RESEARCH ARTICLE | SEPTEMBER 01 2004

Dynamical density functional theory and its application to

spinodal decomposition 
A. J. Archer; R. Evans

J. Chem. Phys. 121, 4246–4254 (2004)

https://doi.org/10.1063/1.1778374


View Export
Online Citation

Articles You May Be Interested In

Nucleation near the spinodal: Limitations of mean field density functional theory
J. Chem. Phys. (October 2004)

Do supercooled liquids freeze by spinodal decomposition?

J. Chem. Phys. (November 2007)

Critical cavities and the kinetic spinodal for superheated liquids

J. Chem. Phys. (June 2004)

06 November 2024 06:40:41

JOURNAL OF CHEMICAL PHYSICS VOLUME 121, NUMBER 9 1 SEPTEMBER 2004

Dynamical density functional theory and its application

to spinodal decomposition
A. J. Archera) and R. Evans
H.H. Wills Physics Laboratory, University of Bristol, Bristol BS8 1TL, United Kingdom
共Received 5 May 2004; accepted 9 June 2004兲
We present an alternative derivation of the dynamical density functional theory for the one-body
density profile of a classical fluid developed by Marconi and Tarazona 关J. Chem. Phys. 110, 8032
共1999兲兴. Our derivation elucidates further some of the physical assumptions inherent in the theory
and shows that it is not restricted to fluids composed of particles interacting solely via pair
potentials; rather it applies to general, multibody interactions. The starting point for our derivation
is the Smoluchowski equation and the theory is therefore one for Brownian particles and as such is
applicable to colloidal fluids. In the second part of this paper we use the dynamical density
functional theory to derive a theory for spinodal decomposition that is applicable at both early and
intermediate times. For early stages of spinodal decomposition our nonlinear theory is equivalent to
the 共generalized兲 linear Cahn–Hilliard theory, but for later times it incorporates coupling between
different Fourier components of the density fluctuations 共modes兲 and therefore goes beyond Cahn–
Hilliard theory. We describe the results of calculations for a model 共Yukawa兲 fluid which show that
the coupling leads to the growth of a second maximum in the density fluctuations, at a wave number
larger than that of the main peak. © 2004 American Institute of Physics.
关DOI: 10.1063/1.1778374兴

I. INTRODUCTION j共 r,t 兲 ⫽⫺⌫ ␳ 共 r,t 兲 ⵜ ␮ 共 r,t 兲 , 共1兲

06 November 2024 06:40:41

Classical density functional theory 共DFT兲1 is a remark- where ⌫ is a mobility constant. An expression for ␮ is ob-
ably successful theory for describing the rich behavior of the tained within DFT by assuming that, as in the case of the
equilibrium structure and thermodynamics of fluids in exter- equilibrium fluid, the chemical potential is given by the func-
tional derivative of the Helmholtz free energy functional
nal potentials. DFT has been used to describe a wide variety
with respect to the density profile.5,6 This assumption, to-
of fluid interfacial, confinement and even freezing phenom-
gether with Eq. 共1兲 and the continuity equation
ena; for example, DFT has been a vital tool in understanding
the wetting behavior and surface phase transitions of fluids ⳵␳ 共 r,t 兲
adsorbed on various substrates.1,2 Given the success of DFT ⫽⫺ⵜ•j共 r,t 兲 , 共2兲
⳵t
for describing static inhomogeneous fluids, it is of great in-
terest to be able to build upon and incorporate these theories provide a basis for the deterministic formulation of dynami-
into a theory for the dynamics of inhomogeneous fluids. cal DFT 共DDFT兲. Equations of this form have been used, for
There have been several approaches to obtaining an example, to study the dynamics of freezing7 and of
solvation.8 More recently, the more systematic approach of
equation of motion for the one-body density profile ␳ (r,t) of
MT has been used with much success to describe the dynam-
a classical fluid. The form that these theories takes depends
ics for several different systems. These applications refer to
somewhat on how the particular authors define ␳ (r,t). For
particles in various external, time-dependent, potentials.9–12
some, ␳ (r,t) is an ‘‘ensemble’’ average over the possible
For the systems considered, the agreement between theory
configurations of the system at time t, given an ensemble of
and the results from Brownian dynamics simulations have
starting configurations at an earlier time t⫽0. Using this
generally been very good.
definition for ␳ (r,t), there is clearly a unique density profile
An alternative approach to obtaining a DDFT is to view
␳ (r,t) at a given time t, and therefore the equation govern-
the fluid one-body density profile 关denoted ¯␳ (r,t) in order to
ing the dynamics of this density profile will be deterministic.
distinguish it from ␳ (r,t), the ‘‘ensemble’’ averaged density
This is the philosophy behind the approach of Marconi and
profile兴 as some sort of spatial and/or time coarse grained
Tarazona 共MT兲 in Refs. 3 and 4. Their approach obtains average of the density operator ␳ˆ (r,t)⫽ 兺 i⫽1 N
␦ (r⫺ri (t)),
more formally some of the results proposed by earlier au- where the ri (t) are the positions of the N particles in the
thors, such as Evans5 and Dieterich et al.,6 where it is as- system. In this case, for Brownian particles, the equation
sumed from the outset that the gradient of the chemical po- governing the dynamics of ¯␳ (r,t) will, of course, still con-
tential ⵜ ␮ (r,t) is the thermodynamic force driving a particle tain a stochastic element. This is the viewpoint of a number
current of theories.13–18 There is some confusion in the literature19
共see also Ref. 20兲 as to what precisely is meant by ␳ (r,t).
a兲
Electronic mail: [email protected] We will attempt to clarify these issues elsewhere.21

0021-9606/2004/121(9)/4246/9/$22.00 4246 © 2004 American Institute of Physics

J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 Dynamical density functional theory and spinodal decomposition 4247

In this paper we adopt the deterministic viewpoint of MT dom, and we have effectively averaged over the momentum
and consider ␳ (r,t) to be an ensemble average, i.e., an aver- degrees of freedom, while keeping the particle coordinates
age over all realizations of the stochastic noise in the preced- fixed. For a colloidal fluid, this thermal equilibration should
ing interval, until time t. We employ the Smoluchowski occur via the solvent, and this approximation should be a
equation as the starting point for an alternative derivation of good one to make. For atomic fluids, this may not be the
the DDFT for classical fluids developed by MT in Refs. 3 case, especially for particles interacting via harshly repulsive
and 4. The present scheme for deriving the DDFT bears 共hard-sphere-like兲 potentials. For fluids interacting with
some similarity in spirit to that given recently in Ref. 22, softer potentials, such as in the Gaussian core model,9 this
where projector–operator techniques are used to obtain first might be a reasonable approximation to make, particularly
the Smoluchowski equation and subsequently an equation of when the fluid is at high densities where each particle inter-
motion for the fluid one-body density profile—the DDFT. acts with a large number of neighbors and so the momentum
Before proceeding with our derivation of the DDFT, in Sec. degrees of freedom can equilibrate faster. Since the particle
II we give a brief introduction to the Smoluchowski equa- number is conserved, we can expect the fluid to obey the
tion, expounding some of the physical assumptions concern- continuity equation
ing the dynamics of the fluid that are implicit in this equa- N
⳵ P 共 rN ,t 兲
tion. In Sec. III we proceed with the derivation of the DDFT
of MT from the Smoluchowski equation. Section IV de- ⳵t
⫽⫺
i⫽1
兺
ⵜi . 关 vi P 共 rN ,t 兲兴 . 共6兲

scribes an important application of the DDFT to analyze the Substituting Eqs. 共4兲 and 共5兲 into 共6兲, one finds
short- and intermediate-time dynamics of spinodal decompo-
N N
sition relevant to colloidal fluids. Finally, in Sec. V we make ⳵ P 共 rN ,t 兲
some concluding remarks. ⳵t
⫽ 兺 j⫽1
兺 ⵜi .⌫i j . 关 k B Tⵜj ⫹ⵜj U 共 rN ,t 兲兴 P 共 rN ,t 兲 .
i⫽1
共7兲
II. THE SMOLUCHOWSKI EQUATION If the potential energy term U(r ,t)⫽0, then ⌫i j ⫽⌫ ␦ i j
N

