Journal of Colloid and Interface Science 269 (2004) 425431

www.elsevier.com/locate/jcis
Effect of electrostatic, hydrodynamic, and Brownian forces on particle
trajectories and sieving in normal ow ltration
Myung-man Kim and Andrew L. Zydney

Department of Chemical Engineering, The Pennsylvania State University, University Park, PA 16802, USA
Received 23 April 2003; accepted 5 August 2003
Abstract
Particle deposition and fouling are critical factors governing the performance of microltration and ultraltration systems. Particle tra-
jectories were evaluated by numerical integration of the Langevin equation, accounting for the combined effects of electrostatic repulsion,
enhanced hydrodynamic drag, and Brownian diffusion. In the absence of Brownian forces, particles are unable to enter the membrane pores
unless the drag associated with the ltration velocity can overcome the electrostatic repulsion. Brownian forces signicantly alter this behav-
ior, allowing some particles to enter the pore even at low ltration velocities. The average particle transmission, evaluated from the probability
of having a particle enter the pore, increases with increasing ltration velocity due to the greater hydrodynamic drag force on the particle.
These results provide important insights into particle behavior in membrane systems.
2003 Elsevier Inc. All rights reserved.
Keywords: Filtration; Particle trajectory; Electrostatics; Hydrodynamic drag; Brownian diffusion
1. Introduction
Particle transport toward and through membrane pores
plays a critical role in determining the performance of both
microltration and ultraltration devices. The accumulation
of solutes or particles in the boundary layer near the mem-
brane, often referred to as concentration polarization, can
limit the ltrate ux through the additional hydraulic resis-
tance provided by the particle cake [1]. Bulk transport can
also control the rate of particle deposition (fouling) on the
membrane surface. Transport into the pores determines the
membrane retention characteristics, which are critical for the
high levels of particle removal required for the production
of ultrapure water [2,3], sterile ltration [4,5], and virus re-
moval [6,7].
Belfort and colleagues [810] evaluated particle trajecto-
ries during cross-ow microltration by solving the Navier
Stokes equations, including the inertial terms to account for
the presence of any inertial lift force. The membrane was
treated as a uniformly porous boundary, without distinct
pores. Particular emphasis was placed on calculating the ef-
fects of inertial lift and transmembrane ow(suction) on par-
*
Corresponding author.
E-mail address: zydney@engr.psu.edu (A.L. Zydney).
ticle motion at realistic particle Reynolds numbers. Results
demonstrate that large particles are transported away from
the membrane surface due to lift forces, allowing one to op-
erate cross-ow ltration devices at high ltration velocities
without signicant fouling or polarization. Small particles
tend to deposit on the membrane surface since the drag force
due to ltration is greater than the inertial lift.
Dagan et al. [11] studied the motion of an individual parti-
cle toward a single (isolated) pore on the membrane surface.
The additional hydrodynamic drag on the particle due to the
presence of the pore boundaries was included in the analy-
sis but no other forces were considered. All particles were
predicted to enter the pore due to the absence of any attrac-
tive or repulsive interactions. Kao et al. [12] extended this
basic approach to consider the effects of van der Waals at-
traction between the particles and the membrane, and they
also employed a more complete analysis of the hydrody-
namic interactions. A similar approach was used by Schmitz
et al. [13] to evaluate the effects of vortex formation within
the pore on the rate of particle deposition at high ltration
velocities.
Bowen and colleagues [1416] extended these analyses
to account for the effects of repulsive electrostatic inter-
actions between charged particles and the charged surface
of most ultraltration and microltration membranes. A -
0021-9797/$ see front matter 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcis.2003.08.004
426 M.-m. Kim, A.L. Zydney / Journal of Colloid and Interface Science 269 (2004) 425431
nite element technique was used to solve the NavierStokes
equations for a stationary particle placed at arbitrary posi-
tions in the ow. The same nite element approach was used
to solve the nonlinear PoissonBoltzmann equation to eval-
uate the electrostatic forces. Particle trajectories were not
determined, but the equilibrium particle position was eval-
uated from an overall force balance. Bowen and Sharif [16]
identied a critical ltration velocity below which the parti-
cles were excluded from the pore and kept from contacting
the membrane, due to the electrostatic repulsion.
Although these studies have provided important insights
into particle motion in membrane systems, the combined ef-
fects of electrostatic repulsion and enhanced hydrodynamic
drag on the particle trajectories have not yet been evaluated.
More signicantly, none of these studies has considered the
effects of Brownian (diffusive) forces on particle motion. In
this study, particle trajectories were evaluated by numerical
integration of the Langevin equation accounting for the elec-
trostatic, hydrodynamic, and Brownian forces. The hydrody-
namic forces were calculated using the detailed uid velocity
proles, evaluated in a particle-free system, with appropriate
expressions used for the enhanced drag coefcients. Sim-
ulations were performed to evaluate the importance of the
different force components on both particle motion and the
overall retention characteristics of the membrane.
2. Theoretical development
In order to simplify the analysis, particle trajectories were
evaluated for normal ow (dead-end) ltration through a
single slit-shaped pore with rounded (cylindrical) edges, as
shown in Fig. 1. This type of single-pore analysis can pro-
vide considerable insights into particle motion in membrane
systems, particularly for ultraltration and microltration
membranes with relatively low surface pore densities. The
slit geometry reduces the system to a two-dimensional prob-
lem, dramatically reducing the complexity and computa-
tional time. The use of a pore with rounded edges eliminates
Fig. 1. Geometry of single-pore system showing typical nite element dis-
cretization.
the mathematical discontinuity at the pore entrance, allow-
ing numerical convergence with a coarser grid.
Particle trajectories were calculated by numerical integra-
tion of the Langevin equation,
(1) m
p
du
p
dt
=6a[K
p
u
p
K
f
u
f
] +F
E
+F
B
,
where u
p
is the velocity vector for the particle, u
f
is the
unperturbed uid velocity (evaluated in the absence of the
particle at the location of the particle center), a is the parti-
cle radius, and m
p
is the particle mass. The rst term on the
right-hand side of Eq. (1) represents the hydrodynamic drag
on the particle, while the second and third terms account for
the electrostatic (F
E
) and Brownian (F
B
) forces. K
p
and
K
f
are diagonal matrices describing the additional hydro-
dynamic hindrance associated with interactions between the
particle and the system boundaries. These hindrance factors
approach a value of one at large distances from the mem-
brane, with the drag force given by Stokes law under these
conditions.
The electrostatic force between the particle and the mem-
brane was evaluated using the approach described by Bowen
et al. [14] based on Derjaguins approximation, yielding
(2) F
E
=
2
r

