Molecular Theory of Surface Tension

Molecular theory of surface tension*
Arthur J. M. Yang, t Paul D. Fleming 111! and Julian H. Gibbs

Brown University, Department of Chemistry, Providence, Rhode Island 02912
(Received 15 August 1975)
A molecular theory of surface tension is developed for a liquid-gas interface of a one component system.
The Helmholtz free energy, the quantity minimized in the van der Waals approach, is obtained here from a
rigorous expansion in powers of derivatives of density p and is minimized by the calculus of variations. The
coefficient A(p) of the term in the square of the density gradient is (kT/6)fdr r 2C(r,p), C being the
direct correlation function. In the case in which p varies in one direction x only, the solution of the Euler-Lagrange
differential equation is analyzed in detail. This describes the cases of a single phase and of two coexisting
phases and leads to the equal area Maxwell construction. The effect of an external field on the solution is
discussed. The Euler-Lagrange differential equation provides a differential statement of Bernoulli's
theorem. In a three dimensional treatment the stress tensor formula is obtained from the corresponding
Euler-Lagrange partial differential equation. A (differential) generalization of the Young-Laplace
equation for a spherical interface is also derived. In addition, metastable regions are described and
interpreted. The stress tensor for a system in the presence of any kind of conservative force field is also
obtained.
I. INTRODUCTION of the theory in three dimensions and consequent es-

1
The goal of a molecular theory of liquid-vapor inter- tablishment of the connection between the van der Waals
faces is to elucidate the density profile in the interfa- and Laplace theories.
cial region and to evaluate the associated surface ten- We consider a one-component system at constant
sion. There have been two basic approaches to this temperature and constant volume. At equilibrium the
goal. The first, that of Laplace, involves a statistical density profile in such a system is that which mini-
mechanical calculation of the intermolecular forces in mizes the total Helmholtz free energy of the system
directions both normal and tangential to the interface. and is asymptotically constant in the interior of indi-
This was put into modern terms by Kirkwood and Buff vidual phases. We express the Helmholtz free energy
and by Buff. 3 Their expression for the surface tension as a definite integral of a local free energy density
was obtained in terms of the intermolecular potential which is a function of the molecular number density
and the pair correlation function. The second, the (and its derivatives). Then the determination of the
Maxwell-van der Waals theory, 4 utilizes a local free equilibrium molecular number density profile becomes
energy density consisting of a uniform term, evaluated a problem in the calculus of variations, i.e., that of
at the local mass density, and a nonuniform term, pro- determining the number density function which ex-
portional to the square of the mass density gradient. tremizes the integral subject to the constraint that the
By minimizing the total Helmholtz free energy of the system is closed. It is through this constraint that the
system, one obtains a simple formula for the surface chemical potential is introduced, as a Lagrange multi-
tension. plier.
Both approaches have been applied to the calculation
In Sec. II the general equations are introduced, the
of interfacial widths and tensions in the vicinity of the
actual microscopic derivation of these equations being
liquid-vapor critical point. Cahn and Hilliard 5 extended
deferred to the Appendix. The case in which the den-
the van der Waals theory to binary mixtures for the
sity varies in one dimension only is analyzed in detail
purpose of discussing the interfacial profile and tension
in this section. The Euler-Lagrange equation provided
near the critical solution temperature. More recent
by the calculus of variations is introduced, its solution
developments in the theory of critical phenomena have
being that which corresponds to the equilibrium number
been utilized by Fisk and Widom, 6 who employed a gen-
density function for the system at any given chemical
eralized van der Waals theory, and by Lovett7 et al.,
potential. From analysis of the one-dimensional dif-
who used both a van der Waals-type and a Laplace-type
ferential equation in Sec. II we obtain descriptions of
theory, in the discussion of liquid-vapor surface ten-
the cases of a single phase and of two coexisting phases
sion in the critical region. In addition, Triezenberg
and also the equal area Maxwell construction.
and Zwanzig8 have considered the van der Waals theory
from the point of view of fluctuation theory. In Sec. Ill the effects of external fields on the one-
dimensional solution are discussed. The differential
In this paper we shall be concerned with four aspects
equation for the pressure is interpreted as the condition
of the van der Waals theory: (1) a microscopic formu-
for mechanical equilibrium. It provides a differential
lation of the theory, in which the free energy is ob-
statement of Bernoulli's theorem (in the absence of
tained from a rigorous expansion in powers of density
flow).
derivatives; (2) discussion, for a case in which the
density varies in one direction only, of the solution of In Sec. IV the three-dimensional theory is discussed.
the (Euler-Lagrange) differential equation which arises As in the one-dimensional case, an Euler-Lagrange
in the theory; (3) incorporation of the effect of an ex- differential equation (here a partial differential equa-
ternal field (e. g. , a gravitational field); (4) treatment tion) is a statement of mechanical equilibrium, i.e. ,
3732 The Journal of Chemical Physics, Vol. 64, No.9, 1 May 1976 Copyright© 1976 American Institute of Physics
Downloaded 10 Feb 2005 to 141.218.142.192. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
Yang, Fleming, and Gibbs: Molecular theory of surface tension 3733
p(x)
F= J d 3rzjJ(r). (II. 1)
When the molecular number density p(r) is sufficient-

ly slowly varying in its spatial dependence on a scale
set by the interaction range, we find that zp(r) is given
by
zp(r)= U(r)p(r)+ zp(p(r))+i A(p(r))Vp(r) 2 • (II. 2)
U(r) is an external potential energy function and zp(p) is
the free energy density of an equilibrium uniform sys-
tem having density p. A(p) is related to the direct cor-
FIG. 1. Density profile for a pure system in the neighborhood
of the gas-liquid interface. relation function C(r, p) of a uniform system having den-
sity p by
V • u= f, where u is the stress tensor of the system and

A(p) = 6kT f d 3r~C(r, p) • (II. 3)
f is the force density. The surface tension emerges as In this section we shall only be concerned with flat
the difference between two components of the stress interfaces (i.e., those associated with a density vary-
tensor, one parallel to the direction of the density gra- ing only in a single linear direction). We will return to
dient and the other normal to it. The spherically sym- discussion of the general equations in Sec. IV. Thus,
metric solution of the three-dimensional partial differ- when the density is a function of x only, (II. 1) becomes
ential equation is a generalized Young-Laplace 9 equa-
tion for a spherical interface. The actual differential
equation is interpretated as a differential form of the
F=a r a
dxzp(x) , (II. 4)
Young-Laplace equation. The analysis of the radial where

part of the differential equation provides interpretation
of metastable states of the system. The three-dimen- zp(x) = U(x)p(x) + zp(p(x) )+ iA(p(x) )p'(x) 2 •
sional treatment also shows that an infinitely large sys-
tem will have a flat interface (radius of curvature ap-
a is the cross sectional area of the vessel, and p' (x)
denotes the derivative of the density with respect to x.
proaching zero), thus reducing the problem to the one-
dimensional situation discussed in Sec. II. This clari- For the remainder of this section we will assume that
fies the ambiguity of treating the system as having a U(x) vanishes. The effects of external fields will be
uniform density in two directions (one-dimensional discussed in Sec. III.
problem) without referring to the effect of gravity. We wish to determine the density function, p(x),
There are four appendices. Appendices A and B, which minimizes the definite integral (II. 4). The whole
which amplify material of the principal sections, are system is closed with total volume equal to V (fixed by
referred to appropriately in these sections. Appendi- the boundaries a, b) and with total number of particles
ces C and D, however, are not closely coupled to the equal toN. (The fixed volume is the basis for working
main sections and are not referred to again. They deal with the Helmholtz free energy rather than Gibbs free
with the critical exponent equality originally suggested energy). Since the system is closed, there is one con-
by Widom. 10 straint on the density function, i.e. ,
II. DENSITY VARIATION IN ONE DIMENSION; THE Ia