The Smoluchowski equation23–27 is a Fokker–Planck ⫽ ␤ D ␦ i j , where D is the diffusion coefficient and Eq. 共7兲
equation 共or generalized diffusion equation兲 for interacting reduces to the diffusion equation, ( ⳵ / ⳵ t) P(rN ,t)
⫽ ␤ ⌫ 兺 i ⵜ 2i P(rN ,t). For a system of interacting particles the

06 November 2024 06:40:41

Brownian particles. A physically intuitive way of arriving at
this equation, e.g., Ref. 24, proceeds as follows: For a fluid diffusion tensor does not, in general, take such a simple
of N Brownian particles, one imagines applying a hypotheti- form.24,28 However, if we neglect the hydrodynamic interac-
cal force on the particles, where the force on the jth particle tions, we can replace ⌫i j by its mean-field value, ⌫ ␦ i j , and
is F j ⫽⫺ⵜj ⌿(rN ) (rN ⬅ 兵 r1 ,r2 ¯rN 其 is the set of position then Eq. 共7兲 reduces to a generalized diffusion equation,
coordinates for the N particles兲. The equilibrium probability termed the Smoluchowski equation,
density function in this situation will be N
⳵ P 共 rN ,t 兲
1 ⳵t
⫽⌫ 兺 ⵜi 关 k B Tⵜi ⫹ⵜi U 共 rN ,t 兲兴 P 共 rN ,t 兲 . 共8兲
P 共 rN 兲 ⫽ exp关 ⫺ ␤ ⌿ 共 rN 兲 ⫺ ␤ U 共 rN 兲兴 , 共3兲 i⫽1
Z
More formally, the Smoluchowski equation is the Fokker–
where Z is a normalization factor, ␤ ⫽1/k B T is the inverse Planck equation for a system of N Brownian particles in the
temperature, and U(rN ) is the potential energy due to the large friction limit.23,26,27 The Langevin equation for a sys-
interparticle interactions and any other external potentials. tem of N Brownian particles of mass m is
Taking the gradient of Eq. 共3兲 we obtain
d2 ri 共 t 兲 ⫺1
dri 共 t 兲
ⵜj P 共 r 兲 m 2 ⫹⌫ ⫽⫺ⵜi U 共 rN ,t 兲 ⫹Xi 共 t 兲 , 共9兲
N
F j ⫽ⵜj U 共 rN 兲 ⫹k B T . 共4兲 dt dt
P 共 rN 兲
where Xi (t)⫽( ␰ xi (t), ␰ iy (t), ␰ zi (t)) is a white noise term with
If F j is switched off, there will be a force ⫺F j driving the the property 具 ␰ ␣i (t) 典 ⫽0 and 具 ␰ i␣ (t) ␰ i␯ (t ⬘ ) 典 ⫽2k B T ␦ i j ␦ ␣ ␯ ␦ (t
diffusion of particle j. We now assume that for time scales ⫺t ⬘ ). When the friction constant ⌫ ⫺1 is large, we may ne-
Ⰷ ␶ B , the Brownian time scale, the velocity of the ith par- glect the second derivative with respect to time in Eq. 共9兲,
ticle is and we obtain the stochastic equation of motion,3,4
N
dri 共 t 兲
vi ⫽⫺ 兺 ⌫i j •Fj
j⫽1
共5兲
dt
⫽⫺⌫ⵜi U 共 rN ,t 兲 ⫹⌫Xi 共 t 兲 . 共10兲

where ⌫i j ⫽ ␤ Di j (rN ) is the mobility tensor and Di j is the The 共generalized兲 Fokker–Planck equation for the distribu-
diffusion tensor. We also assume that we can use the expres- tion function P(rN ,t) corresponding to this Langevin equa-
sion in Eq. 共4兲 for the force F j in 共5兲, with the equilibrium tion is Eq. 共8兲.23,26,27
probability density function P(rN ) replaced by the nonequi-
librium probability density function P(rN ,t) and U(rN ) re-
III. DYNAMICS OF THE ONE-BODY DENSITY PROFILE
placed by U(rN ,t) 共i.e., there may be a one-body time-
dependent external potential兲. It is implicitly assumed that as In this section we derive an equation for the time evolu-
the particles interact, the momentum degrees of freedom tion of the one-body density profile, ␳ (r,t), from the Smolu-
equilibrate much faster than the positional degrees of free- chowski equation, Eq. 共8兲. For a similar approach based
4248 J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 A. J. Archer and R. Evans

solely on pair potentials see Refs. 24, 29, and 30. The one- From equilibrium statistical mechanics one also knows that
body density is merely the integral of the probability distri- c (1) (r) is equal to the functional derivative of the excess
bution function, 共over ideal兲 part of the Helmholtz free energy functional,5

␳ 共 r1 ,t 兲 ⫽N 冕 dr2 ¯ 冕 drN P 共 rN ,t 兲 . 共11兲 c (1) 共 r兲 ⫽⫺ ␤

␦ F ex关 ␳ 共 r兲兴
␦ ␳ 共 r兲
, 共17兲

Similarly, the two-body density is evaluated at the equilibrium density. 关Generally c (1) (r) is a
functional of ␳共r兲.兴 Making the approximation that these
␳ (2) 共 r1 ,r2 ,t 兲 ⫽N 共 N⫺1 兲 冕 dr3 ¯ 冕 drN P 共 rN ,t 兲 , 共12兲 identities, Eqs. 共16兲 and 共17兲, valid for the equilibrium fluid,
hold also for the nonequilibrium fluid and substituting into
and in general the n-particle density is Eq. 共15兲, we obtain the DDFT equation, stated without jus-
tification by Evans 关see Eqs. 共166兲 and 共167兲 of Ref. 5兴, and
␳ (n) 共 rn ,t 兲 ⫽
N!
共 N⫺n 兲 !
冕 drn⫹1 ¯ 冕 drN P 共 rN ,t 兲 . 共13兲 derived much more convincingly, for the case of fluids inter-
acting solely via pair potentials, by MT,3,4 i.e.,
Using Eqs. 共11兲–共13兲, and assuming that the potential energy
function can be expressed in terms of a one-body external
potential acting on each particle, V ext(ri ,t), and that the par-
⌫ ⫺1
⳵␳ 共 r,t 兲
⳵t 冋
⫽ⵜ ␳ 共 r,t 兲 ⵜ
␦ F 关 ␳ 共 r,t 兲兴
␦ ␳ 共 r,t 兲
, 册 共18兲

ticle interactions are a sum of pair potentials, v 2 (ri ,r j ), where F 关 ␳ (r,t) 兴 is the Helmholtz free energy functional,

冕
three-body potentials v 3 (ri ,r j ,rk ), and higher body interac-
tions, F 关 ␳ 共 r,t 兲兴 ⫽k B T dr ␳ 共 r,t 兲关 ln共 ␳ 共 r,t 兲 ⌳ 3 兲 ⫺1 兴

