Fluid Phase Equilibria 407 (2016) 7683

Contents lists available at ScienceDirect

Fluid Phase Equilibria

journal homepage: www.elsevier.com/locate/fluid

Osmotic pressure of aqueous electrolyte solutions via molecular

simulations of chemical potentials: Application to NaCl
William R. Smith a,b, , Filip Moucka c , Ivo Nezbeda c,d
a

Dept. of Mathematics and Statistics, University of Guelph, Guelph, ON N1G 2W1, Canada
Faculty of Science, University of Ontario Institute of Technology, Oshawa, ON L1H 7K4, Canada
Faculty of Science, J. E. Purkinje University, 400 96 st n. Lab., Czech Republic
d
Inst. of Chem. Process Fund., Academy of Sciences, 165 02 Prague 6, Czech Republic
b
c

a r t i c l e

i n f o

Article history:
Received 4 March 2015
Received in revised form 4 May 2015
Accepted 8 May 2015
Available online 16 May 2015
Keywords:
Osmotic pressure
Chemical potential
Molecular simulation
Aqueous electrolytes
NaCl

a b s t r a c t
The osmotic pressure, , is an important thermodynamic property of aqueous electrolyte solutions,
which is intimately related to the activity of the water solvent, and is sensitive to the details of the force
eld used in molecular simulations of such systems. Its calculation in the most important case of discrete water models has received scant attention in the literature; the only existing method involves a
special-purpose molecular dynamics approach implementing virtual semi-permeable membranes separating solution and solvent phases. Here, we develop and demonstrate a new thermodynamically based
approach utilizing simulation results for the salt chemical potential, s , and for the solution specic volume, vm . The methodology may also be used in principle to calculate the activity of water and of the
electrolyte from simulation data for and vm . We demonstrate our approach in the case of aqueous NaCl
solutions at ambient conditions by calculating new results for both and the related osmotic coefcient
property, , from simulation data for NaCl . We compare with experimental data the predictions of two
polarizable force elds (AH/BK3 and AH/SWM4-DP) and of a typical non-polarizable force eld (JC). We
nd that AH/BK3 produces results in good agreement with experiment for both and  over the entire
experimentally accessible concentration range, and that the AH/SWM4-DP results are generally poor.
The JC results are very good at concentrations below about 3 molal, but deteriorate rapidly at higher
concentrations.
2015 Elsevier B.V. All rights reserved.

1. Introduction
Aqueous electrolyte solutions are one of the most important components of environmental, geochemical and biological
systems, and the ability to understand and predict their thermodynamic properties is of considerable interest and importance.
Advances in computational technology have rendered it feasible for molecular-level simulation methodologies based on
molecular dynamics (MD) and Monte Carlo (MC) techniques to
provide useful property predictions and microscopic insight for
many uid systems, including those involving aqueous solutions
[59,5,60,9,36,39,12]. Biocomputing simulations are an important
application area for these methodologies, within which aqueous
electrolytes play a crucial role as the molecular environment for
the protein or other biological molecules of interest. Molecular

Corresponding author at: Dept. of Mathematics and Statistics, University of

Guelph, Guelph, ON N1G 2W1, Canada. Tel.: +1 519 836 7707; fax: +1 519 265 1909.
E-mail address: william.smith@mathtrek.com (W.R. Smith).
http://dx.doi.org/10.1016/j.uid.2015.05.012
0378-3812/ 2015 Elsevier B.V. All rights reserved.

simulations can be used either to predict properties directly, or

to generate pseudo-experimental data that can be used to t the
parameters of macroscopic thermodynamic models.
The requirements for molecular simulation methodology to
provide accurate predictions for a particular uid property for a
given class of systems are: (1) the model should accurately describe
the interactions among the systems molecules; (2) an appropriate
algorithm must be available for the propertys calculation; and (3)
the algorithm must be carefully implemented in computer code.
Concerning requirement (1), information concerning the molecular description is encapsulated in a force eld (FF), a mathematical
model of the molecular interactions implemented within the simulation algorithm. In view of the special nature of water, FFs for its
molecular description have long been of interest and many models
have been proposed, at various levels of molecular-level realism,
ranging from dielectric continuum to non-polarizable and polarizable descriptions [17,61,30,6]. The electrolyte FF used for aqueous
solutions must be tailored to that of water, creating many possibilities. A goal has been to develop a general class of electrolyte
models tailored to the best available water FF, members of which

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

are transferable to broad thermodynamic parameter ranges and to

aqueous environments involving multiple electrolytes.
In general, it is becoming increasingly recognized that FFs based
on the assumption of pairwise additivity are inadequate, both for
pure water and for its solutions. A recent study [44] showed that
electrolyte FFs based on the widely used simple charged LennardJones model with LorentzBerthelot combining rules and tailored
to the commonly used SPC/E water FF [7] are unable to provide
quantitatively accurate predictions of important higher concentration aqueous solution properties. One possibility of incorporating
many-body effects is to add polarizability to the model, which
mimics some of the effects of multi-body molecular interactions.
Many such FFs have been developed for pure H2 O (see, for example the review of Guillot [17]). Two recent promising proposals
of this type are those of Lamoureux and Roux (SWM4-DP) [30]
and of Kiss and Baranyai (BK3) [25]; aqueous alkali halide FFs
tailored to them have also been developed (AH/SWM4-DP [31]
and AH/BK3 [26], respectively). The former mimics polarizability
using a point-charge-on-spring (Drude) model and the latter uses
a Gaussian-charge-on-spring distribution model.
The parameters of aqueous electrolyte FFs have typically been
determined by tting simulation results to approximate high-level
ab initio computational data or to various types of experimental
data, including the innite dilution properties of hydration free
energy, enthalpy and entropy for single ions [22,10,19], chemical
potential derivatives related to experimental values of the solution
compressibility and density at a single concentration in conjunction
with KirkwoodBuff integrals [63,58,16], low-concentration densities [11], and solid salt properties such as lattice constants [22,63].
FFs determined by these means have frequently been unable to
satisfactorily predict more complex and composition-dependent
thermodynamic properties, and thus the composition dependence
of solution chemical potentials [44], and the osmotic pressure,
[35,24,34,56], , have recently been used as alternative approaches
to t FF parameters.
is an important property of aqueous solutions, both in general
and particularly for modeling a range of physiological and physical phenomena (see, e.g., Rsgen et al. [54]). Although it is directly
related to the water chemical potential, H2 O , this connection has
not been computationally exploited in the past in the case of discrete water models, perhaps in part due to challenges in calculating
H2 O for such systems. The chemical potentials of the water solvent and of the electrolyte itself are arguably the most important
thermodynamic properties of aqueous solutions. Their knowledge
as functions of the relevant thermodynamic variables enables the
determination of all other system thermodynamic properties.
Luo et al. [35,34] recently developed a MD simulation method
for directly calculating without the need for chemical potential calculations, which has similarities to earlier work of Murad
et al. [47,48,51,50]. The methodology simulates aqueous solution
and pure water phases separated by virtual membrane walls permeable only to water. is calculated in the simulation from the
force per unit area on the walls. (See the original paper [35] for
details.)
Concerning requirement (2), although many general and welltested MD simulation packages are readily available, they are
not intrinsically well-suited to the calculation of chemical potentials, for which MC methodologies are usually considered to
be more appropriate. The calculation of solubility for aqueous
electrolytes provides a particularly interesting example of the different MD and MC approaches used. All MC methods determine
the solubility by directly calculating the composition at which
the electrolyte solution and crystalline solid chemical potentials
are equal [15,33,55,42,4,49,45,44,38], whereas MD methodologies
determine the solubility indirectly. One MD approach implements
a physical conguration in which the solution phase is equilibrated

