J . FZuid Meeh. (1987), vol. 180, pp.

495-514 495
Printed in Great Britain

Creeping flow in two-dimensional channels

By C. POZRIKIDIS
Research Laboratories, Eastman Kodak Company, Rochester, NY 14650, USA

(Received 25 September 1986)

Creeping flow in two-dimensional periodic channels of arbitrary geometry is con-

sidered. The problem is formulated using the boundary-integral method for Stokes
flow, presently adapted for periodic flows with special geometrical characteristics.
Numerical calculations for steady flow in channels constricted by a plane and a
sinusoidal wall are performed. Detailed streamline patterns are presented and criteria
for flow reversal are established. It is shown that for narrow channels the mechanism
driving the flow has a strong effect on the structure of the flow. The results are
discussed with reference to lubrication, coating and molecular-convective processes.

1. Introduction
Flow in channels is encountered in diverse areas of fluid mechanics. The literature
contains a large number of studies for a variety of flow conditions, corresponding to
different physical situations .
Steady flows between two flexible or solid bodies in relative motion are often
studied with reference to lubrication or coating processes (Langlois 1964). The two
bodies are kept separated by strong pressure forces generated by the flow. In these
processes it is of interest to determine how irregularities of the body surfaces affect
the ratio of the friction force to weight of the bodies, and the flow rate. Pressure-driven
flow in narrow channels with sudden expansions is important in the extrusion and
coating of thin liquid films. The expansions dampen three-dimensional fluid motions
causing product non-uniformities. I n these applications it is important to design the
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

channels to avoid low wall shear stress and onset of recirculating fluid regions. These
may lead to particle or bubble entrapment, increased residence times, and build-up
of chemical reactants. The above two examples demonstrate the importance of flows
in corrugated channels for specific industrial processes. From a fundamental
standpoint, these flows provide useful information for flows in complex media such
as packed bed reactors and porous rocks. The analysis provides relationships for the
flow properties as functions of the macroscopic material characteristics (Taylor 1971;
Deiber & Schowalter 1979).
Unsteady flows are also studied with reference to various physical systems. I n a
simple class of flows, the motion is due to the steady propagation of waves along
the flexible walls of a channel. Examples in this class include the locomotion of
microscopic organisms, peristaltic pumping of sensitive or corrosive fluids, physio-
1ogicFl flows in the human body (Pozrikidis 1987), and the generation of water waves
by wind (Caponi et al. 1982). A second class of flows includes oscillatory flows over
wavy surfaces. Onset of steady streaming motion, and periodic formation and
expansion of eddies are two interesting phenomena aasociated with these flows (Hall
1974; Sobey 1983).
 496 C. Pozrikidis
Apart from their importance in various natural and industrial systems, flows in
corrugated channels are of pure fundamental interest ; they constitute convenient
prototypes for studying the mechanisms of viscous flow under different flow
conditions. Similar systems consisting of walls with cavities, two intersecting planes,
two cylinders, and a plane and a cylinder have been employed by previous authors
(Hasimoto & Sano 1980). As an example, Jeffrey & Sherwood (1980)studied shear
flow over a stationary cylinder tangent to a moving wall and established criteria for
eddy formation at the points of contact.
We saw that there is a wide variety of applications involving flow in corrugated
channels. To study different flow conditions, various methods have been employed.
A common approach is to assume small wall corrugations, small channel widths, and
very small or very large Reynolds numbers, and to write perturbation expansions
in these variables. To consider the more realistic cases of finite amplitudes, numerical
methods have been employed. These include finite difference, finite element, and
spectral solutions, usually at intermediate and low Reynolds numbers. As expected,
these methods are most effective for relatively simple flow geometries. The goal of
the present article is to provide a detailed numerical investigation of flow in
two-dimensional channels, in the limit of creeping motion. Emphasis will be placed
on the fundamental description of the flow as well as on the application of the results
to specific engineering processes.
The analysis is based on the boundary integral method for two-dimensional Stokes
flow, presently extended to two-dimensional periodic flows with special geometrical
features. The basic strength of the method is the simplicity of the required numerical
procedure, allowing an accurate, detailed, and extended investigation.
In $52 and 3 we present the mathematical formulation and develop the numerical
procedure. In $4 we discuss shear and pressure driven flow in periodic channels
constricted by a plane and a sinusoidal wall.