0
g
_
2
p

m
exp([h a]) (
2
p
+
2
m
)
exp(2[h a]) 1
_
,
where is the inverse Debye length,
0
is the permittivity of
free space,
r
is the dielectric constant, and h is the distance
of closest approach measured between the particle center
and the solid surface.
p
and
m
are the surface potentials
of the particle and membrane, respectively. The parameter g
is a steric factor associated with the specic geometry of the
system. For a particle interacting with a single at plate, as
in region I of the membrane system (see Fig. 1), the steric
factor is simply g =1/a. The force in region III was evalu-
ated as the vector sum of the forces associated with the two
parallel plates that form the boundaries of the pore, each of
which was calculated separately using the distance between
the particle and that plate. The force in region II was eval-
uated as the sum of the forces between the particle and the
two cylindrical surfaces that formthe pore entrance, with the
steric factor for each interaction given as [14]
(3) g =
_
1
a
_
1
a
+
1
b
_
,
where b is the radius of the cylindrical surface. The tran-
sition between regions I and II was dened so that the
x- and y-components of the electrostatic force remain con-
tinuous throughout the computational domain. This was
done by evaluating the y-component of the force using the
expression in region I until the point at which it equalled the
y-component of the force in region II. A similar approach
was used for the x-component and for the transition between
regions II and III.
M.-m. Kim, A.L. Zydney / Journal of Colloid and Interface Science 269 (2004) 425431 427
The Brownian force was modeled as a Gaussian white
noise process [17],
(4) F
B
=
_
12ak
B
T
t
,
where k
B
is the Boltzmann constant, T is the absolute tem-
perature, and t is the magnitude of the time step. The pa-
rameter is a Gaussian randomnumber with zero mean and
unit variance. The random direction of the Brownian force
was accounted for by evaluating both the x and y compo-
nents of F
B
at each time step using independent values of
in the two directions.
Equation (1) was nondimensionalized by scaling the par-
ticle and uid velocities using the ltration velocity (V
f
) and
scaling time using the particle relaxation time (t
R
):
(5) t
R
=
2
9
_
a
2

