise additivity. A general statistical expression of the surface not sensitive to the changes in the structure of the liquid
tension15 has been derived by Grant and Desai in 1980 for many around a molecule.19 In the case of water–oil interfaces modified
body interaction potential models. For example, the Kirkwood– by the presence of surfactants for example, the surfactant
Buff expression of eqn (8) has been generalized by Toxvaerd to molecules adsorb at the interface with specific orientations
include the three body contributions.16 and conformations in order to minimize their free energy: the
The third method for calculating g is the test area method of resulting conformation typically has the hydrophilic part of the
Gloor et al.17 From the thermodynamic definition of g, eqn (1), surfactant interacting with the water phase whereas the hydro-
we have phobic part of the surfactant is exposed to the oil phase.
 
DA
g ¼ lim : (9) 1.3 Computer simulation of c
DA!0 DA NVT

The methods of computer simulation are now widely used to


In a reference system (0) containing an interface of A ¼ 2Lx Ly ,
evaluate the averages discussed in Section 1.2. They include the
we make an imaginary change to the area at constant V of
Metropolis Monte Carlo method and molecular dynamics for
DA ¼ A1  A0 as shown in Fig. 2 then,
atomistic models of the intermolecular forces, and Brownian
kB T dynamics and dissipative particle dynamics for more coarse-
g ¼ lim  lnhexpðbDVÞi0 (10)
DA!0 DA grained models.20,21

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

Away from the critical point, the surface tension can be


calculated using any one of the three methods described in
Section 2.1.
If the potential is truncated, a long-range correction has to
be added to g often using a mean-field approach. Recent
studies indicate that because of the inhomogeneity in the
density, it may be better to include the long-range correction
at each step in the simulation and we will return to this point.
Considerable progress has been made in the simulation of
surface tension over the last thirty years. The initial simulation
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

of the gas–liquid interface for atoms has been extended to


molecular fluids including water, and to mixtures, and particularly
ionic liquids. Simulations at the planar interface have been
extended to spherical droplets of atomic and molecular fluids.
Simulations of the solid–liquid and liquid–liquid interfaces have
Fig. 3 (a) Radial density profile r(r) of a spherical interface where r is the also been performed. In all of these cases the calculation of g has
distance from the centre of mass; (b) density profile r(z) of a planar proved a useful way of making contact between the experiment
interface where z is the z-position along the axis normal to the surface.
and simulation. We have tried to capture this chronology and the
corresponding references in Fig. 4.
The extension of these simulation methods from the homo- At this point, we have a variety of different routes to g, there
geneous liquid phase to liquid–vapour and liquid–liquid inter- are powerful and generic simulation packages that can be
faces has been well documented.6–8 For example, to simulate readily adapted to include these methods, computer speeds
the fluid–vapour interface, we would begin with the simulation continue to increase following Moore’s law towards exa-flop
of the bulk fluid close to the coexisting liquid density in a cubic computing in 2020. So what is preventing us from achieving our
box V = Lx3. Additional empty boxes would be added to the top goal of the prediction of g?
and bottom of the equilibrated liquid to create a box, Lz = 5Lx,  There are essentially three main barriers that we wish to
Lx  Lx. The new, extended periodic system is simulated explore in this review. First, the calculation of g is sensitive to
forward until two stable interfaces are established, each of area many of the technical features of the simulation. These include:
A = Lx2. The stability of these interfaces is monitored using the effect of the finite size in the plane of the interface and the
the single particle density profile r(z), see Fig. 3b. Similar corresponding suppression of the long wavelength capillary
techniques for studying liquid droplets in coexistence with waves; the size of the simulation box in the direction normal
their vapour can be employed. A droplet can be set up at the to the surface and the possible interference between liquid
centre of a much larger periodic box and the liquid allowed to slabs in the periodic boundary conditions; the effect of the cut-off
come to equilibrium with its own vapour, or the spherical in the potential functions; and the inclusion of the long-range
droplet and vapour can be surrounded by a soft repulsive wall corrections when the system contains charges and dipoles. We will
of spherical symmetry to confine the droplet. Properties such as demonstrate that, at least in the case of the planar interface, these
the density and pressure profiles, see Fig. 3a, are calculated are now under control.
with respect to the centre of mass of the droplet, which can Secondly, once we are certain that is possible to calculate g
change during the course of the simulation. accurately and within a predictable error bound, we must be
The techniques of establishing a microscopic interface by concerned about the quality of the underlying potential models
simulation works well for temperatures ca. 10% below the that we are using in the simulation. The ability of the potential
critical point of the fluid. Close to the critical point it is not model to predict the coexisting liquid and vapour densities is
possible to stabilise the interface which becomes broader and an essential prerequisite for the accurate prediction of the
more diffuse. Under these circumstances, the technique of surface tension. In some cases, the effective two-body potentials
choice is to simulate a single phase in the grand-canonical that have proved so useful in simulating homogeneous liquid
ensemble and estimate the distribution function of the number phases may not be appropriate for the highly inhomogeneous
of atoms as the gas and liquid appear spontaneously in the gas–liquid interface.
same box. g can be calculated for a series of simulations of Thirdly in the simulation of systems containing surfactants,
different box sizes from the distribution function rmVT(N,L) and polymers and proteins it may not be possible to simulate
extrapolated to the limit of large box lengths.22 The transition- sufficiently large systems for the appropriate length of time to
matrix Monte Carlo (TMMC) method has also been applied establish an equilibrated interface. In these cases, one possible
with the finite-size scaling to calculate the surface tension of solution is to use a more coarse-grained model of the system which
pure Lennard-Jones fluids23 and binary Lennard-Jones mixtures.24 avoids the explicit inclusion of all the atoms. We are just beginning
The liquid–vapor interfaces of a mixture of water and butanol to understand how to use these coarse-grained approaches in the
molecules has been investigated by configurational-bias Monte simulation of g. In the following sections we will address these three
Carlo simulations in the Gibbs ensemble.25 issues for a variety of geometries and systems.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article


Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

Fig. 4 A brief history of the computer simulation of surface tension. MC and MD refer to Monte Carlo and molecular dynamics simulations, LJ to
the Lennard-Jones potential, HS to the hard-sphere potential, LV to the liquid–vapour interface, IK to the Irving–Kirkwood method and Ha to
the Harasima version of the surface tension calculation. For water, we indicate in % the percentage of deviation of the simulated surface
tension from experiments and the water model used. CG and MDPD correspond to the coarse-grained model and many-body dissipative particle
dynamics.

2 Planar interfaces where Lz is the dimension of the simulation cell along z. The
element ab of the molecular pressure tensor is
In the case of a planar interface with z perpendicular to the
interface, the surface tension can be calculated using both * +
XX
1 N1 N    
mechanical and thermodynamic definitions. In this section pab ¼ rkB TI þ rij a f ij b (12)
V i¼1 j 4 i
we extend the formal definitions given in Section 1 to more
operational definitions that can be readily used in a simulation.
where I is the unit tensor, T is the input temperature and r = N/
2.1 The Lennard-Jones fluid V is the mean number density. a and b are the Cartesian
In this section the atoms in the fluid, i and j, interact through coordinates. rij is the vector between the atoms i and j and fij
the Lennard-Jones (LJ) potential and the dependencies of the is the intermolecular force between atoms i and j. Note that the
surface tension on properties such as the cutoff are presented sum over i and j 4 i is over all the atoms in the system. If the
in the context of this potential. system is of the form shown in Fig. 3b this will include two
2.1.1 Kirkwood and Buff definition. The surface tension interfaces and the final result for g obtained from eqn (11) will
gKB was introduced by Kirkwood and Buff.48 In terms of need to be divided by 2.
simulated averages, eqn (7) and (8) of Section 1 become 2.1.2 Irving and Kirkwood definition. The method of Irving
and Kirkwood (IK) expresses the surface tension through the z
gKB ¼ hpN  pT iLz dependent local components of the pressure tensor
*  +
X1 XN   (11) ð Lz =2
1 N rij  rij  3zij  zij dvLJ rij
¼ gIK ¼ ð pN ðzÞ  pT ðzÞÞdz: (13)
A i¼1 j¼iþ1 2rij drij Lz =2

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

The Lz dimension of the box is divided into slabs of width dz simulations of the surface tension in a slab geometry. Whereas
and the z-dependent pressure tensor is calculated as6,53,54 most of the Monte Carlo simulations of atomistic interfaces use
truncated LJ potentials, the molecular dynamics simulations
pab ðzÞ ¼ hrðzÞikB TI
use the corresponding truncated force. The appropriate integration
*    +
X1 X of the truncated force with distance does not give the truncated
1 N N    
1 zi  z z  zj
þ rij a f ij b  y y potential because of the small step function at the cutoff
A i¼1 j 4 i zij zij zij
distance and as a consequence, MC and MD simulations which
(14) purport to use the same potential are often performed with
slightly different potentials making the comparison between
where y(x) is the unit step function defined by y(x) = 0 when
the two methods problematic. Whereas the subtle truncation
x o 0 and y(x) = 1 when x Z 0. A is the surface area normal to
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

effects have little effects on the properties of bulk phases, they


the z axis. r(z) is the total number density calculated in the slab
can significantly affect the interfacial properties. The reader is
at z. The distance zij = zi  zj between two molecular centers of
directed to Trokhymchuk and Alejandre40 and Goujon et al.56
mass is divided into Ns slabs of thickness dz. Following Irving
for a comprehensive discussion of this truncation effect on the
and Kirkwood, the molecules i and j give a local contribution to
surface tension.
the pressure tensor in a given slab if the line joining the atom
There are two ways to obviate this problem. First, one can
i and j crosses, starts or finishes in the slab. Each slab has 1/Ns
use a modified potential and force functions that are contin-
of the total contribution from the i–j interaction.
uous at the cutoff. Using this type of approach Goujon et al.56
2.1.3 Test area definition. The test-area method developed
have shown that the results for the surface tension obtained
by Gloor et al. (TA) is based upon a thermodynamic route and
from MC and MD simulations are identical.56 However, these
expresses the surface tension as a change in the free energy55
results deviate significantly from experiment due to the modification
for an infinitesimal change in the surface area at constant-NVT.
of the underlying LJ potential with a cubic spline function. A
This method uses a perturbation process for which the perturbed
second strategy developed by Trokhymchuk and Alejandre40 in
system (state A þ DA) is obtained from an infinitesimal change
order to match the truncated potential consists in modifying the
DA of the area A of the reference system. The box dimensions

truncated force with a d-function like impulsive contribution.57
ðAþDAÞ ðAþDAÞ
Lx ; Ly ; LzðAþDAÞ in the perturbed system are obtained Even using the same simulation method, truncation of the
using the following transformations potential can also create differences between the mechanical
pffiffiffiffiffiffiffiffiffiffiffi and thermodynamic routes to g when calculating the direct or
LðAþDAÞ
x ¼ LðAÞ
x 1þx intrinsic part of the surface tension in a simulation. The thermo-
pffiffiffiffiffiffiffiffiffiffiffi dynamic route of Section 2.1.3 uses the potential functions,
LyðAþDAÞ ¼ LðAÞ
y 1þx (15) whereas the mechanical route of Sections 2.1.1 and 2.1.2 uses its
. first derivative. However, when truncated potential and forces are
LzðAþDAÞ ¼ LðAÞ
z ð1 þ xÞ
used, the differences between the mechanical and thermodynamic
routes are exactly compensated by the differences associated with
where x { 1, see Fig. 2. The area ðA þ DAÞ of the perturbed
ðAÞ ðAÞ ðAÞ ðAÞ
the long range corrections (see line 1 of Table 1).38,60
state is Lx Ly ð1 þ xÞ and DA is equal to Lx Ly x. These Table 1 demonstrates the equivalence of the Kirkwood Buff,
transformations conserve the volume of the box in the per- Irving Kirkwood and test area methods for calculating g for
turbed state. This means that it is possible to express the three different systems. The first line of the table also shows the
surface tension as a ratio of partition function between the difference in the calculation of the intrinsic part of the surface
perturbed and reference states. V ðAÞ ðrN Þ and V ðAþDAÞ ðr0N Þ are tension by mechanical and thermodynamic routes and the way
the configurational energies of the systems with an area A this is balanced by different long-range corrections.
and a configurational space rN and an area A þ DA and a 2.1.5 Long range corrections. Since most of the molecular
configurational space r 0 N, respectively. The transformations simulations use a pair potential with a spherical cutoff at a
used in the perturbation process lead to the following equality distance rc, long range corrections must be added to the thermo-
dr 0 N = drN. This equality allows us to write the surface tension as dynamic properties such as energy, pressure, stress tensor,
the average of expðDV=kB T Þ at constant-NVT, i.e. chemical potential and surface tension to compensate for the
  missing long-range part of the potential. In the case of bulk
@A
gTA ¼ simulations,20 these long range corrections assume a uniform
@A N;V;T
density r and a radial distribution function equal to unity
 ðAþDAÞ 0N 
kB T V ðr Þ  V ðAÞ ðrN Þ beyond rc. For example, for the homogeneous Lennard-Jones
¼ lim  ln exp 
x!0 DA kB T A
fluid, the contribution of the tail corrections to the total energy
(16) decreases from 12% to 0.4% as the cutoff increases from 2.1s to
6.4s where s represents the effective atomic diameter of the LJ
particle. The usual expressions of the long range corrections to
2.1.4 Truncation effects. It is often appropriate and useful the energy and pressure are not applicable in two-phase systems
to combine and compare Monte Carlo and molecular dynamics and different approaches have been developed. The development