2. Mathematical formulation
We consider slow motion of a Newtonian fluid in a two-dimensional domain which
may be bounded by a number of fluid, solid, and free surfaces or may extend to
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

infinity. At low Reynolds numbers the flow is governed by the Stokes equation and
the continuity equation
aa
2 = 0, (1)
ax,

respectively, where ui, is the stress tensor, u!, = - 8 , P+p(au,/ax,++au,/ax,). These

equations constitute a system of linear, elliptic, partial differential equations, whose
solution may be conveniently expressed as an integral over the flow boundary
(Higdon 1985). The result for points on the flow boundary is

For points in the interior of the flow, the factor 2np in the denominator should be
replaced by 4np. In the above equation, f i = u t k y k is the boundary force, and rk is
 Creeping Pow in two-dimensional channels 497
the unit vector normal to the boundary pointing into the fluid. When the flow extends
to infinity, the boundary is assumed to include a circle or part of a circle of infinite
radius. In this case, care must be taken for the convergence of the integral.
The tensor S,, represents a singular fundamental solution to Stokes equation with
a logarithmic singularity. The velocity, pressure, and stress field associated with this
solution may be written in the form

1
P(x’)= --p,(x, x’) a,,
41cIu I (4)

-
where qjlC l/lx’ -xl as Ix’ -XI -+ 0, and a, is an arbitrary constant. qjkis the stress
tensor corresponding to S,,, defined as q,k= -atk q+p(W,,/az;+ W,/az;). Using
the divergence theorem, one may write (3)in an alternative form,

where a/an denotes differentiation normal to the boundary in the direction of the
fluid (Happel & Brenner 1965, p. 81). Equation (3)is a singular Fredholm equation
of the first kind for the boundary force, and of the second kind for the boundary
velocity. Appropriate forms of this equation for interior or exterior flows are given by
Higdon (1985) and by Lee & Leal (1986).
The choice of the fundamental solution Sg, is important for the efficient imple-
mentation of the solution scheme. Therefore, we would like to discum it in some detail.
The simplest choice is the two-dimensional Stokeslet
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

Pf’T(9)= -2p$
i?

I
where 5 = x’-x, and r = 1 21 (Ladyzhenskaya 1969, p. 67). I n this cme, (4) and (6)
yield the flow at the point x’, associated with a point force of strength la1 located at x,
pointing in the -a-direction.
Other fundamental solutions may be employed to exploit specific geometrical
features. Blake (1971) proposed a fundamental solution which preserves zero velocity
along a plane, placed at y = w. This may be derived from Lorentz formula (seeHappel
& Brenner 1965, p. 87), and is composed of a Stokeslet and a collection of image
Stokeslets, Stokes-Doublets, and potential source-Doublets

syx, x’) = SST(R)-SST(2)-2(y-w) D(3)-2(y-w)SP(S), (7)

 498 C . Pozrikidis
where R = ( 2 , g )= (2'-x, yf +y-2w), and values of the indices 1 and 2 indicate the
x- and y-direction respectively. The matrices D and P are defined as

2# 22
-_ lnp--

lnp-- --
P2 P2
>

with p = 121. This fundamental solution is useful in problems involving the motion of
bodies near a plane wall.
For a flow that is symmetric with respect to a plane a t y = w,it is useful to introduce
a fundamental solution that yields zero velocity normal to this plane at y f = w,

+
qjw = @?($) ( - l)'+'@?(W), (9)
where again i,j = 1 , 2 indicate the x- and y-direction respectively. This solution is
schematically illustrated in figure 1 (a).Similarly, for a flow which is symmetric with
respect to a point, we introduce the solution Ssp,illustrated in figure 1(b).
For flows that are periodic in the x-direction, it is convenient to introduce a periodic
fundamental solution representing an array of point forces along the x-axis. The
velocity and stress tensors may be expressed in closed form by summing ( 5 )
A+k$A,-l
-k$A, 1
-kgAz '
A-k$A,
= - 2pA,,,
PfTP(2)
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