冕
N N
1
U 共 rN ,t 兲 ⫽ 兺 V ext共 ri ,t 兲 ⫹ 2 兺
i⫽1
兺 v 2共 ri ,rj 兲
j⫽i i⫽1
⫹F ex关 ␳ 共 r,t 兲兴 ⫹ dr V ext共 r,t 兲 ␳ 共 r,t 兲 . 共19兲

1
N
The first term is the ideal gas free energy; ⌳ is the de Broglie
⫹ 兺 兺
6 k⫽ j⫽i 兺 v 3共 ri ,rj ,rk 兲 ⫹¯ 共14兲 wavelength. In obtaining Eq. 共18兲 by using Eqs. 共16兲 and

06 November 2024 06:40:41

j⫽i i⫽1
共17兲, which are strictly equilibrium results, we are effectively
then we find that on integrating Eq. 共8兲, one obtains assuming that the correlations between the particles when the
fluid is out of equilibrium are equivalent to those for an
⳵␳ 共 r1 ,t 兲
⌫ ⫺1 ⫽k B Tⵜ 21 ␳ 共 r1 ,t 兲 ⫹ⵜ1 关 ␳ 共 r1 ,t 兲 ⵜ1 V ext共 r1 ,t 兲兴 equilibrium fluid with the same one-body density profile
⳵t ␳ (r,t).

⫹ⵜ1 冕 ␳dr2 (2)

共 r1 ,r2 ,t 兲 ⵜ1 v 2 共 r1 ,r2 兲
We can obtain further insight into the status of Eq. 共18兲
by rewriting it as follows:

冕 冕
⳵␳ 共 r,t 兲
⫹ⵜ1 . dr3 ␳ ⌫ ⫺1 ⫽ⵜ 关 ␳ 共 r,t 兲 ⵜ ␮ 共 r,t 兲兴 , 共20兲
dr2 (3)
共 r1 ,r2 ,r3 ,t 兲 ⳵t

⫻ⵜ1 v 3 共 r1 ,r2 ,r3 兲 ⫹¯ . 共15兲 where ⫺ⵜ ␮ (r,t)⬅⫺ⵜ( ␦ F 关 ␳ 兴 / ␦ ␳ ) is the net driving force
acting on a particle located at (r,t). The chemical potential
We note that if ⳵␳ (r,t)/ ⳵ t⫽0, then Eq. 共15兲 is equivalent to obtained from Eq. 共19兲 has three contributions,
the derivative of the first equation of the YBG hierarchy.31
For a fluid in equilibrium, there is an exact sum rule5 ␦ F 关 ␳ 共 r,t 兲兴
⬅ ␮ 共 r,t 兲 ⫽ ␮ id共 r,t 兲 ⫹ ␮ int共 r,t 兲 ⫹ ␮ ext共 r,t 兲 .
which relates the gradient of the one-body direct correlation ␦ ␳ 共 r,t 兲
function to the interparticle forces acting on a particle 关recall 共21兲
that ⫺k B Tc (1) (r) is the effective one-body potential due to The first contribution is the ideal gas entropic term ␮ id(r,t)
interactions in the fluid兴. If the particles interact solely ⫽k B T ln ⌳3␳(r,t), the second contribution, ␮ int(r,t)
via pair potentials: ⫺k B T ␳ (r)ⵜc (1) (r)⫽ 兰 dr⬘ ␳ (2) (r,r⬘ ) ⫽⫺k B Tc (1) (r), is that due to the interactions with the other
⫻ⵜ v 2 (r,r⬘ ). This result can be generalized straightfor- particles in the fluid, and the final term is simply the external
wardly to fluids where the particles interact via many-body potential, ␮ ext(r,t)⫽V ext(r,t). As noted by MT,3 for nonin-
potentials, Eq. 共14兲, giving teracting particles, ␮ int⬅0, and Eq. 共20兲 reduces to the exact
⫺k B T ␳ 共 r1 兲 ⵜc (1) 共 r1 兲 equation for the diffusion of an ideal Brownian gas, i.e.,
⳵␳ 共 r,t 兲
⫽ 冕 dr2 ␳ (2)
共 r1 ,r2 兲 ⵜ v 2 共 r1 ,r2 兲
⌫ ⫺1
⳵t
⫽k B Tⵜ 2 ␳ 共 r,t 兲 ⫹ⵜ 关 ␳ 共 r,t 兲 ⵜV ext共 r,t 兲兴 .
共22兲
⫹冕 冕 ␳ dr2 dr3 (3)
共 r1 ,r2 ,r3 兲 ⵜ1 v 3 共 r1 ,r2 ,r3 兲 ⫹¯ For an inhomogeneous interacting fluid in equilibrium the
density profile satisfies5

兺冕 冕
⬁
␦ F 关 ␳ 共 r兲兴
⫽ dr2 ¯ drn ␳ (n) 共 rn 兲 ⵜ1 v n 共 rn 兲 . 共16兲 ⫽ ␮ ⫽constant, 共23兲
n⫽2 ␦ ␳ 共 r兲
J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 Dynamical density functional theory and spinodal decomposition 4249

where the Helmholtz free energy functional F 关 ␳ 兴 is given by

Eq. 共19兲 with ␳ (r,t) replaced by ␳共r兲 and V ext(r,t) replaced
by V ext(r), so there is no net driving force on the particles. It
follows that for a time-independent external potential,
V ext(r,t)→V ext(r), as t→⬁, regardless of the initial profile
␳ (r,t⫽0), Eq. 共18兲 will yield the same one-body density
profile in the limit t→⬁ as does the equilibrium DFT 关i.e.,
the solution to Eq. 共23兲兴 for the same external potential. The
present derivation provides, we believe, additional insight to
that of MT3,4 into the physics incorporated into Eq. 共18兲 共the
key DDFT equation兲. Assuming one has an accurate expres-
sion for the Helmholtz free energy functional Eq. 共19兲, and in
particular for the excess Helmholtz free energy functional,
Eq. 共18兲 should provide an accurate description of the dy-
namics of ␳ (r,t) for a system of Brownian particles.

IV. SPINODAL DECOMPOSITION

FIG. 1. Phase diagram for a fluid composed of particles interacting via pair
FROM THE DDFT EQUATION potentials of the form in Eqs. 共38兲–共40兲, calculated from the free energy of
In this section we apply the DDFT derived in the pre- Eq. 共41兲. ␩ ⫽ ␲␳ b ␴ 3 /6 is the packing fraction and k B T ␴ 3 /a is the 共reduced兲
temperature. The path from A to B denotes the quench corresponding to the
ceding section to fluid spinodal decomposition. Since the ba- results in Figs. 2 and 3.
sis for the DDFT is the Smoluchowski equation, an equation
of motion for Brownian particles, we expect our theory to be
particularly relevant to spinodal decomposition in colloidal
fluids, rather than molecular fluids. In colloidal fluids friction ever, in a deep quench far from the spinodal, spinodal de-
with the solvent results in a much faster equilibration of the composition is the mechanism generally expected for phase
momentum degrees of freedom compared with those of the separation.

