G. Nagele
Institut fur
Festkorperforschung
Forschungszentrum Julich
GmbH
Contents
1 Introduction 2
1.1 Model systems and pair potentials . . . . . . . . . . . . . . . . . . . . . . . . 3
Extended version of the article appeared in: Soft Matter: From Synthetic to Biological Materials, 39th
IFF Spring School 2008, Series: Key Technologies Vol. 31, Forschungszentrum Julich Publishing (2008)
B2.2 G. Nagele
1 Introduction
Atomic liquids and colloidal fluids are distinguished from dilute gases by the importance of
shortrange correlations and particle collision processes, and from crystalline solids by the lack
of longrange order. The distinction between a liquid and a gas is only of quantitative nature,
since there is no change in spatial symmetry in going from the gas to the liquid phase.
The most simple liquid systems are monoatomic liquids consisting of spherically shaped atoms
(argon or neon) or quasispherical molecules such as methane. The interactions between the
atoms can be characterized by simple pair potentials which depend only on the distance between
an atomic pair. Colloidal fluids consist of mesoscopically large colloidal particles, typically a
few hundred nanometers in size, dispersed in a lowmolecular solvent such as water. There are
strong local correlations in the positions of these particles (super atoms) so that they form a
colloidal liquid in the liquid (solvent). The omnipresence of colloidal dispersions in chemistry
and biology, and the unsurpassed variety and tunability of their particle interactions explains
their importance both in industrial applications and fundamental research. Wellstudied exam
ples of simple colloidal fluids are suspensions of silica and plexiglass spheres in organic sol
vents, and aqueous suspensions of globular proteins or highly charged polystyrene latex spheres.
While the frictiondominated, diffusive dynamics of colloidal particles is quite different from
the ballistic one of atomic liquids, atomic and colloidal liquids are quite similar in terms of
their equilibrium microstructure, i.e., in terms of the average ordering of particles. There are
just orders of magnitude differences in their characteristic length scales.
In the present lecture, I will discuss a versatile class of liquid state theory methods which al
low to determine theoretically the microstructural and thermodynamic properties both of atomic
and colloidal liquids, from the knowledge of the particle interactions. These socalled integral
equation schemes are based on the OrnsteinZernike (OZ) equation. Some of the most relevant
OZ schemes will be introduced, and we will explore their predictive power in comparison with
computer simulations and experiment. The quantity of central importance calculated with in
tegral equation schemes is the radial distribution function, g(r), of an isotropic liquid, and the
associated static structure factor S(q). The latter property is essentially the Fourier transform of
g(r), and it can be measured using scattering techniques. The radial distribution function (rdf)
quantifies the probability of finding a particle at some distance from any other one. It can be
indirectly measured by radiation scattering experiments and, for very large colloidal particles,
also directly using video microscopy.
Typical atomic and colloidal liquids of spherically shaped particles will be briefly characterized
in section 1 in terms of their interaction potentials. Section 2 provides an overview on general
properties of g(r) and S(q), and on their relation to scattering experiments and thermodynamic
properties. Section 3 introduces the fundamental OrnsteinZernike equation and the associated
concept of direct correlations. As a first application of the OZ equation, a meanfield type
description of critical opalescence in nearcritical liquids will be given. Various approximate
closure relations are discussed which lead to closed integral equations for g(r). Particular atten
tion is paid to the hardsphere fluid, since it serves as a reference system of many liquids, similar
to the harmonic solid in solidstate physics. In the final part of section 3, the OrnsteinZernike
integral equation concept is generalized to fluid mixtures. In section 4, I will explain the con
Theories of Fluid Microstructures B2.3
cept of effective colloid interactions in systems which include, aside from the large colloidal
particles and in addition to the solvent, components of small particles such as free polymer
chains or salt ions. In two important examples, I will describe how a coarsegrained, simpli
fied description is achieved, in the form of a onecomponent suspension of decorated colloidal
particles, by integrating out the smallparticles degrees of freedom. The examples discussed
are a binary mixtures of colloidal hard spheres and small polymer chains, and a suspension of
charged colloidal particles with added salt ions.
provides a fair description of the interaction between pairs of raregas atoms such as in argon,
krypton and xenon, and also of quasispherical molecules such as CH4 . A sketch of the potential
curve is provided in Fig. 1. Two parameters characterize the potential: the collision diameter
where u(r) = 0, and the depth, , of the potential minimum at r = 21/6 . The values of and
have been determined for a large number of atoms using, e.g., atomic scattering techniques.
For argon, Ar = 0.34 nm and Ar /kB = 119.8 K. The shortrange repulsive part of the pair
potential proportional to r12 represents approximately the electronic repulsion of two atoms.
The longerranged van der Waals attraction between two atoms at a distance r is described by
the r6 part.
u(r)
21 / 6
r [nm]

The most simple pair potential one can think of is the potential between hard spheres of diameter
, i.e.
{
for r <
u(r) = . (2)
0 for r >
B2.4 G. Nagele
While there exists no atomic fluid of hard atoms, colloidal suspensions of hard spheres are
realized within good approximation, by coated polymethylmethacrylate (PMMA) spheres dis
persed in a refractive indexmatched nonpolar solvent such as cyclohexane. The coating con
sists of a thin layer, as compared to , of adsorbed polymer chains (cf. Fig. 2). The polymer
brush gives rise to a shortrange repulsion between the colloidal spheres which counterbalances
the remnants of the van der Waals attraction. The sizes of the colloidal hard spheres are in the
range of several hundred to a few thousand nanometers.
PMMA particles in a nonorganic solvent are a paradigm for sterically stabilized dispersions.
u(r)
uel (r)
k BT
0
u vdW(r)
r
Fig. 3: Left: Electrostatic and van der Waals potential contributions to the total effective pair
potential u(r)(red curve). Right: colloidal macroions with counterions and coions.
where the sum is taken over all types of microions, i.e., surfacereleased counterions and salt
ions, of number densities and charge numbers Z . The factor 1/(1 ) corrects for the
free volume accessible to the microions, owing to the presence of colloidal spheres, where
= (4/3)a3 is the colloid volume fraction and the colloid number density. Notice here
that the range and strength of the potential can be controlled by adding or removing small ions,
and by changing the solvent or temperature.
The total effective pair potential, u(r), of chargestabilized colloidal particles is the sum, u(r) =
uel (r) + uvdW , of uel (r) and the attractive van der Waals pair potential, uvdW (r). The van
der Waals attraction between two identical colloidal spheres can be described approximately
by [4, 5]
[ ( )]
Aef f 2 a2 2 a2 4 a2
uvdW (r) = + 2 + ln 1 2 r> (5)
6 r2 4 a2 r r
with uvdW (r) (Aef f 6 /36) /r6 for large r, and uvdW (r) (Aef f /24)/(r ) near
the contact distance r = (cf. Fig. 3). Clearly, the divergence of uvdW (r) at contact distance,
arising from the assumption of ideally smooth spheres surfaces with precisely two surface atoms
overlapping, is unrealistic.
Electrodynamic retardation and nonpairwise additivity effects on the dispersion forces are in
corporated to some extent in the effective Hamaker constant Aef f . For nonmetallic colloidal
spheres, Aef f is typically of the order of a few kB T [4, 5]. Van der Waals forces between iden
tical particles are always attractive. Nonidentical particles can repel each other however, when
the dielectric susceptibility of the solvent is in between that of the two particles.
For dispersions of highly charged colloidal particles at lower salt content, uvdW (r) becomes
completely masked by the electrostatic part uel (r). In this case, one frequently refers to the col
loidal particles, with the associated diffuse layer of associated microions, as Yukawa spheres,
since their microstructural properties are determined by the Yukawalike, exponentially screened
Coulomb potential uel (r).
It is often assumed that the potential energy, U (rN ), of a N particle liquid system can be ap
B2.6 G. Nagele
N
N
U (r )
N
u(ri rj ) = u(rij ) , (6)
i<j i<j
for any configuration rN = {r1 , , rN } of position vectors {ri } pointing to the particles cen
tres. The quality of this pairwiseadditivity assumption depends on the choice of u(r), and
on how certain manybody aspects (e.g., nonadditive dispersion forces, influence of solvent
molecules, electrostatic screening et cetera) are approximately included in the pair potential.
Typically, u(r) is density and temperature dependent. The dependence of the (effective) pair
potential on the thermodynamic state is a remnant of microscopic degrees of freedom which
have been averaged out on the level of coarsegraining where the potential applies. Colloidal
systems where U (rN ) is exactly pairwise additive are scarce. Examples of these exceptional
cases are ideal hard spheres, and suspensions of spheres with a specific shortranged depletion
attraction that describes the integral influence of free polymers added to the system (cf. subsec
tion 4.1).
Under the premise of Eq. (6), the microstructural properties of the fluid and, to some extent,
the thermodynamic properties are solely expressible in terms of u(r) and its associated radial
distribution function g(r). The latter is the most simple and most relevant example for a reduced
distribution function and will be discussed in the following.
eU (r )
N
N
PN (r ) = , (7)
ZN (V, T )
with configurational integral
U (rN )
drN eU (r
N)
ZN (V, T ) = dr1 drN e = , (8)
is the probability density that the centers of the N particles are at the positions r1 , , rN . It
provides far more information than necessary for the calculation of scattering properties and
Theories of Fluid Microstructures B2.7
thermodynamic functions. What is really needed are the reduced distribution functions for a
small subset of n N particles irrespective of the positions of the remaining N n particles.
The correlation length (T ) is a characteristic distance over which two particles are correlated.
For fluids, is typically of the range of u(r) or larger to some extent, but under certain con
ditions (i.e., near a liquidgas critical point) it can become extremely large. For a nparticle
cluster with large mutual distances rij = ri rj  , and N 1,
(n)
n
(1)
N (r1 , , rn ) N (ri ) = n (11)
i=1
since these particles are then uncorrelated. To describe pair correlations in a fluid relative to a
classical ideal gas of uncorrelated particles at the same density and temperature, we define the
pair distribution function, gN (r1 , r2 ), in the N V T ensemble as
(2)
N (r1 , r2 )
gN (r1 , r2 ) := (1) (1)
, (12)
N (r1 ) N (r2 )
so that gN (r1 , r2 ) 1 for r12 . If the system is isotropic as well as homogeneous (i.e.,
(2)
no spatially varying external force field, no crystalline state), N and gN are functions of the
separation r = r12 only. Then,
(2)
N (r) N (N 1)
gN (r) = = dr3 drN PN (rN ) (13)
2 2
is denoted as the radial distribution function (rdf). It plays a central role in onecomponent
fluids, since it is indirectly measurable by radiation scattering experiments. Moreover, thermo
dynamic quantities can be written as integrals over gN (r) and u(r), provided that the particles
interact by pairwise additive forces (cf. subsection 2.4).
As an important observation, we note that gN (r) is the average density of particles a distance
r apart from a given one. In fact, integration of g(r) over the system volume leads to
N (N 1)
dr gN (r) = dr12 dr3 drN PN (rN ) = N 1 (14)
B2.8 G. Nagele
since dr12 = V 1 dr1 dr2 . Likewise, Eq. (14) can be rewritten as
1+ dr [gN (r) 1] = 0 . (15)
V
Notice for the canonical rdf that gN (r ) = 1 N 1 , since there are N 1 particles left
aside from the one at r = 0. For large N , we can identify gN (r) as the conditional probability
of finding a particle a distance r from a given one. To see this we use that in a homogeneous
system, dr/V is the singleparticle probability of finding the given particle in the volume ele
ment dr. Then, dr/V gN (r) is the (unconditional) joint probability for two particles being
separated by the distance r, with one of them located inside dr. The sum (i.e., the integral)
of the joint probability over all accessible volume elements must give the value one. Precisely
this is stated in Eq. (14). The unconditioned singleparticle and twoparticle joint probabilities
can attain values between zero and one only. The conditional probability gN (r), on the other
hand, is the ratio of two unconditioned probalities and can thus attain values also larger than one.