77

Table 1
Aqueous solubility simulation results by various research groups for NaCl at
T = 298.15 K and P = 1 bar using the JoungCheatham SPC/E-compatible FF [22]. m is
the NaCl molality in mol kg1 H2 O. A quantity in parentheses denotes the standard
uncertainty in the nal digits.
Method

Solubility(m)

MD with slab geometry [23]

MC and thermodynamic integration [4]
MD with slab geometry [4]
OEMC [44,45]
MD with slab geometry [27]
MD and thermodynamic integration [27]
MD with MC particle insertion [38]
Experiment [18]

7.27(7)
4.8(3)
5.5(4)
3.64(20)
6.20
6.42
3.59(4)
6.14

in the presence of solid phases (for example, a slab conguration with the solution phase located between two crystalline solid
phases), and solubility is determined as the composition of the
electrolyte in the solution phase from a sufciently long simulation run [23,4,13,27]. Another MD-based methodology seeks to
determine the solubility by examining the composition at which
sufciently large ion clusters emerge in the course of the simulations [3,2,37]. Since these different approaches are expected to
have correspondingly different dependencies on simulation protocols such as system size, length of simulation runs and treatment of
long-ranged forces in inhomogeneous systems, it remains unclear
at the present time whether all currently available implementations can achieve the same result.
Concerning requirement (3), a signicant factor is that the relative dearth of general MC software packages may mean that
algorithms to calculate chemical potentials must be developed
locally, raising potential issues with differences in simulation protocols and in the implementation details of the computer code.
The challenges of requirements (2) and (3) are exemplied by
recent solubility studies which, even when using the same water
and electrolyte FFs, have produced different results. This is illustrated in Table 1, which shows results obtained by several research
groups for the salt solubility at ambient conditions in aqueous NaCl
solutions using the non-polarizable JoungCheatham FF tailored
to SPC/E water (JC) [22] using various MD and Monte Carlo (MC)
approaches. As can be seen, only two of the results are in concordance. As discussed elsewhere [46], the NaCl solubility calculation is
intrinsically difcult, since (1) the solution salt chemical potential
tends to exhibit a small concentration dependence at higher concentrations near the solubility limit; and (2) the solution chemical
potentials are extremely sensitive to the details of the simulation
procedure employed. These difculties are reected in the wide
range of calculated solubility values in Table 1.
In a recent study [46] we independently calculated both water
and salt chemical potentials, and veried their mutual consistency
with respect to the GibbsDuhem equation. To our knowledge,
this was the rst such use of this consistency test in conjunction
with simulations, and it provided support for the correctness of
our calculation procedures. This is further supported by the mutual
agreement of our Osmotic Ensemble Monte Carlo (OEMC) results in
Table 1 and the recent calculations of Mester and Panagiotopoulos
[38], who used a different methodology.
As noted, solution chemical potentials are important for many
purposes. In this paper, we demonstrate their use in calculating
and the related osmotic coefcient property, , by developing a
new thermodynamically based method relating these properties to
molecular simulations of either the water or the electrolyte chemical potential. We demonstrate the approach in the case of aqueous
NaCl solutions at ambient conditions and compare with experiment
the predictions arising from the representative non-polarizable

78

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

JoungCheatham FF [22], and the polarizable AH/SWM4-DP [31]

and AH/BK3 [26] FFs.
In the next section of the paper, we discuss the computational methodology employed, followed by a section describing the
molecular interaction models studied and the simulation protocols
used. The subsequent section describes and discusses our results.
The nal section contains our conclusions and recommendations
for future work.

where  = + +  . is thus related to LR under the incompressibility approximation by

(T, P, m) =

 mMH2 O RT

vH2 O (T, P, m)

LR (T, P, m)

(6)

We remark in passing that the MacMillanMayer osmotic

coefcient, MM , useful in theoretical investigations, is related to
by
= (i [T, P + , m])RTMM (T, P, m)

2. Computational methodology

(7)

where i is the molar solution density of the ions,

We rst briey review the relationship of the osmotic pressure
to the water chemical potential and activity, and then describe our
method to calculate the latter quantity from molecular simulation.
In the following, all chemical potential values are expressed
numerically in terms of the same ideal gas reference states for the
chemical potential as those used in commonly available thermochemical tables (e.g., the NIST-JANAF Thermochemical Tables [8,1]).
We initially consider the case of electrolytes with molecular formula A+ B , where A is the cation and B is the anion, and then
specialize to the case + =  = 1 and nally to NaCl.
The molar chemical potential of water in an electrolyte solution,
H2 O (T, P, m), can be written in terms of the Raoult convention on
the molality concentration scale as
H2 O (T, P, m)

= H2 O (T, P, 0) + RT ln aH2 O (T, P, m)

H

2O

(T, P) + RT ln aH2 O (T, P, m)