06 November 2024 06:40:41

particle positions. However, since we do not explicitly in- In a fluid undergoing spinodal decomposition three dif-
clude the particles of the solvent in which the colloids are ferent regimes can be distinguished. For early times after the
suspended, our theory neglects the hydrodynamic interac- quench, the amplitude of the density fluctuations are small
tions that may be significant to spinodal decomposition in and theories linear in the density fluctuations such as the
some colloidal fluids. well-known Cahn–Hilliard theory34,35 共see also Refs. 29, 32,
When a 共colloidal兲 fluid which exhibits liquid–gas phase 33, 36, and 37兲 provide a good description of this 共early兲
separation 共or more generally fluid–fluid phase separation兲 is stage of spinodal decomposition. At intermediate times the
quenched to a state point in the region of the phase diagram density fluctuations can be large, but sharp interfaces be-
where there is coexistence, the fluid can phase separate in tween domains of gaslike and liquidlike regions have still not
two distinct ways. The first mechanism is that which occurs formed.29 At later stages there are sharp interfaces between
when the state point to which the fluid is quenched is near to domains of liquid and gas and successful theoretical descrip-
the binodal. In this case phase separation generally occurs tions of the later stage dynamics of spinodal decomposition,
via nucleation of droplets of one phase forming in the other such as the Allen–Cahn theory38 共see also Refs. 32, 33, and
phase.32,33 For example, if the fluid is quenched to a state 39兲, focus on the dynamics of the interfaces.
point inside the binodal on the liquid side, then bubbles of First, in Sec. IV A, we shall use the DDFT formalism to
the gas phase can nucleate in the bulk of the metastable derive a generalization of the 共linear兲 Cahn–Hilliard
liquid. theory,34,35 similar to that described in Refs. 29, 36, and 37,
However, if the fluid is quenched to a state point well for the early stages of spinodal decomposition, when the
inside the binodal, then a different mechanism for phase density fluctuations are small. Then in Sec. IV B we will
separation is possible: spinodal decomposition. Spinodal de- proceed to derive a nonlinear theory which we believe may
composition is characterized by the exponential growth of be applicable for the dynamics of spinodal decomposition at
density fluctuations of certain wavelengths.32,33 In 共mean- both short and intermediate time scales. Results of explicit
field兲 theoretical descriptions, liquid–gas phase separation is calculations for a model fluid are given in Sec. IV C.
determined by the occurrence of a van der Waals loop in the
A. Early stages of spinodal decomposition
Helmholtz free energy per particle, f ( v ), where v is the
volume per particle. Spinodal decomposition is predicted to We consider spinodal decomposition in the bulk of a
occur in regions of the phase diagram where ( ⳵ 2 f / ⳵ v 2 ) T fluid, so we set V ext(r,t)⫽0 in Eqs. 共18兲 and 共19兲 and con-
⬍0, i.e., regions where the isothermal compressibility ␹ T is sider small density fluctuations ˜␳ (r,t)⫽ ␳ (r,t)⫺ ␳ b about the
predicted to be negative. The boundary to this region, the bulk fluid density, ␳ b , i.e., we are considering a homoge-
spinodal, is defined by the locus of ( ⳵ 2 f / ⳵ v 2 ) T ⫽0 in the neous fluid which has been rapidly quenched to the region of
phase diagram. Experimentally there is not necessarily a the phase diagram inside the spinodal 共e.g., from A to B in
sharp distinction between regions where phase separation oc- Fig. 1兲 and we seek those wave numbers k for which density
curs via nucleation and via spinodal decomposition. How- fluctuations grow. From Eqs. 共17兲, 共18兲, and 共19兲 we obtain
4250 J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 A. J. Archer and R. Evans

⳵␳
˜ 共 r,t 兲
共 ⌫k B T 兲 ⫺1 ⫽ⵜ 2˜␳ 共 r,t 兲 ⫺ ␳ b ⵜ 2 c (1) 共 r,t 兲
⳵t
⫺ⵜ 关 ˜␳ 共 r,t 兲 ⵜc (1) 共 r,t 兲兴 . 共24兲
This approach is basically that of Refs. 5 and 36. In Ref. 5
Evans writes down Eq. 共18兲, and then linearizes Eq. 共24兲 in ˜␳
by Taylor expanding c (1) about the bulk fluid value, giving

c (1) 共 r兲 ⫽c (1) 共 ⬁ 兲 ⫹ 冕 dr⬘

␦ c (1) 共 r兲
␦ ␳ 共 r⬘ 兲 冏 ˜␳ 共 r⬘ ,t 兲 ⫹O共 ˜␳ 2 兲 ,
␳b
共25兲
where c (⬁)⬅c 关 ␳ b 兴 ⫽⫺ ␤␮ ex and ␮ ex is the excess
(1) (1)

chemical potential. The second term simplifies by recalling5

␦ c (1) 共 r兲 ␦ 2 F ex关 ␳ 兴
⫽⫺ ␤
␦ ␳ 共 r⬘ 兲 ␦ ␳ 共 r⬘ 兲 ␦ ␳ 共 r兲
FIG. 2. The function S(k)⬅(1⫺ ␳ b ĉ(k)) ⫺1 calculated using Eq. 共43兲 at the
⫽c (2) 共 r,r⬘ 兲 ⫽c (2) 共 兩 r⫺r⬘ 兩 ; ␳ b 兲 , 共26兲
state point 共inside the spinodal兲 with ␩ ⫽0.2 and k B T ␴ 3 /a⫽0.05, point B in
for a homogeneous fluid of spherically symmetric particles. Fig. 1. Note that for k⬍k c , where k c ␴ ⯝0.8, S(k)⬍0. In the inset we plot
R(k) ␴ 2 ␤ /⌫⫽⫺k 2 ␴ 2 /S(k), the factor appearing in the exponential in Eq.
For an equilibrium system c (2) (r; ␳ b ) is the Ornstein– 共30兲. In the initial stages of spinodal decomposition density fluctuations with
Zernike pair direct correlation function of the fluid of density wave numbers k⬍k c grow exponentially, whereas for k⬎k c the fluctuations
␳ b . Substituting Eq. 共25兲 into Eq. 共24兲, keeping only terms are damped.
that are linear in the fluctuation ˜␳ , we obtain.5,36
⳵␳
˜ 共 r,t 兲
共 ⌫k B T 兲 ⫺1 ploys a particular approximation, namely Eq. 共43兲, for ĉ(k).
⳵t

06 November 2024 06:40:41

k c is obtained as the solution to the equation ␳ b ĉ(k c )⫽1.
⫽ⵜ 2˜␳ 共 r,t 兲 ⫺ ␳ b ⵜ 2 冋冕 dr⬘ c (2) 共 兩 r⫺r⬘ 兩 ; ␳ b 兲˜␳ 共 r⬘ ,t 兲 . 册 Thus we find that inside the spinodal region there will be
some density fluctuations with a wave number k⬍k c whose
共27兲 amplitude will grow exponentially.5,29,30,34 –37 The deeper the
quench into the spinodal region, the larger the value of k c
Fourier transforming yields an equation for the time evolu- can be.
tion of the different Fourier components This general picture of the exponential growth of certain

␳ˆ 共 k,t 兲 ⫽ 冕 dr exp共 ik"r兲˜␳ 共 r,t 兲 , 共28兲

Fourier components 共modes兲 was obtained originally by
Cahn and Hilliard34,35 who derived an explicit approximation
for the function R(k). Cahn–Hilliard theory for spinodal de-
and one obtains composition is usually derived by considering the continuity
⳵␳ˆ 共 k,t 兲 equation 共2兲, together with the following approximation for
共 ⌫k B T 兲 ⫺1 ⫽⫺k 2 ␳ˆ 共 k,t 兲 ⫹ ␳ b k 2 ĉ 共 k 兲 ␳ˆ 共 k,t 兲 , 共29兲 the current:
⳵t
where ĉ(k)⫽ 兰 dr exp(ik"r)c (2) (r; ␳ b ). The solution of Eq. ␦ F 关 ␳ 共 r,t 兲兴
j共 r,t 兲 ⫽⫺M ⵜ , 共31兲
共29兲 is ␦ ␳ 共 r,t 兲
␳ˆ 共 k,t 兲 ⫽ ␳ˆ 共 k,0兲 exp关 R 共 k 兲 t 兴 , 共30兲 where M is a mobility constant and the functional F is cho-
where R(k)⫽⫺⌫k B Tk (1⫺ ␳ b ĉ(k)). For an equilibrium
2 sen to be of the ␾ 4 Ginzburg–Landau form. Generalizing
fluid, at a state point outside the spinodal, S(k)⬅(1 slightly we assume that the free energy functional has the
⫺ ␳ b ĉ(k)) ⫺1 is the usual static structure factor31,40 and, since square-gradient form5