Eqs. (14) and (15) are valid for a finite system of fixed N , without fluctuations in the number
of particles. In order to be independent of the specific statistical ensemble used in calculating
static properties of a macroscopically large system, it is understood that the thermodynamic
limit (Tlimit, for short) of macroscopically large systems, i.e. N, V with = N/V kept
fixed, is taken at the end of each calculation. Then
g(r) := lim gN (r) lim gN (r) (16)
N,V
Let us summarize general properties of the g(r) for a fluid, which follow from the definition of
gN (r) in Eq. (13):
g(r) 0 , g(r ) = 1 (17)
g(r) 0 , for u(r) 1 (18)
g(r) = eu(r) + O() (19)
g(r) continuous for u(r) piecewise continuous (20)
The typical behavior of g(r) for an atomic liquid with a soft pair potential of LennardJones
type, and for a hardsphere fluid is sketched in Fig. 4. Regions of r with g(r) > 1 (g(r) < 1)
have a larger (lower) probability of finding a second particle from a given one at r = 0, than for
an ideal gas at the same T and . The main features are a smallr region where g(r) = 0 owing
to strong repulsive forces exerted by the particle at the origin, and several peaks representing
increasingly diffuse shells, for increasing r, of next neighbors, second next neighbors and so
on. This shell layering manifests the granularity (noncontinuum nature) of the fluid. The
oscillations in g(r) decrease in amplitude with increasing r. Eventually, g(r) approaches its
asymptotic value one for r > (T, ). The oscillations in g(r) become more pronounced with
increasing . Whereas the LennardJones g(r) is continuous at any r, the hardsphere g(r)
jumps from zero to values 1 at r = , due to the singular nature of the hardsphere u(r).
For hard spheres, g(r) = (r ) + O() according to Eq. (19). Hardsphere systems are
athermal (i.e., T independent) since the probability for a given particle configuration is either
zero or one, independent of , depending only on whether two or more spheres do overlap or
not.
Theories of Fluid Microstructures B2.9
g(r)
r/
1 2
Fig. 4: The g(r) of a LennardJones liquid (left), and a fluid of hard spheres (right).
The quantity on the righthandside can be interpreted as the force on particle 1, if we hold
particle 2 fixed, and average over the positions of all the other particles (as denoted by 1,2 ).
Thus, w(r) is the potential for this force, and it is therefore called the potential of mean force.
Likewise, w(r) can be interpreted as the reversible work required in a process where two par
ticles in the system are moved together, at constant N , V and T , from infinite separation to a
relative distance r. To see this explicitly, recall from statistical mechanics that
F (r12 ) = kB T ln dr3 rN eU (r ) + F0 (N 2, V, T )
N
(24)
is the Helmholtz free energy of a (N 2)particle system in the presence of two fixed spheres
at a distance r12 . The spheres 1 and 2 influence the system through their excluded volume and
longerranged interactions. Here, F0 is an irrelevant configurationindependent part of the free
energy. Consequently
w(r12 ) F (r12 )
= (25)
r12 r12
B2.10 G. Nagele
and
w(r12 ) = F (r12 ) F () , (26)
since w(r) has been defined such that w(r ) = 0. The first and second laws of thermody
namics tell us for a reversible small change of state that
dF = SdT pdV + dN + Wrev . (27)
This implies that dF = Wrev = 12 w(r12 ) dr12 is the reversible (maximal) work done
by the N 2 particle system to achieve an infinitesimal displacement, dr12 , of the two inner
boundary particles at fixed temperature, system volume and particle number.
The function w(r; , T ) is in general a considerably more complicated object than the mere pair
potential u(r), since it involves the effects of particles 1 and 2 on the configurations of the other
particles. It is only in the limit 0 that
w(r) u(r) i.e. g(r) eu(r) , (28)
as will be shown further down. A sketch of w(r) for hard spheres at finite density is shown in
Fig. 5. As seen, two hard spheres effectively attract each other at distances r 1.5. This
manybody effect arises from an unbalance of forces on the two spheres when the gap between
the two particles is depleted from the other ones. Depletion interactions of this kind occur in
any system with strong excluded volume interactions. This is the reason why depletion effects
have attracted considerable interest in the past few years (see also section 4).
g(r)
0
r
r < 2
w(r)
Fig. 5: Potential of mean force of hard spheres (left) and depletion attraction (right).
FS02_8
Theories of Fluid Microstructures B2.11
is the triplet distribution function of an isotropic and homogeneous liquid. Furthermore, g (3)
depends only on the two pair separations r12 , r13 , and the angle between r13 and r12 . The mean
force on particle 1, in presence of particle 2 at a distance r12 , is thus the sum of the direct in
teraction between 1 and 2, and the interaction of 1 with a third particle at r3 , weighted by the
factor g (3) (r1 , r2 , r3 )/g(r12 ). The latter gives the probability of finding a particle at r3 , given
that there are particles at r1 and r2 . The equation given above is the lowest order one in the
socalled YvonBornGreen (YBG) hierarchy of equations relating reduced equilibrium proba
bility density functions of consecutive order. We see now that Eqs. (19) and (28) are limiting
cases of Eq. (30), for 0.
To first order in , the triplet distribution function factorizes according to g (3) (r1 , r2 , r3 ) =
g(r12 )g(r13 )g(r23 ), since the probability of finding three particles simultaneously within the
correlation distance is of order 2 . The substitution of this factorization approximation into
Eq. (30) results into a closed, nonlinear integrodifferential equation for g(r) which can be
solved numerically for a given u(r). This socalled BornGreen integral equation, however,
gives poor predictions for the g(r) of dense liquids [2]. More reliable schemes to calculate g(r)
based on the OrnsteinZernike equation will be discussed in section 3.
ki
laser
q
r i (t) kf
detector
For single and quasielastic scattering from spherical particles [2, 7],
where q = (4/) sin(/2) is the modulus of the scattering wave vector q = kf ki , and N
is the average number of particles in the illuminated volume part of the sample. The socalled
form factor P (q), normalized as P (0) = 1, contains information on the scattering material
distribution inside a particle, i.e., information on the particle size and form.
The most relevant quantity in Eq. (32) including information on interparticle correlations is
called the static structure factor S(q). Its statistical mechanical definition for q > 0 reads
2
N N
1 iq[rl rp ] 1 iqrl
S(q) = lim e = lim e 0 (33)
N N
l,p=1 l=1
where denotes an equilibrium ensemble average. The static structure factor is the Tlimit
of the autocorrelation function,
1
S(q) = lim SN (q) = lim q q , (34)
N N
N
q = eiqrl N q,0 , (35)
l=1
of microscopic density fluctuations of N particles. The prefactor 1/N renders S(q) into an
intensive quantity. We have denoted the structure factor of the finitevolume NVT ensemble as
SN (q), and periodic boundary conditions have been used.
By expanding the double sum into self, l = p, and distinct, l = p, parts, it can be shown that
iqr sin(qr)
S(q) = 1 + dr e h(r) = 1 + 4 dr r2 h(r) , (36)
0 qr
where, h(r), with
h(r) = g(r) 1 (37)
is called the total correlation function. Notice that h(r ) = 0, and S(q ) = 1. As a
result, the static structure factor determines the Fourier transform, h(q), of h(r). Since Fourier
transforms are onetoone mappings, S(q) can be inverted to determine h(r) and thus g(r):
[ ]
1 S(q) 1 1
g(r) = 1 + F (r) = 1 + dq q sin(qr)[S(q) 1] . (38)
2 2 r 0
In principle, this would require to measure S(q) for all wave numbers q where S(q) exhibits
significant oscillations. In light scattering experiments on colloids, this is usually not feasible
since the largest q value accessible is limited by qmax = 4/, which corresponds to backward
scattering.
We have used here the following basic result on Fourier transforms: suppose h(r) is an isotropic
function, with h(r) 0 sufficiently fast for r . The threedimensional Fourier transform,
h(q), of h(r) is then defined as
sin(qr)
h(q) := F{h(r)} = dr e h(r) = 4
iqr
dr r2 h(r) (39)
0 qr
Theories of Fluid Microstructures B2.13
where the third equality follows from performing two angular integrals in spherical coordinates.
Fourier inversion leads then to
1 1 iqr 1 sin(qr)
h(r) = F {h(q)} = 3
dq e h(q) = 2 dq q 2 h(q) . (40)
(2) 2 0 qr
For notational simplicity, Fourier transformed functions are distinguished from their realspace
counterparts by their argument q only.
q=k
/ 2
/ 2 d = 2 / k
Fig. 7: Firstorder Bragg diffraction from a sinusoidal particle density wave of wavenumber
k = 2/d. Here, is the scattering angle, and the wavelength of light in the solvent.
For an intuitive understanding of scattering from a fluidlike particle system we show in the fol
lowing that a measurement at a particular q corresponds to diffusive firstorder Bragg diffraction
from a spatially sinusoidal particle density fluctuation around = N /V of the form
provided that k = q, with a realvalued amplitude factor 0k . At each instant of time, a super
postion of such density fluctuations is thermally induced with wave vectors k of varying length
and direction. Viewed from a coarsegrained level of resolution, the superposition of density
waves
m (r, t) = mk (r, t) , (42)
k
distance d.
Consider now the scattering of a monochromatic radiation beam of wave vector q from such a
melange of density fluctuations. Note that Eq. (34) can be readily recast as
2
1
S(q) = lim dr exp{i q r} m (r) , (44)
N V
On identifying m (r) with its coarsegrained average m (r, 0), one notices from Eq. (44) that
S(q) is nonzero only for a density fluctuation of special orientation and wavelength such that
k = q. The planes of constant phase of this sinusoidal density fluctuation are oriented at an
angle /2 with respect to the incident radiation, with denoting the scattering angle. Diffraction
from these planes is governed by the Bragg condition (see Fig. 7)
2d sin(/2) = m (46)
expresses the internal energy, E, of a onecomponent N particle system with pairwise addi
tive U (rN ) in terms of u(r) and g(r). The internal energyis the sum of a kinetic ideal gas
part, (3/2)N kB T , and an interaction or excess part, U (rN ) . The latter can be understood on
physical grounds as follows: For each particle out of N , there are 4r2 g(r) dr neighbors in
a spherical shell of radius r and thickness dr, and the interaction energy between the central
particles and these neighbors is u(r). Integration from 0 to gives the interaction energy part
Theories of Fluid Microstructures B2.15
of E, with the factor 1/2 included to avoid double counting of particle pairs.
relates the thermodynamic pressure, p, of a liquid with pairwise additive U to an integral over
g(r) and the derivative, u (r), of the pair potential. Here, pid = kB T is the kinetic pressure of a
classical ideal gas. The excess pressure contribution, pex , due to particle interactions can be de
rived along the same lines as the energy equation using, e.g., the classical virial equation [1, 8].
For a repulsive pair potential where u (r) < 0, pint is a positive pressure contribution originating
from the enhanced thermal bombardment of container walls by the mutually repelling particles.
N
N
N
U (r ) = u(rij ) + u3 (rij , rik ) + , (50)
i<j i<j<k
in terms of twobody, threebody and higherorder interaction terms, the energy and pressure
equations generalize to [10]
( )
Eex 2
lim = dr u(r) g(r) + dr dr g (3) (r, r ) u3 (r, r ) + (51)
N 2 6
and
pex 2 u3 (r, r ) (3)
= dr r g(r) u (r) dr dr r g (r, r ) + (52)
pid 6 18 r
The compressibility equation can be derived in the grand canonical ensemble representing an
open system, at constant V and T , which allows for fluctuations in the particle number. This is
perfectly appropriate since S(q) is related to the intensity of quasielastically scattered radiation.