T
:: = 2p div (A, k$Az),
c:: = 2p div ( - k#Az, A ) ,
T:.: = 2p div (A, - k$A,J,
q.: = 2p div (k@A,,A ) ,
TSTP = TSTP = TSTP
Uk kU jkf

with A = 4 In [ ~ ( c o(k$)
s ~ - cos (k$))], (11)
where A is the separation between the point forces, k = 2x/A is the wavenumber, and
A,, A,, indicate derivatives of A with respect to k9 and k$ respectively. One may
readily verify that as la(+O, this solution behaves like a Stokeslet. Similarly, for
semi-inhite periodic flows that are bounded by a plane wall or are symmetric with
Ssw,and SP,
respect to a plane or a point, we introduce the periodic versions of Sw,
denoted as Swp,Sswp, and Sspp respectively. These may be expressed in closed form
by direct summation.
The simplification owing to the use of the appropriate fundamental solution may
 Creeping $ow in two-dimensional channels 499

---------

FIGURE
t
1 . Fundamental solution of Stokes equation for a flow symmetric with respect ( a )to a plane
and ( b ) to a point; the vectors represent Stokeslets.

I- h- I
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

A
k-1-

FIQURE2. Periodic flow in two-dimensional channels; (a) flow constricted by a plane and a
corrugated wall, ( b ) flow constricted by two walls that are symmetric with respect to the point A ;
w is the channel width and a is the amplitude of the corrugations.

be illustrated by considering flow in the channels shown in figure 2. In the first case
(figure 2a)we use the fundamental solution Swp and integrate (3)along the indicated
contour enclosing one fluid period. For zero net pressure drop, the integrals along the
two vertical segments cancel, whereaa the integral along the plane wall is equal to
twice the velocity of this wall. Similarly for flow in the channel shown in figure 2 (b),
we use the solution Sspp, and integrate (3) along the indicated contour. Again, for
zero net pressure drop, the contribution from the straight segments vanishes. Thus,
in both cases, the velocity is expressed simply as an integral over one period of the
wavy wall. Use of the appropriate fundamental solution for the corresponding flow
geometry will be implicit throughout the following discussion.
 500 C.Pozrikidis

3. Numerical procedure
In this section we develop a numerical method for solving (3) for the boundary force
ft(x’)given the boundary velocity ut(x’),with particular reference to the channel flows
depicted in figure 2. It is important to note that the boundary velocity may be
specified either as a total or as a disturbance quantity. When the flow is driven by
the motion of the boundaries, it is convenient to consider the total velocity; when
it is driven by an imposed mean pressure gradient or by the presence of body forces,
it is convenient to consider the disturbance flow owing to the boundary deviation
from a known reference state.
Our numerical procedure employs a collocation method, similar to that suggested
by Higdon (1985). First, we identify a set of N + 1 points along one period of the
appropriate channel boundary (figure 2). The boundary shape is approximated as a
polygonal line composed of N straight segments connecting these points. The
boundary velocity is approximated as a linear function, whereas the boundary force
is approximated as a constant function fr,
over the nth segment. Next, we apply (3)
at the middle of each segment xm, to obtain the following system of linear algebraic
equations for f:,
N N
uj(xm)+ Z B?(x,) = 2 AG(xm).fr, (12)
n-1 n-1
i r
where

Sn denotes integration over the nth segment. The singular integrals in (13) may be
evaluated numerically by subtracting and adding the Stokeslet,
r r r
J S,ds =
J (St,-@T)ds+
J @Tds, I
J qjkut nkds = J ( q j k - + J l’:; utnk ds.J
n n #

T:;)
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

ut nk ds

The second integrals in the right-hand side of (14), associated with the Stokeslet, are
evaluated analytically over each segment. Finally, the linear algebraic system (12)
is solved using standard Gauss elimination.
The efficiency of the above method was tested by solving for simple shear flow in
infinite fluid, considering as a control volume the channel shown in figure 2(a) for
different cell widths w/A and wave amplitudes a / h ; for this flow, the method yields
the exact solution. Indeed, our results were accurate up to the fifteenth decimal place,
limited by the computer round-off error.
Higdon (1985) showed that the relative error in the above approximation is of order
S2, where S is the segment length. Specifically, the relative error in evaluating the
integrals in (12), associated with the polygonal boundary approximation, is of order
S2kZ,where k is the local boundary curvature of order one for smooth boundaries.
Further, the relative error owing to the linear approximation of the boundary
velocities and the stepwise approximation of the boundary forces is of order S2. This
yields a consistent, order 62 total relative error.
Knowing the behaviour of the relative error is beneficial in obtaining high-accuracy
 Creeping flow in two-dimensional channels 501
results while using a moderate number of points. We use successive Romberg
extrapolation from primary runs with n = 20,40,80,and occasionally 160, to increase
the accuracy up to SB. The observed fast convergence confirms the error analysis and
validates our procedure.