b
ap(x) dx =N. (II. 5)
FLAT INTERFACE
This makes the problem one of a constrained variation.
In a one-component system, below its critical tem- Mathematically we can remove the constraint by intro-
perature, liquid and vapor phases can coexist in equi- ducing a Lagrange multiplier. Thus we minimize the
librium. In the presence of gravity the interface sepa- integral
rating the two phases is perpendicular to the direction
of gravity, taken here as the x axis, and the density of
the system varies only in this direction x (Fig. 1). In I=
I a
b
dx[zp(x)- ~p(x)]=- JP
a
b
0 (x) (II. 6)
the two-phase region a density profile such as that
shown in Fig, 1 will render the total free energy of the We choose the symbol ~ because, not surprisingly, the
system a minimum. Lagrange multiplier corresponding to (II. 5) proves to
be the chemical potential.
The determination of such a molecular number den-
sity (or mass density) profile is best formulated in The minimization of (II. 6) is a problem in the calcu-
terms of a local free energy density. The same is true lus of variations. Thus the density profile must be a
of density functions which vary in more than one dimen- solution of the Euler-Lagrange equation
sion. In Appendix A we show that a fixed temperature
8P0(x) d 8P0 (x)
T the Helmholtz free energy of a nonuniform system can
be expressed as the definite integral of a local free en-
ap(x) - dx ap'(x) =0 · (II. 7)
ergy density, Evaluation of the partial derivatives in (11. 7) yields
J. Chem. Phys., Vol. 64, No.9, 1 May 1976
3734 Yang, Fleming, and Gibbs: Molecular theory of surface tension
JJ.(P) g,
FIG. 2. A typical chemical

potential isotherm for a one-
component system.
Po
!J. =!J.(p(x))- A(p(x))p"(x) -!A'(p(x))p'(x) 2 , (II. 8)

where
A'( )= M(p) ~----------------~X

p Cip ' (a) (b)
and FIG. 4. (a) g 1(p) for a chemical potential for whichll(p)-11 has
only one zero at Po • (b) Density profiles for solutions which
!J.(p) = a::) • (II, 9) approach Po asymptotically.
is the chemical potential of a uniform system having

phase.
density p. The derivative terms give corrections to !J.
for a nonuniform system. An equation similar to (II. 9) In order to analyze the solutions to (II. 8), it is use-
[actually, closer to our three-dimensional equation ful to manipulate the equation. We multiply both sides
(IV. 3)] has been obtained by Lebowitz and Percus. 11 of Eq. (II. 8) by p' to obtain
They found the coefficient of the second derivative term 2
to be given by (II. 3) but did not show that the coeffi- ( ( )) } '( ) _ d(! Ap'(x)
{ !J.pX-!J.pX- dx
)
' (II. 10)
cient of the derivative squared term is A'(p). The
proof that this is indeed the case is provided in Appen- or, since p' dx =dp, we have
dix A.
Equation (II. 8) has no explicit dependence on x. One
!Ap' 2 =1p {!J.'(p)-J.L}dp+Ct=gt(p),
p
(II. 11)
interesting property of the solution of such a differen-
where C1 is an integration constant and the lower limit
tial equation is that, if p(x) is a solution of the differ-
ential equation, then p(x+ c) is also a solution, where c
p is unspecified.
is any arbitrary constant. Since the boundaries of the The physical solution of the differential equation
system are fixed at a, b, the transformation from p(x) should have p' and p" approaching zero at the bound-
to p(x+ c) in the case where a phase boundary exists is aries (single phase interiors). In view of Eq. (II. 8),
equivalent to moving the phase boundary (the interface) this means we must also require !J.(p)- !J. to approach
in the direction of minus x by c, and therefore is also zero at the boundaries. Equivalently, we can require
equivalent to increasing the amount of substance in one both Ap' 2 and d(Ap' 2 )/dp, which equals !J.(p)- 11-, to ap-
phase and decreasing the amount in the other. This is proach zero at the boundaries. In view of (ll. 11) we
the familiar statement that the phase equilibrium is in- can achieve this by choosing a value for C1 such that the
dependent of the actual amount of substance in each zero of Ap' 2 falls at the extremum of the plot of ! Ap' 2
(g1(p)) vs p, where d(! A' 2 )/dp =0. A typical chemical
potential vs density isotherm (for temperatures below
JL(Pl-JL 1 the critical temperatures) will have a shape such as that
tcJ
shown in Fig, 2, the van der Waals loop being an ana-
lytic continuation6 • 12 of single phase free energies. M
will be determined by fixing the average density of the
system N/V. Different values of !J. cause !J.(p)- 11- to
I have one, two, or three zeroes (Figs. 3(a)-3(c)].
(a) We discuss the three different cases in turn:
Case A: J.L(p)- J.L has only one zero at Po (Fig. 3(a)].
JL(Pl-JLa
The plot of g 1(p) vs p will have only one extremum at
p0 (Fig. 4(a)]. Therefore, there is only one value of p
which our solution can approach asymptotically. The
plots of several solutions are shown in Fig. 4(b). Since
the physical solution must satisfy the boundary condi-
tion at both ends, (p' approaches zero at both ends), the
(b) (c)
only physical solution in this case is a constant solution,
FIG. 3. !l(p)-11 for three different values of 11· For 11 = 111 , 112, p =p 0 , the one phase state,
or 113 , !l(p)-11 has (a) one, (b) two, or (c) three zeroes, re-
spectively. Case B: J.L(p)- J.L has two zeroes at p1 and p2 (Fig.
J. Chern. Phys., Vol. 64, No. 9, 1 May 1976
g. (ll. 12)
At fixed temperature there is only one J.1., the coexis-

tence chemical potential, which can achieve this goal.
Equation (ll. 12) is the well known equal area construc-
tion13 of Maxwell [Fig. 7(c)]. When we recall that J.J.(p)
= IJI'(p) and that J.1. = J.J.(PLo) = J.J.(pc 0 ) we see that (II 12) can
be rewritten as
PL
0=
I
PG
dp[J.J.(p)- J.J.]
= [¢(PL 0) - J.J.(PL 0 )PL 0 ] - [¢(Pc 0 ) - J.J.(Pc 0 )Pc 0 ] • (ll. 13)

Equation (ll. 13) says that the equilibrium pressure
(P(p) =pJ.J.(p)- ljl(p)) is the same in the coexisting phases.
This is the physical content of the Maxwell construction.
In the above discussion the pressure appeared indi-
rectly and was identified only in the final equation
(II. 13), We can see more directly how the pressure
enters the theory when we note that (ll. 10) can be writ-
ten as
(ll. 14)
Thus we have
(b) P = J.J.p(x)- ljl(p(x))+ i A(p(x))p'(x)2 =canst. (II. 15)

The quantity Pis clearly the pressure since in a single
FIG. 5. (a) g! (p) can be made to be zero at p 1 (left) or at p 2
(right). (b) Density profiles for unphysical solutions which
approach Pt or P2 • g. g.
3(b)].
The plot of g 1 vs p will have one minimum at p 1 and a
point of inflection (with both first and second deriva-
tives of i Ap' 2 equal to zero) at p2 • With only one con-
stant to fix, the zero can only occur at either p 1 or p2
[Fig. 5(a)]. The physical solution in this case will be
either p =p 1 or p =p 2 , depending on the choice of inte-
gration constant, However, only the density p 1 corre-
sponds to a stable state, since the compressibility is
infinite for p = p2 • Thus again p = p 1 is the only physical p<xl
solution [other solutions of the differential equation are
sketched in Fig. 5(b)].
Case C: J.J.(p)- J.1. has three zeroes at Pc, p1 , PL> re-
spectively [Fig. 3(c)].
The plot of g 1 vs p will have two minima and one
maximum [Fig. 6(a)]. Again the integral will be zero
for only one of the three densities Pc, p1 , PL [Fig.
6(b)]. Either Pc or PL may be the stable solution, the
choice depending on which corresponds to the absolute
minimum of the integral. The other solution corre-
sponds to a metastable state. The solution correspond-
ing to p 1 is unstable, the compressibility being negative.
However, if the two minima are at the same height, we (b)
can achieve zeroes at pc 0 and PLo simultaneously [Fig.
FIG. 6. (a) g 1(p) showing two minima at Pc and PL· For differ-
7(a)]. This is the physical solution for the two-phase ent choices of integration constant g 1(p) can be made to be
state [Fig. 7(b)]. The condition for having two minima zero at either Pc (left) or PL (right). (b) Density profiles for
at the same height is unphysical solutions which approach either PL or Pc·
plxl J.L- U(x) = J.L(p(x)) -A(p(x))p"(x)

g,
- i A'(p(x))p'(x) 2 • (III. 1)
1-----""""----- P... When (III. 1) is multiplied by p'(x) we obtain
dP(x) = _ dU(x) p(x) =f(x) (IlL 2)

