Annual Reviews

www.annualreviews.org/aronline

Anrt Rev. Fluid Mectt 1981. 13:425-55

Copyright 1981 by AnnualReviewsInc. All rights reserved

BROWNIAN MOTION OF 8183

SMALL PARTICLES SUSPENDED
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

IN LIQUIDS
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

W. B. Russel
Departmentof ChemicalEngineering, PrincetonUniversity, Princeton,
NewJersey 08544

INTRODUCTION
Intriguing randommotions of small particles suspendedin liquids were
first reported by Robert Brown,a biologist, in 1828. Controversy
concerning the origin of the motion persisted, however, for many
decadesstimulating a series of experimentsby nineteenth-centuryscien-
tists including Pen-in (1910) and attracting notable theorists such
Einstein, Smoluchowski,Langevin, and Lorentz. This early work, re-
viewedby Nelson (1967), eventually confirmedthe molecular nature
matter by relating the particle motionto the thermal fluctuations of
moleculesin the fluid.
This review will focus on morerecent workconcerningsuspensions of
rigid particles small enoughto be affected by Brownianmotion,but still
large enoughfor the fluid to be treated as a continuum.These small
dimensions, ~1 nm-10/~m, render inertia negligible for steady mo-
tions, although acceleration must be retained in somecases because of
the intrinsic transience of the movement.Thedynamicsof small mole-
cules and polymerswill not be discussed, except for the application of
hydrodynamictheories to the former and the behavior of compact
macromoleculessuch as globular proteins. Nor will long-range nonhy-
drodynamicinteractions, such as electrostatic or dispersion forces, be
treated explicitly since their effects are morequantitative than qualita-
tivb.
Even within these boundsthere remain manyinteresting phenomena.
Three general topics are examined in detail: Brownian motion of
isolated particles, the effect of particle-particle andparticle.wall interac-
tions, and the role of Brownianmotionin the theology of suspensions.
 Annual Reviews
www.annualreviews.org/aronline

426 RUSSEL

Controversies surround several of the topics. Most arise from the

molecular origin of Brownianmotion whichrequires that the forces be
appendedto the usual continuumdescriptions of fluid-particle systems.
Of the several existing approachesweopt for the simplest while remain-
ing consistent with morefundamentalmoleculartheories. This reviewis
intended to complementthat of Batchelor (1976b).

BROWNIAN MOTION OF ISOLATED PARTICLES

Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Langevin Descriptions of the Dynamics

The Brownianmotionof particles suspendedin liquids can be described
from an equation of motionbalancing a randomfluctuating force acting
on the particle with its inertia andtl~te fluid resistance. Theapproachof
Langevin(1908), illustrated belowfor an isolated sphere, leads to the
velocity-autocorrelation function characterizing the dynamicsof a heavy
sphere and the translational diffusion coefficient describing the net
displacement. Recent work, noting the limitation of the conventional
approachto heavy particles (Pomeau& Rrsibois 1975), has obtained the
correct velocity autocorrelation for neutrally buoyant particles and
verified that the diffusivity is unaffected. With this backgroundthe
generalization to anisotropic particles and interacting spheres becomes
straightforward.
The Langevin equation for a sphere of mass m and radius a with
center at x and velocity u,
du
(1)
m--~- = -6~r#au+F(t),
includes the assumptionthat the forces that the fluid moleculesexert on
the particle can be separated into rapid fluctuations F(t), with time
scales characteristic of molecular motion(~10-13 s for water), and
muchslower viscous response characterized by the pseudosteadyStokes
drag. The Brownianforces are random
<F(t)> (2)
anduncorrelatedon the tir~e scales of particle motion,i.e.
(V(t)F(t + ~)) = F/~(,). (3)
Theassumptionthat at thermal equilibrium kinetic energyis partitioned
equally amongthe three translational modesof the particle

m(u(t)u(t)) (4)

serves to determineF.
 Annual Reviews
www.annualreviews.org/aronline

BROVCNIANMOTION 427

Successive integrations of (1) with x(0)=x0 and dx/dt(-oo)ffi 0 lead

to x(t) from whichthe velocity autocorrelation function
R(~)-- (u(t)u(t +
_ 12~_~exp(_ 6~r#a~.)rn
(5)

follows. From(4)
1:--- 12~rl~akTl, (6)
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

comprisinga fluctuation-dissipation theoremrelating the strengths of

the randomBrownianfluctuations to the steady frictional forces and
thereby reflecting their common origin in the interactions betweenthe
particles and the solvent molecules.
The autocorrelation function indicates that the energyimparted to a
particle by each thermal impulse decays exponentially on the viscous
time scale m/6~rl~a(,~10-gs for a neutrally buoyant 0.1/~msphere in
water). Thesubsequentrandomforcing has no coherent effect.
A parallel macroscopicanalysis of the diffusion process in terms of
Ficks law showsthat
| d
lim ~ -d-~((x-x0)(x-x0) ) = R(,)d,=DoI. (7)
t--~oO f0
With(5) this providesthe Stokes-Einsteinrelation Do-- kT/6~rtta for the
diffusion coefficient of an isolated sphere.
The assumptionsimplicit in the Langevinequation have been estab-
lished by Mazur & Oppenheim (1970) and Albers, Deutch
Oppenheim (1971). Theybegan with the Liouville equation for both the
particles and the fluid moleculesand integrated over the positions and
momenta of the latter to obtain the equationof motionfor the particle.
Theconventionalform (1) results whenall relaxation times associated
with the fluid are short comparedwith those of the particle. According
to their papers this means m/mf>>lwith mf the mass of a fluid
molecule. In fact, the vorticity of the continuumfluid must diffuse
faster than the particle loses inertia as well, i.e.
a 2 m
--<<
u 6er/~a
indicating that the particle also must be muchdenser than the fluid.
Thus, as noted by Hauge& Martin-Lrf (1973), the appropriate limit
actually rn/rnr-~oowith a fixed ratio of particle to molecularsize. The
normalLangevinformulation, therefore, does not apply to the neutrally
buoyantparticles of frequent interest.
 Annual Reviews
www.annualreviews.org/aronline

~,28 RU$SEL

This limitation waspointed out by Lorentz in 1911(Hauge& Martin-

L6f 1973) but attracted little attention until numericalsimulations of
molecular motions in liquids by Rahrnan(1964) and Aider & Wainwright
(1967) producedvelocity correlations with a long tail decayingas -a/2,
t
rather than exponentially as predicted by (5). Alder & Wainwright
(1970) recognized the hydrodynamic origin of the effect and solved the
transient Navier-Stokesequations numericallyto predict the asymptotic
decay.
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

Morerecently three separate analyses havepredicted correctly the full

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

velocity correlation as well as the diffusion coefficient for comparable

partite and fluid densities. Assumptions(2) and (3), concerning
Brownianforces, remain valid but the evaluation of F and the friction
law requires somecare. Whena2/v--m/6~rl~a, vorticity generated by
the suddenacceleration, due to the. Brownianimpulse, diffuses awayon
the sametime scale that the particle decelerates. Asa result, the fluid
inertia remains important, preventing the viscous force from reaching
the pseudosteadylimit before the motiondies away. Linearity is pre-
served, however, because the velocities remain small enoughto render
negligible the convective terms in the equations of motion. Hence
classical solutions to the unsteady Stokes equations suffice for the
time-dependent drag.
The approaches differ in the way they determine F. Hauge& Martin-
L6f (1973) substituted the unsteady drag for an incompressiblefluid
equation (1) and calculated F from the theory of Brownianfluctuations
in a fluid continuum. Hinch (1975) espoused a purely continuum
approachbut applied the Langevinequation to the fluid as well as the
particles with the drag determinedby the instantaneousstress field. This
methodaccounts directly for thermal fluctuations in the fluid, thereby
permitting g to be calculated from the equipartition of kinetic energy
for both fluid and panicles. Thevelocity autocorrelation for a neutrally
buoyantparticle resulting from both analyses
2 kT
R()--- -~ --~-Im{aeerfc aV~,r (8)

with

a~-V~ +i, "r= 3vt

24a
can be interpreted as the responseof an initially stationary sphere to an
impulse of magnitudekT at ~=0. The long-time limit

lira g(~-)--- ~1 kTz_3/~ (9)