4. Results and discussion

We consider steady flow in a periodic channel constricted by a plane wall at y = w,
and a sinusoidal wall at y = --a cos (kz).
This flow may be driven by two independent
mechanisms: the steady, horizontal translation of the plane wall, and the presence
of a mean pressure gradient 0 = - dP/dx. I n the trivial case of zero wave amplitude,
these reduce to plane Couette and plane Poiseuille flow, respectively. The two cases
will be considered independently with a linear superposition possible owing to the
linearity of Stokes equation. To clarify the computational procedure, we note that
in the shear-driven case we calculate the total flow, whereas in the pressure-driven
case we calculate the disturbance flow owing to the amplitude of the wavy wall.
4.1. 'Shear-driven flow '
First, we consider flow in narrow channels, taking as a typical example the case
w / A = 0.100. Figure 3 (a)illustrates the changes in the streamline pattern as the wave
amplitude a/A is increased from 0.025 to 0.090. When a / A = 0.025, the streamlines
are smooth, adjusting to the curvature of the wavy wall. Doubling the amplitude of
the wave causes a strong expansion of the fluid at the trough of the wavy wall, leading
to development of adverse pressure gradients and to flow reversal. The developed
eddies occupy almost the entire space of the sinusoidal corrugations, leaving only a
narrow gap for longitudinal fluid transport. As the wave amplitude becomes very
close to the channel width the channel reduces into a series of independent cells, and
the flow within each cell is essentially flow in a cavity driven by a moving lid (Pan
& Acrivos 1967). The structure of the flow may be better visualized by considering
the velocity profiles along the centre line z = 0 (figure 4a). We observe an almost
linear profile for small amplitudes, and a parabolic profile for large amplitudes, with
significnt regions of recirculating flow.
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

The above calculations demonstrate the onset of reversed-flow regions, when the
wave amplitude exceeds a critical value. Flow reversal is indicated by a change in
sign of the vorticity or the shear stress along the wavy wall. The distribution of shear
stress along the wavy wall for the three flows discussed above is presented in
figure 5 ( a ) . As the wave amplitude is increased, the shear stress at the trough
decreases rapidly, and touches the zero axis when a/A = 0.030. This defines the
critical wave amplitude for flow reversal.
The magnitude of the wall shear stress is important in applications involving
molecular-convective processes at high Prandtl or Schmidt numbers (Higdon 1985).
The pronounced maximum at the crest of the wavy wall indicates increased local
transport, which may be responsible for a number of effects. For instance, a dissolving
wavy wall will tend to level out under the high shear rate at the crests. Similarly,
a chemical reaction between the wall material and the fluid, with deposition of
products, will cause a fast increase in the wall amplitude and possible clogging of the
channel.
It is interesting to consider in some detail the structure of the developed viscous
eddies. We saw that the size of these eddies is a function of wave amplitude. To
illustrate this clearly, we consider the eddy height, h, defined as the vertical distance
 502 C.Pozrikidis
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

FIGURE 3. Streamline patterns for Couette flow in channels of (a)width w/A = 0.100, and wave
amplitude a/A = 0.025,0.050,0.090; (b) w / A = 0.500, and a/A = 0.100,0.200,0.300,0.400; (c)
w/A = 1.O00, and a/A = 0.200,0.500,0.800;( d ) w / A = 2.000, and a/A = 0.200.
 Creepingflow in two-dimensional channels 503
1.a

XW (

-1.1 I 1

-0.5 0 0.5 1.o

1.c
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

-Y (
W

-1.1
-0.2 0 0.5 1.o
ulV
FIQORE 4. Velocity profile along the vertical at x = 0, for Couette flow in a channel of ( a )width
w / h = 0.100 and wave amplitude (i) a / h = 0.025, (ii) a / h = 0.050, (iii) a/h = 0.090; (b)
w / h = 0.500, and (i) a / h = 0.100, (ii) a/h = 0.200, (iii) a/A = 0.300, (iv) a / h = 0.400.
 504 C. Pozrikidis
10.0

