See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/257216572

Brownian walker in a confined geometry leading to a space-

dependent diffusion coefficient

Article  in  Physica A: Statistical Mechanics and its Applications · February 2002

DOI: 10.1016/S0378-4371(01)00510-6

CITATIONS READS

49 275

4 authors, including:

Pascal Lançon Ghassan George Batrouni

University of Nice Sophia Antipolis University of Nice Sophia Antipolis
14 PUBLICATIONS   578 CITATIONS    234 PUBLICATIONS   7,181 CITATIONS   

SEE PROFILE SEE PROFILE

Nicole Ostrowsky
University of Nice Sophia Antipolis
43 PUBLICATIONS   1,137 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

BIOMAG View project

developing many body methods View project

All content following this page was uploaded by Ghassan George Batrouni on 05 April 2019.

The user has requested enhancement of the downloaded file.

 Physica A 304 (2002) 65 – 76
www.elsevier.com/locate/physa

Brownian walker in a con!ned geometry leading

to a space-dependent di%usion coe'cient
P. Lan)cona , G. Batrounib , L. Lobrya , N. Ostrowskya; ∗
a Laboratoire de Physique de La Mati ere Condensee (CNRS UMR 6622),
Universite de Nice Sophia-Antipolis, Parc Valrose, 06108 Nice Cedex, France
b Institut Non Lin eaire de Nice (CNRS UMR 6618), Universite de Nice Sophia-Antipolis,
1361 Route des Lucioles, 06560 Sophia Antipolis, France

Abstract
Brownian motion in a con!ned geometry has been studied using both light scattering and
digital video microscopy. The particles were trapped between two nearly parallel walls making
their con!nement position dependent. Consequently, not only did we measure a di%usion coef-
!cient which depended on the particles’ position, but also report and explain a new e%ect: a
drift of the particles’ individual positions in the direction of the di%usion coe'cient gradient, in
the absence of any external force or concentration gradient.  c 2002 Elsevier Science B.V. All
rights reserved.

PACS: 05.40.Jc; 82.70.Dd; 67.40.Hf

Keywords: Brownian motion; Colloids; Hydrodynamics in speci!c geometries; Flow in narrow channels

1. Introduction

Brownian motion, although a very old subject, is still fascinating our physics com-
munity, helping us to understand many of the new phenomena which have emerged in
recent years. The old model, however, has been “spiced up” by subjecting the Brownian
walker to various constraints.
For example, Brownian motion of spherical colloidal particles in the vicinity of a
wall has been extensively studied, both theoretically [1–3] and experimentally [4,5].
It has been shown that the di%usion coe'cients parallel or perpendicular to the wall
were greatly reduced when the particles were close enough to the obstacle, i.e. within

∗ Corresponding author. Tel.: +33-4-92-07-67-81; fax: +33-4-92-07-67-54.

E-mail address: nostro@naxos.unice.fr (N. Ostrowsky).

0378-4371/02/$ - see front matter
c 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 4 3 7 1 ( 0 1 ) 0 0 5 1 0 - 6
66 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76

distances comparable to or less than their radius. When the particles are trapped in a
more con!ned geometry, such as a porous medium, the theory is far more complicated,
and a few experimental studies have been recently reported [6,7]. We have set up a
series of experiments in a model geometry, where the particles are trapped between
two nearly parallel walls [8,9] and we have been able to study the Brownian motion
either perpendicular or parallel to the walls, for di%erent controllable con!nements.
We shall brieKy recall in Section 2 our experimental results on the dependence of the
Brownian walker di%usion constant with the con!nement and compare it with existing
theories. We shall then show (Section 3) that when the di%usion constant becomes
position dependent, D(x), the Brownian walkers behave in a counter-intuitive fashion:
they exhibit a drift, i.e., their average displacement increases with time, but no particle
Kux is observed and the particle concentration does not vary in time. Experimental
evidence of this puzzling e%ect will be presented in Section 4.