B2.16 G. Nagele
The radiation beam samples only a fraction of the system volume, and in this subvolume the
number of particles, while macroscopically large, fluctuates.
To derive the compressibility equation, we need thus the definition of the radial distribution
function in the grand canonical ensemble. For a given N , we know from Eq. (13) that
N gN (r) = N (N 1) dr3 rN PN (rN )
2
N 1
= N (N 1) dr3 rN PN (r ) dr1 dr2 (r r12 )
V
N
N (N 1) 1
= (r r12 )N = (r rij ) (55)
V V i=j
N
where N denotes the fixed N , canonical ensemble average, and N = N/V . The canonical
gN (r) has been formulated by the most right equality in terms of an ensemble average invoking
Dirac functions. To obtain the grandcanonical rdf, denoted by gV (r), we merely have to
replace the canonical average by the grand canonical one, denoted by gc :
( )N = drN PN (rN )( ) ( )gc := P (N )( )N . (56)
N
We have introduced here the grand canonical probability, P (N ), of finding a system with ex
actly N particles. The grandcanonical pair distribution function is thus given by
N
1
2
gV (r) := (r rij ) = P (N )2N gN (r) (57)
V i=j N =2
gc
with = N gc /V . Note here that gV (r) = gN (r) + O(1/N gc ). For a fixed N , we know from
Eq. (14) that
2 N (N 1)
N dr gN (r) = . (58)
V
For an open system of volume V , this leads to
[ ]
N (N 1) gc N 2
gc N 2
gc N 2
gc
2 dr [gV (r) 1] = = 1 (59)
V V N gc
The variance of the number fluctuations in an open, macroscopic system is related to the isother
mal compressibility by the thermodynamic relation [6, 8]
( )2
N N gc
gc
kB T T = . (60)
N gc
This completes our derivation of the compressibility equation, on noting that g(r) = lim gV (r)
in the thermodynamic limit of N gc and V with fixed. The contradiction
between the compressibility equation and the canonical ensemble result in Eq. (15) is only
apparent. On first sight the latter might suggest that limq0 S(q) = 0 is exactly valid. However,
Theories of Fluid Microstructures B2.17
Eq. (15) applies only to a closed, finitevolume system with zero particle number fluctuations.
Notice further, for N and V finite and fixed (at given density ), that physically allowed wave
numbers are restricted to q > V 1/3 : It makes no sense to consider particle density fluctuations
of wave lengths ( q 1 ) larger than the system size. Therefore, the thermodynamic limit of a
macroscopic system should be performed first, followed by the limit q 0. In summary,
lim S(q) = lim lim SN (q) = lim SN (q = 0) = 0 , (61)
q0 q0
where SN (q) is the NVT static structure factor defined in Eq. (34). In the integral of the
compressibility equation which relates to a thermodynamic quantity, one must use the radial
distribution function in the thermodynamic limit (see also [9]).
The particles of a liquid near the triple point of gasliquidsolid coexistence are densely packed
so that T is very small. Furthermore, the compressibility is nearly zero for a classical crystal
near T = 0, since there are hardly any vibrations of the atoms around their equilibrium posi
tions. In contrast, T diverges at a critical point which is the terminal point of the gasliquid
coexistence line. The divergence is accompanied by a long distance tail in h(r) which causes
the phenomenon of critical opalescence observed in light scattering studies near critical points.
We will study critical opalescence in section 3.2. Summarizing, we note
T 0, fluid near triple point
0, ideal crystal (62)
id
T
, fluid at critical point .
The reader should be warned that Eqs. (51) and (52) for E and p, and also the compressibility
equation as stated in Eq. (53), apply to stateindependent, i.e., (, T )independent pair inter
actions only. Special considerations and care are required for statedependent pair potentials
u(r; , T ) and, more generally, for a statedependent U (rN ) [1117]. Effective potentials of
this kind occur in sizeasymmetric colloidal mixtures, and in liquid metals, in the process of av
eraging out the particle degrees of freedom associated with all but a single component of large
particles, which may be colloidal particles in the case of colloidpolymer mixtures, or metal
ions in case of liquid metals. The mixture is hereby mapped onto a onecomponent system of
pseudoparticles, or decorated particles, whose interactions are governed by an effective, state
dependent interaction energy U ef f (rN ). This mapping on an effective onecomponent system
will be explained in section 4, and two prominent examples will be discussed.
In many practical situations, a mixture of large colloidal particles and small colloidal or poly
meric particles, dispersed in a solvent is in osmotic equilibrium with a reservoir of the small
particles and solvent by means of a semipermeable membrane impermeable to the large parti
cles. Since this fixes the chemical potentials, { }, of the small particles and the solvent, it is
most convenient to treat them grandcanonically in order to derive U ef f (rN ; { }), so that, in
case of a phase separation, U ef f remains in the same form in both phases [18]. It is remarkable,
that the osmotic compressibility of the noncontractedout large particles component is deter
mined by their pair correlations alone, independent of the largesmall and smallsmall particle
correlations, and independent of the socalled volume term (see section 4), through the osmotic
compressibility equation of Kirkwood and Buff [19]
( )
1 1
= . (63)
kB T T,N,{ } S(q 0)
B2.18 G. Nagele
Here, is the osmotic pressure of large particles relative to the reservoir of small particles and
solvent, and S(q) is the structure factor describing the largelarge particle correlations. More
over, the { } denote the externally controlled chemical potentials of the small particles and sol
vent in the reservoir. To obtain the thermodynamics and phase behavior of the whole mixture,
however, information on the pair correlations of all particulate components is required, includ
ing the largesmall and smallsmall ones. Eq. (63) can be obtained from the multicomponent
extension of the compressibility equation in Eq. (53) in combination with the thermodynamic
GibbsDuhem relation for the osmotic pressure of the large particles.
It introduces the direct correlation function, c(r), as a new function and can be viewed as the
definition of c(r) in terms of the total correlation function, h(r) = g(r) 1, of two particles a
distance r = r12 apart. Eq. (64) can be recursively solved for h(r12 ) to give
h(r12 ) = c(r12 ) + dr3 c(r13 )c(r23 ) + 2
dr3 dr4 c(r13 ) c(r24 ) c(r34 ) + O(c4 ) . (65)
This leads to the following physical interpretation of the OZ equation: the total correlations be
tween particles 1 and 2, described by h(r12 ), are due in part to the direct correlations, c(r12 ), of
these particles but also to an indirect correlation propagated by direct correlations via increas
ingly large numbers of intermediate particles.
Once information is available about c(r) in form of a closure relation involving u(r), the OZ
equation can be viewed also as a closed integral equation for h(r). Some information on c(r)
derives from Eq. (64) in the lowdensity limit 0 where c(r) h(r). With Eq. (19) follows
then
c(r) f (r) for 0 , (66)
where
f (r) := eu(r) 1 (67)
Theories of Fluid Microstructures B2.19
for r . Without proof we note that the longdistance asymptotic result (68) holds true for
a wide class of pair potentials even at finite densities. The range of c(r) is thus comparable
with that of u(r), and the fact that h(r) is generally longer ranged than u(r) can be ascribed
to indirect correlation effects. One word of warning: we refer here and in what follows to
onecomponent liquids of electrically neutral particles. Ionic fluids must be distinguished from
neutral fluids in that the effect of screening (cf. subsection 4.2) in such systems is to cause
h(r) to decay exponentially at large r, whereas c(r) still has the range of the infinite Coulomb
potential and therefore decays as r1 . In principle, ionic liquids consist of at least two charged
components of opposite sign to enforce overall charge neutrality.
using r = r12 , r = r23 and r r  = r13 . Fourier transformation of both sides of the OZ
equation leads to
h(q) = c(q) + c(q) h(q) (70)
or
c(q)
h(q) = , (71)
1 c(q)
where c(q) is the threedimensional Fourier transform of c(r). From noting that S(q) = 1 +
h(q), we obtain S(q) in terms of c(q):
1
S(q) = 0, (72)
1 c(q)
from which we learn that c(q) 1.
In the derivation of Eq. (70), we have employed the convolution theorem of Fourier transfor
mation theory. The convolution (german: Faltung), f1 f2 , of two integrable functions f1 (r)
and f2 (r) is defined as
(f1 f2 )(r) := dr f1 (r )f2 (r r ) = dr f2 (r )f1 (r r ) .
(73)
i.e., the Fourier transform of the convolution of two functions is equal to the product of their
Fourier transforms.
Using Eq. (72), we finally obtain the compressibility equation in terms of c(q):
1
= 1 c(q 0) = 1 4 dr r2 c(r) . (75)
kB T T 0
B2.20 G. Nagele
V T
Fig. 8: Schematic p V and p T phase diagrams of a LennardJones system (Argon).
critical temperature Tc and pressure pc . It is the location of a continuous (i.e., 2nd order) phase
transition. On approaching the critical point along the coexistence line, the (density) differ
ence between liquid and gas phases ceases to exist. At the critical point, strong and longliving
density fluctuations occur (cf. Eq. (60)) such that T becomes arbitrarily large in the thermo
dynamic limit (cf. Fig. 8):
( )
p
= 0 i.e. T = . (76)
V T,N
which means that S(q) becomes very large for small q as the critical point is approached. Then,
regions of larger and smaller densities develop. As the size of these regions approaches the
wavelength of visible light, there is in fact so much scattering that the fluid appears cloudy or
opalescent. This phenomenon is therefore called critical opalescence. It is due to the occurrence
of longrange spatial correlations between particles in the vicinity of a critical point such that
the volume integral over h(r) diverges in the limit of an infinite volume. As T Tc ,
1
c(q 0) = 4 dr r2 c(r) (78)
0
Theories of Fluid Microstructures B2.21
which means that, contrary to h(r) and g(r), c(r) remains shortranged with finite second mo
ment.
Since c(q) is well behaved at Tc , we assume that it can be expanded in a truncated Taylor series
around q = 0 up to O(q 2 ). Using sin(x)/x 1 x2 /6 + O(x4 ), c(q) is thus approximated, for
small q, by
sin(qr)
c(q) = 4 dr r2 c(r) = c0 c2 q 2 + O(q 4 ) (79)
0 qr
with expansion coefficients
c0 = 4 dr r2 c(r) 1 for T Tc (80)
0
2
c2 = dr r4 c(r) . (81)
3 0
The coefficient c2 is positive valued provided the attractive tail of u(r) is sufficiently long
ranged. We restrict here our attention to small q and thereby to large distances r. Note that
c0 = c(q = 0) 1. Substitution of this truncated expansion into Eq. (72) then gives
1 1 1
S(q) = (82)
1 c0 + c2 q 2 c2 2 + q 2
as an approximation of S(q) for small q, more precisely for qRU 1. This is the smallq
approximation for the nearcritical structure factor originally proposed by Ornstein and Zernike
around 1914. The range of the interaction potential is denoted as RU , and we have further
introduced the correlation length
( )1/2 ( )1/2
c2 1/2 T
(T ) := = [c2 S(0)] = c2 id , (83)
1 c0 T
by assuming the fourth moment, c2 , of c(r) to exist and to be positive. With the compressibility
diverging for T Tc in a powerlaw fashion as T (T Tc ) , it follows that the corre
lation length diverges as (T Tc )/2 . The numerical value of the critical exponent is
predicted as = 1 in a simple Landautype meanfield approximation. However, the accepted
value of for the liquidgas transition is 1.24, as derived from renormalization group calcula
tions of critical phenomena and verified by highprecision experiments on critical fluids [20,21].