5.0

rrw
P
0

- 5.0
-0.5 0 0.5
x/A

10.0

5.0

uw
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

I
- 5.0 I

-0.5 0.5

FIQWRE 5. Shear stress distribution along the wavy wall for Couette flow in a channel of (a)width
w/A = 0.100 and wave amplitude (i) u/A = 0.025, (ii) a / A = 0.050, (iii) a/A = 0.090; ( b )
w/A = 0.500, and (i) a/A = 0.100, (ii) a/A = 0.200, (iii) a / A = 0.300, (iv) a/A = 0.400.
 Creeping flow in two-dimensional channels 505
1.:

h
z;;
0.:

0 0.5 1
alw
FIGURE6. Eddy size at trough aa a function of wave amplitude, for Couette flow in a channel
of (i) w/A = 0.100, (ii) w / h = 0.500

between the dividing streamline enclosing these eddies and the wavy wall at x = 0.
I n figure 6, we plot the reduced eddy height h/2a as a function of wave amplitude
a / w . This is equal to zero for non-reversing flow, a / w < 0.301, but then, it increases
rapidly, reaches a maximum, and finally, as a / w + 1, it tends to unity. The maximum
height, h/2a,is greater than one, indicating that eddies may protrude above the crests
of the wavy wall. I n contrast, the strength of the eddies, expressd by the x-velocity
on the dividing streamline at z = 0, increases in a monotonic, smooth fashion. These
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

results indicate that near the critical conditions for eddy formation, the flow is very
sensitive to small changes in the boundary geometry. As a consequence, the
fabrication of channels should be monitored very accurately to avoid large regions
of recirculating fluid.
Proceeding to wider channels, we concentrate on the case w/A = 0.500 (figure 3b).
As previously, for small wave amplitudes, the streamlines follow the curvature of the
sinusoidal w4ll, whereas for larger amplitudes flow reversal occurs. I n the present
case, as the wave amplitude becomes close to the channel width, secondary eddies
develop at the trough of the corrugations. The velocity profile at x = 0 for each of
the above patterns is presented in figure 4b. We note the weak fluid motion at the
trough of the corrugations for large wave amplitudes. The shear stress distribution
along the wavy wall is shown in figure 5b. As in the narrow channel case, the shear
stress reaches a maximum at the wave crests and drops to very low values along the
rest of the wavy wall. The curve for all\ =0.400 crosses the zero axis twice,
indicating the presence of secondary eddies. The height of the primary eddies shows
behaviour similar to that for narrow channels (figure 6).
Onset of secondary eddies inside deep corrugations is predicted by a number of
previous studies (Higdon 1985). I n particular, Moffatt (1964)predicted the develop-
 506 C. Pozrikidis
0.2

)(; cr
0.1

0 1.o 2.0

0.3

0.2
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

0.1

..................
....-.......

0 1.o 2.0
WlA

FIGURE 7. Critical wave amplitude for flow separation as a function of channel width for (a)Couette
and (b) Poiseuille flow. The straight solid line defines the physical boundary, a < w.In (a)dashed
line shows predictions of first-order perturbation theory for small w l A ; the dotted line shows
predictions of second-orderperturbation theory for small a/A; circles show experimental results
by Munson et d.(1986). In (b) dotted line shows predictions of first-order perturbation theory for
small w / A ; the dashed line shows predictions of second-order perturbation theory for small alh.
 Creeping JEowin two-dimensional channels 507

We
3
3

1
0 0.5 1.o
a/w

FIGURE 8. Drag force on the plane wall as a function of wave amplitudefor Couette flow in a channel
of (i) 0 , w / h = 0.100, (ii) 0 , w / h = 0.500. The dashed line shows predictions of first-order
perturbation theory for small wlh, and dotted lines predictions of second-order perturbation
theory for small alw.

