Solvent activity in electrolyte solutions from molecular simulation of the osmotic

pressure
Maximilian Kohns, Steffen Reiser, Martin Horsch, and Hans Hasse

Citation: The Journal of Chemical Physics 144, 084112 (2016); doi: 10.1063/1.4942500
View online: http://dx.doi.org/10.1063/1.4942500
View Table of Contents: http://aip.scitation.org/toc/jcp/144/8
Published by the American Institute of Physics
 THE JOURNAL OF CHEMICAL PHYSICS 144, 084112 (2016)

Solvent activity in electrolyte solutions from molecular simulation

of the osmotic pressure
Maximilian Kohns, Steffen Reiser, Martin Horsch,a) and Hans Hasse
Laboratory of Engineering Thermodynamics, University of Kaiserslautern, Erwin-Schrödinger Str. 44,
D-67663 Kaiserslautern, Germany
(Received 22 October 2015; accepted 8 February 2016; published online 25 February 2016)

A method for determining the activity of the solvent in electrolyte solutions by molecular dynamics
simulations is presented. The electrolyte solution is simulated in contact with the pure solvent.
Between the two phases, there is a virtual membrane, which is permeable only for the solvent.
In the simulation, this is realized by an external field which acts only on the solutes and confines
them to a part of the simulation volume. The osmotic pressure, i.e., the pressure difference between
both phases, is obtained with high accuracy from the force on the membrane, so that reliable data
on the solvent activity can be determined. The acronym of the new method is therefore OPAS
(osmotic pressure for activity of solvents). The OPAS method is verified using tests of varying
complexity. This includes a comparison of results from the OPAS method for aqueous NaCl solu-
tions to results from the literature which were obtained with other molecular simulation methods.
Favorable agreement is observed not only for the solvent activity but also for the activity coefficient
of NaCl, which is obtained by application of the Gibbs-Duhem equation. C 2016 AIP Publishing
LLC. [http://dx.doi.org/10.1063/1.4942500]

I. INTRODUCTION method for the computation of solubility limits in electrolyte

solutions. That method employs a semi-grand ensemble to
Electrolyte solutions are important in many industrial
compute the salt molality m at a given chemical potential of the
and natural processes. The thermodynamic behavior of
salt µ± in a Monte Carlo (MC) simulation. In a recent paper,
electrolyte solutions is dominated by the strong Coulombic
Mester and Panagiotopoulos10 conducted molecular dynamics
interactions between the ions as well as between ions and
(MD) simulations in the isothermal-isobaric (NpT) ensemble
solvent molecules, which results in a strong structuring
to determine the chemical potential of the salt µ± at a given
of the solution. The consequences of this on macroscopic
salt molality m by performing fractional insertion of ion pairs.
properties of the solutions are difficult to describe, so that
Using the Gibbs-Duhem equation, the solvent activity asolv can
phenomenological models of electrolyte solutions regularly
be derived from the results of either of these two methods. Both
employ a large number of adjustable parameters and are
methods have in common that advanced sampling methods
useful mainly for correlating experimental data.1–4 Another
were used to improve the statistics, as particle insertion moves
shortcoming of these models is that their parameters are
for ions are very unlikely to be accepted.6
not transferable when different solvents are used. Molecular
In previous work of our group,11,12 a set of molecular
simulations provide an opportunity to gain insight into the
models for alkali and halide ions in conjunction with the
inner workings of electrolyte solutions on the molecular level.
SPC/E13 water model has been developed. Simulation results
In contrast to phenomenological models, molecular simulation
using these models have shown good agreement with available
usually takes into account only binary interactions between
experimental data for various properties such as liquid
species. Additionally, properties of mixtures can be predicted
solution densities, radial distribution functions, diffusion
if accurate molecular models for all components are available.
coefficients, and electric conductivities.14 Furthermore, the
One of the most important properties of an electrolyte
model parameters are transferable so that methanolic and
solution is the activity of the solvent. It is directly related
ethanolic electrolyte solutions can also be described.15,16 The
to phenomena such as vapor pressure lowering and freezing
most important property that has not yet been investigated
point depression, and it is connected via the Gibbs-Duhem
with these models is the solvent activity.
equation to the activity coefficient of the salt, which controls
In the present work, a method is introduced for
the solubility limit of the salt. However, the reliable determi-
molecular dynamics simulation of the osmotic pressures
nation of the solvent activity and related properties of elec-
for determining the activity of solvents (OPAS). The OPAS
trolyte solutions by molecular simulations is challenging and
method circumvents the problems regarding particle insertion.
subject to recent research. In a series of papers, Smith and
OPAS simulations exploit the relationship between the solvent
co-workers5–9 developed the Osmotic Ensemble Monte Carlo
activity and the osmotic pressure by performing simulations of
a)Author to whom correspondence should be addressed. Electronic
coexisting phases in osmotic equilibrium, i.e., a pure solvent
mail: martin.horsch@mv.uni-kl.de. Telephone: +49-631/205-3227. Fax: phase and an electrolyte solution phase in direct contact. The
+49-631/205-3835. phases are separated by a semipermeable membrane, which is

0021-9606/2016/144(8)/084112/10/$30.00 144, 084112-1 © 2016 AIP Publishing LLC

084112-2 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

only permeable for the solvent but not for the solutes (here:
the ions). The idea of such direct coexistence simulations goes
back to a series of papers by Murad and co-workers,17–20 who
used Lennard-Jones (LJ) particles to which they applied a
position constraint to simulate a semipermeable membrane.
The present OPAS simulation method follows an approach
which is similar to that of Luo and Roux,21 who use a
virtual semipermeable membrane. The OPAS method adopts
the general simulation scenario of Luo and Roux and extends
their work regarding the simulation methodology and the
evaluation of the results.
In Section II, the OPAS method is presented. In
Section III, the tests are described. Their results are presented
and discussed in Section IV, before we conclude in Section V.