2. Brownian motion of particles trapped between two nearly parallel walls

2.1. Brownian motion perpendicular to the walls

The particles were trapped in a wedge-shaped cell (see Fig. 1a) where the con!ne-
ment was adjusted by changing the height of the incident laser beam probing by light
scattering the Brownian dynamics of the particles. The top view of the cell (Fig. 1b)
explains the wave vector geometry. By looking at the scattered light in a direction very
close to the reKected beam, the incident (k̃i ) and scattered (k̃s ) wave vectors are sym-
metric with respect to a plane parallel to the wall. The scattering wave vector q̃ is thus
perpendicular to the wall, which enables us to selectively study the component of the
particles motion perpendicular to the wall. As mentioned above, the closer a particle
is to a wall, the slower its di%usion motion. We thus expect the particle’s di%usion

Fig. 1. (a) Light scattering cell with a height-dependent con!nement (wedge shaped cell). (b) Enlarged top
view showing the directions of the incident k̃i , scattered k̃s , and scattering q̃ wave vectors. (c) Sketch of the
expected variation of D⊥ as a function of the distance z from the left wall. Only the average value can be
measured in the light scattering experiment.
 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76 67

Fig. 2. Variation of the di%usion constant D⊥ as a function of the relative con!nement e=2a. Measurements
were performed on particles with radius a = 0:11
m (white triangle) and a = 0:039
m (black triangles).
The various !ts are explained in the text.

coe'cient to depend on its distance z from say the left wall to vary as sketched in
Fig. 1c, with a positive gradient when it is close to the left wall, and a negative gradient
when it gets close to the right wall.
In our experiment, the scattered light came from particles distributed equally in
the whole z range. We thus measured for each con!nement, e, an average value

D⊥ (z)=D0 z which is plotted in Fig. 2 as a function of the relative con!nement e=2a
(where 2a is the particle diameter). The theoretical curves were computed by simply
adding the e%ect of each wall, to second order (for the dotted line) or using the ex-
act formula (for the dashed line). The solid line included the e%ect of the backKow
resulting from the simultaneous presence of the two walls [8].
We emphasize that in this experiment, the di%usion coe'cient was con!nement
dependent, but because of the ensemble average over all the particles in the scattering
volume, we could not study the e%ect of a space-dependent di%usion coe'cient on a
Brownian walk.

2.2. Brownian motion parallel to the wall

To study the Brownian motion parallel to the wall, we monitored the trapped particles
using digital video microscopy. As the two walls were not quite parallel but made a
slight angle with each other (see Fig. 3), the con!nement, e, was dependent on the
particle’s position, x, thus leading to a space-dependent di%usion coe'cient D== (x). The
con!nement, e, must be su'ciently small (on the order of the particles size) to modify
the di%usion coe'cient, and the angle between the walls large enough for the di%usion
coe'cient to vary signi!cantly over the explored range. To achieve these conditions in
68 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76

Fig. 3. Brownian walker trapped between two nearly parallel walls. The step length is a function of the
con!nement e, and thus of the position x.

Fig. 4. Left-hand side: experimental setup for the video microscopy. The rectangular inset is an enlarged top
view showing the center of the cell with the circular excluded volume and the observation frame. The round
inset explains the two contributions to the change in di%usion coe'cient when a particle moves a distance
d x. Right-hand side: D== =D0 with respect to relative con!nement e=2a. Open squares are the experimental
data, black dots were calculated by the collocation method, and dotted and solid lines follow analytical
approximate solutions.

the middle of the cell and not on the edge, as is the case for the wedge cell mentioned
above, we have changed the geometry of the cell. The suspension was con!ned between
a Kat disk and a planar convex of curvature radius R = 15:5 mm separated by an elastic
O-ring (see left-hand side of Fig. 4). The spacing, e, between the Kat and curved
walls depended on the distance, r, from the center of the cell as e = r 2 =2R. The
contact between the two walls as well as the dependence of the con!nement e on the
distance r were carefully measured by monitoring the Newton rings observed under
the microscope. We used as a light source a new super-radiant diode 1 whose coherence
length is less than 100
m. This was important as this coherence length was long
enough to observe the desired Newton rings, but short enough to avoid any other
interference patterns due to all the cell interfaces, which were visible with an ordinary
diode laser and completely masked the relevant signal (see Fig. 5).
Polystyrene spheres, of radius a = 1
m, were suspended in a mixture of H2 O + D2 O
so as to cancel any sedimentation e%ects. A surfactant is added (2:2 g=l of SDS) into
the suspension to prevent aggregation to the walls. The horizontal Brownian motion of

1 Kindly supplied by EPF (Lausanne) and MITEL Semiconductors (Stockholm).

 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76 69

Fig. 5. Newton rings observed with light sources having di%erent coherence lengths. (a) Diode laser;
(b) Superradiant diode; (c) White light + !lter and (d) Microcavity LED.