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

Table 1 Values of surface tension (mN m1) calculated from Monte Carlo chlorine69 and alkanes.60,70 The long range corrections calculated
simulations using a truncated Lennard-Jones potential and a truncated by this expression are also given in Table 2 for comparison. These
potential modified by a polynomial function (see ref. 58). Different definitions
two tail corrections are added to the intrinsic part of the surface
are used to calculate the intrinsic part of the surface tension, gI, i.e. the part
within the simulation cut off. gLRC is the long range correction to the surface tension at the end of the simulation but are calculated with a
tension; and each definition, IK, KB and TA has its own expression for the tail density profile that corresponds to the truncated potential without
correction to the surface tension. The reader is directed to ref. 58 for a any tail corrections. By themselves they can only provide a upper
description of the methodology of the calculation. For the MARTINI model, limit to the true tail correction.
the reader is redirected to ref. 59. The subscripts give the accuracy of the last
A third approach used by Alejandre and coworkers64,65,71
decimal(s), i.e., 65.814 means 65.8  1.4
consists in modelling the full LJ potential by calculating the
gKB gIK gTA dispersion term of the LJ potential as a lattice sum such as
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

gI gLRC gTOT gI gLRC gTOT gI gLRC gTOT using the Ewald method,28 the particle-particle-mesh Ewald
Truncated LJ potential
method72,73 or one of the multilevel summation methods74 that
9.512 3.61 13.113 9.512 3.81 13.313 10.611 2.61 13.211 have been proposed to reduce the computational cost of the full
lattice sum. In these methods atoms in the central box interact
Cubic spline modified LJ potential
8.58 0 8.58 8.58 0 8.58 8.58 0 8.58
with one another and with all of their periodic images. Recent
studies72,73 have established that the simulations of the liquid–
MARTINI force field – n-hexadecane–water liquid–liquid interface vapour interface of the LJ fluid can be performed accurately
43.210 0 43.210 43.210 0 43.210 43.210 0 43.210
with both the Ewald and the PPPM approach. Changes in the
intrinsic surface tension with the cutoff are balanced by
of these long range corrections for the two-phase simulations and changes in the long range part of the dispersion interaction
the interfacial properties has been an area of active research for the which is calculated in reciprocal space.
past two decades.38,41,42,61–66 These tail corrections are important A fourth and important set of approaches is based on the
because they range from 50% to 7% of the intrinsic surface tension introduction of the local configurational energy38,41,63,66,76–80 or
as the cutoff changes from 2.1–6.4s. An accurate estimate of these local force77,78 in each slab, dz of the simulation box. As first
tail contributions of the surface tension can be critical in obtaining proposed by Guo and Lu41,63,76 the tail contribution to the
accurate simulated values of g when the Lennard-Jones potential is configurational energy depends on the position of the molecule
involved. in the direction normal to the interface, and is specifically
The first operational expression for correcting g was developed included at each step of the Monte Carlo simulation. The tail
from the work of Fowler67 and used in simulations from 1973 correction changes the Markov chain of states generated during
onwards.42,61,68 This expression is based on the use of a step the simulation. This long range correction to the configurational
function for the density profile r(z) and thus an interface of energy comprises two contributions: the first one is similar to that
zero-thickness. Values of g calculated by this route are given corresponding to a bulk (homogeneous) system with a local
in Table 2 as a function of the cutoff radius at two different density r(z). The second contribution corrects for the density
temperatures. inhomogeneity by considering an integral over a series of
A second approach based upon the Kirkwood–Buff expression48 density differences. This second contribution has usually been
for the surface tension, proposed initially by Chapela et al. and neglected due to the high computational cost of its calculation.
corrected later by Blokhuis et al., assumes that the density profile The approximate approach of Guo and Lu, with the inclusion of
r(z) can be fitted to a hyperbolic tangent function with an interface only the first term at each step, has been widely used in the
of non-zero thickness. The resulting operational expression has calculation of the surface tension of different fluids.56,58,70,81–85
been widely used with the truncated Lennard-Jones potential for An alternative approach based on a local long range correction to
the calculation of the surface tension of water,18,28,29,38 hexane and the energy was proposed by Janec̆ek66 in 2006. The original
formulation66 has been revisited by MacDowell and Blas79 in
order to avoid the need for the recalculation of the complete
Table 2 Values of the reduced surface tensions (intrinsic, gI, and long-
density profile at each step in the simulation. The method of
range correction contributions, gLRC), calculated from configurations Janec̆ek has also been widely applied to different liquid–vapor
obtained with the LJ potential truncated at rc = 3.2s at T* = 0.67 and systems: LJ fluids,79,80,86 cyclic hydrocarbons,87 alkanes,88
T* = 0.83. The long-range corrections to the surface tension are calcu- water and CO238 and argon.89
lated from different expressions: gFOW
LRC is calculated from the expression of
Recently,75 the full version of the Guo and Lu method (both
Fowler,67 gBLO
LRC from Blokhuis et al.,
62 JAN
gLRC from Janec̆ek et al.66 and gGL
LRC
from Guo and Lu.41 The subscripts give the accuracy of the last decimal
terms) was compared directly with the Janec̆ek method by studying
places, i.e., 0.9239 means 0.923  0.009 the cutoff dependence of the surface tension and the results are
presented in Fig. 5. The reduced surface tension is the sum of
rc gI gFOW
LRC gBLO
LRC gJAN
LRC gGL
LRC
the intrinsic and long-range parts. In the figure we have also
T* = 0.67 included the results for g using a truncated LJ potential and
3.2 0.9239 0.327 0.296 0.293 0.297 shifted LJ potential (where the Lennard-Jones potential is shifted
T * = 0.83 to zero at rc). For the shifted Lennard-Jones potential, there is no
3.2 0.59511 0.270 0.222 0.221 0.225 long range correction to the surface tension.

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

potential. For more accurate two-body potentials (for example,


the Barker Fisher Watts potential89), the decay at large r is
faster than r6 and the conclusions that we make for the LJ
fluid are equally valid. The use of a truncated and shifted
potential for the surface tension does not produce a converged
result for any of the cutoffs and should be avoided. Shifting the
potential to zero at rc certainly avoids the long range corrections
but the results show that this truncated and shifted potential
produces values of surface tensions that are dependent on the
cutoff. This type of potential is not able to produce quantitative
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

predictions because the cutoff is an adjustable parameter.


Our overall conclusion is that an efficient accurate method
for adding the long-range correction in these slab simulations
is to use the Janec̆ek method66 to estimate the long-range
correction at each step of the simulation with an update of
Fig. 5 Values of surface tension (g* = gs/e) as a function of the reduced the density profile at every ten Monte Carlo cycles. The long
cutoff distance (rc* = rc/s) for different potential models. For T* = 0.80 , range correction to g should also be calculated at each step
Janec̆ek method;38 For T* = 0.83 , simplified version of Guo and Lu;75 , using the Janec̆ek method and added to the intrinsic calculation
full version of Guo and Lu;75 , Janec̆ek method;75 , truncated LJ of g. A cutoff of between 3.0 and 4.0s can be employed.
potential;75 n, truncated and shifted LJ potential.75 For T* = 0.85 ,
2.1.6 Size effects. It is now well-established that the calculation
dispersion interaction calculated with the Ewald method;72 , dispersion 39,60,91
interaction calculated with the PPPM method.72 of the surface tension
 is dependent on both the surface
area A ¼ Lx Ly and the box dimension (Lz) along the normal
to the surface. Indeed, several studies have shown that the
Some limited results have also been included for the Ewald surface tension exhibits an oscillatory behaviour as a function
and PPPM lattice sums,72 which used a cutoff of 4s. The resulting of Lx (for Ly = Lx). The precise nature of these oscillations in g(Lx)
simulation time is increased by a factor of 5 for the Ewald method depends on the potential model employed.60 Typical oscillations
and a factor of 2 for the PPPM approach as compared to a are shown in Fig. 6 for several potential models at different
simulation with a simple real-space cutoff, rc = 4s. reduced temperatures.
The full correction of Guo and Lu and the method of Janec̆ek The origin of these oscillations has been attributed to the stress
give almost identical results for the interfacial properties when anisotropy39,60,92 in the liquid phase induced by the use of periodic
applied to the same set of configurations.75 Although not boundary conditions and small box dimensions. Interestingly, the
shown in the plot, the convergence of the coexisting densities modification of the truncated LJ potential using a cubic spline
as a function of the potential cutoff is slower than for the significantly reduces the stress anisotropy in the liquid phase and
surface tension. We would recommend using a cutoff of at least as a result the dependence of the surface tension on the surface
4.5s to obtain accurate values of these densities even with area. When the long range correction to the energy is included at
the long-range correction included at each step. The surface each step, we see that one peak of oscillation is removed and the
tension and coexisting densities can be calculated without amplitude of these oscillations is reduced. We also note the same
using the long-range correction at each step. In this approach oscillatory behaviour60 of the surface tension as we change the Lz
we would recommend using rc Z 5.5s. Even though the dimension of the box. We conclude that in order to avoid any size-
intrinsic part of the calculation of g converges for rc Z 4s, effects dependence of the surface tension, a surface area of 11 
there is still the issue of accurately adding the long-range 11s2 and a value of Lz = 60s must be used for simulations carried
correction by the Blokhuis method62 and this requires an out with the truncated LJ potential. These box dimensions can be
accurate estimate of the coexisting densities and rc Z 5.5s. reduced to A = 99s2 and Lz = 50s when the long range corrections
Even in the case where the electrostatic interactions play a to the energy are included directly in the simulation. With these
major role, the long-range correction to the dispersion inter- box dimensions, the vapour and liquid densities show no size-effect
action remains important18 and it is necessary to add a correc- dependence.71
tion for each term in the site–site dispersion potential. These For the planar liquid–vapour interface, the surface tension
conclusions will also be valid for fluid mixtures. However, in also depends on the slab thickness, D, of the liquid phase.93 As
this case the density profile of a particular component may not the slab thickness decreases, the surface tension decreases
be well represented by a hyperbolic tangent function, e.g. where from its limiting value (D - N) as shown for a number of
one component is adsorbed at the gas–liquid interface of the systems in Fig. 7. For the LJ-like potentials, we recommend a
other component as is the case for CO2/H2O in ref. 90. In this value of D* Z 10 to ensure that the two interfaces are quite
case, we would recommend avoiding the Blokhuis correction independent. When the potential is more strongly attractive,
and using the Janec̆ek correction during the course of the run. the surface tension is higher and the interface sharper. In this
We note that the method of Janec̆ek can be readily extended to case, smaller values of D* can be used as shown for the slabs of
more complicated functional forms than the Lennard-Jones the metallic liquids, copper and tin94 in Fig. 7.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

phases differs by several orders of magnitude and the interface


is well-defined. As the temperature approaches the critical
temperature (T 4 0.9Tc), the two-phase simulations with explicit
interface must be carried out with caution to obtain an accurate
estimate of g.35,89 The previous recommendations concerning
the box dimensions and the cutoff radius are no longer valid.
Potoff and Panagiotopoulos use the grand canonical Monte
Carlo simulations and histogram-reweighting techniques to
calculate the free energy barrier height at different system-sizes.
The surface tension can then be obtained by using the methodology
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