冕 冋 册
for an equilibrium fluid S(k)⬎0 for all values of k, it fol- 1
lows that outside the spinodal R(k)⬍0 for all values of k. F sg关 ␳ 共 r,t 兲兴 ⫽ dr f 0 共 ␳ 共 r,t 兲兲 ⫹ K 兩 ⵜ ␳ 共 r,t 兲 兩 2 , 共32兲
2
From Eq. 共30兲, all Fourier components will decay implying,
of course, that the fluid is stable. where f 0 ( ␳ )⬅ ␳ f ( ␳ ) is the Helmholtz free energy density for
On approaching the spinodal from the single phase re- the homogeneous fluid of density ␳ and we treat K as a
gion one finds that S(k⫽0)→⬁; at the spinodal (1 positive constant. From Eq. 共2兲 one then finds
⫺ ␳ b ĉ(k⫽0))⫽0. Within the mean-field 共van der Waals-
type兲 theory of fluids to be described below, we find that
inside the spinodal S(k) can take negative values40 and thus
the quantity R(k) can have positive values for k⬍k c , where
⳵␳ 共 r,t 兲
⳵t
⫽M ⵜ 2 冋
⳵ f 0 共 ␳ 共 r,t 兲兲
⳵␳ 共 r,t 兲
⫺Kⵜ 2 ␳ 共 r,t 兲 . 册 共33兲

the value of k c depends upon how far into the spinodal re- This equation is then linearized about the bulk density ␳ b to
gion one has quenched, see for example, Fig. 2, which em- obtain32,36
J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 Dynamical density functional theory and spinodal decomposition 4251

⳵␳
˜ 共 r,t 兲
⳵t
⫽M ⵜ 2 冋冉 冊 ⳵2 f 0
⳵␳ 2 ␳b
册
⫺Kⵜ 2 ˜␳ 共 r,t 兲 . 共34兲
order parameter, rather than a constant, in a Cahn–Hilliard
treatment—see Eq. 共10兲 in Ref. 43; we shall discuss this
further in Sec. IV D. In order to proceed further we must
On Fourier transforming Eq. 共34兲 one obtains ⳵␳ˆ (k,t)/ ⳵ t assume a particular form for the Helmholtz free energy func-
⫽R sg(k) ␳ˆ (k,t), where R sg(k)⫽⫺M k 2 (Kk 2 ⫹( ⳵ 2 f 0 / tional, from which we can obtain ĉ(k) and thus solve Eq.
⳵␳ ) ␳ b ), the solution to which is Eq. 共30兲, with R(k) re-
2 共37兲 numerically.
placed by R sg(k). Note that ( ⳵ 2 f 0 / ⳵␳ 2 ) ␳ b is negative inside
the spinodal. It is clear from this argument that Cahn–
Hilliard theory can therefore be viewed as a special case of C. Results for a model fluid
the more general linear theory presented in the earlier part of We consider a fluid composed of particles interacting via
this section.41 By employing a free energy functional more pair potentials of the form
accurate than 共32兲 one should be able to incorporate short
wavelength density fluctuations as well as the long wave- v 2 共 r 兲 ⫽ v hs共 r 兲 ⫹ v at共 r 兲 , 共38兲
length, k→0, fluctuations accounted for by a square-gradient
approach. where

B. Spinodal decomposition at intermediate times v hs共 r 兲 ⫽ 再 ⬁,

0,
r⭐ ␴
r⬎ ␴
共39兲
In order to go one step beyond the lowest order 共linear兲
theories described above, we consider an approximate excess is the hard-sphere pair potential, and the attractive part of the
Helmholtz free energy functional, obtained by Taylor ex- pair potential has a Yukawa form
panding the excess Helmholtz free energy about the uniform
density. By integrating Eq. 共25兲 and omitting terms beyond a␭ 3 exp共 ⫺␭r 兲
v at共 r 兲 ⫽⫺ , 共40兲
O(˜␳ 2 ) we obtain 4 ␲ ␭r

冕
F ex关 ␳ 兴 ⫽F ex关 ␳ b 兴 ⫹ ␮ ex dr ˜␳ 共 r,t 兲
where a and ␭ are positive constants. Provided the decay

06 November 2024 06:40:41

length ␭ ⫺1 is sufficiently large this model fluid exhibits

冕 冕
stable, with respect to freezing, liquid–gas phase separation.
k BT
⫺ dr dr⬘ ˜␳ 共 r,t 兲˜␳ 共 r⬘ ,t 兲 c (2) 共 兩 r⫺r⬘ 兩 ; ␳ b 兲 . Pair potentials of this form are often used as crude models
2 for simple fluids but they could be used to model the effec-
共35兲 tive 共depletion兲 potential between the colloids in a colloid–
polymer mixture solution44 by choosing the pair potential
This truncated quadratic density expansion is often used in
parameters in Eqs. 共39兲 and 共40兲 to mimic the well-known
the theory of inhomogeneous fluids in equilibrium.1 Using
Asakura–Oosawa potential.44 – 47 In calculations for our
Eq. 共35兲 in Eq. 共24兲 we find
model system we approximate the excess Helmholtz free en-
共 ⌫k B T 兲 ⫺1
⳵␳
˜ 共 r,t 兲
⳵t
冋
⫽ⵜ 2˜␳ 共 r,t 兲 ⫺ⵜ 共 ␳ b ⫹˜␳ 共 r,t 兲兲
ergy functional by

⫻ 冕 dr⬘ ˜␳ 共 r⬘ ,t 兲 ⵜc (2) 共 兩 r⫺r⬘ 兩 ; ␳ b 兲 . 册 F ex关 ␳ 兴 ⫽F ex

hs
关␳兴⫹
1
2
冕 冕dr dr⬘ ␳ 共 r,t 兲 ␳ 共 r⬘ ,t 兲v at共 兩 r⫺r⬘ 兩 兲 ,
共41兲
共36兲 hs
where F ex 关 ␳ 兴 is the reference hard-sphere Helmholtz excess
The first two terms on the right-hand side are those entering free energy functional and attractive interactions are treated
the linear theory 共27兲 while the third term is the only nonlin- in a mean-field fashion.1 If we employ the Rosenfeld Funda-
ear one which arises for the functional 共35兲. Fourier trans- mental Measure Theory48 –50 for F ex hs
关 ␳ 兴 , this nonlocal func-
forming Eq. 共36兲 we obtain tional generates the Percus–Yevick pair direct correlations
⳵␳ˆ 共 k,t 兲 functions in a hard-sphere fluid. Thus, using Eq. 共26兲 we
共 ⌫k B T 兲 ⫺1 obtain the following simple 共random phase兲 approximation
⳵t
⫽⫺k 2 ␳ˆ 共 k,t 兲 ⫹ ␳ b k 2 ĉ 共 k 兲 ␳ˆ 共 k,t 兲 c (2) 共 r; ␳ b 兲 ⫽c PY共 r; ␳ b 兲 ⫺ ␤ v at共 r 兲 , 共42兲