II. OPAS SIMULATION METHOD

The OPAS molecular dynamics simulation method
directly computes the osmotic equilibrium in mixtures using
virtual semipermeable membranes. The method is developed
for studying electrolyte solutions, but can, in principle, also FIG. 1. Snapshot from an OPAS simulation of an aqueous sodium chloride
be used for studying mixtures of molecular species. As it is solution. The vertical lines indicate the positions of the two semipermeable
membranes. Na+ cations and Cl− anions are shown as yellow and green
described here, the OPAS method is applicable to mixtures of spheres, respectively. The outer compartment of the simulation box is not
one solvent with an arbitrary number of solutes. An extension accessible to the ions. The oxygen atom of water is shown in blue and the
to mixed solvents is possible but would require adaptions. hydrogen atoms of water are shown in white. The size of the water molecules
To simulate the osmotic equilibrium, two planar, has been reduced for better visualization.
semipermeable membranes M1 and M2 are introduced in
a cubic simulation box. For the simulations of the present resulting pressure pout is hard to guess. Therefore, an OPAS
work, they are located at the positions zM1 = 0.25 L and simulation consists of a two-step procedure.
zM2 = 0.75 L, where L is the box length, and lie parallel to First, a pseudo-isothermal-isobaric (NpT) ensemble
the x, y-plane in the coordinate system we choose here. simulation is carried out for determining the box volume
The membranes divide the simulation volume into two ⟨V ⟩ of the system in osmotic equilibrium. To avoid the
compartments. The outer compartment (here: 0 ≤ z/L < 0.25 necessity of a specially designed barostat for that purpose, the
and 0.75 < z/L ≤ 1) contains only solvent molecules, while entire system is coupled to a barostat that yields the effective
the inner compartment (here: 0.25 ≤ z/L ≤ 0.75) contains pressure
both solvent and solute molecules. The conventional periodic
peff = (poutVout + pinVin)/(Vout + Vin). (4)
boundary conditions are used. A snapshot of this simulation
setup is shown in Figure 1. The effective pressure peff corresponds to an average over
If a solute molecule i leaves the inner compartment at the virial of the entire system. The pressure in the outer
either M1 or M2, a force compartment pout is fixed to pset and pin can be estimated
on the fly from pin = pout + Π, where Π is obtained by
Fi = c(zM1 − z i ) for z i < zM1, (1) Eq. (3). As a result, the desired pressure pset is obtained
Fi = c(zM2 − z i ) for z i > zM2 (2) in the outer compartment and a number for the box volume
⟨V ⟩ is obtained. Furthermore, this pseudo-NpT run serves as
is exerted on it in the z direction, where z i is the z-component a pre-equilibration of the system. Note that for the specific
of the center of mass position of ion i. In the present study, the choice of compartment volumes Vout = Vin in the present work,
value of the force constant was taken from Luo and Roux21 and Eq. (4) simplifies to
is c = 4.184 × 1024 J/m2. The osmotic pressure Π is given by
the pressure difference pin − pout between both compartments Π
peff = pout +. (5)
and is obtained from the membrane force as 2
 The volume ⟨V ⟩ is then used in a canonical (NVT) ensemble
Ftot i |Fi |
Π = pin − pout = = , (3) simulation, which yields the osmotic pressure ⟨Π⟩ = ⟨Ftot⟩/Atot
Atot AM1 + AM2 of the system. The numbers found in that simulation for pout
where Ftot is the total net force exerted from the ions on both are usually very close to the desired number. Note that setting
membranes and Atot is the total area of both membranes. pout is only a matter of convenience and not directly related
It is desirable to simulate the osmotic equilibrium at a to the determination of the osmotic pressure, i.e., the pseudo-
specified pressure pout = pset in the outer compartment, i.e., in NpT run could, in principle, be omitted or replaced by other
the pure solvent phase (typically pset = 1 bar). A direct run schemes. The osmotic pressure obtained in the NVT run is
in the canonical (NVT) ensemble is not convenient as the directly related to the solvent activity asolv via
084112-3 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