of Binder.96 This method works well for predicting the surface


tension of temperatures above 0.7Tc. For lower temperatures,
Potoff and Panagiotopoulos concluded that this methodology is
too time-consuming. The grand-canonical transition-matrix
Monte Carlo (TMMC) method23 developed by Errington is then
applicable to the entire liquid–vapour coexistence curve and
Fig. 6 Reduced surface tensions calculated for different values of Lx for
different potential models and temperatures. For T* = 0.8 , truncated LJ performs better than equivalent Gibbs ensemble calculations.22
potential;60 , truncated LJ potential modified by local range corrections 2.1.8 Potential model. Now that the methodological questions
of Guo and Lu;60 , truncated and shifted potential;39 , truncated LJ have been addressed, we can ask if a simple Lennard-Jones potential
potential modified by a cubic spline function from ref. 60; T* = 0.85 m, can reproduce the surface tension of liquid argon. In fact, it is hard
truncated LJ potential;39 T* = 0.9 , truncated LJ potential.91
to get within 15% of the experimental values using any reason-
able, effective two-body potential. It is possible to approach the
experimental values of g using a truncated and shifted Lennard-
Jones potential.35 In this case, the simulated surface tensions
are strongly cutoff-dependent. The shifted potentials bear little
resemblance to the true intermolecular potentials between
argon atoms and these shifted potentials are not transferable.
A more appealing approach is to include a three-body
interactions97
 
  n 1 þ 3 cos yi cos yj cos yk
vð3BÞ ri ; rj ; rk ¼ (17)
rij3 rik3 rjk3

where yi is the internal angle of the triangle, ijk, centered on


atom i. This potential must be added to the accurate two-body
interactions such as the BFW potential98 or the NLD potential99
for argon–argon interactions. This three-body contribution to
the surface tension can be readily calculated in a Monte Carlo
simulation.89,100,101 The intrinsic part of the surface tension has
Fig. 7 Ratio of the surface tension to the surface tension glim calculated at been calculated using the Kirkwood–Buff approach16 developed to
the largest value of D* = D/s. UT,JAN represents the truncated LJ potential
include the three body contributions. The reader is directed to
modified by the addition of the long range corrections to the energy
calculated with the Janec̆ek method; UT and UTS denote the truncated and ref. 89 for a comprehensive description of the calculation. Care
truncated and shifted LJ potentials. UEAM corresponds to Embedded needs to be taken to properly include the corresponding long
Atoms Models applied to slabs of liquid copper.94 The EAM potential has range corrections to the surface tension form both the two-body
the advantage to be decomposable into pairwise contributions for the and three-body potentials. This approach leads to a good agree-
calculation of the pressure. The description of the methodology for the
ment for g with experiments as shown in Fig. 8.
calculation of the surface tension can be found in ref. 95.
Whereas the two-body effective LJ potential overestimates
the surface tension along the liquid–vapour equilibrium curve
In the case of atoms interacting though the Lennard-Jones with a maximum difference of 33% approaching the critical
type potentials, the impact of the size-effects is now well- point, the best agreement with the experiment is obtained with
understood and can be controlled. As systems approach to the NLD + AT potential which shows a relative deviation of less
within 10% of the critical point, the interfaces become more than 3% over the whole temperature range. This is a good
delocalised and the periodic slabs will interact making this example of how simulation can be used predictively in model-
approach to g inappropriate. ling the surface tension. The ability of a particular model to
2.1.7 Close to the critical point. The calculation of the reproduce g along the coexistence curve is a useful test, since
surface tension of liquid–vapour interfaces using a slab geometry these interfacial properties are not normally used in fitting the
is accurate when the density between the vapour and the liquid parameters of a given model.

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

the surface tensions over this range. Considerable attention has


been paid to polarisable models of water, and in Fig. 9 we show
the surface tensions of two polarisable models POL4108 and
SWM4-DP106 at different temperatures. These models show
their limits in the prediction of g at temperatures greater than
400 K. Tables 3 and 4 also shows some values of surface
tensions calculated using the TIP4P/2005 model and confirms
the quantitative agreement with the experiments independently
of the method used for considering the long range corrections
to the energy41,66 and to the surface tension. These results are
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

in line with the conclusions drawn by Mı́guez et al. from a


cutoff of 3s. To conclude, the best performance in reproducing
the surface tension of water over a large range of temperature is
Fig. 8 Surface tensions calculated from the two-body Lennard-Jones
given by the TIP4P/2005 model.
potential,89 the three-body BFW98 and NLD99 potentials.

2.2 Molecular systems


2.2.1 Molecular forces. Many simulations of molecular
systems are based on additive site–site potentials, where the
total force on molecule i from j, fij, is given by
nj nj  
Xni X Xni X
riajb dv riajb
f ij ¼ fiajb ¼  (18)
a¼1 b¼1
r
a¼1 b¼1 iajb
driajb

where v(riajb) is the potential between site a in molecule i and


site b in molecule j and ni and nj are the number of sites in
molecules i and j, respectively.
The definition of the molecular pressure tensor20,102–105 for
Fig. 9 Simulated values of the surface tension of water. The experimental
molecular systems with forces given by eqn (18) is given by a values are shown as a solid black line. , TIP4P/2005 using both the
straightforward extension of eqn (12). Recall that this approach mechanical and thermodynamic definitions18 and with the Guo and Lu
is only valid for pair-additive potentials. For molecular systems methodology.41 , TIP4P/2005 water using the Ewald summation tech-
containing partial charges the electrostatic interactions are nique for the calculation of the dispersion interaction;71 , SPCE model;29
, TIP4P water model calculated using the test-area method;17,28 E,
usually calculated using an Ewald sum. In this case, the
TIP4P/2005 water model calculated using the test-area method;28 ,
contribution to the potential in the real space is pairwise- TIP4P/2005 water model calculated using the test-area method and the
additive whereas the contribution in the reciprocal space is Janec̆ek methodology;38 , the Drude oscillator model SWM4-DP;106,107
not. The appropriate expressions for the Lennard-Jones and , the polarizable POL4 model.108
electrostatic contributions to the pressure tensor have been
considered in detail.18,29,85
Whereas the size effects have been thoroughly investigated
for the LJ liquid–vapor interfaces, the study of the impact of the Table 3 The surface tension from simulation: CH4 using the truncated LJ
potential UT,58 with the LRC of Guo and Lu,58 and with the LRC of
cutoff value and size effects of molecular systems is not so
Janec̆ek;38 H2O using the truncated LJ potential,28 with the LRC of Guo
extensive. Mı́guez et al.38 have established the dependence of and Lu,18 and with the LRC of Janec̆ek;38 CO2.18,38 The subscripts give the
the surface tension of water on rc when the electrostatic accuracy of the last decimal(s), i.e., 65.814 means 65.8  1.4
interactions are calculated by the reaction field and the long
UT + ULRC
range corrections to the dispersion interactions are included
during the course of the simulation. For rc Z 3s, the surface UT Guo and Lu – LRC Janec̆ek-LRC
tension of water is independent of the cutoff radius when long T (K) gKB gTA gKB gTA gKB gTA Exp.
range corrections to the surface tension are properly applied. Methane
Fig. 9 shows the temperature dependence of the surface 120 13.51 13.51 13.112 13.212 13.81 13.81 11.3
tension of water calculated using different methodologies and
Water – TIP4P/2005
models such as SPC/E,109 TIP4P110 and TIP4P/2005.111 Whereas 388 55.840 55.540 55.9
the TIP4P model shows significant deviations from experi- 400 52.57 52.27 52.010 52.010 53.6
ments with values of surface tensions underestimated by 17% 418 49.720 51.020 49.8
to 45% over the temperature range from 300 K to 500 K, the SPC/ CO2
E and the TIP4P/2005 models exhibit quantitative predictions of 270 4.82 5.77 5.610 5.12 5.5

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

Table 4 Contributions of surface tensions (mN m1) derived from the recent work Zubillaga et al.49 which reports the values of surface
Lennard-Jones contribution (gLJ), the real part (gR) and the two terms tensions calculated for 61 molecular systems with the OPLS/AA
(gK,1,gK,2) of the reciprocal space part of the Ewald contribution, the LRC
force field.115,116 Again, in this study, the simulated surface
term (gLRC). The terms are calculated from the pressure tensor.18 The total
(g) is the sum of all of these contributions calculated with a real-space tensions are within 10% to 20% of the experimental values.
cutoff radius of 12 Å However, other force fields87 can lead to significant difference
from experiments. These differences cannot be understood
gLJ gR gK,1 gK,2 gLRC g
a priori, that is before the simulation has been conducted and
H2O-TIP4P/2005 (T = 478 K) more systematic work is required in this area. Interestingly,
87.6 112.1 6.4 12.0 3.5 33.6
when the coexisting liquid and vapour density are accurately
CO2 (T = 238 K) predicted by a model, in most cases, the surface tension is also
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

6.1 2.0 0.8 1.7 2.5 11.5 accurately predicted.


H2S (T = 187 K)
In most of the simulations reported in Fig. 10, the electro-
21.0 8.0 1.5 2.8 8.0 38.3 static interactions are calculated using the Ewald summation
technique.29,118 Table 4 lists different contributions to the
C6H6 (T = 409 K)
7.5 3.5 0.3 0.5 2.7 13.9 surface tension calculated in systems differing in the relative
magnitudes of the dispersion–repulsion and electrostatic inter-
actions. Since the simulation boxes for these slabs are not
We now focus on the ability of the two-phase simulations to normally cubic, the Ewald summation requires a large number
reproduce the temperature dependence of surface tension of of k-vectors in the longer z-direction.29 Interestingly, the contribution
various molecular systems as shown in Fig. 10. As shown in the to the dispersion–repulsion term is negative for water due to
legend of Fig. 10, various molecular systems such as the linear the strong electrostatic interactions in this system. When the
alkanes (n-pentane70 and n-heptadecane112), branched alkanes electrostatic interactions become less strong, the dispersion–
(2,3-dimethylpentane70), cyclic alkanes (cyclohexane87), aromatics repulsion term of the surface tension is again positive for CO2,
(benzene87 and phenol113), acid gases (CO218,38 and H2S18) H2S and C6H6.
incondensable gases114 (O2, SO2, and N2) have been modelled We note that the reaction field has also been employed in
by using different force field models. Table 3 shows the the calculation of the surface tension.38,119 Since the dielectric
calculation of the surface tension of CO2 from simulations constant depends on the z-position along the normal to the
using truncated LJ potentials, and truncated potentials modified surface, the use of the dielectric constant corresponding to a
by the addition of long range corrections according to the Guo and homogeneous bulk phase is not ideal. However, the authors
Lu and Janec̆ek methods. Within the statistical fluctuations, the report the results of surface tensions38,119 that compare well
simulated surface tensions are equal and indeed independent with those calculated using the Ewald method.
of the definition used for the calculation. We can conclude that The Ewald sum is the method of choice for including long-
the prediction of the surface tension of many molecular systems range interactions in simulations. Its accuracy can be improved
can be calculated to within 15% of the experimental values by extending the number of k-vectors in the reciprocal space
along much of the orthobaric curve. This view is supported by sum and balancing the real and reciprocal contributions to the
sum. The easiest way to use the Ewald method for a 3-D system
with 2-D periodicity is to surround the slab (both the liquid and
vapour phases or the liquid between solid walls) with a number
of empty boxes along the z-direction and to make this extended
box periodic. This empty space helps to prevent any interaction
between periodic slabs in the z-direction. The Ewald sum can
then be applied to the extended cell. As always, there is an
additional term in the energy since the infinite array of images
in the Ewald sum is eventually surrounded by a vacuum. In this
case, there is a correction proportional to Mz (the z component
of the overall dipole of the central box).120 As the system size
increases beyond ca. 500 charges it is often more efficient to
assign the charges to a fine regular mesh and use a direct fast
Fourier transform to calculate the reciprocal sum. These techniques
such as the particle mesh Ewald method have been applied to
slab geometries121 with good effect. Recently a new technique
Fig. 10 Surface tensions of molecular systems calculated from two-phase has been developed to periodically reproduce the spherical
simulations. The reader is directed to the original paper for a description
cut-off sphere around a charge. The isotropic periodic sum
of the methodology used for the calculation of g. CO2,18,38 H2S,18 n-C5H12,
n-C10H22,70 N2, O2,114 C6H6, C6H12,87 2,3-DMP,70 C6H5OH,113 n-C17H36,112
method has been applied to bulk fluids, liquid/liquid and
and monoethanolamine (MEA).117 The experimental surface tensions, liquid/vapour interfaces, and lipid bilayers and monolayers
represented in dotted lines, can be found in the original papers. and is capable of outperforming the mesh lattice methods.121

