You are on page 1of 21

Available online at www.sciencedirect.

com

Applied Mathematical Modelling 33 (2009) 978–998


www.elsevier.com/locate/apm

Mathematical modeling for colloidal dispersion


undergoing Brownian motion
Chaocheng Huang
Department of Mathematics and Statistics, Wright State University, Dayton, OH 45435, United States

Received 1 January 2006; received in revised form 1 December 2007; accepted 14 December 2007
Available online 10 January 2008

Abstract

An averaged motion approach for modeling Brownian dynamics for suspension systems of electrically charged particles
in liquid is developed. The continuum model for the motion of particles consists of a system of integral equations coupled
with a degenerate parabolic equation. Existence and uniqueness of global solution for the coupled system are established,
and numerical results for the non-Newtonian viscosity of the mixture in terms of shear rate or Pechlet number are
obtained. The model reveals some non-Newtonian properties such as the well-known shear thinning phenomenon for
the viscosity of colloidal dispersions.
Ó 2008 Elsevier Inc. All rights reserved.

Keywords: Brownian dynamics; Colloidal dispersion; Particle suspension

1. Introduction

Understanding the role of viscosities in colloidal dispersions of particles suspended in fluids is very impor-
tant in many industrial applications [1]. In order to compute the viscosity of the particle–fluid mixture, we first
need to fully understand the motion of the particles in the liquid. In typical colloidal dispersion systems, par-
ticles are often electrically charged and they interact with themselves as well as with the carrier fluid. The par-
ticles may also undergo the Brownian motion. It is well documented that, due to the dispersion effect on the
carrier fluid, the hydraulic property of the mixture is changed dramatically and consequently that the colloidal
dispersion systems usually behave as non-Newtonian fluids [1]. When the Reynolds number of the carrier fluid
is not very small and the particles’ size is in the range 107–109 m, the motion of the individual particles may
be modeled by the Brownian dynamics (BD). In BD models, the dynamics of the carrier fluid is often assumed
to be independent of the particles, whereas the motions of the particles depend on the motion of the fluid only
through Stokes’ drag. Since one needs to trace individual particles, a BD mode may involves a large number of
equations. For more detailed discussion of BD, we refer to [2–4]; we also refer to [5, Chapter 15] for numerical
simulations of BD.

E-mail address: chuang@gauss.math.wright.edu

0307-904X/$ - see front matter Ó 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2007.12.027
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 979

Nomenclature

a particle radius
A Hamaker constant
A0 upper bounds for bðx; y; tÞ
F ðx; tÞ interparticle force
kB Boltzmann constant
L size of container
m particle mass
n spatial dimension
N number of particles
P ðx; tÞ particle density
pe Pechlet number
T temperature
ua attractive potential
ur repulsive potential
V0 volume particles occupied
V ðx; tÞ velocity of carrier fluid
w Brownian
pffiffiffi motion
a ð 3=pÞm2n
b Stokes’ friction
d Dirac measure
e ¼ m=ð6pl0 aÞ
e0 solvent dielectric
er relative dielectric
c_ shear rate of carrier fluid
c rescaled shear rate
~c rescaled shear rate of carrier fluid
CðkÞ Gamma function
Cðn; g; s; x; y; tÞ fundamental solution
j inverse Debye constant
k time scaling parameter
l0 viscosity of carrier fluid (water in our case)
l viscosity of mixture
g0 g0 a is the distance between two neighbouring particles
w0 surface potential of particle
r12 shear stress of the mixture
s12 shear strain
ffi of the mixture
pffiffiffiffiffiffiffiffiffiffi
m ¼ 2k B T is diffusion coefficient
~m dimensionless diffusion coefficient

In this paper, we introduce a continuum model for the colloidal dispersions based on a spatial averaging
argument for BD. This approach, which is also referred as to homogenization, has been used in many appli-
cations of upscaling. We assume that particles are fairly densely populated in fluids so that probability density
functions P ðx; tÞ can be properly defined to describe particle positions. Thus, instead of dealing with individual
particles, we propose to analyze the dynamics of the density functions and their effects on the viscosities of the
mixtures. Towards this end, we shall first develop continuum models for interparticle forces, for hydrody-
namic forces and for the Brownian forces. We shall then use these continuum representations to determine
the governing equations for the density functions P ðx; tÞ of particles with prescribed initial data. Since all
the forces acting on a particle at location x depend also on other particles, these force terms generally are
980 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

functionals of the particle densities P ðx; tÞ. Consequently, the continuum model consists of system of partially
differential equations and integral equations for P ðx; tÞ. This integro-differential system will then be analyzed
mathematically and numerically. Some numerical results based on our model will be presented and be com-
pared with Brownian dynamics simulations.
The paper is organized as follows. In the next section we introduce the Brownian dynamic model for a N-
particle system. In Section 3 we present the averaged motion approach to develop the continuum model for
interparticle forces. The BD model will then be upscaled to the averaged Brownian dynamics (ABD) that con-
sists of a stochastic differential system for random processes fðtÞ ¼ ðf1 ðtÞ; f2 ðtÞÞ where f1 ðtÞ and f2 ðtÞ represent
the positions and the velocities of the particles, respectively. The coefficients in the stochastic differential equa-
tion depend on the averaged interparticle force F that is actually a functional of the entire family of solutions
emanating at t ¼ 0 from points n (with initial density P 0 ðnÞ) and prescribed initial velocity w0 ðnÞ. In other words,
the continuum model is described by a stochastic differential equation strongly coupled with an integral equa-
tion. In Section 4 we use Ito’s formula to derive the degenerate parabolic equation associated with the stochastic
differential equation. We then introduce the fundamental solution of the degenerate parabolic equation assum-
ing that F is a given function. In Section 5 we formulate the deterministic model for ABD that governs the
dynamics of particles’ density P ðx; tÞ and the averaged interparticle force F ðx; tÞ. We then set up a coupled inte-
gro-differential system for ðP ; F Þ in a dimensionless form. In Sections 6 and 7 we mathematically analyze this
system and show that the system has a unique solution for all t > 0. In Section 8 we consider the case where
the diffusion coefficient m of the Brownian motion vanishes, and show that our model reduces to the averaged
motion model studied in [6,7]. In Section 9 we return to the BD model introduced in Section 2 with represen-
tative physical constants and carry out a dimensional analysis. This yields an ABD model with representative
size parameters. This model is then studied numerically in Section 10, where we reproduce the well-known shear
thinning phenomenon for the viscosities of colloidal dispersions described in [1,2, Chapter 15, 3,4].
2. Brownian dynamics

Consider a N-particle system in a liquid. We assume that all the particles are spherical with radius a and
mass m. Let r be the variable vector in Rn and ri the center of the ith particle ði ¼ 1; . . . ; N Þ. Set
ri  rj
rij ¼ jri  rj j; ^rij ¼ :
jri  rj j
The Brownian dynamics is described by, for 1 6 i 6 N ,
m€ri ¼ FPi þ FNi þ FBi ðLangevin’s equationÞ; ð2:1Þ
where FPi , FNi and FBi represent the sum of interparticle forces, the sum of hydrodynamic forces and the sum of
Brownian forces, respectively, acting on the ith particle. The hydrodynamic forces are assumed to be the drag
forces:
FNi ¼ bð_r  Vðr; tÞÞ; ð2:2Þ
where Vðr; tÞ is the velocity of the fluid (assumed to be known), b ¼ 6pl0 a (Stokes friction), and l0 is the vis-
cosity of the fluid. The Brownian forces FBi are stochastic in nature and satisfy [1, p. 66]
hFBi ðtÞi ¼ 0; hFBi ðtÞ  FBi ðsÞi ¼ 2k B T bdðt  sÞ ð2:3Þ
where h  i denotes the expected value, k B is Boltzmann’s constant, T the temperature, and d is the Dirac mea-
sure. For electrically charged particles, the interparticle forces are given by
XN
ouðrij Þ
FPi ¼  ^rij ; ð2:4Þ
j¼1;j–i
orij

where uðsÞ is a potential function. In the case where the particle surface potential is low and remains constant
during interaction between particles, we assume that potential functions have the following form (see [2,3,5,
Chapter 15])
u ¼ ua þ u r ; ð2:5Þ
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 981

where ua represents the attractive force and ur is the repulsive force. The specific choices for ua and ur depend
on physical assumptions. For many applications (see [1,5, Chapter 15]), ua and ur are taken to be
  
