Journal of Membrane Science 160 (1999) 255265

Simulation of an ultraltration process of bovine

serum albumin in hollow-ber membranes
Argimiro R. Secchi*, Keiko Wada1, Isabel C. Tessaro2
Departamento de Engenharia Qumica, Universidade Federal do Rio Grande do Sul,
Rua Eng. Luiz Englert, s/n - Zip Code 90040-040 Porto Alegre, Brazil

Accepted 4 March 1999

Abstract

This work presents a numerical simulation of an ultraltration process of bovine serum albumin in solution, using hollow-
ber membranes. Such membranes are constituted of tiny polymer cylinders disposed in a tube-and-shell arrangement. The
concentrate ows through the interior of the bers and the pure solvent is recovered in the shell, assuming perfect solute
rejection. In modeling the process, the ow of concentrate inside the bers was considered to be laminar, with constant
density, viscosity and solute diffusivity. Axial diffusion and angular effects were ignored. The model combines the effect of
concentration polarization and adsorption, which are the two main limiting phenomena in ultraltration processes. The
pressure on the shell side was considered constant and inside the bers a linear pressure prole, dependent on the axial
position, was adopted. The solution of the problem was achieved with the method of orthogonal collocation, with adequate
choice of the weight function in the radial direction. In the axial direction, a nite-difference method was used. The numerical
results were compared with experimental data available in the literature. # 1999 Elsevier Science B.V. All rights reserved.

Keywords: Hollow-ber membranes; Ultraltration; Membrane process simulation; Concentration polarization; Adsorption

1. Introduction permeate is collected on the outside of the bers; (ii)

For ultraltration processes, the commercial mod-
ules often used are hollow-ber and spiral-wound.
Hollow-ber modules are constituted of tiny polymer
cylinders disposed in the tube-and-shell arrangement,
as shown in Fig. 1. The modules contain about 50
3000 bers and can be operated in two different ways:
(i) the feed solution enter inside the bers, whereas the
*Corresponding author. Tel.: +55-51-316-3110; fax: +55-51-
316-3277; e-mail: arge@enq.ufrgs.br
1
keiko@enq.ufrgs.br
2
isabel@enq.ufrgs.br
the feed solution is introduced on the shell side and the
permeate passes into the bers.
The main advantage of hollow-ber modules in
ultraltration processes is the high packing density,
followed by easiness in the operation and maintenance
and low energy consumption. The performance of this
process depends on the characteristics of the module
and the operating conditions. The characteristics of
the module are the nature of the polymeric material,
the structure of the bers, the porosity and pore size
distribution, the diameter, thickness and length of the
bers, and the packing density. The operational para-
meters are the following: the characteristics of the

0376-7388/99/$ see front matter # 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 6 - 7 3 8 8 ( 9 9 ) 0 0 0 9 1 - 5
256 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265
Fig. 1. Schematic representation of the hollow-fiber module.
solution which depend on the chemical nature of the
solute, the solvent and solute concentration; physical
properties as density, viscosity and osmotic pressure
(which are functions of the solute concentration); feed
ow and transmembrane pressure drop.
In the present work the experimental data were
obtained from the work of Ma et al. [2], where an
Amicon model H1P10 hollow-ber cartridge (Ami-
con, Danvers, MA) was used in their experiments.
Each of the 1000 bers in the cartridge was 200 mm
I.D., anisotropic, non-cellulosic membranes rated for
10 000 molecular weight cutoff, with the feed solution
introduced inside the bers, and the permeate recover
in the shell side. As reported by Ma et al. [2], the
measured active length of the bers between the epoxy
headers was 170.2 cm. In this way of operation, the
pressure drop for the solution along the bers inu- C; R; 0 1
2. Mathematical model
Under the operating conditions of the process in
study, with pressure drop around 20 kPa, permeation
velocity close to 103 cm/s and axial velocity, inside
the bers, about 5 cm/s, the concentrate ow is essen-
tially laminar (with Reynolds number about 10). For
the development of the mathematical model, the phy-
sical properties of the solution, such as density, visc-
osity and solute diffusivity are assumed to be constant.
Axial diffusion and angular effects are ignored. The
pressure on the shell side was considered constant and
for the pressure inside the bers a linear prole,
dependent on axial position, was adopted. This last
consideration is due to the experimental observations
of Ma et al. [2].
In dimensionless form and in cylindrical coordi-
nates, the equation of continuity combined with the
convectivediffusive ux equation for the solute
becomes
 
