4

The GvdW Theory: A Density Functional

Theory of Adsorption, Surface Tension,
and Screening
STURE NORDHOLM Department of Chemistry, Göteborg University,
Göteborg, Sweden
ROBERT PENFOLD Food Materials Science Division, Institute of Food
Research, Norwich, England

I. INTRODUCTION

The debt owed to J. van der Waals for revealing the essence of fluid behavior in his famous
analysis of the binding energy and excluded volume mechanisms [1,2] is widely recognized
as enormous. Less known is the fact that van der Waals also extended his theory to
nonuniform fluids [3]. This work was a beginning of the development of so-called density
functional theories which have only recently gained prominence in both quantum chem-
istry [4,5] and statistical mechanics [6,7]. Here we shall review the origin of the generalized
van der Waals (GvdW) theory with the aim to show that the remarkable accuracy of
simple mechanistic analysis applies also to a wide range of nonuniform fluid phenomena
such as adsorption, surface tension, and electrolyte screening. Thus the traditional equa-
tion of state of van der Waals as well as the Debye–Hückel and Poisson–Boltzmann
theories of electrolyte screening are simple special cases of the GvdW theory which
unify and considerably extend these well-known traditional approximations to account,
in particular, for nonlocal entropy and short-range packing structure.
The key idea of the GvdW theory is a greatly simplified treatment of correlations in
the fluid. Note that if there were no interactions then the ideal gas partition function would
be a sum over all possible configurations of the fluid. In a realistic interacting fluid these
configurations would occur with a weighting factor for each. This weighting factor would
reflect the potential energy VðRÞ associated with the configuration R through the
Boltzmann factor exp½VðRÞ. This innocent-looking weighting factor introduces enor-
mous complexity which will continue to be the subject of analysis and approximation in
liquid state theory for all the foreseeable future. van der Waals introduced the idea that for
bulk fluids the weighting factor be allowed only two possible values, 0 or 1. In this way the
ensemble of configurations and weights was dramatically simplified. The configurations
which showed overlap between hard repulsive parts of the pair potentials acting between
the particles were simply eliminated by the so-called ‘‘excluded volume effect.’’ The aver-
age potential energy was then calculated by the mean field approximation, i.e., over the
truncated but otherwise unweighted ensemble of configurations. In this way entropy,
internal energy, and free energy could be obtained by greatly simplified calculations.

83

Copyright © 2001 Marcel Dekker, Inc.

 84 Nordholm and Penfold

This approach certainly does not always yield thermodynamic properties of high
accuracy. It is, however, uniquely well suited to the treatment of simple fluids. We shall see
that it works very well for Lennard–Jones models of simple fluids which have quite short-
range interactions. It also works very well, in principle better, for primitive models of
electrolyte solutions. We are able by this GvdW theory to resolve both short-range pack-
ing structure at hard walls and long-range screening mechanisms in electrolyte solutions.
We feel therefore that this theory should be of great utility in the field of electrochemistry
where both short-range fluid properties and long-range electrolyte screening mechanisms
meet and mingle. Perhaps the greatest advantage of the theory is that it retains great
simplicity, while offering generality and numerical efficiency through the variational nat-
ure of the density functional formulation. Thus, it need not remain in the hands of a few
specialists but will, hopefully, be useful for nearly any practitioner of chemistry.

II. THE VAN DER WAALS LEGACY: GENERALIZED

Every chemist knows the van der Waals equation of state,

kB T a
P¼  2 ð1Þ
v  v0 v
In it the pressure P is found in terms of temperature T and volume per particle v. The
interactions enter through two parameters: v0 , the excluded volume per particle, and a, the
binding energy per particle at unity density in a bulk fluid. The equation of state arose as
an extension of the ideal gas law,
kB T
P¼ ð2Þ
v
It was assumed that the volume V was reduced by the presence of ‘‘other particles’’ to the
free volume V  Nv0 where N is the number of particles. In arriving at the binding energy
effect the mean field approximation was used, which says that the soft (negative) part of
the pair potential was sampled in an uncorrelated manner as if the system was an ideal gas.
The corresponding free energy per particle in the bulk fluid is
 
n
f ¼ 3kB T ln  þ kB T ln  kB T  an ð3Þ
1  nv0
where  is the thermal wavelength which does not depend on volume and n is the particle
density N=V. The third term on the RHS, kB T, is the so-called communal entropy term
which reflects the large density fluctuations in the bulk ideal gas. If we instead restrict the
ideal gas particles to singly occupied cells, each of volume v ¼ V=N, then the communal
entropy term disappears and we get
 
n
fc ¼ kB T ln  an ð4Þ
1  nv0
Here we have limited ourselves to the configurational free energy per particle fc , i.e., we
have omitted the kinetic part which in classical statistical mechanics just yields the first
term 3kB T ln . The first minor generalization of van der Waals’ original derivation is the
omission of the communal entropy. We will be dealing with interacting, often dense, fluids
which do not display the same density fluctuations as do the corresponding ideal gases.