A 4a2 4a2 2a2
ua ¼  þ 2 þ 2 ln 1  2 ; ð2:6Þ
12 r2  4a2 r r
ur ¼ 2per e0 w20 a lnð1 þ ejs Þ; ð2:7Þ

where r is the distance between the centers of two neighboring particles, s ¼ r  2a is the separation between
these two particles, A is the Hamaker constant, er is the relative dielectric, e0 is the solvent dielectric, w0 is the
surface potential of the particles, and 1=j is the Debye’s screening length. The repulsive potential (2.7) is de-
rived under the assumption that all particles have the same potential on their surfaces and that each particle is
surrounded by a ‘‘cloud” of ions forming a buffer which prevents other particles from reaching it [1, p. 120].
This implies that particles do not collide, that is, r > 2a. For technical reasons, we further assume that
r > 2a þ 2g0 a; g0 > 0; ð2:8Þ

where 2g0 , the relative separation, is assumed to be independent of a. This assumption implies that ua is a
bounded smooth function. We shall use the choice (2.6) and (2.7) just as an example; all of our analysis will
be carried out under general assumptions. We also remark that in some of the analytical results as well as in
the numerical computations later on, we can actually remove the restriction g0 > 0; see Remark 7.1 (in Section
7) and Section 10.

3. Averaged Brownian dynamics

In this section, we want to replace the discrete BD model for individual particles by a continuum model for
this density function. In view of (2.4)–(2.8), the continuum model for the interparticle force has the
representation
Z
x  x0
F ðx; tÞ ¼ Gðjx  x0 jÞ P ðx0 ; tÞdx0 ; ð3:1Þ
Rn jx  x0 j

where P ðx; tÞ is the density of particles and GðsÞ is a function satisfying


GðsÞ ¼ 0 if s 6 2g0 ; GðsÞ is bounded for 0 6 s < 1: ð3:2Þ
Let xðtÞ ¼ xðt; xÞ (x is in a probability space) be a random process describing particle positions. Then (2.1)
leads to
 
d2 x dx dw
2
¼ F ðx; tÞ  K  V ðx; tÞ þm ; ð3:3Þ
dt dt dt
pffiffiffiffiffiffiffiffiffiffiffiffiffi
where m ¼ m1 2k B T b, wðtÞ is the standard Wiener process, and KðyÞ is a function of y in Rn . To represent
Stokes’ drag we should take KðyÞ ¼ bm1 y for all y. However, for technical reasons, we shall assume that
KðyÞ is truncated:
KðyÞ ¼ bm1 y if jyj 6 S 0 for some large S 0 ; and ð3:4Þ
KðyÞ is a bounded smooth function for y 2 Rn : ð3:5Þ
In most applications, this is not a serious restriction since the expression for the Stokes’ drag is applicable
only in certain limited ranges of the relative velocity dx=dt  V ðx; tÞ. In the numerical results, we do not need
to use this truncation. The system of (3.1)–(3.5) models the dynamics of particles, and is the stochastic descrip-
tion of average Brownian dynamics (ABD).
In the following section we shall recall the construction of the fundamental solution of the degenerate
parabolic equation associated with (3.3). This will be needed in the subsequent sections for establishing a
deterministic continuous model.
982 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

4. Fundamental solutions

Consider the degenerate parabolic equation


ov 1
þ g  rn v þ bðn; g; sÞ  rg v þ m2 Dg v ¼ 0 ð4:1Þ
os 2
for vðn; g; sÞ, where n; g vary in Rn , 0 < s < 1, m is a positive constant and bðn; g; sÞ is a function in Rn
satisfying
m 6 C 0 ; kbkL1 6 A0 ð4:2Þ
for some positive constants. By [8], there exists a unique fundamental solution Cðn; g; s; x; y; tÞ for (4.1), i.e.,
for any fixed ðx; y; tÞ, Cðn; g; s; x; y; tÞ satisfies (4.1) in ðn; g; sÞ for 0 < s < t, and
Cðn; g; s; x; y; tÞjs¼t0 ¼ dðn  xÞdðg  yÞ;
where the product of the two Dirac measures is understood as the measure defined by
Z
gðn; gÞdðn  xÞdðg  yÞdn dg ¼ gðx; yÞ; for any g 2 C 1
0 :
R2n
Furthermore, for any bounded smooth function f ðx; yÞ, the function
Z
vðn; g; sÞ ¼ Cðn; g; s; x; y; tÞf ðx; yÞdx dy
R2n
is the unique bounded solution of (4.1) in 0 < s < t, satisfying
vðn; g; tÞ ¼ f ðn; gÞ:
Finally, for fixed ðn; g; sÞ, Cðn; g; s; x; y; tÞ satisfies the adjoint differential equation
ou 1
  y  rx u  ry  ðbðx; y; tÞuÞ þ m2 Dy u ¼ 0
ot 2
with respect to the variables ðx; y; tÞ for s < t < 1.
Remark 4.1. In [8], it was assumed that bðx; y; tÞ is locally Lipschitz continuous. However, the construction of
the fundamental solution remains unchanged for any uniformly bounded bðx; y; tÞ, and Cðn; g; s; x; y; tÞ is then
a weak solution of (4.1).
We now briefly review the construction of C in order to trace how the various estimates depend on b and m.
Let
 2
jy  gj
2
6n  x þ yþg
2
ðt  sÞ
Rðn; g; s; x; y; tÞ ¼ 2 þ : ð4:3Þ
2m ðt  sÞ m2 ðt  sÞ3
A direct computation shows that for s < t, the function
a pffiffiffi n
W ðn; g; s; x; y; tÞ ¼ expðRðn; g; s; x; y; tÞÞ; a ¼ 3=p m2n ð4:4Þ
ðt  sÞ2n
is the fundamental solution for (4.1) with b  0. In other words, W satisfies, for fixed ðx; y; tÞ,
oW 1
þ g  rn W þ m2 Dg W ¼ 0; 0 < s < t;
os 2
W ðn; g; t  0; x; y; tÞ ¼ dðn  xÞdðg  yÞ:
Furthermore, W has the following semigroup property:
Z
W ðn; g; s;  g; sÞW ð
n;  g; s; x; y; tÞd
n;  g ¼ W ðn; g; s; x; y; tÞ:
n d ð4:5Þ
R2n

We shall construct Cðn; g; s; x; y; tÞ in the form


Z t Z
Cðn; g; s; x; y; tÞ ¼ W ðn; g; s; x; y; tÞ þ ds W ðn; g; s; n; g; sÞQðn; g; s; x; y; tÞdn dg; ð4:6Þ
s R2n
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 983

where Q is the solution to the integral equation


Qðn; g; s; x; y; tÞ ¼ bðn; g; stÞ  rg W ðn; g; s; x; y; tÞ
Z t Z
þ bðn; g; sÞ  ds rg W ðn; g; s; n; g; sÞQðn; g; s; x; y; tÞdn dg: ð4:7Þ
s R2n