@C 1 @ @C @C @C
R V U (1)
@ PeR @R @R @R @y
with the following boundary conditions
(2)

ences signicantly the process, because the pressure @C
0 (3)
difference among the concentrate solution and the @R R0
permeate is the driving force for the permeation. 
Another factor of fundamental importance in @C
Vm yC; 1; yPe (4)
the performance of ultraltration processes is the @R R1
phenomenon of concentration polarization. This phe- and initial condition
nomenon refers to the reduction in the permeation rate
due to the existence of high solute concentration at C0; R; y 1 (5)
the membrane surface, because the solute is retained Following the work of Ma et al. [2], the velocity
by the membrane whereas the solvent permeates proles were based on the results obtained by Yuan
through the membrane. This zone with high concen- and Finkelstein [5]. Yuan and Finkelstein used a
tration constitutes a resistance to permeation, perturbation technique to solve the NavierStokes
which favors the phenomenon of adsorption, and equations to obtain the steady-state velocity proles
increases the osmotic pressure of the solution, for laminar ow through porous tube. For small values
which, in turn, reduces the effective transmembrane (1) of the dimensionless number
pressure.
rm v m w
The main objective of this work is the modeling and  (6)
simulation of an ultraltration process of bovine w
serum albumin, using a hollow-ber module where the axial and radial velocity proles can be expressed
the effects of pressure drop and osmotic pressure on as (in the studied cases (<2103)
the effective transmembrane pressure are of funda-
mental importance. UR; y 21 2Vm y1 R2 (7)
 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265 257

VR; y Vm 2R R3 (8) where the mass ratio, !, can be expressed as a function

of the solute concentration at the membrane surface,
where Vm is the dimensionless membrane permeation
cm, by the relationship
velocity, which is a function of the pressure difference
between the internal ber side (Pm), the shell side (Pc), cm w
!cm (13)
and the osmotic pressure (m) at the membrane sur- 1 cm  s
face. The permeation velocity is evaluated from the where w1.0 cm3/g and s0.733 cm3/g are the
expression specic volume of the solvent and solute, respectively.
vm y KPm Pc m (9) For the consideration of linear pressure prole
along the ber length, the pressure drop through the
The membrane permeability constant, K, is usually membrane, PmPc, in each axial position can be
determined from permeation ux versus pressure drop expressed by
tests with pure solvent. The use of the clean membrane z
permeability alone does not take into account possible Pm z Pc
Pro
Pax (14)
L
effects of the solute on the membrane permeability,
such as pore plugging, and other surface effects such where
Pax is the pressure drop along the ber and
as solute adsorption and gel formation, which can be
Pro is the pressure drop through the membrane at the
exacerbated by concentration polarization. entrance of the ber (
Pro
Pax
Pre).
In this work a resistance model proposed by Gekas
et al. [1] was adopted, and it is represented by the
following equation 3. Solution of the model

1 The solution of the model was obtained by the use

m
ad (10)
K of orthogonal collocation in the radial direction and
nite differences in the axial direction. This differs
where
m is the resistance of the membrane due to the
from the work of Ma et al. [2], which used nite
permeation of pure solvent and
ad is the resistance
differences in non-uniform mesh in both radial and
due to solute adsorption. Considering the adsorption
axial directions, needing at least 72 points in the radial
as an instantaneous phenomenon, then
ad is the
direction and 30 points in the axial direction. The
resistance due to the adsorption at the equilibrium
resultant system of algebraicdifferential equations
conditions, that can be expressed by isotherms of
was solved by multiple-steps integration, with variable
adsorption. The model used, that presented good
step and variable order BDF (Backward Differentia-
agreement with experimental data, was the Langmuir
tion Formula), by the use of the computational pack-
isotherm
age DASSLC of Secchi and Pereira [4].
k1 cm Due to the symmetry of the system in the radial