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 85

Thus we prefer to make the cell model our reference for density fluctuations. Density
fluctuations are thereby truncated but in a well-defined way open to correction.
The second generalization is the reinterpretation of the excluded volume per particle
v0 . Realizing that only binary collisions are likely in a low-density gas, van der Waals
suggested the value 2 3 =2 for hard spheres of diameter  and for particles which were
modeled as hard spheres with attractive tails. Thus, for the Lennard–Jones fluid where the
pair potential actually is
" #
 12  6

ðrÞ ¼ 4 12  6 ð5Þ
r r

the interaction is typically approximated as a hard sphere potential for r < . The estimate
of van der Waals is chosen to give the correct second virial coefficient B2 ðTÞ at low
densities. At higher densities the excluded volume per particle decreases due to the overlap
of excluded volume spheres centered on each particle. Thus the effective excluded volume
v0 is density dependent and it decreases monotonically with density. If we are to treat a
wide range of fluid densities it is then better to determine a constant excluded volume v0 in
the high-density limit. In fact, in the simplest form of GvdW theory denoted GvdW(S) [8],
v0 is taken to be  3 to give the pressure divergence at the close packing limit in the simple
cubic crystal structure.
We shall use the Lennard–Jones fluid as the illustrative example as we review the
GvdW theory. In the GvdW(S) theory the equation of state is just like the vdW equation
of state above with the parameters given by v0 ¼  3 and a ¼ 16 3 =9. This equation of
state works moderately well. However, we do not wish to deal only with bulk thermo-
dynamic properties. We wish to treat surface tension and adsorption, screening phenom-
ena in electrolyte solutions, etc. To do this we need a free energy density functional, i.e., an
expression for the configurational free energy as a function of the particle density which is
now considered to be a function of position. If all particle–particle interactions were local
in the fluid then the functional would have been of the form
ð ð
Fc ¼ drnðrÞfc ðnðrÞÞ þ drnðrÞVext ðrÞ ð6Þ

where fc ðnÞ is the free energy per particle in the corresponding bulk fluid. Here Vext ðrÞ is an
external potential.
This type of fully local potential has some limited use, e.g., to consider adsorption in
a slowly varying external potential. It fails, however, to describe the most important
phenomena such as surface tension and adsorption at most types of interfaces. These
phenomena reflect in a fundamental way the nonlocal interactions in the fluid. The
most obvious nonlocality of the free energy arises due to the range of the attractive or
soft interactions represented by the second term in the equation of state, a=v2 . The
corresponding potential energy can be obtained by the functional
ð ð
1
EV ¼ drnðrÞ dr 0 nðr 0 Þ
s ðr; r 0 Þ ð7Þ
2

Here
s is the soft and normally attractive part of the pair potential. This simple bilinear
form of functional lacks correlation effects except that which is introduced by the trunca-
tion of the integral at the onset of the inner hard part of the potential. We are then using
an extended form of the mean field approximation as did van der Waals in his original

Copyright © 2001 Marcel Dekker, Inc.

 86 Nordholm and Penfold

work. We can now write down the first truly useful free energy density functional in the
form
ð   ð ð ð
nðrÞ 1
Fc ¼ kB T drnðrÞ ln þ drnðrÞ dr 0 nðr 0 Þ
s ðr; r 0 Þ þ drVext ðrÞ ð8Þ
1  nðrÞv0 2
This type of functional, which we refer to as ‘‘coarse-grained,’’ can be used to calculate
both surface tension and adsorption isotherms to quite good accuracy for many fluids and
interfaces. It can also be used for screening problems in the theory of electrolytes.
We shall illustrate the applicability of the GvdW(S) functional above by considering
the case of gas–liquid surface tension for the Lennard–Jones fluid. This will also introduce
the variational principle by which equilibrium properties are most efficiently found in a
density functional theory. Suppose we assume the profile to be of step function shape, i.e.,
changing abruptly from liquid to gas density at a plane. In this case the binding energy
integrals in EV can be done analytically and we get for the surface tension [9]
3 4 27
¼  ðnl  ng Þ2 ¼ aðnl  ng Þ2 ð9Þ
4 64
This expression is not particularly accurate, as shown in Fig. 1, but it tells a large part of
the story of surface tension. It arises predominantly due to the loss of binding energy of
the liquid-phase particles at the interface. It is roughly proportional to the product of the
binding energy constant a and the range of the attractions . It is also in this approxima-
tion proportional to the square of the density difference nl  ng . We shall see, however,
that this proportionality is more nearly to the fourth power of the density difference.
A more accurate result can be obtained by allowing the density profile at the inter-
face to relax. The optimal profile is the one which minimizes the free energy. The surface
tension is computed by first calculating the free energy for the interface (over a sufficiently
wide slice of the liquid and gas phases) and then subtracting the corresponding free energy
obtained from the fully local functional. The fully optimized profile can be obtained by
iteratively refining it point by point to yield the best possible shape, i.e., that correspond-
ing to the lowest free energy. More conveniently, and very nearly as accurately, we can

FIG. 1 The calculated surface tension of an argon fluid represented as a Lennard–Jones fluid is
shown as a function of temperature. The GvdW(HS-B2 )-functional is used in all cases. The filled
squares correspond to step function profile and local entropy, the filled circles to tanh profile with
local entropy, and the open circles to tanh profile with nonlocal entropy. The latter data are in good
agreement with experiment.