the polystyrene balls was observed through a microscope equipped with a long-range
objective of magni!cation 50×, followed by a CCD camera coupled to the microscope
via an eye piece of magni!cation 8×. The video signal was processed in real time by
a computer, which recorded, every 3 s, the horizontal position, size and shape of all
objects in a rectangular frame 65
m × 100
m. This time interval was long enough
to allow the image analysis of all the particles present in the frame and small enough
so that the particle’s average displacement was only a fraction of their diameter. As
the particles’ con!nement, e, was related to their distance, r, to the center of the
cell, we were able to explore di%erent con!nement regions by moving the observation
frame in the horizontal plane. The explored con!nements varied from 2.5 to 11
m.
The vertical Brownian motion of the particles over this short vertical range could not
be monitored. However, we took that motion into account when interpreting the data
by averaging the particle’s vertical position over the con!nement range. The volume
fraction of polystyrene balls, of the order of 1%, was chosen so as to be able to follow
a fairly large number of particles at the same time (around 30 for e=3
m) to improve
the statistics in the data analysis. Following the particles’ positions from one frame to
another, the program was able to analyze a great number of trajectories (more than 105
for each run). When two particles got closer than twice their diameter, the program
treated them as “dimers”, their trajectories as “monomers” were ended at that time and
70 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76

were no longer used to determine the di%usion coe'cient or the drift. In that way,
the role of particle interactions, which have a range smaller than a couple of particle
diameters, could be safely ignored.
The con!nement dependence of the di%usion coe'cient was determined as follows.
A given observation frame was divided into 3 zones (see rectangular inset in Fig. 4),
each corresponding to an approximately constant con!nement e. For each zone we
averaged the particle’s displacement squared, either in the x, or in the y directions, as
a function of time, and checked that indeed they followed the well-known di%usion
law:

x2  =
y2  = 2D== (e)t : (1)
By moving the frame to di%erent locations, we were able to explore a range of con-
!nement extending from e=2a = 1:2–11. The bulk di%usion constant D0 was determined
using the well-known relation:
kT
D0 = ; (2)
6a
where  = 0:99 × 10−3 SI is the viscosity of the mixture of water and heavy water,
yielding D0 = 1:92 × 10−13 m2 =s. The experimental values for D== =D0 are shown on
the right-hand side of Fig. 4 (white squares) and !t remarkably well the available
numerical predictions (black dots) using the collocation method [10] averaged over all
the possible vertical positions z of the particle for a given con!nement e, i.e., with
a 6 z 6 (e − a). For comparison, we also plotted as a solid line the analytical solution
obtained using the Faxen expression [3] for the position-dependent drag of a particle
moving parallel to a single wall, then adding the e%ect of each of the two walls, and
averaging over the vertical position z. This solution clearly overestimates the reduction
of the di%usion coe'cient of a particle trapped between two parallel walls, particularly
as the relative con!nement  = e=2a reaches its lower limit 1. We also plotted (dashed
lines) Faxen’s analytical expression [3] for the drag of a particle trapped just in the
middle of two parallel planes. The impossibility of averaging over z an expression
only known for z = e=2 explains the observed discrepancy, which goes to zero as the
relative con!nement approaches its limit 1, where the z-average becomes irrelevant.
We point out that in this situation, there is a one-to-one correspondence between the
con!nement e and the particle’s position x. This setup is thus a controlled experimental
realization of an old theoretical problem, the Brownian walker with a space-dependent
di%usion coe'cient. We will now show that this constrained Brownian motion may
be represented by a random walk, provided the steps are biased in the direction of
the gradient of the di%usion constant. This leads to an apparent measurable drift of the
particle’s positions which is not accompanied by any Kux, as the particles concentration
remains in equilibrium.

3. Brownian walker with a space-dependent diusion coecient

As in our experiment the di%usion coe'cient varies in only one direction, we brieKy
sketch a heuristic derivation of the 1D Brownian walker algorithm.
 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76 71

The velocity of a 1D Brownian particle subjected to a random force and a viscous

drag follows the Langevin equation [11]:
dv
= −v + (t) ; (3)
dt
where −1 is the velocity relaxation time and (t) the random force per unit mass,
with zero mean and a correlation function proportional to a  function:

(t) = 0 and
(t)(t  ) = q(t − t  ) : (4)
Using the equipartition theorem it can be shown that q is related to the temperature T
and the particle’s mass m by the standard relation q = 2kT=m.
Discretizing the random function (t) over time intervals Rt−1 allows us to drop
in Eq. (3) the inertial term, dv=dt, and to replace v by Rx=Rt. Choosing

for (t) the
simplest random function which obeys relations (4), i.e., (t) = ± q=Rt, leads to the
well-known Brownian walker algorithm:

1 q
x(t + Rt) = x(t) ± Rt
 Rt
√ kT
= x(t) ± 2DRt with D = : (5)
m
When the di%usion coe'cient D, i.e., when the temperature T and=or the drag coe'-
cient  become position dependent, the above classical algorithm may still be used, but
it needs to be clari!ed. During each time interval Rt, the walker makes a√step to the
right or to the left, but should the length of this position-dependent step, 2DRt, be
computed at the departure point x(t) = x, the arrival point x(t + Rt) = x + Rx or at any
point in between? These mathematical choices, often referred to as the Ito=Stratonovitch
convention [11], model di;erent physical situations and the choice of the convention
is dictated only by the physics as we shall now explain. We denote by D(x + Rx)
the di%usion coe'cient appearing in (5) where  = 0, 1=2 and 1 correspond to the
Ito, Stratonovitch and isothermal choices, respectively. As we will show, this last case
models a situation where the temperature, T , is uniform but the drag coe'cient, , is
space dependent. Using in (5) the standard limited expansion:
dD

D(x + Rx) ≈ D(x) +  Rx with Rx = ± 2D(x)Rt (6)
dx
yields for the algorithm of a Brownian walker with a position-dependent di%usion
coe'cient:

dD
x(t + Rt) = x(t) ± 2D[x(t)]Rt +  Rt : (7)
dx
Depending on the value of , this algorithm has very di%erent implications concerning
the equilibrium distribution of the Brownian walkers, their individual drift
x(t) − x(0)
and their net Kux. This algorithm has been used [12] with  = 1 to model a Brownian
walker with a space-dependent di%usion coe'cient. We shall justify this choice in the
case where the di%usion coe'cient gradient does not come from a temperature gradient.
72 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76

Fig. 6. Particles Kux for Brownian walkers with a step length depending on arrival position ( = 1) or
departure position ( = 0).

Averaging Eq. (7) over a large number of walkers shows that the average position
of a Brownian walker is no longer zero since the di%usion gradient term acts as an
external force leading to particle drift. If this gradient is assumed to be constant, this
drift increases linearly with time as
dD

x(t) − x(0) =  t: (8)
dx
A !rst intuitive but misleading idea would be to conclude that the particles will migrate
in the direction of the di%usion gradient, leading therefore to a concentration gradient.
This is actually incorrect as we will now show. To illustrate this idea, let us suppose
we start from a uniform particle distribution 0 . To check if this corresponds to an
equilibrium state, let us determine the particles Kux through an imaginary surface S
placed perpendicular to the di%usion coe'cient gradient, at coordinate x (see Fig. 6).
During a time interval Rt, all the particles crossing S from the left (or right) are half
of those included in the volume SLright (or SLleft ) where Lright (or Lleft ) is the right
(left) step, terminating at x, taken by a walker during that time interval. The net
particle Kux to the right will thus be
0 SLright − SLleft
J= : (9)
2 SRt
Algorithm (5) allows computing the length of these two steps both of which end at
the same point x:

dD
L right = 2D(x)Rt ± ( − 1) Rt (10)
dx
left

leading to the particle Kux:

dD
J = −0 (1 − ) : (11)
dx
As a result, in the situation of maximum drift where  = 1 this Kux will vanish (see
left part of Fig. 6), meaning that the uniform particle distribution corresponds to an
equilibrium. According to Boltzmann, this should correspond to an isothermal situation,
the di%usion coe'cient gradient arising only from a pure hydrodynamic e%ect, the
spatial dependence of the coe'cient . For all the other values of , the Kux will be
negative, leading to a concentration gradient of the particles in the direction opposite
to that of the di%usion coe'cient gradient. The maximum Kux is obtained for  = 0,
as shown in the right part of Fig. 6.
 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76 73

Fig. 7. Schematic pictures of particles with a space-dependent di%usion coe'cient. The arrows represent the
particles displacements during a given time interval. Averaging these displacements over all the particles in
a given volume V (either the gray particles or the white particles) results in a drift in the direction of the
gradient of the di%usion constant. The net Kux of particles (marked with an ×) across the surface S is zero.