ad cm (11)
1 k2 cm direction, a variable change was proceeded in the
radial coordinate, in the way that xR2. This was
where the constants k1 and k2 are estimated with done before the application of the method of ortho-
experimental data available. gonal collocation. In this new system of coordinates,
The osmotic pressurecomposition relationship for Eq. (1) assumes the form
the bovine albumin solution, was determined from the  
work of Scatchard et al. [3]. In the operating condi- @C 4 @ @C @C @C
x W U (15)
tions of the experimental data (of Ma et al. [2]), @ Pe @x @x @x @y
temperature and pH values were, respectively, 378C
Eqs. (7) and (8) become, respectively,
(310.15 K) and 6.7. The empirical relationship for the
osmotic pressure, in kPa, has the following polyno- Ux; y 21 2Vm y1 x (16)
mial form Wx; y 2Vm 2x x2 (17)
2 3
m ! 37:20! 243:98! 487:96! (12) and the boundary conditions given by Eqs. (3) and (4)
258 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265

become

@C 
finite (18)
@x x0

@C  Pe
Vm yC; 1; y (19)
@x x1 2
The nite value in Eq. (18) results
pfrom the chain
rule applied to Eq. (3): @C=@R 2 x@C=@x, and
at x0 (or R0) it is sufcient that (@C/@x) be nite.

3.1. Selection of the weight function

Fig. 2. Approximation of the S(x) functions with the weight
The method of orthogonal collocation is based on function w(x). The subscript 1 refers to the use of Eq. (1).
the family of the Jacobi polynomials, where the
parameters  and  dene the weight function asso- curves by the least-square method, the following
ciated with this family equations were obtained as better approximations
wx x 1 x (20) 0 (25)
For non-linear problems, the major difculty to use 
1 0:76PeVmj ; PeVmj < 4
the method of orthogonal collocation is the proper  (26)
PeVmj ; PeVmj > 4
choice of the weight function. To help in this choice,
the following methodology was adopted: (i) only the As Vmj varies in the range of [0,1],  can assume
steady-state solution was considered; (ii) discretiza- values of 1 to Pe. It is interesting to notice that, if
tion was made by nite differences in the axial Eq. (1) is used without variable change, the values of
direction, resulting in the following system of ordinary the parameters would be
differential equations  0:87; (27)
 
4 d dCj dCj Cj Cj1  0:69 0:29Pe Vmj (28)
x Wj Uj (21)
Pe dx dx dx
yj
Fig. 2 shows how the weight function of the Jacobi
where the term Cj1 can be seen as an external polynomials has an enormous capacity to approximate
disturbance for the equation j, because there is no the functions S(x)Sj(x)/Mj.
axial diffusion; (iii) placing the problem, without From the results of Eq. (26), it would be necessary
external disturbance, in the auto-adjoint form, it to use an appropriate weight function (and conse-
results in quently, a Jacobi polynomial) for each axial position.
   2   Besides, Vmj is unknown, and could be extrapolated by
d x dCj
x exp PeVmj x Sj xCj (22) the values calculated in previous positions. However,
dx 4 dx
this calculation can eliminate the advantage of the use
where of the orthogonal collocation in relation to other
  2  discretization techniques.
x
Sj x Mj exp PeVmj x 1 x (23) It was decided, therefore, to use xed values for the
4
weight function parameters for the whole extension of
and the ber, with the commitment of needing a slightly
Pe larger number of collocation points. As shown in
Mj 81 2Vmj yj (24) Fig. 3, the value of 2 seems to be appropriate