(1)

where T is the absolute temperature, P is the pressure, m is the

molality of the solution in mol kg1 H2 O, R is the universal gas constant and aH2 O (T, P, m) is the water activity. Superscript * denotes
a property of pure water. For future reference, we note that
m=

ns
nH2 O MH2 O

(2)

where ns and nH2 O are the number of moles of salt and water in the
solution, respectively, and MH2 O is the molecular weight of water
in kg mol1 .
The osmotic pressure, , arises when water in the solution
phase at a pressure P + is in osmotic and thermal equilibrium
with a pure water phase at pressure P. Equating the water chemical
potentials in the two phases gives as the solution of the equation
(see also Robinson and Stokes [52], Simonen [57] and Lee [32]):


RT ln aH2 O (T, P, m) +

P+

vH2 O (T, z, m)dz = 0

(3)

where vH2 O (T, z, m) is the partial molar volume of water at temperature T, pressure z and molality m. Under the (typically excellent)
incompressibility approximation that the water partial molar volume is independent of pressure, this gives the result (Simonen
[57] discusses the general case when this dependence cannot be
neglected):
(T, P, m) =

RT ln aH2 O (T, P, m)

vH2 O (T, P, m)

(4)

Thus, if vH2 O (T, P, m) is known, or aH2 O may be calculated one

from the other without the use of additional approximations; here,
we are interested in calculating from aH2 O .
A related quantity of interest is the osmotic coefcient. The
LewisRandall, practical or molal osmotic coefcient, LR ,
used herein and measured in experiments, is [18,52,32]
LR (T, P, m) =

ln aH2 O (T, P, m)
 mMH2 O

(5)

i =

 ns
=
V (T, P + , m)

nH2 O
V (T, P + , m)

(8)

 mM H2 O

where V is the total solution volume. Eq. (7) is an implicit equation

for , but under the incompressibility approximation LR and MM
are related by (see also Simonen [57] and Lee [32]):
LR (T, P, M) =

vH2 O (T, P, m)MM (T, P, m)

vH2 O (T, P, m) + mM H2 O vNaCl (T, P, m)

(9)

where vNaCl is the partial molar volume of NaCl. vH2 O and vNaCl can
be calculated from simulations of the system specic volume, vm ,
as described below. As noted by Lee [32], LR and MM can take on
quite different values.
The mean molar ionic chemical potential of the salt, s (T, P, m),
is typically expressed experimentally in terms of the Henry convention on the molality concentration scale, which for a 1:1 electrolyte
is given by

s (T, P, m) = s (T, P) + 2RT ln m + 2RT ln  (T, P, m)

(10)

s (T, P)

where
is the mean molar electrolyte standard state chemical potential at unit molality and the (T, P) of the solution and
 (T, P, m) is the electrolyte activity coefcient in the convention
of Eq. (10). ln  can be expressed accurately in the convenient
empirically based analytical form: [18]



A m
3
2
ln  (T, P, m) = ln(10)
(11)
+ bm + cm + dm
1+B m
where the parameters (A, B, b, c, d) depend on T and P and the particular electrolyte. The rst term in Eq. (11) arises from the extended
DebyeHckel theory, and b, c, d are empirical constants. A is related
to the dielectric permittivity of pure water for the FF at the solution (T, P) and B may be treated as an empirical constant. Values
of the constants describing the experimental data for various salt
solutions at ambient conditions are available in the compilation of
Hamer and Wu [18].
aH2 O (T, P, m) in Eq. (1) can be obtained by integrating the GibbsDuhem equation in conjunction with Eqs. (10) and (11), to give
[46]:
ln aH2 O (T, P, m) = 2mMH2 O MH2 O ln(10)

4
3
2A
4
cm3 + dm +

3
2
B3 + B4 m


2A m
2A
4A ln(B m + 1)

+
B3
B2
B3

bm2 +

(12)

We now turn to the calculation of vH2 O in Eq. (4) from molecular

simulations. A convenient and accurate empirical expression for
the specic volume of an aqueous solution of 1:1 salt is [20,14]

vm (T, P, m) = vm (T, P) exp

Ae mMs
1 + mMs

vm (T, P) exp(Ae ws )
(13)

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

79

Table 2
Parameters in Eqs. (11) (15)a tting the NaCl chemical potential and specic volume data in Tables 3 and 4 for aqueous NaCl solutions at ambient conditions (T = 298.15 K,
P = 1 bar) as functions of molality, m (mol kg1 H2 O) for the indicated force elds. Expt denotes values obtained from tting to experimental data, taken from the indicated
sources. A quantity in parentheses denotes the standard uncertainty in the nal digits.
Parameter

AH/SWM4-DP

AH/BK3

JC

Expt

b (kg mol1 )
c (kg2 mol2 )
d (kg3 mol3 )
vm (m3 kg1 )
Ae

0.038951
0.0115923
0.0012310
0.0010015(19)
0.6307(232)

0.188599
0.0290473
0.0017209
0.0009990(6)
0.6586(38)

0.019088
0.0215307
-0.0013356
0.0010019(7)
0.7363(47)

0.020442 [18]
0.0057927 [18]
-0.0002886 [18]
0.001003 [8,1]
0.6938 [14]

For all force elds, A = 0.5108, B = 1.4495 [18] in Eqs. (11) and (12).

where vm (m3 kg1 solution) is the specic volume, Ae is an empirical constant dependent on the electrolyte and (T, P), Ms is the
molecular weight of the salt in kg mol1 and ws is the mass fraction
of the salt in the solution. (We note in passing that Eq. (13) using the
italicized values of Ae and vm at T = 298.15 K and P = 1 bar in Table 2
predicts the experimental vm data [53,29,40] to within 0.03% up to
m = 6.)
The partial molar volumes predicted by Eq. (13) are


vH2 O (T, P, m)

V
nH2 O

vs (T, P, m)

V
ns

uij = uLJ,ij + uPC,ij

where

uPC,ij =
ns ,T,P

Ae mMs
1 + mMs

nH O ,T,P
2

= Ms vm (T, P, m) 1

Ae
1 + mMs

(14)

(15)