Eq. (9) may be generalized to the case where the particle distribution is position
dependent:
dD d
J = −(1 − )(x) −D (12)
dx dx
which leads to the well-known generalizations of Fick’s law when dealing with a
space-dependent di%usion coe'cient [13].
To con!rm these results, we performed simulations of the Brownian walkers follow-
ing algorithm (5). We con!rmed that only the =1 case leads to a uniform distribution
of particles with no net Kux through any given surface while, at the same time, the
average individual positions exhibit a drift in the direction of the di%usion coe'cient
gradient according to Eq. (8). This situation of “drift without Kux” may be compared
with the equilibrium situation of the Brownian particles subjected to an external force,
such as their weight. If one follows the motion of individual particles in a given volume
V , an average downward drift is observed; however, there is no net Kux across a given
surface S because of the vertical concentration gradient. The individual downward drift
of the sedimenting particles leads to a downwards Kux which is exactly compensated
by the upward Kux linked to the equilibrium concentration gradient. In our isothermal
case, the drift of individual particles in a given volume V from lower to larger D(x)
region does not lead to a net Kux across a surface S because particles in the larger D(x)
region di%use further than particles in the lower D(x) region. This physical situation
imposes the choice of  = 1 in algorithm (7), so that a particle coming from a low
D(x) region makes a right step just equal to the left step of that particle coming from
a high D(x) region and arriving at the same point (see left part of Fig. 6).
It should be emphasized that the drift is measured averaging the positions of all the
particles contained in the volume V , whereas the net Kux involves only those particles
which cross the surface S from either side (see Fig. 7). In the absence of gradient,
the fact that the drift and the net Kux are measured on di%erent populations is rarely
mentioned, as it does not lead to any surprising result. In the presence of a gradient,
be it a concentration gradient or, as in our case, a di%usion coe'cient gradient, it does
74 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76

lead to a counter intuitive idea: A drift in the individual particle positions does not
always result in a net Kux, and may be observed in an equilibrium situation where the
particles concentration does not change with time.
Physically, the origin of the Ito=Stratonovitch problem is that we are neglecting the
dv=dt term in the Langevin equation (3), under the assumption that dt−1 . As a
consequence, we lose information at very small time scales. We are, therefore, obliged
to make a choice (Ito, Stratonovitch or isothermal) depending on the physics.

4. Experimental evidence of the drift

As pointed out in Section 2.2, our video microscopy experiment is a controlled ex-
perimental realization of Brownian walkers with a space-dependent di%usion coe'cient,
due to a purely hydrodynamic e%ect, and in the absence of any temperature gradient.
We thus expect to be in the  = 1 case described above, where the drift is maximum
and the Kux zero.
To demonstrate the existence of an individual drift of the particles, we !xed the
center of the observation frame at a position y = 0 and x = 300
m, corresponding to
an average relative con!nement e=2a = 1:5 so that all particles present in the frame
were outside the excluded volume (i.e., e ¿ 2a), and had a di%usion coe'cient with
the largest x dependence, but no y dependence (to !rst order). For the determination
of
x(t) − x(0) and
y(t) − y(0), each trajectory was segmented into independent
paths lasting a time t, each contributing to the evaluation of the average drift during
time t. The results are shown in Fig. 8, and reveal a drift in the Brownian walker
position along the x direction, and none in the y direction along which the di%usion
coe'cient may be considered as constant. The statistics of the results clearly deteriorate
as the time t increase: After recording trajectories for typically a dozen hours, more
than a hundred thousand independent segments contributed to the determination of
the drift at short times, whereas only up to a few thousand independent segments
were left for t = 200 s. This is due to the fairly high particle concentration which
lowers the lifetime of a “monomer” (time during which a particle does not approach

Fig. 8. Average position of the walkers as a function of time, along (black dots) and perpendicular (open
squares) to the di%usion gradient.
 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76 75