yj
for the operating conditions simulated. It is observed
The last step, (iv), is to nd  and  to approximate that high values of  tend to better approximate the
the weight function to the function Sj(x)/Mj. Fitting the initial region of the ber (Vmj1), but they lose quality
 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265
Fig. 3. Dimensionless axial steady-state profile for concentration
at the center of the fiber for several weight functions with n10
(precise values are obtained with n15).
at the end of the ber (Vmj0). The inverse happens for
small values of . When using values for the para-
259
k1 1:707  0:013  105 kPa s cm2 =g
k2 0:82  0:05 cm3 =g
The results obtained for 10 different experimental
conditions are shown in Table 1. It can be observed
that the results predicted by the model used in this
work (listed in the last column) present a good agree-
ment with the experimental data, and better to the
model without adsorption (listed in the sixth column).
The evolution of the concentration prole in both
axial and radial directions for several times, using the
data of the experiment number 7 (Table 1) is shown in
Fig. 4. It can be observed that the steady-state regime
is reached in a short time, about 12 s. The effect of
concentration polarization is observed by the sudden
increase of the solute concentration from the center of
the ber to the membrane surface, reaching a max-
imum in the exit of the ber, in disagreement with the
results predicted by Ma et al. [2], where the maximum
is located inside the ber (that corresponds to a non-
stationary prole, as illustrated in Fig. 4). It is worth
noting that Ma et al. [2] used the method of Crank
Nicholson for the numerical integration.
meter  different than zero, a larger deterioration of
the approximation occurs.
4. Results and discussion
The results of the present simulation will be com-
pared with experimental data available in the litera-
ture, and with the results predicted by Ma et al. [2]. It
is to be reminded that Ma et al. [2] used a nite-
difference method with non-uniform mesh in both
radial and axial coordinates, and did not consider
adsorption effects. In the present simulations a
nite-differences method was used in the axial direc-
tion with a mesh of 40 points (being 20 small incre-
ments over the rst 10% of the ber length and the
other 20 large increments over the remaining 90%).
For the radial direction the Jacobi polynomial was
In Figs. 5 and 6 the distribution of the velocity eld
is presented along the radial and axial coordinates, for
two xed values of pressure drop in the exit of the
concentrate side. Both gures show that the radial
velocity at membrane surface decreases in the axial
direction (given by the size of the arrows). This effect
is due to the pressure drop in the axial ow and the
combined effect of concentration polarization and
adsorption that cause the increase of the osmotic
pressure.
Fig. 6 illustrates the phenomenon of reverse ow,
represented by the area inside of the curve drawn at the
upper right position, which is caused by the osmotic
pressure which becomes dominant in the effective
transmembrane pressure.
The variation of the permeate rate (left axis) and the
recovery, 100 qp/q, (right axis), as a function of the
feed rate, is shown in Fig. 7. The dotted line considers
used with n9, 2 and 0.
For the Langmuir adsorption isotherm, the para-
meters
m, k1, and k2 were estimated with the avail-
able experimental data, and the following values were
obtained
m 0:1764  0:0027  105 kPa s=cm
the pressure drop in the axial direction independent of
the entrance velocity in the ber and equal to 6.80 kPa.
The solid line considers the pressure drop in the axial
direction dependent on the entrance velocity by

Pax 1:59uo kPa. The comparison between the
curves shows the importance of this effect for high
feed rates.
260 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265

Table 1
Experimental data from literature [2] and predicted by the model

Exp. No.
Pax (kPa)
Pre (kPa) q (cm3/s) qexp:

p (cm3/s) qp a (cm3/s) qp (cm3/s)

1 6.13 12.40 1.24 0.40 0.42 0.41

2 5.33 12.27 0.99 0.38 0.36 0.38
3 4.93 13.33 0.84 0.39 0.34 0.38
4 4.27 19.47 0.83 0.46 0.43 0.44
5 5.73 24.00 1.20 0.58 0.58 0.56
6 7.07 25.46 1.47 0.58 0.57 0.61
7 6.80 20.66 1.35 0.54 0.51 0.54
8 6.67 18.40 1.35 0.52 0.49 0.52
9 6.80 12.67 1.37 0.43 0.43 0.43
10 4.93 24.66 0.85 0.50 0.50 0.50

Sum of square deviation (103): 6.10 1.90

a
Predicted by Ma et al. [2].