Π All simulations of this study were carried out with an

ln(asolv) = − , (6)
ρsolv RT extended version of the molecular simulation program ms2.22
More simulation details are given in Appendix C.
where R is the universal gas constant, and ρsolv is the molar
density of the pure liquid solvent at the temperature T, which
is assumed to be independent of pressure. This property is III. TESTS OF THE OPAS METHOD
available from separate molecular simulations. A derivation
of Eq. (6) is given in Appendix A. A. Approach
During the simulation, the molality of the electrolyte in Tests of varying complexity were carried out in order
the inner compartment fluctuates as solvent molecules are free to verify the OPAS simulation method. In all cases, profiles
to move and hence, the composition in the inner compartment of properties as functions of the coordinate z (normal to the
is a simulation output. Therefore, a density profile for the membranes) were determined: component densities, pressure,
solvent ρsolv(z) is recorded during the NVT run by dividing the and, where applicable, chemical potential of the solvent.
simulation volume into equally sized bins which lie parallel All profiles were determined by dividing the simulation
to the membranes. In order to exclude artifacts present in volume into equally sized bins (for more details, see
the vicinity of the membranes, only a part of this profile in Appendix C). The OPAS simulations must fulfill several con-
the inner compartment should be considered for evaluation. ditions:
In this work, this region is chosen to be 0.4 ≤ z/L ≤ 0.6
and referred to as the inner sampling volume. The averaged • The system size must be sufficient, so that the density
density of the solvent ρsolv in the inner sampling volume is of the solvent in the inner sampling volume is not
then used to compute the mole fraction of the solvent in the influenced by the membranes.
inner compartment via • The osmotic pressure Π obtained from sampling
the membrane force must be equal to the pressure
N
solutes
−1 difference ∆p = pin − pout between both compartments.
x solv = ρsolv*. ρsolv + ρi +/ , (7) • The chemical potential of the solvent must be constant
, i=1 - over the coordinate z.
where the densities of the solutes ρi are known, as the number The first condition can be checked by inspection of the
of solute molecules is an input of the simulation and the solute component density profiles ρi (z). For the second condition,
molecules are always in the inner compartment. sampling volumes for the evaluation of the pressure profile
When considering electrolyte solutions, activity coeffi- p(z) have to be defined. In this work, the inner sampling
cients and related quantities are usually expressed in terms of volume for the pressure was chosen to be the same as that for
the salt molality m instead of mole fractions. The conversion the densities 0.4 ≤ z/L ≤ 0.6, and accordingly, for the outer
from the ion-based mole fraction of the solvent x solv, obtained sampling volume the choice 0 ≤ z/L < 0.1 or 0.9 < z/L ≤ 1
by Eq. (7), to the salt molality and a corresponding estimation was made. By averaging over both sampling volumes, an
of the statistical uncertainty of this quantity is given in estimate of the pressure difference ∆p = pin − pout is obtained.
Appendix B. For the third condition, a profile of the reduced residual
Altogether, for a specific state point of interest, the OPAS chemical potential of the solvent µ̃solv(z) was recorded by
method has the following set of simulation parameters: applying Widom’s test particle insertion23 individually in each
bin of the simulation volume. The reduced residual chemical
• the membrane positions zM1 and zM2, or, respectively, potential is defined as24
the compartment volume ratio Vin/Vout;
• the number of particles N; µ̃solv = [µsolv − µid
solv(T)]/(kT), (8)
• the choice of the inner sampling volume for the where
evaluation of the density profile;
• the membrane force constant c. µsolv = µid
solv(T) + kT ln x solv
− kT ln⟨V exp(−ψsolv/kT)⟩/N. (9)
Obviously, a symmetric arrangement of the membranes
is convenient, as it leads to a simple modification of a Here, k is Boltzmann’s constant, ψsolv is the potential energy
standard barostat, cf. Eq. (5). It might be advantageous to of a solvent test molecule, and µid solv(T) is the part of the
use a box elongated in z-direction; however, this would ideal chemical potential of the solvent that only depends
automatically lead to a larger number of particles. In the on temperature, which is in fact an arbitrary normalization
present work, it was found that N = 4000 is sufficient to obtain for the chemical potential. Furthermore, in the fourth test,
smooth density profiles in a broad inner sampling volume a comparison against literature data for the solvent activity
of 0.4 ≤ z/L ≤ 0.6, cf. Section IV. Therefore, a different and the activity coefficient of the solute NaCl obtained with
arrangement is unnecessary. The impact of the membrane independent simulation methods is carried out.
force constant c was studied in a number of preliminary runs
and was found to be small. Furthermore, the good agreement
B. Molecular models
of OPAS results with results of other methods from the
literature9,10 shows that the choice of c adopted from Luo and In the tests for the OPAS method, different molecular
Roux21 is appropriate, cf. Section IV. model classes are used. The first two tests are carried out by
084112-4 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

studying liquid LJ mixtures. In the third test, a liquid mixture are free to pass the membrane, while the solute molecules
of argon and O2 is investigated, and the fourth test is a study of of component B are not. For this test, only the NVT run
aqueous sodium chloride solutions. Table I gives an overview has been performed to independently validate this step of
of the tests and the employed molecular models. All molecular an OPAS simulation. The state point used for this test
models used in the present study can be described as special is T = 1.2 ϵ/k and ρ = 0.6 σ −3. The system consists of
cases of the potential given in NA = 3600 solvent molecules and NB = 400 solute molecules.
Therefore, initially, the component densities in the outer
U = ULJ + UCC + UQQ
compartment are ρinitA,out = 0.6 σ
−3
and ρinit
B,out = 0.0 σ
−3
as
n LJ n LJ
N
 −1  N  i  j  ( σ
i j ab
) 12 (
σi j ab 6
) well as ρA,in = 0.48 σ and ρB,in = 0.12 σ in the inner
init −3 init −3

= 4ϵ i j ab  − compartment. The cut-off radius is r cut = 6.5 σ.