~-*~ 8 V-~-~m
 Annual Reviews
www.annualreviews.org/aronline

BROW~-L~ MOTION 429

agrees with the molecular-dynamics simulations. The short-time limit

2 kT
lim R(r) -- - (10)
--*0 3 m

illustrates that the initially rapid acceleration causes the impulse to be

1
distributed between the particle and the fluid added mass ~ m.
One mechanismby which the initial impulse, received solely by the
particle, can be redistributed over the added mass was defined by
Zwanzig & Bixon (1975) following earlier work by Chow& Hermans
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

(1973). For a slightly compressible fluid, i.e. a/c<<a2/~,, the velocity

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

correlation falls from the true initial value kT/m to (I0)

2 kT ~/~ ct
R(t) -~--~ 1 + e-/2c/") cos 2 a V~sm-~-- . (11)

This short initial transient, due to acoustic dampingwith no effect of

viscosity, is followed by the slower viscous decay predicted by (8). The
complete time dependenceof the velocity correlation plotted in Figure 1
differs significantly from the exponential form predicted with the pseu-
dosteady force law. Note also that only for ac/v>>lO2, e.g. spheres
>0.1# in water, do the acoustic and viscous time scales separate
completely.
Substitution of the correct R(t) for neutrally buoyant particles into
(7) generates the same diffusivity as does the exponential form valid
only for heavy particles. Indeed, as recognized by the above authors and

"m
I0 I0-~ I0" I lO

Figure I me velocity autocorre]ation function for an isolated sphere: ---- with

pseudosteady friction law, - incompressible with time-dependent friction ......
compressible 2.
but invisid with a/e-- 10
 Annual Reviews
www.annualreviews.org/aronline

430 RUSSEL

~
hO

~. ~ ~"~~

0.6

Do 0.4
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Fibre 2 ~e FouEer tr~fo~ of ~ vc]ocRy aut~lation f~ction for ~ isolat~

sphe~: ..... ~ pseudoste~dy f~cEon,~-.~mpressib]~ ~ t~de~de~t f~cEoo.

discussedby Batchdor(]976b), a different result ~ou]dbe ~fficuh

reconcile. ~e physical explanation stems from the relation

~R(t) at = 2u~(0) (12/

where ~(~) is the Fourier transfo~ of the velocity co.elation. Ap-

parently the low-frequency components, representing ~most steady
motion, provide the largest net displacements and do~nate ~e diffu-
sion process. ~e two autoeo~elation functions necessarily share a
commonzero-frequency li~t (Figure 2) and therefore a~ on
cliff.ion coefficient; the remainderof the spectra, while interesting, is
superfluous.
~ese results indicate clearly that the diffusion coefficients for bodies
of arbitra~ shape, obtained directly from their pseudosteadyhydrody-
n~e mobflifies (Brenner 1974), predict accurately the increase
m~n-squaredispla~ment and rotation for t>>12/v(1~ the characteris-
tic dimension). Given the difficulty in meas~ngsmall spati~ fluct~-
fions, at least by li~t scattering, at t~e scales less than 10-* s (Berne
& Peeora 1976) t~s ~fo~ation fully charaete~es the cu~ently ob-
seeable dynamics.

Diff~ion in Biological Membranes

~11 membranesare generally thou~t to be lipid bilayers of t~c~ess
~2 nm which, thou~ fl~d, are 102-10a times more viscous ~an the
adjacent aqueousphases. Interest in their structure and function has
stimulated measurementsof rotational and translafion~ diff~ion for
 Annual Reviews
www.annualreviews.org/aronline

aaow~n~ MO~O~ 431

both the lipids themselvesand large proteins intercalated in the bilayer

(Edidin 1974). The data, wheninterpreted correctly, provide informa-
tion on the membranefluidity significant to the understanding of
molecular clustering, rearrangement, and other aspects of biological
organization. The intrinsic two-dimensionalityof the process, however,
modifies the hydrodynamicmobilities and complicates the interpreta-
tion in somesubtle and interesting waysdescribed below.
Saffman & Delbriick (1975) model the diffusant as a cylinder
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

radius a spanning a planar membraneof thickness h. Since the mem-

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

brane viscosity # greatly exceeds that of the surrounding fluidl the

molecule can protrude from the membranewithout significant conse-
quences. The problem presents some intriguing physics due to the
two-dimensionalnature of the motion within the membrane.The pseu-
dosteadyrotational mobility suffices to determinethe rotational diffu-
sion coefficient
kT
D r-- ~ (13)
4 rtta2h

but the translational mobility for Stokes flow in an infinite membraneis

infinite. Theauthors proposeseveral possible resolutions to the paradox
based on additional constraints or physical processes: a finite mem-
brane size, couplingwith adjacent fluids of finite viscosity, or unsteady
inertial effects within the membrane.They conclude that the correct
result dependson the length scales characterizingthe individual effects,
with the second probably controlling for biological membranes.
A finite membrane forces the fluid velocity due to steady translation
to be zero at the boundaryrather than diverging logarithmically. For a
characteristic membrane radius R>>athis leads to

kT
Do= ln", (14)
4~r/th a

introducing a weakbut troublesome dependenceon the membranesize

plus a smaller geometricaleffect.
Thesingular nature of the planar problemindicates that a small, but
finite, viscosity in the adjacent fluids could generate a critical three-
dimensionality. Saffman(1976) treated the surrounding phases as New-
tonian with viscosity # and the membrane as having viscosity # in the
plane and an infinite viscosity in the normaldirection becauseof the
highly oriented molecularstructure of thin membranes. WhenI.ta/t~h<<1
the adjacent fluids only influence the membrane flow in the outer region
 Annual Reviews
www.annualreviews.org/aronline

432 Rt~SS~L

r~h#/l~. The resulting three-dimensionalmotionremovesthe logarith-

mic singularity, leavinga finite translational diffusivity
kT /in ~_~h_0.5771).
(15)

For large membranes boundedby truly inviscid fluids (perhaps gases)

the mobility for steady translation, of course, becomesfinite at finite
Reynoldsnumber,but the resulting velocity-dependent mobility
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

In ~ - 0.077 l
4~r#h
invalidates the linear analysis leading to (7). Saffman(1976) avoided
this difficulty by resorting to a Langevinanalysis with unsteady, but
linear, friction, muchlike the developmentof Zwanzig& Bixon0975).
The diffusivity, deducedfrom the indefinite form of (7), depends
time as
kT [ 4/tt _1.5771)

and diverges as t--~ as expected from the steady solution.

For diffusion processes in biological membranes Saffman& Delbriick
(1975)
R
,,,104
concludedthat
t~t ~10~5, t~h ,,,102
a 2
oa #it
so that the smallest diffusivity (15) controls. The theory has proven
useful for interpreting experimental data (e.g. Wuet al 1979) but few
sets are sufficiently completeto test it. As cited in the original paper,
Cone(1972) and Poo& Cone(1974) measuredboth rotational (,--,5
s -~) and translational (~4x 10-~3 m2/s) diffusivities of rhodopsin
(a~2x 10.9 -~
m) in disk membranesfrom the frogs retina (/~,--~10
Ns/m #~10 -3 Ns/m~, h~2 x 10"9 m). The observed ratio
~,
2D--~0.2 - 2.0 x 10- ~7 m

brackets the value 1.6X10-~7 m~ predicted by (13) and (15) although

the individual values deviate significantly, perhapsbecauseof the tmcer-
tainty in membrane viscosity. The correspondingratio for the equiva-
lent spherein an infinite fluid,

~a
~,~5 x 10-~7 ~,
m
 Annual Reviews
www.annualreviews.org/aronline

BROWNIA.N MOTION 433

showsthis theory, still usedat times to interpret data, to be numerically

inaccurate as well as physically inappropriate.
In addition to the effect on membrane-bound species, diffusion
within biological membranesalso maybe significant within a larger
context. Evenwith the lower diffusivity implied by /-t/g-~,10 3, the
partitioning of a reactant into a membranecan greatly enhance the
reaction rate due to the lower dimensionality of the diffusion process
therein (Adam& Delbriick 1968) and the molecular orientation im-
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

posed by the planar geometry(Poo & Cone1974).