We applied the following methodology for a given FF to calculate and LR in the case of aqueous NaCl solutions at ambient
conditions:
1. We calculated several (NaCl , m) points over the experimentally accessible composition range using the OEMC algorithm,
described in detail in our previous papers [33,42,41], which
determines the electrolyte concentration corresponding to a
specied externally imposed chemical potential.
2. We tted the (NaCl , m) simulation data to Eqs. (10) and (11),

treating (NaCl , b, c, d) as tting parameters (see our previous

paper [46] for the details of the methodology.) The experimental
values of A and B were used; the precise values of A and B were
found to have only a small effect on the overall quality of the
tted curve (see a related discussion on p. 1050 of Hamer and
Wu [18].)
3. We calculated several (m, vm ) points over the experimentally
accessible concentration range using NPT MC simulations.
4. We tted the (m, vm ) simulation data to Eq. (13) to determine
the parameters Ae and vm .
5. We calculated from Eqs. (4), (12) and (14), and LR from Eqs.
(5) and (12).
3. Molecular models and simulation details
We briey describe the FFs used in our study; detailed descriptions and parameter values are found in the original papers cited.
The four-site AH/SWM4-DP FF [31] is based on the TIP4P water
geometry [21], with OH bonds and HOH angle set to their gas
phase experimental values, the site charges are qM = 2qH , and
polarizability is modeled by a charge of magnitude qD attached
to the oxygen site (which has a counterbalancing charge of magnitude qD ) by a simple harmonic spring of force constant kD .
A Lennard-Jones (LJ) potential with parameters and models



uLJ,ij = 4

= MH2 O vm (T, P, m) 1 +

the nonpolar interaction between oxygen sites. The complete pair

potential between two water molecules is expressed as [30]

|riO rjO |

(16)

12

|riO rjO |

6 

1 qia qjb
4
0
|ria rjb |

(17)

(18)

a,b

and i, j refer to molecules, a, b refer to molecular interaction sites,

and the summation over a, b runs over all ve charge sites. The
electrolyte is modeled by polarizable LJ particles of charge e interacting with the water solvent model using LorentzBerthelot rules.
The polarizability of the ions is also represented by a charge-onspring model.
The AH/BK3 FF [26] is built on the BK3 water FF [25], which
also assumes the TIP4P water geometry with the gas phase
experimental values of the OH bonds and HOH angle, the nonelectrostatic and electrostatic interactions are modeled differently.
The oxygenoxygen interaction is represented by the EXP6 potential:
uEXP6,ij = A exp(B|riO rjO |) C|riO rjO |6

(19)

and Gaussian charge distributions are centered on the sites:

ia (r) = qia exp 2ia (r ria )2

(20)

where ia is the charge density. The interactions are expressed by

the electrostatic potential energy of two Gaussians,
uGC,ij =



1 qia qjb
|erf iajb |ria rjb |
4
0
|ria rjb

(21)

a,b

where
iajb =

jb

 ia

(22)

2ia 2jb

Similarly as for the SWM4-DP water interaction, polarizability is

represented by a charge-on-spring model, but without introducing
additional charged sites: the positions of all three charges are not
xed, but attached to each site by springs. The total waterwater
interaction energy is similar to that of the SWM4-DP potential.
uij = uEXP6,ij + uGC,ij

(23)

but with different forms of the individual site-site potentials. For

aqueous NaCl, AH/BK3 models each ion by a polarizable EXP6 interaction with Gaussian distributed charges attached to the particle,
and Kong combining rules [28] with respect to the EXP6 part of the
solvent FF.
As a typical example of a non-polarizable FF, we consider the
SPC/E-compatible JoungCheatham (JC) FF that models the ions as

80

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

Table 3
Simulation results for the molality, m (mol kg1 H2 O), of aqueous NaCl at ambient
conditions (T = 298.15 K, P =1 bar) as a function of the molar electrolyte chemical
potential, NaCl (kJ mol1 ) for the indicated force elds. A quantity in parentheses
denotes the standard uncertainty in the nal digits.
AH/SWM4-DP
0.07(7)
0.05(5)

AH/BK3

JC

0.0004(4)
0.10(10)
0.1(1)
0.0003(3)
0.28(20)
0.4(2)
0.0072(72)
0.80(30)
1.02(40)
2.10(40)
2.60(50)

0.83(20)
0.033(33)
1.06(40)
2.09(30)
0.16(10)

AH/SWM4-DP

AH/BK3

JC

Expt [53]

0.0000
0.2056
0.4112
0.8224
1.0279
1.6447
2.0559
2.2615
2.8783
3.4950
4.1118
4.7285
4.9342
5.3453
5.7565
5.9620
6.5789

1.0034(4)

0.9992(70)

0.9872(4)
0.9708(4)

0.9841(90)
0.9696(10)

1.0034(2)
0.9938(2)
0.9857(2)

0.9476(4)
0.9327(5)

0.9416(10)
0.9301(10)

0.9177(5)

0.9102(10)

0.8944(9)

0.8781(10)

0.8731(9)

0.8612(10)

0.8656(10)

0.8474(10)

0.8538(10)

0.8324(10)

1.0030
0.9946
0.9866
0.9713
0.9641
0.9436
0.9309
0.9249
0.9076
0.8917
0.8769
0.8615
0.8588
0.8503
0.8421
0.8382
0.8267

0.9595(2)
0.9383(3)
0.9178(3)
0.9005(3)
0.8847(3)
0.8681(3)
0.8546(3)
0.8403(3)
0.8308(3)

3.0(4)
0.46(15)
5.9(3)
1.06(15)
2.08(15)
3.08(15)
3.77(15)
4.72(20)
5.52(20)

charged Lennard-Jones spheres and LorentzBerthelot combining

rules [22].
The simulation protocols were as follows. Each simulation
employed a cubic simulation box of 270 water particles with periodic boundary conditions, and the number of ion pairs depended
on the concentration. We used the same value of the cut-off
for the LJ, EXP6 and Coulombic interactions, and the
radius, rc = 9 A,
long-ranged Coulombic interactions were treated using the Ewald
summation method [6]. The simulations were initialized by placing
the particles on a face-centered cubic lattice, and rotations of the
molecules were generated randomly. For all polarizable FF simulations, we generated congurations using the Multi-Particle-Move
Monte Carlo (MPM-MC) method [43].