_
.
The nondimensional formof the Langevin equation becomes
(6)
du
p
d
=K
p
u
p
K
f
u
f
+F
E
+

Pe
_
2a
2
Dt
,
where D is the Brownian diffusion coefcient given by the
StokesEinstein equation:
(7) D =
k
B
T
6a
.
Equation (6) is expressed in terms of two key dimensionless
groups: the Peclet number (Pe = V
f
a/D), which describes
the relative importance of hydrodynamic drag and Brownian
diffusion, and the electrostatic interaction parameter,
(8) =
2
r

m
3V
f
,
which describes the relative importance of the electrostatic
and hydrodynamic forces. F
E
is the dimensionless electro-
static force, which is evaluated directly from the nondimen-
sionalization using Eq. (2). The term involving the square
root (in Eq. (6)) depends on the magnitude of the integration
time step, with typical values on the order of 10
1
.
The hindrance factors in region I (Fig. 1) were evaluated
using the expressions developed by Goldman et al. [18] for
the motion of a sphere parallel to a plane wall,
(9) K
px
=
1
1
9
16
_
a
h
_
+
1
8
_
a
h
_
3

45
256
_
a
h
_
4

1
16
_
a
h
_
5
,
and for linear ow past a stationary sphere located near a
plane wall,
(10) K
f x
=1 +
9
16
_
a
h
_
,
where h is again the distance between the particle center and
the surface. Wakiya [19] evaluated the drag on a sphere mov-
ing perpendicular to a plane wall as
(11) K
py
=
1
1
9
8
_
a
h
_
+
1
2
_
a
h
_
3
.
Drummond and Deen [20] developed an approximate ex-
pression for K
fy
based on numerical results for ow past
a stationary sphere perpendicular to a circular disk:
(12) K
fy
=
1
_
a
h
_
3.8
1
9
8
_
a
h
_
+
1
2
_
a
h
_
3
.
Faxen [21] evaluated the hindrance factor for a particle
located between two parallel plates, analogous to the geom-
etry in region III, as
(13) K
py
=
1
11.004
_
a
lp
_
+0.418
_
a
lp
_
3
+0.21
_
a
lp
_
4
0.169
_
a
lp
_
5
,
where l
p
is the pore half-width. Equation (13) was developed
for a particle at the pore centerline; this expression was used
at all positions within the pore since more complete analyt-
ical results are unavailable and the particle tends to migrate
toward the pore centerline due to the repulsive interactions
with the pore walls. Model calculations performed using the
area-average hindrance factor in a slit shaped pore [22] gave
results that were indistinguishable fromthose determined us-
ing Eq. (13).
Deen [23] presented an expression for the hindrance fac-
tor for a stationary particle at the centerline of a slit-shaped
pore,
(14) K
fy
=
1
1
3
_
a
lp
_
2
11.004
_
a
lp
_
+0.418
_
a
lp
_
3
+0.21
_
a
lp
_
4
0.169
_
a
lp
_
5
,
where Eq. (14) was again assumed to be valid at all positions
in the pore. Ganatos et al. [24] evaluated the hindrance fac-
tor for a particle moving perpendicular between two plates,
but the results were only presented in graphical form. In-
stead, K
px
and K
py
were evaluated from the expressions for
a single plate using the distance to the closer plate. The hin-
drance factors in region II were linearly interpolated between
the values at the boundaries of regions I and III to provide
a smooth transition in K
px
, K
py
, K
f x
, and K
fy
throughout
the entire domain.
In order to solve Eq. (1), the uid velocity was rst eval-
uated in a particle-free system by solving the steady-state
NavierStokes equation for an incompressible uid without
the inertial terms:
(15) p +
2
u
f
=0,
(16) u
f
=0.
No-slip boundary conditions were applied at all surfaces
of the membrane, with the feed velocity and pressure as-
sumed to be uniform far from the membrane surface. Equa-
tions (15) and (16) were solved numerically using a -
nite element method implemented in FLUENT 6.0 (Fluent,
Inc., Lebanon, NH) with a typical discretized grid shown
in Fig. 1. The particle trajectory was then evaluated using
a trapezoidal integration scheme for Eq. (1) that was also
implemented within FLUENT.
Limited calculations were also performed to verify the
expressions used for the hindrance factors. In this case the
428 M.-m. Kim, A.L. Zydney / Journal of Colloid and Interface Science 269 (2004) 425431
particle was xed at the pore axis (x = 0) at a given dis-
tance above the membrane, with the uid velocity evaluated
for the particlemembrane system using no-slip boundary
conditions on both the particle and membrane surfaces. The
hydrodynamic force was then evaluated as
(17)
F
H
=2a
2