⫹
1
共 2␲ 兲3
冕 dk⬘ k"k⬘ ␳ˆ 共 k ⬘ ,t 兲 ĉ 共 k ⬘ 兲 ␳ˆ 共 兩 k⫺k⬘ 兩 ,t 兲 , 共37兲
where c PY(r) is the Percus–Yevick 共PY兲 approximation31 for
the hard-sphere pair direct correlation function. With this
choice the Fourier transform of c (2) (r) can be carried out
which should be compared with Eq. 共29兲; now there is an analytically
additional term on the right-hand side is nonlinear in ␳ˆ (k,t).
This term describes the coupling between the different Fou- ␤ a␭ 2
rier components of the density fluctuations 共modes兲.42 An ĉ 共 k 兲 ⫽ĉ PY共 k 兲 ⫹ , 共43兲
␭ 2 ⫹k 2
equation almost equivalent to Eq. 共37兲 was derived recently
by considering the mobility M to be a linear function of the where ĉ PY(k) is given by51
4252 J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 A. J. Archer and R. Evans

ĉ PY共 k 兲 ⫽⫺4 ␲␴ 3 冋冉 ␣ ⫹2 ␤ ⫹4 ␥ 24␥

q3 冊
⫺ 5 sin共 q 兲
q

⫹ ⫺ 冉 ␣ ⫹ ␤ ⫹ ␥ 2 ␤ ⫹12␥ 24␥
q2
⫹
q4 冊
⫺ 6 cos共 q 兲
q

⫹ 冉 24␥ 2 ␤
q6
⫺ 4
q 冊册 , 共44兲

where q⫽k ␴ , and the coefficients

共 1⫹2 ␩ 兲 2
␣⫽ ,
共 1⫺ ␩ 兲 4
⫺6 ␩ 共 1⫹ ␩ /2兲 2
␤⫽ , 共45兲
共 1⫺ ␩ 兲 4
␩ 共 1⫹2 ␩ 兲 2
␥⫽
2 共 1⫺ ␩ 兲 4 FIG. 3. Plot of ␳ˆ (k,t) for a quench to the state point k B T ␴ 3 /a⫽0.05 and
depend upon ␩ ⫽ ␲␳ b ␴ /6, the packing fraction. The phase
3 ␩ ⫽0.2, point B in Fig. 1. ␳ˆ (k,t) is shown for times t * ⫽k B T⌫ ␴ 2 t
⫽45,50,55 共in the inset兲 and t * ⫽60,65,67,69 共in the main figure兲. The
diagram for our model fluid is displayed in Fig. 1. The results obtained from the linear theory 关Eq. 共29兲兴 are denoted by a dashed
liquid–gas binodal is calculated by performing the common line, and those from the nonlinear theory 关Eq. 共37兲兴 by a solid line. The
tangent construction on the Helmholtz free energy per par- effect of the coupling between Fourier components 共modes兲 with different
wave numbers, described by the nonlinear theory, becomes increasingly
ticle for the bulk fluid, f ( ␳ b )⫽ f PY( ␳ b )⫺ ␳ b a/2, obtained
significant at later times, whereas for earlier times 共see inset兲, the results
from Eq. 共41兲; f PY( ␳ b ) is the PY compressibility approxima- from the two theories are indistinguishable.
tion for the hard-sphere Helmholtz free energy.31 The spin-
odal is the locus of points for which ⳵ 2 f / ⳵ v 2 ⫽0, where v

06 November 2024 06:40:41

⫽1/␳ b . For simplicity we choose to set the inverse length function corresponds to wave numbers k for which density
scale parameter in the attractive part of the pair potential fluctuations grow exponentially in the early stages of spin-
equal to the hard-sphere diameter, i.e., ␭ ⫺1 ⫽ ␴ . The quantity odal decomposition.
a⫽ 兰 dr v at(r) determines the energy scale. We expect the Some typical plots of ␳ˆ (k,t) at early times are displayed
present model fluid to exhibit a freezing transition for large in the inset to Fig. 3, for a quench to point B in Fig. 1. The
␩, but we do not consider this here. plots are for the reduced times t * ⫽k B T⌫ ␴ 2 t⫽45, 50, and
We are now in a position to use our approximate form 55 in the inset and 60, 65, 67, and 69 in the main figure. The
for ĉ(k), obtained from Eqs. 共43兲 and 共44兲, together with Eq. results obtained from the linear theory, Eq. 共29兲, and the
共37兲, to calculate ␳ˆ (k,t) during the early and intermediate nonlinear theory, Eq. 共37兲, are indistinguishable for the three
times of spinodal decomposition. We assume that at t⫽0 earliest times 共see inset兲. However, at later times the linear
␳ˆ (k,t) takes a small, constant, positive value for all values of theory 共dashed line兲 continues to give just a single peak in
k and choose ␳ˆ (k,t⫽0)⫽10⫺8 . The later time dynamics are ␳ˆ (k,t) that grows exponentially, whereas the nonlinear
insensitive to the specific choice of value for ␳ˆ (k,t⫽0) be- theory 共solid line兲, which includes the effect of coupling be-
cause for short times, when Eq. 共30兲 describes the dynamics tween Fourier components with different wave numbers,
of spinodal decomposition, Fourier components 共modes兲 shows that components with wave numbers different from
with wave number k⬎k c do not grow and are exponentially that predicted by the linear theory can also grow. We see that
damped, whereas components with k⬍k c grow exponen- the effect of the coupling incorporated into the nonlinear
tially with the initial value ␳ˆ (k,t⫽0) as a prefactor. Thus theory becomes increasingly significant at intermediate
choosing a different value for ␳ˆ (k,t⫽0) is effectively a shift times, producing first a shoulder which grows into a bump in
of the time axis. For a quench to the state point k B T ␴ 3 /a ␳ˆ (k,t) at a larger wave number, k ␴ ⯝0.8 than that of the
⫽0.05 and ␩ ⫽0.2 共i.e., the quench to point B, in Fig. 1兲 the main peak which first appears in ␳ˆ (k,t) at early times.
early time growth gives a single peak in ␳ˆ (k,t) at k ␴ ⯝0.6
corresponding to where the maximum of R(k) in Eq. 共30兲 D. Discussion
occurs—see the inset to Fig. 2. We find that R(k) obtained
using Eqs. 共43兲 and 共44兲 for ĉ(k) has a similar form to that In simulation studies of spinodal decomposition the
extracted from molecular dynamics simulation results for a quantity that is often measured in order to characterize the
Lennard-Jones fluid deep inside the spinodal region.52 More- fluid is the structure factor
over, the overall shape of R(k) and the values of k c ␴ are in 1
N

keeping with results for the Lennard-Jones fluid 共for similar S 共 k,t 兲 ⫽ 兺
N i, j⫽1
具 exp共 ik关 ri 共 t 兲 ⫺r j 共 t 兲兴 兲 典 , 共46兲
state points兲 obtained by Evans and Telo Da Gama,36 using a
theory equivalent to the present, and by Abraham,37 using a where 具¯典 is an ensemble average over different realizations
perturbation theory approach. We also plot in Fig. 2 the func- of the stochastic noise in the interval up to time t. For small
tion S(k)⬅(1⫺ ␳ b ĉ(k)) ⫺1 ; the negative portion of this values of k one finds that29
J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 Dynamical density functional theory and spinodal decomposition 4253