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

Finally there are a number of specialised techniques to deal constant-NVT statistical ensemble.57,128–131 The constant-NPT
with 2-D periodic charges such as the methods of Lekner122,123 ensemble is not designed for modelling a two-phase binary
and Hautman and Klein.124 Both these methods require that system because the tangential component, pT, of the pressure
the parameter zij/sij between two charges i and j is small since tensor is not constant along the surface normal, normally
this is used as an expansion parameter in a series expansion of exhibiting a strong minimum in the interfacial region. The
the force. The methods are best applied to physisorbed mono- most appropriate ensemble is at constant-NpN AT where pN is
layers or very thin liquid films. the normal pressure and the A interfacial area. The reader is
directed to the work of Biscay et al.90,132 for a comprehensive
2.3 Mixtures description of the methodology. In this case, the expressions
As discussed in the previous section, the two-phase simulations for the calculation of the long range corrections to the surface
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

of liquid–vapour interfaces of pure systems (i.e. a single component) tension through the thermodynamic and mechanical routes are
is well-understood from a methodological viewpoint and the also available.90,132 Fig. 12 shows the surface tension of different
reproduction of the temperature dependence of g for these binary mixtures as a function of the pressure. Fig. 12 shows the
systems is good. For mixtures we need to extend these methods value of the constant-NpN AT simulation method in predicting
to consider additional parameters such as composition and the surface tension. The results of these slab simulations over a
pressure. For example, the long range corrections to the surface large range of pressures for the coexisting densities compare
tension and configurational energy for a mixture is straight- well with the simulated phase diagrams calculated from Gibbs
forward and has been achieved with the approach of Guo and ensemble Monte Carlo (GEMC) simulations.134,135
Lu.76 To our knowledge, the method of Janec̆ek has not been Another challenging aspect of the computer modelling of
applied to the liquid–vapour equilibrium of mixtures but this the surface tension of complex systems is the prediction of the
extension poses no fundamental problems. Fig. 11 shows the dependence of the surface tension on salt concentration. This
surface tension of different mixtures as a function of the calculation may be quite sensitive to the choice of standard
composition. In the case of the liquid–vapour interface of an (non-polarisable) or polarisable models of water and the ions.
alcohol–water mixture, the simulations accurately predict the Fig. 13 shows the increase of the relative surface tension
significant decrease of the surface tension with the alcohol defined by Dg = gsalt  gwater where gsalt and gwater are the
mole fraction.125,126 In the water–alcohol mixtures, the long surface tensions of the salt solution and pure water. Different
range corrections to the surface tension contribute between 5% methods have been employed to include electronic polarisation
and 10% of the total surface tension depending on the alcohol into simple atom–atom potentials. Three of the most widely-used
concentration and must be accurately included. The molecular techniques are the induced dipole model,136–138 the fluctuation
description of the mixture in terms of density profiles and charge model139,140 and the classical Drude oscillator model.106
hydrogen bond profiles has been used convincingly to interpret The Drude oscillator model, with the simple functional form of
the decrease of the surface tension with the alcohol concentration. the pairwise additive interactions and the ability to account for
The dependence of the surface tension on the composition is well the electronic polarisability, is particularly useful since it intro-
reproduced for other mixtures127 where the variation is less duces only a small computational overhead. Different polarisable
pronounced with the composition.
The situation is still more delicate when we examine the
pressure dependence of binary systems. To date, most of the
simulations of binary systems have been performed in the

Fig. 12 Surface tensions of different binary mixtures as a function of the


Fig. 11 Surface tensions of water–methanol and water–propan-2-ol pressure. Water–methane132 (T = 373 K), water–CO290 (T = 383 K) and
mixtures as a function of the mole fraction of alcohol126 and surface water–H2S90 (T = 393 K), water–N285 (T = 298 K), decane–CO284 (T = 344
tensions of hexane–decane and CO2–decane mixtures as a function of K), pentane–H2S84 (T = 344 K), hexane–N2133 (T = 333 K), and ethanol–
the liquid mole fraction.127 The dotted lines are the experimental values. N2133 (T = 333 K) systems.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

In particular, the arbitrariness in the choice of contour


connecting two atoms that contribute to the stress leads to an
ambiguity in the definition of the local elements of the pressure
tensor in a spherical geometry. Indeed, in relation to the planar
interfaces, the curvature complicates the definition of the local
stress149 and amplifies the difference in the different routes to g
(see Section 3.1). Despite this ambiguity, the Irving–Kirkwood
(IK)48 and Harasima150 definitions are widely used in the
calculation of p and g for droplets. Recently, Nakamura
et al.151 have also proposed a novel method for computing
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

the pressure tensor along the radial direction of a molecular


system with spherical symmetry. The method uses the slice-
averaged pressure to improve the numerical stability and the
precision of the calculation significantly. Although this approach is
numerically useful, the ambiguity in the contour remains.
Fig. 13 Relative surface tensions defined as Dg = gsalt  gwater as a function Density functional theory calculations (DFT) provide a
of the molal concentration of NaCl for different nonpolarisable and straightforward, if approximate, estimate of g(Re) with respect
polarisable models for ions and water at T = 298 K. The dashed lines to the drop size (Re).148,149,152–154 Fig. 14 compares the results of
represent the experimental linear dependence of the surface tension on
g for the simulation of spherical Lennard-Jones droplets from
molality. , TIP4P2005 + OPLS models;107 , SPC/E + ion models;143 .,
polarisable Drude models;107 J, nonpolarisable + electronic continuum simulation and from the DFT methods. The main conclusions are:
models;144 and , polarisable POL3 model.145 (i) there is a maximum in g with respect to the drop size in
DFT but a variety of different behaviours for this curve from
water models: SWM4-DP;106 SWM4-NDP;141 and SWM6142 based on different simulation approaches;
classical Drude oscillators, have been proposed. Fig. 13 shows that the (ii) a number of different simulation approaches produce
nonpolarizable models perform well in reproducing a monotonic g(Re) values of different magnitudes to the DFT results, while
increase of the surface tension with the molality. some simulation results are quite similar to DFT;
The polarisable force field based upon the Drude oscillator (iii) DFT predicts a negative Tolman length (d) while negative
model leads to some scattering in the predicted surface tensions and positive values can be found in different simulations of the
and the POL3 model145 underestimates the increase of the same system using different approaches.
surface tension with the salt concentration. The electronic
continuum model (EC),146,147 an alternative method of adding
electronic polarisation into a simple nonpolarisable model,
produces a fluctuating change in the surface tensions with the
molality. The structure of the interface is significantly different
between the polarisable and non-polarisable models: without
polarisation the structure corresponds to a profile resembling a
simple LJ fluid; with polarisation the ions occupy preferential
layers at the aqueous interface.

3 Spherical interfaces
While the calculation of surface tension by simulation for
planar interfaces is essentially under control, for spherical
droplets there are still open questions about the dependence
of the surface tension on the drop size, and the validity of the
mechanical definition of the local pressure.6,10,33,148,149 The
methodology is still being developed and there are very few
comparisons with real experimental systems. Fig. 14 Curvature dependence of the liquid–vapour surface tension of
the LJ fluid at a temperature T* = 0.8. g/gN is plotted against the radius of
Drop size can be defined from the radius of the equimolar
the droplet where gN is the surface tension of the planar limit. , FMT
dividing surface (Re) which can be derived from the radial (a nonlocal DFT using functional measure theory).33 Thermodynamic
density profile r(r) routes: , grand canonical Monte Carlo simulations;152 , Monte
ð1 Carlo simulations using a thermodynamic route of Homman et al.;32 ,
1 drðrÞ
Re3 ¼ drr3 (19) the simulated surface tension calculated from TA ellipsoidal deforma-
ðrv  rl Þ 0 dr tions.33 Mechanical routes: , molecular dynamics simulations using the
mechanical route;35 , Monte Carlo simulations using the IK definition.32
where rv and rl are the vapour and liquid densities. The inset shows a magnification in the region of the maximum of g(Re)/gN.

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

These disagreements have been noted in many studies over where x is a small, dimensionless variable and Z represents the
the last 30 years.155 Although the DFT method allows a direct box length and the atom coordinate vectors, so that the
and unambiguous calculation of g, its application to realistic transformation preserves the spherical symmetry of the droplet.
systems with complex interactions is more difficult and it does V 0r and V 1r are the local configurational energy in the radial
not provide exact results. On the other hand molecular simulation element dr at r when the local volume is changed from V 0r to V1r .
is exact, can be readily applied to complex intermolecular forces The expression for the tangential component of the local
but the calculation of surface tension for a well-defined model pressure157 is
remains ambiguous. More recently Sampayo et al.,33 Lau et al.156
kB T
and Homman et al.32 have provided a macroscopic thermodynamic pT ðrÞ ¼ lim
x!0 xVr0
route to compute the surface tension through a revised version of  1 0 
V r ðr Þ  V 0r ðrÞ
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

N
the test-area method (TA) (see Section 2.1.3). These TA results for  ln ð1 þ xÞ exp  (24)
kB T mVr0 T
g(Re), shown in Fig. 14, are in much better agreement than the
results from the mechanical route via the pressure tensor with where m is the chemical potential. Eqn (24) provides the
the results obtained from surface free energy change,154 Landau r-dependent tangential component, pT. The normal component
free energy approach152 and density functional theory.153 is calculated through the condition of mechanical equilibrium.
In the following sections we shall consider these results A more convenient form of eqn (24) for use in a simulation of N
more carefully. molecules is
3.1 Mechanical routes to the local pressure tensor and surface X
N X
N
kB T
tension pT ðrÞ ¼ lim Fðr  ri Þ
x!0
i j4i
xVr0
As for the planar interfaces, two methods have been used to *  1 0   ! +
compute the local elements of the pressure: the Irving–Kirkood Nvr rij  v0r rij
 ln ð1 þ xÞ exp 
definition (IK)48 and the method of Harasima (Ha).150 For the kB T
mVr0 T
Ha method the transverse component is calculated directly and
(25)
the radial (normal) component is deduced from the condition
of mechanical equilibrium. Whereas, in the IK method, pN where F is the Heaviside step function and v(rij) the pair
is calculated directly and pT is deduced from rp = 0. potential.
More, recently, a thermodynamic route has been proposed for Lau et al.156 have recently extended the test-area method
droplets based on the calculation of the local energy, to with g  DA=DA, where A is the free energy and the surface area
compute the local components of the pressure tensor.157 to compute the surface tension of spherical interfaces.33,156 In
Several thermodynamic and statistical mechanics definitions order to conserve the volume of phase space, a spherical to
are available to compute the surface tension.6,10,34,158 Rowlinson6 elliptical transformation was employed. The expression for g is
and Lovett et al. suggest a mechanical route that is invariant to the  
5 b 2 
form of pressure tensor. From the condition of the mechanical g¼ hbi  a (26)
8pR2 2
stability, rp = 0, they showed that
ð1 where
g¼ drðRs =rÞðpN ðrÞ  pT ðrÞÞ; (20)  
0 1 @ 2 DV   1 @DV 2
hbi ¼ ; and a2 ¼
where Rs is the radius of tension6 2 @e2 2 @e
ð1 ð 1
and DV ¼ V 1r ðr0 Þ  V 0r ðrÞ. For the operational version of eqn (26)
Rs ¼ drrðdpN ðrÞ=drÞ drðdpN ðrÞ=drÞ: (21)
0 0 the reader is directed to the original paper.156 An entropic
  (gentropic) and energetic (genergetic) component of the total surface
From eqn (20), Lovett159 showed that to O R2 , where R is the tension (g = genergetic  gentropic) have been identified (see
size of the droplet, g is independent of Rs and is given by eqn (25) and (26) of Lau et al.156). By applying the test-area
ð1 method with the elliptical transformation, a small curvature
g¼ dr½ pN ðrÞ  pT ðrÞ: (22) dependence of the surface tension of the LJ system was found
0
while a strong curvature dependence of g was observed for small
water clusters.
3.2 Thermodynamic routes to the local pressure and surface
tension 3.3 Surface tension from the surface of tension
To obtain a the local pressure for a droplet, we assume a local The surface tension of the droplet can also be calculated in
partition function in the grand canonical ensemble.44,58 An isotropic simulation from the surface of tension (which is the radius at
change in the volume of the system is made by modifying the which the tension acts6) through the Laplace equation6,36,160
coordinates and the box dimensions in Cartesian space
ðDpÞRs
g¼ (27)
Z1a = Z0a(1 + x)1/3 (23) 2