Formally a solution of (4.7) can be written in the form


X1
Q¼ Qk ; ð4:8Þ
k¼0

where
Q0 ðn; g; s; x; y; tÞ ¼ bðn; g; sÞ  rg W ðn; g; s; x; y; tÞ
and, for k P 1,
Z t Z
Qkþ1 ðn; g; s; x; y; tÞ ¼ bðn; g; sÞ  ds rg W ðn; g; s; n; g; sÞQk ðn; g; s; x; y; tÞdn dg:
s R2n

We now show convergence of the series in (4.8). Note that at ðn; g; s; x; y; tÞ


 
 abeR  5A0 a CA0 a
  1 R
jQ0 j ¼  r g R 6 eR R2 6 e 2 ; ð4:9Þ
ðt  sÞ2n  mðt  sÞ2nþ12 mðt  sÞ
2nþ12

where and in what follows, C is always a universal constant. Next


Z Z    
CA2 t a Rðn; g; s; n; g; sÞ a Rðn; g; s; x; y; tÞ 
jQ1 j 6 2 0 ds 2nþ12
exp   1 exp  dn dg
m s R2n ðs  sÞ 2 ðt  sÞ2nþ2 2

and, by changes of variables  n ¼ 2~ g ¼ 2~


n;  g, by (4.4) and (4.5),
2  Z t  
CA n g x y 1 1 CA20 a Rðn; g; s; x; y; tÞ
jQ1 j 6 2 0 W pffiffiffi ; pffiffiffi ; s; pffiffiffi ; pffiffiffi ; t ds 6 exp  :
m 2 2 2 2
1 1
s ðs  sÞ2 ðt  sÞ2 m2 ðt  sÞ2n 2

Proceeding by induction, we can obtain


k  
C kþ1 Akþ2
0 aðt  sÞ
2
Rðn; g; s; x; y; tÞ
jQkþ1 j 6
exp  ;
mkþ2 ðt  sÞ2n C k2 2
where CðsÞ is the Gamma function. This yields the convergence in (4.8) and
1
!
CA0 a Rðn; g; s; x; y; tÞ CA0 ðt  sÞ2
jQj 6 2nþ1
exp  þ : ð4:10Þ
mðt  sÞ 2 2 m

Using (4.9) and (4.10) in (4.6), we easily obtain:


Lemma 1. There exists a constant C independent of A0 and m such that
R 1
! !
aeR Cae2 CA0 ðt  sÞ2
jCðn; g; s; x; y; tÞj 6 þ exp 1 ; ð4:11Þ
ðt  sÞ2n ðt  sÞ2n m

where R ¼ Rðn; g; s; x; y; tÞ defined in (4.3).

Remark 4.2. In the case when bðx; y; tÞ is also locally Lipschitz, the fundamental solution C can also be con-
structed in the slightly different form:
Z t Z
Cðn; g; s; x; y; tÞ ¼ W ðn; g; s; x; y; tÞ þ ds Qðn; g; s; n; g; sÞW ðn; g; s; x; y; tÞdn dg; ð4:12Þ
s R2n
984 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

where Q is the solution to a slightly different integral equation


Qðn; g; s; x; y; tÞ ¼ ry  ðbðx; y; tÞW ðn; g; s; x; y; tÞÞ
Z t Z
 ds Qðn; g; s;  g; sÞry  ðbðx; y; tÞW ðn; g; s; x; y; tÞÞdn dg:
n;  ð4:13Þ
s R2n

In Section 10 we shall only need to calculate Cðn; 0; 0; x; y; tÞ, rather than Cðn; g; s; x; y; tÞ, and this is much eas-
ier to do by using (4.12) and (4.13) (by taking g ¼ 0; s ¼ 0) than by using (4.6) and (4.7).

5. Deterministic formulation of ABD

Throughout the paper, we assume that the initial particle density P 0 ðnÞ satisfies
Z
0 6 P 0 ðnÞ 6 C 1 ; P 0 ðnÞdn < 1; ð5:1Þ
Rn

where C 1 is a positive constant. We also assume that the initial particle velocity w0 ðnÞ satisfies

jrw0 ðnÞj 6 C 2 ; detðI þ trw0 ðnÞÞ P h0 for all t > 0; ð5:2Þ

where I is the unit n  n matrix and h0 > 0. The last condition is satisfied if, for instance, the matrix rw0 is
non-negative definite. We introduce the stochastic process vector
 
1 2 def dxðtÞ
fðtÞ ¼ ðf ðtÞ; f ðtÞÞ ¼ xðtÞ; ;
dt

where f1 ðtÞ is the position of the particle and f2 ðtÞ is its velocity at time t. By (3.3), fðtÞ satisfies the stochastic
differential system
df1 ðtÞ ¼ f2 ðtÞdt; ð5:3Þ
2 1 2 1
df ðtÞ ¼ ðF ðf ðtÞ; tÞ  Kðf ðtÞ  V ðf ; tÞÞdt þ m dwðtÞ:
Consider this system for t > s, with initial condition
f1 ðsÞ ¼ n; f2 ðsÞ ¼ g ð5:4Þ
and denote the solution by fn;g;s ðtÞ. Let Cðn; g; s; x; y; tÞ be the fundamental solution of the degenerate parabolic
equation
ou 1
þ g  rn u þ ðF ðn; sÞ  Kðg  V ðn; sÞÞÞ  rg u þ m2 Dg u ¼ 0: ð5:5Þ
os 2
Since Cðn; g; s; x; y; tÞ is the probability density of ff1n;g;s ðtÞ ¼ x; f2n;g;s ðtÞ ¼ yg, it is natural to formulate the aver-
aged Brownian dynamics as follows:

Problem (ABD): Find a function P ðx; tÞ satisfying


Z
P ðx; tÞ ¼ Cðn; w0 ðnÞ; 0; x; y; tÞP 0 ðnÞdn dy; ð5:6Þ
R2n

where Cðn; g; s; x; y; tÞ is the fundamental solution of (5.5) in which the coefficient F ðx; tÞ satisfies
Z
x  x0
F ðx; tÞ ¼ Gðjx  x0 jÞ P ðx0 ; tÞdx0 : ð5:7Þ
Rn jx  x0 j
We observe that the auxiliary function
Z
Pb ðx; y; tÞ ¼ Cðn; w0 ðnÞ; 0; x; y; tÞP 0 ðnÞdn; ð5:8Þ
Rn
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 985

satisfies the parabolic system


1
Pb t þ y  rx Pb þ ry  ððF ðx; tÞ  Kðy  V ðx; tÞÞÞ Pb Þ  m2 Dy Pb ¼ 0 ð5:9Þ
2
for t > 0, and the initial condition
Pb ðx; y; 0Þ ¼ P 0 ðxÞdðy  w0 ðxÞÞ: ð5:10Þ
Obviously,
Z
P ðx; tÞ ¼ Pb ðx; y; tÞdy: ð5:11Þ
Rn

Remark 5.1. Problem (ABD) can be formulated for general measures P P 0 . Consider the special case where P 0 is
a finite linear combination of Dirac measures at points nj ; P 0 ¼ M
j¼1 pj dðn  nj Þ. Then

X
M Z
P ðx; tÞ ¼ pj Cðnj ; w0 ðnj Þ; 0; x; y; tÞdn dy:
j¼1 Rn

We can also write


X
M
P ðx; tÞ ¼ pj Prðf1j ðtÞ ¼ xÞ;
j¼1

where Pr is the probability measure, and fj ðtÞ ¼ ðf1j ðtÞ; f2j ðtÞÞ is the solution of the stochastic differential equa-
tion (5.3) with

X
M
x  f1j ðtÞ
F ðx; tÞ ¼ < Gðjx  f1j ðtÞjÞ >
j¼1 jx  f1j ðtÞj

and

f1j ð0Þ ¼ nj ; f2j ð0Þ ¼ w0 ðnj Þ:


One can prove (by successive iterations, as in [9, Chapter 5, Section 5]) that this stochastic differential sys-
tem has a unique solution.