Fourier inversion of the OZ approximation for S(q) gives the asymptotic form of h(r) for large
pair separation r RU
1 iqr 1 er/
h(r) = dq e [S(q) 1] . (84)
(2)3 4 c2 r
At the critical temperature, becomes infinite with
1
h(r)T =Tc for r RU (85)
r
which corresponds to
1
S(q)T =Tc for q RU 1 . (86)
q2
B2.22 G. Nagele
The total correlation function decays thus algebraically slow, and not exponentially fast, at the
critical point.
1
I(q)
q2
Fig. 9: Experimental OrnsteinZernike plot of argon. From Ref. [22].
Experiments performed very close to Tc reveal in fact deviations at small q from the OZ approx
imation prediction for S(q). Very close to the critical point the data for 1/S(q) versus q 2 bend
slightly downwards for q 2 0. Improved modern renormalization group theory calculations of
critical phenomena lead to a corrected asymptotic scaling behavior of h(r) at Tc , given in three
dimensions by [20, 21]
1
h(r)T =Tc 1+ (88)
r
or, equivalently,
1
S(q)T =Tc 2 , (89)
q
with a small but nonzero critical exponent . The accepted theoretical value for is 0.04. The
reason for the failure of the OZ approximation very close to Tc is that, by assuming a truncated
smallq 2 expansion of c(q) in the neighborhood of q = 0 to hold true, one does not account for
Theories of Fluid Microstructures B2.23
the full spectrum of correlation fluctuations existing at all length scales. In the thermodynamic
limit, fluctuations of very long wavelengths become possible so that at a critical point the free
energy and related twopoint correlation functions such as c(r) become nonanalytic in their
dependence on the thermodynamic variables and r respectively. It should be noticed here that
the higherorder coefficients cn , with n = 4, 6, , appearing in the regular smallq expansion
of c(q), cease to be finite when u(r) has a van der Waals tail. We finally remark that the liquid
gas transition of simple liquids at Tc belongs to the same universality class of secondorder
phase transitions as the ferromagneticparamagnetic transition in uniaxial ferromagnets (cf. the
threedimensional Ising model). All members of the same universality class show the same
critical exponents.
For a system with a hardcore excluded volume part in u(r), any closure relation should be
consistent with the exact condition
which states that two spheres of hardsphere diameter can not interpenetrate, and the asymp
totic result
c(r) = u(r) , for r (91)
valid for a wide class of pair potentials.
The exact asymptotic form of c(r) forms the basis of the socalled meanspherical approxima
tion, first introduced into liquid state theory by Lebowitz and Percus (1966). In MSA, c(r) is
assumed to be given approximately by the closure relation
for all nonoverlap distances r > . Together with Eqs. (90) and (92), the OZ equation (69)
becomes a linear integral equation determining g(r) for r > , and c(r) for r < . The most
attractive feature of the MSA closure, as compared to other ones, is that analytic solutions exist,
even in the manycomponent case, for various pair potential models, namely for the hard and
sticky hardsphere potentials, the square well potential, the Coulomb potential, attractive and
repulsive Yukawatype potentials, and for the dipolar hardsphere potential. These potentials
B2.24 G. Nagele
are of particular interest for molten salts, electrolyte solutions and in colloid science. No ana
lytic MSA solution exists for the LennardJones potential.
While the MSA is well suited for shortrange attractive and repulsive potentials, it can predict
nonphysical negative values for g(r) close to contact distance in case of dilute systems of
strongly repelling particles. At very low density (more precisely, small volume fractions ), the
MSA predicts that
with a negative g(r) for u(r) > 1. This unphysical prediction should be contrasted with
the exact zerodensity form of g(r) given in Eq. (19). Recall here, that the volume fraction
= (/6) 3 is defined as the fraction of the system volume filled by the spherical particles.
Rescaled MSA:
For fluids of (colloidal) particles, where the physical hard core is masked by strong and long
range repulsive forces, there exists an improved variant of the MSA which preserves the positive
semidefiniteness of g(r). This variant is called the rescaled MSA (Hansen and Hayter, 1982).
It is based on the fact that the g(r) of such systems is continuous at all distances r. Moreover
and most importantly, two particles in these systems are virtually never closer to each other than
a certain distance > , so that g(r) 0 for r < .
g(r)
T
r/
Fig. 10: Radial distribution function of a chargestabilized Yukawatype dispersion. Compari
son between RMSAg(r) and Monte Carlo computer simulation results. From Ref. [3].
In RMSA, the actual system is replaced by a fictitious system consisting of particles of enlarged
diameter > , at the same number density and, for r > , with the same pair potential
u(r) than the original one. The (densitydependent) effective diameter, , is determined from
the continuity of g(r) at r = , by demanding that
g(r = ; , ) = 0 , (94)
Theories of Fluid Microstructures B2.25
with g(r) calculated in MSA for a larger rescaled volume fraction = ( /)3 > . The
RMSAg(r) is positive semidefinite since the volume fraction, , of the fictitious system is so
much larger than that Eq. (93) does not apply any more.
Fig. 10 includes the RMSAg(r) for a dilute aqueous suspension of highly charged polystyrene
spheres interacting by the Yukawatype DLVO potential of Eq. (3). The charge number,
ZRM SA = 257, employed in the RMSA calculation has been selected such that the height
of the principal peak of the RMSA g(r) is coincident with that of the exact g(r), gener
ated by Monte Carlo (MC) computer simulations using a smaller charge number ZM C = 205.
Since ZRM SA > ZM C , the RMSA underestimates the structural ordering in systems of strongly
correlated particles. However, once Z has been adjusted to fit the actual peak height, the over
all shape of g(r) is remarkably well predicted by the semianalytical RMSA solution. The
RMSA has been extended to multicomponent systems of mutually repelling Yukawa particles
(cf. Ref. [3]). The accuracy of the RMSA in its prediction for the effective charge number Z
can be further improved by correcting for the penetrating background of a uniformly assumed
microion density which maintains electroneutrality [23].
Aside from the linear MSA, there exist a variety of nonlinear integral equation schemes. The
PercusYevick approximation (Percus and Yevick, 1958) is among the most popular ones. To
introduce the PY closure relation, we reformulate the OZ equation as
[ ]
c(r) = g(r) 1 + dr c(r ) {g(r r ) 1} =: g(r) gind (r) . (95)
The term in brackets, gind (r), describes the indirect part of the pair correlations. Since g(r) =
exp[w(r)], one can approximate gind (r) by
gind (r) = 1 + (c h) (r) e[w(r)u(r)] , (96)
or, equivalently, c(r) by
[ ]
c(r) g(r) 1 eu(r) = g(r) y(r) = f (r)y(r) . (97)
This is the PY closure relation for c(r). We have introduced here the socalled cavity function,
y(r), defined as
y(r) := eu(r) g(r) . (98)
Contrary to g(r), which for hard spheres has a jump discontinuity at r = due to the factor
exp[u(r)], y(r) is continuous for all r. It agrees with g(r) for all r where u(r) = 0. In PY
approximation, c(r) is thus assumed to be zero at distances where the pair potential vanishes.
The continuity of y(r) is easy to see, using Eq. (13), from noting that
[ ]
V2 N
y(r12 ) = lim dr3 drN exp u(rij ) (99)
ZN
i<j
for pairwise additive forces, with the pair {i, j} = {1, 2} omitted from the sum. Hence y(r) is
a smooth continuation of g(r) into the overlap region (cavity) r < .
B2.26 G. Nagele
Substitution of Eq. (97) into the OZ equation gives the nonlinear PY integral equation
[ ][ ]
u(rr ) u(r )
y(r) = 1 + dr e y(r r ) 1 e 1 y(r ) (100)
for y(r) or, likewise, g(r). This equation can be solved analytically in three dimensions for
the important case of hard spheres (cf. section 3.4), and by numerical methods for arbitrary
pair potentials. In contrast to the HNC approximation, the PY approximation can not exclude
unphysical negative values in g(r). Although the PYg(r) of mondisperse hardsphere fluids is
nonnegative at all distances r, negative values of certain PY partial radial distribution functions
of strongly sizeasymmetric hardsphere mixtures can be found at some intermediate distances.
However, the PY predicted contact values in these asymmetric systems remain positive valued.
The PY approximation is exact to first order in the particle density. For a proof expand the YBG
equation in Eq. (30) to linear order in , using
with Mayers f function, and the definitions i := ri and ij := rij . This gives
[ ]
1 [w(12) u(12)] = lim 1 d3 f (13) f (23) + O(2 ) (102)
so that
[w(r) u(r)] = (f f ) (r) + O(2 ) . (103)
The Tlimit of an infinite system should be taken after differentiation with respect to r1 , with
the consequence that the constant terms coming from the factorization of (f 1)(f 1) drop
out. Expansion of the PYclosure in Eq. (96) to linear order in gives the same result, showing
that g(r) in PY approximation is exact to first order in density. Since
[ ]
g(r) eu(r) = eu(r) e(w(r)u(r)) 1 eu(r) [w(r) u(r)] + (2 ) , (104)
for the lowdensity form of the direct correlation function. The common factor f (r) in the virial
expansion of c(r) indicates that its range for nonionic systems is essentialy that of u(r).
Another frequently used approximate integral equation scheme is the hypernettedchain ap
proximation (van Leeuwen et al., 1959). The name stems from its diagrammatic derivation.
The HNC closure relation in terms of c(r) is
Exponentiation gives
g(r) eu(r)+h(r)c(r) , (108)
showing that in HNC approximation the positive definiteness of the exact g(r) is preserved at
any density. Since from Eq. (84)
the HNC approximation leads further to the correct asymptotic behavior of c(r) for arbitrary .
On the other hand the PYc(r) gives the correct longdistance behavior in general only for small
densities.
This is the nonlinear HNC integral equation for g(r). It can be solved only numerically even
for hard spheres. Like in the PY approximation, the HNC approximation predicts g(r) correctly
to first order in the density.
The PY is quite successful for hard spheres or, more generally, for particles with shortrange
hardspherelike interactions. In contrast to the MSA, however, it does not work so well for
systems with attractive tails. The HNC is complementary to the PY in the sense that it is un
satisfactory for hard spheres but appears to account satisfactorily for the effects of soft cores
and, in particular, for longrange repulsive potential tails as given in ionic fluids and dispersions
of Yukawa particles. All three integral equation schemes have severe deficiencies close to a
gasliquid critical point.
HNC results for a Yukawa system (cf. Eq. (3)) of moderately charged colloidal particles (Z =
107) of diameter = 160 nm and fixed screening parameter in an organic solvent ( = 10)
are shown in Fig. 11. The charge Z was determined from a fit of the HNC peak height of S(q)
to the experimentally given one. There is then rather good agreement between the theoretical
and experimental S(q). The deviations at small q and around the minimum can be attributed
to polydispersity effects, that is to a spread in the experimental particle sizes. With increasing
volume fraction , there is increasing ordering visible through more pronounced undulations
in g(r), and the system becomes less compressible (decreasing S(0)). The particles avoid each
other as much as possible because of the strong and longerranged electrostatic repulsion so that
g(r < 1.5) = 0. In monodisperse systems with longrange repulsion, the position, qm , of the
principal peak of S(q) increases with volume fraction approximately as qm 1/3 . Away from
a critical point, the location, rm , of the main peak of g(r) is approximately related to qm by
2
rm . (112)
qm
B2.28 G. Nagele
= 0.101
S(q)
g(r)
r/ q [107 m1 ]
Fig. 11: HNC radial distribution function g(r) at various volume fractions (left), and static
structure factor S(q) (right) of chargestabilized dispersions of silica spheres. Open circles:
light scattering results of S(q). From Ref. [24].