the AH/SWM4-DP simulations exhibited premature (below the

experimental solubility limit) NaCl precipitation, and we have thus
terminated this curve at m = 3.
The ln aH2 O , results, calculated from Eq. (12) and the parameters in Table 2, are shown in Fig. 2. As in Fig. 1, the AH/SWM4-DP
curve is terminated at m = 3. Also shown for comparison for the JC
FF are simulation data for ln aH2 O from Mester and Panagiotopoulos [38], which are seen to be in mutual agreement with the JC
curve of this study. Fig. 2 shows that the AH/BK3 results are superior overall in their agreement with experiment, although it is

-380

NaCl / kJ mol-1

NaCl
434.000
424.000
414.024
414.000
410.000
409.024
409.000
405.000
404.024
404.000
402.000
401.000
400.000
399.024
398.000
395.000
394.024
393.000
389.024
388.000
384.024
379.024
376.024
374.024
371.024
369.024

Table 4
Simulation results for the specic volume, vm (105 m3 kg1 solution), of aqueous
NaCl as a function of the molality, m (mol kg1 ), at ambient conditions (T = 298.15 K,
P = 1 bar) for the indicated force elds and from experiment. A quantity in parentheses denotes the standard uncertainty in the nal digits.

-390

-400

4. Results and discussion

We considered aqueous NaCl solutions at ambient conditions
(T = 298.15 K, P = 1 bar) and carried out the steps indicated in the
Computational Methodology section. The raw simulation data are
shown in Tables 3 and 4 and the parameter values obtained from
tting the electrolyte chemical potential and solution specic volume simulation data to the relevant equations are given in Table 2.
Table 2 also shows parameter values corresponding to the experimental results.
The NaCl results are shown in Fig. 1 for each FF, where they are
compared with the experimental data [18]. The larger uncertainty
bars on the AH/SWM4-DP and AH/BK3 curves are due to the much
more time-consuming nature of the calculation in comparison with
the case of non-polarizable force elds, resulting in reduced precision for a given computation time. Also shown for comparison for
the JC FF are simulation data points of Mester and Panagiotopoulos
[38], obtained using a gradual particle insertion method in conjunction with MD simulations; these results are seen to be in mutual
agreement with our results.
The JC FF results are in closest agreement with the experimental curve up to about m = 3, and thereafter the AH/BK3 FF is
superior. We found that near m = 3 and at higher concentrations

-410

m / mol kg-1
Fig. 1. Simulation results for the electrolyte mean molar chemical potential of an
aqueous NaCl solution at T = 298.15 K, P = 1 bar as a function of concentration using
different force elds and from experiment. Blue circles denote simulation data points
for the (non-polarizable) JC FF [22] and dark green triangles for the (polarizable)
AH/SWM4-DP FF [31], taken from our previous papers [46,43]. Red circles denote
simulation results of this paper for the (polarizable) AH/BK3 FF [26]. Shown for
comparison purposes are cyan lled squares denoting the JC simulation results of
Mester and Panagiotopoulos [38]. The solid black curve indicates the experimental
results [18], calculated using Eqs. (10) and (11) with parameters from Table 2 and
 = 393.133 [62], and the black horizontal dotted line indicates the experimental
solid chemical potential value [8], shown for reference purposes. The colored curves
denote the tted NaCl results for each FF from Eqs. (10) and (11) using the relevant
parameters in Table 2. The blue long-dash curve denotes the JC result, the dark green
dash-dot curve the AH/SWM4-DP result and the red short-dash curve the AH/BK3
result. The uncertainty bars indicate one standard deviation where shown; otherwise, they lie within the symbols. The uncertainties of the curves are approximately
equal to the uncertainty bars for the simulation data points. (For interpretation of
the references to color in this gure legend, the reader is referred to the web version
of this article.)

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

81

0.0
1.805

-1

1.800

10 vH2O / m mol

-0.2

ln (aH2O )

-0.1

-0.3

1.795

1.790

1.785

-0.4

1.780

m / mol kg

Fig. 2. The water activity dened by Eq. (1) for an aqueous NaCl solution at
T = 298.15 K, P = 1 bar as a function of concentration using the indicated force elds
and from experiment. The curves are calculated from Eq. (12) using the parameters
in Table 2. See the caption to Fig. 1 for denitions of the notations for the curve and
point symbols.

somewhat problematic to discern differences between the curves

at low concentrations, in view of the overlapping uncertainties.
The vm simulation results and the curves of Eq. (13) using the
parameters in Table 2 are shown in Fig. 3, in terms of their per cent
deviations from the experimental values. The quantity vm is
vm (m) = vm (predicted) vm (expt)

(24)

where predicted denotes the simulation data point or the result

of Eq. (13) for the FF in question, and expt denotes the experimental value predicted by Eq. (13) using the italicized parameters
in Table 2. AH/SWM4-DP simulation points at m 3 are shown in
the gure, but we did not use them in our tting procedure, for
the reasons discussed in the context of Fig. 1. Below m = 3, the
AH/SWM4-DP results lie up to 1% above the experimental curve.
Fig. 3 indicates that the AH/BK3 FF gives the best overall agreement
with experiment, predicting results to within better than about
0.6% of the experimental values. The JC curve shows an increasingly
negative deviation from the experimental results with increasing

m / mol kg-1

-1

Fig. 4. The partial molar volume of H2 O, vH2 O , for an aqueous NaCl solution at
T = 298.15 K, P = 1 bar as a function of concentration, using different force elds and
from experiment. The curves are calculated from Eq. (14). See the caption to Fig. 1
for denitions of the curve notations.

concentration, although they still agree with experiment to within

about 1.3%.
vH2 O is shown in Fig. 4, as calculated from Eq. (14) using the
parameters in Table 2. As in Figs. 1 3, the AH/SWM4-DP curve is
terminated at m = 3. Differences among the FF results arise primarily from their different values of vm in Table 2, which give rise to
different intercept values of the curves in the gure, which from
Eq. (14) are given by MH2 O vm . Each FF curve deviates from its corresponding pure uid and from the experimental result by less than
2%, and thus vH2 O can be replaced by MH2 O vm in Eqs. (4) and (6) to
an excellent approximation.
The results for the osmotic pressure, , calculated for each FF
from Eq. (4) are shown in Fig. 5, using ln aH2 O from Eq. (12), vH2 O
from Eq. (14) and the parameter values in Table 2. As in Figs. 14, we
have terminated the AH/SWM4-DP curves at m = 3. Also shown for
comparison for the JC FF are data points calculated by combining
the simulation results for ln aH2 O of Mester and Panagiotopoulos
[38] with our vH2 O results from Eq. (14). These data points are
seen to be in agreement with the JC curve within their mutual
600