dx dx '
1---------+---Px. where
P(x) = J.Lp(x)- 1/J(x) + A(p(x))p'(x) 2
1----------=-Po, = J.Lp(x)- 1/J(p)- U(x)p(x) + i A(p(x))p'(x) 2 •
.___........
~..!--~~-~P ....___ _ _ _ _ _ _ ~ X Equation (III. 2) is the statement of mechanical equilib-
rium for a system in an external field. It says that the
(b) internal force, dP(x)/dx, is balanced by the external
force f(x).
JJ.(P)- JL,.u
This equation is also a differential statement of the
Bernoulli theorem. In the liquid phase region the den-
sity p(x) is essentially a constant ("incompressible"),
and (III. 2) implies that
P(x) + U(x)p = P =con st. (III. 3)
This is the usual statement of Bernoulli's theorem for
an incompressible fluid (in the absence of flow) in an
external potentiaL U(x)p is the potential energy density
at point x.
(c) When the spatial variations of p(x) are important, the
FIG. 7. (a) g 1 (p)whenJ.t is chosen so that the two minima are at differential form (III. 2) must be retained. The differ-
the same height. (b) The density profile for the physical two- ential form can be conveniently rewritten as
phase solutions. (c) Maxwell equal area construction.
dP(x)+p(x)dU(x)=O. (III. 4)
The integrated form is
phase interior, where p'- 0 and J.L(p)- J.L, it is
P = P(p) = J.L(p)p- 1/J(p) • (IL 16)
P(x)+ ix
xo
dx ddU(x) p(x)=P(x0 ) .
X
(III. 5)
Equation (II. 14) is simply the statement that the system The most important external potential for real fluid
is in mechanical equilibrium in the absence of external systems is that due to gravity, which for sufficiently
forces. In Sec. III we will see that (II. 14) is appropri- small xis
ately altered in the presence of an external field. GMm -GMm
U(x)=- R--:::: R +mgx, (IlL 6)
We note that when (II. 15) is integrated over the total o+X o
volume of the system, it can be written as where G is the gravitational constant, M is the mass of
F = J.LN- P V + a0 a, (II. 17) the earth, m is the molecular mass, R 0 is the distance
of some reference point (e. g., the center of mass of the
where the surface tension a0 is given by system or the Gibbs surface) from the center of the
earth. xis the distance from the reference point and
a0=Jba dxA(p(x))p'(x) 2
• (II. 18)
g=GM/R~ (III. 7)
Thus we see that, even though we have not treated is the gravitational acceleration at distance R 0 •
the interface as a true discontinuity, the surface con-
tribution to the free energy (i.e., the contribution The differential equation (III. 1) becomes
which grows only as the cross sectional area rather J.L + GMm/R 0 - mgx = J.L(p(x))- A(p(x))p"(x)
than as the volume) emerges in a natural way. Thus a
unique expression for the surface tension is obtained
- i A'(p(x))p'(x)2 • (III. 8)
even though the density is everywhere continuous. Thus, as is well known, whenever a specific reference
point in the fluid is chosen, the chemical potential
Ill. EXTERNAL FIELDS changes whenever the system is raised or lowered, We
will return to this point in a subsequent paper, in which
In this section we consider the effects of an external
we shall discuss the properties of the solution of Eq.
field on the density profile. We again treat only density
(III. 8) in detail.
variations in one dimension. When the potential energy
term in (II. 4) is retained, the differential equation The differential equation for the pressure in the
takes the form gravitational field is
J. Chem. Phys., Vol. 64, No. 9, 1 May 1976
dP(x)/dx=- mp(x)g, (ill.9) the condition for mechanical equilibrium. The internal
force V • a{r) precisely balances the external force f(r).
or equivalently
The expression for the stress tensor allows us to uti-
dP(x)+ mp(x)gdx= 0 , (ill.10)
lize an analysis of Laplace type to obtain an expression
where for the surface tension. We define a unit vector which
is everywhere in the direction of the density gradient
P(x) = (j:L- mgx)p(x) -1/'(p)+ t A(p(x))p'(x) 2 •
A Vp(r)
The integrated form is n(r)= !Vp(r)l · (IV. 5)
P(x)+rnJ"' dxp(x)g=P(O). (ill.ll) n

In the interfacial region clearly becomes the unit nor-
a mal to the interfacial "surface." The stress tensor can
This is the correct statement of Bernoulli's theorem. be written in terms of as n
It says that the pressure at height x is diminished (in-
a (r) =P0(r)l + a(r)n(r)n(r) , (IV. 6)
creased if x < 0) by the total weight per unit area of fluid
between the points zero and x. where
We shall note that in the presence of gravity even the a(r) =A(p(r))(Vp(r))2 •
pure bulk phase is not homogeneous [i.e., p'(x) *0], Thus we see that a(r), the difference between the
although it is usually treated as such (since (p' /p)X« 1, stress normal to the interface and the stress tangential
where X is the height of the sample). Thus for a single to the interface, is a surface tension per unit length.
bulk phase in a gravitation field there is a nonzero con-
The surface tension at a particular point r 0 on the inter-
tribution to the "tension" given in (II. 19). This contri- face is obtained by integrating a(r) along the line which
bution, however, is in general negligible in comparison is everywhere parallel to n(r) and which passes through
with the interfacial contribution (except in the neighbor- "r0 . " Thus
hood of a critical point).
IV. THREE-DIMENSIONAL THEORY aro = Jdrna(r) ' (IV. 7)

where
The simple one-dimensional treatment in Sec. II is
physically reasonable only if we consider it as applying rn=r• n(r).
to a system in an infinitesimal but nonzero gravitation-
This is just the result which would have been obtained
al field, for, in the total absence of gravity, a gas-liq-
from the Laplace approach. This expression can be
uid system will not form a flat interface. Instead, it
shown to be equivalent to the Buff2 expression for the
will form a spherical interface, as dictated by spatial
surface tension when the molecules interact pairwise.
isotropy. Thus, one phase will be a sphere imbedded
This follows from the relationship between the direct
in the other phase. In this section we discuss this is-
correlation function and the pair correlation function.
sue in terms of the three-dimensional theory. 14
Our formulation makes no restriction to pairwise addi-
We consider the variation of the functional tive interactions and should therefore be generally valid
for classical fluids (so long as the interaction is short
(IV. 1) ranged).
We must remember that amongst the set of zero field
where
solutions to the partial differential equation (IV. 3),
P 0(r) = !J-p(r) -1/'(r) . there is only one solution corresponding to the equilib-
rium state. This is the solution which makes the def-
The requirement that the integral (IV. 1) be stationary
inite integral of Helmholtz free energy density an ab-
leads to a three-dimensional Euler-Lagrange equation
solute minimum. Among the remaining solutions, some
ap0(r) _ V. 8P0(r) =0 (IV. 2) correspond to local minima (metastable states), while
ap(r) avp(r) • others correspond to local maxima or points of inflec-
Thus we have tion (unstable states). If we could solve the differential
equation exactly, we could substitute the solutions into
IJ-(p) -11- + U=AV 2p+t A'(Vp) 2 • (IV. 3) the integral and determine which is the equilibrium so-
lution and which are metastable solutions. Since we
We show in Appendix A that this differential equation
cannot solve the partial differential equation, we must
can be derived by direct expansion of the chemical po-
examine the properties of the physical solution and de-
tential in powers of Vp(r). The three-dimensional gen-
eralization of (ill. 2) is obtained when we multiply (IV. 3) termine whether, in fact, such a solution exists.
by Vp(r t The equation can then be written as In the absence of gravity, for a two-phase system, we
V• a=-pVU=f,
know that one phase should be a sphere (or system of
(IV. 4)
spheres) imbedded in the second phase. The density
where function will approach one constant inside the spherical
interface and another constant outside the interface.
a= {11-P -1/J(p)- pU} l +A{VpVp -t(Vp)2 1}
Knowing this, we can divide space into several regions
is the stress tensor. 15 This is the general statement of such that each region contains only one sphere. Since
fL!Pl-fL p!r)
~~,------------~
~~---------------Pt:
FIG. 8. The physi-