6. Some estimates

Lemma 2. If ðF ; P Þ is a solution of Problem (ABD) for 0 < t < T , then


Z Z
P ðx; tÞdx ¼ P 0 ðxÞdx ð6:1Þ
Rn Rn

and
jF ðx; tÞj 6 kP 0 kL1 sup jGj; 0 6 t < T ; ð6:2Þ
where G is defined in (3.2).

Proof. Representing the function f ¼ 1 in terms of the fundamental solution, we have


Z
Cðn; g; 0; x; y; tÞdx dy ¼ 1 ð6:3Þ
R2n
986 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

and (6.1) then follows from (5.6). Next, by (3.2) and (5.7),
Z Z
0 0
F ðx; tÞ 6 ðsup jGjÞ P ðx ; tÞdx ¼ ðsup jGjÞ P 0 ðxÞdx: 
jxx0 j>2g0 Rn

Lemma 3. If ðF ; P Þ is a solution of Problem (ABD) for 0 < t < T , then


 pffi 
A1 t
P ðx; tÞ 6 C exp for 0 < t < T ; ð6:4Þ
m
where C and A1 are positive constants independent of T (and m).

Proof. From (5.6) and Lemma 1 we obtain


 pffi  Z  
Ca A1 t 1
P ðx; tÞ 6 2n exp exp  Rðn; w0 ðnÞ; 0; x; y; tÞ dn dy; ð6:5Þ
t m R2n 2
where by (6.2), A1 is a constant independent of T. To estimate the last integral we change the variables as
 
 y þ w0 ðnÞ
n¼nxþ t; y ¼ y  w0 ðnÞ:
2
Note that by (5.2),
   !
oðn; y Þ I þ 2t rw0 2t I
det ¼ det ¼ detðI þ trw0 Þ P h0 : ð6:6Þ
oðn; yÞ rw0 I
It follows that the integral in (6.5) is bounded by, after changes of variables,
Z ! Z ! Z !
1 jy j2 6j
nj2  1 6jnj2  jy j2
exp  2  2 3 dn dy ¼ exp  2 3 dn exp  2 dy
h0 R2n 4m t m t h0 R n mt Rn 4m t
Z Z !
2
1 n 3n 2 n jzj
¼ m t2 n
expð6jzj Þdzm t2 exp  dz 6 Cm2n t2n :
h0 Rn Rn 4

Hence, by recalling the definition of a (see (4.4)),


Z  
a 1 0
exp  Rðn; w ðnÞ; 0; x; y; tÞ dn dy 6 C: ð6:7Þ
t2n R2n 2
Substituting this in (6.5), the assertion (6.4) follows. h

Remark 6.1. If we allow g0 ¼ 0 in assumption (2.8), then, for the potentials ua , ur as in (2.6), (2.7), the function
GðsÞ in (3.1) will have a singularity like s1n at s ¼ 0. All the estimates in this section then remain unchanged
except that in (6.4) A1 must be replaces by CkP ð; tÞkL1 .

7. Existence and uniqueness for Problem (ABD)

Theorem 1. If (5.1) and (5.2) hold, then there exists a unique solution ðF ; P Þ to Problem (ABD) for all t > 0.

Proof. Let T be any positive number. Denote by AT the set of all functions F ðx; tÞ defined in
XT ¼ fðx; tÞ : x 2 Rn ; 0 6 t < T g
having finite norm
kF kL1  kF kL1 ðXT Þ 6 sup jGj:
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 987

For every F 2 AT , we define a function P ðx; tÞ by (5.6), where C is the fundamental solution for (5.5) corre-
sponding to this F, and a function F  by
Z
 x  x0
F ðx; tÞ ¼ Gðjx  x0 jÞ P ðx0 ; tÞdx0 : ð7:1Þ
Rn jx  x0 j
From the proof of Lemma 2, we see that (6.1) is still valid and, consequently, F  ðx; tÞ still satisfies (6.2) in XT .
Hence, F  2 AT . Thus the mapping S defined by
SðF Þ ¼ F 
maps AT into itself. We shall next prove that
S is a contraction mapping if T is small; ð7:2Þ
this will establish existence and uniqueness for a small time interval.
To prove (7.2), we take any F 1 , F 2 in AT and denote the corresponding P ; C; Q; Qj in (5.6), (4.7), (4.8) by
P 1 ; C1 ; Q1 ; Q1j and P 2 ; C2 ; Q2 ; Q2j , respectively. We shall first estimate
Z t Z
C1 ðn; g; s; x; y; tÞ  C2 ðn; g; s; x; y; tÞ ¼ ds W ðn; g; s; n; g; sÞðQ1  Q2 Þðn; g; s; x; y; tÞdn dg: ð7:3Þ
s R2n

As in Section 4, we have the estimates


Ca R
jQ10  Q20 j 6 kF 1  F 2 kL1 jry W j 6 kF 1  F 2 kL1 2nþ12
e 2
ðt  sÞ
and
Z Z  Z t Z 
 t   
jQ1kþ1  Q2kþ1 j 6 kF 1  F 2 kL1  ðrg W ÞQ1k  þ C  ds ðrg W ÞðQ1k  Q2k Þ:
s R2n s R2n

Proceeding by induction and using the estimates for Qk in Section 4, we deduce that
k R
C kþ2 aðt  sÞ2 e2
jQ1kþ1  Q2kþ1 j 6 kF 1  F 2 kL1 2n
mkþ2 ðt  sÞ Cðkþ1
2
Þ
where C is a constant independent of T and m. It follows that
1
!
X
1
Ca Cðt  sÞ2
1 2
jQ  Q j 6 jQ1kþ1  Q2kþ1 j 6 kF 1  F 2 kL1 2nþ12
exp
k¼0 mðt  sÞ m

and then, from (7.3) (as in the derivation of (4.11)),


  pffiffiffiffiffiffiffiffiffiffi 
Ca C ts R
jC1  C2 j 6 kF 1  F 2 kL1 2n
exp  1 e 2 :
ðt  sÞ m
Using (6.7) we then obtain
Z  pffi 
C t
jðC1  C2 Þðn; w0 ðnÞ; 0; x; y; tÞjdn dy 6 C exp  1 kF 1  F 2 kL1
R2n m
and consequently, from the representation (5.6) for P 1 and P 2 ,
Z  pffi 
0 C t
jP 1 ðx; tÞ  P 2 ðx; tÞj 6 jðC1  C2 Þðn; w ðnÞ; s; x; y; tÞjP 0 ðnÞdn dy 6 Cðexp  1ÞkF 1  F 2 kL1 ;
R 2n m
where in the last inequality we used the first part of the assumption (5.1). By definition of the mapping S, it
then follows that
pffiffiffiffi
C T
kSðF 1 Þ  SðF 2 ÞkL1 6 kF 1  F 2 kL1 ; 0 < t < T
m
so that S is a contraction mapping if T is small enough, as asserted in (7.2).
988 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