Suppose we can separate the pair potential of a liquid system into a shortrange reference part,
u0 (r), and a soft and longrange perturbational part ul (r) (cf. Fig. 12),
where ul (r) is Fourierintegrable. Let us further assume that the direct correlation function,
c0 (r), of the reference system (where u = u0 ) is known exactly or to a good approximation. For
a reference system of hard spheres, e.g., we could use the analytic PY solution for c0 (r) derived
in section 3.4. The true direct correlation function, c(r), of the system can then be approximated
u 0 (r )
1 r
u1 (r)
Fig. 12: Pair potential consisting of hardsphere reference part, u0 (r), and a longerranged,
attractive perturbational part ul (r).
by
c(r) c0 (r) ul (r) , r > 0 (114)
Theories of Fluid Microstructures B2.29
which is asymptotically correct at long pair separations. For historical reasons, this closure
relation is referred to as the random phase approximation. In this approximation one obtains
1 1
S(q) = . (115)
1 c(q) 1 c0 (q) + ul (q)
The static structure factor is thus expressed in terms of the structure factor, S0 (q), of the refer
ence system and the Fourier transform,
ul (q) = dr eiqr ul (r) , (116)
0
of the perturbational part of u(r). Eq. (115) can be rewritten in the mnemonic form
1 1
+ ul (q) . (117)
S(q) S0 (q)
The RPA is the most simple perturbation theory for fluid microstructures, usually suited only for
small wave numbers. It has been successfully used, also in its multicomponent extension, for
the calculation of (partial) structure factors in dense systems with (ultra) soft potentials where
u0 (r) = 0, such as polymer blends and star polymers [25, 26]. For an ideal gas as reference
system, one has c0 (r) = 0 and S0 (q) = 1. The RPA reduces then to the MSA for pointlike
particles, referred to in the literature as a version of the DebyeHuckel approximation. One
should notice that the perturbation must be sufficiently weak, or the density sufficiently low, to
ensure that ul (q)S0 (q) > 1 with S(q) > 0. Furthermore, the RPA does not ensure that
g(r < ) = 0 in case of a hardsphere reference system. This nonphysical feature of the RPA
is related to an ambiguity in the choice of the perturbation potential ul (r) in Eq. (116) for r < .
The true g(r) should not depend on this choice.
In the optimized random phase approximation (ORPA), ul (r) is extended into the hardcore
regime r < such that g(r < ) = 0. While the ORPA is a considerable improvement of the
RPA, there is a price to pay in form of a much larger numerical effort to calculate S(q).
in PY approximation, i.e. the hardsphere direct correlation is approximately set equal to zero
for nonoverlap distances. As a matter of fact, the true c(r) has a small but nonvanishing tail
for r > . As can be noticed here, the PY closure is identical to the MSA closure (cf. Eq. (92))
in the case of a hardsphere fluid.
B2.30 G. Nagele
Upon inserting the hardsphere potential into Eq. (100), one obtains a quadratic integral equa
tion for y(r) of the form
y(r) = 1 + dr y(r ) dr y(r ) y(r r ) . (120)
r < r <, rr >
It is required to solve this integral equation within r < for c(r) = y(r), since c(r > ) = 0
is known already. Following Wertheim [27], we use a thirdorder polynomial
c(r) = a0 + a1 r + a2 r2 + a3 r3 (121)
as a trial solution of c(r < ), with yet unknown densitydependent expansion coefficients {ai }.
This ansatz is suggested from the low density form of c(r), which is a third order polynomial
in case of hard spheres. For a proof of this statement use Eq. (65) to show that to first order in
density, y(r) is given by
y(r) = 1 + dr f (r ) f (r r ) + O(2 ) . (122)
For hard spheres, the Mayerf function is f (r) = 1 for r < and zero otherwise. The
convolution integral in Eq. (122) is then equal to the volume of overlap of two spheres of equal
radii with centres separated by r. As a consequence
[ ]
3 1 3
y(r) = 1 + 8 1 x + x (2 x) + O(2 ) , (123)
4 16
with x = r/. The overlap volume is zero for x > 2 as expressed by the unit step function
(2 x). Recall that the PY approximation describes g(r) exactly to first order in . In using
the polynomial ansatz in Eq. (121), it is assumed that the functional form of c(r) is valid for all
volume fractions.
The four expansion coefficients, {ai }, are determined by employing the continuity of y(r) and
its first two derivatives at r = . Their continuity follows from Eq. (120) and its first two
derivatives. A fourth condition follows from the PY integral equation (120) evaluated at r = 0:
y(0) = 1 + dr y(r ) . (124)
r <
After inserting Eq. (121) in Eq. (120) and making use of the four boundary conditions to
determine the {ai }, a lengthy calculation gives the following PY result for the hardsphere c(r):
[ ( ) ]
1
c(r < ) = 1 1 + x + 2 x 3
(125)
2
Theories of Fluid Microstructures B2.31
with
(1 + 2)2 6 (1 + 0.5)2
1 = , 2 = . (126)
(1 )4 (1 )4
The PY result for c(r) reduces, for small , to the correct first order density form of y(r)
given in Eq. (123).
g(r)
c(r)
0 1 2 3 4 5
r/ r/
Fig. 13: PercusYevick direct correlation function (left) and radial distribution function (right)
of hard spheres.
Fourier transformation of c(r) leads with Eq. (72) to an analytic expression for S(q). This
expression reads explicitly [3]
1
S(y) = 2 (127)
X (y) + Y 2 (y)
with
X(y) = 1 12 [ A f1 (y) + B f2 (y) ] (128)
Y (y) = 12 [ A f3 (y) + B f4 (y) ] , (129)
where
1 + 2 1 + 0.5
A= , B= , (130)
(1 )2 (1 )2
and
y sin(y) cos(y) 1
f1 (y) = , f2 (y) = (131)
y3 y2
f2 (y) 1
f3 (y) = + , f4 (y) = yf1 (y) . (132)
y 2y
We have introduced here the reduced wave number y = q.
The reduced isothermal compressibility follows as
(1 )4
lim S(q) = , (133)
q0 (1 + 2)2
which is a monotonically decreasing function in . For given analytical S(q), the hardsphere
g(r) can be determined in principle by numerical Fourierinversion. However, to avoid prob
lems caused by the jump discontinuity in g(r), it is safer to calculate first the function (r) :=
B2.32 G. Nagele
S(q)
r/
q
Fig. 14: PercusYevick static structure factor of hard spheres.
6
= 0.49
5
MD
4 PYVW
g(r) PY
3 RY
0
2 3 4 5
r/a
Fig. 15: Hardsphere g(r) for a volume fraction = 0.49 close to the freezing transition.
Comparison between PY, RY, VerletWeis corrected PY and MD computer simulations (filled
circles). From Ref. [28].
h(r) c(r) by Fourierinverting (q) = c(q)h(q) = [S(q) 1]2 /(S(q)). The hardsphere
g(r) follows then in PY approximation from g(r) = y(r) = 1 + (r) for r > . Contrary to
g(r), (r) and its first two derivatives are continuous also at r = as one can deduce from
the OZ equation. Notice here that the identity y(r) = 1 + (r) holds true only within PY ap
proximation. The relation between y(r) and (r) is approximated in the HNC approximation
by y(r) = exp{(r)}, which agrees with the corresponding PY relation for small (r) only.
While the full PYg(r) of hard spheres can not be represented analytically, one can derive closed
expressions for the contact values of g(r) and its first derivative from the continuity of y(r) and
its derivative at r = :
1 + 0.5
g(r = + ) = , (134)
(1 )2
and
dg 4.5 (1 + )
(r = + ) = . (135)
dr (1 )3
PY results for the hardsphere c(r) and g(r), and for the static structure factor, are shown in
Figs. 13 and 14, respectively, for various volume fractions. For = 0.1, we further show the
Theories of Fluid Microstructures B2.33
cavity function with y(r) = c(r) for r < . The PY approximation provides a quite good
representation of the true hardsphere S(q) and g(r) for volume fractions 0.35. At larger
values of it underestimates the contact value of g(r) and its oscillations are slightly out of
phase, as can be seen from Fig. 15 in comparison with Molecular Dynamics (MD) computer
simulation results. The PY approximation further fails to predict the liquidsolid freezing tran
sition which occurs for hard spheres at f = 0.494. This failure is not restricted to the PY
approximation: The OrnsteinZernike integral equations discussed in this lecture have been de
signed to describe the homogeneous fluid state, and are thus not suited to describe the density
inhomogeneities and the symmetry change in a firstorder liquidcrystal phase transition.
On the basis of the analytic PY solution, Verlet and Weis [29] provide a simple prescription to
obtain results for the hardsphere g(r) and S(q), which are in good agreement with computer
simulation results even up to the freezing volume fraction (see also [30]).
For a given and , the VerletWeis corrected g(r) is given by
(1)x
e
gV W (x; ) = gP Y (x ; ) + A( ) cos [ (x 1)] , (136)
x
with x = r/, and a rescaled volume fraction and diameter = ( /)1/3 . The three
parameters A, , and are determined on demanding that gV W ( + ) = (pCS /pid 1) /4 =
(1 0.5 ) / (1 )3 and SV W (q = 0) = kB T (/pCS )T in agreement with the Carnahan
Starling equation of state for the hardsphere pressure given in Eq. (140), and from minimizing
the integral over gM D (r/; ) gP Y (r/ ; ) for an interval ranging from r/ = 1.6 to 3.0.
Here, gM D (r) is the exact rdf obtained from Molecular Dynamics (MD) simulations. This
leads to = (1 /16), and to A and expressed as simple functions of [29]. Fig. 15
compares the rdf predictions of various integral equation schemes, including the VWcorrected
PY and RY schemes, for a hardsphere system at volume concentration = 0.49 close to
freezing. The principal peak height of S(q) at wave number qm , predicted by the VWcorrected
PY scheme, is well described by the expression [7, 31]
(1 0.5)
S(qm ) = 1 + 0.644 , (137)
(1 )3
and conforms with the empirical HansenVerlet freezing criterion [32]. According to this cri
terion, a threedimensional fluid freezes into a solid when the principal peak height, S(qm ), of
S(q) is located in between 2.8 to 3.1. The precise height depends to some extent on the range
of the pair potential, and amounts to S(qm ; f = 0.494) = 2.85 in the case of hard spheres.
For dispersions of strongly repulsive colloidal particles, this criterion becomes equivalent to a
dynamic freezing criterion [33], as has been shown on the basis of a dynamic mode coupling
theory [34]. For an alternative analytic parametrization of the rdf of hardsphere fluids see [35].
To illustrate the thermodynamic inconsistency of the PY approximation for the case of hard
spheres we integrate the compressibility in (133) with respect to . This yields the compress
ibility equation of state
pc 1 + + 2
= . (138)
pid (1 )3
Using instead the pressure (or virial) equation of state, Eq. (49), one obtains
pv 1 + 2 + 32
= 1 + 4 g(r = + ) = , (139)
pid (1 )2
which agrees with the pressure, pc , derived from the compressibility equation only up to third
order in the volume fraction. The pressure Eq. (49) is referred to also as virial equation of
state, since it can be derived from the virial theorem of classical mechanics. The first equality
in Eq. (139) relating the pressure to the contact value of g(r) is an exact relation stating that p =
kB T [1 + 4 g( + ; )]. There is thus only a trivial temperature dependence of the pressure in
the equation of state of hard spheres which arises solely from the ideal gas contribution.
The energy equation, Eq. (48), leads for hard spheres to the exact result E = Eid = N kB T ,
with zero excess internal energy coming from particle interactions. Thus, the excess pressure
can not be derived using the energy equation, since the internal energy of hard spheres is of
purely kinetic origin. As a consequence, the phase diagram of hard spheres is temperature in
dependent and depends only on (athermal system). Note that the pressure and the elastic bulk
and shear moduli are measures of the energy density and thus scale for hard spheres as kB T
times a function of alone.