cal three-dimen-
sional solution given
as superposition of
spherically sym-
metric solutions.
1 - - - - - - - - - P.
~o--------+ r
FIG. 10. Isotherm (left) and density profile (right) when the
radius of the sphere is small.
the (asymptotic) densities at the boundaries of the vari- The physical solution should also approach another
ous regions are the same, the solutions for these regions constant as r approaches 0 (the center of the sphere).
join continuously at the boundaries. We m'ay therefore We need to know whether or not this constant is another
assume that the true solution is the sum of the solutions zero of IJ.(p)- IJ.. We examine the behavior of the solu-
in the different regions (Fig. 8). In each region we can tion in the neighborhood of r=O. Equation (IV.9) is sin-
assume the solution of the three-dimensional partial gular at r= 0. Therefore, if p' remains nonzero as r
differential equation to be spherically symmetric: approaches zero, p" will diverge, and the solution will
be singular there. Therefore, for a bounded solution,
p(r, IJ, ¢) =p(r) • (IV. 8) p' must approach zero as r approaches zero. Since the
Substituting this into the partial differential equation we differential equation is second order, the solution will
obtain a second-order ordinary differential equation for be determined by the initial conditions p(r0 ) and p'(r0 ). 16
p which contains r explicitly, We choose the reference point as r 0 = 0, where p'(O)
equals zero. Then, once we determine the value for p
IJ.(p)- 1J. =i A'(p)p'(d+Ap"(r)+ (2/r)A(p)p'(r) • (IV. 9) at r= 0, p(O), the whole curve is uniquely determined.
Ap' /r has a removable singularity at r= 0, since p'- 0
Comparing Eq. (IV. 9) with Eq. (II. 8), we notice that
as r-0. We can find the limit of Ap' /r as r-0 by
Eq. (IV. 9) has the additional term, (2/r)Ap'. This
L'Hospital's rule,
term makes the analysis more complicated than that in
Sec. II. lim Ap' = lim d(Ap') . (IV. 10)
In analyzing the solution, we first examine the behav- r•O r r•O dr
ior of the solution as r-oo and r- 0. As r gets large, As r- 0 the differential equation (IV. 9) becomes
the contribution from the term Ap' / r becomes smalL
(IV. 11)
Then the solution will behave similarly to the solution in
the one-dimensional treatment. Thus, one of the 2
where A'p' +Ap" is limr.a,p•·o(Ap' /r). Since p'- 0 as
boundary conditions is that p( r) approaches Po asymp- r-0, we have from Eq. (IV. 11)
totically as r-oo, where p0 is a zero of IJ.(p)- IJ.,
limp" =t (IJ.(p)- iJ.)/A • (IV. 12)
r•O
Since A is positive, for IJ.(p)- 1J. > 0 p(r) is concave up-

p!r) ward in the neighborhood of r= 0 and for IJ.(p)- ll < 0 it
fL!P)-fL > 0 is concave downward {Fig. 9).
In most cases p" will be very small as r- 0 (nonos-
cillatory solution). From Eq. (IV. 11) we know that
Ap" =HIJ.(p)- !J.) as r-0. Therefore, t (IJ.{p)- !J.) will
FIG. 9. The be-
be small as r- 0. This implies that as r- 0 the density
fL(P)-J-L <0 havior of solutions
will asymptotically approach a constant which is very
1 - - - - - - - - - - - Pt: in the neighborhood
of r= 0 for different nearly one root of IJ.{p)- 1J. = 0 (either PL or Pa). But in
values of p(O). the extreme case shown in Fig. 10, the density in the
neighborhood of r= 0 may have a value which is very
different from Pc or PL, and the radius of the sphere
will be very small.
In the discussion in Sec. II, we have seen that even
fL!P)-J-L< 0 when IJ.(p)- IJ. has three zeroes, there will not be a two-
~--------------------~r
0 phase system unless 1J. is the coexistence chemical po-
p(r)
~----------------~~=====PL
(a)
0~----------------~r
FIG. 11. Physical solutions for different values of p(O).

(b)
FIG. 13. Isotherm (a) and density profile (b) for a liquid drop.
tential, JJ.mu> which satisfies the Maxwell equal area

construction. The reason is that the relation obtained f(2Ap' /r) dp is determined by the sign of p'. When
from the differential equation (II. 11) enabled us to write
IPL(JJ.(p)-JJ.)dp>O (JJ.<!J.max), (IV.17)
!Ap'2l+oo-!Ap'21.oo= IPL(JJ.(p)-JJ.)dp' (IV. 13) PG
PG
we know from Eq. (IV, 16) that f(2Ap' /r) dp> 0 for a
where PL and Pa are two roots of JJ.(p)- JJ. = 0. If physical solution, and therefore p' must be greater than
0, This means that the density increases as r in-
(IV. 14) creases, which indicates that the low density phase (gas
phase) is inside the sphere and the high density phase
then! Ap'2 can only be fixed to be zero at one end [Fig. (liquid phase) is outside the sphere. This corresponds
6(b)]. The presence of the additional term Ap' /r in the to a gas bubble in a uniform liquid phase (Fig. 12). By
radial differential equation (IV. 9) allows us to find a the same argument we see that when
physical solution for JJ. unequal to JJ.max by choosing p"
at r =0 [or p(O)] in such a way that the integral f(2Ap' I
r) dp compensates the value of (IV. 18)
IPL (JJ.(p)- JJ.) dp (Fig. 11) ' (IV. 15) p' will be negative. This corresponds to a liquid drop
PG in a uniform gas phase (Fig, 13),
i.e.' Since A is small, the principal contribution of
PL iPL 2Ap1 f(2Ap' /r) dp comes from the region where p' is signifi-
!Ap' 2
1
PL
-iAp' 2
1
PG I
=
PG
(JJ.(p)-JJ.)dp-
PG
--dp=O.
r cantly large, i, e., the interfacial region. (In the re-
gion where r is very small, L'Hospital's rule gave
(IV, 16)
Ap' /r-Ap", which is a small quantity for a nonoscil-
The dependence of the solution on the chemical poten-
lating solution). The value of f(2Ap' /r) dp then depends
tial can be seen from Eq. (IV. 16). Both A and rare
on the position of the interface on the r axis, i.e., the
positive quantities, and p' will always have the same
radius of the sphere r 0 • Therefore, by (IV, 16), the
sign between PL and Pa· Thus, the sign of the integral
smaller I J:<f(JJ.(p)- JJ.) dp I is, the greater r 0 will be.
In fact, as JJ.- IJ.max>
JLIP)-JL pCr)
(IV. 19)
and r 0 must approach infinity, so that f(Ap' /r) dp- 0.

This removes the apparent ambiguity in the one-dimen-
sional treatment, where we assumed the system had a
flat interface without reference to gravity. From the
analysis of the problem in three dimensions, we can
see that the one-dimensional treatment is a special
case where one phase forms a sphere of infinite radius.
The surface of an infinite sphere has zero curvature and
is therefore flat,
0~--------------------~r
We now derive an equation for the pressure variation
(b) across a spherical interface. We integrate (IV. 9) over
FIG. 12. Isotherm (a) and density profile (b) for a gas bubble. the interfacial region to obtain
{i/J(p)- f.1.P- ~ Ap'