To extend the solution uniquely to all t > 0, it suffices to extend it to 0 6 t 6 T  , where T  is any positive
number. Suppose that the solution ðF ; P Þ exists for all 0 6 t 6 t0 where 0 < t0 6 T  . We shall prove that it can
then be extended uniquely to 0 6 t 6 t0 þ T , where T is independent of t0 (although it may depend on T  ); this
clearly will complete the proof of the theorem.
From the semigroup property (4.5) for W and from (4.6), one can easily derive the semigroup property for C :
Z
Cðn; g; s; ~ g; sÞCð~
n; ~ g; s; x; y; tÞd ~
n; ~ g ¼ Cðn; g; s; x; y; tÞ:
n d~
R2n

This property can then be used to deduce from (5.8) the relation
Z
Pb ðx; y; tÞ ¼ Cðn; g; s; x; y; tÞ Pb ðn; g; sÞdn dg:
R2n

By (5.11) we then have, for t > t0 ,


Z
P ðx; tÞ ¼ Cðn; g; t0 ; x; y; tÞ Pb ðn; g; t0 Þdn dg dy; ð7:4Þ
R3n

where Pb ðn; g; t0 Þ is given by (5.8).


Using the relation (7.5) for P 1 and P 2 and proceeding as in the proof of (7.2), we derive, for t > t0 , the
estimates
Z
jP 1 ðx; tÞ  P 2 ðx; tÞj 6 jðC1  C2 Þðn; g; t0 ; x; y; tÞj Pb ðn; g; t0 Þdn dg dy
R3n
Z   pffiffiffiffiffiffiffiffiffiffiffi   
Ca C t  t0 R b
6 kF 1  F 2 kL1 2n
exp  1 exp  P ðn; g; t0 Þdn dg dy;
R3n ðt  t 0 Þ m 2

where R ¼ Rðn; g; t0 ; x; y; tÞ, and as in the proof of Lemma 3,


Z  
Ca Rðn; g; t0 ; x; y; tÞ
2n
exp  dn dy 6 C:
R2n ðt  t 0 Þ 2
Hence
  pffiffiffiffiffiffiffiffiffiffiffi Z
C t  t0
jP 1 ðx; tÞ  P 2 ðx; tÞj 6 CkF 1  F 2 kL1 exp 1 sup Pb ðn; g; t0 Þdg
m Rn n

and, by (5.8), (4.11) and (6.7), the last integral is bounded by a constant independent of t0 . It follows that S is a
contraction if t  t0 < T for some small T > 0 independent of t0 . This completes the proof of Theorem 1 h.

Remark 7.1. It can be shown, by a standard potential theory argument, that the function Cðn; g; s; x; y; tÞ is
Hölder continuous in t and, by means of (5.6), that P ðx; tÞ is Hölder continuous in t. From (5.7) it then follows
that F ðx; tÞ is Hölder continuous in both x and t. Consequently the fundamental solution Cðn; g; s; x; y; tÞ sat-
isfies the parabolic equation in the classical sense (cf. Remark 4.1).
We conclude this section by deriving a bound on P ðx; tÞ for jxj ! 1. From (5.6) and Lemma 1, we have
Z Z  
C 0 eCt Rðn; w0 ðnÞ; 0; x; y; tÞ
P ðx; tÞ 6 2n P 0 ðnÞdn exp  dy; ð7:5Þ
t Rn Rn 2
where the constant C 0 may depend on m. Setting
y þ w0 ðnÞ
X ¼ x  n; Y ¼ :
2
We can write
   
1 3 t2 2 2 2 3 h 2 2
Rðn; w0 ðnÞ; 0; x; y; tÞ ¼ 2 3 jY  w0 j þ jX j  2X  Yt þ jY j t2 P 2 3 jX j þ ejY j t2  C h t2
2 mt 3 mt 4
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 989

for any 0 < h < 1, where C h and e are positive numbers depending on h. It follows that
Z    Z  
Rðn; w0 ðnÞ; 0; x; y; tÞ 3h 2 3C h 3e 2
exp  dy 6 exp  2 3 jX j þ 2 exp  2 jY j dy
Rn 2 4m t mt Rn mt
 
3h 2 3C h p ffi
6 C 0 exp  2 3 jX j þ 2 t:
4m t mt
Substituting this into (7.5), we obtain
 Z  
C0 3C h 3h 2
P ðx; tÞ 6 2n1 exp Ct þ 2 exp  2 3 jx  nj P 0 ðnÞdn: ð7:6Þ
t 2 mt Rn 4m t
This yields a decay of P ðx; tÞ as jxj ! 1 (for any fixed t > 0). In particular, we have the following:
Theorem 2. If P 0 ðnÞ is compactly supported in Bd , the ball of radius d centered at the origin, then, for all t > 0,
  !
2
C0 C0 C 1 ðjxj  dÞ
P ðx; tÞ 6 1 exp Ct þ exp  if jxj P d; ð7:7Þ
t2n2 t t3

where C 0 and C 1 are constants depending only on m and initial data.

8. Case m ¼ 0

In [6] the authors introduced a model of averaged motion of charged particles:

d2 wðx; tÞ
¼ H ðwðx; tÞ; tÞ; ð8:1Þ
dt2
dwðx; 0Þ
wðx; 0Þ ¼ 0; ¼ w0 ðxÞ; ð8:2Þ
dt
where H ðx; tÞ is the averaged electric force given by
H ðx; tÞ ¼ ruðx; tÞ; ð8:3Þ

u is the solution of

Duðx; tÞ ¼ P ðx; tÞ in Rn ðn P 3Þ; ð8:4Þ


uðx; tÞ ! 0 if jxj ! 1;

P ðx; tÞ is the density of particles at time t and, by conservation of mass,

P ðx; tÞ ¼ P 0 ðw1 ðx; tÞÞJ ðw1 Þðwðx; tÞÞ; ð8:5Þ

where P 0 ðxÞ is the initial density and J ðÞ is the Jacobian determinant.
It was proved in [6] that system (8.1)–(8.5) has a unique classical solution for small time t > 0, and that a
global solution does not exist in general.
Theorem 3. The model of averaged Brownian dynamics formally reduces, in the case m ¼ 0 and KðxÞ ¼ 0, to the
model (8.1), (8.2), and (8.5) with force H ðx; tÞ ¼ F ðx; tÞ.