_
0
_

rz
sin +(P
s
+
zz
) cos
_
r=a
sin d.
The uid velocity, pressure, and stress integrals were all
evaluated in FLUENT using the nite element method.
3. Results and analysis
Finite element calculations were repeated on successively
ner grids to ensure numerical convergence. Most of the
simulations were performed for a model system in which
the pore half-width was 12.5 nm and the pore spacing was
200 nm, corresponding to a membrane with a surface pore
density of 12.5%. Typical uid streamlines in the single pore
system are shown as the solid curves in Fig. 2. The stream-
lines are plotted on dimensionless coordinates, where both x
and y have been scaled by the particle radius (a = 10 nm).
The streamlines in this system are completely independent
of the particle properties and position, since the Navier
Stokes equations were solved in the absence of the particle
and without inertial terms or electrical stresses. The converg-
ing ow into the pore accelerates the uid, with V
y
at x =0
increasing from0.001 m/s at y =10 to 0.012 m/s at the pore
exit (y =10). The x-component of the velocity attains its
maximum absolute value (V
x
= 0.0019 m/s) at y = 0 and
x =1.25.
The dashed lines in Fig. 2 represent contours with con-
stant values of the total electrostatic force evaluated from
Eq. (2) for a 10-nm particle with
p
=
m
=25.7 mV at
an ionic strength of 0.94 mM, with the latter corresponding
to a Debye length
1
= 10 nm. The electrostatic repul-
sion is greatest right near the membrane surface and de-
cays exponentially as h increases. The contour lines at large
values of y are horizontal, with the presence of the pore
having minimal effect on the electrostatic interactions. The
contour lines near the membrane bend down and into the
pore, with the repulsive force having components in both
the x- and y-directions. The x-component of the electrosta-
tic force vanishes at the pore centerline since the particle is
repelled from both walls of the pore.
Particle trajectories were evaluated by integration of
Eq. (1) using FLUENT 6.0 (Fluent Inc., Lebanon, NH). Cal-
culations were performed using successively smaller time
steps until a converged solution is achieved. Figure 3 shows
the particle trajectories (in the absence of any Brownian
forces) for a particle initially located at a dimensionless
position of x = 7 for three sets of conditions: (1) no
electrostatic interactions and no additional hydrodynamic
hindrance (K
f
= K
p
= 1), (2) with the electrostatic repul-
sion given by Eq. (2) using
p
=
m
= 25.7 mV and

1
=10 nm but with no hydrodynamic hindrance, and (3)
in the presence of electrostatic repulsion and with the hin-
drance factors given by Eqs. (9)(14). Results are shown
for V
f
=0.001 m/s, where V
f
is the uniform feed velocity
at y = 100 nm (y = 10), which corresponds to an elec-
trostatic force parameter of = 31. In the absence of any
electrostatic or additional hydrodynamic hindrance the par-
ticle follows the uid streamlines, moving into and through
the pore. The situation is very different in the presence of
electrostatic repulsion. In this case the particle begins by
following the uid streamline, but the trajectory deviates
fromthe streamline as the particle approaches the membrane
due to the electrostatic force. Signicant deviations are rst
seen at a dimensionless length of about 6, i.e., when the
particle is within six Debye lengths of the surface. The elec-
trostatic force causes the particle to move toward the pore
centerline (x = 0), with the particle becoming trapped at a
dimensionless distance y = 3.17 above the pore due to the
large value of the electrostatic force as indicated by the large
Fig. 2. Fluid streamlines (solid curves) and electrostatic force contours (dashed curves) for a =1 and
p
=
m
=25.7 mV. The labels on the contour lines
show the dimensionless electrostatic force F
e
/(4
0