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

where Rs is the radius of the surface of tension, g the surface The Tolman length calculated by this approach is normally
tension at Rs and Dp the pressure difference between the vapour small and negative153,162 but some carefully performed simulations
and liquid phases separated by the surface of tension. Usually, still suggest a positive value.162,163 This remains an unsolved issue.
Rs is determined from eqn (27) once the surface tension has
been determined by, say a mechanical route. However, Rs can
be calculated directly from a simulation using the approach of 4 Coarse-grained (CG) models
Hill.36 In this method, the system is modelled by a sphere of
liquid (a) surrounded by a vapour phase (b). The two phases are All the computer simulations discussed up until this point have
separated by a dividing surface of radius R. Thermodynamic been performed using atomistic force fields with different UA
equilibrium implies that (United Atom) and AA (All Atom) models. These atomistic
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

   models, often associated with Monte Carlo (MC) or molecular


@A 
 dynamics (MD) methods, have been extensively used to study
 ¼0 (28)
@R Ra ;Rb ;N;T  the behaviour of simple and complex systems on time scales
R¼Rs
extending over hundreds of nanoseconds. They are predicated
where Ra and Rb are radii inside the two phases at which the on atomistic force fields that perform well in terms of accuracy
density has reached its limiting values and Rs corresponds to and computational cost for the prediction of the surface tension
the minimum in the free energy.32 Simulations in the canonical of the liquid–vapour interfaces of pure systems and mixtures.
ensemble are performed using different trial values of Rt for the This approach is less obvious when we plan to use these force
surface of tension. The free energy derivative in eqn (28) is fields to model liquid–liquid interfaces or more complex systems
estimated using a perturbation approach, another application such as vesicles, membrane-bound proteins, polymers, or micelles.
of the test area method. In this case Rt is displaced with Ra and These systems often relax over times that are longer than a
Rb fixed. The system on one side of Rt expands while on the microsecond. In this case, we need to simplify the interactions
other side it contracts with an overall conservation of volume. by employing a coarse-grained (CG) model, in which the CG
The full expression for the derivative is given in eqn (12) of particle (often called the ‘bead’) may correspond to many atoms
Homman et al.32 Rt = Rs when the free energy derivative is zero. or molecules. Reducing the number of degrees of freedom in
The extraction of the surface tension from Laplace’s equation this way averages the structure of the component monomers but
follows once Dp = pl  pv is computed. There are no ambiguities in it is normally possible to preserve the geometry (architecture)
calculating the limiting pressures in the two bulk phases.34 As and the overall charge distribution of the polymer and to
shown in Fig. 14 the surface tensions calculated from thermo- increase the time-step of the simulation by an order of magnitude.
dynamic approaches are in good agreement with analytical theory. The reader is directed to a number of discussions of the
advantages and drawbacks of the CG models.164–169
3.4 Tolman length
There are two approaches to developing coarse-grained force
The Tolman length (d) is a measure of the deviation of the fields. The top-down approach derives the parameters from
surface tension with respect to the planar limit. It can be macroscopic properties (compressibility, diffusion, surface tension,
expressed as the limit in the planar case (R - +N) of the bulk density, and enthalpy of vaporisation)170–175 and requires
difference between the surface of tension Rs and the Gibbs the use of an extensive, experimental database. The bottom-up
equimolar dividing surface Re: approaches use an integration over the atomistic degrees of
d ¼ lim dðRÞ  ze  zs ; dðRÞ ¼ Re  Rs : (29) freedom and the corresponding atomistic force-field to develop
R!1 the mesoscopic model.176–181 Useful bottom-up CG strategies
where ze and zs, respectively, are the positions of the Gibbs include the iterative Boltzmann inversion (IBI)164,182,183 and
equimolar surface and the surface of tension of planar interface force matching (FM)184,185 schemes.
(with the surface normal along the z). While ze can be obtained Different CG force fields171,172,174,175,179,186–189 use different
by a fit of the density profile using an error function,161 zs is mappings as they develop beads from the underlying atoms.
more difficult to extract. These mappings can range from 4 : 1 (a CG particle represents
A straightforward modification of the previous approach for four heavy atoms plus associated hydrogen atoms)186 to 20 : 1
planar interfaces allows us to obtain the Tolman length32 for (one CG particle corresponds to 20 united atoms).179–181 Some
the droplet. This approach is based on the evaluation of the free models such as the MARTINI force field186–188,190–193 developed
energy derivative with respect to R using a test area approach. by Marrink et al. model the dispersion and overlap interactions
This method gives a Tolman length of d = 0.04  0.01. This with a 12-6 shifted Lennard-Jones potential, while Klein and
value is in fair agreement with the DFT calculation obtained co-workers171 prefer a 9-6 LJ potential, and Chiu et al.172 have
by van Giessen and Blokhuis (d = 0.1).162 An alternative employed a Morse potential to study alkanes and water. The
simulation approach to calculate d is to fit to the defining MARTINI model has been developed from the calculation of the
equation.153,157 free energy of hydration, the free energy of vaporization, and
the partitioning free energies between water and a number of
g 2d organic phases. Again, Avendano et al.174,175 have developed
¼1 : (30)
g1 Rs the (SAFT-g) CG field from the SAFT equation of state,194,195

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

using the Mie-generalized Lennard-Jones potential.196 All these potential, is not sufficiently steep to sustain a liquid–vapour
force fields can be readily used with standard MC and MD equilibrium. It is necessary to adapt the model so that coefficient a
methods. Other CG potentials can be used with mesoscopic depends on the local density to form a stable liquid–vapour
methods such as dissipative particle dynamics (DPD).26,197–199 interface (MDPD). A direct comparison between atomistic MC
The first simulations of the liquid–vapour interface of water and MDPD simulations has been performed to compare the
with the MARTINI CG force field187 gave values of surface surface tension, coexisting densities and profiles of the normal
tensions of 45 and 30 mN m1 at 298 K depending on the and tangential components of the pressure tensor along the
system-sizes (see Fig. 15). For the polarisable version of the surface normal for water. This MDPD method has also been
water MARTINI model,191 the simulated surface tension is extended to the case of the liquid–vapour interface of mixtures201
30 mN m1. This significant deviation from experiments (gexp. = and electrolytes.173 Calculations of g and the rl for water at room
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

73 mN m1) can be explained by the absence of significant temperature agree with the results of atomistic simulation and
ordering of the dipoles of the CG water particles in the inter- experiment to within ca. 5% and the coarse grained parameters are
facial region191 unlike the behaviour observed for the atomistic transferable over a range of 50 K.
models.202 The MARTINI model performs better in the case As discussed, the computer modelling of micellar systems203–207
of the liquid–vapour interface of alkanes.187 These types of is a particularly challenging problem due to the microsecond
molecules are the best candidate for a coarse-grained approach time-scale for the micellar relaxation and the length scale (the
since they are formed by a number of identical groups that simulation of tens of micelles of C12E6 in aqueous solution
interact predominantly through dispersion–repulsion inter- might require ca. a million atoms). Although leading-edge
actions. Another coarse-graining strategy, based on the models computers can perform simulations of this size and length, it
of Mie196 segments, has been successfully applied to calculate is difficult to use atomistic simulation in a routine way to fit
the surface tension of liquid–vapour interfaces174,175 of CO2, SF, models to experimental data and a coarse-grained approach is
CH4, and long linear alkanes including n-C10H22 and n-C20H42. required. In the case of the study of water–surfactant–oil
The surface tension of water26 has also been calculated complex systems, the prediction of the interfacial tension of
from the many-body dissipative particle dynamics (MDPD) the oil–water liquid–liquid interface represents a good test for
method199 by considering a top-down approach for the devel- these CG models. Fig. 15 shows the interfacial tension of some
opment of the potential parameters. We note the standard water–alkanes, water–benzene and water–chloroform calculated
version of DPD has a short-ranged conservative force usually from the CG MARTINI force field.59 This figure shows that we
represented by a linear ramp cannot expect a prediction of the interfacial tension of some
liquid–liquid systems with the CG MARTINI force field to be
fij = a(1  rij/rc)r̂ij rij r rc (31)
better than 10 mN m1. The calculation of the surface tension
where a determines the interaction between the DPD beads. with the MARTINI force field is less accurate when the interfaces
While this model gives rise to a perfectly satisfactory liquid– are characterized by strong local orientational ordering or
liquid interface, this force, and the corresponding quadratic hydrogen bonding.59 The definition of the CG element must
also consider if the atomic groups representing the bead are
undergoing internal changes.
Nevertheless, the interfacial tension of oil–water systems is
reasonably reproduced with the MARTINI model.59 The DPD
mesoscale simulations of water–benzene and water–cyclohexane200
systems predict interfacial tensions to an accuracy of 10 mN m1.
These CG models (MARTINI and DPD) also reproduce accurately
the dependence of the interfacial tension of these oil–water systems
with temperature200 and alkane chain length.59
The simulation of oil–surfactant–water mixtures needs large
systems, so that the results do not depend on the area of the
surface, and also long simulations since the equilibration of the
surfactant between the bulk and surface regions is slow. The
modelling of oil–surfactant–water systems has been investigated
by DPD simulations209–216 and using the MARTINI models.208,217
Fig. 15 The interfacial tensions, g of: n-hexane–water, n-octane–water,
These simulations accurately reproduce the dependence of the
n-decane–water, n-dodecane–water, n-hexadecane–water, cyclohexane– interfacial tension as a function of the surfactant concentration
water, benzene–water, chloroform–water, octan-1-ol-water;59 cyclohexane– at the interface. Fig. 16 shows the calculated interfacial tension of
water and benzene–water calculated using the DPD method;200 mixtures such the dodecane–water modified by the addition of SDS (sodium
as water–methanol, hexane–benzene, and water + NaCl salt calculated with
dodecyl sulfate) surfactants. Three different models have been
the DPD model;173,201 the liquid–vapour interfaces of pure water and pure
dodecane calculated using the MARTINI force field;187 the liquid–vapour
compared: the standard MARTINI force field187 with explicit
interfaces of the CF4 SF6 and n-eicosane systems;175 the liquid–vapour interface water molecules,191 an implicit-solvent-version of the MARTINI
of water using a CG model with the Morse potential.172 model,217 and CG potentials206 used with the DPD method.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

systems involving the self-assembly of amphiphiles. So far this


work has not been extended to nanoscopic droplet interface,
to design new surfactants of new surfactants for industrial
applications, of to understand how solid surface modifies g at
a neighbouring liquid–vapour interface.

5 Conclusions
The slab geometry is a useful technique for modelling liquid–
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

liquid and liquid–vapour planar interfaces and for calculating


the surface tensions from the triple point to within ca. 10%
of the critical point. For the system close to the critical point
the best route is to employ grand canonical Monte Carlo
Fig. 16 g/g0 ratio as a function Cs, the number of surfactants at the simulations and histogram-reweighting techniques to calculate g.
surface divided by the surface area. g0 is the interfacial tension of the There are a number methods for accurately calculating the
water–dodecane interface (with no surfactant). The values of the inter- surface tension from a slab geometry simulation. The mechanical
facial tensions are calculated with the DRY MARTINI model;208 the MAR- route to g involves the calculation of the pressure tensor, p. For the
TINI model with explicit solvent molecules and the DPD method.
planar interface, the transverse component of p cannot be deter-
mined uniquely as a function of z but all reasonable definitions
In order to allow the comparison between the different methods, we will lead to the same surface tension. In addition, the test area
report the results of interfacial tension as the ratio g/g0 where g0 is method is a simple and reliable thermodynamic method for the
the interfacial tension of the oil–water system without any calculation of g which involves scaling the surface area of the slab
surfactants. All the simulated interfacial tensions as a function at constant volume. These two routes are consistent once the
of surface coverage fall on one simple curve in Fig. 16; i.e., the appropriate long-range corrections are added to the intrinsic
interfacial tension decreases in a concave down fashion as the surface tension. The test area method has the advantage that it
surfaces are covered by more dodecyl sulphate anions. g reaches can be used with non-pair additive potentials.
a minimum finite value corresponding to the interfacial tension For the planar interface the computational procedures used
at the point of surfactant micellisation. At this point, the interface in the Monte Carlo and molecular dynamics simulations of the
is fully saturated with surfactant molecules. The mesoscopic (DPD) slab geometry are now under control and we have given some
simulations and MARTINI coarse-grained force fields correctly clear recommendations in this review which should lead to
reproduce the dependence of interfacial tension on surfactant accurate calculations of g for a particular model.
concentration. Fig. 17a shows a typical configuration of the oil/ In particular, the long-range corrections to the pressure are
water/SDS surfactant mixtures below the critical micelle concen- particularly important and can amount to as much as 30% of gI
tration (cmc) calculated from the MARTINI CG model whereas for a cut-off rc = 3.0s for the Lennard-Jones fluid. The Janec̆ek66
Fig. 17b presents a configuration of the same system near the cmc method provides an accurate calculation of the long range
where we observe the formation of surfactant micelle in the bulk correction for the pressure tensor when applied at each step
water phase. in the calculation and in terms of its implementation and
Mesoscopic simulation of CG models can be used to make speed, is the most promising approach. However, it is also
quantitative prediction of interfacial properties of complex possible to calculate these corrections by applying the Ewald
sum in the slab geometry for the dispersion term of the
Lennard-Jones potential.
Once these methodological issues are properly addressed,
the simulations will be predictive if the underlying potential
models are accurate. Most straightforward effective pairwise
potentials are not fitted to the surface tension so that its
calculation can be a severe test of such potentials. For example
for modelling g for argon, simulations with the LJ potential
exhibit systematic differences between experiment and simula-
tion of 20% for g and it requires the introduction of the 3-body
forces and an accurate 2-body potential to reduce this difference
to 2% along most of the orthobaric curve. One can argue that
given the accuracy in the experimental determination of g the
Fig. 17 Snapshots of the n-dodecane–water + surfactants at different
SDS concentrations (a) obtained from the MARTINI model at concentra-
simulation is predictive for the liquid–vapour interface of argon.
tions well below the cmc for the model. (b) For the DPD model at However, even for an approximately spherical molecule such as
concentrations well above the cmc for the model. CH4, where the polarisability is low, the agreement between the

