Simple model of foam drainage

M. A. Fortes, and S. Coughlan

Citation: Journal of Applied Physics 76, 4029 (1994); doi: 10.1063/1.357350

View online: https://doi.org/10.1063/1.357350
View Table of Contents: http://aip.scitation.org/toc/jap/76/7
Published by the American Institute of Physics
Simple model of foam drainage
M. A. Fortes and S. Coughlan
Departarnento de Engenharia de Materiais, Institute Superior Tknico, 1000 Lisboa, Portugal
(Received 15 December 1993; accepted for publication 27 June 1994)
A simple model of foam drainage is introduced in which the Plateau borders and quadruple junctions
are identified with pools that discharge through channels to pools underneath. The flow is driven by
gravity and there are friction losses in the exhausting channels. The equation of Bernoulli combined
with the Hagen-Poiseuille equation is applied to describe the flow. The area of the cross section of
the exhausting channels can be taken as a constant or may vary during drainage. The predictions of
the model are compared with standard drainage curves and with the results of a recently reported
experiment in which additional liquid is supplied at the top of the froth.

I. INTRODUCTION gas in the foam) and g is the acceleration due to gravity. ‘We
define the ratio of the absolute values of the two pressure
In a gravitational field the liquid phase of a three- gradients as
dimensional soap froth drains vertically, originating a gradi-
ent in the liquid fraction, with the amount of liquid increas- KE---y -41/r) (1)
ing from top to bottom. The froth is formed by polyhedral /z I dz I ’
cells which contact at faces (two per face), at edges or Pla-
teau borders (three cells at an edge) and at vertices or qua- The radius r can be related to the area S of the cross section
druple junctions (four cells at a vertex). The liquid phase is of the Plateau border, a more accessible experimental ‘quan-
distributed in the faces, edges and vertices of the cells but is tity, by
a simply connected phase in the sense that two points in the
S= m2, (21
liquid can be joined by a continuous curve lying entirely in
the liquid. Drainage can therefore be regarded as liquid flow where CYis a constant that depends on the assumed shape of
in this interconnected network of faces, edges, and vertices, the Plateau borders. We shall take cu=O.161 as obtained by
Flow through a film transfers liquid to Plateau borders, while Kann’ for a border. with triangular cross section. Then
flow through the Plateau borders. transfers liquid from one
vertex to another.
Most previous models of drainagere3 assume that Pla- K=o*2 IK!gI.
; s-1/2
(3)
teau border flow dominates with virtually all liquid circulat-
ing in the Plateau borders; however, as recently discussed by Using y=50 mJ m’-‘, p= lo3 kg m-3, and gE9.8 m s -‘, we
Narsimhan3 this may not be true when the film thickness is obtain ~=0.03 for Id lnSldz]=O.l cm-’ (a 10% increase in
relatively large, for example, in the initial stage of drainage S within 1 cm) and S=O.l mm’. The capillarity contribution
of a foam. In the model introduced in the~.presentarticle we to the driving force will only be significant for very thin
consider the two contributions to now. Plateau borders and in the final stages of drainage. We thus
Another characteristic of previous models is that they discard the capillary contribution and discuss drainage driven
ignore the discrete nature of. the foam structure and use av- exclusively by gravity.
erage foam properties that vary continuously with z (the co- Experimentally foam drainage can be studied by simply
ordinate in the direction of gravityj and time t. The model measuring the volume V, of liquid that drains at the bottom
that we introduce is a discrete model in which the role of of a tube containing the froth.4 The Vd(t) curves (where t is
films, Plateau borders, and also of vertices is explicitly time) are typically of the type shown in Fig. 1, with V,
treated, and therefore allows a better visualization of the ac- tending to a limiting value. Simultaneously with drainage,
tual drainage process. the foam can age or coarsen due to diffusion of the gas from
Gravity will be assumed as the dominant driving force the smaller to the adjacent larger cells, with elimination of
for drainage. As drainage proceeds, a gradient of film and the smaller cells. It may also happen, particularly at the later
Plateau border cross-sectional areas develops which gives stages, when the cell walls become very thin, that ruptures
rise to a pressure gradient in the liquid, opposing gravity. occur. We neglect these complications and assume that the
This gradient has been taken into account in the models in geometry and topology are unchanged during drainage, ex-
which drainage is assumed to be controlled by Plateau border cept for the amount of liquid in the cell walls, Plateau bor-
flow.‘Z3 If r is the radius of curvature of the surface of a ders, and quadruple junctions.
Plateau border, the pressure gradient is fiId( l/r)/&], where A recent experiment related to drainage and reported by
y is the liquid surface tension. This ignores any effect of the Weaire et u1.s is also of interest in the context of this article.
vertices and applies to a fictitious continuous Plateau border The experiment consists of supplying, at the top of the froth,
running vertically. The pressure gradient due to gravity is pg the liquid of which the froth is made. A fairly sharp interface
where p is the liquid density (we neglect the density of the between wet (liquid rich) and dry froth is then formed which