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Application of Hydrodynamic Theories to Molecular Motion

The study of molecular motion in liquids is a venerable field amply
reviewed in the chemical physics literature (e.g. Pomeau& Rrsibois
1975, Hynes1977). The topic arises here becauseof the recurrent use,
and frequent success, of hydrodynamicmodelsin explaining Brownian
motions of molecules ranging from argon in the condensed state to
macromoleculesin solution. Recently molecular dynamicssimulations
mentionedabove and data obtained with newexperimental techniques,
such as depolarized light scattering (Bauer et al 1976), have demon-
strated both the potential and someof the limitations of the approach,
thereby generating further theoretical and experimentalactivity. This
section will not delve into the detailed moleculartheories nowemerging
but will review briefly the hydrodynamic theories and illustrate their
comparisonwith experimental data.
Einsteins (1956)original predictions for the translational and rota-
tional diffusion coefficients apply to spherical moleculeswith a no-slip
boundarycondition at the solvent-moleculeinterface. This assumption
was immediately questioned with respect to events at the molecular
scale (Sutherland 1905) and a slip condition proposed instead. The
alteration increases the translational diffusivity only by a factor of 3/2
but reducesthe rotational friction coefficient to zero allowingthe sphere
to reorient inertially with
3 ( 5kT )t/2 (17)

Indeedsomespherical moleculesapproachthis limit (Baueret al 1974).

Rotation of nonspherical molecules, however,drives fluid motion even
with slip at the surface, but the friction coefficient remainsconsiderably
smaller than without slip, except for extremegeometries. Therotational
diffusion coefficient thus provides a sensitive indication of the ap-
propriate boundarycondition once the molecular geometryand size are
known.
 Annual Reviews
www.annualreviews.org/aronline

434 RUSSEL

The appropriate friction coefficients with slip were calculated by Hu

& Zwanzig (1974) for spheroids and by Youngren & Acrivos (1975)
ellipsoids. For a general ellipsoid the coefficients characterizing rotation
about each of the three principle axes are nonzero and unequal. With
axial symmetry, however, two become equal and the third is zero. Only
in the limits of a needle or a disk does the friction coefficient with slip
equal that without slip.
Diffusion coefficients for the individual rotational modes can be
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

extracted from reorientation times detected by depolarized light scatter-

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

ing and ~3C NMRaccording to

1
-- +,o (18)
(Berne & Pecora 1976). Time scales measured for a variety of systems
indeed vary linearly with the solvent viscosity as implied by 08) with an
intercept ~0 which correlates with the inertial time scale (Bauer et al
1974). These results support the validity of the hydrodynamic model
and have been compared with qlaan~litative predictions based on known
molecular volumes and shapes.
Data for small molecules, ranging, in molecular volume from chlorine
(2.6 i0 -3 nm3) to valeric acid (0.209 nm3), dissolved in low-viscosity
organic liquids which do not exhibit strong solute-solvent interactions
provides one limit. For these systems the no-slip theory predicts relaxa-
tion times two to ten times too long, but as illustrated in Figure 3 the
theory for spheroidal particles with slip provides generally acceptable
values. Two exceptions are the aromatics, benzene and nitrobenzene,
which have geometries a bit too complex to be approximated by a
spheroid. Youngren & Acrivos 0975) found that a more detailed
representation of benzene as six hemispheres (hydrogen atoms) attached
to an oblate spheroid (carbon ring) resulted in the much improved
prediction denoted by the asterisk in Figure 3.
Solutions with strong solute-solvent interactions, such as hydrogen
bonding, pose more difficult probI~ems. The results depend strongly on
the particular solvent with the no-slip boundary condition apparently
appropriate in some cases (Millar et al 1979, von Jena &Lessing 1979),
while others fall intermediate betweenthe two limits (Bauer et al 1974).
Ultimately, a satisfactory theory for the motion of small molecules
must blend a detailed statistical-mechanical description of the short-time
collisional dynamics with a hydrodynamic forraulation of the subse-
quent large-scale motions (Hynes 1977). Only then will the true nature
of these apparent transitions fron~L no-slip to slip behavior with decreas-
ing molecular size and decreasing intermolecular attraction be fully
understood.
 Annual Reviews
www.annualreviews.org/aronline

BROW~L~IOnON 435
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

0.3 0.4 0.5 0.6

MOLECULAR SIZE

Figure 3 Ratio of the measuredrotational diffusivity to that predicted for an ellipsoid

with perfect slip as a functionof molecularsize: o nitrobenzcneand benzene(Bauerct al
1974); bcnz~n(Youngrcn&Activos 1975); chlorine (Topalian t al 1979).

EFFECTS OF INTERACTIONS
Generalized Description of the Dynamics
The Browrfian motion of particles suspended at concentrations for
whichhydrodynamic interactions becomesignificant has attracted con-
siderable attention in recent years (e.g. Ermak& McCammon 1978,
Hess & Klein 1978). Twocomplications enter the Langevinformulation,
even in the low-frequency limit with pseudosteady hydrodynamics.
Clearly, the friction coefficients--now configuration dependent--
couple the motions of the interacting particles. Less obvious is the
coupling betweenthe fluctuating Brownianforces at separations on the
order of the particle size. Thegeneral analysis for Ninteracting spheres
in a volumeV sketched belowrepresents a generalization of several
existing treatments, illustrating the continuum-mechanics approachto
the problemand the differences fromthe single particle limit.
The coupled Langevin equations for N identical spheres without
external couples, written in matrix form, become
du
m ~ =-Z-u+F (19)

u=0at t= - oo
 Annual Reviews
www.annualreviews.org/aronline

436 RUSSEL

and x=xo at t=0. The matrices

u={ui}
F-- (F~};= 1 ..... N (20)
account for the velocities and the Brownianforces on all N particles
while the elementsof the generalizedfriction tensor
Z= (~,~.} i,j=l ..... N (21)
determinethe force on the ith particle due to the velocity of the jth.
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Eachcomponentdependson all the positions Xk(k----1 ..... N), thereby

couplingthe set of equations. Onlyfor well-separatedspheres, i.e.
j:/:i
lira ~ii= [[ 0 (22)
Ixk-xil-,oo 6~r#aI
j--i
k4~i
do the particles moveindependen, tly. The Brownian forces remain
random
(v)
and uncorrelated in time
(F(t)FT(t)) Fog(t- t ), (23)
while the kinetic energyimpartedto the particles is partitioned equally
amongthe translational modesas

m(uur ~ = ~ k TI. (24)

This system can be solved exactly as was the equation for a single
particle, provided the configuration {x/} does not changesignificantly
on the viscous time scale. As a result
Fof2kTZ
and
R($) = (u(t)ur(t + $)) 2rn } (25)
so that
D -- fon()d~- kTZ- (26)

whereZ- ~ = {~%}is the generalized mobility tensor. From(25) one can

readily verify that the changein the rdative position, %=[x,.-xj [, of
any two particles during the viscous relaxation time
<(Arij)2> /z
2(mkT)

remainssmall for all conditions of interest (Batchelor 1976a).

 Annual Reviews
www.annualreviews.org/aronline

~ROWNIS~ MOTtO~ 437

The generalized fluctuation-dissipation theorem in (25) reveals two

interesting facets of the Brownianforces in a coupled system. Since
(Fi( t )Fj( t) ) -- 2kT~ij(x~,)$( t-t) (27)
the forces acting on two interacting particles (i#j) are coupled and
their magnitudes (i=j) depend on the configuration (x~,}. Other deriva-
tions, perhaps more fundamental but also more involved, starting with
the Liouville equations for the particles and the molecules arrive at the
same conclusion (Lax 1966, Zwanzig 1969, Deuteh & Oppertheim 1971).
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

The elements of D comprise multicomponent diffusion coefficients

t ,s =kr,0,j(x,) (28)
indicating that the flux of a particular particle may be hindered by
hydrodynamic interactions, since Dli <kT/6,rl~aI, but also becomes
coupled to the diffusion of its neighbors since D,.s#0 for i#j.
The tracer or self-diffusion coefficient, characterizing the wandering
of a tagged particle in a uniform environment of untagged neighbors,
reflects the former effect. For untagged spheres at volume fractions
<<1 without long-range interaction potentials, the configurational aver-
age of (28) yields

(D,,) = ~k-~T (1 - 1.83,)1 (29)

o~r/~a
(Batchelor 1976a, Anderson & Reed 1976a). Figure 4 illustrates the
uncertainties which plague comparisons of this result with measure-
ments for small macromolecules such as bovine serum Mbumin. The
 Annual Reviews
www.annualreviews.org/aronline

438 RtISSEL

effective hydrodynamicradius must be extracted from the data, which

introduces considerable uncertainty into the slope for the range of a
,-,3.28-3.75 nmfound in the literature. In addition, weaklong-range
forces, which maypersist even at the isoelectric point or high salt
concentrations, can alter the slope significantly (Anderson & Reed
1976a). Nonetheless,the results provide reasonablecorroborationof the
theory.