i=1 j=i+1 a=1 b=1
 r i j ab r i j ab 

e n e
ni j
2. Test 2: Binary LJ mixture II
 
  1 qic q j d Q ic Q j d
+ + 5 · f (ωi ,ω j ) . (10)
c=1 d=1
4πε 0 r i j c d ri j c d The second test for the OPAS method is similar to the
Here, ε 0 is the vacuum permittivity, ϵ i j ab and σi j ab are the first one. Both solvent A and solute B are LJ species sharing
Lennard-Jones energy and size parameters, r i j ab and r i j c d are the same size parameter, but with different energy parameters,
site-site distances, and qic , q j d , Q ic , and Q j d are the magnitude i.e., σ = σA = σB and ϵ B/ϵ A = 0.66. As for the first test,
of the electrostatic interactions, i.e., the point charges and only the NVT run has been performed. The state point used
quadrupole moments. Therein, f (ωi ,ω j ) is a dimensionless is T = 1.0 ϵ A/k and ρ = 0.7 σ −3. The system consists of
angle-dependent expression in terms of the orientation (ωi ,ω j ) NA = 3600 solvent molecules and NB = 400 solute molecules.
of the point quadrupoles.25 The cut-off radius is r cut = 6.5 σ.
The modified Lorentz-Berthelot combining rules are
used26,27 for the interaction between unlike Lennard-Jones 3. Test 3: Argon + O2 mixture
sites,
σiiaa + σ j j bb The third test for the OPAS method is a liquid mixture
σi j ab = , (11) of the solvent argon and the solute O2. The molecular models
√ 2 developed by Vrabec et al.29 were used for both species.
ϵ i j ab = ξ i j ϵ iiaa ϵ j j bb . (12)
The binary interaction parameter ξAr,O2 = 0.988 is taken from
The binary interaction parameter ξ i j is set to 1 for all Vrabec et al.28 Again, only the NVT run was performed.
combinations except for the mixture of argon and O2, where The state point used is T = 140 K and ρ = 26 mol l−1. The
the value ξAr,O2 = 0.988 adjusted by Vrabec et al.28 is used. system consists of NAr = 3600 solvent and NO2 = 400 solute
molecules. The cut-off radius is r cut = 22.75 Å.
C. Description of tests
4. Test 4: Aqueous NaCl solution
In the following, an overview of the tests is given. Results
are presented and discussed below. The last test for the OPAS method is an aqueous NaCl
solution, where the Na+ and Cl− ion models of Joung and
Cheatham III30 (JC) are used together with the SPC/E13
1. Test 1: Binary LJ mixture I
water model. The full OPAS method was run at the state
The first test for the OPAS method is a liquid binary point T = 298.15 K and pout = 1 bar. The system consists
LJ mixture. The solvent A and the solute B are equal of NW = 3920 water molecules and NNa+ = NCl− = 40 ions,
LJ species, i.e., σ = σA = σB and ϵ = ϵ A = ϵ B. The only respectively. The cut-off radius is r cut = 15 Å. Note that for
difference is that the solvent molecules of component A this simulation only the density and pressure profiles were

TABLE I. Overview of tests for the OPAS simulation method. The osmotic pressure ∆p is obtained from the pressure profile, and Π is obtained by sampling
the membrane force. Statistical uncertainties are given in parentheses.

Test Molecular models State point Osmotic pressure ∆p Osmotic pressure Π

Binary LJ mixture I LJ σ = σ A = σ B and ϵ = ϵ A = ϵ B T = 1.2 ϵ/k, ρ = 0.6 σ −3, 0.146(6) ϵσ −3 0.153(2) ϵσ −3

NA = 3600, NB = 400

Binary LJ mixture II LJ σ = σ A = σ B and ϵ B/ϵ A = 0.66 T = 1.0 ϵ A/k, ρ = 0.7 σ −3, 0.150(6) ϵ Aσ −3 0.148(2) ϵ Aσ −3
NA = 3600, NB = 400

Ar + O2 mixture Ar and O2 models by Vrabec T = 140 K, ρ = 26 mol l−1, 6.1(2) MPa 6.33(6) MPa
et al.,29 ξ Ar,O2 = 0.988 adjusted by Vrabec et al.28 NAr = 3600, NO2 = 400

Aqueous NaCl solution SPC/E13 for water, JC30 for Na+ and Cl− T = 298.15 K, p out = 1 bar, 7(2) MPa 6.0(2) MPa
NW = 3920,
NNa+ = NCl− = 40
084112-5 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