J. Appl. Phys. 76 (7), 1 October 1994 0021-8979/94/76(7)/4029/7/$6.00 Q 1994 American Institute of Physics 4029
 12
I
- IO-
“E
2 8
m Channel
:
6 hol i

4
-Ai--
2
/ l!i
0;.
250
1
500 750 1000 t
Time Is) 3

FIG. 1. Experimental drainage curve for a froth prepared by blowing nitro-

gen through an aqueous detergent solution. Average size of bubbles 0.3 cm.
A cylindrical container 5cm diameter and 30 cm long was used.

moves downwards at a constant velocity. Furthermore, this N

velocity was found to be proportional to Q I”, where Q is the
Aow of liquid supplied at the top.
The rate of drainage ~of a. froth is affected by various
parameters such as average bubble size, froth density (i.e.,
FIG. 2. The model system consistingof various pools in a column that
the volume fraction of liquid), and by the basic properties of discharge through channels to the pools underneath.
the liquid in the froth, such as its viscosity, surface tension,
etc. The effect of these parameters has been discussed by
various authors using different approaches. For example,
Krotov’ and Kann7 discuss their effect on the global perme- gas pressure is, on average, independent of the level in the
ability of a foam, while other authorste3 have elaborated froth. Again we ignore any pressure differences due to wall
more detailed models of drainage from which the effect of curvature change along a Plateau border.
those parameters can be predicted. It will be assumed that the vertical velocity of the liquid
Unfortunately there is very limited experimental data on in a pool is negligeable compared to the velocity u i of the
the effects of foam parameters on the drainage rate. We escaping liquid through a channel. This corresponds to as-
therefore give special relevance, as a test of our model, to the suming a,eAi. The fact that the interconnecting channels
direct experimental drainage curves and to the experimental (cell walls and edges) in a froth are randomly inclined will
observations of Weaire et al.,5 both of which can be simu- be taken into account by identifying Lj with the vertical dis-
lated with the model. tance between pools (edges and quadruple junctions, respec-
tively). We shall assume that Li is independent of i and of
the order of the average edge length. It was shown by Kann’
II. MODEL
that the rate of drainage through an arbitrary cross section of
the foam column can be obtained if the real structure of
It will be assumed that the liquid in a draining froth randomly oriented channels is replaced by a system of verti-
flows from a Plateau border to other borders, located under- cal channels (with the same cross section as in the real froth),
neath, to which it is connected by cell faces. In 3D froths the number of which is l/3 of the true number of channels in
(but not in a 2D froth with horizontal borders) there is also the actual froth. The number of columns of pools and chan-
flow along the Plateau borders, from a quadruple junction to nels in the model is then l/3 of the number of channels in the
quadruple junctions underneath. Flow is driven by gravity. actual froth. It should be noted that the relation between the
We neglect the 3D structure of the foam and use essentially a number of films (faces) and the number of Plateau borders
one-dimensional model froth, formed by parallel columns, (edges) depends on the topology of the froth.
each of which has N pools vertically arranged (see Fig. 2). As drainage occurs, the area ai of the exhausting chan-
The liquid in pool i falls. to the pool underneath (i+l) nels will in general decrease. Initially we assume ai is con-
through a channel of length Li and cross-sectional area ai. stant; we then elaborate the model to take into account the
The cross-sectional area of pool i is Ai and the height of change in ai during drainage. In either case it is assumed that
liquid in it is hi. The pools represent either the Plateau bor- the channel walls are rigid, in the sense that the liquid is at
ders or the quadruple junctions and the channels represent rest at the wall. The effect on drainage of nonrigid walls in
the cell faces or the Plateau borders, respectively. Plateau borders was discussed by Desai and Kumar’ and
The pressures at the top of a pool and at the exit of its incorporated in the model of Narsimham3 however, other
exhausting channel are identical and independent of i. This authors’ have assumed rigid walls on their models, since
is, in fact, what happens approximately in a froth where the experimenta results suggest a high surface viscosity due to