3.5
3.0

500

400

2.0

/ bar

100 vm/vm(expt)

2.5

1.5
1.0
0.5

300

200

0.0

100

-0.5
-1.0

0
0

-1.5
0

m / mol kg-1
Fig. 3. Deviations of the simulation results from the experimental values for the
specic volume of an aqueous NaCl solution at T = 298.15 K, P = 1 bar as a function of
concentration using different force elds. vm = vm (sim) vm (expt). The points are
individual simulation data points and the curves denote results obtained from Eq.
(13) tted to the simulation data. The open AH/SWM4-DP points are likely 2-phase
states and were not used in the data tting (see the text). See the caption to Fig. 1
for denitions of the notations for the point and curve symbols.

m / mol kg-1
Fig. 5. The osmotic pressure, , dened by Eq. (4), for an aqueous NaCl solution
at T = 298.15 K, P = 1 bar as a function of concentration using the indicated force
elds and from experiment. Shown for comparison purposes are lled cyan squares
calculated by combining in Eq. (4) the ln aH2 O simulation data of Mester and Panagiotopoulos [38] and our simulation data for vH2 O , as described in the text. See the
caption to Fig. 1 for denitions of the curve and other symbol notations. (For interpretation of the references to color in this gure legend, the reader is referred to the
web version of this article.)

82

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

3.5
3.0
2.5

LR

2.0
1.5
1.0
0.5
0.0
0

m / mol kg-1
Fig. 6. The practical molal osmotic coefcient, LR , dened by Eq. (5) for an aqueous
NaCl solution at T = 298.15 K, P = 1 bar as a function of concentration using the indicated force elds and from experiment. Shown for comparison purposes are lled
cyan squares calculated from Eq. (5) using the data of Mester and Panagiotopoulos
[38]. See the caption to Fig. 1 for denitions of the curve and other symbol notations. (For interpretation of the references to color in this gure legend, the reader
is referred to the web version of this article.)

uncertainties. Since Eq. (4) shows that is simply ln aH2 O divided

by the water partial molar volume, and the latter quantity is reproduced well by all the FFs, the relative agreements with respect to
experiment of the different FF results in Figs. 2 and 5 follow the
same pattern, with the AH/BK3 values being superior overall, and
the JC showing comparable agreement for m 3. We also remark
that for m 3, Fig. 5 shows good agreement with the AH/SWM4DP FF results of Luo et al. [35,34], who studied the renement of
this FFs parameters to improve its predictions of the experimental
values of .
The results for the osmotic coefcient, LR for each FF, as calculated from Eq. (5) and the parameters in Table 2, are shown in Fig. 6.
Also shown for comparison for the JC FF are data points calculated
from the values of ln aH2 O in Table 1 of Mester and Panagiotopoulos [38]. As in the cases of Figs. 1, 2 and 5, these data points are
consistent with the JC curve calculated here. The simulation point
uncertainty bars and curve uncertainties in Fig. 6 are much larger
at low concentrations for LR than those for shown in Fig. 5, in
view of the factor m in the denominator of Eq. (5). As in Figs. 15,
the AH/SWM4-DP curve has been terminated at m = 3; its results
for m < 3 are poor except at the lowest concentrations. The AH/BK3
FF produces the best agreement with experiment for LR , and the JC
results show comparable agreement with experiment up to about
m = 3.
Finally, we note that, although the approach of this paper
demonstrates the calculation of via Eq. (4) from simulation
results for s and for vm , it can also be used in principle for the calculation of ln aH2 O and the salt activity from simulation results for
and for vm . For example, the methodology of Luo et al. for determining [35], augmented by calculations of vH2 O from simulations
of vm as described in this paper, can be employed for this purpose.
This would be the molecular simulation counterpart to methods
for determining ln aH2 O developed by experimentalists early in the
last century [52].
5. Conclusions
We have developed a new thermodynamically based molecular
simulation methodology to calculate the concentration dependence of the osmotic pressure, , and the osmotic coefcients, LR
and MM , for aqueous electrolyte solutions, and have applied the