1 V. CONCLUDING REMARKS
S 共 k,t 兲 ⯝A 共 t 兲 ⫹ 关 ␳ˆ 共 k,t 兲兴 2 , 共47兲
N This paper falls broadly into two parts. In the first part
we provided an alternative derivation of the DDFT devel-
oped by MT.3,4 Our derivation elucidates the physical ap-
where A(t) is a 共small兲 wave number independent baseline. proximations made in order to construct the theory. The start-
ing point is the Smoluchowski equation 共8兲, the 共generalized兲
Thus, to a reasonable approximation S(k,t)⬀ 关 ␳ˆ (k,t) 兴 2 for
Fokker–Planck equation for the probability distribution
small values of k 共see Refs. 29 and 36兲. Simulation studies of
function P(rN ,t) corresponding to the Langevin equation of
the early stages of spinodal decomposition in model colloidal
motion, Eq. 共10兲. This stochastic basis for the theory means
fluids such as that in Ref. 53, where the authors consider a
that the theory should be applicable to colloidal fluids where
fluid composed of particles interacting via the Lennard-Jones
because of interactions with the solvent particles the momen-
pair potential, display the growth of a single peak in S(k,t). tum degrees of freedom of the colloids equilibrate much
This is, of course, a general feature of the very early stages faster than the positional degrees of freedom. In atomic flu-
of spinodal decomposition and is found in many other ids the equilibration time scale for the momentum degrees of
systems.29,32 Thus, our results for short times showing the freedom can be of the same order as that for the relaxation of
growth of a single peak in ␳ˆ (k,t), and therefore also in positional degrees of freedom and therefore the present
S(k,t), are in keeping with simulation studies. theory may break down for atomic fluids. However, because
The development of a second peak 共shoulder兲 in S(k,t) the correct equilibrium limit is built into this theory, i.e.,
at intermediate times after the quench, at a larger value of k when ⳵␳ (r,t)/ ⳵ t⫽0 the theory is equivalent to Eq. 共23兲, it is
was observed in the two-dimensional calculations of Mao feasible that the present theory would give reasonable results
et al.,43 where they considered a nonlinear extension of for atomic fluids, assuming that one also has a reliable ap-
Cahn–Hilliard theory for polymer mixtures. The order pa- proximation for the excess Helmholtz free energy functional,
rameter for their theory is the deviation of the concentration F ex关 ␳ 兴 . The fact that the correct equilibrium limit is built in
c(r,t)⫽C(r,t)⫺C 0 关i.e., c(r,t) replaces ˜␳ (r,t) in our is, we believe, one of the most appealing features of the
theory兴. They used the approximation that the mobility M theory.
⫽M 0 ⫹M 1 c(r,t), i.e., a linear function of c(r,t), together The further approximation 共beyond assuming that the

06 November 2024 06:40:41

with a square-gradient Helmholtz free energy functional in Smoluchowski equation holds兲 made in deriving the DDFT,
Eq. 共31兲. By linearizing ␦ F/ ␦ C about the average concen- Eq. 共18兲, is to assume Eqs. 共16兲 and 共17兲, exact equilibrium
tration C 0 in the manner leading to Eq. 共34兲 along with the results, are also valid for the nonequilibrium fluid. This ap-
continuity equation 共2兲, Mao et al.43 obtain a nonlinear proximation is equivalent to assuming that the two-particle,
theory that is very similar in structure to that which we ob- three-particle, and higher order correlations in the nonequi-
tain using Eq. 共1兲 with a constant mobility together with the librium fluid are equivalent to those in an equilibrium fluid
truncated functional 共35兲. The main difference between the with the specified one-body density profile. We expect that
two approaches, other than the choice of approximation for this approximation is a reasonable one to make and will not
the Helmholtz free energy functional,54 is that Mao et al.43 result in significant failures of the theory, as long as other
have an adjustable parameter, the ratio M 1 /M 0 , that allows approximations can be justified. We therefore believe that
Eq. 共18兲 may well provide a reliable account of the dynamics
a tuning of the mobility for their polymer blend, whereas in
for a variety of different fluids, provided of course, that one
our approach we are effectively restricted to the choice
has an accurate approximation for the appropriate F ex关 ␳ 兴 .
M 1 /M 0 ⫽1/␳ b . Their results also show that the second peak
The second part of the present paper is concerned with
共shoulder兲 in S(k,t) results from considering a theory that is
the application of the DDFT to the problem of spinodal de-
nonlinear 共second order兲 in the order parameter. The other
composition. Our key result is Eq. 共37兲. For early times after
significant difference from Ref. 43 is that the present theory quenching the fluid, when density fluctuations are small in
uses a microscopic nonlocal functional 共35兲, and therefore amplitude, our theory reduces to the linear theory of Evans
includes the effects of interparticle correlations more accu- and Telo Da Gama.36 Furthermore, this linear theory reduces
rately than a gradient 共Ginzburg–Landau兲 theory such as that to the Cahn–Hilliard theory if we were to use a square gra-
used by Mao et al.43 Nevertheless, in the early and interme- dient Helmholtz free energy functional Eq. 共32兲 rather than
diate times of spinodal decomposition, where sharp inter- the more accurate nonlocal functional, Eq. 共35兲. Cahn–
faces between gaslike and liquidlike domains have not yet Hilliard theory32–35 is known to provide a successful quali-
formed, we should expect good qualitative agreement be- tative description of the early stages of spinodal decomposi-
tween our results and those from the nonlinear Cahn– tion. Since the present theory incorporates all the effects that
Hilliard gradient theory of Mao et al.43 Mao et al. also found Cahn–Hilliard theory describes, and provides a more accu-
that the presence of the nonlinear terms in their theory re- rate treatment of short wavelength correlations36 it should be
sulted in a significant change in the connectivity of contour a reliable quantitative theory for the early stages of spinodal
plots of the order parameter from that found in the absence decomposition.
of these terms (M 1 ⫽0)—see Figs. 3 and 5 of Ref. 43. On However, the key feature of our theory is that it incor-
the basis of our present analysis we would expect this obser- porates a ‘‘mode-coupling’’ term 关final term on the right-
vation to be a general feature of spinodal decomposition at hand side of Eq. 共37兲兴 which describes the coupling between
intermediate time scales. different density fluctuation modes, an effect which becomes
4254 J. Chem. Phys., Vol. 121, No. 9, 1 September 2004 A. J. Archer and R. Evans