2
} outside Waals free energy and differential equation arose nat-
urally from a rigorous expansion of the free energy and
chemical potential in powers of density derivatives,
(IV. 20) The expression for the full three-dimensional stress
tensor made possible the identification of the relation-
We recognize that- i/J(p)+ f.1.P+~ Ap' 2 is just the pres- ship between the van der Waals and Laplace theories.
sure obtained in Sec. II [Eq, (II. 16)]. Thus we can re- For a fluid of pairwise-interacting molecules, the
write Eq. (IV. 20) as choice between the two approaches is merely one of
treating the direct correlation function as the fundamen-
P Inside - P outside= J~ Ap' dr ,
2
(IV. 21)
tal correlation, as in the van der Waals theory, or
treating the pair correlation function as the fundamental
correlation, as in the Laplace theory. Since either
If we expand 1/r around r 0 , the radius of the inter- correlation function determines the other, results
face (e. g., the radius at which Ap' 2 is maximum), we obtained from the two approaches must be identi-
obtain caL
Plnslde -Poutslde=...!_
ro
J 2
2Ap' dr- 2 r-ro
ro
2
A p ,zdr, For a flat interface, the requirement that the density
asymptotically approach single phase densities on ei-
ther side of the interface led to the necessary introduc-
(IV. 22) tion of the Maxwell equal area construction for the
That this is a generalization of the Young-Laplace chemical potential as a function of the density. The
equation is easily seen, If the interfacial thickness is equal area condition requires that the equilibrium pres-
much smaller than the radius of the sphere r 0 , we can sure be the same in the two phases (usual statement of
drop the second term on the right hand side of Eq. mechanical equilibrium).
(IV. 22) and obtain
In the presence of an external field the pressure has
P inside - Pouts ide = - 2
ro
f Ap 12 dr , (IV. 23)
been rendered a function of position, The density is a
function of position even in a single bulk phase. This
leads to the surprising conclusion that even a single
Using the definition of the surface tension obtained in bulk phase has a nonvanishing "surface" tension in the
Sec, II [Eq. (II, 19)], we obtain the ordinary Young- presence of an external field such as gravity. This
Laplace equation for a spherical interface, contribution is generally negligible compared with an
interfacial contribution except in the critical region.
(IV. 24) This may be one source of the well-known difficulty in
identifying the critical point in the presence of grav-
Equation (IV. 24) is the usual statement of mechanical ity.
equilibrium for a spherical surface. The differential
equation (IV. 9) can be written as For a spherical interface the theory has provided a
differential form of the Young-Laplace equation for the
2
dP/dr=- 2a/r. (IV. 25) pressure variation across a spherical interface. The
generalization of this equation to arbitrarily curved in-
This is a differential statement of the Young-Laplace terfaces was seen in Appendix B to arise naturally from
equation. It is valid even when its integrated form, the differential equation for the stress tensor. The
(IV. 24), is not. In Appendix B the general forms of physical spherically symmetric solutions provide a
(IV. 24) and (IV. 25) for interfaces of arbitrary curva- thermodynamic interpretation of nucleation of a stable
ture will be obtained. internal phase from the metastable external phase,
The one-dimensional (flat interface) solution has been
We have obtained an interpretation of metastable seen to arise as the limit of a spherical interface (ei-
states. At a given temperature a metastable phase ther vapor internal or liquid internal) of infinite ra-
(liquid or vapor) can be an external phase in mechanical dius.
equilibrium with a spherical bubble or drop of a stable
phase. The radius of the sphere which must satisfy the A final note of caution is in order here. Although we
mechanical equilibrium condition is the Young-Laplace have obtained the local free energy (II. 2) from a rigor-
radius defined by (IV. 24). ous expansion of the exact free energy, (A. 20), the ex-
pansion cannot converge in general [e. g., the moment
As is well known, 17 this Young-Laplace radius pro- f d3rr4 C(r) does not exist for any potential behaving as
vides a critical nucleation size for a given tempera- 1/ r 6 for large r]. The expansion can at best only be
ture. asymptotic, but, as pointed out by Widom, 12 may nev-
ertheless yield a numerically accurate approxima-
tion.
V. DISCUSSION
We have provided a microscopic formulation of the ACKNOWLEDGMENT
van der Waals theory of surface tension. The van der We thank Dr. Alan Dunker for helpful discussions.
J. Chern. Phys., Vol. 64, No.9, 1 May 1976
APPENDIX A: LOCAL FREE ENERGY DENSITY functional F', 19 Indeed, (A5) is the Legendre transfor-
mation which transforms from the potential u(r) as a
In this Appendix we derive the local free energy den-
variable to the density p(r) as a variable. Thus, if we
sity from microscopic theory. The derivation is simi-
can express the one-body potential u(r) as a functional
lar to that given by Lebowitz and Percus. 11 Our basic
of p(r), we can obtain F' by functional integration from
approach is taken from a rigorous treatment of static
a known reference point.
properties of classical fluids given by Martin. 18
For a system with no interaction, the density p(r) is
We assume that the system is described by a classi-
given by
cal Hamiltonian of the form
H=K. E.+ Jd rU(r)pM(r)+ Pot. E ,

3
(A1)
(AS)
where 'i..=(h2/2mrr) 112 is the thermal wavelength. Thus,

where
N
for this trivial case u(r) is a simple functional of the
density:
PM(r) = L o(r- raJ
u(r)=lnp(r)X3 • (A9)
is the microscopic density and U(r) is an external po-
The free energy in the absence of interaction is
tential. We have not specified the form of the interac-
tion potential or even that the molecules interact pair-
wise. The only assumption that we require is that the
F'=F~= J d 3rp(r){lnp(r)'i..3 -1}. (A10)
interaction potential is short ranged.
Equations (AS) and (A9) suggest that the effects of in-
The grand partition function for the system is given teraction may be treated in terms of an effective one-
by body potential. Thus we set
ew=Trc,e-B<H-~NJ' (A2) eu(r)+C(r)
p(r) = 3 (All)
;>.,
where the symbol Tr cl has its usual meaning of
or
t
N=O
_1_
N!
IT J
c>=l
d3p"'ad3r"' •
h
u(r) = lnp(r)X3 - C(r) • (A12)
The internal potential is seen to be a functional of the
his Planck's constant and {3= 1/kT, k being Boltzmann's
density p(r). Clearly, it vanishes for zero density (for
constant. We define a dimensionless one-body potential
finite range of interaction), since the effects of interac-
u(r) = {3JJ.- {3U(r) , (A3) tion become negligible at sufficiently low densities
(large mean separation between molecules). Thus,
where J.l is the chemical potential. The thermodynamic
once C is known as a functional of the density, F' can
potential, W, is a functional of u(r) as well as the two-
be obtained by integration of (A12).
or higher-body potentials. Here we will only be con-
cerned with variations in u(r). We shall derive an expression for the local free en-
ergy density by formally carrying out the functional in-
The functional derivative of the grand partition func-
tegration. First we note that C(r) can be derived from
tion with respect to u(r) gives
a potential
(A4) acJ>
C(r) = op(r) ' (A13)
Thus W acts as a generating function, generating the
where
local equilibrium density upon differentiation with re-
spect to u(r). cJ>=FfJ-F'. (A14)
We define a dimensionless free energy by Hence all higher functional derivatives of C(r) with re-
F' = J d 3ru(r)p(r)- W. (A5)

spect to the density are symmetric:
oc(r1) o2cJ>
C(r1, r2) = op(r2) = op(r1)op(r2) = C(r2, r1) ' (A15a)
F' is simply related to the Helmholtz free energy F by
1 ac(r )
{3F= /3(JJ.N- PV) = {3JJ.N- W C(r1, r 2, r 3 ) = op(r2)op(r 3
) = C(r2, r 1, r 3 ) , etc., (A15b)
=F'+ J d 3r{3U(r)p(r). (A6) ac(r 1)

C(r1, r 2 , r 3 , r 4 ) = op(r )op(ra)llp(r ) = C(r2, r 1, r 3 , r 4 ) , etc.
The functional derivative ofF' with respect to the den- 2 4
(A15c)
sity is
The second derivative of F' with respect to the density,
oF' - ( )
llp(r)- u r. +
J d3 I ( ' ) llu(r')
· r P r llp(r) -
f d3r ou(r')
0 w ou(r')
I
op(r) c·l(r r ) a2F' - ou(r1)
~> 2 1lp(rl)ap(r2) - op(r2 )
=u(r). (A7)
Thus the density p(r) is the natural variable for the (Al6)
is related to the two point fluctuation function by
Jd r G- (r
3
2
1
1, r 2 )G(r2 , r 3 ) = 1i(r1 - r 3 ) , (A17)
(A20)
where
where
is the density fluctuation function,

Since ci> vanishes at zero density, we have
with
<P(p) = <P(iJ)- <P(o)
_ Il sdsJor! ds'
= t ds :s ci>(sp) = J 3
d r1[ dsC(r 1 , sp)p(r 1 ) , (A18)
C(r, r 2 ) = 2
0
C(r, r 2 , ss' ji) •
For a potential of finite range, /(r) is local; that is, it

where the functional dependence of ci> and C on the den- depends significantly only on densities within a force
sity is indicated by p. Similarly, range of the field point r. We shall see that (ll. 2) is
C(r 1 , p) = Jd r 1 dsC(r
3
2
1
1, r 2 , sp)p(r 2 ) • (A19)
obtained by expansion of J(r) in powers of the force
range.
Thus, the free energy is given formally by First we examine the expansion of C,
(A21)
where tion functions for a uniform system having density p.
C(r,r2 ,p)=2 £ sds £

0
1
0
1
ds'C(r,r2 ,ss'p),
In accordance with translational invariance, the cor-
relation functions depend only upon the independent dif-
C(r,r2 ,r3,p)=2 £ sds £0

1
0
1
ds'C(r,r 2 ,r3 ,ss'p),
ferences:
(A22a)
and
C(r,r2 ,r3 ,r4 ,p)=C0(r 2 -r,r3 -r,r4 -r,p), (A22c)

C(r, r 2 , p), C(r, r 2 , r 3, p), and C (r, r 2, r 3, r 4 , p) are, respec- along with a similar relation for the barred correlation
tively, the two-, three-, and four-body direct correla- functions. Substituting into (A20) we have
/(r) =p(r)(lnp(r)X3 - 1)-! p(r) J d 3r 2C0(r2 , p{r ))p(r+ r 2 )-! p(r) J d3r d r CO(r
2
3
3 2, r 3 , p(r))p(r+ r 2 )
x {p(r+ r 3 ) -p(r)}-t p{r) Jd r 3