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 87

choose a plausible profile shape with a width parameter  and then find the optimal -
value. A favored profile shape is the tanh profile,
n l  ng
nðxÞ ¼ ng þ ð10Þ
1 þ expðxÞ
The corresponding result for the surface tension [9] provides quite reasonable accuracy for
a Lennard–Jones fluid or an inert gas fluid, except helium which displays large quantum
effects. Thus we can conclude that the leading mechanisms of surface tension in a simple
fluid is the loss of binding energy of the liquid phase at the gas–liquid interface and the
second most important mechanism is likely to be the adsorption–depletion at the interface
which creates a molecularly smooth density profile instead of an abrupt step in the density.
The next most important mechanism affecting the surface tension at a single com-
ponent simple fluid gas–liquid interface is, we believe, associated with the nonlocality of
the repulsive interactions. To account for this mechanism, observe that it enters by way of
the excluded volume effect. In the coarse-grained GvdW(S) theory above, the free volume
factor ðrÞ is given by
ðrÞ ¼ 1  nðrÞ 3 ð11Þ
for a Lennard–Jones fluid. The role of the density in this expression is to describe the
number of other particles within repulsive range. If the particle density is varying signifi-
cantly on the length scale  then the local particle density nðrÞ should clearly be replaced by
some nonlocal average. The natural domain for such an average is the sphere of radius ,
the range of the hard repulsion, centered on the location of our test particle at r. Thus, we
get in the so-called ‘‘fine-grained’’ GvdW(S) theory [10]
ðrÞ ¼ 1  n ðrÞ 3 ð12Þ
where n ðrÞ is the coarse-grained particle density defined by
ð
3
n ðrÞ ¼ dsn ðsÞ ð13Þ
4 3 jrsj<
This is clearly only the simplest of possible nonlocal density averages which accounts for
the range of the repulsion. Within the GvdW theory we have adhered to this simplest form
which works very well as it stands. There are, however, a number of different suggestions
available in the density functional literature which arise from matching the GvdW
approach to the modern liquid state theory based on pair-correlation analysis. The func-
tionals obtained this way are generally referred to as ‘‘weighted-density functionals’’ [6,7].
As can be seen in Fig. 1 the nonlocal entropy in the form of a nonlocal excluded
volume effect systematically reduces the surfaces tension of a Lennard–Jones or inert gas
fluid and brings the GvdW result close to corresponding simulation or experimental
results. One can understand this in terms of the easing of excluded volume constraints
on the dominant liquid phase close the interface. The nonlocal entropy also brings the
possibility of describing the characteristic packing structure of a simple fluid around one
of its own particles or at a hard wall. In the ‘‘fine-grained’’ GvdW theory (because it
distinguishes between the fine-grained particle density nðrÞ and the coarse grained n ðrÞ
entering the free volume factor) these ubiquitous oscillations can be understood as a
consequence of the ‘‘overlapping excluded volume’’ mechanism. At a hard wall the
excluded volume spheres of neighboring hard spheres are partially buried in the already
excluded volume associated with the wall [10–12]. This provides excess free volume for the
rest of the fluid and thereby attracts the first shell of particles towards the wall. They then

Copyright © 2001 Marcel Dekker, Inc.

 88 Nordholm and Penfold

make up a structured wall for a second layer to be attracted to, thus forming another wall
attracting a third layer, and so on.
In order to accurately describe such oscillations, which have been the center of
attention of modern liquid state theory, two major requirements need be fulfilled. The
first has already been discussed above, i.e., the need to accurately resolve the nonlocal
interactions, in particular the repulsive interactions. The second is the need to accurately
resolve the mechanisms of the equation of state of the bulk fluid. Thus we need a mechan-
istically accurate bulk equation of state in order to create a free energy functional which
can accurately resolve nonuniform fluid phenomena related to the nonlocality of interac-
tions. So far we have only discussed the original van der Waals form of equation of state
and its slight modification by choosing a high-density estimate for the excluded volume,
v0 ¼  3 for a fluid with effective hard sphere diameter , instead of the low-density
estimate v0 ¼ 2 3 =3 suggested by van der Waals. These two estimates really suggest
that the excluded volume per particle is density dependent. Several different forms of
equations of state have been developed within the GvdW effort which account for this
density dependence of the excluded volume [13,14]. Here we shall describe only one simple
extension of this type which nevertheless produces substantial improvements in the treat-
ment of surface tension. We call it the GvdW(HS-B2 ) equation of state [14] because it is
designed to reproduce the hard sphere second virial coefficient B2 in the low-density limit
while retaining close packing at the simple cubic limiting density of n ¼  3 . Its form is as
follows:


a  2  1  3 k T
k T 3 B
P¼ B 3 ð14Þ
v v2

Here the correction to obtain the exact hard sphere B2 is added linearly while at high
density v0 acts nonlinearly and defines the close-packing limit. The corresponding free
energy density functional is
ð      
nðrÞ 2
Fc ¼ kB T drnðrÞ ln þ  1  n ðrÞ
3
1  nðrÞ 3 3
ð ð Z ð15Þ
1
þ drnðrÞ dr 0 nðr 0 Þ
s ðr; r 0 Þ þ drVext ðrÞ
2

This functional gives a good account of both the equation of state and the surface tension
of a Lennard–Jones fluid. The surface tension is shown in Fig. 1. The critical parameters
obtained from the equation of state are shown in Table 1.

TABLE 1 Critical Parameters for the Lennard–Jones Model are Given in Reduced Units such
that  ¼ 1 and  ¼ 1

Source P
c n
c Tc

GvdW(HS-B2 ) 0.156 0.333 1.25

Simulation 0.13–0.17 0.32–0.36 1.32–1.36
Argon (experiment) 0.116 0.316 1.26

The simulation results have been extracted from the literature: see, e.g., B. Smit. J. Chem. Phys. 96:8639 (1992).

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 89

III. SURFACE TENSION IN MIXTURES

The GvdW theory has been applied also to mixtures of Lennard–Jones fluids [15,16]. The
extension to mixtures is straightforward with respect to the binding energy and interaction
with an external field but not quite so straightforward with respect to the excluded volume
effect. The GvdW(S) theory produces for a mixture of c components an equation of state
of the form
Xc c X
X c
kB T aij
P¼  ð16Þ
v  v0i i¼1 j¼1 v2i
i¼1 i

Here we have taken the pair potential acting between particles of species i and j, respec-
tively, to be
!
ij12 ij6

sij ðrÞ ¼ 4ij 12  6 ð17Þ
r r

The choice of appropriate potential parameters to use in the study of a certain mixture can
be a significant problem. Traditionally the Lorentz–Berthelot rules,
1
ij ¼ ðii þ jj Þ ð18Þ
2
pffiffiffiffiffiffiffiffiffi
ij ¼ ii jj ð19Þ

are frequently used to allow between-species Lennard–Jones potentials to be found from

within-species potentials which are better known. The former (Lorentz) rule is generally
found adequate but the latter (Berthelot) rule is often found to overestimate the strength
of between species attraction. Thus one often introduces a correction factor kij
pffiffiffiffiffiffiffiffiffi
ij ¼ kij ii jj ð20Þ

where kij is typically 0.95 or so, i.e., a little below unity.