4030 J. Appl. Phys., Vol. 76, No. 7, 1 October 1994 M. A. Fortes and S. Coughlan
the adsorbed monolayer of tensiactive molecules. We also 12pL
ignore coarsening of the froth, i.e., we assume that the ge- hf=p6dr us (9b)
ometry and topology are unchanged during drainage, except
for the variation of hi and ai. Equations (9a) and (9b) can be written in the form
We first make a mass balance for pool i. Let ui be the
escape velocity from pool i defined such that aiuI is the PVO
hf.:, u, ;
volume of liquid leaving pool i per unit time. The volume
entering from pool i-l is u~-~u~-~. The height hi in pool i
where ua is the initial velocity, for hi=h,. This velocity can
changes with time and the rate of change of the volume of
be calculated from Eq. (8). For a circular section
liquid in the pool is Ai(dhi/dt), since we assume Ai to be
independent of time. Mass conservation (incompressible
liquid-volume conservation) leads to (114
dhi
ui-laill-uiai=Ai -, i>l, and for parallel plates
dt

for 1 <i=GN. For the first pool the equation is

dh ,
Q-ulal=Al ~7 (5) An estimate of the numerical values of p and u. can be
obtained as follows. For p=l g cmw3, ,!~u=lO-~ N m-* s
where Q is the flow of liquid supply at the top of the column (density and viscosity of water), L =0.5 cm, r=0.02 cm, and
of pools. The flow of liquid leaving the last pool N, i.e., the g=980 cm sm2 we obtain from Eq. (lla) /3ue=lOO cm s-l.
flow of drained liquid, is uNaN and the total volume drained Using Eq. (8) and taking h,<L, we obtain uo=5 cm s-l and
up to time’t is therefore p=20. Similar calculations for d=0.005 cm using
Eqs. (llb) and (8) give uo=0.2 cm s-l and p=12 000. In
this example, the initial flow of liquid through the circular
exhaust channels (Plateau borders of cross section rrr”) is
comparable to that through the rectangular exhaust channels
In an actual froth. there are 4 columns per unit area of the (cell walls of cross section Ld).
cross section. If the froth cross section is S, the total amount We now combine Eqs. (8) and (10) with the mass con-
of liquid accumulated by drainage can be calculated from servation equations (4) and (5). Two cases will be ,treated
separately. In the first, the parameter pi is taken as a constant
‘,= f@vN, (7) independent of i and independent of time. In the second case,
where the factor l/3 takes into account the equivalence pre- pi varies during drainage and will be related.to hi. In both
viously discussed between the actual froth and the model cases we assume that A,=A, is a constant independent of i.
froth. This is the quantity that is directly measured in stan-
-.
dard drainage experiments.
We now relate the average velocity u through the exr III. COMANT EXHAUST AREA
haust channel (omitting the subscript i) to the height h, using il
well-known equations for tlie laminar flow of a newtonian We differentiate Eq. (8) with respect to time, and corn-.
liquid of viscosity ,u and density p. We combine Bernoulli’s bine with Eq. (10) to obtain
equation (energy conservation) with the equation of Ragen-
Poiseuille for the friction loss in a parabolic velocity profile dhi dui
g yjy-‘~PUOfU’) --g. 02)
(see, for example, Ref. 9). Assuming that the pressures at the
extreme sections are identical and the velocity’at the top is
The mass conservation equations (4) and (5) are written in
negligeable, we have
the form 1 .- .
u2+2ghf=2g(h+L). 63)
dh.
The term g(h+L) is the change in potential energy per unit L=X(Ui-l-Uf), (134
dt
mass (g is the gravitational acceleration) and u”/2 is the
change in kinetic energy per unit mass. The friction loss is dhl
-= %UI, Wb)
given in terms of hf. For a circular cross section of radius r dt A0
(Plateau borders acting as exhaust channels) the equation for
hf. is where

(94
while for flow between two parallel plates separated by a is. another constant independent of i. We now eliminate
distance d (cell walls acting as exhaust channels) dhJdt between Eqs. (12) and (13),