2 d 3 r 3 d 3 r 4CO(r2 , r 3 , r 4>p(r))p(r+ r 2 ){p(r+ r 3) -p(r)}p(r+ r 4 ) - p(r)}, (A23)
The only contribution to the integrals in (A23) comes from points within a force range of the field point r. Thus we
can effectively expand p(r+ r 2 ) in a Taylor series about the point r:
p(r+r2 )=p(r)+r2 • Vp(r)+Hr2 • V)2 p(r)+••• • (A24)
Substituting into (A23) and dropping all terms of higher than second order in gradients, we obtain
f{r)=f0(p(r))-!p(r) Jd r {r 3
2 2• Vp{r)+Hr 2 • V)2 p{r)}{C0(r2 ,p(r))+p(r) Jd r;C (r ,r ,p(r))}
3 0
2 3
-! p{r) f asr2 d3ra(ra. Vp{r))(r3 ° Vp{r)){CO(rz, rs, p(r)) +! p(r) f d3r4CO(rz, r3, r4, p(r))} ' (A25)
where fo(p) = p (lnp X3

-1) -~ p2 J asraCO<ra, p). In accordance with the rotational invariance of the uniform system
correlation functions, we see that (A25) can be written as
(A26)
where
and
j 2(p(r)) = p~r) Jdar 2 d 3 r 3 r 2 • r 3{CO(r2 , r 3 , p)+~ Jd r C (r 3

4
0
2, r 3 , r 4 , p)}.
With use of the symmetry properties of the correlation functions, f 2(p) can be rewritten
f 2(p) = :v J d 3r 1 tfr2 d 3r 3 (r 2 - r 1) • (r3 - r 1{c(r 1 , r 2 , r 3 , p)+~ Jd r C(r 3

4 1, r2 , r 3, r 4, p)~
= pV
6
f tfr tfr
1 2 d 3r 3(r12 - r 1r 2 ) {C(r
-
1 , r 2 , r 3 , p) +
E.
2
f - t!'r4 C(r 1 , r 2 , r 3 , r 4 , p)( l
=
1 ~V Jd r r d r (r r C(r r r p) +~ Jd r C(r
3
1d
3
2
3
3 1-
2
2) { 1, 2, 3,
3
4 1, r 2 , r 3 , r 4 , p)}
= :v Jdar2 dar3 ~{c0(r2 ,ra,p)+~ Jdar C (r ,r ,r ,p)}. 4

0
2 3 4 (A27)
In order to obtain a free energy density of the form of (II. 2), we must eliminate the Laplacian term in (A26). We
note that it can be rewritten as
f(r) = / 0(p(r)) +! A(p(r))Vp(r) 2 - V • {/1(p(r))Vp(r)} , (A28)
where
! A(p) =jf(p)-fz(p) •
The divergence term in (A28) does not contribute to the free energy (A20). From (A15) and (A21) we have
OCb(rz, p) =
()p
f daraCO(rz, r a• p) (A29a)
and
5CO(r2, ra, p) =
-~""-.il.i.-"""-
()p
J d ar 4-C0( r 2 , ra, r 4 , p ) • (A29b)
Substituting into f( and combining with/2 , we obtain
(A30)
Now
1 52
2 ()p2 2 (1
{p C 0(r2 , p)}= .,t sds Jo ds'
(1 1)2
()p 2
2 0
{p C (r2 , ss' p)}=
11 1 d
dsp ds P
111 ds' ds' d 2 0
{ss' p CO(r2 , ss' p)}= C (r2 , p) • (A31)
0 0
Hence A(p) is given by (II. 3). Equation (II. 2) is obtained by combination of (A31), (A28), (A20), and (A6).
The general partial differential equation (IV. 3) can be derived by direct expansion of the chemical potential in
powers of the density gradient. We expand the right hand side of (A12). We note that C(r) can be expanded as
C(r) = C(p(r))+ J dar1C 0(r1> p(r)){p(r +r 1)- p(r)} + ~ J dar 1 dar2 C0(r 1 , r 2 , p(r)){p(r+ r 1 ) - p(r)}{p(r+ r 2 ) - p(r)}+ ·" ,
(A32)
where C(p) is the internal potential of the uniform sys-

tem having density p. If we use (A24) we obtain
C(r) = C(p(r)) + ,8A(p(r))V2 p(r) + ,8A 3 (p(r))Vp(r)2 ,

=_!_Jtf r!l5Co(rl,p)=.!.A'()
12 r1 1 5p 2 P (A34)
(A33)
where
If this is substituted into (A12), we obtain
.BAa(p) =i J 0
dar 1 t!'r2 r 1 • r 2 C (r1 , r 2 , p) • u(r) = f3SJ.(p(r))- ,8A(p(r))V2p(r)
-! ,BA'(p(r))Vp(r)2 • (A35)
By a set of manipulations similar to those in (A27) we
see that A3 can be written as Equation (A35) is (IV. 3) multiplied by {3.
APPENDIX B: GENERAL YOUNG-LAPLACE (Bl)

EQUATION rn is everywhere the component of r normal to surfaces
of constant density. Then Vp(r) can be written as
In this Appendix we shall derive the Young-Laplace
equation for interfaces of arbitrary curvatureo In Vp(r) =n(r) op(r) . (B2)
terms of the unit vector (IVo 5) we define the normal orn
component of the position vector r by The divergence of the stress tensor is given by
V • a(r) = [ll - IJ.(p(r))- U(r) + A(p(r))v;(r) + t A'(p(r))(ap(r))

arn
2
n ap(r)-
]
arn
VU(r)p(r) = apn(r) n(r)
arn
+ A(p(r))V~p(r) a~(r) n(r)- V.LU(r)p(r) , (B3)

arn
where APPENDIX C: CRITICAL EXPONENT EQUALITY

In this Appendix we wish to discuss the surface ten-
sion of a flat interface in the vicinity of the critical
point. In particular, we obtain an equality for the sur-
face tension exponent in terms of bulk exponents. This
equality, which was previously suggested by Widom, 10
and is
Jl+IJ=y+2{3, (C 1)
Next we note that where J1 is the surface tension exponent, v is the cor-
relation length exponent, y is the compressibility ex-
~ V~(r) Vp(r)• V.LIVp(r)l
V.L. n(r)= ap(r)/ar" ponent, and f3 is the order parameter exponent. These
!Vp(r)! 2
exponents are defined for the two-phase region at tem-
V~p(r) peratures near the critical temperature. For T suffi-
(B4)
IVp(r)l ciently near Tc they are defined by
But from differential geometry, 20 ao=Aao(Tc- T)JJ.' (C2)
V.L• n(r)=-2H(r)=2/R(r), (B5) ~=A~(Tc- rrv, (C3)
where H(r) is the invariant mean curvature, and R(r) is
the mean inverse radius of curvature of the constant x =_!(ap)
P ap r
=A (T -
x c
T)"r
,
(C4)
density surface passing through the point r. Thus,
and
from (B3) and (IV" 4) we have
PL- PG =Al>.(Tc- T)S • (C5)
ap"(r) =-A( (r))2 ap(r)2/ar" _ aU(r) (r)" (BB)
arn p R(r) arn p That is, these exponents characterize how the surface
tension and the density difference vanish and how the
Equation (B6) is a differential statement of the Young- correlation length ~ and compressibility x diverge at
Laplace law for an interface of arbitrary curvature. the critical point. Widom 's derivation of (C 1) was
We obtain the usual form by integrating (B6) along a based on a three-step procedure involving assumptions
line passing through a point rn= r 0 {defined as point at about the singular part of the free energy and the spe-
which [ a p(r)/ar~Jrn=ro = o},
2
cific heat exponent. Here we show that the equality can
be obtained directly from the theory presented here.
Pn(ro- ~) -Pn(ro+ ~) = ar/R , (B7)
The argument to be presented below reveals the basis
where of the satisfaction of (C 1) in the Fisk-Widom 6 calcula-
tion of the density profile from a van der Waals theory
and a scaling equation of state.
To obtain the equality we consider an approximate

R=R(r)jrn=ro' density profile of the form
and where we have assumed that ~ « R and that
p(x) = PL x<--~
2'
(B8)
p(x) = p(O) + p'(O)x (C6)