The form of the free volume factor ðrÞ is so far generally taken to be
X
c
i ðrÞ ¼ 1  n i ðrÞii3 ð21Þ
i¼1

This form is particularly simple, since ðrÞ is independent of the type of species i. It
behaves as if each particle excluded a volume equal to the smallest cube into which its
effective hard sphere shape could be placed. The alternative form
X
c
i ðrÞ ¼ 1  n i ðrÞij3 ð22Þ
i¼1

may seem more realistic, since it refers the excluded volume to the type of particle. Thus a
large particle will be more squeezed by its neighbors than a small particle. Reasonable as
this may sound, it also brings some less physical features, such as total exclusion of large
particles from a solvent of small particles still not at close packing. As a result the former
linear mixture rule for the free volume factor has most often been preferred in the GvdW
applications. In the GvdW(HS-B2 ) form of the theory the correction term brings in the
binary type of excluded volume mixing rule. The free energy density functional becomes

Copyright © 2001 Marcel Dekker, Inc.

 90 Nordholm and Penfold
" ! #
c ð
X Xc  
ni ðrÞ 2 3
Fc ¼ kB T drni ðrÞ ln P þ 3
  jj nj ðrÞ
i¼1
1  cj¼1 nj ðrÞjj3 j¼1
3 ij
ð23Þ
c ð ð
1X c X
þ drni ðrÞ dsnj ðsÞ
sij ðjr  sjÞ
2 i¼1 j¼1

The corresponding GvdW(S) functional is obtained by dropping the second term in the
square brackets of the entropy part of the functional above.
The minimization of this functional presents a problem which for many component
mixtures can be quite timeconsuming if the truly optimal form of the interface and free
energy is to be found. One may use an iterative method of solution much like the famous
scheme used to solve for the Hartree–Fock wave function in electronic structure calcula-
tions [4]. An alternative, much to be preferred when sufficiently accurate, is to use a simple
parametrized form for the particle densities through the interface and then determine the
optimal values of these parameters. The simplest possible scheme is, of course, to take the
profile to be a step function,

ni ðxÞ ¼ nli for x < 0

ð24Þ
¼ ngi for x  0

This will be the form of the profile if we used a fully local free energy density functional in
our calculation. In the case of the HS-B2 -approximation the fully local functional would
be
" ! #
X c ð X c  
ni ðrÞ 2 3
Fc;loc ¼ kB T drni ðrÞ ln P þ   jj nj ðrÞ
3

i¼1
1  cj¼1 nj ðrÞjj3 j¼1
3 ij
ð ð25Þ
X c X c Xc ð
 aij drni ðrÞnj ðrÞ þ drni ðrÞVext ðrÞ
i¼1 j¼1 i¼1

Note that the external field Vext vanishes in the normal surface tension calculation. In the
fully local approximation there is no surface tension. Thus we can obtain the surface
tension  associated with a planar interface of area A by the expression

 ¼ Fc ðn1;opt ; . . . ; nc;opt Þ  Fc;loc ðn1;B ; . . . ; nc;B Þ =A ð26Þ

Here n1;opt ; . . . ; nc;opt are the correct particle densities obtained by optimizing the nonlocal
free energy, while n1;B ; . . . ; nc;B imply that bulk particle densities have been used on both
sides of a step function interface.
The main problem is to find the free energy of the real interface with nonlocal
energetic and entropic effects. For a general multicomponent interface the minimization
of the nonlocal HS-B2 -functional is a nontrivial numerical problem. Fortunately, the
variational nature of the problem lends itself to a stepwise solution where simple para-
metrization of the density profiles through the interface upon integration of the functional
yields the free energy as a function of the parameters. In fact, if we take the profile to be a
step function as in the case of local free energy then with local entropy we get the result

27 Xc X c
¼ a ðn  nig Þðnjl  njg Þ ð27Þ
64 i¼1 j¼1 ij il

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 91

which is the simplest estimate of the surface tension of a mixture. It will be an over-
estimate, since nonlocal entropy, smooth interfacial profiles and adsorption to the inter-
face are all surface-tension-reducing mechanisms. In order to approach the true surface
tension but keep the calculation simple we can use an extension to mixtures of the tanh
profile applied with great success to single component fluids. It may seem then that we
could take each component to be of tanh shape but this will not be recommended here. It
turns out that it is better to take the total particle density ntot and the mole fraction of each
component xi ðzÞ ¼ ni ðzÞ=ntot ðzÞ to be of tanh shape, i.e.,
ntot;l  ntot;g
ntot ðzÞ ¼ ntot;g þ ð28Þ
1 þ expðzÞ
xi;l  xi;g
xi ðzÞ ¼ xi;g þ i ¼ 1; . . . ; c ð29Þ
1 þ expðzÞ

The reason for this is subtle and subject to verification in specific cases but related to the
more fundamental nature of relative concentrations than component particle densities in
mixtures. If we use this tanhðn; xÞ scheme then we find, for example, that the volatile
component will naturally show adsorption to the interface. This is illustrated in Fig. 2.
A study of the surface tension of Lennard–Jones mixtures has recently been com-
pleted [17]. Very good agreement was obtained in comparison with both experiment and
simulation. In Fig. 3 we show results obtained by this tanhðn; xÞ scheme for an Ar–
methane mixture in comparison with simulation. We note that if we make the comparison
at the same reduced temperature T
¼ T=Tc , then there is quite reasonable agreement with
the surface tension obtained by simulation. The slight but significant deviation between
the critical temperatures in the simulation and in the HS-B2 approximation will otherwise
give rise to a similar deviation in surface tension. In the study of the Ar–Kr mixture we
found that neither the simulation nor the GvdW theory appear to be able to reproduce a
slight dip in the surface tension below linear interpolation as a function of composition
unless it is assumed that the Berthelot rule should be corrected by a factor slightly less than
unity, i.e., 0.97 in the case of the GvdW theory, which lowers the surface tension most for

FIG. 2 The tanhðn; xÞ profiles for a liquid–gas interface in a binary mixture are illustrated. The full
and dashed lines correspond to mole fractions x1 ðzÞ and x2 ðzÞ, while dash–dotted and dotted lines
correspond to particle densities n1 ðzÞ and n2 ðzÞ of a nonvolatile (1) and volatile (2) component. Note
that there is adsorption of the volatile component to the interface.