Equilibrium Analysis of Gradient Diffusion

Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Diffusion coefficients, but not the detailed dynamicsof colloidal par-

ticles, can be derived from the analysis of an equilibrium system in
whichan external potential acts on the particles to create a nonuniform
concentration. In his classic paper Einstein obtainedthe diffusivity for
an isolated sphere by balancing the flux due to the external potential
against diffusion downthe concentration gradient and recognizing the
gradient in chemicalpotential as the appropriate driving force for the
latter. Recently Batchelor (1976a) extended Einsteins thermodynamic
argument to finite concentrations; his detailed calculation for pair
interactions betweenhard spheres corrects the diffusion coefficient for
hydrodynamic and potential interactions in the dilute limit. Thealterna-
tive statistical-mechanical approachsketched below(Russel et al 1980)
reaches the same conclusion without invoking thermodynamicargu-
ments about the effective driving force for diffusion. The resulting
generalized Stokes-Einsteinrelation, valid at arbitrary concentrations,
directly relates the gradient diffusion coefficient to the sedimentation
coefficient and the osmoticcompressibility.
Consider a closed system of N spheres suspended in a Newtonian
liquid and subjected to the external potential U(x). At equilibrium the
sedimentationand diffusion fluxes must balance locally as

- (x) v V(x)O. (30)

In addition, the Boltzmanndistribution

1 1 V(x, .....
xN)+ _~ U(x;) (31)

determinesthe probability of finding the N particles in configuration

(Xk}in terms of the interparticle potential V andthe external potential.
Q normalizesthe distribution so that

fP~vd3x~.., d3xN--N!,
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN MOTION 439

while

/~(Xl) =
(U-l)! fP~,d3x2..,a3x. (32)

is the local concentration.Thesedimentationcoefficient K(g,) in princi-

ple can be calculated from PNand creeping flow solutions for N
interacting spheres; exact O(q,) corrections to Stokes law are available
for hard-sphere repulsions (Batchelor 1972) and longer-rangepotentials
(Anderson & Reed 1976b).
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

The diffusivity D follows from (30) once Vn(x) has been related
V U. For slowly varying potentials one can showthat

Vn(x,)= n(Xl)kT VU(x,)(1 +4~rn(x,)f0~r2[g(r)-I ]dr). (33)

where
l fexp(__~)dx3...d3x
~
1 V
g(r)=.2(Xl)(u-2)--~.
__ (34)

is the radial distribution function without the external potential. The

bracketedterm in (33) can be identified as the osmoticcompressibility
kTdn/dv (Reed & Gubbins 1973) leaving

D= K(O) (35)
6~l~atin"
This result is rigorous and exact for a slowlyvaryingexternal potential,
i.e. aVU/U<< I. The independent roles of hydrodynamicand thermody-
namic forces in the diffusion process are best illustrated with the
rigorous results available in the dilute limit. For hard-sphererepulsions,
i.e.
0 r12 > 23,
V= (36)
oo r~<2a.

Batchelor(1976a) notes that

dn--kT(1+S)
and
r()= 1 - 6.55, (37)
D
so that ~ -- 1.45~.
 Annual Reviews
www.annualreviews.org/aronline

440 RUSSEL

The thermodynamicenhancementrepresents an osmotically driven

expansion into a region of lower pressure and slightly overcompensates
for the hydrodynamicretardation of the sedimentation process. Both
effects dependstrongly on the interparticle potential; long range repul-
sions increase the osmoticcompressibility but decrease the sedimenta~
tion coefficient (Reed& Anderson1980). Thenet effect can be dramatic
increases in the diffusivity (Andersonet al 1978). Thephysical situation,
therefore, differs markedlyfrom se, lf-diffusion in whichneighboring
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

spheres merelyprovide a passive resistance to motion.

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Thehard-spheretheory has proven, controversial for two reasons. Part

of the confusion arises from the published predictions for the O(~)
coefficient in DID o ranging from -:2.6 to + 8.0 (Pyun & Fixman1964,
Altenberger & Deutch 1973, Phillie, s 1973, Anderson& Reed 1976a,
Harris 1976). Recently Felderhof (1978) and Wills (1979) independently
confirmed (37) and, more important, discussed at length the approxi-
mate hydrodynamicsand, in somecases, incomplete physics responsible
for the other values.
Attempts to verify (37) experimentally have been confounded
residual effects of long-rangeinterparticle potentials with small macro-
molecules such as bovine serum albumin (Fair & Jamieson 1980) and
the difficulty in interpreting dynamiclight-scattering data for large
particles.
The data of Newman et al (1974) appears to be free of both problems.
Their aqueous solutions of fd bacteriophage DNAwere quite monodis-
perse with a molecular weight of 1.86_ 0.06 106 and a hydrodynamic

7.5

I
0
~
(~QI~CENTRATION
) t k~,m
Figure 5 ~n~n~afion de~nden~ of ~e ~a~ent-~f~ion ~ffi~ent for fd
bacteriophage DNA ~ water: ~t ~tte~ng ~m of Ne~ et ~ (1974); Eq~fion
(3D.
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN
MOTION 441

T~bleI O()coefficients (fromNewman

et al 1974)

Hardsphere
theory Experiment
K(sedimentation) - 6.55 - 6.7_+0.8
I d~r(osmoticcompressibility)
8.0 7.6_ 3.9
kT dn
D 1.45 1.2_+
0.4
D-~(diffusion)
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

radius of 31.6_ + 0.6 nm. They measured independently the sedimenta-

tion and diffusion coefficients and the osmotic pressure as functions of
concentration. The dynamic light-scattering data for D in Figure 5 was
independent of scattering angle since the wavelength of light signifi-
cantly exceeded the molecular size. As illustrated in Table 1 the ob-
served concentration dependence of all three quantities fell within
experimental error of the hard-sphere predictions. The osmotic pressure
establishes that long-range interparticle forces are negligible and the
sedimentation coefficient verifies the hydrodynamictheory, leaving the
diffusion data as a clear confirmation of the generalized Stokes-Einstein
equation, at least for the dilute limit.

Dynamic Light Scattering

The development of dynamic light scattering as a rapid and accurate
technique for measuring translational and rotational diffusion coeffi-
cients of colloidal particles and dissolved macromolecules has greatly
advanced the study of Brownian motion. In addition to making basic
studies of the phenomenaeasier, the technique has proven quite valua-
ble for the characterization of biological and synthetic macromolecules
with respect to molecular size and shape (e.g. Bloomfield 1977,
McDonnell & Jamieson 1977).
Since suspended particles or macromolecules generally scatter far
more light than does a low-molecular-weight fluid, the scattering inten-
sity directly reflects their size, shape, and dynamical properties. For
example, the spectrum of light scattered from a volume containing N
identical rigid spheres of mass m (Berne & Pecora 1976)

I(k, ~)=Nm2p(k)RefoG(k, t)e-i~tdt (38)

depends on the single-particle scattering function P(k) and the co-
herent-structure factor

~(k, t)= ~--(,,*(k, t),,*(k,O)).

 Annual Reviews
www.annualreviews.org/aronline

442 RUSSEL

Here

(40)
is the spatial transformof the fluctuating numberdensity and
k = the wavevector of the scattered light,
4~r. 0
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

h = wavelengthof the light in the medium,

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

0 = scattering anglerelative to the incident beam.