Proof. Setting f ¼ w ¼ ðw1 ; w2 Þ, we have, by (3.3),


dw1 ¼ w2 ; dw2 ¼ F ðw1 ; tÞ; ð8:6Þ
also,
w1 ðx; 0Þ ¼ x; w2 ðx; 0Þ ¼ w0 ðxÞ:
990 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

We can rewrite (8.6) in the form


 
dw y
¼ Mðw; tÞ; Mðx; y; tÞ ¼ : ð8:7Þ
dt F ðx; tÞ
With Pb defined as in (5.8), we formally have, from (5.9) with m ¼ 0,
d b o Pb ow ow o Pb
ð P ðw1 ; w2 ; tÞÞ ¼ þ rx Pb  1 þ ry Pb  2 ¼ þ w2  rx Pb þ F  ry Pb ¼ 0
dt ot ot ot ot
so that, by (5.10),
Pb ðw1 ; w2 ; tÞ ¼ Pb ðx; y; 0Þ ¼ P 0 ðxÞdðy  w0 ðxÞÞ
or
Pb ðx; y; tÞ ¼ P 0 ððw1 Þ1 Þdððw1 Þ2  w0 ððw1 Þ1 ÞÞ; ð8:8Þ
1 1 1
where w ¼ ððw Þ1 ; ðw Þ2 Þ. For any continuous function gðxÞ with compact support, we have
Z Z
P ðx; tÞgðxÞdx ¼ Pb ðx; y; tÞgðxÞdy dx:
Rn R2n

By changing variables ðn; gÞ ¼ w1 ðx; y; tÞ and using (8.8), we find that the right-hand side is equal to
Z
P 0 ðnÞdðg  w0 ðnÞÞgðw1 ðn; g; tÞÞJ ðwðn; g; tÞÞdn dg:
R2n

Since, by (8.7) and direct computation,


d
J ðwÞ ¼ J ðwÞtraceðrMÞ ¼ 0;
dt
we obtain
J ðwðn; g; tÞÞ ¼ J ðwðn; g; 0ÞÞ ¼ 1:
We thus conclude that
Z Z Z
0
P ðx; tÞgðxÞdx ¼ P 0 ðnÞdðg  w ðnÞÞgðw1 ðn; g; tÞÞdn dg ¼ P 0 ðnÞgðw1 ðn; w0 ðnÞ; tÞÞdn: ð8:9Þ
Rn R2n Rn

Changing variables x ¼ w1 ðn; w0 ðnÞ; tÞ, we find that the right-hand side of (8.9) is equal to
Z
P 0 ðw1 1
1 ÞJ ðw1 ÞgðxÞdx
Rn

where w1 0
1 is the inverse of the mapping n#w1 ðn; w ðnÞ; tÞ for fixed t. Since g is arbitrary, (8.5) follows. h

Remark 8.1. If we denote by Pb m the density Pb ðx; y; tÞ corresponding to m > 0, then, by Ito’s formula,

d b m2
ð P m ðw1 ; w2 ; tÞÞ þ ðDy Pb m Þðw1 ; w2 ; tÞ ¼ 0;
dt 2
where ðw1 ; w2 Þ is a solution of (8.6) (assuming KðxÞ  0). Repeating the proof of Theorem 3, one can show that
formally, if Pb m ðx; y; tÞ ! Pb ðx; y; tÞ as m ! 0, then P ðx; tÞ satisfies (8.5).

9. Dimensional analysis

In this section we non-dimensionalize the ABD model in R3 using representative numbers for various phys-
ical constants. The related dimensionless coefficients and potentials in the ABD model developed in Section 5
will then be calculated explicitly in preparation for numerical simulations of the next section.
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 991

We take the fluid flow as a shear flow with velocity


V ¼ ð_cx2 ; 0; 0ÞT
where c_ > 0 is the shear rate. The shear stress r12 of the fluid of the particle–liquid system is related to the
shear rate c_ by
r12 ¼ l_c;
where l is the shear viscosity. For Newtonian fluids, l is independent of c_ . Since colloidal dispersions behave
as non-Newtonian fluids, we expect the shear viscosity l to depend on c_ , and thus we shall write l ¼ lð_cÞ. The
shear stress r12 consists of two parts. The first part is due to the carrier (Newtonian) fluid with a constant vis-
cosity l0 > 0. The second part, s12 , is caused by the particle motion and thus depends on the particle density P.
Therefore
s12
r12 ¼ l0 c_ þ s12 or l ¼ l0 þ : ð9:1Þ
c_
For isolated particles

1 X N
rij;x1 rij;x2 ouij
s12 ¼ :
2V 0 i;j¼1;j–i rij orij

where rij;xk is the kth component of ri  rj and V 0 is the volume that the particle–fluid system occupies; see
[2,3]. From (9.1) it follows that

1 XN
rij;x1 rij;x2 ouij
l ¼ l0 þ : ð9:2Þ
2_cV 0 i;j¼1;j–i rij orij

In applications, one is interested in understanding how the viscosity l depends on the shear rate c_ .
We recall that the discrete interparticle force FPi is defined by (2.4) and (2.5) where, by (2.6) and (2.7),
!
dua 2Aa2 r 1 1
¼ 2
þ 3 3 ; ð9:3Þ
dr 3 ðr2  4a2 Þ r r  2a2 r
dur 1
¼ 2per e0 w20 ja jðr2aÞ : ð9:4Þ
dr e þ1
Consider a volume element DV in R3 with a point x0 2 DV , and denote by P the density of particles. The
number of particles contained in DV is given by
3
N ðDV Þ ¼ jDV jP ðx0 ; tÞ; jDV j ¼ volume of DV :
4pa3
Therefore, for any smooth function f ðxÞ,
X Z
3 3
f ðxj Þ  f ðx 0 ÞjDV jP ðx 0 ; tÞ  f ðxÞP ðx; tÞdx:
xj 2DV
4pa3 4pa3 DV

Let L be the characteristic length scale of the fluid region (for instance, the size of a container). Assuming that
f ðxÞ ¼ f ðx1 ; x2 Þ is independent of x3 , it follows that:
X N Z Z
3 0 3L
f ðxj Þ  3
f ðxÞP ðx; tÞdx dx 3 ¼ 3
f ðxÞP ðx; tÞdx0 ; x0 ¼ ðx1 ; x2 Þ;
j¼1
4pa D3
L
2pa D 2
L

k
where DkL ¼ ðL; LÞ is the fluid region for k ¼ 3. Since the velocity is independent of x3 , it is reasonable to
assume that particle motion takes place only in x0 -plane and that P ðx; tÞ is also independent of x3 . Hence,
by (9.3), for small a, the attractive force
992 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

XN
dua i xi  xj
 ðjx  xj jÞ i
j¼1
dr jx  xj j
j–i

can be approximated by
Z !
LA ri 1 1: xi  x
 þ  P ðx; tÞdx; ð9:5Þ
pa D2L ðr2i  4a2 Þ2 r3i r3i  2a2 ri jxi  xj

where ri ¼ jxi  xj. Analogously by (9.4), the repulsive force


X dur xi  xj
 ðjxi  xj jÞ i
j¼1
dr jx  xj j
j–i

can be approximated by
Z
3L 1 xi  x
er e0 w20 ja 3 jðr 2aÞ
P ðx; tÞdx: ð9:6Þ
a D2L e i þ 1 jxi  xj

Therefore, in (5.7) the potential G is given by GðsÞ ¼ Ga ðsÞv þ Gr ðsÞv, where


!
LA s 1 1
Ga ðsÞ ¼  þ  ;
pam ðs2  4a2 Þ2 s3 s3  2a2 s
3L 1
Gr ðsÞ ¼ er e0 w20 ja 3 jðs2aÞ
ame þ1
and v is the characteristic function of D2L n D22að1þg0 Þ .
Let e be the inertial time scale, and k the real time for computation, scaled on e [1, p.162]. By [1, p. 442],
m m
e¼ ¼ : ð9:7Þ
b 6pl0 a
We scale the phase variables and the time respectively, by

x ¼ L~x; n ¼ L~
n; y ¼ N~y ; g ¼ N~
g; t ¼ ke~t; s ¼ ke~s; ð9:8Þ

where
L
N¼ ð9:9Þ
ke
is the scale for the velocity. In the new variables, ABD model (5.5)–(5.7) becomes
Z
Pe ð~x; ~tÞ ¼ e ~
Cð ~ 0 ð~
n; w nÞ; 0; ~x; ~y ; ~tÞP 0 ð~
nÞd~
n d~y ; ð9:10Þ
R2 R2
Z 0
Fe ð~x; ~tÞ ¼ e x  ~x0 jÞ ~x  ~x Pe ð~x0 ; ~tÞd~x0 ;
Gðj~ ð9:11Þ
R2 j~x  ~x0 j