recorded, as the profile of the chemical potential of water

in an aqueous electrolyte solution cannot be obtained with
standard test particle insertion methods.6
The OPAS method was also used to simulate osmotic
pressures for aqueous solutions at different molalities of NaCl,
using the SPC/E + JC model combination and a cut-off radius
of r cut = 15 Å. The state point T = 298.15 K and pout = 1 bar
was used. This system has been investigated recently by
Mester and Panagiotopoulos10 and by Moucka et al.,9 thus
allowing for a comparison of results obtained with different
methods. The density ρSPC/E = 55.27 mol l−1 of pure SPC/E
water at T = 298.15 K and p = 1 bar is taken from Guevara-
Carrion et al.31
Using a normalization according to Raoult, the chemical
potential of the solvent is given by
µW = µ0W + RT ln(aW), (13)
where µ0W is the chemical potential of pure water. The OPAS
method provides results for the solvent activity, which were
correlated using the following functional form:
ln(aW) = −2mMW/m0 − MW ln(10)
(
2A
× β(m/m0)2 + √
B3 + B4 m/m0
√ √
4A ln(B m/m0 + 1) 2A m/m0 2A
)
+ − − 3 . (14)
B3 B2 B
Here, MW = 18.016 is the molecular mass of water, m0
= 1 mol kg−1, and A, B, and β are fit parameters. This
expression results from the semi-empirical correlation for
salt activity coefficients by Hamer and Wu32 when applying
the Gibbs-Duhem equation. As in the work of Mester and
Panagiotopoulos,10 the parameter A was fixed to A = 0.5677,
since it is related to the dielectric constant of the solvent (here:
SPC/E) via Debye-Hückel theory. The remaining parameters
B and β were fit to the OPAS simulation results, cf. Section IV.
The chemical potential of the salt is usually expressed based
on a “Henry-like” normalization according to which
FIG. 2. Component density, pressure, and chemical potential profiles from
OPAS simulation of a binary LJ mixture. Top: profiles of the reduced density
µNaCl = µ0NaCl + 2RT ln(m/m0) + 2RT ln(γNaCl), (15) for the solvent (component A, solid line) and the solute (component B, dashed
line). Middle: profile of the reduced pressure. Bottom: profile of the reduced
where γNaCl is the activity coefficient of the salt and the residual chemical potential of the solvent. All profiles are plotted versus the
standard state, corresponding to µ0NaCl, is a hypothetical reduced coordinate normal to the membrane. The vertical lines indicate the
solution of NaCl at infinite dilution, for which the salt activity positions of the two semipermeable membranes. The gray areas represent the
equals one. Since the correlation for the solvent activity, cf. inner and outer sampling volumes for the evaluation of the profiles. The box
length is L ≈ 18.8 σ.
Eq. (14), was derived from the correlation for the activity
coefficient of the salt as proposed by Hamer and Wu32
( √ ) osmotic equilibrium, the pressure in the inner compartment is
−A m/m0
ln(γNaCl) = ln(10) √ + βm/m0 , (16) higher than in the outer compartment. The resulting pressure
1 + B m/m0 difference ∆p = pin − pout is in good agreement with the
the parameters obtained by fitting the solvent activity can be osmotic pressure Π derived from the membrane force, as can
transferred. be seen in Table I. The values for Π and ∆p only differ within
their statistical uncertainties, which are quantified here as the
doubled standard deviation obtained by block averaging, cf.
IV. RESULTS AND DISCUSSION Appendix C. This shows that systematic differences between
Π and ∆p which could occur due to finite-size effects in
A. Test 1: Binary LJ mixture I
computing the pressure profile33 are not important here. The
Figure 2 shows the simulation results for the binary advantage of the direct computation of Π via the membrane
LJ mixture I. The density profile shows a constant density force is that it is computationally much cheaper than the
of the solvent in the inner sampling volume. Due to the evaluation of the pressure profile and yet yields results with
084112-6 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

smaller statistical uncertainties. The statistical uncertainty pressure profile is in good agreement with the osmotic pressure
using the sampling of the membrane force is about three times Π derived from the membrane force, cf. Table I. Again, the
lower than using the pressure profile. In contrast to the density sampling of the membrane force yields three times lower
and pressure profiles, the profile of the chemical potential of statistical uncertainties for the osmotic pressure than the
the solvent remains constant over the membranes. pressure profile. The profile of the chemical potential of the
solvent remains constant over the membranes.
B. Test 2: Binary LJ mixture II
C. Test 3: Argon + O2 mixture
Figure 3 shows the simulation results for the binary LJ
mixture II. As for the first test, the density profile shows a Figure 4 shows the simulation results for the mixture
constant density of the solvent in the inner sampling volume. of argon and O2. All three profiles show the same behavior
Furthermore, the pressure difference ∆p obtained from the as for the tests with LJ particles. A constant density of the

FIG. 3. Component density, pressure, and chemical potential profiles from FIG. 4. Component density, pressure, and chemical potential profiles from
OPAS simulation of an unlike LJ mixture. Top: profiles of the reduced density OPAS simulation of mixture of argon and O2. Top: profiles of the reduced
for the solvent (component A, solid line) and the solute (component B, dashed density for the solvent argon (solid line) and the solute O2 (dashed line). Mid-
line). Middle: profile of the reduced pressure. Bottom: profile of the reduced dle: profile of the reduced pressure. Bottom: profile of the reduced residual
residual chemical potential of the solvent. All profiles are plotted versus the chemical potential of the solvent. All profiles are plotted versus the reduced
reduced coordinate normal to the membrane. The vertical lines indicate the coordinate normal to the membrane. The vertical lines indicate the positions
positions of the two semipermeable membranes. The gray areas represent the of the two semipermeable membranes. The gray areas represent the inner and
inner and outer sampling volumes for the evaluation of the profiles. The box outer sampling volumes for the evaluation of the profiles. The box length is
length is L ≈ 17.9 σ. L ≈ 63.5 Å.
084112-7 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

solvent argon is reached in the inner sampling volume. The

osmotic pressures derived from the two independent methods
are in acceptable agreement (cf. Table I), see the discussion
above, and the chemical potential of the solvent argon remains
constant over the box length. These results confirm that the
method is valid not only for simple LJ systems but also for
models of real fluids with electrostatic interactions.