r
a
2
p
).
M.-m. Kim, A.L. Zydney / Journal of Colloid and Interface Science 269 (2004) 425431 429
Fig. 3. Particle trajectories for (1) no drag modication or electrostatic forces, (2) no drag modication and electrostatic forces, and (3) additional hydrodynamic
hindrance and electrostatic forces. Conditions: a =1,
p
=
m
=25.7 mV, and V
f
=0.001 m/s.
Fig. 4. Effect of ltration velocity on the particle trajectories for a =1 and
p
=
m
=25.7 mV for (1) V
f
=0.001 m/s, (2) V
f
=0.01 m/s, (3) V
f
=
0.05 m/s.
value of . This is the equilibrium position for the particle,
i.e., the location at which the hydrodynamic drag force is ex-
actly balanced by the electrostatic repulsion. In the absence
of Brownian diffusion, all particles in the feed suspension
(with initial dimensionless position at y = 10) move to the
same nal equilibrium position. The additional hindrance
associated with the hydrodynamic interactions between the
particle and the surface causes only a small change in the
trajectory under the conditions examined in Fig. 3, with the
nal equilibriumposition shifted to y =2.80. The additional
hydrodynamic hindrance allows the particle to move slightly
closer to the membrane surface, which is a direct result of
the increase in the drag force.
In order to verify the analytical expressions used for the
hindrance factors, the detailed velocity and pressure proles
were determined by numerical solution of the NavierStokes
equations for the actual membrane system incorporating the
uid ow around a particle xed at y = 2.80, with no-slip
boundary conditions applied at both the particle and surface
boundaries. The hindrance factor was then evaluated using
Eq. (17), giving K
f x
=1.62. This value is only 2.5% larger
than the value (K
f x
= 1.58) determined using the analyt-
ical expression given by Eq. (12). Similar agreement was
obtained at other particle locations, demonstrating that the
analytical approximations developed from the literature re-
sults for systems with simpler geometry (Eqs. (9)(14)) pro-
vide a good description of the hydrodynamic hindrance in
the membrane system.
The effects of the ltration velocity on the particle tra-
jectories are examined in Fig. 4, with the simulations cor-
responding to values of the electrostatic force parameter of
=31, 3.1, and 0.62. As the ltration velocity increases, the
particle trajectory follows the uid streamline over a greater
distance since the hydrodynamic force remains much larger
than the electrostatic force until the particle gets closer to
the membrane surface. However, a 10-fold increase in the
ltration velocity (from 0.001 to 0.01 m/s) causes the equi-
libriumparticle dimensionless position to shift only from2.8
to 1.08 due to the exponential dependence of the electrosta-
tic force on the separation distance. At ltration velocities
above 0.049 m/s, which corresponds to < 0.64, the par-
ticle is able to overcome the electrostatic repulsion and en-
430 M.-m. Kim, A.L. Zydney / Journal of Colloid and Interface Science 269 (2004) 425431
ter the pore. This critical ltration velocity is similar to
the critical ux concept discussed by Field et al. [25]; the
membrane surface remains completely particle-free for all
ltration velocities below this critical value. At small val-
ues of , corresponding to very high ltration velocities or
weakly charged particles/membranes, the electrostatic force
becomes negligible and the particles simply follow the uid
streamlines through the membrane.
Bowen and co-workers [1416] have dened the criti-
cal ltration velocity as the uid velocity within the mem-
brane pore at which the net force acting on a particle, lo-
cated at a xed position above the pore, vanishes. Bowen
and Sharif [15] performed calculations for a spherical parti-
cle above a cylindrical pore, with the uid velocity and drag
force evaluated by numerical solution of the NavierStokes
equations for the particlepore system. It is difcult to com-
pare our results with those of Bowen and Sharif because of
the differences in the pore geometry (slit-shaped pore ver-
sus cylindrical pore) and the differences in the calculation
procedures (actual particle trajectories versus a full solution
for the hydrodynamic forces on a particle at a xed location).
However, Bowen and Sharif [15] do present results for a par-
ticle with
p
=
m
=25.7 mV and a = 3. Under these
conditions, a particle located 10 nm above the pore (i.e., at
y =1) experienced no net force at a critical ltration velocity
of 0.01 m/s. Corresponding calculations performed as part