Conventionaltotal-intensity light scattering measures

f I(k, ~)do~ = m~P(k)(n*2(k,O)), (41)

thereby detecting for a knownconcentration only the particle size. The

full spectrum, on the other hand, also contains information on the
dynamicsof the concentration fluctuations. Several books (Cummins
Pike 1974, Chu1974) describe the experimentaltechniques. This section
will focus on the theoretical problemof relating n*(k, t) to the thermo-
dynamicaland mechanicalproperties of the suspension.
The Fourier transform of the usual conservation equation
t)
0t ---ik.J*(k, t) (42)
suffices to determinen* oncethe transformof the flux J* is specified.
For noninteracting spheres Ficks law yields
J*= -ikDon* (43)
leading to
-k2Do
t)=e (44)
and

I(k, ~) =Nm2P(k) Dk2 (45)

~2 + D~k4
Theformer indicates that concentration fluctuations with length scale
k-~ disappear on the diffusion time scale (k~Do)- ~. Optical mixingor
beating techniques whichdetect decay times of ,--,10 -6 S and slower,
therefore, can measurediffusion coefficients for particles larger than a
few nanometers.
At finite concentrations, however,(44) fails becauseof interactions
betweenparticles. As noted in the previoussection, the diffusion coeffi-
 Annual Reviews
www.annualreviews.org/aronline

~Ro~ MORON 443

cient becomesconcentration dependent, a moderate effect for hard-

sphere interactions but a dramatic one for strong electrostatic repul-
sions. In addition, experimentswith large particles (Colbyet al 1975,
Fijnaut et al 1978)or long-rangeelectrostatic effects (Schaefer &Berne
1974, Brownet al 1975) reveal decay times for the correlation function
that deviate from the expectedk-2 dependence.Theeffective diffusion
coefficient depends,therefore, on the wavelengthof the fluctuation as
well as the meanconcentration, considerably complicatingthe interpre-
tation.
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Suchobservations stimulated several attempts to generalize the theory

to include interactions (Altenberger & Deutch 1973, Harris 1976,
Ackerson1976, 1978, Felderhof 1978, Wills 1979, Altenberger 1979). As
with the theories mentionedin the previous section the results vary
considerably, in part due to different levels of approximationfor the
hydrodynamics.As a result opinions differ about which coefficient
(gradient diffusion or self-diffusion) is measuredby dynamiclight
scattering. Altenberger (1979) approaches the correct answer via
complexroute, apparently failing only in the detailed hydrodynamics,
and also discusses the shortcomingsof the previous efforts. Our much
simpler approach outlined below (Russel & Glendinning1980) follows
the conventional treatment of the light scattering based on (38)-(40)
and employs the exact two-sphere hydrodynamicsto predict the con-
centration and wave-numberdependence of the effective diffusion
coefficient for hard spheresin the dilute limit.
The transformed flux ,I* in (42) arises from diffusion downcon-
centration gradients with spatial periodicity k. As for the ak<<1 limit
discussedin the previoussection, the effective diffusivity can be derived
for an equilibrium system, rather than the more complex non-
equilibrium one implied by (42). This requires the construction of
spatially periodic potential U(x)-U*(k)e i~ to balance the diffusion
flux and maintain steady concentration variations of the same wave-
length. Then

J* = - ikD(k, (p)n*(k) ~noikU*(k

) (46)

provided the deviations from the meannumberdensity no remain small.

NowK(k, ~) must be calculated and n*(k) related to U*(k).
For long wavelengthsthe sedimentationcoefficient must asymptoteto
(37). Then all particles have the same velocity, and hydrodynamic
interactions produce three distinct O(~) effects on the motion of
individual particle: a passive retardation by force-free neighbors
(- 1.83~,), a backflowand local pressure gradient from the net motion
 Annual Reviews
www.annualreviews.org/aronline

444 RUSSEL

of all the particles (-5.0e~), and enhancementdue to nearby particles

moving in the same direction (+0.28q,). As ak becomes O(1) the
large-scale motionis suppressedby the periodicity, and the velocities of
nearbyparticles shift out of phase..Asa result, the last two contribu-
tions decayto zero in an oscillatory fashionleaving, for ak>>1, no effect
of the motion of other particles but only the passive resistance of
neighbors(- 1.835) as in the self-diffusion process.
The gradient in the meannumberdensity can be evaluated from a
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

Boltzmann distribution reflecting the small, spatially varying potential.

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

The resulting proportionality constant between n*(k) and -U*(k)/

(nokT)is the static-structure factor
S(k)= 1 +4~rn0f (g-
oo 1)-~--r ,"., .
sin kr
(47)

whichfor k = 0 reduces to the osmoticcompressibility. For hard spheres

3~ (2cos2ak__Sina2kak).
, (48)
S(k)-~l+(ak~
as with the sedimentation coefficient the O() term decays from -8~,
for ak<<1 to zero for ak>>1 in an oscillatory fashion. Thefinal result for
the effective diffusion coefficient from(46) and(48)
kT
D(k, q,) = ~ { 1 + Kt~q~
} + O(ep~") (49)

whereKo is the function of ak plotted in Figure 6.

IO-z oI0 210

iO-~ 101

Figure 6 The concentration dependence of the effective diffusion coefficient as a

function of the wavenumberof the gradients.,
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN MOTION 445

Finally, the conservationequation (42) can be integrated to determine

the correlation function

G(k, t) -= -g23(k*)t
S(k)e (50)
and the spectrum

I(k, ~0) =Nm2P(IOS(k)~2D(~: ~) (51)

o~2 +k4D:(k,)
"

Clearly, at finite concentrations dynamiclight scattering only detects

Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

gradient diffusion whenak~ 0.2 while for ak>~4the process closely

resemblesstir-diffusion. For the visible spectrumand scattering angles
between 45 and 155 these translate into a~<20nmand a~>0.20/~m,
respectively.
While the theory seems to answer some troublesome questions, no
definitive data are available for testing the predictions. Figure 7 shows
data from Fijnaut et al (1978) for polymethylmethacrylate spheres with
a=0.12 #m in benzene. The comparison is only qualitative because
mostof the volumefractions lie far beyondthe range of validity of (49);
in addition, the highly swollen latices maydeviate significantly from
hard-spherebehavior. Nonetheless, the trends are reasonably consistent
with the predictions.
The hydrodynamicportion of the theory is complete but the light-
scattering analysis remainsless certain. Berne&Pecora(1976)carefully

Figure 7 Comparisonof theory for D(k, ) with the data of Fijnaut et al (1978):
0.06, (o) 0.11, (*) 0.25, ([]) 0.33, (/x)
 Annual Reviews
www.annualreviews.org/aronline

446 RUSSEL

note that conventionaltreatments of dynamicscattering pertain only to

particles small relative to the wavelengthof light and neglect multiple
scattering. For independentparticles the latter is certainly negligible
and the single-particle scattering function P(k), available from Mie
theory (Kerker 1969), accounts for the former effect. But when#,NO(l)
and ak~O(1),exactly the region of :interest, multiple scattering maybe
significant and unaccountedfor within the current theory.

Hindered Diffusion in Pores

Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Several transport processes of biological or industrial interest involve

diffusion and convection of small particles or macromoleculesthrough
packed beds, small capillary tubes, or membranescontaining small
pores. For steady-state transport through porous membranes,either
physiologicalor artificial, the key questions center on the variation in
the rejection characteristics andflux with the flowrate, particle size, and
membrane structure (Bean 1972). For intrinsically transient separation
techniques such as gel permeation and hydrodynamicchromatography
(Silebi & McHugh 1978), the performance depends on the propagation
time and dispersion of a pulse. Oneidealized model, applicable in some
degree to both processes, considers the movement of spheres of radius a
through a cylindrical pore of radius r 0 (Anderson & Quinn 1974,
Brenner& Gaydos1977). In fact, thin sheets of track-etched mica (Bean
1972) with very uniform pores provide model membraneshaving almost
exactly this geometry.
For dilute concentrations within the pores, i.e. <2a/3ro (Anderson
& Quinn1974), the conservation equation for the particles is (Zwanzig
1969, Brenner & Gaydos1977, Felderhof 1978)

~--~- +V.U(x)PfV.D(x). VP+ VV .

The equation holds both within ~mdwithout the pore with P, the
single-particle probability density or the local numberdensity, becom-
ing spatially uniformin the surroundingfluid. Withina pore, hydrody-
namicinteractions with the wall retard the motion, reducing the sphere
velocity U(x) belowthe local fluid velocity and generating the aniso-
tropic diffusivity tensor Dwith elementsDij(x)<~D o. Potential interac-
tions with the pore wall producethe force - V pr on the particle.
Therole of entrance and exit effects can be illustrated by scaling the
radial position on r 0 and the axial position on the membrane thickness L
within the pore so that

( r0] ~ Peu~z--\~0! 7~ rde-~/kr (Pe~/k~) )

dll 2~z
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN MOTION z~47

but on r o in the vicinity of the pore mouthsso that

OP
0-~- + Pe*V. uP = V. de- ~/~r. VpeV/kr. (54)

The pore and entrance Peclet numbersare related by Pe--UbL/Do--

(L/ro)Pe* with ub the average fluid velocity in the pore. Within the
pore the dimensionlessparticle velocity u and diffusivities, d~_anddll,
and the interaction potential dependonly on r, but outside u, t, and V
vary in two dimensions.
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

For long pores P exp(V/kT)=n(z) at steady state within the pore.