The phase transitions in a hardsphere fluid are purely entropy driven, since Aex = T Sex , so
that a stable phase at a given has maximal entropy. A system of monodisperse hard spheres
is in a supercritical fluid state for < f 0.494, where f is the freezing volume frac
tion, followed by a fluidcrystal coexistence region for f < < m 0.54 where m is
the melting volume fraction. In thermodynamic equilibrium, a fully crystalline facecentered
cubic (fcc)phase, characterized by 12 nearest neighbors, is found for m < < cp . Here,
cp = / 18 0.7404 is the maximal packing density allowed for monodisperse spheres,
as conjectured by Johannes Kepler already in the year 1611. However, the mathematical proof
that fcc is the densest packing of monodisperse, nondeformable spheres was given no earlier
than in 1998 by Thomas Hales, about three hundred years past its conjecture. For comparison,
the closestpacked volume fractions
for a simple cubic (sc) and a bodycentered cubic (bcc)
crystal are /6 0.52 and 3/8 0.68, respectively. A hexagonal cubic crystal has the
same closest packed volume fraction as a fcc crystal. However, a fcc hardsphere crystal is
thermodynamically more stable than a hc crystal [36]. While the pressure in the fluid branch
of a hardsphere fluid is well described by the CarnahanStarling equation of state given in Eq.
(140), a decently good parametrizations of the fcc branch, valid for > m , has been provided
by Wood [37] and Hall [38]. The orientationaveraged pair distribution function, gf cc (r), of the
fcc hardsphere phase has been parameterized by Kincaid and Weis [39].
Real suspensions of colloidal hard spheres are always polydisperse to a certain degree, and nu
cleation often becomes exceedingly slow so that for 0.58 < < rcp 0.634 the system is
Theories of Fluid Microstructures B2.35
trapped in a glasslike, nonequilibrium phase without longrange ordering. Here, rcp is the
volume fraction of random close packing [40] where a sphere has on the average six contacts
with neighboring ones. Both at random close packing and crystal close packing the particles get
immobilized so that the pressure at concentrations close to these two concentrations diverges
like p 3kB T /( rcp ) and p 3kB T /( cp ), respectively.
Contrary to the threedimensional case where rcp < cp , and consistent with the occurrence
of a glassy phase, the local rule of starting from a single sphere and placing the next one such
that the (areal) density is maximised leads precisely to the areal volume fraction, 2D , of a
twodimensional hexagonal lattice. The latter is the one of the maximal packing configuration
with 2D
cp = / 12 0.9069.
p
1
pid
Fig. 16: Hardsphere compressibility and pressure (virial) equations of states in PY and HNC
approximations. Dashed line: exact pressure curve. From Ref. [1].
The PY and HNC compressibility and pressure (virial) equation of states for hard spheres are
plotted in Fig. 16, in comparison with the exact pressure curve obtained from computer
simulations. The exact hardsphere equation of state is very well described in the fluid regime
( 0.494) by the CarnahanStarling (CS) formula [1, 2]
[ ]
pCS 1 1 2 1 + + 2 3
= pv + pc = . (140)
pid pid 3 3 (1 )3
As can be noticed from Fig. 16, the exact pressure is bracketed by pc and pv , with the difference
between pc and pv increasing for increasing . The PY solution is obviously a better approxi
mation for hard spheres than the HNC. The fluidsolid coexistence pressure obtained from the
CS formula is pCS /pid (f ) = 12.48 corresponding to g( + ; f ) = 5.81.
FS02_22
B2.36 G. Nagele
Several hybrid integral equation schemes have been proposed in the past which partially restore
thermodynamic consistency. Out of these schemes, we discuss here only the one proposed
by Rogers and Young [41], which interpolates between the PY and HNC approximations and
removes part of their thermodynamic inconsistencies. It is based on the observation (cf. Fig.
17) that computer simulation data for the S(q) of particles with purely repulsive pair potentials
are bracketed, near qm , by the PY and HNC structure factors. The RY closure relation is given
by { }
u(r) 1 [ f (r)[h(r)c(r)] ]
g(r) e 1+ e 1 (141)
f (r)
with a mixing function,
f (r) = 1 er , (142)
including a mixing parameter {0, }. The closure relation is constructed in such a way
that for
r or 0 : f (r) 0 RY PY (143)
r or : f (r) 1 RY HNC .
Hence the RY y(r) reduces to its PY value for r 0 and to its HNC value for r . This
3
PY
RMSA
HNC
2 RY
MC
g(r)
0
1 2 3
r/
Fig. 17: Radial distribution function of a chargestabilized, Yukawalike suspension with =
0.2 and Z = 100. Comparison of RY, HNC, RMSA and PY predictions with MC simulations.
From Ref. [28].
is consistent with the observation that, while the HNC closure is correct at large separations,
the PY approximation is expected to be more reliable at small r, at least for strongly repulsive
potentials. The parameter determines the proportion in which HNC and PY are mixed at
intermediate r. Its numerical value follows from requiring partial thermodynamic consistency
by demanding the equality, pT = cT , of the compressibilities obtained from the pressure and
compressibility equations od state. Since T is directly related to the longwavelength limit of
the static structure factor, one may expect that the RY approximation will provide quite reliable
results for S(q) at finite q.
Theories of Fluid Microstructures B2.37
The RY mixing scheme has been found to perform very well for threedimensional liquids with
purely repulsive pair potentials, unless the system is very close to the freezing transition line.
Its predictions of the pair structure are less precise for twodimensional systems. Moreover, the
RY scheme is less accurate for ultrasoft potentials such as the one describing star polymers,
since its closure is PYlike at small particle separations [26].
Fig. 17 includes a comparison of RY, HNC and PY results for g(r) with Monte Carlo (MC) sim
ulations of a threedimensional colloidal dispersion of Yukawa spheres. The true MonteCarlo
g(r) is strongly overestimated by the PY approximation, whereas the fluid g(r) is somewhat
underestimated by the HNC approximation, and to an even larger extent by the RMSA. The RY
approximation, on the other hand, reproduces the MC data quite well.
u(r) u (r)
g(r) g (r)
c(r) c (r)
h(r) h (r) = g (r) 1 (144)
are needed for a generalization to a mcomponent mixture. There are m(m + 1)/2 partial pair
potentials u (r), radial distribution functions g (r), total correlation functions h (r), and
direct correlation functions c (r) (with , {1, . . . , m}) necessary to characterize the fluid
pair structure. Here, u (r) is the pair potential of two particles belonging to component
and , respectively. Furthermore, g (r) gives the relative conditional probability of finding
an type particle a distance r apart from a given type particle. The partial pair potentials
are obviously symmetric in the two component indices, i.e. u (r) = u (r). The remaining
functions in Eq. (144) are thus also symmetric in their component indices.
m
h (r) = c (r) + dr c (r r ) h (r ) . (145)
=1
B2.38 G. Nagele
m
h (q) = c (q) + c (q) h (q) . (146)
=1
The total correlation function, h , between two particles of components and is thus written
as the sum of a direct correlation part, c , and an indirect correlation part mediated through all
other particles of components = 1, , m with relative density weight .
Since there are m(m + 1)/2 unknown functions, c (r), in the OZ equations one needs the
same number of closure relations to obtain a complete set of integral equations determining the
partial radial distribution functions. We quote here only the multicomponent generalizations
of the MSA relations. All the other integral equation schemes discussed previously can be
generalized accordingly to mixtures. The multicomponent MSA closure relations are [3]
and the OZ equations, one obtains a closed set of integral equations for the h (r). In place of
a single static structure factor, there are now m (m + 1) / partial static structure factors related
to the partial rdfs by
1/2
S (q) = + ( ) dr eiqr h (r) , (149)
Fig. 18: Left: AsakuraOosawaVrij (AOV) model of colloidal hard spheres in osmotic equilib
rium with a solution of ideal, freely overlapping polymer coils. Right: lensshaped depletion
zones overlap volume V2 (r) (shadowed) of two colloidal spheres.
of large particles in the suspension volume V , and the chemical potential of small particles are
fixed.1 The semigrand total free energy, F (N, V, T, S ), of the mixture is then given by [13,46]
{ }
dr N (zS )NS
eF = eULL (r ) dxNS e [ ULS (r ,x )+USS (x ) ]
N N NS NS
N ! 3N NS =0
N S !
drN ef f N
= 3N
eULL (r ) , (150)
N!
where U = ULL + ULS + USS is the total potential energy of the mixture, zS = exp{S }/3S
is the activity coefficient of a small particle at fixed reservoir chemical potential S , and
and S are the thermal wavelengths of large and small particles, respectively. We have used
the grandcanonical average over the set of position vectors, xNS , of smallparticle systems of
particle number NS ranging from zero to infinity.
In the second equation, I have introduced the effective interaction energy,
ef f N
ULL (r ; V, T, S ) = ULL (rN ) + (rN ; V, T, S ) , (151)
of the N large particles in the mixture. It is the sum of the bare potential contribution, ULL , and
a grandfree energy contribution, , with
{ }
(zS )NS
NS [ ULS (rN ,xNS )+USS (xNS ) ]
(r ; V, T, S ) = kB T ln
N
dx e , (152)
N =0
NS !
S
of an inhomogeneous dispersion of small particles in the external field of N large particles fixed
at positions rN . Similar to the discussion of Eq. (30), we can convince ourselves that
( )
(zS )NS
NS 0 NS ! dx NS
U
r1
eU
N
F1 (r ) = (zS )NS
= 1 ULL
ef f N
(r ) (153)
dx N S e U
NS 0 NS !
1
Different ensembles become equivalent in the Tlimit, and one can change then from one description to the
other by a Legendre transformation. However, it is conceptually advantageous to select that ensemble compatible
with the experimentally controlled variables and the physical constraints.
B2.40 G. Nagele
is the mean force on a large sphere 1, for the positions of the N large particles fixed, and grand
canonically averaged over the positions of the small particles. Here again, the total potential en
ef f
ergy in the mixture is denoted by U (rN ; xNS ). Hence, ULL plays the role of a statedependent
ef f
N particle potential of mean force. Once ULL is known, any configurational average over a
function, f (rN ), of largeparticle positions, such as the colloidal pair distribution function, can
ef f ef f
be formally expressed as a onecomponent average involving ULL only. However, ULL is state
dependent, since it includes a freeenergy contribution, , of the small particles. Therefore, and
ef f
even for a pairwiseadditive ULL , ULL contains in general the whole sequence of manybody
contributions.
ef f
In the framework of the semigrand canonical treatment, the ULL of a homogeneous system
can be uniquely decomposed in a sum of independent manybody contributions [13],
N
N
ef f N
ULL e0 (N, V, T, S ) +
(r ; V, T, S ) = U e2 (rij ; S ) +
u e3 (rij , rik ; S ) + (154)
u
i<j i<j<k
Here, ue2 (rij ; S ) is the grandfree energychange required to bring two colloidal spheres in the
system from infinity to positions r1 and r2 . Likewise, u e3 (rij , rik ; S ) is the free energy change
to bring three infinitely distant particles to (r1 , r2 , r3 ), minus the sum of the three pairwise work
contributions, and so on. The manybody terms are constructed such that u en 0, whenever
the distance of a pair of particles in a ncluster is very large. It is crucial to realize that, pro
vided the LS and SS interactions embodied in ULS and USS are shortranged as compared to
ef f
the LL interactions, one can expect that the threebody and higherorder contributions to ULL
are quite small. This allows then to map the mixture on an onecomponent system of large par
ticles interacting pairwise via the statedependent pair potential u e2 . Incidentally, if the degrees
of freedom of the large particles would be integrated out instead of the small ones, then the
ef f
resulting USS (x) would include nonnegligible manybody contributions of any order, which
renders such a contraction rather useless.