D. Test 4: Aqueous NaCl solution

Figure 5 shows the simulation results for the test of
an aqueous NaCl solution. The density of the solvent water
is constant in the inner sampling volume. The same holds
true for the densities of the two ions Na+ and Cl−, which
coincide due to the electroneutrality condition. From the
pressure profile in Figure 5 and from Table I it can be clearly
seen that the computation of the osmotic pressure via the
membrane force is more accurate than the evaluation of the
pressure difference, as the former method leads to much
smaller statistical uncertainties.
Figure 6 shows the simulation results for aqueous NaCl
solutions at varying salt concentrations with the SPC/E + JC
model combination. The numerical simulation results are
given in Table II.

FIG. 6. Osmotic pressures (top), water activity (middle), and activity coef-
ficient of NaCl (bottom) over the salt molality for aqueous NaCl solutions
using the SPC/E + JC30 model combination. Present simulation results are
shown as open circles, the statistical uncertainties are within symbol size, and
the correlation to these results is shown as the dashed-dotted line. Correlations
fitted to simulation results obtained by Mester and Panagiotopoulos10 (dashed
line) and by Moucka et al.9 (dotted line) using the same molecular models as
well as the correlation to experimental data by Hamer and Wu32 (solid line)
are also shown for comparison.

Figure 6 also contains the results of Mester and

Panagiotopoulos10 and of Moucka et al.,9 who used the
same molecular models, but different simulation methods,
as well as the correlation to experimental data by Hamer
and Wu.32 First, it is clear that the SPC/E + JC model
FIG. 5. Component density and pressure profiles from OPAS simulation of combination overestimates the osmotic pressures in aqueous
an aqueous NaCl solution using the SPC/E + JC30 model combination. Top:
NaCl solutions at all salt molalities, and the deviations are
profiles of the density for the solvent water (solid line) and the ions Na+
(dashed line) and Cl− (dotted line). These two lines for the density of the more pronounced for high molalities. Consequently, the water
ions are undiscernable. The salt molality evaluated from the density profile is activity is underestimated, but this is irrelevant for the present
m ≈ 1.14 mol kg−1. Bottom: profile of the pressure. Both profiles are plotted test. The results of all three molecular simulation methods
versus the reduced coordinate normal to the membrane. The vertical lines
indicate the positions of the two semipermeable membranes. The gray areas
are in very good agreement, which indicates that the OPAS
represent the inner and outer sampling volumes for the evaluation of the simulation is reliable. By fitting the water activity to the
profiles. The box length is L ≈ 49.0 Å. functional form in Eq. (14), the parameters B = 1.768 and
084112-8 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

TABLE II. Osmotic pressure and water activity at T = 298.15 K and varying presented. The advantage of the method is its simplicity. In
salinity obtained by OPAS simulation for aqueous NaCl solutions using contrast to other methods for computing solvent activities in
the SPC/E + JC30 model combination. Statistical uncertainties are given in
parentheses.
electrolyte solutions,5–10 the OPAS method does not require
using sophisticated insertion schemes. Although the OPAS
m/mol kg−1 Π/MPa ln(a W) method is mainly intended for studying electrolyte solutions,
it is also applicable to any kind of mixture.
0.3358(2) 1.7(1) −0.0122(8)
The method was carefully tested using systems of
0.5628(4) 2.7(1) −0.020(1)
1.141(1) 5.8(2) −0.043(1)
varying complexity and successfully passed all tests. The
1.738(2) 9.6(2) −0.070(2) simulation results from the OPAS method agree well with
2.350(3) 14.3(4) −0.105(2) results obtained with other molecular simulation methods.9,10
2.985(4) 19.5(4) −0.142(4) Favorable agreement is also observed when the results are
3.636(4) 24.9(4) −0.182(4) transferred to the activity coefficient of the salt by means of
5.018(7) 37.9(6) −0.277(4) the Gibbs-Duhem equation, and when these data are used for
6.49(1) 55.3(8) −0.404(6) solubility calculations.
From the results of OPAS simulations, also data on other
properties related to the solvent activity can be obtained,
such as the lowering of the solvent vapor pressure or the
β = 0.106 are obtained (A = 0.5677 was not adjusted, see freezing point depression. Additionally, by employing the
above). Note that Mester and Panagiotopoulos10 and Moucka Gibbs-Duhem equation, the computation of salt solubilities
et al.8 use more parameters to fit their data, but the three- is possible if the chemical potential of the pure solid salt is
parameter equation employed here is sufficient to describe the known.
data well.
Finally, the fit parameters are transferred to the correlation
for the activity coefficient of the salt, cf. Eq. (16) and ACKNOWLEDGMENTS
Figure 6. As for the water activity, it is clear that the The authors gratefully acknowledge financial support
SPC/E + JC model does not provide a good representation within the Reinhart Koselleck Program (No. HA1993/15-1)
of the activity coefficient of the salt, especially at higher of the German Research Foundation (DFG). The present work
molalities. The activity coefficients of NaCl computed with was conducted under the auspices of the Boltzmann-Zuse
the three independent molecular simulation methods show Society of Computational Molecular Engineering (BZS) and
some deviations. However, the shape of the curves is in the simulations were carried out on the Regional University
favorable agreement and the deviations are probably within Computing Center Kaiserslautern (RHRK) under the grant
the combined statistical uncertainties of the three independent TUKL-MSWS as well as the High Performance Computing
results. Furthermore, the activity coefficient of the salt is Center Stuttgart (HLRS) under the grant MMHBF. The authors
considerably more sensitive than the chemical potential of the would like to thank Stephan Deublein, Rajat Srivastava, Colin
salt, since the latter property is mainly dominated by the term Glass, and Jadran Vrabec for fruitful discussions.
2RT ln(m), cf. Eq. (15). Therefore, we conclude that results
from OPAS simulations can also be reliably transferred to the
activity coefficient of the salt. APPENDIX A: DERIVATION OF SOLVENT ACTIVITY
The activity coefficient of the salt in aqueous solution
can be used for computing the salt solubility in water. For Let ′ denote the pure solvent phase and ′′ denote
this calculation, numbers for the reference chemical potential the solution phase in osmotic equilibrium at a constant
of the salt and the chemical potential of the solid salt are temperature T. Equality of the chemical potential of the solvent
needed. Taking µ0NaCl = −391.6 kJ mol−1 from Mester and in both phases, together with a normalization according to
Panagiotopoulos,10 µsolid
′
Raoult, for which asolv = 1, yields
NaCl = −384.37 kJ mol
−1
from Moucka
et al. for the SPC/E + JC model (as suggested by Mester
7
pure pure
and Panagiotopoulos10), and the OPAS results for ln(γNaCl), a µsolv (T, p′) = µsolv (T, p′′) + RT ln asolv
′′
(T, p′′, x ′′), (A1)
solubility of 3.34 mol kg−1 of NaCl in water is obtained. This
where x denotes the vector of mole fractions describing
is in the range of the results of the other groups, who obtained
the composition of the phase. The difference in chemical
4.74 mol kg−1 (Moucka et al.7), 3.64 mol kg−1 (Moucka
potentials of the pure solvent can be calculated from the
et al.34), and 3.59 mol kg−1 (Mester and Panagiotopoulos10), pure pure
molar volume of the pure solvent vsolv = 1/ρsolv , so that from
where the experimentally determined value for NaCl in water
Eq. (A1) it follows that
is 6.15 mol kg−1 (Lide35). Therefore, we conclude that results
from OPAS simulations can also be used as one step of p
′′