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

simulated and experimental values of g is not satisfactory is better described by the more realistic polarisable model.
(although in this case the difference may be due to the neglect These polarisable model do not do such a good job on the
of the octopole–octopole electrostatic interaction). Even for surface tension. The underlying reasons for this are not well
atomic liquids the intrinsic planar surface is not smooth and understood and further experimental work and further work on
g will be a function of the wave vector describing the periodicity model building for these systems is required.
of the surface. g(k) can be calculated in simulations218 but to Another challenging topic is the calculation of the interfacial
date this k-dependence has not been observed experimentally tension of chemically reacting mixtures such as CO2 + NO2/
to the best of our knowledge. The calculation of interfacial N2O4221 by coupling the reaction ensemble Monte Carlo (RxMC)
properties of intrinsic surfaces will be an interesting area for method222–224 with the explicit modeling of the interface.
future simulations. A knowledge of g for surface tensions of droplet on surfaces
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

For a large range of molecular liquids the prediction of the is critical for the determination of spreading and wetting.149 In
surface tension is to within 15% of experiment. This includes reality all interfaces are curved, the planar interface is the limit
small molecules such as N2, H2S, alcohols, alkanes (up to C12), of the spherical droplet with an infinite radius of curvature and
and aromatics. In these simulations, it is necessary to introduce we expect to see important differences in g for small droplets.
the electrostatic interaction either as point multipoles, distributed However, there are still fundamental problems in defining the
multipoles or partial charges. For long-range interactions (where local components of the pressure tensor in an isolated spherical
the potential falls off more slowly than the dimensionality) these droplet. This results in a number of reasonable yet different
are included using an Ewald sum or one of the other techniques definitions of g through the mechanical route. As is clear from
for handling long range force. The real and reciprocal-space Fig. 14 there are also differences between the mechanical and
contributions to the pressure tensor are well understood and can thermodynamic routes as exemplified by the extensions of the
be calculated as a function of z through the interface. The first test area method. Although this is not a proof of the superiority
simulations of water were performed in 198031 and since then of one approach over the other, the thermodynamic route in
there have been over 1000 papers reporting simulations of various simulation is in much better agreement with the density
water interfaces. Currently one of the best predictive model for the functional-like theories than the mechanical route. The test
surface tension is TIP4P2005111 with an average deviation of 5% area method itself is quite complicated to apply to spherical
along the coexistence curve and a maximum deviation of 15% at droplet because we need to identify a linear transformation that
600 K (Tcritical = 645 K). It is important to have this model under is volume conserving. However, at this time, we would recommend
control since water is an important solvent in many industrial the TA33 approach for the calculation of g for droplets. The
applications of surfactancy. methodological problems for the small spherical droplets have
The simulation of mixtures allows us to extend the variables not been sufficiently resolved to perform meaningful, routine
that we consider from temperature to concentration and pressure. simulation on molecular systems and their mixtures in this
(The Gibbs rule tells us that for a single component systems at geometry. For example it would be difficult to use simulation to
each temperature there is a single point on the coexistence line but ask if there is a significant difference in the Tolman length of a
for mixtures there are many coexistence points on higher water or CO2 droplet. The whole area of long-range corrections
dimensional surfaces.) The development of good potentials for for these finite systems, which is so important for the calculation
complex mixture requires not just an understanding of the of g in the planar slab, has not been addressed. An understanding
single component interactions in a molecular fluid but also of the effects of system-size and cut-off dependence is difficult
the mixing rules between the unlike components. Most simulations mainly because we expect the properties of the droplet to change
to date have been performed using the venerable Lorentz–Berthelot with size and droplet radius. These difficulties mean that many
mixing rule219,220 and future work will involve accurately para- simulations of the droplet are run with truncated and shifted
metrising cross interactions between important components Lennard-Jones potentials, which we know to be problematic for
such as CO2 and H2O. This is true for the important dispersion planar interfaces.
interaction, while on the other hand the electrostatic interac- The simulation of coarse-grained models for the calculation
tions are quite transferable between like and unlike species. of interfacial properties is an important area of active research.
The preferred ensemble for these simulations is constant- We are still seeking of the correct level of coarse-graining for
NpN AT. For two component mixtures such as C5H12/H2S and different problems (such as surface wetting, polymer micelle
H2O/CO2 good predictions of g as a function of pressure are interaction and lubrication). An important feature of the problem
obtained for pressure up to 40 MPa. The results are important is to build a mesoscale model from an underpinning atomistic
for the storage of CO2. The risk of leakage through the cap rock model. This is not straightforward because techniques such as
is governed by the water acid gas interfacial tension through the iterative Boltzmann inversion or force-matching while applicable
Laplace equation.90 The simulation of the interfacial properties of to homogeneous fluid are difficult to use in interfacial simulations
the salt solution is also a challenging and interesting problem. because the effective potential may depend strongly on the local
The driving force for this work is an understanding of water density. In multi-scale simulation we have a loose coupling of the
purification by osmosis. We observe that the increase in g with levels by transfer of parameters, such as g calculated at, say, the
molality is well described by non-polarisable effective pair atomistic to the mesoscale (i.e. the levels are independent). In an
potentials but the arrangement of the molecules at the interface alternative approach it is possible to have a close-coupling of the

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

levels where the behaviour at, say, the mesoscale demands that the 5 S. Hartland, Surface and Interfacial Tension: Measurement,
simulation be carried forward at the atomistic scale. To date, we Theory, and Applications, Marcel Dekker, New York Basel,
are unaware of simulations of surface tension that involve this 2004.
second kind of close coupling and this will be an important 6 J. S. Rowlinson and B. Widom, Molecular Theory of Capillarity,
development for the future. Another interesting topic that could Clarendon Press, Oxford, 1982.
be addressed by using the theoretical background of the spherical 7 D. Nicholson and N. Parsonage, Computer Simulation and
interfaces and the use of GC models157 is the question of the the Statistical Mechanics of Adsorption, Academic Press,
prediction of the spontaneous curvature and elasticity of inter- New York, 1982.
facial films225,226 in the case of phase equilibria of water-in-oil 8 C. Croxton, Statistical Mechanics of the Liquid Surface,
microemulsions. Wiley, New York, 1980.
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

Simulations of g are also important in the oil and gas industry 9 J.-P. Hansen and I. R. McDonald, Theory of Simple Liquids,
because we can predict the temperature and pressure dependence Academic Press, London, 2nd edn, 1986.
in regions that are not readily accessible in laboratory experiment. 10 P. Schofield and J. R. Henderson, Proc. R. Soc. London, Ser.
The general extension of these methods to multi-component A, 1982, 379, 231–246.
mixtures is straightforward. Many important detergency problems 11 J. H. Irving and J. Kirkwood, J. Chem. Phys., 1950, 18, 817–829.
in industry are tackled using 5–7 component mixtures of 12 A. Harasima, Adv. Chem. Phys., 1958, 1, 203–237.
surfactants and simulation will be an important part in the 13 F. P. Buff, Z. Elektrochem., 1952, 56, 311–313.
optimisation of these products. In the future, we will need a 14 F. P. Buff, J. Chem. Phys., 1955, 23, 419–427.
large system to include a significant number of molecules of 15 M. Grant and R. C. Desai, J. Chem. Phys., 1980, 72, 1482–1486.
each species and longer times to ensure equilibration of the 16 S. Toxvaerd, Prog. Surf. Sci., 1972, 3, 189–220.
species between the phases, but the principles discussed in the 17 G. J. Gloor, G. Jackson, F. J. Blas and E. de Miguel, J. Chem.
review remain the same. Fortunately computer power continue Phys., 2005, 123, 134703.
to increase and it is likely that leading-edge machines of 1018 18 A. Ghoufi, F. Goujon, V. Lachet and P. Malfreyt, J. Chem.
Flops will become available in the next five years and more Phys., 2008, 128, 154716.
general machine of this power available five years beyond that. 19 A. Marchand, J. H. Weijs, J. H. Snoeijer and B. Andreotti,
These speed increases will occur because of the parallelisation Am. J. Phys., 2011, 79, 999–1001.
of the algorithms and all of the techniques discussed in this 20 M. P. Allen and D. J. Tildesley, Computer Simulation of
review can be run on parallel machines using domain decom- Liquids, Clarendon Pr, Oxford, 1987.
position and other techniques. We can envisage a powerful code 21 D. Frenkel and B. Smit, Understanding Molecular Simula-
for the calculation of the surface tension and other interfacial tion: From Algorithms to Applications, Academic Press, 1996.
properties of multi-component mixtures that will widely be used 22 J. J. Potoff and A. Panagiotopoulos, J. Chem. Phys., 2000,
to guide the experimental design of multi-component surfactant 112, 6411–6415.
mixtures. While the quantitative prediction of the exact surface 23 J. R. Errington, J. Chem. Phys., 2003, 67, 012102.
tension of multi-component may still be a difficult problem, 24 V. K. Shen and J. R. Errington, J. Chem. Phys., 2006,
these simulation codes will enable us to study the changes and 124, 024721.
trends in g with the addition of minor components. 25 B. Chen, J. I. Siepmann and M. L. Klein, J. Am. Chem. Soc.,
2002, 124, 12332–12337.
26 A. Ghoufi and P. Malfreyt, Phys. Rev. E: Stat., Nonlinear, Soft
Acknowledgements Matter Phys., 2011, 83, 051601.
27 A. E. Ismail, G. S. Grest and M. J. Stevens, J. Chem. Phys.,
We wish to thank Florent Goujon, University Blaise Pascal,
2006, 125, 014702.
Véronique Lachet, IFPen, Emeric Bourasseau, CEA, Massimo
28 C. Vega and E. de Miguel, J. Chem. Phys., 2007, 126, 154707.
Noro and Patrick Warren, Unilever Research, for useful discussions
29 J. Alejandre, D. J. Tildesley and G. A. Chapela, J. Chem.
in the preparation of this review.
Phys., 1995, 102, 4574–4583.
30 M. Matsumoto and Y. Kataoka, J. Chem. Phys., 1988, 88,
References 3233–3245.
31 C. Y. Lee and H. L. Scott, J. Chem. Phys., 1980, 73, 4591–4596.
1 K. Kosswig, Ullmann’s Encyclopedia of Industrial Chemistry, 32 A.-A. Homman, E. Bourasseau, G. Stoltz, P. Malfreyt,
Wiley-VCH, New York, 2000, pp. 431–503. L. Strafella and A. Ghoufi, J. Chem. Phys., 2014, 140, 034110.
2 M. Bavière, Basic Concepts in Enhanced Oil Recovery Processes, 33 J. G. Sampayo, A. Malijevský, E. A. Müller, E. de Miguel and
Elsevier, 1991. G. Jackson, J. Chem. Phys., 2010, 132, 141101.
3 D. Bonn, J. Eggers, J. Indekeu, J. Meunier and E. Rolley, 34 S. M. Thompson, K. E. Gubbins, J. P. R. B. Walton,
Rev. Mod. Phys., 2009, 81, 739–805. R. A. R. Chantry and J. S. Rowlinson, J. Chem. Phys.,
4 M. Fuerstenau, Froth Flotation a Century of Innovation, 1984, 81, 530–542.
Society for Mining, Metallurgy, and Exploration, Littleton, 35 J. Vrabec, G. K. Kediaa, G. Fuchs and H. Hasse, Mol. Phys.,
Colorado, 2007. 2006, 104, 1509–1527.