e ~
where Cð g; ~s; ~x; ~y ; ~tÞ is the fundamental solution of
n; ~

o~
u 1
g  r~n þ ð Fe ð~
þ~ e ð~
n; ~sÞ  K g  Ve ð~ u þ ~m2 D~g ~u ¼ 0;
nÞÞÞ  r~g ~ ð9:12Þ
o~s 2
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 993

where

ec_ ¼ ke_c ¼ L c_ ;
N
e ð~ keb ~
K nÞ ¼ n ¼ k~
n ðfor j~nj < L1 S 0 Þ;
m !
ec_ n
~2
Ve ð~
nÞ ¼ ðsince g  V ðnÞ ¼ N ð~ g  Ve ð~
nÞÞÞ;
0

w ~ ¼ 1 w0 ðnÞ;
~ 0 ðnÞ
N
e
GðsÞ ¼G e a ðsÞ~ vþGe r ðsÞ~
v
and
!
e keL2 s 1 1
G a ðsÞ ¼ Ga ðLsÞ ¼ A1 þ  ;
N ðs2  4a2 L2 Þ2 s3 s3  2a2 L2 s
2
e r ðsÞ ¼ keL Gr ðLsÞ ¼ U1 j1 1
G 1 Þ ;
N e jLðs2aL þ1
v is the characteristic function of D21 n D22aL1 ð1þg0 Þ , and
~
kem2 2kek B T b
~m2 ¼ ¼ ; ð9:13Þ
N2 m2 N 2
kAe
A1 ¼ ; ð9:14Þ
paNm
3keL3
U1 ¼ er e0 w20 3 ; ð9:15Þ
a mN
particle radius
j1 ¼ ja ¼ ð9:16Þ
double layer thickness
are dimensionless constants. The relationship between the scaled and unscaled quantities for P, F, C are
Pe ð~x; ~tÞ ¼ P ðx; tÞ;
ke
Fe ð~x; ~tÞ ¼ F ðx; tÞ;
N
e ~
Cð n; ~g; ~s; ~x; ~y ; ~tÞ ¼ L2 N 2 Cðn; g; s; x; y; tÞ:
2
Physically, ae2m is the inertial energy, A is the dispersion energy, er e0 w20 a is the electrostatic energy, and 2k B T is
the thermal energy (see [1, p. 465]). We choose the initial density such that
Z
Pe 0 ðxÞdx ¼ /0 jD31 j ¼ 8/0 ð/0 is volume fractionÞ: ð9:17Þ
D31

The relation (9.2) becomes


Z 0 0
m 3L 0 ðx1  x1 Þðx2  x2 Þ
l ¼ l0 þ Gðjx  x jÞ P ðx; tÞP ðx0 ; tÞdx dx0
2_cV 0 2pa3 D4L jx  x0 j

or in the dimensionless form,


Z 0 0
l
¼ 1 þ l1 e  x0 jÞ ðx1  x1 Þðx2  x2 Þ Pe ðx; tÞ Pe ðx0 ; tÞdx dx0 ;
Gðjx ð9:18Þ
l0 D41 jx  x0 j

where, since V 0 ¼ ð2LÞ3 ,


994 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

 2 1
ke 3L keL
l1 ¼   m  L5 ð9:19Þ
2l0 ec_ ð8L3 Þ 2pa3 N
is also dimensionless.
We now compute the above constants by using representative physical constants. In the SI unit system,
these constants are chosen and or calculated as follows:

Symbol Name Expression Value ðk ¼ 103 Þ Reference


4
l0 Viscosity of water – 8:91  10 [1, p. 507]
a Radius of particle – 107 –
m Mass of particle 8pa3 =3 8:38  1021 –
kB Boltzmann’s constant – 1:38  1023 [1, p. xvii]
T Temperature – 300 –
A Hamaker’s constant 24k B T 9:94  1020 [1, p. 260]
j1 Debye’s length a 107 [1, p. 214]
er e0 w20 a Electrostatic 150k B T 6:21  1019 [1, p. 472]
b Stokes friction 6pl0 a 1:68  109 (??)
e Inertial time m=b 5  1012 (9.7)
L Length scale – 106 –
N Velocity scale L=ðkeÞ
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2:0042  102 (9.9)
m Brownian
m 2bk B T =m 4:45  105 (3.3)

Using the above quantities, we obtain

Symbol Expression Value (k ¼ 103 ) Reference


23 3 2 2
~2
m 2:46  10 k L 2:46  10 (9.13)
A1 9:40  1016 k2 L1 9:40  104 (9.14)
U1 5:53  107 k2 L2 5:53  101 (9.15)
j1 ja 1 (9.16)
l1 5:63  1013 L2 k1 ðec_ Þ1 5:63  102 ðec_ Þ1 (9.19)

Remark 9.1. We observe that, by the change of variables x ¼ aL1 y,

Z Z !
e a ðjxjÞdx ¼  A1 L jyj 1 1
G 2 2
þ 3
 3
dy
D21 a D2L=a nBð2þ2g0 Þ ðjyj  4Þ jyj jyj  2jyj
Z !
3 jyj 1 1
¼ 9:4  10 2 3 2
þ
dy 3

D2L=a nBð2þ2g0 Þ ðjyj  4Þ jyj  2jyj jyj
Z !
La1
jyj 1 1
¼ 9:4  103 2 2
þ 3 3 dy;
ð2þ103 Þ ðjyj  4Þ jyj jyj  2jyj
Z 2 Z
e r ðjxjÞdx ¼ U1 j1 a 1
G dy
D21 L2 D2L=a nBð2þ2g0 Þ expðjyj  2Þ þ 1
Z
1
¼ 5:53  101 dy
D2L=a nBð2þ2g expðjyj  2Þ þ 1

Z L=a
1
¼ 2p  5:53  101 dy:
2þ104 expðjyj  2Þ þ 1
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 995

The two integrands on the right-hand sides are integrable away from r ¼ 2. The ratio between the factors of
the attractive and the repulsive forces is of order 102 . The first integral depends on the relative separation g0 .
Some numerical values for sup F a and sup F r , for Pe 0 ¼ 2:4~v and several choices of g0 , are shown in the follow-
ing table:

g0 5 1 101 102 103 104


sup Fe a 8:2  104 1:2  102 0.2 1.7 17.8 177.1
sup Fe r 5:3  102 2.4 5.3 5.8 5.8 13.9

As in [1, p. 69], we introduce the diffusion coefficient D0 of an isolated spherical particle of radius a in a liquid
with viscosity l0 by D0 ¼ k B T =ð6pl0 aÞ, and the Pechlet number [1, p. 464] (before scaling) Pe ¼ a2 c_ =D0 . The
Pechlet number quantifies the relative importance of the Brownian force to the shear forces. If Pe  1, then
Brownian forces dominate. If Pe 1, then the shear forces dominate. By the scaling in (9.8), we obtain
a2 ec_ 27pl20 aec_
Pe ¼ ¼ : ð9:20Þ
D0 ke 2k B T k
The range of interests for (dimensionless) ec_ is the range corresponding to the Pechlet number in the interval
101 < Pe < 10. Since, by (9.20),

ec_ ¼ 2k B T k Pe ¼ 1:23  106 Pe;


27pl20 a
this range corresponds to
1:23  107 < ec_ < 1:23  105 : ð9:21Þ
In the next section we demonstrate an example of gðec_ Þ for ec_ in the range (9.21).