another one within 2 particles’ diameter—see above), but which was chosen as a
compromise to have good statistics at short times while allowing us to follow each
particle during a reasonable time, !xed at 200 s. It should be pointed out that in order
to avoid any bias in the statistics, for a trajectory segment to be valid and included
in the statistics, the position of a walker at instant t = 0 had to be inside an internal
frame, 15
m away from the edges of the observation frame. This condition ensured
that after di%using for 200 s, the walker had less than 0.5% chance to have covered
15
m, and was thus still present in the observation frame. Failure to impose this
condition resulted in the observation of a spurious drift, in the opposite direction, due
to an arti!cial selection of walkers because of the experimental boundary conditions
(limits of the observation frame).
To compare our experimental results with the theoretical predictions (Eq. (7)), we
evaluated the di%usion coe'cient gradient encountered by the walkers present in the
observation frame. It is important to realize that as a walker moves a distance d x (see
round inset in the left-hand side of Fig. 4), its di%usion coe'cient D== varies !rst
because the con!nement e = x2 =2R varies (path (1) parallel to the bottom wall, i.e.,
at constant z), and second because at constant con!nement, the particle’s altitude z
changes (path (2)). Adding both contributions and averaging over the vertical position
z of the walker yields:
     
dD== (e; z) x @D== (e; z)  x @D== (e; z) 
=  +  : (13)
dx z R @e z z 2R @z e z

Using our experimental data and the collocation numerical results, we found as a
numerical result:
dD== =d xz ≈ 2:2 × 10−9 m=s. This value of the slope is used to plot
the straight dotted line in Fig. 8. The experimental data are thus in good agreement
with the predicted drift corresponding to the expected  = 1 value.
Finally, to claim drift without Kux, it is not su'cient only to demonstrate the drift:
we must also demonstrate the absence of Kux. If a Kux was due to the observed drift
dD=d x, we would expect a radially outwards Kux of dD=d x particles, which would
empty our observation screen in less than a day. Furthermore, if this Kux were to be
balanced by a concentration gradient, one can show that a concentration change by 30%
over a distance of 60
m would be necessary. Experimentally, we observed no Kux and
no concentration gradient over a period of a week or more, which is consistent with
the Boltzmann requirement of a uniform concentration in the absence of a temperature
gradient.
In conclusion, we have shown that our video microscopy setup is an experimental
realization of an old theoretical problem, the Brownian Walker with a space-dependent
di%usion coe'cient. This situation is often encountered in the studies of particle suspen-
sions in con!ned media, such as porous media [14], entangled polymer suspensions [15]
or particles trapped in vesicles [16]. It further illustrates the classical Ito–Stratonovitch
dilemma [17] where a multiplicative Langevin equation needs a proper interpretation
rule to describe a given phenomenon. It is worth noting that the rule naturally im-
posed by our physical conditions (no temperature gradient) is neither the Ito nor the
Stratonovitch convention, but a third choice, rarely mentioned in the literature.
 76 P. Lanc3on et al. / Physica A 304 (2002) 65 – 76

References

[1] H. Brenner, J. Fluid Mech. 12 (1962) 35.

[2] H. Brenner, Chem. Eng. Sci. 16 (1961) 242.
[3] H. Faxen, Arkiv. Mat. Astron. Fys. 27 (1923) 17.
[4] M.I.M. Feitosa, O.N. Mesquita, Phys. Rev. A 44 (1991) 6677.
[5] L.P. Faucheux, A.J. Libchaber, Phys. Rev. E 49 (1994) 5158.
[6] B. Lin, J. Yu, S.A. Rice, Phys. Rev. E 62 (2000) 3909.
[7] E.R. Dufresne, D. Altman, D.G. Grier, Europhys. Lett. 53 (2001) 264.
[8] L. Lobry, N. Ostrowsky, Phys. Rev. B 53 (1996) 12 050.
[9] P. Lan)con, G. Batrouni, L. Lobry, N. Ostrowsky, Europhys. Lett. 54 (2001) 28.
[10] P. Ganatos, S. Weinbaum, R. Pfe%er, J. Fluid Mech. 99 (1980) 755.
[11] H. Risken, The Fokker–Planck Equation, Springer, Berlin.
[12] D.L. Ermak, J.A. McCammon, J. Chem. Phys. 69 (1978) 1352.
[13] M.J. Schnitzer, Phys. Rev. E 48 (1993) 2553.
[14] G. Viramontes-Gamboa, J.L. Arauz-Lara, M. Medina-Noyola, Phys. Rev. Lett. 75 (1995) 759.
[15] P. Tong, X. Ye, B.J. Ackerson, L.J. Fetters, Phys. Rev. Lett. 79 (1997) 2363.
[16] A.D. Dinsmore, D.T. Wong, P. Nelson, A.G. Yodh, Phys. Rev. Lett. 80 (1998) 409.
[17] N.G. van Kampen, J. Stat. Phys. 24 (1981) 175.

View publication stats