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

36 T. L. Hill, J. Phys. Chem., 1952, 56, 526–531. 66 J. Janec̆ek, J. Chem. Phys., 2006, 131, 6264–6269.
37 R. C. Tolman, J. Chem. Phys., 1949, 17, 333–337. 67 R. H. Fowler, Proc. R. Soc. London, Ser. A, 1937, 159, 229–246.
38 J. M. Mı́guez, M. M. Piñeiro and F. J. Blas, J. Chem. Phys., 68 K. S. C. Freeman and I. R. McDonald, Mol. Phys., 1973, 26,
2013, 138, 34707–34716. 529–537.
39 P. Orea, J. Lopez-Lemus and J. Alejandre, J. Chem. Phys., 69 J. Alejandre, D. J. Tildesley and G. A. Chapela, Mol. Phys.,
2005, 123, 114702. 1995, 85, 651–663.
40 A. Trokhymchuk and J. Alejandre, J. Chem. Phys., 1999, 111, 70 F. Biscay, A. Ghoufi, F. Goujon, V. Lachet and P. Malfreyt,
8510–8523. J. Phys. Chem. B, 2008, 112, 13885–13897.
41 M. Guo and B. Lu, J. Chem. Phys., 1997, 106, 3688–3695. 71 J. Alejandre and G. A. Chapela, J. Chem. Phys., 2010,
42 G. A. Chapela, G. Saville, S. M. Thompson and J. S. Rowlinson, 132, 014701.
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

J. Chem. Soc., Faraday Trans. 2, 1977, 73, 1133–1144. 72 R. E. Isele-Holder, W. Mitchell and A. E. Ismail, J. Chem.
43 K. S. Liu, J. Chem. Phys., 1974, 60, 4226–4230. Phys., 2012, 157, 174107.
44 A. Ghoufi, F. Goujon, V. Lachet and P. Malfreyt, Phys. Rev. 73 R. E. Isele-Holder, W. Mitchell, J. R. Hammond,
E: Stat., Nonlinear, Soft Matter Phys., 2008, 77, 031601. A. Kohlmeyer and A. E. Ismail, J. Chem. Theory Comput.,
45 A. P. Lyubartsev, A. A. Martsinovski, S. Shevkunov and 2013, 9, 5412–5420.
P. N. Vorontsovvelyaminov, J. Chem. Phys., 1992, 96, 1776–1783. 74 D. Tameling, P. Springer, P. Bientinesi and A. E. Ismail,
46 K. Binder, Phys. Rev. A: At., Mol., Opt. Phys., 1982, 25, J. Chem. Phys., 2014, 140, 024105.
1699–1709. 75 F. Goujon, A. Ghoufi, P. Malfreyt and D. J. Tildesley,
47 C. H. Bennett, J. Comput. Phys., 1976, 22, 245–268. J. Chem. Theory Comput., 2015, 11, 4575–4585.
48 J. G. Kirkwood and F. P. Buff, J. Chem. Phys., 1949, 17, 76 M. Guo and B. C. Y. Lu, J. Chem. Phys., 1998, 109, 1134.
338–343. 77 M. Mecke and J. Winkelmann, J. Chem. Phys., 1997, 107,
49 R. A. Zubillaga, A. Labastida, B. Cruz, J. C. Martinez, 9264–9270.
E. Sanchez and J. Alejandre, J. Chem. Theory Comput., 78 M. Mecke, J. Winkelmann and J. Fischer, J. Chem. Phys.,
2013, 9, 1611–1615. 1999, 110, 1188–1194.
50 P. Jungwirth and D. J. Tobias, J. Phys. Chem. B, 2001, 105, 79 L. G. MacDowell and F. Blas, J. Chem. Phys., 2009, 131,
10468–10472. 074705.
51 B. J. Alder and T. E. Wainwright, J. Chem. Phys., 1957, 27, 80 F. J. Martinez-Ruiz, F. J. Blas, B. Mendiboure and A. I. M.-V.
1208–1209. Bravo, J. Chem. Phys., 2014, 141, 184701.
52 W. W. Wood and F. R. Parker, J. Chem. Phys., 1957, 27, 81 F. Goujon, P. Malfreyt, A. Boutin and A. H. Fuchs, Mol.
720–733. Simul., 2001, 27, 99–114.
53 J. P. R. B. Walton, D. J. Tildesley, J. S. Rowlinson and 82 F. Goujon, P. Malfreyt, A. Boutin and A. H. Fuchs, J. Chem.
J. R. Henderson, Mol. Physiol., 1983, 48, 1357–1368. Phys., 2002, 116, 8106–8117.
54 J. P. R. B. Walton, D. J. Tildesley, J. S. Rowlinson and 83 N. Ferrando, V. Lachet, J. Pérez-Pellitero, A. D. Mackie,
J. R. Henderson, Mol. Physiol., 1983, 50, 1381. P. Malfreyt and A. Boutin, J. Phys. Chem. B, 2011, 115,
55 A. Ghoufi and P. Malfreyt, Mol. Phys., 2006, 104, 2929–2943. 10654–10664.
56 F. Goujon, P. Malfreyt, J. M. Simon, A. Boutin, B. Rousseau 84 J. C. Neyt, A. Wender, V. Lachet and P. Malfreyt, J. Phys.
and A. H. Fuchs, J. Chem. Phys., 2004, 121, 12559–12571. Chem. C, 2012, 116, 10563–10572.
57 J. L. Rivera, C. McCabe and P. T. Cummings, Phys. Rev. E: 85 J. C. Neyt, A. Wender, V. Lachet, A. Ghoufi and P. Malfreyt,
Stat., Nonlinear, Soft Matter Phys., 2003, 67, 011603. J. Chem. Phys., 2013, 139, 024701.
58 C. Ibergay, A. Ghoufi, F. Goujon, P. Ungerer, A. Boutin, 86 V. K. Shen, R. D. Mountain and J. R. Errington, J. Phys.
B. Rousseau and P. Malfreyt, Phys. Rev. E: Stat., Nonlinear, Chem. B, 2007, 111, 6198–6207.
Soft Matter Phys., 2007, 75, 051602. 87 J. Janec̆ek, H. Krienke and G. Schmeer, J. Phys. Chem. B,
59 M. Ndao, J. Devemy, A. Ghoufi and P. Malfreyt, J. Chem. 2006, 110, 6916–6923.
Theory Comput., 2015, 11, 3818–3828. 88 J. Benet, L. G. MacDowell and C. Menduiña, J. Chem. Eng.
60 F. Biscay, A. Ghoufi, F. Goujon, V. Lachet and P. Malfreyt, Data, 2010, 55, 5465–5470.
J. Chem. Phys., 2009, 130, 184710. 89 F. Goujon, P. Malfreyt and D. J. Tildesley, J. Chem. Phys.,
61 E. Salomons and M. Mareschal, J. Phys.: Condens. Matter, 2014, 140, 244710.
1991, 3, 9215–9228. 90 F. Biscay, A. Ghoufi, V. Lachet and P. Malfreyt, J. Phys.
62 E. M. Blokhuis, D. Bedaux, C. D. Holcomb and Chem. B, 2009, 113, 14277–14290.
J. A. Zollweg, Mol. Physiol., 1995, 85, 665–669. 91 J. R. Errington and D. A. Kofke, J. Chem. Phys., 2007,
63 M. Guo, B. Y. Peng and C. Y. Lu, Fluid Phase Equilib., 1997, 127, 174709.
130, 19–30. 92 M. E. Velazquez, A. Gama-Goicochea, M. Conzalez-
64 J. Lopez-Lemus and J. Alejandre, Mol. Phys., 2002, 100, Melchor, M. Neria and J. Alejandre, J. Chem. Phys., 2006,
2983–2992. 124, 084104.
65 J. Lopez-Lemus and J. Alejandre, Mol. Phys., 2003, 101, 93 S. Werth, S. V. Lishchuk, M. Horsch and H. Hasse, Physica
743–751. A, 2013, 392, 2359–2367.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

94 G. Filippini, E. Bourasseau, A. Ghoufi, F. Goujon and 123 J. Lekner, Physica A, 1991, 176, 485–498.
P. Malfreyt, J. Chem. Phys., 2014, 141, 081103. 124 J. Hautman and M. Klein, Mol. Physiol., 1992, 75, 379–395.
95 E. Bourasseau, A. Homman, O. Durand, A. Ghoufi and 125 T. M. Chang and L. X. Dang, J. Phys. Chem. B, 2005, 109,
P. Malfreyt, Eur. Phys. J. B, 2013, 86, 251. 5759–5765.
96 K. Binder, Phys. Rev. A: At., Mol., Opt. Phys., 1982, 25, 126 F. Biscay, A. Ghoufi and P. Malfreyt, J. Chem. Phys., 2011,
1699–1709. 134, 044709.
97 B. M. Axilrod and E. Teller, J. Chem. Phys., 1943, 11, 299–300. 127 E. A. Müller and A. Meja, Fluid Phase Equilib., 2009, 282,
98 J. A. Barker, R. A. Fisher and R. O. Watts, Mol. Physiol., 68–81.
1971, 21, 657–673. 128 A. R. van Buuren, S. J. Marrink and H. J. C. Berendsen,
99 A. E. Nasrabad, R. Laghaei and U. K. Deiters, J. Chem. Phys., J. Phys. Chem., 1993, 97, 9206–9212.
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

2004, 121, 6423–6434. 129 Y. Zhang, S. E. Feller, B. R. Brooks and R. W. Pastor,


100 S. Werth, M. Horsch, J. Vrabec and H. Hasse, J. Chem. J. Chem. Phys., 1995, 23, 10252–10266.
Phys., 2015, 140, 244710. 130 J. L. Rivera and J. Alejandre, Colloids Surf., A, 2002, 207,
101 F. Goujon, P. Malfreyt and D. J. Tildesley, J. Chem. Phys., 223–228.
2015, 142, 107102. 131 S. A. Patel and C. L. Brooks, J. Chem. Phys., 2006, 124, 204706.
102 G. Ciccotti and J. P. Ryckaert, Comput. Phys. Rep., 1986, 4, 132 F. Biscay, A. Ghoufi, V. Lachet and P. Malfreyt, J. Chem.
345–392. Phys., 2009, 131, 124707.
103 R. L. C. Akkermans and W. J. Briels, J. Chem. Phys., 2001, 133 J. M. Garrido, L. Cifuentes, M. Cartes, H. Segura and
114, 1020–1031. A. Meja, J. Supercrit. Fluids, 2014, 89, 78–88.
104 W. K. den Otter, M. Kröhn and J. H. R. Clarke, Phys. Rev. E: 134 A. Z. Panagiotopoulos, Mol. Physiol., 1987, 61, 813–826.
Stat., Nonlinear, Soft Matter Phys., 2001, 65, 016704. 135 A. Z. Panagiotopoulos, N. Quirke, M. Stapleton and
105 R. L. C. Akkermans and G. Ciccotti, J. Phys. Chem. B, 2004, D. J. Tildesley, Mol. Phys., 1988, 63, 527–545.
108, 6866–6869. 136 P. T. van Duijnen and M. Swart, J. Phys. Chem. A, 1998, 102,
106 G. Lamoureux, A. D. MacKerell and B. Roux, J. Chem. Phys., 2399–2407.
2003, 119, 5185–5197. 137 S. Y. Noskov, G. Lamoureux and B. Roux, J. Phys. Chem. B,
107 J. C. Neyt, A. Wender, V. Lachet, A. Ghoufi and P. Malfreyt, 2005, 109, 6705–6713.
Phys. Chem. Chem. Phys., 2013, 15, 11679–11690. 138 D. Elking, T. Darden and R. J. Woods, J. Comput. Chem.,
108 L. Viererblova and J. Kolafa, Phys. Chem. Chem. Phys., 2011, 2007, 28, 1261–1274.
13, 19925–19935. 139 J. Gasteiger and M. Marsili, Tetrahedron Lett., 1978,
109 H. J. C. Berendsen, J. R. Grigera and T. P. Straatsma, 3181–3184.
J. Phys. Chem., 1987, 91, 6269–6271. 140 A. K. Rappe and W. A. Goddard, J. Phys. Chem., 1991, 95,
110 W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey 3358–3363.
and M. L. Klein, J. Chem. Phys., 1983, 79, 926–935. 141 G. Lamoureux, E. Harder, I. V. Vorobyov, B. Roux and
111 J. L. F. Abascal and C. Vega, J. Chem. Phys., 2005, 123, J. A. D. MacKerell, Chem. Phys. Lett., 2006, 418, 245–249.
234505. 142 W. Yu, P. E. M. Lopes, B. Roux and A. D. MacKerell,
112 D. A. Hernandez and H. Dominguez, J. Chem. Phys., 2013, J. Chem. Phys., 2013, 138, 034508.
138, 134702. 143 D. Bhatt, J. Newman and C. J. Radke, J. Phys. Chem. B, 2004,
113 F. Biscay, A. Ghoufi, V. Lachet and P. Malfreyt, J. Phys. 108, 9077–9084.
Chem. C, 2011, 115, 8670–8683. 144 J. C. Neyt, A. Wender, V. Lachet, A. Szymczyk, A. Ghoufi
114 J. C. Neyt, A. Wender, V. Lachet and P. Malfreyt, J. Phys. and P. Malfreyt, Chem. Phys. Lett., 2014, 595, 209–213.
Chem. B, 2011, 115, 9421–9430. 145 R. D’Auria and D. J. Tobias, J. Phys. Chem. A, 2009, 113,
115 W. L. Jorgensen and J. Tirado-Rives, Proc. Natl. Acad. Sci. 7286–7293.
U. S. A., 2005, 102, 6665–6670. 146 I. Leontyev and A. Stuchebrukhov, J. Chem. Phys., 2009,
116 C. Caleman, P. J. van Maaren, M. Hong, J. S. Hub, 130, 085102.
L. T. L. T. Costa and D. van der Spoel, J. Chem. Theory 147 I. Leontyev and A. Stuchebrukhov, Phys. Chem. Chem.
Comput., 2012, 8, 61–74. Phys., 2011, 13, 2613–2626.
117 J. Alejandre, J. L. Rivera, M. A. Mora and V. de la Garza, 148 E. M. Blokhuis and D. Bedeaux, J. Chem. Phys., 1992, 97, 3576.
J. Phys. Chem. B, 2000, 104, 1332–1337. 149 A. Malijevsky and G. Jackson, J. Phys.: Condens. Matter,
118 S. Nosé and M. Klein, Mol. Physiol., 1983, 50, 1055–1076. 2012, 24, 464121.
119 J. M. Mı́guez, D. G. Salgado, J. L. Legido and M. Piñeiro, 150 A. Harasima, Adv. Chem. Phys., 1952, 1, 203.
J. Chem. Phys., 2010, 132, 184102. 151 T. Nakamura, W. Shinoda and T. Ikeshoji, J. Chem. Phys.,
120 I. C. Yeh and M. Berkowitz, J. Chem. Phys., 1999, 111, 2011, 135, 094106.
3155–3162. 152 M. Schrader, P. Virnau and K. Binder, Phys. Rev. E: Stat.,
121 R. M. Venable, L. E. Chen and R. W. Pastor, J. Phys. Chem. Nonlinear, Soft Matter Phys., 2009, 79, 061104.
B, 2009, 113, 5855–5862. 153 B. J. Block, S. K. Das, M. Oettel, P. Virnau and K. Binder,
122 J. Lekner, Physica A, 1989, 157, 826–838. J. Chem. Phys., 2010, 133, 154702.