10. A simple numerical example

In this section, we compute a simple example to illustrate our model is consistent with particle method.
Extensive numerical schemes and numerical analysis for our ABD model will be discussed in separate papers.
I choose the initial data

8/0 for x 2 D1=2 ;
w0 ¼ 0; P 0 ðxÞ ¼ ð10:1Þ
0 otherwise;
where /0 is the volume fraction of particles, and compute the viscosity l as a function of c_ . This initial data
represent the situation that particles stay away from the boundary. The choice of the constant for P 0 is con-
sistent with (9.17). By Theorem 2, one sees that the density P ðx; tÞ decays exponentially for large jxj. It follows
that, in the dimensionless form (9.10)–(9.12), the corresponding density Pe is very small for jxj > 1 when L is
reasonably large. Therefore, we shall assume that P ðx; tÞ ¼ 0 for jxj > 1. Since we assumed that the velocity
V ðx; tÞ ¼ V ðx1 ; x2 ; tÞ as well as all the quantities we need to compute are independent of x3 . The simulation will
actually take place only in the domain D21 ¼ ½1; 1
. We shall first compute the density Pe of the dimensionless
2

ABD model. For convenience, we shall drop all the tildes in (9.10)–(9.12). By (4.12) and (4.13), formula (9.10)
(without tildes) can be rewritten as
Z Z 1 Z 1
P ðx; tÞ ¼ b
Cðn; 0; 0; x; y; tÞP 0 ðnÞdn dy ¼ P 0 ðx; tÞ þ H ðx; y 1 ; y 2 ; tÞdy 1 dy 2 ; ð10:2Þ
D41 1 1

where
Z 1 Z 1 Z 1 Z 1
Pb 0 ðx; tÞ ¼ W ðn; n; 0; 0; x; y 1 ; y 2 ; tÞP 0 ðnÞdn1 dn2 dy 1 dy 2 ; ð10:3Þ
1 1 1 1
996 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

150

A = 24
100
φ 0 = 0.3

μ
μ0

50

0 -1 0 1
10 10 10
Pe
Fig. 1. Hamaker’ number A = 24kBT, 30% volume fraction of particles, time step = 0:01.

W ðn; g; s; x; y; tÞ is defined in (4.4), and H ðx; y; tÞ defined by


Z t Z 1 Z 1 Z 1 Z 1 Z 1 Z 1
H ðx; y; tÞ ¼ ds Qðn; 0; 0; n; g; sÞW ðn; g; s; x; y; tÞP 0 ðnÞdn dg dn;
0 1 1 1 1 1 1

where Q is defined by Eq. (4.13). By (4.13) and integration by parts, we find that H ðx; y; tÞ satisfies

150

A = 50
100
φ0 = 0.3

50

0 -1 0 1
10 10 10
Pe
Fig. 2. Hamaker’ number A = 50kBT, 30% volume fraction of particles, time step = 0:01.
C. Huang / Applied Mathematical Modelling 33 (2009) 978–998 997

250

200
A = 24
φ0 = 0.38
150
μ
μ0
100

50

0
10 -1 0.2 10 0 10 1
Pe
Fig. 3. Hamaker’ number A = 24kBT, 38% volume fraction of particles, time step = 0:01.

250

200
A = 50
φ0 = 0.38
150

100

50

0 -1 0 1
10 0.2 10 10
Pe
Fig. 4. Hamaker’ number A = 50kBT, 38% volume fraction of particles, time step = 0:01.
Z t Z 1 Z 1 Z 1 Z 1
H ðx; y; tÞ ¼ ds ðbð g; sÞ  rg W ðn; g; s; x; y; tÞÞH ðn; g; sÞdn dg
n;  ð10:4Þ
0 1 1 1 1
Z t Z 1 Z 1 Z 1 Z 1 Z 1 Z 1
þ ds bð g; sÞ  rg W ðn; g; s; x; y; tÞ
n;  W ðn; 0; 0; n; g; sÞP 0 ðnÞdn dn dg;
0 1 1 1 1 1 1

where
bðx; y; tÞ ¼ F ðx; tÞ  kðy  V ðxÞÞ: ð10:5Þ
998 C. Huang / Applied Mathematical Modelling 33 (2009) 978–998

The ABD model then reduces to (9.11) (without tildes) and Eqs. (10.2)–(10.5).
We shall use an explicit iteration scheme to solve the system of integral equations as follows. We start with
P ðx; 0Þ ¼ P 0 ðxÞ defined in (10.1). The force term F ðx; tÞ is evaluated by (9.11) in terms of P ðx; tÞ. We then
advance time by the Euler forward scheme to compute H ðx; y; t þ DtÞ from the integral equation (10.4), and
P ðx; t þ DtÞ from (10.2). All the spatial integrations are evaluated numerically using the 5 points Gauss quad-
rature in [-1, 1]. The viscosity ll0 is finally calculated by (9.18). The results with the time step Dt ¼ 0:01 for
/0 ¼ 30% and 38% are presented in Figs. 1 and 3. Figs. 2 and 4 give similar results when the Hamaker con-
stant is chosen as 50k B T (instead of 24k B T ). The four graphs show the shear thinning phenomenon in the range
of Pechlet number 101 < Pe < 10: the viscosity first decreases rapidly and then decreases more gradually until
it levels off. The graphs are similar to those in [5, p. 166] by the particle method.

11. Conclusions

Within the microscopic range where Brownian dynamics is valid, colloidal dispersions can be described by
the ABD model presented in Section 5. This continuous model has a unique solution for all time. The model
reduces to the system of integral equation (10.2)–(10.5) where F is defined by (9.11) (without tildes). The com-
putational time for solving this system could be much less than required for the particle method, especially
when particles are concentrated in a compact subdomain. A numerical example obtained by the ABD scheme
exhibit the same shear thinning phenomenon as computed by the particle method (as in [1,2, Chapter 15, 3,4]).

References

[1] W.B. Russel, D.A. Saville, W.R. Schowalter, Colloidal Dispersions, Cambridge University Press, Cambridge, England, 1989.
[2] H.A. Barnes, M.F. Edwards, L.V. Woodcock, Applications of computer simulations to dense suspension rheology, Chem Eng. Sci. 42
(1987) 591–608.
[3] D.M. Heyes, Rheology of molecular liquids and concentrated suspensions by microscopic dynamical simulations, J. Non-Newtonian
Fluid Mech. 22 (1988) 47–85.
[4] D.M. Heyes, J.R. Mclrose, Brownian dynamics simulations of models of hard sphere suspensions, J. Non-Newtonian Fluid Mech. 46
(1993) 1–28.
[5] A. Friedman, Mathematics in Industrial Problems, Part 5, IMA, vol. 49, Springer-Verlag, 1992.
[6] A. Friedman, C. Huang, Averaged motion of charged particles under their self-induced electric field, Indiana Univ. Math. J. 43 (1994)
1167–1225.
[7] A. Friedman, C. Huang, Averaged motion of charged particles in a curved strip, SIAM J. Math. Anal. 57 (6) (1997) 1557–1587.
[8] M. Weber, The fundamental solution of a degenerate partial differential equation of parabolic type, Trans. AMS Math. Soc. 71 (1951)
24–37.
[9] A. Friedman, Stochastic Differential Equation and Applications, vol. 1, Academic Press, New York, 1975.
[10] W.B. Russel, Dynamics of concentrated colloidal dispersions. Statistical mechanical approach, in: M.C. Roco (Ed.), Particulate Two-
phase Flow, Butterworths, 1991.

You might also like