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

The axial flux then consists of an O(D o exp[- Vo/kT]An/L ) contribu-

tion from diffusion and O(ub exp[- Vo/kT]An)from convection. Here
V0 denotes the characteristic potential within the pore and An the
concentration difference throughthe pore.
Theconcentrations within the entrance and exit regions are invaria-
bly assumedto be in equilibrium with the bulk, i.e. Pexp(V/kT)=n*--
constant (Anderson& Quinn1974). Accordingto (54) this requires
convection into the pore be weak,leaving diffusion to accommodate the
axial flux through the pore. Equatingthe internal and external fluxes
determines
An* r0 1 + Pe
An L 1 + Pe*
establishing that n*~ constant and equilibrium is approachedif roll
<<1 alongwithPe*<<1.
Withclosely spaced pores, rather than isolated ones as considered
above, the external transport process becomesone-dimensional and
controlled by the convectivemass-transfercoefficient k, leading to
An*_~ fDo (1 + Pe)
An kL
with f the void fraction of the membrane.For thin highly porous
membranes,entrance effects thus can be significant even with rolL<<1
(Malone & Anderson1978).
Anderson& Quinn (1974) examinedsteady-state transport through
long pore with equilibrium at the ends, neglecting dispersive effects
arising from coupling betweenradial diffusion and the parabolic veloc-
ity profile. ThenPexp(V/kT)= n(z) for all Pe so that integration of (53)
over the cross section gives

Pe(u)~- = (dtt) d2---~-n

~ (55)
dz
with n=no at z=O and n--n~ at z=l. The dimensionless particle
 Annual Reviews
www.annualreviews.org/aronline

4~8 RUSSEL

velocity

(u)
f ue- Z/~T da
(56)
f e- v/kr da

and pore diffusivity

e v/~r da
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

f dll
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

(d,)
f e - v/~r da

are normalized to characterize the motion of an average particle within

the pore. The dimensionless axial flux, normalized on diffusion over the
full cross section, is
dn + Pe(u)n)

e<P~>n-n~
--~<Pe)(dll) (Pe)-I (57)

where (Pc) = Pc(u)/(dll ) ; the steric

relates the average concentration in the pore mouths to the bulk fluid
concentrations no and n~.
Brenner & Gaydos (1977) treated the unsteady situation including
dispersive effects by extending the classical Taylor-Aris theory to allow
for the variable mobilities and interaction potential crucial to pore
transport. Their result for the dispersion coefficient

D*--D0
(
(dll ) -~- (dr)) pe*2
(59)

indicates the effect to be significant for transient processes whenPc* > 1.

At steady state, however, since Pe~.Pe* the diffusion flux and also the
effect of dispersion are then insignificant relative to convection (Ander-
son & Quinn 1974).
Brenner & Gaydos 0977) calculated all three hydrodynamic coeffi-
cients from the best available approximate solutions for a rigid sphere
moving in a tube. Their asymptotic expansions for the mobilities, valid
for a/ro<<1, matchedexact solutions for motion near a plane wall in the
inner region (ro-r)/ro<<l with approximations from the method of
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN MOTION 449

reflections for (ro-r)/a>>1. Withoutlong range interaction potentials

the sphere is merelyexcludedfromr 0- a < r < r 0 and
9 aln a _ 1.539 a +o(a
b_dll_=l+~(-:) ro ro
ro ~ ro !

(6O)
[ a \2 [ a \2
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

(dv>--l-3.862--a+r0 14.40~)
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

with = (1- a/r o)z. The interactions with the wall reducethe diffusive
flux substantially and the convective slightly, e.g. ~(dll),~0.6 and
q~(u)~0.9 for a/ro~.,O.1. The corresponding meanvelocity (u)
increased, however,becauseof the exclusionof particles fromthe slowly
movingfluid near the wall.
Several groups have attempted to test these predictions with the
track-etched membranesand either bovine serum albumin or poly-
styrene latices. Significant hindrancehas beenobservedbut comparison
with the theory remainsinconclusive due to difficulties with adsorption
onto the pore walls (Wong& Quinn1976) and electrostatic interactions
with the pore walls (Malone& Anderson1978).

THE EFFECT OF BROWNIAN MOTION ON THE

RHEOLOGY OF SUSPENSIONS
Suspensionsof particles in Newtonianliquids often exhibit markedly
non-ideal theological behavior. Withlarger particles for whichhydrody-
namiceffects dominate, the stress maynot be Newtonianand mayvary
with the history of the samplebut mustremain a linear function of the
rate of strain, dependentonly on the volumefraction and geometryof
particles and their instantaneous configuration (Batchelor 1970). For
smaller particles subject to Brownlanmotion, however,the constitutive
relation becomesnonlinear in the rate of strain and shows a fading
memory.The additional strain rate and time dependenceboth scale on
the diffusion time #a~/kT, with a the characteristic dimensionof the
particle; the data of Krieger (1972) in Figure 8 for polystyrenespheres
of severalsizes in three different fluids illustrate the formerscalingquite
clearly. Other nonhydrodynamic forces, particularly long-range col-
loidal interactions, can affect the rheologysignificantly (Russel 1980)
but will not be includedhere.
The physical mechanismresponsible depends on the geometry of the
particle. For example,a shear flow tends to orient a rigid rod in the
 Annual Reviews
www.annualreviews.org/aronline

4~0 RUSSEL
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

I0
"4
I0 I0 ,

Figure8 Shear-ratedependenceof the effective viscosity for monodispers suspensions

of polystyrene
spheresat ~=0.50in (--) wal:er, (o) benzylalcohol,(e): m-cresol(Krieger
1972).

direction of flow, causing gradients in the orientational distribution

function and thereby generating ,m opposing rotational diffusion pro-
cess. As a result Brownianmotion is significant even at infinite dilution
(Leal & Hinch 1971, Hinch & Leal 1972). With spheres, however,
orientations are irrelevant and positions only become significant at
finite concentrations. Then hydrodynamic and potential interactions
can cause spatial nonuniformities in the probability density which drive
translational diffusion. In both cases Brownian motion exerts two
influences, a direct contribution to the bulk stress tensor from the
Browniantorque or force acting on the particle and an indirect effect on
the viscous stress through the oricntational or spatial distribution func-
tion. Batchelor (1976b) recently has reviewed in detail the salient
features of both problems. This section will concentrate on the nature of
the direct Brownianstress and its role in the rheology of suspensions of
spheres.
The existence of a direct thermodynamic contribution to the bulk
stress in a suspension has been questioned at times (e.g. Leal & Hinch
1971). The conventional continuum, formulation ignores fluctuating
stresses due to thermal motion of the fluid and therefore misses the
direct Brownian contribution. A thermodynamic stress, including the
effect of Brownianmotion as well as any interparticle forces or torques,
must be appended to the bulk x~iscous stresses defined by Batchelor
(1970). The derivation, generally attributed to Giesekus (1962)
rotational motion and outlined by Batchelor (1977) for translational:~
diffusion of spheres, resembles the.. developmentof the stress tensor for,
the theory of rubber elasticity (Bird et al 1977).
 Annual Reviews
www.annualreviews.org/aronline

BROWbr[AN MOTION451

For a volume ~containing N axisymmetric particles with positions x;

and orientations e~ the Helmholtz free energy A depends on the parti-
tion function PN(Xk,ek)and the total interaction potential V(xk, e~,)
A =kTln PN+ V. (61)
For an arbitrary homogeneous deformation K perturbing the system
slightly from equilibrium
dA =~K: (o)th~=krdln PN+dV. (62)
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

With the chain rule for the differentiation and the relation between K
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

and particle displacements, the latter becomes

N
i~=l(XiK"
~-~-+(eiI-e, (kTlnP~v+V)~x
eiei):K~-~i) .
i

For arbitrary deformations K then

1N
(o) t~erm=~.i~l(X~---~ii+(e~l--eieiel). ~_~)(kTlnPN+V
)