Copyright © 2001 Marcel Dekker, Inc.

 92 Nordholm and Penfold

FIG. 3 The calculated surface tension of an argon–methane mixture as function of composition

(mole fraction of argon) is shown as open circles in comparison with corresponding simulation
results obtained by Mecke, Winkelmann, and Fischer [J. Chem. Phys. 110:1188 (1999)]. The
GvdW(HS-B2 ) functional was used with tanhðn; xÞ profiles.

x ¼ 0:5 or so. Such corrections are commonly found in many other applications of the
Lorentz–Berthelot mixing rules.

IV. GVDW THEORY OF ELECTROLYTE SOLUTIONS

Electrolyte solutions have a theoretical description which is quite distinct from that of
Lennard–Jones fluids. The reason is that the latter are short-range simple fluids for which
the virial equation of state provides a good description of the onset of nonideality. The
ionic interactions are long-range and as a consequence electrolyte solutions are predomi-
nantly correlated fluids for which the virial equation of state is inapplicable. The second
virial coefficient is not definable for the Coulomb interaction, since its falloff with 1=r gives
it essentially infinite range. Instead the electrolyte solutions are governed by the screening
mechanism which neutralizes every charge by adsorption of counterions and desorption of
coions. The screening mechanism is a correlation mechanism. Thus the GvdW theory in its
very simplest form fails to capture the essence of a bulk electrolyte solution. Let us
consider first the primitive restricted model of an electrolyte solution wherein the ions
are hard spheres of equal size in which are embedded charges q1 and q2 , respectively, and
the solvent is accounted for only by a relative dielectric permittivity . If we were to apply
the HS-B2 -functional to the bulk electrolyte solution in this model we would simply end up
with a hard sphere fluid, since the uncorrelated motion of counterions and coions would
leave the charge density ðrÞ equal to zero everywhere around any one of the ions. Thus
the mean-field approximation, which ignores correlations among the ions and therefore
neglects charge screening, is not appropriate for bulk electrolytes.
It may seem that the prospects are bleak for the GvdW approach to electrolytes but,
in fact, the reverse is the case. We need only follow Debye and Hückel [18] into their
analysis of the screening mechanism, almost as successful as the van der Waals analysis of
short-range fluids, to see that the mean-field approximation can be applied to the correla-
tion mechanism with great advantage. In fact, we can then add finite ion size effects to the
analysis and thereby unify these two most successful traditional theories.

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 93

A. Recovering the Poisson–Boltzmann Equation

The GvdW density functional for a binary primitive electrolyte with the hard sphere
diameters set to zero will have the form
ð
Fðn1 ; n2 Þ ¼ kB T dr½n1 ðrÞ ln n1 ðrÞ þ n2 ðrÞ ln n2 ðrÞ
ð ð ð ð30Þ
1 1
þ drðrÞ dsðsÞ þ drext ðrÞðrÞ
2 40 jr  sj
Here the charge density is defined as
ðrÞ ¼ q1 n1 ðrÞ þ q2 n2 ðrÞ ð31Þ
and ext is an external electric field which is to be screened. This functional can be
minimized by the Lagrange multiplier method which readily yields the result
n1 ðrÞ ¼ n1B expðq1 ðrÞ=kB T Þ
ð32Þ
n2 ðrÞ ¼ n2B expðq2 ðrÞ=kB T Þ
The electrostatic field  is now the sum of the external field ext and an internal field
created by the ions,
ð
1
ðrÞ ¼ ext ðrÞ þ dsðsÞ ð33Þ
40 jr  sj
Recalling now the Poisson equation relating a charge density and a corresponding field we
get
r2 ðrÞ ¼ ðrÞ=0  ð34Þ
in the part of space where the ions make up the only charges. Finally, inserting the results
for the ion densities in ðrÞ on the right we get
1
r2 ðrÞ ¼  ½q n expðq1 ðrÞ=kB T Þ þ q2 n2B expðq2 ðrÞ=kB T Þ ð35Þ
0  1 1B
This is the famous Poisson–Boltzmann equation.

B. . . . and the Debye–Hückel Theory

In their analysis of electrolyte screening Debye and Hückel assumed that the screening
mechanism was weak, i.e.,
 
qi ðrÞ=kB T   1 ð36Þ
for all ions i and positions r. Then the exponentials in the Poisson–Boltzmann equation
can be expanded as
expðqi ðrÞ=kB T Þ ’ 1  qi ðrÞ=kB T ð37Þ
and the Poisson–Boltzmann equation takes the form
" #
1 q21 n1B þ q22 n2B
r ðrÞ ¼ 
2
ðrÞ ð38Þ
0  kB T

Defining the inverse screening length 0 by

Copyright © 2001 Marcel Dekker, Inc.

 94 Nordholm and Penfold

q21 n1B þ q22 n2B

20 ¼¼ ð39Þ
0 kB T
we have

r2 ðrÞ ¼ 20 ðrÞ ð40Þ

which is the basic equation of the Debye–Hückel theory of electrolyte screening.