The form of the pressure drop equation was deter-
mined considering a HagenPoiseuille ow, which
predicts the proportionality between the pressure drop
and the mean ow velocity. The equation for pressure
drop as a function of the entrance velocity was
obtained by a linear tting with the experimental data,
with a correlation coefcient of 0.996.
Fig. 8 shows the variation of the solute concentra-
tion at the membrane surface along the ber length.
The abrupt increase of concentration for small values

Fig. 4. Dimensionless evolution of the solute concentration profile inside of the fiber.
 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265 261

Fig. 5. Velocity field for transmembrane pressure
Pre20.66 kPa Fig. 7. Permeate rate and recovery as a function of the feed flow
with dimensionless coordinates. rate for two models of the pressure drop in the axial direction,
Pax
(
uo is the bulk axial velocity at z0).

of Z is due, essentially, to the high permeation rate of

solvent. However, as Z increases, the permeation rate The curves shown in the Figs. 810 were obtained
is reduced since the difference of effective transmem- by keeping the overall pressure drop along the ber
brane pressure decreases. Even with the reduction in constant. In Fig. 9, the variation of the radial velocity
the permeation rate along Z the solute concentration at at the membrane surface with the axial position shows
the membrane surface continues increasing for the the effect of the effective transmembrane pressure,
curves with
Pre>100 mm Hg (13.33 kPa). This hap- that takes into account the effects of the pressure drop
pens due to the concentration process. However, for
low transmembrane pressure, characterized in this
gure by
Pre<100 mm Hg (13.33 kPa), the osmosis
phenomenon and the radial diffusion become signi-
cant.

Fig. 6. Velocity field for transmembrane pressure
Pre0 with Fig. 8. Dimensionless concentration versus dimensionless axial
dimensionless coordinates. coordinate for different values of transmembrane pressure.
262 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265
Fig. 9. Dimensionless radial velocity as a function of the
dimensionless axial coordinate for different values of the
transmembrane pressure.
in the axial ow and the osmotic pressure. The last
effect is shown more clearly in Fig. 10.
For small and moderate transmembrane pressure
differences, as indicated by the values of
Pre0 and
Fig. 10. Osmotic pressure as a function of the dimensionless axial
coordinate for different values of the transmembrane pressure.
Fig. 11. Permeate rate and recovery versus transmembrane
pressure.

Pre10 mm Hg (1.33 kPa), the osmosis phenom-
enon can be better observed, i.e., solvent ows from
the permeate side to the concentrate side, character-
ized by the negative velocities of permeation as shown
in Fig. 9.
Fig. 11 shows the inuence of the transmembrane
pressure on the permeation rate and on the recovery. It
is observed that for regions with low
Pre the varia-
tion is approximately linear, in agreement with the
HagenPoiseuille equation applied to ow through
pores. As
Pre increases, the variation rate is reduced
approaching a condition where the increase of the
permeation rate by the increase of
Pre is much less
signicant, due, mainly, to the effect of concentration
polarization.
Figs. 1214 were obtained considering the equation

Pax 1:59uo , but keeping constant the value of

Pre. This means that the pressure difference through
the membrane at the entrance of the ber is increasing
as the feed ow rate increases. These gures represent
the effects of the concentration polarization and the
osmotic pressure for several values of feed ow rate.
As can be observed in Fig. 12, as the ow velocity
increases, the thickness of the polarization layer is
reduced, and Cm decreases until a ow rate of about
2 cm3/s. The smooth increase of concentration at the
membrane surface for Z>0.5 is due to the increase of
the bulk solute concentration in the tube side. For feed
ow rates varying from 4 to 8 cm3/s it is veried that
the concentration at the membrane surface reaches a
maximum upstream of the hollow-ber exit. This
 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265
Fig. 12. Dimensionless concentration at the membrane surface as a
function of the dimensionless axial coordinate for different feed
flow rates.
effect can be explained by the increase of the polar-
ization in the entrance of the ber due to a larger
pressure difference through the membrane in this
region. Ma et al. [2] also predicted this phenomenon
but at lower feed ow rates.
Fig. 13. Dimensionless radial velocity as a function of the
dimensionless axial coordinate for different feed flow rates.
263
Fig. 14. Osmotic pressure as a function of the dimensionless axial
coordinate for different feed flow rates.
The reduction of concentration after the maximum
point is due to the fast reduction of the effective
transmembrane pressure difference. For high feed
ow rates, the increase of concentration in the con-
centrate stream, when passing through the ber, is not
very signicant. Thus, the concentration gradient in
the radial direction is larger for higher ow rates,
favoring the diffusion of the solute at the membrane
surface to the bulk stream. This effect contributes
essentially to the reduction of the solute concentration
at the membrane surface as shown by the curve
q8.0 cm3/s.
In Fig. 13, the area under the curve vm versus Z is
proportional to the permeation rate. It can be observed
that as the ow increases the permeation rate
increases.
5. Conclusions
A low pressure, hollow-ber ultraltration system,
considering total rejection of the solute, was simu-
lated. Effects of pressure drop on the concentrate
stream, adsorption and osmotic pressure on the per-
meation rate were considered.
The adsorption term incorporated in the model had
a reasonable improvement in the predicted results.
264 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265