For a binary mixture of neutral particles, the thermodynamically extensive volume term is given
by [13]
e0 = V S (S ) + N u
U e0 (S ), (155)
where S (S ) is the osmotic pressure of small particles in the system void of colloids, which
is equal to the reservoir polymer pressure, and u e0 (S ) is the grandfree energy change arising
from inserting a single colloidal particle to the polymer solution. Since U e0 /V is here linearly
dependent on L = N/V , it only adds a constant to the pressure and chemical potential of the
effective onecomponent system, with zero influence on its phase behavior.
Theories of Fluid Microstructures B2.41
The grand free energy of the ideal gas of polymers in the system volume V in presence of N
hard spheres fixed at positions rN , and in contact with a reservoir of ideal polymers, is given
by [51]
id (rN ; V, p ) = id N
p (p ) Vf ree (r ) . (157)
Here, the T dependence has been hidden for conciseness. According to Fig. 18,
4 N N
Vf ree (r ) = V N (a + ap ) +
N 3
V2 (rij ) + V3 (rij , rik ) + (158)
3 i<j i<j<k
is the free volume in the mixture accessible to the polymer sphere centers. Moreover, id p =
kB T zp is the osmotic pressure in the reservoir of the idealgas polymers at the given reservoir
activity zp , which for ideal polymers that do not
( r interact
) among themselves is equal to the reser
r 3
voir polymer density, p , with p = kB T ln p p . Note here that the polymer concentration
in the reservoir is in general different from the one in the mixture (see Fig. 18 and [51,52]). The
reason why the reservoir osmotic pressure of polymers appears in Eq. (157), and not the os
motic polymer pressure in the mixture, is that work has to be done against the polymer pressure
of the reservoir, when the free volume in the system decreases due to a change in the positions
of the colloidal spheres.
Each colloidal sphere excludes a volume (4/3) (a + ap )3 from the polymer centers. When
two or more exclusion (depletion) volumes overlap, more free volume becomes available for
the polymers with a corresponding loss in the grandfree energy. The overlap volumes, V2 ,
V3 , and so on of two, three and more excluded volumes correct for this gain in free volume.
In summary, the suspension can lower its free energy by the clustering of colloidal particles,
which implies that the spheres feel an effective attractive interaction commonly referred to as
depletion attraction. For the present model, where only temperatureindependent excluded vol
ume effects are operative, this attraction is purely entropic, i.e., the loss in free energy is due to
the gain in entropy caused by overlapping depletion zones.
B2.42 G. Nagele
For size ratios q = p / < 2/ 3 1 0.155, threebody and higherorder excluded volume
overlaps are exactly zero.2 When the ratio q is increased beyond this threshold, threebody and
higherorder interaction terms come successively into play. If three excluded volume zones can
overlap, the total gain in three volume is smaller than would be estimated on basis of the pair
interactions alone. Thus, threebody interactions lead to a repulsive correction to the depletion
interaction. The fourbody interactions, in turn, are attractive again.
Quite notably, for q < 0.155, the AOV model of a colloidpolymer mixture is mapped on an
effective onecomponent colloid system with exactly pairwise additive effective interactions.
ef f
On noting that Ucc = Ucc + id , and using Eq. (158), we obtain the result
[ 3]
N
ef f N
Ucc (r ; V, p ) = kB T zp V 1 (1 + q) + [ uHC (rij ) + uAOV (rij ) ] , (159)
i<j
for the effective colloid interaction energy, with a purely attractive AOV depletion pair potential
given by [44, 45]
uAOV (r; p ) = id
p (p ) V2 (r) (160)
3 [ ]
(1 + q) 3 3r r3
= id
p (p ) 1 + , (161)
6 2 (1 + q) 2 (1 + q)3 3
for < r < (1 + q). The AOV pair potential vanishes for nonoverlap distances, r >
(1 + q), of two excludedvolume spheres. Fig. 19 shows uAOV (r) for various (reduced)
osmotic polymer pressures, red = id 3
p , at a fixed q = 0.1 (left), and for various polymer
colloid size ratios at fixed red = 50 (right).
The hardcore potential associated with is denoted by uHC (r). It prevents the colloid spheres
from overlapping. The volume term is seen to be linear in the colloid concentration. Therefore,
it has no effect on the phase behavior of the colloids in the mixture. Since p is externally con
trolled, the AOV potential can be considered as stateindependent. The range of the attractive
AOV potential in this simplifying model is equal to p , independent of the polymer concentra
tion, but its depth can be controlled by changing the density and thus the osmotic pressure of
the polymer solution in the reservoir.
For interacting (selfavoiding) polymers, which deform at nonzero concentrations, the range
and depth of the depletion potential is smaller than for the ideal polymers considered here. In
addition, there is a slight repulsive barrier in front of the attractive well, caused by the corre
lations between the interacting polymers. At high enough polymer concentration, depletion
induced attraction can cause the suspension to phaseseparate into colloidpoor and colloidrich
phases. Depending on the range of attraction, the colloidal particles in the colloidrich phase
can be either in liquidlike arrangements for longerranged attractions, or in a crystalline state
(for very shortranged attractions) [52]. For very shortrange attractions of range typically less
than 5 % of the colloid diameter, a firstorder isostructural solidsolid transition is predicted in
simulations [54, 55].
To make contact with subsection 2.2, note that the potential of mean force associated with the
rdf of large particles, wLL (r) = kB T ln gLL (r), contains in general manybody contributions.
2
The value q = 2/ 3 1 characterizes a small polymer sphere that fits exactly into the space in between three
colloid spheres touching each other in an equilateral configuration (see [53]).
Theories of Fluid Microstructures B2.43
10 0.1
30 q = 0.1 red = 50
0.2
red = 50 q = 0.3
Fig. 19: AOV depletion potential in units of kB T for values of reduced osmotic polymer pres
sure, red = id 3
p , as indicated (left), and polymercolloid size ratios, q = p /, as indicated
(right).
Different from ue2 (r), its definition involves also an average over the positions of all large parti
cles with the exception of two kept fixed at the distance r. Therefore, it reduces to u e2 (r) only
in the zero concentration limit of the large particles.
Fig. 20: Primitive Model of charged colloidal hard spheres (macroions) of radius a and charge
Ze and pointlike counterions and coions (microions).
The surfacereleased counterions and added salt ions (i.e., the microions) are much smaller than
the colloidal particles (macroions). Therefore we can assume all microions to be pointlike, i.e.,
of zero diameter. For simplicity, and although we want to study a system with added salt, let
us restrict for the time being to a twocomponent system of macroions and pointlike, surface
B2.44 G. Nagele
1 Z + c zc = 0 , (162)
where, typically, zc  = 1 and Z zc . According to Eq. (146), there are thus three coupled
OrnsteinZernike equations
for components 1 and 2, i.e., for the colloidal macroions of density 1 , and for the pointlike
counterions of density 2 = c . The subscript c is used here to label the counterions, and not the
colloidal particles. The three OZ equations include information on macroionmacroion (11),
macroioncounterion (12) and counterioncounterion (22) pair correlations. However, we are
only interested here in the macroionmacroion correlations (11). To eliminate explicit reference
to the counterions, we define an effective onecomponent OrnsteinZernike equation for the
macroions alone as
h11 (q) = cef f (q) + 1 cef f (q) h11 (q) . (164)
By demanding h11 to be the same as in the original twocomponent OZ equations, the effective
direct correlation function, cef f (q), is determined by
c c12 (q)2
cef f (q) = c11 (q) + . (165)
1 c c22 (q)
It is thus related to all three partial direct correlation functions. Eq. (164) is formally identical
with the OZ equation of a genuinely onecomponent system. It describes the microstructure
of particles 1 (macroions) immersed in a bath of particles of component 2 (counterions). The
counterions do not appear explicitly in Eq. (164). Their effects are hidden in cef f (q).
Suppose cef f (q) and its Fourierinverse, cef f (r), would be already known from Eq. (165). We
can associate an effective pair potential, uef f (r), describing the interaction of two counterion
dressed macroions from noting that
valid at large pair separation. In fact, according to a general theorem due to Henderson [56], at
a given state point (1 , T ), there is a onetoone correspondence,
between a given pair distribution function g11 (r) and an effective onecomponent system with
exactly pairwise interactions described by a pair potential uef f (r; 1 ), that exactly reproduces
g11 (r) irrespective of the underlying manybody interactions. In general, however, the knowl
edge of uef f (r) alone is insufficient to give higherorder distribution functions such as g111 (r, r ),
(3)
and to obtain the volume term. The potential appearing in the asymptotic relation in Eq. (166)
is identical to the uniquely determined effective pair potential associated with g11 (r). The form
of uef f (r) may be different at different state points (densities) and must be reestablished in
each case. For a given pair distribution function and state point, the associated effective pair po
tential can be determined by an inversion procedure using, e.g., an appropriate onecomponent
OZ closure relation where u(r) is now the searchedfor quantity, or using computer simulations
Theories of Fluid Microstructures B2.45
However, saltfree macroion suspensions are an exemption to the zerodensity rule in Eq. (168)
since, in three dimensions and due to Eq. (162), the counterions cease to screen the Coulomb
interaction between a pair of macroions when 1 0. A saltfree macroion suspension is thus
not weakly coupled even in the infinite dilution limit (see [47] for a lucid discussion of this
subtle point).
We can easily generalize Eqs. (163) and (165) to a more than a twocomponent ionic system by
including the effect of pointlike added salt ions of number density s and charge number zs .
For an overall electroneutral system,
s zs = 0 , (169)
s
Our task is to calculate first all partial direct correlation functions c of the mixture, from solv
ing the coupled OrnsteinZernike equations with appropriate closure relations. In a next step,
cef f (r) can be determined using the manycomponent extension of Eq. (165), with uef f (r)
deduced from its longdistance behavior. To determine cef f (r), we will use here the many
component linear MSA closure because of its analytical simplicity.
The direct correlation functions of pointlike salt and counterions (i.e., components > 1) are
approximated in the MSA by
where
z z
u (r) = LB , r>0, (171)
r
and z {zc , zs }. Fourier transformation leads to
z z
c (q) = 4LB . (172)
q2
Notice in this context that in the limit of pointlike ions, or for ionic systems at infinite dilution,
the MSA reduces to what is known in electrolyte theory as the DebyeHuckel (DH) approxi
mation. When dealing with direct correlation functions of charged particles, which have short
range (e.g., excluded volume) interaction contributions aside from the longrange Coulomb
interactions, it is helpful to partition c (r) into a shortrange part, cs (r), and a longrange
Coulomb part according to
z z
c (r) = cs (r) LB , (173)
r
B2.46 G. Nagele
The screening parameter is determined by the concentrations and charges of all microion com
ponents in the system.
1
The larger asymptotic behavior of cef f (r) is determined by the factor [2 + q 2 ] in Eq. (175),
since cs1 (q) and cs11 (q) are Fourier transforms of shortrange functions. Fourier inversion leads
thus to the intermediate result
1 iqr er
cef f (r) = dre c ef f (q) g r (177)
(2)3 r
with a yet undetermined interaction strength g. The magnitude of g is determined by the pre
1
factor of [2 + q 2 ] which includes the so far unspecified colloidmicroion direct correlations
cs1 (q). Thus, the effective macroion pair potential is asymptotically of the Yukawatype, with
a screening parameter determined by the DebyeHuckel relation in Eq. (176). The microions
distribute themselves around the macroions to screen the colloidcolloid Coulomb repulsion.
The only approximation used to obtain the asymptotic result in Eq. (177) has been the DH
approximation for the microionic direct correlations.