C. Hole-Corrected Debye–Hückel Theory

We have seen above that GvdW theory yields Poisson–Boltzmann and Debye–Hückel
theory by neglect of ion size and excluded volume effects. Let us now consider how
such effects might be brought into a corrected form of Debye–Hückel theory. We note
that the ion size enters in two fundamentally different ways: (1) reducing the entropy by
the introduction of excluded volume and (2) altering the electrostatic energy by keeping
the ions from approaching closer than the relevant hard sphere diameter ij . We shall show
first that if we retain the restricted primitive model and take the charges to be symmetrical,
q1 ¼ q2 , then the excluded volume effect vanishes in the linearized Debye–Hückel limit.
This follows by reformulating the free energy in terms of total ion density nðrÞ ¼ n1 ðrÞ þ
n2 ðrÞ and charge density ðrÞ ¼ q1 n1 ðrÞ þ q2 n2 ðrÞ, which in the symmetrical case is equal to
q½n1 ðrÞ  n2 ðrÞ. In the linearized theory ðrÞ is decoupled from nðrÞ, since a linear response
to an electric field will preserve nðrÞ, changing the ion densities in opposite directions to the
same extent. Thus the excluded volume effect is absent. This is a great simplification and
undoubtedly a reason for the considerable success of the simple Debye–Hückel theory.

1. Local Hole Correction of the Electrostatic Energy

While the excluded volume effect does not affect the screening mechanism in the linearized
theory for symmetrical primitive electrolyte models, there is still an effect on the electro-
static energy. Note that in the Poisson–Boltzmann or Debye–Hückel theories the ions are
point particles without size. The Coulomb integrals do not account for the truncation at
r ¼  which is essential in the van der Waals calculation of binding energy. Putting this
truncation in the Coulomb integrals turns out to be a nontrivial task so we shall approach
it stepwise by first considering a local approximation [19] which is justified if 0   1, i.e.,
if the Debye length 1=0 is much greater than the ion size . In this limit the charge density
ðrÞ is essentially constant on the length scale of  and the hole correction to the electro-
static energy can be approximated as
ð ð ð ð
1 1 1 1
drðrÞ dsðsÞ ’ dr ðrÞ
2
ds
2 s< 4 0 jr  sj 2 s< 4 0 jr  sj
ð ð41Þ
2
¼ dr ðrÞ
2
40 
This is the electrostatic energy arising from ions approaching within  of each other. When
subtracted from the free energy functional above the corrected Debye–Hückel equation
becomes

r2 ðrÞ ¼ 2 ðrÞ ð42Þ

with  defined by

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 95
 
2 ¼ 20 = 1  20  2 =2 ð43Þ

Thus we have found that the screening should be moreqefficient

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
than in the Debye–Hückel
theory. The Debye length 1= is shorter by the factor 1  20  2 =2 due to the hard sphere
holes cut in the Coulomb integrals which reduce the repulsion associated with counterion
accumulation. A comparison with Monte Carlo simulation results [20] bears out this view
of the ion size effect [19].

2. Nonlocal Hole Correction of the Electrostatic Energy: CDH Theory

The hole correction of the electrostatic energy is a nonlocal mechanism just like the
excluded volume effect in the GvdW theory of simple fluids. We shall now consider the
charge density around a hard sphere ion in an electrolyte solution still represented in the
symmetrical primitive model. In order to account for this fact in the simplest way we shall
assume that the charge density i ðrÞ around an ion of type i maintains its simple expo-
nential form as obtained in the usual Debye–Hückel theory, i.e.,
8
< 2 q i 1 exp½ðr  Þ r > 
i ðrÞ ¼ 4ð1 þ Þ r ð44Þ
:
0 r 
The -value is, however, now to be determined by minimizing the free energy including
nonlocal corrections of the electrostatic energy. The integrals required are elementary but
large in number and one obtains finally a free energy

kB T 3 qi 
Fc;i ðÞ ¼ TS0 þ  ½2 þ 3 þ expðÞ ð45Þ
16nB ð1 þ Þ2 160  ð1 þ Þ2

wherein the first term is a background free energy of the bulk electrolyte solution, the
second term is related to the cost in entropy to create an exponential screening charge
density, and the third term is the electrostatic energy. The total particle density is denoted
nB , i.e., nB ¼ n1B þ n2B . The minimization yields a -value, which in turn allows the inter-
nal electrostatic energy to be calculated, and by Debye charging over coupling parameter 
one can readily obtain the osmotic coefficient
,

¼ P=nB kB T ¼ ðPð ¼ 0Þ þ
PÞ=nB kB T ð46Þ
and the activity coefficient  defined by the relation
X X
ln  ¼ xi ln i ¼ xi ð

i ð ¼ 0Þ þ

Di Þ=kB T ð47Þ
i i

Here

i ð ¼ 0Þ is the change in chemical potential of an ion of type i due to its hard
sphere size and

Di is the corresponding change due to its electrostatic interactions. The
results are compared with Monte Carlo results for 1-1 or 2-2 salt solutions by Van Megen
and Snook [20] in Figs. 4–9 and discussed in the original publication [21]. We note that the
1-1 data are very well reproduced up to more than 2M solutions. The 2-2 results show
significant deviations due mainly to a nonlinear response to strong electrostatic coupling
but the trends and magnitudes are well reproduced. The upturn in osmotic pressure and
activity are related mainly to the hard sphere excluded volume effect which in these
calculations by both the DH and the CDH theories was accounted for by the GvdW(I)
equation of state for hard spheres which yields

Copyright © 2001 Marcel Dekker, Inc.

 96 Nordholm and Penfold

FIG. 4 The calculated internal energy of a 1-1 salt (line) is compared with the corresponding
simulation results (open circles) obtained by Van Megen and Snook (Ref. 20). The Debye–Hückel
(DH, dashed line) and corrected Debye–Hückel (CDH, full line) theory were used together with a
GvdW(I) treatment of the uncharged hard-sphere mixture. The ion diameter was 4.25 Å, the tem-
perature was 298 K and the dielectric constant  was 78.36.