J. Appt. Phys.,.Vol. 76, No. 7, 1 October 1994 M. A. Fortes and S. Coughlan 4031
 dui
(PUO+Ui) dt=Xg(Ui-l-Ui),

(15)
dul
CPuo+ud dt=h

It is convenient to introduce new dimensionless variables as

follows:

ui
Xi=-,
uo
Hi+, T= 2 t, (16aj 10000
T (time)
and define

czg$, Q*--c. FIG. 3. Reduced height Hi in pools 1-5 (constant exhaust area) as a func-
tion of reduced time T (log scale) for Q*=O (simple drainage) and for two
a0u0
values of p.
Equations (15) then become
dxi The time interval AT between successive inversions of
(p+Xi) dT=X”-l-Xiy
slope was obtained. The results were then compared to those
(17) obtained experimentally by Weaire et aL5 in which a moving
dxl interface between wet (liquid rich) and dry froth was ob-
(p+xl)‘-=12*-x1.
dT served. In the experiment, this interface was found to move
with a constant velocity, the velocity being proportional to
These equations can be integrated numerically, from T=O,
Q’“. The moving interface can be identified with the inver-
xi= 1 (and Hi= l), for any values of p and Q *. Other initial
sion of slope of the Hi(T) curves such as in Fig. 4. The
states could, of course, be considered. Simple drainage cor-
values of AT between changes of slope in successive pools
responds to Q*=O. In the experiment of Weaire et al5
for given Q* were found to be fairly constant, indicating a
Q *#O. The heights Hi can be calculated from Xi through Eq.
fairly constant velocity of the interface; however, this veloc-
(8) which can be written in the form
ity could not be related to Q * through a power law, although
it increases as Q* increases. The model was therefore elabo-
rated to take into account a variable exhaust area.

In the initial state xi= Hi= 1 and

V. VARIABLE EXHAUST AREA

(19) As liquid drains, the area ai of the exhaust channels

decreases. In order to take this feature into account we as-
For given p and L/h, the value of ~=gh,lu~ can be obtained sume that the area ai is proportional to hi. This is equivalent
from Eq. (19). The parameters of the problem are Q*, fl, and to assuming that the volume of liquid in the channels is
L/ho. The total volume drained from a column of pools can proportional to the volume of liquid in the associated pools,
be obtained from Eqs. (6) and (7) with t defined by Eq. since we maintain the assumption that Ai=Ao is a constant.
Wb), We write

XN dT. (20)

0.8
IV. RESULTS FOR CONSTANT EXHAUST AREA

The flow equations (17) were integrated for N=30 pools

and for various values of Q*, fi, and Llho=3. The results are
given in the form of H,(T) curves where Hi is calculated
from Xi through Eq. (18).
Figure 3 shows Hi(T) curves for various pools for
0.0; I I I
Q*=O (simple drainage) and for /?=O (no viscosity) and 1000 2000 3000 4000 5000
/3=600. In other computations the value Q * =0 was used for
T (time1
T in the interval 0, To. At To the value of Q* was changed to
some nonzero value. Figure 4 shows an example for p=600,
FIG. 4. Reduced height Hi in pools l-5 as a function of reduced time T.
To= 150, Q * =l. The Hi(T) curves show a change in the The pools drained (Q*=O) from T=O to Te=lSO. At this stage liquid was
sign of slope which occurs sequentially [the same happens in supplied at the top (pool 1) with’Q*=l. Note the inversion of slope of the
the Xi(T) curves]. Hi(T) curves which occurs progressively.

4032 J. Appl. Phys., Vol. 76, No. 7, 1 October 1994 M. A. Fortes and S. Coughlan
 hi
a;=u() -,
ho 0.8
where a0 is the initial area of the channels, independent of i.
This equation can be replaced by a more general power de- 0.6
pendence of ai on hi, but we shall not consider this possi- i’
0.4
bihty in detail.
The mass conservation equations (4) and (5) become 0.2
dHi
Xi-1Hi-l-XiHi=~ do,