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.
View Article Online

Review Article Chem Soc Rev

154 H. E. Bardouni, M. Mareschal, R. Lovett and M. Baus, 185 W. G. Noid, J. W. Chu, G. S. Ayton, V. Krishna, S. Izvekov,
J. Chem. Phys., 2000, 113, 9804. G. A. Voth, A. Das and H. C. Andersen, J. Chem. Phys., 2008,
155 Y. A. Lei, T. Bykov, S. Yoo and X. C. Zeng, J. Am. Chem. Soc., 128, 244114.
2005, 127, 15346–15347. 186 S. J. Marrink, A. H. de Vries and A. E. Mark, J. Phys. Chem.
156 G. V. Lau, I. J. Ford, P. A. Hunt, E. A. Müller and G. Jackson, B, 2004, 108, 750–760.
J. Chem. Phys., 2015, 142, 114701. 187 S. J. Marrink, H. J. Risselada, S. Yefimov, D. P. Tieleman
157 A. Ghoufi and P. Malfreyt, J. Chem. Phys., 2011, 135, 104105. and A. H. de Vries, J. Phys. Chem. B, 2007, 111, 7812–7824.
158 S. J. Hemingway, J. R. Henderson and J. S. Rowlinson, 188 L. Monticelli, S. K. Kandasamy, X. Periole, R. G. Larson,
Faraday Symp. Chem. Soc., 1981, 16, 33. D. P. Tieleman and S. J. Marrink, J. Chem. Theory Comput.,
159 R. Lovett and M. Baus, J. Chem. Phys., 1996, 106, 635. 2008, 4, 819–834.
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

160 J. Gibbs, The Collected Works of J. Willard Gibbs, Yale 189 K. A. Maerzke and J. I. Siepmann, J. Phys. Chem. B, 2011,
University Press, New Haven, 1948. 115, 3452–3465.
161 S. Senapati and M. L. Berkowitz, Phys. Rev. Lett., 2001, 190 C. A. Lopez, A. J. Rzepiela, A. H. de Vries, L. Dijkhuizen,
17, 176101. P. H. Hünenberger and S. J. Marrink, J. Chem. Theory
162 A. E. van Giessen and E. M. Blokhuis, J. Chem. Phys., 2009, Comput., 2009, 5, 3195–3210.
131, 164705. 191 S. O. Yesylevskyy, L. V. Schäfer, D. Sengupta and
163 M. J. Haye and C. Bruin, J. Chem. Phys., 1994, 100, 556–559. S. J. Marrink, PLoS Comput. Biol., 2010, 6, 1–17.
164 F. Müller-Plathe, ChemPhysChem, 2002, 3, 754–769. 192 D. Sergi, G. Scocchi and A. Ortona, J. Chem. Phys., 2012,
165 J. W. Chu, S. Izbeko and G. A. Voth, Mol. Simul., 2006, 32, 137, 094904.
211–218. 193 S. J. Marrink and D. P. Tieleman, Chem. Soc. Rev., 2013, 42,
166 C. Peter and K. Kremer, Soft Matter, 2009, 5, 4357–4366. 6801–6822.
167 G. A. Voth, Coarse-Graining of Condensed Phase and Bio- 194 W. Chapman, K. Gubbins, G. Jackson and M. Radosz, Fluid
molecular Systems, CRC Press, 2009. Phase Equilib., 1989, 52, 31–38.
168 C. Peter and K. Kremer, Faraday Discuss., 2010, 144, 9–24. 195 W. Chapman, K. Gubbins, G. Jackson and M. Radosz, Ind.
169 Y. Li, B. C. Abberton, M. Kröger and W. K. Liu, Polymers, Eng. Chem. Res., 1990, 29, 1709–1721.
2013, 5, 751–832. 196 G. Mie, Ann. Phys., 1903, 316, 657–697.
170 R. D. Groot and K. L. Rabone, Biophys. J., 2001, 81, 725–736. 197 P. J. Hoogerbrugge and J. M. V. A. Koelman, Europhys. Lett.,
171 R. DeVane, W. Shinoda, P. B. Moore and M. L. Klein, 1992, 19, 155–160.
J. Chem. Theory Comput., 2009, 5, 2115–2124. 198 R. D. Groot and P. B. Warren, J. Chem. Phys., 1997, 107,
172 S. W. Chiu, H. L. Scott and E. Jakobsson, J. Chem. Theory 4423–4435.
Comput., 2010, 6, 851–863. 199 P. B. Warren, Phys. Rev. E: Stat., Nonlinear, Soft Matter
173 A. Ghoufi and P. Malfreyt, J. Chem. Theory Comput., 2012, 8, Phys., 2003, 68, 066702.
787–791. 200 E. Mayoral and A. G. Goicochea, J. Chem. Phys., 2013,
174 C. Avendano, T. Lafitte, A. Galindo, C. S. Adjiman, G. Jackson 138, 094703.
and E. A. Müller, J. Phys. Chem. B, 2011, 115, 11154–11169. 201 A. Ghoufi and P. Malfreyt, Eur. Phys. J. E: Soft Matter Biol.
175 C. Avendano, T. Lafitte, A. Galindo, C. S. Adjiman, G. Jackson Phys., 2013, 36, 10.
and E. A. Müller, J. Phys. Chem. B, 2013, 117, 2717–2733. 202 V. P. Sokhan and D. J. Tildesley, Mol. Phys., 1997, 92,
176 T. Spyriouni, C. Tzoumanekas, D. Theodorou, F. Müller- 625–640.
Plathe and G. Milano, Macromolecules, 2007, 40, 3876–3885. 203 Z. Li and E. E. Dormidontova, Macromolecules, 2010, 43,
177 V. A. Harmandis and K. Kremer, Macromolecules, 2009, 42, 3521–3531.
791–802. 204 A. Vishnyakov, M. T. Lee and A. V. Neimark, J. Phys. Chem.
178 T. Mulder, V. A. Harmandaris, A. V. Lyulin, N. F. A. van der Lett., 2013, 4, 797–802.
Vegt, K. Kremer and M. A. J. Michels, Macromolecules, 205 M. T. Lee, A. Vishnyakov and A. V. Neimark, J. Phys. Chem.
2009, 42, 384–391. B, 2013, 117, 10304–10310.
179 G. Maurel, B. Schnell, F. Goujon, M. Couty and P. Malfreyt, 206 Z. Mai, E. Couallier, M. Rakib and B. Rousseau, J. Chem.
J. Chem. Theory Comput., 2012, 8, 4570–4579. Phys., 2014, 140, 204902.
180 G. Maurel, F. Goujon, B. Schnell and P. Malfreyt, RSC Adv., 207 X. Tang, P. H. Koenig and R. G. Larson, J. Phys. Chem. B,
2015, 5, 14065–14073. 2014, 18, 3864–3880.
181 G. Maurel, F. Goujon, B. Schnell and P. Malfreyt, J. Phys. 208 S. Wang and R. G. Larson, Langmuir, 2015, 31, 1262–1271.
Chem. C, 2015, 119, 4817–4826. 209 R. D. Groot, Langmuir, 2000, 16, 7493–7502.
182 D. Reith, M. Pütz and F. Müller-Plathe, J. Comput. Chem., 210 L. Rekvig, M. Kranenburg, J. Vreede, B. Hafskjold and
2003, 24, 1624–1636. B. Smit, Langmuir, 2003, 19, 8195–8205.
183 A. P. Lyubartsev and A. Laaksonen, Phys. Rev. E: Stat. Phys., 211 L. Rekvig, M. Kranenburg, B. Hafskjold and B. Smit, Europhys.
Plasmas, Fluids, Relat. Interdiscip. Top., 1995, 52, 3730–3737. Lett., 2003, 63, 902–907.
184 S. Izvekov and G. A. Voth, J. Phys. Chem. B, 2005, 109, 212 Y. Li, P. Zhang, F. L. Dong, X. L. Cao, X. W. Song and
2469–2473. X. H. Cui, J. Colloid Interface Sci., 2005, 290, 275–280.

Chem. Soc. Rev. This journal is © The Royal Society of Chemistry 2016
View Article Online

Chem Soc Rev Review Article

213 Y. Li, X. He, X. Cao, G. Zhao, X. Tian and X. H. Cui, 220 D. Berthelot, C. R. Hebd. Seances Acad. Sci., 1898, 126,
J. Colloid Interface Sci., 2007, 307, 215–220. 1703–1855.
214 Z. Chen, X. Cheng, H. Cui, P. Cheng and H. Wang, Colloids 221 E. Bourasseau, V. Lachet, N. Desbiens, J.-B. Maillet, J.-M. Teuler
Surf., A, 2007, 301, 437–443. and P. Ungerer, J. Phys. Chem. B, 2008, 112, 15783–15792.
215 Y. Li, Y. Guo, M. Bao and X. Gao, J. Colloid Interface Sci., 222 W. R. Smith and B. Triska, J. Chem. Phys., 1994, 100, 3019–3027.
2011, 361, 573–580. 223 J. Johnson, A. Panagiotopoulos and K. Gubbins, Mol. Phys.,
216 V. V. Ginzburg, K. Chang, P. K. Jog, A. B. Argenton and 1994, 81, 717–733.
L. Rakesh, J. Phys. Chem. B, 2011, 115, 4654–4661. 224 C. H. Turnera, J. K. Brennan, M. Lisal, W. R. Smith,
217 C. Arnarez, J. J. Uusitalo, M. F. Masman, H. I. Ingolfsson, J. K. Johnson and K. E. Gubbins, Mol. Simul., 2008, 34,
D. H. de Jong, M. N. Melo, X. Periole, A. H. de Vries and 119–1146.
Published on 08 January 2016. Downloaded by University of Sussex on 09/01/2016 08:21:39.

S. J. Marrink, J. Chem. Theory Comput., 2014, 11, 260–275. 225 R. Leung and D. O. Shah, J. Colloid Interface Sci., 1987, 120,
218 E. Chacón and P. Tarazona, Phys. Rev. Lett., 2003, 96, 166103. 320–329.
219 H. A. Lorentz, Ann. Phys., 1881, 248, 127–136. 226 D. Langevin, Annu. Rev. Phys. Chem., 1992, 43, 341–349.

This journal is © The Royal Society of Chemistry 2016 Chem. Soc. Rev.

You might also like