FIG. 5 As in Fig. 4 but the osmotic coefficient is shown.

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 97

FIG. 6 As in Fig. 4 but the logarithm of the mean activity coefficient is shown.

FIG. 7 As in Fig. 4 but the internal energy of a 2-2 salt is shown.

Copyright © 2001 Marcel Dekker, Inc.

 98 Nordholm and Penfold

FIG. 8 As in Fig. 4 but the osmotic coefficient of a 2-2 salt is shown.

FIG. 9 As in Fig. 4 but the logarithm of the mean activity coefficient for a 2-2 salt is shown.

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 99

X  
2
Fc ð ¼ 0Þ ¼ kB T Ni lnð1  niB  3 Þ  ln niB ð48Þ
i
3
" #
2 nB  3
Pð ¼ 0Þ ¼ nB kB T 1 þ ð49Þ
3 1  nB  3
" #
2 nB  3 3


1 ð ¼ 0Þ ¼ k T  lnð1  nB  Þ ð50Þ
3 B 1  nB  3

Thus we can see that a combination of van der Waals treatment of hard sphere excluded
volume and Debye–Hückel treatment of screening with a minor generalization to account
for hole correction of electrostatic interactions yields quite accurate bulk thermodynamic
data for symmetrical salt solutions.

3. Hole Correction of the Debye–Hückel Coion Density Profile: The DHH Theory
So far in our revision of the Debye–Hückel theory we have focused our attention on the
truncation of Coulomb integrals due to hard sphere holes formed around the ions. The
corresponding corrections have redefined the inverse Debye length  but not altered the
exponential form of the charge density. Now we shall take note of the fact that the
exponential form of the charge density cannot be maintained at high -values, since this
would imply a negative coion density for small separations. Recall that in the linear theory
for symmetrical primitive electrolyte models we have
ðrÞ ¼ q1 n1 ðrÞ þ q2 n2 ðrÞ ¼ q½n1 ðrÞ  n2 ðrÞ ð51Þ

and, if the central ion is of type 1 and q1 ¼ q > 0, then

8
< n  1 ðrÞ ¼ n  2 1 exp½ðr  Þ r>
n1 ðrÞ ¼
1B 2q 1B 8ð1 þ Þ r ð52Þ
:
0 r 

Now we see that if  is large enough and  small enough then n1 ðrÞ will become negative
for small r before the truncation arises at r ¼ . This is, of course, impossible and we must
either abandon the use of the linearized DH (Debye–Hückel) approach altogether or
eliminate the negative portion of the coion density profile.
It turns out that nature, as usual, is benign and allows us to retain the simple DH
form of profile and still get good accuracy for many strongly coupled electrolytes as long
as we remove the unphysical negative particle densities [22]. A particularly illustrative case
is offered by the so-called one-component plasma (OCP) model wherein one component is
made up of charged point particles and the other is a uniform neutralizing background
charge density. The Debye–Hückel theory now leads to the charge density

q20 1
ðrÞ ¼  expðrÞ ð53Þ
4 r
which will always produce an unphysical ion density

20 1
nðrÞ ¼ nB  expð0 rÞ ð54Þ
4 r
for sufficiently small r. Here the DH -value is given by

Copyright © 2001 Marcel Dekker, Inc.

 100 Nordholm and Penfold

nB q2 3
20 ¼ ¼ 2 ð55Þ
0 kB T a
where a ¼ ð3=4nB Þ1=3 and the coupling strength  is defined as q2 =ð40 kB TaÞ. The DH
theory then leads to the result for the electrostatic potential energy per ion u
u ð3Þ3=2
u^ ¼ ¼ ð56Þ
kB T 6
the electrostatic contribution to the free energy
2 ð3Þ 3=2
f^ ¼
F=NkB T ¼ u^ ¼  ð57Þ
3 9
and the osmotic coefficient
P 1 ð3Þ3=2

0 ¼ ¼ 1 þ u^ ¼ 1  ð58Þ
nB kB T 3 18
The accuracy of the DH results for u^ , f^, and
0 is good for low coupling strength,
 < 1, but for higher coupling strength the errors become very large. By correcting the ion
density so that it vanishes rather than going negative for small r while maintaining full
neutralization, i.e.,
8

< n 1  h exp½ ðr  hÞ r>h
nðrÞ ¼
B r 0
ð59Þ
:
0 r h
we get a greatly improved representation of the screening and the thermodynamic proper-
ties. The hole can be found to have the size h given by
1 h 1=3 i !  1
h¼ 1 þ ð3Þ3=2 1 ¼ ð60Þ
0 0
where we have introduced the auxiliary variable ! ¼ ð1 þ ð3Þ3=2 Þ1=3 . The thermodynamic
properties can then be readily found to be
1
u^ ¼  ð!2  1Þ ð61Þ
4
" !  #
1 ! 2
þ ! þ 1 2 !  1
f^ ¼  !  1  ln
2
þ pffiffiffi arctan pffiffiffi ð62Þ
4 3 3 3ð! þ 1Þ

1 2

¼1 ð!  1Þ ð63Þ
12
The results of the simple DHH theory outlined here are shown compared with DH
results and corresponding Monte Carlo results in Figs. 10–12. Clearly, the major error of
the DH theory has been accounted for. The OCP model is greatly idealized but the same
hole correction method can be applied to more realistic electrolyte models. In a series of
articles the DHH theory has been applied to a one-component plasma composed of
charged hard spheres [23], to local correlation correction of the screening of macroions
by counterions [24], and to the generation of correlated free energy density functionals for
electrolyte solutions [25,26]. The extensive results obtained bear out the hopeful view of
the DHH approximation provided by the OCP results shown here. It is noteworthy that in

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 101

FIG. 10 The calculated internal energy of a one-component plasma as a function of coupling

strength is compared with corresponding simulation results (open circles) by Brush, Sahlin, and
Teller (J. Chem. Phys. 45:2102 (1966). The Debye–Hückel (DH) and hole-corrected Debye–
Hückel (DHH) theories were used with results as shown (indicated lines).