ment of an infinite sequence of self-similar eddies at the corner between two

stationary solid planes, for sufficiently small angles of inclination. It is of interest to
examine whether the two eddies shown in the lower frame of figure 3 ( b ) may be
described by Moffatt’s similarity solution. For this purpose, we evaluate their
strength ratio equal to 922, and calculate the angle of inclination corresponding to this
value according to Moffatt. This is equal to in,very close t o the angle formed by the
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

tangents to the wavy wall at the lower points of separation. This demoristrates that
the flow inside the corrugations tends to attain a self-similar state, even at moderate
wave amplitudes. Thus, onset of eddies may be understood in terms of Moffatt’s local
analysis.
To show the behaviour for wider channels, we present streamline patterns for
w / h = 1.OOO and 2.000 (figure 3c, d ) . The first demonstrates the onset of a series of
alternating eddies for very deep corrugations in agreement with Moffatt (1964). The
second is characteristic of simple shear flow over a wavy wall, recovered from our
calculations for large w/a.
We saw that for every channel width w / h there is a critical wall amplitude, (a/&,
for flow reversal. I n figure 7(a),we plot (alh),, as a function of w/h, along with
previous experimental results (Munson, Rangwella & Mann 1985)’ predictions of
second-order perturbation theory for small amplitudes a / w (Appendix), and predic-
tions of first-order perturbation theory for small channel widths w / h (Hasegawa &
Izuchi 1984). We observe a good agreement between the numerical and experimental
predictions; the small differences are attributed to the cylindrical geometry used by
the above authors. Perturbation expansions fail to predict flow reversal with
reasonable accuracy, except for very narrow channels, and therefore, they are of
17 RLY 180
 508 C.Pozrikidis
1.o

2Q
- 0.5
wv

0 0.5 1.o
alw
FIGURE 9. Flow rate as a function of wave amplitude for Couette flow in a channel of (i) 0 ,
w / A = 0.100, (ii) 0 , w / A = 0.500. The dashed line shows predictions of first-order perturbation
theory for small w/A, and dotted lines predictions of second-order perturbation theory for small
afw.

limited practical interest. At large amplitudes, the incipient flow-reversal curve tends
to an asymptotic value, characteristic of simple shear flow over a sinusoidal wall,
Looking at figure 7 ( a )from a different perspective, we consider the wave amplitude
a/A fixed, and vary the channel width w/A. This indicates that eddies may be
produced for any wave amplitude by sufficiently decreasing the channel width.
Narrower channels cause stronger adverse pressure gradients, responsible for flow
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

reversal.
Concluding this section, we would like to discuss the effect of the channel geometry
on certain bulk properties of the flow, namely the drag force on the moving plane
and the flow rate. The first is important in lubrication processes, whereas the second
is important in coating or pumping operations. These are presented in figures 8 and
9, along with predictions of perturbation theories. Considering first the drag force,
we note the parabolic increase for small a/v and the singular behaviour as alw tends
to unity. Perturbation analysis predicts that this singularity is of the order
w/(w-ap, as compared with A/wfor zero wave amplitude (Hasegawa & Izuchi 1984).
Thus, measuring the drag force may provide a potential method for estimating wall
roughness. It is interesting to note that the drag force is approximately doubled when
the wave amplitude becomes equal to half the channel width. Figure 9 shows that
the flow rate decreases at a parabolic rate for small amplitudes, and at an almost
linear rate for large amplitudes. It vanishes when the gap between the two walls
becomes zero. The onset of eddies does not affect the drag force or the flow rate in
any apparent fashion.
 Creeping$ow in two-dimensional channels 509

FIQURE 10. Streamline patterns for Poiseuille flow in a channel of ( a )width w / h = 0.600 and wave
amplitude a / h = 0.300; (b) w / h = 1.OOO and a/A = 0.500; (c) w / h = 2.000 and a/A = 0.200.

4.2. Pressure-driven $ow

https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

In this section we consider flow driven by an imposed mean pressure gradient

G = -dP/dx. We discuss our results with respect to those for shear-ddven flow,
indicating the differences between the two configurations. First, we not;i: that for very
wide channels, the structure of the flow in the vicinity of the wavy wall is independent
of the type of flow. It may be substantially different for narrow channels, reflecting
variations in the driving mechanism.
Our calculations show that in a narrow channel w / h = 0.100, the streamlines adjust
to the curvature of the wavy wall for any wave amplitude. This constitutes a
fundamental difference from the shear-driven case for which the flow reverses when
a / h = 0.030. Physically, in the pressure-driven case, the imposed mean pressure
gradient is able to defeat any developing local adverse pressure gradients.
Proceeding to wider channels, we consider a channel with w / h = 0.500 and
a / A = 0.300, figure lo@).A comparison with figure 3 ( b ) (third frame) shows that
in pressure-driven flow, the fluid is able to penetrate much deeper into the sinusoidal
corrugations, causing a reduction in eddy size. Calculations for larger wave ampli-
tudes show that the developing eddies never fill up the entire channel ;there is always
a substantial region of non-reversing flow attached to the plane wall. In the limit as
17-2
 510 C. Pozrikidis