of this study, with the feed velocity adjusted to give a max-
imum uid velocity at the pore exit of 0.01 m/s, yielded an
equilibrium particle position of slightly under y =1.5. This
value is somewhat larger than that determined by Bowen and
Sharif [15], which is a direct result of the difference in pore
geometries. The cylindrical pore geometry causes a much
greater reduction in the ow area, leading to a much larger
increase in the uid velocity as one moves from the bulk so-
lution into the pore. This causes the drag force to increase
very rapidly as the particle approaches the membrane, lead-
ing to a smaller value of the equilibrium height.
Although previous studies of particle trajectories in mem-
brane systems have neglected the effects of Brownian mo-
tion, Brownian forces will become comparable to the hydro-
dynamic and electrostatic forces as the particle size becomes
signicantly less than a micron. In this case, the particle
trajectories are no longer deterministic due to the random
nature of the Brownian forces. Figure 5 shows a series of par-
ticle trajectories at V
f
=0.001 m/s, which gives =31 and
Pe = 0.46 for a 10 nm particle (with the particle diffusion
coefcient evaluated from Eq. (7)). These simulations were
thus performed with very large electrostatic interactions and
with the Brownian forces being comparable to the hydrody-
namic forces (Pe on the order of 1). In the absence of Brown-
ian diffusion, the particle attains an equilibrium position at
y =3.2 where there is no net force on the particle. The ran-
dom nature of the Brownian force eliminates the possibility
of an equilibrium position. Instead, the Brownian force al-
lowed particle 4 to pass over the electrostatic force barrier,
moving into the pore and through the membrane even though
Fig. 5. Effect of Brownian forces on the particle trajectories for a = 1,
V
f
=0.001 m/s, and
p
=
m
=25.7 mV. Curve 1 shows results with
no Brownian forces.
the ltration velocity is below the critical ux. In contrast,
particle 3 passed out of the single pore system, moving back
into the feed suspension, while particle 2 was temporarily
captured in the inlet region (but would eventually either pass
into the pore or leave the system after a sufcient number of
time steps). Simulations at much larger values of Pe show a
much smaller effect of Brownian motion.
The results in Fig. 5 suggest that a certain fraction of the
particles will enter the pore under any given set of condi-
tions, with the remaining particles passing out of the single
pore system and back into the bulk suspension. This phe-
nomenon was examined in more detail by evaluating the
number of particles that enter the pore from 100 identical
simulations, with a random number generator used to obtain
a unique initialization of the Brownian forces. The particle
transmission was evaluated for several values of the dimen-
sionless surface potential
m
=
p
, where
m
=e
m
/kT ,
where e is the electronic charge and kT/e = 25.7 mV.
In each case, the particle transmission was dened as the
fraction of particles that enter the pore at a given set of con-
ditions. As seen in Fig. 6, the particle transmission increases
from zero at low ltration velocities to a value of one at
high ltration velocities due to the increase in the hydro-
dynamic drag force toward the membrane pore. The particle
transmission rst becomes measurable (S > 0.01) when the
electrostatic interaction parameter is =4.7, 12.2, and 14.4
at
m
= 0.1, 0.5, and 1.0, respectively. The value of is
smallest for the lowest surface potential because the Brown-
ian forces, which are greater for simulations at low
m
due
to the smaller values of the Peclet number at low V
f
, are
able to assist the particles in overcoming the electrostatic re-
pulsion barrier. The particle transmission reaches 100% at
high feed velocities corresponding to = 0.1, 0.4, and 0.8
at
m
=0.1, 0.5, and 1.0, respectively. These values of are
approximately equal to the values determined from the crit-
ical ltration velocity evaluated in the absence of Brownian
diffusion since the Brownian forces are relatively small due
to the large Peclet number for these simulations (Pe ranges
M.-m. Kim, A.L. Zydney / Journal of Colloid and Interface Science 269 (2004) 425431 431
Fig. 6. Particle retention evaluated from multiple simulations as a function
of the ltration velocity for several values of the dimensionless membrane
and particle potential with a =1.
from about 3 at
m
= 0.1 and S = 1 to more than 20 at