(63)
= ~ (xiF~ +L~. e~.eg)

where F~ and L~ are the net nonhydrodynamic forces and torques,

respectively, acting on the ith particle. Clearly, gradients in the partition
function, due to either an applied flow or interpardele forces, generate
bulk effects equivalent to those from steady forces,
Din
~X P~v
i
-kT~

and torques,
kTe i
i 0~e
In P~v
- (64)

on the individual particles. The complete direct Brownianstress consists

of this thermodynamiccontribution plus the viscous stresses caused by
these forces and torques.
Calculation of these stresses requires knowledgeof PNin the presence
of flow. For dilute suspensions of spheres without long range interaction
potentials the conservation equation for the pair density P2 is

3z + V-uP 2 -- V-D" VPz (65)

with D=2(DII-D~2) from (28) and u(r) the relative velocity

spheres with separation r in a homogeneous shear flow E. The most
 Annual Reviews
www.annualreviews.org/aronline

452 RUSSEL

: J
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

I I
0 ,05 .10 A5 .20

Figure 9 Comparisonof the predicted low-shearlimiting viscosity for a dilute suspension

of hard spheres (~) with data from Saunders(1961) (o) for polystyrene latices.

significant Brownianeffects occur fc,r i~a3y/kT<<

1 for which

P2(r)= n(1 Ixa3kT r._~E.r f(~)).r 2 ~ (66)

The shear thus increases the pair density wherer. u < 0 and reduces it
where r.u> 0 causing Brownian forces which oppose the motion and
increase the stress (Batchelor1977).
Batchelorsresult for the low-shearlimiting viscosity

-- = 1 + 2.5~ + 6.2 q~2+ O(q~3) (67)

includes a direct Browniancontribution of 0.90q,2. Unfortunately, the
corresponding high-shear limit, wheredirect Brownianstresses become
negligible, has not beendeterminedfor shear flow. AnO(~,2) coefficient
less than 6.2 wouldclearly demonstratethat the decrease in viscosity
with increasing shear rate illustrated in Figure 8 arises from the di-
minishedrole of Brownianmotion. In Figure 9 (67) is plotted with the
data of Saunders(1961) for monodispersepolystyrene latices in water.
In viewof the experimentaldifficulties in suppressinglong-rangerepul-
sions without inducing aggregation and in measuringsmall viscosity
incrementsaccurately, the comparisonis quite good.

CONCLUDING REMARKS
In recent years several uncertainties in analyses of Brownianmotionof
small particles in liquids have been resolved. Existing hydrodynamic
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN MOTION 453

theories now appear to describe accurately the detailed motion and

collective behavior of dilute suspensions of colloidal particles, compact
macromolecules, and even small molecules, with a modified surface
boundary condition. Long-range interaction potentials, which can play
a dominant role, have yet to be included quantitatively in some cases.
Likewise, the difficult problem of multiparticle interactions at higher
concentrations remains. Nonetheless, the techniques for incorporating
Brownian effects into continuum treatments of suspensions have been
established.
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org
Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

ACKNOWLEDGMENTS

I am indebted to E. J. Hinch for sharing his thoughts on Brownian

motion with me and to my colleagues D. A. Saville and W. R. Scho-
walter for their support and comments.

Literature Cited
Ackerson,B. J. 1976. Correlations for in- Colloil Interface Sci., ed. M. Kerker.
teracting Brownianparticles. J. Chem. 4:502-12. NewYork: Academic.587 pp.
Phys. 64:242-46 Batchelor, G. K. 1970. Thestress systemin
Ackerson,B. J. 1978. Correlations for in- a suspensionof force-free particles. J.
teracting Brownian particles. II. J. Chem. Fluid Mech.41: 545- 70
Phys. 69:684-90 Batchelor, G. K. 1972. Sedimentationin a
Adam,G., Delbriick, M. 1968. In Struct- dilute dispersion of spheres. J. Fluid
ural Chemistryand MolecularBiology, ed. Mech. 52:245-68
A. Rich, N. Davidson,pp. 198-215. San Batchelor, G. K. 1976a.Brownishdiffusion
Francisco: Freeman,906 pp. of particles with hydrodynamic interac-
Aibers, J., Deutch, J. M., Oppenheim,I. tion. d. Fluid Mech.74:1-29
1971.GeneralizedLangevinequations, d. Batchelor, G. K. 1976b. Developmentsin
Chem.Phys. 54:3541-46 microhydrodynamics. In Theoretical and
Aider,B. J., Wainwright, T. E. 1967.Veloc- Applied Mechanics,ed. W.Koiter, pp.
ity autocorrelafions for hard spheres. 33-55. Amsterdam:North Holland. 260
Phys. Rev. Lett. 18:988-90 PP.
Aider, B. J., Wainwright,T. E. 1970. Decay Batchelor, G. K. 1977. The effect of
of the velocity autocorrelation function. Brownianmotionon the bulk stress in a
Phys. Rev. A 1:18-21 suspensionof spherical particles. J. Fluid
Aitenberger, A. R., Deuteh, J. M. 1973. Mech. 83:97-117
Light scattering from dilute maeromolec- Bauer, D. R., Brauman,J. I., Peeora, R.
ular solutions. J. Chem.Phys. 59:894-98 1974. Molecularreorientation in liquids.
Aitenberger, A. R. 1979. Onthe wavevec- Experimental test of hydrodynamic
tor dependentmutualdiffusion of inter- models. J. Am. Chem.Soc. 96:6840-43
acting Brownianparticles. J. Chem.Phys. Bauer, D. R., Brauman,J. I., Pecora, R.
70:1994-2002 1976. Depolarizedlight scattering from
Anderson,J. L., Quinn, J. A. 1974. Re- liquids. Ann. Rev. Phys. Chem.27:443-
stricted transportin small pores.Biophys.
J. 14:130-49 Bean, C. P. 1972. In Membranes--A Series
Anderson, J. L., Rauh, F., Morales, A. of Advances,ed. G. Eisenman.1:1-54.
1978. Particle diffusion as a function of NewYork: Dekker. 333 pp.
concentrationandionic strength. J. Phys. Berne, B. J., Pecora, R. 1976. Dynam/c
Chem. 82:608-16 Light Scattering. NewYork: Wiley. 376
Anderson,J. L., Reed,C. C. 1976a. Diffu- PP.
sion of spherical macromoleeules at finite Bird, R. B., Hassager, O., Armstrong,R.
concentration. J. Chem.Phys. 64:8240- C., Curtiss, C. F., 1977. Dynamicsof
50 Po~,mericLiquids. Vol 2 Kinetic Theory.
Anderson, J. L., Reed, C. C. 1976b. In NewYork: Wiley. 250 pp.
 Annual Reviews
www.annualreviews.org/aronline

454 RUSSEL

Bloomfield, V. A. 1977: Hydrodynamics in Giesekus,H. 1962.Elasto-viskoseFliJssig-

biophysical chemistry. Ann. Rev. Phys. ke.!ten, fiir die in station/iren Sehichts-
Chem. 28:233-59 tromungen s/imtliehe Normalspan-
Bremaer, H. 1974. R.heology of a dilute nungskomponenten versehieden gross
suspension of axisymmetric Brownish sind. Rheol. Acta 2:50-62
particles. Int. J. MultiphaseFlow.1:195- Harris, S. 1976. Diffusioneffects in solu-
341 tions of Brownianparticles. J, Phys. A:
Brenner, H., Gaydos,L. J. 1977. The con- Math. Gen. 9:1895-98
strained Brownishmovementof spheri- Hauge, E. H., Martin.Ltf, A. 1973.
cal particles in cylindrical pores of com- Fluctuating hydrodynamics and Brow
parable radius. J. Colloid Interface Sci. nian motion.J. Stat. Phys. 7:259-81
58:312-56 Hess, W., Klein, R. 1978. Dynamicalprop-
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

Brown,J. C., Pusey, P. N., Goodwin, J. W., erties of colloidal systems.Physica94A:

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

Ottewill, R. H. 1975. Light scattering 71-90

study of dynamicand time averagedcor- Hinch,E. J., Leal, U G. 1972. Theeffect of
relations in dispersions of charged par- Brownian motion on the rheological
ticles. J. Phys. A: Math.Gen.8:664-82 properties of a suspension of non-
Chow,T. S,, Henna.as,J. J. 1973. Brownish spherical particles. J. Fluid Mech.52:
motionof a spherical particle in a com- 683-712
pressible fluid. Physics 65:156-62 Hinch, E. J. 1975. Applicationof the Lan-
Chu, B. 1974. Laser Light Scattering. New gevin equation to fluid suspensions. J.
York: Academic.317 pp. Fluid Mech.72:499- 511
Colby, P. C., Narducci,L. M., Bluemel,V., Hu, C., Zwanzig,g. 1974. Rotational fric-
Baer, J. 1975. Light scattering measure- tion coefficients for spheroids with the
ments from dense optical systems. Phys. slipping boundary condition. J. Chem.
Rev. A12:1530-38 Phys. 60:4354-57
Cone, R. A. 1972. Therotational diffusion Hynes,J. T. 1977. Statistical mechanicsof
of rhodopsinin the visual receptor mem- molecular motion in dense fluids. Ann.
brane. Nature NewBiol. 236:39-43 Rev. Phys. Chem.28:301-21
Cttmmins. H. Z., Pike, E. R., eds. 1974. Keller, K. H., Canales, E. R., Yum,S. I.
Photon Correlation and Light-Beating 1971. Tracer and mutual diffusion of
Spectroscopy. NewYork: Plenum. 584 proteins. J, Phys. Chem.75:379-87
PP- Kerker, M. 1969. The Scattering of Light
Deutch, J. M., Oppenheim, I. 1971. Molec- and Other Electromagnetic Radiation.
uiar theory of Brownian motion for NewYork: Academic,414 pp.
several particles: J. Chem.Phys. 54:3547 Kitchen, R. G., Preston, B. N., Wells,J. D.
-55 1976. Diffusion and sedimentation of
Edidin, M, 1974,. Rotational and transla- serum albumin in concentrated solu-
tional diffusion, in membranes. Ann.Re~. tions. J. Po~vm.Sci. Syrup. 55:39-49
Biophys. Bioeng. 3:179-201 Krieger, I. M. 1972. Rheolog~of monodis-
Einstein, A. 1956.Investigationson the The- perse latices. Adv. ColloidInterface Sci.
ory of Brownian Movement.NewYork: 3:111-36
Dover.122 pp.. Langevin, P. 1908. Sur la th~orie du
Ermak, D, L., McCammon,J. A. 1978, mouvementbrownien. C. R. Acad. Sci.
Brownian dynamics with hydrodynamic 146:530-33
interactions, or. Chem.Phys. 69:1352-60 Lax, M.1966. Classical noise IV: Langevin
Fair, B. D., Jamieson,A. M. 1980. Effect methods. Rev. Mod.Phys. 38:541-66
of electrodyrtamic interactions on the Leal, L. G., Hineh,E. J. 1971.Theeffect of
translational diffusion of bovine serum weakBrownishrotations on particles in
albuminat finite concentration. J. Col- shear flow. J. Fluid Mech.46:685-703
loid Interface Sci. 73:130-35 Malone, D. M., Anderson, J. L. 1978.
Felderhof, B. U. 1978. Diffusion of inter- Hindereddiffusion of particles through
acting Browuianparticles. J. Phys. A: small pores. Chem.Eng. Sci. 33:1429-40
Math. Gen. 11:929-37 Mazur,P., Oppettheim,I. 1970. Molecular
Fijnant, H. M., Pathmamanoharan, C., theory of Brownian motion. Physica
Nieuwenhuis,E. A., Vrij, A. 1978. Dy- 50:241-58
namiclight scattering from concentrated McDonnell,M. E., Jamieson, A. M. 1977.
colloidal suspensions. Chem.Phys. Lett. Quasielastic light scattering measure-
59:351-55 mentsof diffusion coefficients in poly-
 Annual Reviews
www.annualreviews.org/aronline

BROWNIAN MOTION 455

styrene solutions. J. Macromol.Sci.-- Saffman, P. G. 1976. Brownianmotion in

Phys. B13:67-88 thin sheets of fluid. J. Fluid Mech,
Millar, D. P., Shah, R., Zewail,A. H. 1979. 73:593-602
Picosecond saturation spectroscopy of Saffman, P. G., Delbriick, M. 1975.
cresyl violet: rotational diffusion by a Brownian motion in biological mem-
"sticking" boundary condition in the branes. Proc. Natl. Acad.Sci. 72:3111-13
liquid phase. Chem.Phys. Left. 66:435- Saunders, F. L. 1961. Rheological proper-
ties of monodisperselatex systems I.
Nelson, E. 1967. DynamicalTheories of Concentration dependence of relative
BrownianMotion. Princeton Univ. Press. viscosity. J. ColloidSci. 16:13-22
142pp. Schaefer, D. W., Berne, B. J. 1974. Dy-
Newman, J., Swinney,H. L., Berkowitz,S. namics of charged macromolecules in
Annu. Rev. Fluid Mech. 1981.13:425-455. Downloaded from www.annualreviews.org

A., Day, L. A. 1974. Hydrodynamic solution. Phys. Rev. Left. 32:1110-13

Access provided by Universidade de Brasilia on 12/12/16. For personal use only.

properties and molecular weight of fd Silebi, C. A., McHugh,A. J. 1978. An

bacteriophage DNA.Biochemistry 13: analy.sis of flow separation in hydrody-
4832-38 namic chromatography of polymer
Perrin, M. J. 1910. BrownianMotion and latices. AIChEJ. 24:204-11
Molecular Realffy. London: Taylor & Suthefland, G. B. B. M.1905. A dynamical
Francis. 93 pp. theory of diffusion for non-electrolytes
Phillies, G. D. J. 1973. Effect of inter- and the molecular mass of albumin.
molecularinteractions on diffusion. I. Philos. Mag.9:781-85
Two-componentsolutions. J. Chem. Topalian, J. H., Maguire,J. F., MeTague,
Phys. 60:976-82 J. P. 1979. Liquid CI2 dynamicsstudied
Pomeau,Y., R~sibois, P. 1975. Timede- by depolarized Raman and Rayleigh
pendent correlation functions and mode- scattering. J. Chem.Phys. 71:1884-88
mode coupling theories. Phys. Rep. yon Jena, A., Lessing, H. E. 1979. Rota-
19c:64-139 tional diffusion anomaliesin dye solu-
Poo, M., Cone,R. A. 1974. Lateral diffu- tions from transient dichroismexperi-
sion of rhodopsin in the photoreceptor ments. Chem.Phys. 40:245-56
membrane.Nature 247:438-41 Wills, P. R. 1979. Isothermaldiffusion and
Pyun, C. W., Fixman,M. 1964. Frictional quasi-elastic light-scattering of macro-
coefficient of polymermoleculesin solu- molecular solutes at finite concentra-
tion. J. Chem.Phys. 41:937-44 tions, J. Chem,Phys. 70:5865-74
Rahman,A. 1964. Correlations in the mo- Wong,J. H., Quian, J. A. 1976. In Colloid
tions of atoms in liquid argon. Phys. and Interface Science, ed. M. Kerker.
Rev. A136:405-11 V:169-80. NewYork: Academic. 507
Reed,C.C., Anderson,J. L. 1980. Hindered PP.
settling of a suspensionat low Reynolds Wu,E. S., Jacobson,K., Szoka,F., Portis,
number.AIChEJ. In press A. 1979. Lateral diffusion of a hydro-
Reed, T. M., Gubbins,K. E. 1973. Applied phobic peptide in phospholipid multi-
Statistical Mechanics. NewYork: Mc- layers. Biochemistry17:5543-50
Crraw-Hill.506pp. Youngren,G. K., Acrivos, A. 1975. Rota-
Russel, W.B. 1980. A reviewof the role of tional friction coefficientsfor ellipsoids
colloidal forces in the theologyof sus- and chemical molecules with the slip
pensions. J. Rheol. 24:287-317 boundary condition. J. Chem. Phys.
Rnssel, W.B., GlendinningA. B. 1980. The 63:3846-48
effective diffusion coefficient detectedby Zwanzig,R. 1969. Langevintheory of poly-
dynamiclight scattering. J. Chem.Phys. mer dynamicsin dilute solution. Adv.
Submitted Chem.Phys. 15:325-32
Russel, W.B., Hinch, E. J., Rallison, J. Zwanzig,R., Bixon, M.1975. Compressibil-
1980. Ongradient diffusion of rigid ity effects in the hydrodynamic theory of
spheres in concentratedsuspensions. J. Brownianmotion. J. Fluid Mech.69:21-
Fluid Mech.Submitted 25