This indicates the need to consider the polarization U dimensionless axial velocity, u=uo
effect together with the effect of adsorption for the u axial velocity (cm/s) R
r
problem in study. Although the model need two more u bulk axial velocity, r22 0m rudr (cm/s)
m
adjustable parameters, the inclusion of the resistance V dimensionless radial velocity (v/vmi)
due to solute adsorption in the membrane permeability v radial velocity (cm/s)
is of fundamental importance to better elucidate this W 2RV
phenomenon. New experiments are being planned in w weight function
order to verify the actual contribution of the surface x R2
effects over the membrane permeability. y dimensionless axial coordinate, zvmi =uo rm
The results can be considered very close to the
y axial discretization interval
experimental data available in the literature. However, Z dimensionless axial coordinate (z/L)
it is been planned to build a pilot equipment for tests, z axial coordinate (cm)
from where more experimental data will be obtained
for comparison effects.
The use of orthogonal collocation in the radial Greek letters
direction reduced the two-dimensional mesh of
2160 points, used by Ma et al. [2] with nite differ-  weight function parameter
ences, to a two-dimensional mesh of 360 points. The  weight function parameter
methodology used to select the weight function shows  dimensionless number (rmvor/m)
to be very useful to determine an adequate polynomial  viscosity (g/cm s)
approximation with reduced order for non-linear  osmotic pressure (kPa)
PDEs. This methodology has been successfully  density (g/cm3)
applied to other systems by the authors.  dimensionless time (tvmi/rm)
 specific volume (cm3/g)

resistance to permeation (kPa s/cm)
6. Nomenclature ! mass ratio, g protein/g solvent

C dimensionless solute concentration (c/co)

c solute concentration (g/cm3) Subscript
D solute diffusivity in the solvent
(9.0107 cm2/s) ad adsorption
K permeability constant (cm/s kPa) ax global axial parameter
k1 Langmuir isotherm constant (kPa s cm2/g) c value in the shell side
k2 Langmuir isotherm constant (cm3/g) e value in the exit of the fiber
L length of the fiber (cm) i initial value
M axial discretization constant j index of the axial discretization
N number of fibers m value at the membrane surface
n degree of the Jacobi polynomial o value in the fiber entrance
P fluid pressure (kPa) r transmembrane parameter

P pressure drop (kPa) s solute
Pe initial Peclet number (vmirm/D) w solvent
Pn Jacobi polynomial of degree n
q feed flow rate, Nrm 2
uo , (cmR3/s)

L
qp permeate flow rate, 2Nrm 0 vm dz (cm3/s)
References
R dimensionless radial coordinate (r/rm)
S radial coordinate function [1] V. Gekas, P. Aimar, J.P. Lafaille, V. Sanchez, Simulation
r radial coordinate (cm) study of adsorptionconcentration polarization interplay in
t time (s) protein ultrafiltration, Chem. Eng. Sci. 48 (1993) 2753.
 A.R. Secchi et al. / Journal of Membrane Science 160 (1999) 255265 265

[2] R.P. Ma, C.H. Gooding, W.K. Alexander, A dynamic model
for low-pressure, hollow-fiber ultrafiltration, AIChE J. 31
(1985) 1728.
[3] G. Scatchard et al., Preparation and properties of serum and
shapes proteins. VII: osmotic equilibrium in concentrated
solutions of serum albumin, J. Am. Chem. Soc. 68 (1946)
610.
[4] A.R. Secchi, F.A. Pereira, DASSLC: User's Manual - Version
2.2. ENQLIB, http://www.enq.ufrgs.br/enqlib/numeric/,
1997.
[5] S.W. Yuan, A.B. Finkelstein, Laminar pipe flow with
injection and suction through porous wall, Trans. ASME 78
(1956) 719.