approach to aqueous NaCl solutions at ambient conditions. This

employs simulation results for the concentration dependence of
the electrolyte chemical potential, s and the solution specic volume, vm . The latter are used to calculate the water partial molar
volume, vH2 O .
We have calculated results predicted by the AH/SWM4-DP
polarizable force eld of Lamoureux and Roux [31], the AH/BK3
polarizable force eld of Kiss and Baranyai [25] and the commonly
used non-polarizable force eld of Joung and Cheatham [22], and
compared the results of all three force elds with the experimental data. The AH/BK3 force eld produces the best results, showing
good agreement of for vH2 O , and LR with the experimental data
over the entire concentration range. The JoungCheatham force
eld is competitive up to about m = 3. The SWM4-DP force eld
is unsatisfactory at concentrations above m = 3, since it shows premature salt precipitation for m 3.
Our methodology can also be used in principle to calculate the
water and salt activities from simulation data for and vm .
6. Acknowledgements
Support for this work was provided by the Natural Sciences
and Engineering Research Council of Canada (Discovery Grant
OGP1041), the SHARCNET (Shared Hierarchical Academic Research
Computing Network) HPC consortium (http://www.sharcnet.ca),
the Czech National Science Foundation (Grant No. GP13-35793P),
and the Czech Ministry of Education, Youth and Sport (Project
No. LH12019). Access to computing and storage facilities owned
by parties and projects contributing to the National Grid Infrastructure MetaCentrum, provided under the program Projects of
Large Infrastructure for Research, Development, and Innovations
(LM2010005), is greatly appreciated.
References
[1] NIST Chemistry Webbook, 2011 http://webbook.nist.gov [accessed 01.02.15].
[2] J. Alejandre, G.A. Chapela, F. Bresme, J.-P. Hansen, The short range anion-H
interaction is the driving force for crystal formation of ions in water, J. Chem.
Phys. 130 (2009) 174505.
[3] J. Alejandre, J.-P. Hansen, Ions in water: from ion clustering to crystal
nucleation, Phys. Rev. E 76 (2007) 061505.
[4] J.L. Aragones, E. Sanz, C. Vega, Solubility of NaCl in water by molecular
simulation revisited, J. Chem. Phys. 136 (2012) 244508.
[5] P. Aufnger, T.E. Cheatham III, A.C. Vaiana, Spontaneous formation of KCl
aggregates in biomolecular simulations: a force eld issue? 2007, pp.
18511858.
[6] A. Baranyai, P.T. Kiss, A transferable classical potential for the water molecule
J. Chem. Phys. 133 (2010) 144109.
[7] H.J.C. Berendsen, J.R. Grigera, T.P. Straatsma, The missing term in effective pair
potentials, J. Phys. Chem. 91 (1987) 62696271.
[8] M.W. Chase Jr., NIST-JANAF Thermochemical Tables, J. Phys. Chem. Ref. Data,
Monograph No. 9, Am. Chem. Soc. and Am. Inst. of Physics, Woodbury, New
York, 1998.
[9] A.A. Chen, R.V. Pappu, Parameters of monovalent ions in the AMBER-99
forceeld: assessment of inaccuracies and proposed improvements, J. Phys.
Chem. B 111 (2007) 1188411887.
[10] L.X. Dang, J. Am. Chem. Soc. 117 (1995) 69546960.
[11] S. Deublein, J. Vrabec, H. Hasse, A set of molecular models for alkali and halide
ions in aqueous solution, J. Chem. Phys. 136 (2012) 084501.
[12] R.O. Dror, R.M. Dirks, J.P. Grossman, H. Xu, D.E. Shaw, Biomolecular
simulation: a computational microscope for molecular biology, Annu. Rev.
Biophys. 41 (2012) 429452.
[13] J.R. Espinosa, E. Sanz, C. Valeriani, C. Vega, On uidsolid direct coexistence
simulations: the pseudo-hard sphere model, J. Chem. Phys. 139 144502.
[14] W.R. Fawcett, Liquids, Solutions, and Interfaces, Oxford, New York, 2004.
[15] M. Ferrario, Solubility of KF in water by molecular dynamics using the
Kirkwood integration method, J. Chem. Phys. 117 (2002) 49474956.
[16] M.B. Gee, N.R. Cox, Y. Jiao, N. Bentenitis, S. Weerasinghe, P.E. Smith, A
KirkwoodBuff derived force eld for aqueous alkali halides, J. Chem. Theory
Comput. 7 (2011) 13691380.
[17] B. Guillot, A reappraisal of what we have learnt during three decades of
computer simulations on water, J. Mol. Liq. 101 (2002) 219260.
[18] W.J. Hamer, Y.-C. Wu, Osmotic coefcients and mean activity coefcients of
un0-univalent electrolytes in water at 25 C, J. Phys. Chem. Ref. Data 1 (1972)
10471099.

W.R. Smith et al. / Fluid Phase Equilibria 407 (2016) 7683

[19] D. Horinek, S.I. Mamatkulov, R.R. Netz, Rational design of ion force elds based
on thermodynamic solvation properties, J. Chem. Phys. 130 (2009) 124507.
[20] A.L. Horvath, Handbook of Electrolyte Electrolyte Solutions, Ellis Horwood,
Chichester, 1985.
[21] W.L. Jorgensen, J. Chandrasekhar, J.D. Madura, R.W. Impey, M.L. Klein,
Comparison of simple potential functions for simulating liquid water, J. Chem.
Phys. 79 (1983) 926935.
[22] I.S. Joung, III. Cheatham, E. Thomas, Determination of alkali and halide
monovalent ion parameters for use in explicitly solvated biomolecular
simulations, J. Phys. Chem. B 112 (2008) 90209041.
[23] I.S. Joung, III. Cheatham, E. Thomas, Molecular dynamics simulations of the
dynamic and energetic properties of alkali and halide ions using
water-model-specic ion parameters, J. Phys. Chem. B 113 (2009)
1327913290.
[24] Y.V. Kalyuzhnyi, V. Vlachy, K.A. Dill, Aqueous alkali halide solutions: can
osmotic coefcients be explained on the basis of ionic sizes alone? Phys,
Chem. Chem. Phys. 12 (2010) 62606266.
[25] P.T. Kiss, A. Baranyai, A systematic development of a polarizable potential of
water, J. Chem. Phys. 138 (2013) 204507.
[26] P.T. Kiss, A. Baranyai, A new polarizable force eld for alkali and halide ions, J.
Chem. Phys. 141 (2014) 114501.
[27] K. Kobayashi, Y. Liang, T. Sakka, T. Matsuoka, Molecular dynamics study of
salt-solution interface: solubility and surface charge of salt in water, J. Chem.
Phys. 140 (2014) 144705.
[28] C.L. Kong, M.R. Chakrabarty, Combining rules for intermolecular potential
parameters. iii. Application to the exp 6 potential, J. Phys. Chem. 77 (1973)
26682670.
[29] B.S. Krumgalz, R. Pogorelsky, Y.A. Iosilevski, A. Weiser, K.S. Pitzer, Ion
interaction approach for volumetric calculations for solutions of single
electrolytes at 25 C, J. Sol. Chem. 23 (1994) 849875.
[30] G. Lamoureux, A.D. MacKerell, B. Roux, A simple polarizable model of water
based on classical Drude oscillators, J. Chem. Phys. 119 (2003) 51855197.
[31] G. Lamoureux, B. Roux, Absolute hydration free energy scale for alkali and
halide ions established from simulations with a polarizable force eld, J. Phys.
Chem. B 110 (2006) 33083322.
[32] L.L. Lee, Molecular Thermodynamics of Electrolyte Solutions, World Scientic,
Singapore, 2008.
[33] M. Lsal, W.R. Smith, J. Kolafa, Molecular simulations of aqueous electrolyte
solubility: 1. The expanded-ensemble osmotic molecular dynamics method
for the solution phase, J. Phys. Chem. B 109 (2005) 1295612965.
[34] Y. Luo, W. Jiang, H. Yu, A.D. MacKerell Jr., B. Roux, Simulation study of ion
pairing in concentrated aqueous salt solutions with a polarizable force eld,
Faraday Discuss. 160 (2013) 135149.
[35] Y. Luo, B. Roux, Simulation of osmotic pressure in concentrated aqueous salt
solutions, J. Phys. Chem. Letters 1 (2010) 183189.
[36] E.J. Maginn, J.R. Elliott, Historical perspective and current outlook for
molecular dynamics as a chemical engineering tool, Ind. Eng. Chem. Res. 49
(2010) 30593078.
[37] F.F. Mendoza, J. Alejandre, The role of ion-water interactions in the solubility
of ionic solutions, J. Mol. Liq. 185 (2013) 5055.
[38] Z. Mester, A.Z. Panagiotopoulos, Mean ionic activity coefcients in aqueous
NaCl solutions from molecular dynamics simulations, J. Chem. Phys. 142
(2015) 044507.
[39] M. Meunier, Industrial Applications of Molecular Simulations, CRC Press, Boca
Raton, 2011.
[40] M.V. Mironenko, G.E. Boitnott, S.A. Grant, Experimental determination of the
volumetric properties of NaCl solutions to 253 K, J. Phys. Chem. B Lett. 105
(2001) 99099912.
[41] F. Moucka, M. Lsal, W.R. Smith, Molecular simulation of aqueous electrolyte
solubility. 3. Alkali-halide salts and their mixtures in water and in
hydrochloric acid, J. Phys. Chem. B 116 (2012) 54685478.