(22) POOL NUMBER

FIG. 6. Reduced height Hi as a function of pool number in simple drainage

keeping the dimensionless variables defined in Eq. (16). (Q*=O) in the case of variable exhaust area (circular channels).
The flow equation now depends on whether the cross
section of the channels is circular or rectangular {in which
the width is much larger than thickness d). In the first case r? from which 5 can be obtained for given p and L/ho. There
is proportional to hi, in the second case di is proportional to are three parameters as before (including Q*).
fZi. Substituting into Eq. (8) and using Eq. (9) we find the The total volume drained from one column of pools .is
general form of the flow equation, given by [cf. Eq. (20)]

x:+2,0 $=25 HNxN dT. (27)

f
where n = 1,2. 12= 1 corresponds to a circular exhaust cross VI. RESULTS FOR VARIABLE EXHAUST AREA
section for which
Equations (22) and (23) were integrated for N=30 pools
(24)
and for various values of Q”, p, and Llh,=3. The integra-
tion was carried out both for II = 1 and n =2. The results are
and n=2 corresponds to a rectangular exhaust cross section again given in the form of curves Hi(T).
for which Figure 5 shows Hi for the first five pools as a function of
T for Q * =0 (simple drainage), IZ= 1 and for p=O (no vis-
12pL cosity), and p=600. In Fig. 6, Hi is plotted as a function of
Buo= -32-Y W pool number i for various values of T for Q ‘=O, n = 1, and
PO
p=600. At large times a constant gradient of Hi (and there-
where r. and do are, respectively, the initial radius and thick- fore of liquid fraction) develops in the froth. The accumu-
ness of the exhaust channels [cf. Eq. (ll)]. lated drained volume from a column is plotted in Fig. 7 as a
Equations (22) and (23) can be integrated numerically function of T for p=600.
from T=O, Xi= Hi= 1. The flow equation (23) gives for T=O In other computations the value Q * =O was used for T in
the interval 0, To. At To the value of Q * was changed to
(261 some nonzero value. Figure 8 shows an example for II =l
and p=600, T,=625, Q*= 100. The H,(T) curves show a
change in the sign of slope which occurs sequentially [the

30 -

T(time)
T (time)

FIG. 5. Reduced height Hi in pools 1-5 as a function of reduced time T (log

scale) in simple drainage (Q * =O) for two values of p in the case of variable FIG. 7. The quantity (l/h,,A,)VN(t), proportional to the volume drained, as
exhaust area (circular channels). a function of time for p=600 iQ*=O, n=l).

J. Appl. Phys., Vol. 76, No. 7, 1 October 1994 M. A. Fortes and S. Coughlan 4033
 10 r

0.1; / I 1 0.0; I
200 400 600 800 200 400 600
T (time) (a) P

FIG. 8. Inversion of slope in the H,(T) curves (log scale) for the first five
pools due to liquid supply with Q*=lW at To=625 (p=600; n=l).

same happens in the xi(T) curves]. The time interval AT

between successive inversions of slope was obtained and the
results are plotted in Fig. 9. It can be seen that after an initial
transient behavior the values of AT are approximately con-
stant, though slowly decreasing with the propagation of the
interface between the wet and dry pools. It should be noted,
however, that locally AT can show both an increase or de-
crease in value. (b) P
We then look for a similar relationship to that obtained
by Weaire et aL5 between AT and Q*. We have therefore FIG. 10. (a) The best-fit exponent tn and (b) the best-fit constant b in Eq.
attempted a power relationship of the form (25) as a function of ,f? for circular and rectangular exhaust channels.

&=bQ*“, (28)
shown in Fig. 10(a). The values of m for rectangular chan-
where m and b are constants. Weaire et al. experimentally nels were find to be smaller than those for circular channels.
obtained nt = l/2. We take for AT the average value in the At large fl, the m values are in both cases close to l/2. There
range where AT is slowly decreasing. In all cases it was is, therefore, a reasonable agreement with experiment.
possible to find values of m and b such that Eq. (28) was The dependence of m on the time To at which liquid is
satisfied. It is found that the value of m varies with p, as added was also examined. It was found that there is a weak
dependence of m on To; however, this effect is small com-
0.05
pared to differences in m that occur with a change in p.
Finally, the constant b in Eq. (28) was found to decrease
0.04 - (i.e., wet/dry front velocity increases) as p increases [Fig.
F. f---
0.03 Nb)l.
si A comparison can also be made with the experimental
0.02 values obtained by Weaire et al. For example, an interface
0.01 velocity of 1.8 cm s-l was experimentally measured for a
i/ flow rate Q of 1 cm3 s-r. Using these results, along with
0.00~‘~~~~~~~~~~~~~~“““~‘
10 20
~~‘~”
30 some reasonable assumptions of the dimensions of the tube
(4 POOL NUMBER containing the froth and the cell sizes, we can obtained the
reduced flow rate Q* and reduced time interval AT that cor-
respond to the above experimental results. The values ob-
tained are as follows: AT = 18 for Q * =20. It can be seen that
these figures are of the same order of magnitude as those
6
I- obtained using the current model for simple drainage.
a
4