To determine g analytically, we apply the MSA closure additionally to the colloidcolloid and
colloidmicroion direct correlation functions:
Z2
c11 (r) = u11 (r) = LB , r > = 2a (178)
r
Zz
c1 (r) = u1 (r) = LB , r>a. (179)
r
In MSA, the shortrange direct correlation parts are thus zero outside the overlap region, i.e.
cs11 (r) = 0 for r > = 2a and cs1 (r) = 0, for r > a. As a consequence, the MSA predicts the
microioncolloid shortrange direct correlation part to be a linear function of the colloid charge
number. This linearity in Z is exactly valid for the true cs1 only in the limit of weak particle
charges. For larger macroion charges, the MSA underestimates the accumulation of counterions
close to the surface of a colloidal macroion.
Theories of Fluid Microstructures B2.47
Using these properties of the MSA closure leads, in the limit 1 0, to the final result
( )2
ea er
uef f (r) = lim cef f (r) = LB Z 2
, r> , (180)
1 0 1 + a r
for the effective macroion pair potential at infinite dilution, as quoted already in Eq. (3). Note
here that is nonzero for a finite salt ion concentration. Within the linearization approximation
of weak particle charges,
which is consistent with the DHMSA approximation of direct correlations, uef f (r) can be
identified, for nonzero salt content, with the macroion  macroion potential of mean force
w11 (r). From the meanspherical approximations made in its derivation, one might expect that
this form of the effective pair potential applies only at long distances r, and for weakly charged
colloidal particles. However, it has been empirically found that the range of applicability of
Eq. (180) can be substantially extended when Z is treated not as the bare macroion charge, but
as an effective or dressed colloid charge smaller than the bare one. In this way one accounts
approximately for the stronger screening by counterions close to strongly charged macroion
surfaces. Methods to determine the effective macroion charge by means of socalled cellmodel
and jellium approximations are discussed in [59,60]. For an interesting alternative derivation of
the effective macroion pair potential, and volume energy, of charged colloidal spheres based on
linear response theory, see [61, 62]. This work gives also some justification for the freevolume
correction factor 1/(1 ) in the screening constant appearing in Eq. (4).
Due to the approximate nature and, usually, nonperturbative character of integral equation
schemes, one can not make decisive a priori statements on their accuracy. The accuracy of a
specific scheme depends in general on the range and the attractive or repulsive nature of u(r),
the system dimensionality, and on the degree of thermodynamic consistency. The extent of ther
modynamic inconsistency is particularly influential when critical phenomena are studied. The
existence and location of the spinodal instability line, e.g., depends on fine details of the un
derlying integral equation scheme. For a more recent discussion of liquidgas criticality in the
context of the OZ integral equation approximations, see [63]. Partially selfconsistent integral
equation schemes such as the RY approximation, in which different routes to the same thermo
dynamic properties are enforced, are superior to standard integral equation schemes, however,
for the price of a somewhat larger numerical effort. Yet, even this enlarged numerical effort
is usually quite small as compared to computer simulation studies, which makes the integral
B2.48 G. Nagele
equation schemes well suited to study liquid state properties over a broad range of system pa
rameters.
The importance of OZ integral equation schemes goes well beyond the calculation of g(r) and
S(q) from a given pair potential. In combination with powerful density functional theory meth
ods [6466] and perturbation schemes, one can study firstorder liquidsolid phase transitions,
and calculate structural properties of inhomogeneous fluids in external electric or magnetic
fields, and near confining walls or liquidgas interfaces (e.g., layering and wetting phenomena).
Furthermore, integralequationcalculated distribution functions play an important role as static
input to theories dealing with the dynamics of fluids.
Integral equation approaches form also the basis of inversion schemes to deduce information on
the (effective) pair potential from an experimentally determined g(r) or S(q). From knowing
g(r), c(r) can be determined using the OZ equation. The pair potential of the effective one
component system (cf., Eq. 167) follows then directly, but approximately, from using a closure
relation such as the HNC closure. I should also add that the OZ equation for twoparticle cor
relation functions, that has been discussed in this lecture, can be generalized to higherorder
correlation functions using functional calculus methods [67]. The OZ equations for higher
order correlation functions, in turn, are the basis for extended integral equation schemes that
allow to calculate, e.g., the triplet distribution function g (3) (r, r ) [68].
My lecture has been restricted to fluids of spherical particles with spherically symmetric pair
interactions. The integral equation schemes and their closure relations have been broadened
in the past to deal also with fluids of nonspherical rigid and flexible particles, such as more
complex molecules, rodshaped viruses and polymers. For an example, the socalled refer
ence interactionsite model (RISM) method has been used to calculate the sitesite distribution
function of rigid molecules, whose interactions are modelled by an interactionsite potential.
The molecule is hereby represented by a discrete set of interaction sites located at the places
of the atomic nuclei [1]. The sitesite distribution of nonrigid molecules like polymer chains
and polyelectrolytes has been successfully determined on the basis of the polymer reference
interactionsite model (PRISM) [69]. Furthermore, the microstructure of liquids adsorbed in a
porous medium, such as in a gel or in an arrested particle matrix, can be approximately predicted
using socalled replica OrnsteinZernike equations [70, 71].
Theories of Fluid Microstructures B2.49
References
[1] J.P. Hansen and I.R. McDonald, Theory of Simple Liquids (2nd Edition, Academic Press,
London, 1986)
[2] P.A. Egelstaff, An Introduction to the Liquid State (2nd Edition, Clarendon Press, Oxford,
1992)
[4] W.B. Russel, D.A. Saville and W.R. Schowalter, Colloidal Dispersions (Cambridge Uni
versity Press, Cambridge, 1989)
[5] J. Mahanty and B.W. Ninham, Dispersion Forces (Academic Press, New York, 1976)
[7] G. Nagele, The Physics of Colloidal Soft Matter, Lecture Notes 14, Institute of Funda
mental Technological Research (Polish Academy of Sciences publication, Warsaw, 2004)
[8] J.L. Barrat and J.P. Hansen, Basic Concepts for Simple and Complex Fluids (Elsevier,
Amsterdam, 2003)
[11] A.A. Louis, Phil. Trans. R. Soc. Lond. A 359, 939 (2001)
[13] M. Dijkstra, R. van Roij and R. Evans, J. Chem. Phys. 113, 4799 (2000)
[14] J. Dobnikar, R. CastanedaPriego, H.H. von Grunberg and E. Trizac, New J. of Phys. 8,
277 (2006)
[15] E. Trizac, L. Belloni, J. Dobnikar, H.H. von Grunberg and R. CastanedaPriego, Phys.
Rev. E 75, 011401 (2007)
[16] C.F. Tejero and M. Baus, J. Chem. Phys. 118, 892 (2003)
[17] F.H. Stillinger, H. Sakai and S. Torquato, J. Chem. Phys. 117, 288 (2002)
[18] H.N.W. Lekkerkerker, W.C.K. Poon, P.N. Pusey, A. Stroobants and P.W. Warren, Euro
phys. Lett. 20, 559 (1992)
[19] J.G. Kirkwood and F.P. Buff, J. Chem. Phys. 19, 774 (1951)
[20] P.M. Chaikin and T.C. Lubensky, Principles of Condensed Matter Physics (Cambrigde
University Press, Cambrigde, 1995)
[21] J.J. Binney, N.J. Dowrick, A.J. Fisher and M.E.J. Newman, The Theory of Critical Phe
nomena (Oxford University Press, New York, 1993)
B2.50 G. Nagele
[22] H.E. Stanley, Introduction to Phase Transitions and Critical Phenomena (Oxford Univer
sity Press, New York, 1971)
[24] G. Nagele, O. Kellerbauer, R. Krause and R. Klein, Phys. Rev. E 47, 2562 (1993)
[25] P.G. de Gennes, Scaling Concepts in Polymer Physics (Cornell University Press, Ithaca,
1979)
[30] W.R. Smith, D.J. Henderson, P.J. Leonard, J.A. Barker and E.W. Grundke, Mol. Phys.
106, 3 (2008)
[31] A.J. Banchio, G. Nagele and J. Bergenholtz, J. Chem. Phys. 113, 3381 (2000)
[32] J.P. Hansen and L. Verlet, Phys. Rev. 184, 151 (1969)
[33] H. Lowen, T. Palberg and R. G. Simon, Phys. Rev. Lett. 70, 1557 (1993)
[34] R. Pesche, M. Kollmann and G. Nagele, J. Chem. Phys. 114, 8701 (2001)
[35] A. Trokhymchuk, I. Nezbeda, J. Jirsak and D. Henderson, J. Chem. Phys. 123, 024501
(2005); Erratum, J. Chem. Phys. 124, 149902 (2006)
[39] J.M. Kincaid and J.J. Weis, Mol. Phys. 34, 931 (1977).
[41] F.J. Rogers and D.A. Young, Phys. Rev. A 30, 999 (1984)
[45] H. de Hek and A. Vrij, J. Colloid Interface Sci. 84, 409 (1981)
[46] M. Dijkstra, J.M. Brader and R. Evans, J. Phys.: Condens. Matter 11, 10079 (1999)
[48] B. Zoetekouw and R. van Roij, Phys. Rev. E 73, 021403 (2006)
[50] G. Ruiz, J.A. Anta and C.F. Tejero, J. Phys.: Condens. Matter 15, S3537 (2003)
[51] H.N.W. Lekkerkerker and G.J. Vroege, in Fundamental Problems in Statistical Mechan
ics, Vol. VIII, edited by H. van Beijeren and M.H. Ernst (Elsevier Science B.V., Amster
dam, 1994)
[52] A. Vrij and R. Tuinier, in Fundamentals of Interface and Colloid Science, Vol. IV, edited
by J. Lyklema (Elsevier, Amsterdam, 2005)
[53] A.P. Gast, C.K. Hall and W.B. Russel, J. Colloid Interface Sci. 96, 251 (1983)
[54] P.G. Bolhuis and D. Frenkel, Phys. Rev. Lett. 72, 221 (1994)
[55] D. Frenkel in Soft and Fragile Matter: Nonequilibrium Dynamics, Metastability and Flow,
Proceedings of the 53d Scottish Universities Summer School in Physics, St. Andrews,
1999, edited by M.E. Cates and M.R. Evans (SUSSP Publications, Edinburgh, 2000)
[57] N.G. Almarza and E. Lomba, Phys. Rev. E 68, 011202 (2003)
[58] C.P. Royall, A.A. Louis and H. Tanaka, J. Chem. Phys. 127, 044507 (2007)
[59] E. Trizac, L. Bocquet, M. Aubouy and H.H. von Grunberg, Langmuir 19, 4027 (2003)
[60] S. Pianegonda, E. Trizac and Y. Levin, J. Chem. Phys. 126, 014702 (2007)
[63] See articles in New Approaches to Problems in Liquid State Theory, Eds.: C. Caccamo,
J.P. Hansen and G. Stell, (Kluwer, Dordrecht, 1998)
[64] H. Lowen, Coarsegraining Colloids and Polymers, 33. IFFFerienkurs 2002, Schriften
des Forschungszentrums Julich, Vol. 10 (2002)
[68] J.L. Barrat, J.P. Hansen and G. Pastore, Mol. Phys. 63, 747 (1988)
[69] T. Hofmann, R.G. Winkler and P. Reineker, J. Chem. Phys. 114, 10181 (2001)
[70] J.A. Given and G. Stell, J. Chem. Phys. 97, 4573 (1992)
[71] B. Hribar, V. Vlachy and O. Pizio, Mol. Phys. 100, 3093 (2002)
Mult mai mult decât documente.
Descoperiți tot ce are Scribd de oferit, inclusiv cărți și cărți audio de la editori majori.
Anulați oricând.