83

[42] F. Moucka, M. Lsal, J. Skvor,

J. Jirsk, I. Nezbeda, W.R. Smith, Molecular
simulation of aqueous electrolyte solubility. 2. Osmotic ensemble Monte
Carlo methodology for free energy and solubility calculations and application
to NaCl, J. Phys. Chem. B 115 (2011) 78497861.
[43] F. Moucka, I. Nezbeda, W.R. Smith, Computationally efcient Monte Carlo
simulations for polarisable models: multi-particle move method for water
and aqueous electrolytes, Mol. Simul. 39 (2013) 11251134.
[44] F. Moucka, I. Nezbeda, W.R. Smith, Molecular force eld development for
aqueous electrolytes: 1. Incorporating appropriate experimental data and the
inadequacy of simple electrolyte force elds based on Lennard-Jones and
point charge interactions with LorentzBerthelot rules, J. Chem. Theory
Comput. 9 (2013) 50765085.
[45] F. Moucka, I. Nezbeda, W.R. Smith, Molecular force elds for aqueous
electrolytes: SPC/E-compatible charged LJ sphere models and their
limitations, J. Chem. Phys. 138 (2013) 154102.
[46] F. Moucka, I. Nezbeda, W.R. Smith, Molecular simulation of aqueous
electrolytes: water chemical potential results and GibbsDuhem equation
consistency tests, J. Chem. Phys. 139 (2013) 124505.
[47] S. Murad, J.G. Powles, A computer simulation of the classic experiment on
osmosis and osmotic pressure, J. Chem. Phys. 99 (1993) 72717272.
[48] S. Murad, J.G. Powles, B. Holtz, Osmosis and reverse osmosis in solutions:
Monte Carlo simulations and van der Waals one-uid theory, Mol. Phys. 86
(1995) 14731483.
[49] A.S. Paluch, S. Jayaraman, J.K. Shah, E.J. Maginn, Erratum: A method for
computing the solubility limit of solids: application to sodium chloride in
water and alcohols [J. Chem. Phys. 133 (2010) 124504], J. Chem. Phys. 137
(2012) 039901.
[50] F. Paritosh, S. Murad, Molecular simulations of osmosis and reverse osmosis
in aqueous electrolyte solutions, Am. Inst. Chem. Eng. J. 42 (1996) 29842986.
[51] J.G. Powles, S. Murad, B. Holtz, A novel osmotic pressure route to the activity
coefcient of a molecule in a solution, Chem. Phys. Lett. 245 (1995) 178182.
[52] R.A. Robinson, R.H. Stokes, Electrolyte Solutions: Second Revised Edition,
Dover Publications, New York, 2002.
[53] P.S.Z. Rogers, K.S.J. Pitzer, Volumetric properties of aqueous sodium chloride
solutions, J. Phys. Chem. Ref. Data 11 (1982) 1581.
[54] J. Rsgen, B.M. Pettitt, D.W. Bolen, An analysis of the molecular origin of
osmolyte-dependent protein stability, Protein Sci. 16 (2007) 733743.
[55] E. Sanz, C. Vega, Solubility of KF and NaCl in water by molecular simulation, J.
Chem. Phys. 126 (2007) 014507.
[56] A. Saxena, A.E. Garcia, Multisite ion model in concentrated solutions of
divalent cations (MgCl2 and CaCl2 ): osmotic pressure calculations, J. Phys.
Chem. B 119 (2015) 219227.
[57] J.-P. Simonen, Study of experimental-to-McMillanMayer conversion of
thermodynamic excess functions, J. Chem. Soc. Faraday Trans. 92 (19) (1996)
35193523.
[58] P.E. Smith, The effect of urea on the morphology of NaCl crystals: a combined
theoretical and simulation study, Fluid Phase Equilib. 290 (2010) 3642.
[59] D.N. Theodoru, Progress and outlook in Monte Carlo simulations, Ind. Eng.
Chem. Res 49 (2010) 30473058.
[60] P. Ungerer, C. Nieto-Draghi, B. Rousseau, G. Ahunbay, V. Lachet, Molecular
simulation of the thermophysical properties of uids: From understanding
toward quantitative predictions, Mol. Simul. 134 (2010) 7189.
[61] C. Vega, J.L.F. Abascal, Simulating water with rigid non-polarizable models: a
general perspective, Phys. Chem. Chem. Phys. 13 (2011) 1966319688.
[62] D.D. Wagman, W.H. Evans, V.B. Parker, R.H. Schumm, I. Halow, S.M. Bailey,
K.L. Churney, R.L. Nuttall, The NBS tables of chemical thermodynamic
properties: selected values for inorganic and C1 and C2 organic substances in
SI units, J. Phys. Chem. Ref. Data (Suppl. 2) (1982).
[63] S. Weerasinghe, P.E. Smith, A Kirkwood-Buff derived force eld for sodium
chloride in water, J. Chem. Phys. 119 (2003) 1134211349.