important in the intermediate stages of spinodal decomposi- 14

T. R. Kirkpatrick and D. Thirumalai, J. Phys. A 22, L149 共1989兲.
tion. The results of calculations including our ‘‘mode-
15
T. Munakata, J. Phys. Soc. Jpn. 58, 2434 共1989兲.
16
T. Munakata, J. Phys. Soc. Jpn. 59, 1299 共1989兲.
coupling’’ term are in qualitative agreement with those from 17
T. Munakata, Phys. Rev. E 50, 2347 共1994兲.
a recent nonlinear extension to Cahn–Hilliard theory.43 We 18
D. S. Dean, J. Phys. A 29, L613 共1996兲.
believe that further work needs to be done in testing the 19
H. Frusawa and R. Hayakawa, J. Phys. A 33, L155 共2000兲.
predictions of our theory. In particular, a comparison with
20
K. Kawasaki, J. Phys.: Condens. Matter 12, 6343 共2000兲.
21
A. J. Archer and M. Rauscher, cond-mat/0405603.
Brownian dynamics simulation results for spinodal decom- 22
T. Munakata, Phys. Rev. E 67, 022101 共2003兲.
position would be very useful; we do not know of relevant 23
P. N. Pusey, in Liquids, Freezing and Glass Transition, edited by J-P.
simulations which go beyond the very early stages of spin- Hansen, D. Levesque, and J. Zinn-Justin 共North-Holland, Amsterdam,
odal decomposition. Of course, running simulations that get 1991兲.
24
into the stage of spinodal decomposition for which our re- P. N. Pusey and R. J. A. Tough, in Dynamic Light Scattering, edited by R.
Pecora 共Plenum, New York, 1985兲.
sults exhibit deviations from those of the linear theory could 25
C. A. Croxton, Liquid State Physics—A Statistical Mechanical Introduc-
be computationally expensive. tion 共Cambridge University Press, Cambridge, 1974兲.
26
We conclude by mentioning one important conceptual H. Risken, The Fokker-Planck Equation: Methods of Solutions and Appli-
issue regarding the input to our theory of spinodal decompo- cations, 2nd ed. 共Springer-Verlag, Berlin, 1996兲.
27
C. W. Gardiner, Handbook of Stochastic Methods for Physics, Chemistry
sition. Clearly Eqs. 共29兲 or 共37兲 require ĉ(k) as input. As and the Natural Sciences 共Springer-Verlag, Berlin, 1996兲.
emphasized earlier, for an equilibrium state ĉ(k) is simply 28
R. Verberg, I. M. de Schepper, and E. G. D. Cohen, Phys. Rev. E 61, 2967
the Fourier transform of the pair direct correlation function 共2000兲.
and, as such, it can be obtained from integral equation
29
J. K. G. Dhont, J. Chem. Phys. 105, 5112 共1996兲.
30
J. K. G. Dhont, A. F. H. Duyndam, and B. J. Ackerson, Langmuir 8, 2907
theories31 or from simulations. In Ref. 36 results for ĉ(k) 共1992兲.
calculated using Percus–Yevick theory for a Lennard-Jones 31
J-P. Hansen and I. R. McDonald, Theory of Simple Liquids, 2nd ed. 共Aca-
fluid were extrapolated into the unstable spinodal region.55,56 demic, London, 1986兲.
32
Such a procedure is fraught with uncertainty and is difficult J. D. Gunton, M. San Miguel, and P. S. Sahni, in Phase Transitions and
Critical Phenomena, edited by C. Domb and J. L. Lebowitz 共Academic,
to justify. The present DFT formulation of the theory avoids
London, 1988兲, Vol. 8, p. 267.
such problems. c (2) is defined as the second functional de- 33
A. Onuki, Phase Transition Dynamics 共Cambridge University Press, Cam-
rivative of ⫺ ␤ F ex关 ␳ 兴 and provided one makes a division of bridge, 2002兲.

06 November 2024 06:40:41

F ex关 ␳ 兴 into a repulsive reference contribution plus an attrac- 34
J. W. Cahn and J. E. Hilliard, J. Chem. Phys. 31, 688 共1959兲.
tive 共mean-field-like兲 contribution, as in Eq. 共41兲, there is no
35
J. W. Cahn, Acta Metall. 9, 795 共1961兲.
36
R. Evans and M. M. Telo Da Gama, Mol. Phys. 38, 687 共1979兲.
difficulty in calculating c (2) (r; ␳ b ) inside the spinodal. We do 37
F. F. Abraham, J. Chem. Phys. 64, 2660 共1976兲.
not need to make any extrapolation. One could envisage 38
S. M. Allen and J. W. Cahn, Acta Metall. 27, 1085 共1979兲.
treating attractive interactions in a more sophisticated 39
A. J. Bray, Adv. Phys. 43, 357 共1994兲.
40
fashion1 than in Eq. 共41兲 but provided the basic division is Note that inside the spinodal S(k) should not be interpreted as the static
structure factor 关one can define a time-dependent structure factor S(k,t),
maintained one should expect to obtain similar results for
as in Eq. 共46兲兴. Rather S(k) is defined for all state points as (1
ĉ(k). ⫺ ␳ b ĉ(k)) ⫺1 , where ĉ(k) is defined as the Fourier transform of the second
functional derivative of F ex关 ␳ 兴 , i.e., via 共26兲. Thus S(k) can take negative
ACKNOWLEDGMENTS values inside the spinodal region.
41
Starting from the square gradient functional 共32兲 it is straightforward to
The authors are grateful to Christos Likos, Markus Raus- show 共Ref. 5兲 that 1⫺ ␳ b ĉ(k)⫽ ␤ ␳ b 关 ( ⳵ 2 f 0 / ⳵␳ 2 ) ␳ b ⫹Kk 2 兴 .
cher, and Jan Dhont for useful discussions and comments on 42
We note that there is some similarity between the structure of Eq. 共37兲 and
the present work. A.J.A. acknowledges the support of the equations of mode coupling theory—see, for example, E. Zaccarelli,
EPSRC under grant No. GR/S28631/01. G. Foffi, P. De Gregorio, F. Sciortino, P. Tartaglia, and K. A. Dawson, J.
Phys.: Condens. Matter 14, 2413 共2002兲—used to describe the glass tran-
1 sition.
For example, R. Evans, in Fundamentals of Inhomogeneous Fluids, edited 43
by D. Henderson 共Dekker, New York, 1992兲, Chap. 3. Y. Mao, T. C. B. McLeish, P. I. C. Teixeira, and D. J. Read, Eur. Phys. J.
2
S. Dietrich, in Phase Transitions and Critical Phenomena edited by C. E 6, 69 共2001兲.
Domb and J. L. Lebowitz 共Academic, London, 1988兲, Vol. 12, p. 1.
44
See, e.g., C. N. Likos, Phys. Rep. 348, 267 共2001兲.
3
U. Marini Bettolo Marconi and P. Tarazona, J. Chem. Phys. 110, 8032
45
S. Asakura and F. Oosawa, J. Chem. Phys. 22, 1255 共1954兲.
共1999兲.
46
S. Asakura and F. Oosawa, J. Polym. Sci. 33, 183 共1958兲.
47
4
U. Marini Bettolo Marconi and P. Tarazona, J. Phys.: Condens. Matter 12, For a recent summary see J. M. Brader, R. Evans, and M. Schmidt, Mol.
A413 共2000兲. Phys. 101, 3349 共2003兲.
5
R. Evans, Adv. Phys. 28, 143 共1979兲.
48
Y. Rosenfeld, Phys. Rev. Lett. 63, 980 共1989兲.
49
6
W. Dieterich, H. L. Frisch, and A. Majhofer, Z. Phys. B: Condens. Matter Y. Rosenfeld, D. Levesque, and J.-J. Weis, J. Chem. Phys. 92, 6818
78, 317 共1990兲. 共1990兲.
7
B. Bagchi, Physica A 145, 273 共1987兲.
50
Y. Rosenfeld, J. Chem. Phys. 93, 4305 共1990兲.
8
A. Yoshimori, T. J. F. Day, and G. N. Patey, J. Chem. Phys. 108, 6378
51
N. W. Ashcroft and J. Lekner, Phys. Rev. 145, 83 共1966兲.
52
共1998兲. F. F. Abraham, M. R. Mruzic, and G. M. Pound, J. Chem. Phys. 70, 2577
9
J. Dzubiella and C. N. Likos, J. Phys.: Condens. Matter 15, L147 共2003兲. 共1979兲.
10
F. Penna and P. Tarazona, J. Chem. Phys. 119, 1766 共2003兲.
53
J. F. M. Lodge and D. M. Heyes, J. Chem. Phys. 109, 7567 共1998兲.
11
F. Penna, J. Dzubiella, and P. Tarazona, Phys. Rev. E 68, 061407 共2003兲.
54
In Ref. 43 the Flory–Huggins–de Gennes form is used for F 关 C(r,t) 兴 .
12
J. Chakrabarti, J. Dzubiella, and H. Löwen, cond-mat/0403475. 55
C. Ebner, W. F. Saam, and D. Stroud, Phys. Rev. A 14, 2264 共1976兲.
13
K. Kawasaki, Physica A 208, 35 共1994兲. 56
M. M. Telo Da Gama and R. Evans, Mol. Phys. 38, 367 共1979兲.