m
= 1.0 and S = 1). The particle transmission decreases
with increasing surface potential due to the corresponding
increase in the electrostatic repulsion. The breadth of the
sieving curve decreases with increasing
m
since the higher
feed velocities reduce the effects of the Brownian forces.
4. Conclusions
The simulations presented in this manuscript provide the
rst available results for the effects of the additional hydro-
dynamic hindrance, electrostatic interactions, and Brownian
forces on the particle trajectories in normal ow ltration.
In the absence of Brownian forces, the membrane remains
completely free of particles below a critical ltration veloc-
ity, analogous to the critical ux concept used to describe
fouling in many membrane systems. Below the critical l-
tration velocity, the particles attain an equilibrium position
above the membrane at a point where the hydrodynamic
drag force is exactly balanced by the electrostatic repulsion,
which occurs when the electrostatic interaction parameter
is approximately equal to one. This equilibrium location
is determined primarily by the magnitude of the electrosta-
tic force and the ltration velocity, with the hydrodynamic
hindrance factors causing only a small displacement in the
equilibrium position.
Brownian forces have a large effect on the particle tra-
jectories at small to moderate values of the Peclet number,
allowing particles to enter the pore even under conditions in
which the electrostatic repulsion is greater than the hydrody-
namic drag force. This effect has been largely unappreciated
in previous studies of particle motion in membrane systems,
even in applications where fouling is due to the deposition
of relatively small colloids for which the Brownian forces
are substantial. The probability of a particle entering the
pore was evaluated from repeated particle trajectory calcu-
lations accounting for the random nature of the Brownian
force. The fraction of particles entering the pore increased
with increasing ltration velocity, with this dependence be-
ing very similar to that observed in membrane systems due
to the concentration polarization phenomenon [1]. However,
in this case there is no direct accumulation of particles at
the membrane surface. Instead, the increase in transmission
arises because the hydrodynamic drag force is able to over-
come the electrostatic repulsion, pushing the particle closer
to the membrane surface and making it easier for the Brown-
ian forces to overcome the energy barrier. Thus, considerable
care must be exercised in interpreting the ux dependence of
the particle sieving coefcient for systems in which electro-
static, hydrodynamic, and Brownian forces all have a signif-
icant effect on particle transport.
Acknowledgments
Support for this work was provided in part by Grant
CTS-0091552 from the National Science Foundation. The
authors also thank Professor Ali Borhan for assistance with
FLUENT.
References
[1] L.J. Zeman, A.L. Zydney, Microltration and Ultraltration: Princi-
ples and Applications, Dekker, New York, 1996.
[2] R.Y. Ning, Desalination 151 (2003) 67.
[3] T. Ikeda, H. Muragishi, R. Bairinji, T. Uemura, Desalination 98 (1994)
391.
[4] X.W. Wang, Agr. Eng. 25 (1994) 53.
[5] Y.F. Maa, C.C. Hsu, Biotechnol. Bioeng. 50 (1996) 319.
[6] R. Iranpour, Water Environ. Res. 70 (1998) 1198.
[7] M. Otaki, K. Yano, S. Ohgaki, Water Sci. Technol. 37 (1998) 107.
[8] F.W. Altena, G. Belfort, Chem. Eng. Sci. 39 (1984) 343.
[9] G. Belfort, N. Nagata, Desalination 53 (1985) 57.
[10] J.R. Otis, F.W. Altena, J.T. Mahar, G. Belfort, Exp. Fluids 4 (1986) 1.
[11] Z. Dagan, S. Weinbaum, R. Pfeffer, Chem. Eng. Sci. 38 (1983) 583.
[12] J. Kao, Y. Wang, R. Pfeffer, S. Weinbaum, J. Colloid Interface Sci. 121
(1988) 543.
[13] P. Schmitz, D. Houi, B. Wandelt, J. Membrane Sci. 71 (1992) 29.
[14] W.R. Bowen, A.N. Filippov, A.O. Sharif, V.M. Starov, Adv. Colloid
Interface Sci. 81 (1999) 35.
[15] W.R. Bowen, A.O. Sharif, Chem. Eng. Sci. 53 (1998) 879.
[16] W.R. Bowen, A.O. Sharif, Colloids Surf. A 201 (2002) 207.
[17] A. Li, G. Ahmadi, Aerosol Sci. Technol. 16 (1992) 209.
[18] A.J. Goldman, R.G. Cox, H. Brenner, Chem. Eng. Sci. 22 (1967) 653.
[19] S. Wakiya, J. Phys. Soc. Jpn. 8 (1953) 254.
[20] M.C. Drumond, W.M. Deen, J. Biomed. Eng. 117 (1995) 414.
[21] H. Faxen, Ann. Phys. 68 (1922) 89.
[22] Y. Pawar, J.L. Anderson, Ind. Eng. Chem. Res. 32 (1993) 743.
[23] W.M. Deen, AIChE J. 33 (1987) 1409.
[24] P. Ganatos, S. Weinbaum, R. Pfeffer, J. Fluid Mech. 99 (1980) 739.
[25] R.W. Field, D. Wu, J.A. Howell, B.B. Gupta, J. Membrane Sci. 100
(1995) 259.