1.a

-YW a

-I.(
-0.5 0 0.2
2pu/6wa

FIGURE1 1 . Velocity profile along the vertical at x = 0, for Poiseuille flow in a channel of
w/A = 0.500, and (i) a / A = 0.100, (ii) a / h = 0.200, (iii) a / A = 0.300, (iv) a / h = 0.400.

a +w,the reduced eddy size h/(2a)tends to 0.25, as opposed to unity for shear driven
flow.
The above results demonstrate that the structure of the flow may be critically
affected by the mechanism driving the flow. Velocity profiles at x = 0 for pressure-
driven flow are presented in figure 11, and may be compared with those for
shear-driven flow, figure4(b). We notice the rapid decrease of the velocity with
increasing wave amplitude. This is a result of our non-dimensionalization (with
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

respect to the mean pressure gradient), and is attributed to the rapid increase in wall
drag force, and therefore in flow rate, with increasing the wave amplitude. The
strength of the eddies is substantially smaller than that for shear-driven flow, owing
to the fact that in the present case, the eddies reside more deeply into the
corrugations. Considering the shear stress distribution (figure 12), we note the
pronounced maximum at the wave crests, familiar from shear-driven flow. However,
now we observe a smoother decrease around the maximum, suggesting a milder effect
on molecular-convective processes. The shear stress at the wave crest is a function
of wave amplitude, reaching an extreme value approximately when a / h = 0.200. This
is expected, since for a specified pressure gradient, increasing a / h , causes a local
acceleration of the flow above the crests, simply for mass conservation, but also a
reduction in the flow rate; the combined effect yields a maximum. Interpreting this
behaviour, we conclude that for a specified pressure drop, there is an optimum wave
amplitude for fastest convective transport.
Flow reversal in pressure driven flow may be understood by considering the
individual mechanisms dominating different flow regions. For large wave amplitudes,
the flow within the corrugations is dominated by the similarity solution of Moffatt
(1964),describing an infinite sequence of eddies. On the other hand, as a/w+ 1, the
 Creeping$ow in two-dimensional channels 511

2a
-
wC

-0.5 0 0.5
x/A
FIGURE12. Shear stress distribution along the wavy wall for Poiseuille flow in a channel of
w/A = 0.50, and (i) a / h = 0.100, (ii) a/A = 0.200, (iii) a/h = 0.300, (iv) a/A = 0.400.

flow in the vicinity of the wave crests is dominated by the Jeffery-Hamel similarity
solution (Fraenkell962). This describes flow in a wedge confined by two planes, with
a point source or point sink at the apex. I n the limit of zero Reynolds number, and
when the wedge angle 2 a is less than 1.43x, the velocity field in polar coordinates
is given by the similarity solution

u, = 2 -Q
sin2a - sin28
T sin 2a -2 a cos 2a
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