VII. CONCLUSIONS

A simple model for foam drainage has been proposed in

POOL NUMBER
order to study both drainage and the effect on the foam of the
(W continuous addition of liquid from the top. The model asso-
ciates. the Plateau borders and the quadruple junctions with
FIG. 9. ‘IIme interval AT between inversions of slope in Hi(T) curves for
successive pools for n=l, Q*=lOOz (a) for p=O (Ta=1.56); (bj for p=600 vertically stacked pools that are connected by channels
(To=625). through which the liquid drains. The initial volume of liquid

4034 J. Appl. Phys., Vol. 76, No. 7, 1 October 1994 M. A. Fortes and S. Coughlan
in the pools and the length of the exhausting channels can be lar cross section for the exhaust channels and varies (weakly)
related to the initial volume fraction of liquid in the froth and with the time at which liquid is added from the top.
to the average cell size, respectively. -The model examined, although ignoring additional re-
Equations have been derived for the flow of liquid finements (such as a pressure gradient through the liquid),
through the channels and pools. These have been numeri- does show remarkable similarities to experimental results
cally integrated and the results obtained have been compared and suggests potential areas of further investigation; for ex- .
to the standard drainage curves for liquid foams and to the ample,’the effects of the viscosity of the liquid and the scale
recent experimental results obtained by Weaire et al.’ of the froth on the experimental value of m and the question
In its simplest form the model assumes that the exhaust of whether the interface velocity is indeed constant or varies
channels are constant in cross-sectional area; however the slowly as it moves through the froth.
results obtained are not in agreement with experiment.
Therefore, a more elaborate model was used in which the ACKNOWLEDGMENTS
cross-sectional area was related to the amount of liquid in the The authors acknowledge Professor D. Weaire for access
pool. The drainage curves obtained are in good agreement to his experimental results prior to their publication and Pro-
with those obtained from experiment. fessor Palhoto de Matos for helpful discussions. Figure 1
In an actual froth, the liquid drains both through the was taken from unpublished work of B. M. S.’Gonqalves,
Plateau borders and the cell walls. The model shows that the L. M. S. Teixeira, and T. T. P. Gal&o. The research was
relative importance of the two paths depends on their relative supported by the EC Science Programme, Contract No. SC-
dimension and no general conclusion can be drawn on which CT92-0777.
one, if any, dominates.
The effect of continuously adding liquid from above was ‘K. B. Kann, Fluid Dyn. 21, 420 (1986) (Engl. trans.).
also studied. An interface between wet and dry pools. was ‘I. I. Gol’dfarb, K. B. Kann, and I. R. Shreiber, Fluid Dyn. 23, 244 (1988)
observed. After an initial transient, this interface moves with (Engl. transl.).
‘G. Narsimhan, AIChE Symp. Ser. 86, 76 (1990).
a slowly changing velocity (as opposed to the constant ve- “J. J. Bikermann, Foams (Springer, Berlin, 1973).
locity reported by Weaire et al.). The average velocity was ‘D. Weaire, N. Pittet, S. Hutzler, and D. Pardal, Phys. Rev. L&t. 71, 2670
found to be proportional to Q*m, where ti varied signifi- (1993).
cantly with the parameter /I. However, m was found to be of 6V. V. Krotov, Colloid J. U.S.S.R. 42, 912 (1980) (Engl. trans.).
7K. B. Kann, Fluid Dyn. 25, 236 (1990) (Engl. trans.).
the order of 10-r and was close to l/2 for large values of the “D. Desai and R. Kumar. Chem. Eng. Sci. 37, 1361 (1982).
parameter ,B, such as occur in “normal” froths. The value of “F. M. White, Fluid Mechar@s, 2nd ed. (McGraw-Hill, New York, 1986),
m also depends on the assumption of a circular or rectangu- Chap. 6.

J. Appl. Phys., Vol. 76, No. 7, 1 October 1994 M. A. Fortes and S. Coughlan 4035