Thus again we can explicitly recognize the limitations
of the (integrated) Young-Laplace equation. The dif-
ferential form (B6) is valid regardless of the sharpness
of the interface.
p(x) = Pc x > 1,
p This expansion, along with the corresponding identi-

fication of A(p), was first discussed by Ornstein and
Zernike. 21 This identification was employed in a con-
text similar to the present one by Fisk and Widom 6 in
their analysis of a scaling equation of state.
Substituting (C 14) into (CB), we obtain
ao a( Tc - T),.-2,..28+v =( T c- T)"+28-v • (C 15)
Thus, we finally have
------~--~----~--------- X
-{12 0 {12
11- =y+ 2{3- v' (C 16)
FIG. 14. Schematic density profile employed for study of in-
terfacial phenomena in the critical region. which is the desired equality.
The equality has emerged naturally from the theory.
where ~ is the interfacial width (Fig. 14), In Appendix Of course, we have not actually evaluated the integral
D we show that ~ is proportional to the correlation expression for the surface tension, but the approximate
length. A more precise determination of the density expression (C8) certainly characterizes the surface ten-
profile is unnecessary, because all that is really im- sion properly in the critical region.
portant is its width ~. We see that p'(O) is related to ~
and PL- PG by APPENDIX D: RELATION BETWEEN INTERFACIAL
WIDTH AND CORRELATION LENGTH
p'(O) =- (pL- PG)/ ~ • (C7)
In this Appendix we show that the interfacial width ~.
r
Thus, from (ll. 19) we have approximately employed in the approximate density profile (C6), must
be proportional to (i.e. , diverge with the same exponent
Uo"" M(p(O))~L ~ PG =A(p(O)) (pL ~ PG)2 (C8) as) the correlation length in the critical region. To see
this we first note' that an assumed density profile of the
Next we examine A(p). We recall that the inverse of form (C6) amounts to considering a restricted varia-
the structure factor [cf. (A16)] tional problem. The true density profile which solves
S(k, p) = f d 3 f e·•t•r ((pM(r)- p)(pM(Q)- p)) (C9)

the equation (ll. 8) arises from considering all possible
density profiles. We therefore expect that the impor-
tant features of the true density profile can be obtained
is from (C6) by variation ofF- JlN with respect to PL• PG•
s- 1(k, p) = 1/p- C(k, p) , (C10) and~.
where From (ll. 4) we have, using (C6),
is the Fourier transform of the direct correlation func-

tion. We see that for small k
+ £l/2
-t/2
d.x[lJ!(x)- /lp(x)] • (D1)
(Note that a+ ~/2 <0.)

1 {apf3)
s-1(k, p) = P \""ap r -21 dka
d
2
I
C(k, p) -.o ka' (Cll)
The interfacial term can be written by a change of
where we have used the long wavelength limit, variable as
s-t(o p} = J:.(apf3) • (C12)

' P \ ap r
The second derivative of C(k, p) is
= ~ Lt;2a ds [ 1/!~PL; PG + (pG- PL)s)
(C13)
Thus we have
+ t A(PL; PG +(pG _ PL)s) (PL ~ PG ) 2
]- 11-~~PL; PG ).
(D2)
s-t(k) =1.{8!{3) + /3A(p)!z2
p\ p T Variation of (F- gN} with respect to Pv and PL yields
(C14)
(1/!'(po)- IJ./b-
~
1)+
2
a(f.- IJ.N.)
apG
=0 (D3a)
where and
(ap) _&!_
~ (p) = P ap r A(p) = 1/J"(p)
2 - (1/J'(pL)- g) (a+~)+ a(fs ~~N,) =0 . (D3b)
is the square of the correlation length. Thus, for ~/b« 1 and Ua« 1 we have
(D4a) with respect to the position of the interface we would

have obtained the Maxwell construction (IL 13) and con-
and
sequently that 11, PL, and Pv must be their coexistence
(D4b) values.]
Therefore, for a sufficiently large system PL and Pv More important is the variation with respect to ~.
must be zeroes of Jl(p)- Jl. [If we had varied F- J1N We see that
requires
~= J1/2 dsA(_pL; PG + (pL- PG)~ ( PL ~ PGY = 2 J1/2 ds q/PL2+ PG + (pL- PG)s) -1/J(pL) -1/J(pG) . (D6)
s -1/2 \ s -1/2 \
Thus, solving (D6) for ~ 2 yields
2
~2 _ f_lft~ ds A((pL + PG)/2 + (pG- PL)s) (pL- PG)
(D7)
- I_lj ; ds[21f!((pL + PG)/2 + (pG- PL)s) -1/J(pL) -1/J(pG)] ·
1
Next we note that the terms in the integrand of the denominator can be written as
1/J((pL + PG)/2+ (pG- PL)s) -1/J(pG) = - {

12
ds' d~' 1/J~PL; PG + (pG- PL)s') = t/ 2
ds'1/J'(PL; PG + (pG- PL)s') (pG- PL)
(DSa)
and
1/!~L; PG + (pG- PL)s) -1/J(pL) =Ls ds'1f!'(PL; PG + (pG- PL)s') (pG- PL) • (DSb)
12
We also observe that
(D9a)
and
r
-1/2
ds1/J'(pL)=(S+~)1/J'(pL)= Jl(S+~). (D9b)
Thus we can write
112
[1/J'~L; PG -1/J'(PL~]= (pL- PG){£
12
+ Ls ds' + (pG- PL)s') ds' [ ds"1f!"(PL; PG + (pG- PL)s")
12
+Is ds' Is' ds"1f!" (PL + PG + (pG- PL)s")] • (D 10)

-1/2 -1/2 2
Integrating (D10) over sand noting that the term linear ins gives no contribution to the integral, we obtain
I 1/2 ds [ 21/! ( PL; PG + (pG- pL)s) -1/J(pL) -1/J(pG)]

-1/2
=11/2 ds [!1/2 ds' 11/2 ds"1f!" ( PL 2+ PG
-1/2 s s'
+ (pG- PL)s")
(Dll)
When (Dll) is substituted into (D7), that equation becomes
1/2 dsA ~ PL+PG +(pG-PL)s) { !1/2 ds [J1/2 ds' !1/2 ds"1f!" ( PL;PG +(pG-PL)s")
~2 =
J
-1/2 2 -1/2 s s'
+ fs-1/2 ds'fs•1/2 ds"1f!"(PL+PG