solubility calculations. pure

vsolv dp = −RT ln asolv
′′
(T, p′′, x ′′), (A2)
p′

V. CONCLUSIONS pure
or, assuming that vsolv is independent of pressure
A new molecular simulation method for the computation pure
of solvent activities in mixtures, called OPAS simulation, is vsolv (p′′ − p′) = −RT ln asolv
′′
(T, p′′, x ′′). (A3)
084112-9 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

The pressure difference (p′′ − p′) is the osmotic pressure Π estimated with the block average method by Flyvbjerg and
and therefore Petersen.36
Π vsolv
pure The total number of particles was 4000 throughout. In all
′′
ln asolv(T, p′′, x ′′)
=− . (A4) MD simulations, Newtons equations of motion were solved
RT with a gear predictor-corrector scheme of fifth order. The long-
In the present work, we assume that the influence of pressure range interactions were considered by Ewald summation.37
on the activity of the solvent in the studied liquid mixtures The calculation of the pressure profiles is based on the method
is negligible. This is usually the case as the ratio of solvent proposed by Kirkwood and Buff.38
activities at different pressures (but constant temperature and For the simulations of electrolyte solutions, at first
composition) can be expressed as simulations in a modified isothermal-isobaric (NpT) ensemble
p
′′ were carried out (details see main text). A physically
asolv(T, p′′, x) 1 pure reasonable configuration was obtained after equilibrating
ln = (vsolv − vsolv ) dp, (A5)
asolv(T, p′, x) RT for 2 000 000 time steps, followed by a production run of
p′
10 000 000 time steps. The resulting box volume V was
where vsolv is the partial molar volume of the solvent in the then used for a run in the canonical (NVT) ensemble,
pure
mixture. As vsolv ≈ vsolv , the influence of the pressure on the in which equilibration and production took 3 000 000 and
solvent activity is small. If desired, Eq. (A5) can be used to 10 000 000 time steps, respectively. The time step was
correct for the pressure influence. ∆t = 1.2 fs.
For the simulations of the tests (binary LJ mixtures and
argon + O2 mixture), simulations were directly carried out in
APPENDIX B: CONVERSION the canonical (NVT) ensemble. The number of time steps was
OF CONCENTRATION UNITS the same as for the electrolyte solution, and the reduced time
Application of Eq. (7) of the main text yields the ion-based step was ∆t ∗ = 0.0005. Widom’s test particle insertion23 was
mole fraction of the solvent, so that in case of a 1-1 electrolyte, used to calculate the chemical potential of the solvent, and
5400 test particles were inserted at every production time
x solv + 2x ion = 1. (B1) step.
In order to obtain the molality of the salt, the ion-based mole
fraction of the solvent has to be converted to the salt-based 1K. S. Pitzer, J. Phys. Chem. 77, 268 (1973).
2K. S. Pitzer and G. Mayorga, J. Phys. Chem. 77, 2300 (1973).
mole fraction of the solvent, 3K. S. Pitzer and G. Mayorga, J. Solution Chem. 3, 539 (1974).
x ion 4K. S. Pitzer and J. J. Kim, J. Am. Chem. Soc. 96, 5701 (1974).
x̃ solv = 1 − . (B2) 5M. Lisal, W. R. Smith, and J. Kolafa, J. Phys. Chem. B 109, 12956
1 − x ion
(2005).
The mole fraction of the solvent can then be converted to the 6F. Moucka, M. Lisal, J. Skvor, J. Jirsak, I. Nezbeda, and W. R. Smith, J. Phys.
salt molality Chem. B 115, 7849 (2011).
7F. Moucka, M. Lisal, and W. R. Smith, J. Phys. Chem. B 116, 5468
x̃ salt 1 − x̃ solv
m= = , (B3) (2012).
8F. Moucka, I. Nezbeda, and W. R. Smith, J. Chem. Phys. 139, 124505
x̃ solv Msolv x̃ solv Msolv
(2013).
where Msolv is the molecular mass of the solvent in kg mol−1. 9F. Moucka, I. Nezbeda, and W. R. Smith, J. Chem. Theory Comput. 11, 1756