FIG. 11 As in Fig. 10 but the osmotic coefficient is shown.

Copyright © 2001 Marcel Dekker, Inc.

 102 Nordholm and Penfold

FIG. 12 As in Fig. 10 but the reduced free energy is shown.

the case of hard sphere ions the electrostatic hole is at high coion density and weak
coupling preempted by the hard sphere hole. The latter were accounted for in the CDH
theory above as a general reduction of Coulomb interactions due to hard sphere exclusion.
The DHH theory brings in electrostatic hole formation between ions of the same charge.
In both cases we have applied simple correlation mechanisms in the spirit of the original
van der Waals analysis which applies, with some generalization, as well to electrolytes as
to simple fluids.

V. CONCLUSION

Our review of the generalized van der Waals theory has taken us from the simple cubic
equation of state, which is known to all chemists, to the surface tension problem, which
can be regarded as the most important nonuniform fluid property, to the electrolyte
screening mechanism in salt solutions. Along this route we went from bulk behavior
through the interfacial properties to the correlation dominated Coulomb fluids retaining
the simple physical van der Waals approach to fluids. We thereby hope to have demon-
strated that this analysis combining volume exclusion correlations with mean-field neglect
of correlations is capable of very fruitful generalization and extension to widespread
phenomena and very useful accuracy. Thus we have found within the confines of the
variational formulation of the GvdW theory not only the traditional van der Waals theory
but also the Poisson–Boltzmann and Debye–Hückel theories and all the applications of
these theories to fluids of both short range and long range interactions. Bulk thermody-
namic properties, adsorption, surface tension and screening phenomena can be simply and
efficiently addressed. A key role in the efficiency is played by the variational density
functional approach which permits simple parametrized density profiles to be used. This
is particularly well illustrated by the tanh profiles in the surface tension applications and

Copyright © 2001 Marcel Dekker, Inc.

 The GvdW Theory 103

by the use of Debye–Hückel charge densities with optimized -values in the electrolyte
applications.
Electrochemistry is, with its combination of simple fluids and ions, a natural area of
application for the GvdW theory. Here there is an obvious need to understand both the
short-range fluids, the long-range electrolyte solutions and their properties in bulk and at
interfaces. We hope this review can provide a useful tool for electrochemists and many
others with communal interests and stimulate further progress in a field rich in fluid
phenomena and mechanisms.

ACKNOWLEDGMENT

We would like to thank S. Sarman for help in the preparation of this review and the
Swedish Natural Science Research Council for financial support.

REFERENCES

1. J. D. van der Waals, Dissertation, Leiden, 1873.

2. A. Ya. Kipnis, B. E. Yavelov, and J. S. Rowlinson, Van der Waals and Molecular Science,
Clarendon Press, Oxford, 1996.
3. J. D. van der Waals. Z. Phys. Chem. (Leipzig) 13:657 (1894).
4. R. G. Parr and W. Yang, Density-Functional Methods of Atoms and Molecules, Oxford
University Press, New York, 1989.
5. A. D. Becke. J. Chem. Phys. 88:1053 (1988); ibid. 98:5648 (1993).
6. P. Tarazona. Mol. Phys. 52:81 (1984); Phys. Rev. A 31:2672 (1985).
7. W. A. Curtin and N. W. Ashcroft. Phys. Rev. A 32:2909 (1985).
8. S. Nordholm and A. D. J. Haymet. Aust. J. Chem. 30:2013 (1980).
9. S. Nordholm and J. Gibson. Aust. J. Chem. 34:2263 (1981).
10. S. Nordholm, M. Johnson, and B. C. Freasier. Aust. J. Chem. 33:2139 (1980).
11. B. C. Freasier, C. E. Woodward, and S. Nordholm. J. Chem. Phys. 90:5657 (1989).
12. S. Nordholm and R. Penfold. J. Chem. Phys. 96:3022 (1992).
13. M. A. Hooper and S. Nordholm. Aust. J. Chem. 34:1809 (1981).
14. S. Abbas and S. Nordholm. J. Colloid Interface Sci. 166:481 (1994).
15. S. Nordholm, P. R. Harrowell, and K. Cheung. Aust. J. Chem. 35:247 (1982).
16. P. R. Harrowell and S. Nordholm. Aust. J. Chem. 35:673 (1982).
17. H. Greberg, G. Paolini, J. Satherley, R. Penfold, and S. Nordholm. J. Colloid Interface Sci.
(submitted).
18. P. Debye and E. Hückel. Phys. Z. 24:185 (1923).
19. S. Nordholm. J. Chem. Phys. 78:5759 (1983).
20. W. Van Megen and I. Snook. Mol. Phys. 39:1043 (1980).
21. S. Nordholm. Aust. J. Chem. 37:1 (1984).
22. S. Nordholm. Chem. Phys. Lett. 105:302 (1984).
23. R. Penfold, S. Nordholm, B Jönsson, and C. E. Woodward. J. Chem. Phys. 92:1915 (1990).
24. R. Penfold, S. Nordholm, B. Jönsson, and C. E. Woodward. J. Chem. Phys. 95:2055 (1991).
25. R. Penfold and S. Nordholm. J. Chem. Phys. 96:3102 (1992).
26. R. Penfold, B. Jönsson, S. Nordholm, and N. Nichols. Phys. Rev. E (to appear).

Copyright © 2001 Marcel Dekker, Inc.