2
+(pG-PL)s1l}-1
'J (D12)
From comparison of (D 12) with (C 14) it is clear that edited by C. Domb and M. S. Green (Academic, New York,
the divergence of the interfacial width is the same as 1972), Vol. 2, Chap. 3.
13A similar derivation of the Maxwell construction has been
that of the correlation length. We can estimate the
given by N. G. van Kampen, Phys. Rev. Sec. A 135, 362 (1964).
proportionality constant if we assume that A(p) is reg-
The generality of his argument, there applied to the original
ular for p near the critical density [actually, to be con- van der Waals equation, is seen by the discussion presented
sistent with (C14) we must assume A(p)=(Tc- T)"- 2 v here.
g(p), where g(p) is finite and nonzero at the critical 14 Recently, K. W. Sarkies and N. E. Franke1 [Phys. Rev. A
point] and note that the multiple integ.rals in the denom- 11, 1724 (1975)] have considered nucleation theory within the
inators have their principle contributions at Pc and PL, context of van der Waals theory. Their approach, similar
respectively. From the assumed regularity of A we to ours although specialized to analysis with the Fisk-
must have Widom6 scaling equation of state and to density-independent
A, is based on the earlier work of Cahn and Hilliard [J. Chern
(D13) Chern. Phys. 31, 688 (1975)].
15 This expression for the stress tensor was obtained by Lovett1
sufficiently near the critical point. Hence, we have in his Ph. D. thesis by use of the same variational principle
2- HA(pL)+A(pG)) applied to the expanded free energy (II. 2) [his II. 1)]. How-
(D14)
~ - Hlfi"(pL) + 1/J"(pG)) ever, as was the case with Lebowitz and Percus, 1 he did not
reproduce Eq. (IV. 3) [his II. 3)] by direct expansion of the
or, using (C14), chemical potential and was led to conclude that the van der
2
~2- ~ 2 (pL)lf/'(pL)+ ; (pc)lf/'(pG) Waals free energy density was incorrect. Accordingly, he
3 (D15) did not publish this expression for the stress tensor, for
- 1/J"(pL)+I/J"(pG)
which he should nonetheless be credited.
Thus, in this approximation the interfacial width dif- 16 All the higher order derivatives are determined by the dif-
fers by a trivial factor ,f3 from a rms correlation ferential equation (IV. 9).
length. Since both HpL) and ~(pc) diverge with expo- 11 H. Reiss, Methods of Thermodynamics (Blaisdell, New York,
nent v, ~ must diverge with exponent v as well. A 1965), pp. 202-203.
more correct treatment of the integrals in (Dl2) should 18 p, C, Martin (lecture notes, unpublished).
19 That (AS) is associated with a stationarity ofF' under varia-
not alter this conclusion. 22
tion of the density p(r) has been demonstrated by C. De
Dominicis and P. C. Martin, J. Math. Phys. 5, 14 (1964),
*Work supported by the Materials Research Laboratory at This general variational principle forms the basis for the
Brown University, funded by the National Science Founda- approximate (expanded) variational principle employed here
tion, and by the United States Public Health Service, Grant to obtain (IV. 3) from (II. 2). In contrast to Lebowitz and
GM10906-12. Percus, 11 we have shown that the same differential equation
tPresent address: Fu Jen University, Chemistry Department, results from either extremizing the expanded free energy or
Taipei, Taiwan. from expanding the general stationarity condition (A. 8),
tPresent address: Research and Development Department, 20 See, for example, I. S. Sokolnikoff, Tensor Analysis, Theory
Phillips Petroleum Company, Bartlesville, OK 74004. and Applications to Geometry and Mechanics of Continua
1This paper is based in part upon a thesis (Brown University, (Wiley, New York, 19xx), pp. 176-186.
May, 1975), presented by Arthur J. M. Yang in partial full- 21 L. S. Ornstein and F, Zernike, Proc. Akad. Sci. (Amster-
fillment of the requirements for the Ph. D. degree. A pre- dam) 17, 793 (1914); reprinted in The Equilibrium Theory of
liminary report of this work was presented by Dr. Yang at Classical Fluids, edited by H. L. Frisch and J. L. Lebowitz
the March 1975 American Physical Society Meeting in Denver (Benjamin, New York, 1964).
[see A. J. M. Yang, J. H. Gibbs, and P. D. Fleming, Bull. 22 For the special case of the Fisk-Widom6 scaling equation of
Am. Phys, Soc. 20, 305 (1975)]. state we have
2J. G. Kirkwood and F. P. Buff, J. Chern. Phys. 17, 338
(1949). >11" {(pL + Pc)/2+ (pG- PL)s"]
3F. P. Buff, Z. Elektrochem. 56, 311 (1952).
4J. D. van der Waals, Z. Phys. Chern. 13, 657 (1894),
5J. W. CahnandJ. E. Hilliard, J. Chern. Phys. 28,258
=(Tc-T)Y[(1;f3 z-~j(Z)+Z(Z;1) f(Z)],
(1958). where Z = I - 2s" 1 118 , and j(Z) is the Fisk-Widom scaling
6s. Fisk and B. Widom, J. Chern. Phys. 50, 3219 (1969). function. Evaluation of the integrals in (D 12) for this case
1
F. P. Buff, R. A. Lovett, and F. H. Stillinger, Phys. Rev. yields
Lett. 15, 621 (1965); R. A. Lovett, Ph. D. thesis, Univer-
1;2 1;2 2
sity of Rochester, 1965; R. A. Lovett, P. W. De Hav~,
!; 2(pL) = !; 2(pr) = f3 2w(O)'
James J. Vieceli, and F. P. Buff, J. Chern. Phys. 58, 1880
(1973). where w(t) is defined by Fisk and Widom to be
8D. G. Triezenberg and R. Zwanzig, Phys. Rev. Lett. 28,
1183 (1972).
9This equation has also been called the Laplace-Kelvin equa-
w(t) = 2 jt1 z28- 1(1-Z) ~ dZ.
tion, the Kelvin equation, the Laplace equation, or the Pla- Recall that for the Fisk-Widom equation of state !; 2(pL)
teau equation. The confusion is furthered by the usual ref- =!; 2(pr) (=L 2 in their notation). For one choice of scaling
erence to J. W. Gibbs, Scientific Papers (Dover, New York, function [(1-Z)j{Z)/j(1) = £(1-z 4 13)], w(O) =~and (D 12) be-
1961), Vol. 1, p. 229, Eq. 500. comes (using f3 = !l
10 B. Widom, J. Chern. Phys. 43, 3892 (1965).
11 !;/!; (pL) = 2v'3"',
J. L. Lebowitz and J. K. Percus, J. Math. Phys. 4, 116
(1963). Again the interfacial width differs by a trivial factor from the
12 B. Widom in Phase Transitions and Critical Phenomena,
correlation length.

Molecular Theory of Surface Tension

Uploaded by

Copyright:

Available Formats

Molecular Theory of Surface Tension

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Molecular Theory of Surface Tension

Uploaded by

Copyright:

Available Formats

Molecular theory of surface tension*

Arthur J. M. Yang, t Paul D. Fleming 111! and Julian H. Gibbs

I. INTRODUCTION of the theory in three dimensions and consequent es-

When the molecular number density p(r) is sufficient-

V • u= f, where u is the stress tensor of the system and

Young-Laplace equation. The analysis of the radial where

II. DENSITY VARIATION IN ONE DIMENSION; THE Ia

ergy density, Evaluation of the partial derivatives in (11. 7) yields

J. Chem. Phys., Vol. 64, No.9, 1 May 1976

FIG. 2. A typical chemical

!J. =!J.(p(x))- A(p(x))p"(x) -!A'(p(x))p'(x) 2 , (II. 8)

A'( )= M(p) ~----------------~X

is the chemical potential of a uniform system having

J. Chern. Phys., Vol. 64, No. 9, 1 May 1976

At fixed temperature there is only one J.1., the coexis-

= [¢(PL 0) - J.J.(PL 0 )PL 0 ] - [¢(Pc 0 ) - J.J.(Pc 0 )Pc 0 ] • (ll. 13)

(b) P = J.J.p(x)- ljl(p(x))+ i A(p(x))p'(x)2 =canst. (II. 15)

J. Chern. Phys., Vol. 64, No. 9, 1 May 1976

plxl J.L- U(x) = J.L(p(x)) -A(p(x))p"(x)

dP(x) = _ dU(x) p(x) =f(x) (IlL 2)

P(x) = J.Lp(x)- 1/J(x) + A(p(x))p'(x) 2

1----------=-Po, = J.Lp(x)- 1/J(p)- U(x)p(x) + i A(p(x))p'(x) 2 •

J. Chem. Phys., Vol. 64, No. 9, 1 May 1976

P(x)+rnJ"' dxp(x)g=P(O). (ill.ll) n

IV. THREE-DIMENSIONAL THEORY aro = Jdrna(r) ' (IV. 7)

J. Chern. Phys., Vol. 64, No. 9, 1 May 1976

FIG. 8. The physi-

Since A is positive, for IJ.(p)- 1J. > 0 p(r) is concave up-

J. Chem. Phys., Vol. 64, No. 9, 1 May 1976

FIG. 11. Physical solutions for different values of p(O).

tential, JJ.mu> which satisfies the Maxwell equal area

and r 0 must approach infinity, so that f(Ap' /r) dp- 0.

J. Chern. Phys., Vol. 64, No. 9, 1 May 1976

{i/J(p)- f.1.P- ~ Ap'

J. Chern. Phys., Vol. 64, No.9, 1 May 1976

H=K. E.+ Jd rU(r)pM(r)+ Pot. E ,

where 'i..=(h2/2mrr) 112 is the thermal wavelength. Thus,

F' = J d 3ru(r)p(r)- W. (A5)

=F'+ J d 3r{3U(r)p(r). (A6) ac(r 1)

J. Chem. Phys., Vol. 64, No. 9, 1 May 1976

is related to the two point fluctuation function by

is the density fluctuation function,

For a potential of finite range, /(r) is local; that is, it

where tion functions for a uniform system having density p.

C(r,r2 ,p)=2 £ sds £

C(r,r2 ,r3,p)=2 £ sds £0

C(r,r2 ,r3 ,r4 ,p)=C0(r 2 -r,r3 -r,r4 -r,p), (A22c)

x {p(r+ r 3 ) -p(r)}-t p{r) Jd r 3

where fo(p) = p (lnp X3

J. Chern. Phys., Vol. 64, No.9, 1 May 1976

j 2(p(r)) = p~r) Jdar 2 d 3 r 3 r 2 • r 3{CO(r2 , r 3 , p)+~ Jd r C (r 3

f 2(p) = :v J d 3r 1 tfr2 d 3r 3 (r 2 - r 1) • (r3 - r 1{c(r 1 , r 2 , r 3 , p)+~ Jd r C(r 3

= :v Jdar2 dar3 ~{c0(r2 ,ra,p)+~ Jdar C (r ,r ,r ,p)}. 4

Substituting into f( and combining with/2 , we obtain

where C(p) is the internal potential of the uniform sys-

C(r) = C(p(r)) + ,8A(p(r))V2 p(r) + ,8A 3 (p(r))Vp(r)2 ,

J. Chern. Phys., Vol. 64, No.9, 1 May 1976

APPENDIX B: GENERAL YOUNG-LAPLACE (Bl)

V • a(r) = [ll - IJ.(p(r))- U(r) + A(p(r))v;(r) + t A'(p(r))(ap(r))