During the simulation, the statistical uncertainty of the (2015).

10Z. Mester and A. Z. Panagiotopoulos, J. Chem. Phys. 142, 044507 (2015).
average density of the solvent ρsolv is monitored. It is assumed 11S. Deublein, J. Vrabec, and H. Hasse, J. Chem. Phys. 136, 084501 (2012).
to be three times the standard deviation obtained with the block 12S. Reiser, S. Deublein, J. Vrabec, and H. Hasse, J. Chem. Phys. 140, 044504
average method by Flyvbjerg and Petersen.36 The uncertainty (2014).
13H. J. C. Berendsen, J. R. Grigera, and T. P. Straatsma, J. Phys. Chem. 91,
of the salt molality is derived by applying error propagation
through all conversion steps in Eqs. (7) and (B1)–(B3). 6269 (1987).
14S. Reiser, M. Horsch, and H. Hasse, J. Chem. Eng. Data 59, 3434 (2014).
15S. Reiser, M. Horsch, and H. Hasse, J. Chem. Eng. Data 60, 1614 (2015).
16S. Reiser, M. Horsch, and H. Hasse, Fluid Phase Equilib. 408, 141

APPENDIX C: SIMULATION DETAILS (2016).

17J. G. Powles, S. Murad, and P. V. Ravi, Chem. Phys. Lett. 188, 21 (1992).
18S. Murad and J. G. Powles, J. Chem. Phys. 99, 7271 (1993).
All simulations of this study were carried out with
19S. Murad, J. G. Powles, and B. Holtz, Mol. Phys. 86, 1473 (1995).
an extended version of the simulation program ms2.22 In 20F. Paritosh and S. Murad, AIChE J. 42, 2984 (1996).
ms2, thermophysical properties can be determined for rigid 21Y. Luo and B. Roux, J. Phys. Chem. Lett. 1, 183 (2010).
molecular models using MC or MD simulation techniques. For 22C. W. Glass, S. Reiser, G. Rutkai, S. Deublein, A. Koester, G.

all simulations, the LJ interaction partners are determined for Guevara-Carrion, A. Wafai, M. Horsch, M. Bernreuther, T. Windmann, H.
every time step and MC loop, respectively. Interaction energies Hasse, and J. Vrabec, Comput. Phys. Commun. 185, 3302 (2014).
23B. Widom, J. Chem. Phys. 39, 2808 (1963).
between molecules and/or ions are determined explicitly for 24J. Vrabec and H. Hasse, Mol. Phys. 100, 3375 (2002).
distances smaller than the cut-off radius r cut. The thermostat 25C. G. Gray and K. E. Gubbins, Theory of Molecular Fluids: Fundamentals

incorporated in ms2 is velocity scaling. The pressure is (Clarendon Press, Oxford, 1984), Vol. 1.
26H. A. Lorentz, Ann. Phys. 248, 127 (1881).
kept constant using Andersen’s barostat in MD, and random 27D. Berthelot, Comptes Rendus de l’Academie des Sciences Paris 126, 1703
volume changes evaluated according to Metropolis acceptance (1898); 126, 1857 (1898).
criterion in MC, respectively. The simulation uncertainties are 28J. Vrabec, Y. Huang, and H. Hasse, Fluid Phase Equilib. 279, 120 (2009).
084112-10 Kohns et al. J. Chem. Phys. 144, 084112 (2016)

29J.Vrabec, J. Stoll, and H. Hasse, J. Phys. Chem. B 105, 12126 (2001). 34F. Moucka, I. Nezbeda, and W. R. Smith, J. Chem. Phys. 138, 154102
30I.Joung and T. E. Cheatham III, J. Phys. Chem. B 112, 9020 (2008). (2013).
31G. Guevara-Carrion, J. Vrabec, and H. Hasse, J. Chem. Phys. 134, 074508 35CRC Handbook of Chemistry and Physics, edited by D. R. Lide, Internet

(2011). Version 2005 (CRC Press, Boca Raton, 2005).

32W. J. Hamer and Y. C. Wu, J. Phys. Chem. Ref. Data 1, 1047 (1972). 36H. Flyvbjerg and H. G. Petersen, J. Chem. Phys. 91, 461 (1989).
33S. Werth, S. V. Lishchuk, M. Horsch, and H. Hasse, Physica A 392, 2359 37P. P. Ewald, Ann. Phys. 64, 253 (1921).

(2013). 38J. G. Kirkwood and F. P. Buff, J. Chem. Phys. 17, 338 (1949).