u(j = 0,

where Q is the strength of the point source or point sink, and 9 = 0 along the axis
of the wedge. Note that the radial velocity decreases uniformly as the inverse of the
distance from the origin, but flow reversal does not occur. This solution may be
extended to describe flow in corners with curved walls, provided a local coordinate
system is used (Langlois 1964,p. 177). For a corner formed by a plane and a sinusoidal
wall, the centre of the local coordinate system is located at the intersection between
the plane wall and the tangent to the wavy wall. Flow reversal along the wavy wall
is determined by the relative strength of the Jeffery-Hamel and the Moffatt similarity
solution. When the first dominates, for small wave amplitudes, the streamlines bend
to follow the curvature of the wavy wall ; when the second dominates, for large wave
amplitudes, flow reversal occurs.
Considering wide channels, we plot the streamline patterns for w/h = 1.OOO
and 2.000 (figure l o b , c ) , and compare them to the corresponding ones for shear
driven flow presented in figures 3(c) (middle panel) and 3(d). The comparison for
w/h = 1.000 clearly illustrates the deeper penetration of the fluid into the sinusoidal
corrugations for pressure-driven flow. On the other hand, the comparison for
 512 C. Pozrikidis
w/A = 2.000 indicates a similarity in the flow structure for very wide channels. This
is expected, as when w/a becomes large, the curvature of the parabolic profile in the
vicinity of the wavy wall becomes less important, and the parabolic flow is locally
approximated as a simple shear flow.
The incipient flow reversal curve for pressure-driven flow is shown in figure 7 (b),
along with predictions of perturbation analyses. Note the insufficiency of second-
order perturbation expansion for small a/w (Appendix) to predict flow reversal.
First-order perturbation expansion for small w / A (Hasegawa & Izuchi 1983) is
accurate only in a qualitative sense. At large channel widths, the separation curve
tends to an asymptotic value which is identical with that for shear-driven flow, and
is characteristic of simple shear flow over a sinusoidal wall. Again, an interesting
interpretation arises by considering the wave amplitude constant and varying the
channel width. This shows that at small amplitudes, flow reversal does not occur for
any channel widths. Next, there is a range of moderate amplitudes where the flow
reverses only within wide channels; the eddies produced may be suppressed by
sufficiently decreasing the channel width. Finally, flow reversal always occurs for
large wave amplitudes, independently of the channel width. This behaviour is
fundamentally different from that for shear-driven flow, for which eddies may be
produced for any wave amplitude by sufficiently decreasing the channel width.

The support of Eastman Kodak company is acknowledged.

Appendix
In this Appendix we outline a perturbation analysis of shear- and pressure-driven
flow in a channel constricted by a plane wall a t y = w , and a sinusoidal wall at
y = --a cos (kz),valid for small wave amplitudes E = a/w. First, following Wang
(197&),we introduce the stream function $, and expand

JL
vw
= $o+E$’1+€2$2+... .
The zeroth-order term is equal to $o = -?jPfor Couette, and $o = 3 2 Y-3) y2 for
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

Poiseuille flow, where Y = y/w.In the first case V is the velocity of the plane wall;
in the second case V is defined with respect to the mean pressure gradient
G = -dP/dx, as V = Gw2/2p. In the limit of creeping motion, $ satisfies the
biharmonic equation and thus

v4+-( =0 (i = 0,1,2, ...). (A 2)

The boundary conditions satisfied by the stream function are the no-slip and
no-penetration condition on the wavy and the plane wall,
@=-=O w at Y = - ~ c o s ( k z ) ,
(A 3)
aY
@=O atY=l,

and ”= v
-- at Y = 1 for Couette flow,
aY
or -= 0 a t Y = 1 for Poiseuille flow.
aY
 Creeping Pow in two-dimensional channels 513
Substituting (A 1)into these equations and collecting first-order terms, we may solve
for the first-order correction. Details are given by Wang (1978); the result is identical
for Couette and Poiseuille flow,
1
=- c o ~ ( k ~ ) - [ - 2 q ( e Q ~ - e ~ ~ ~ ) + ( e - ~ Q +2q-l)
YeQY+(e2Q-2q-1) Ye-QY],
R
(A 4)
where &' = e2Q+ e-2Q-2-4q2,
and q = kw.This is a purely periodic term in phase with the wave and thus, does not
affect the flow rate or the mean drag force on the solid surfaces. Working in a similar
fashion, we may evaluate the second-order correction (see also Munson et al. 1985),
+I.a
=-~+GY-~YZ+(Ae2QY+Be-2QY+CYe2QY+DYe-2 QY)
sin(2kz), (A5)
(8q2-ee-4Q+1 +4q- 16Fq)
where A= 9
4E
B =+-A,

C=
2q( -4q- + +
1 e-4Q) 4F( - 1+ 4q e-4Q)+ 9
4E

D=
2q( -4 q + 1-e4*)+4F(- 1-4q +e4Q) 9
4E
with E = 16q2-ee-44-e4Q+2,

F = - q(4q +e-29-e2*)
R 9

and G = F, for Couette flow,

G = F-+, for Poiseuille flow.
The non-periodic part of (A 5) yields an order-c2contribution to the drag force and
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press

the flow rate. The wall shear stress may be evaluated in a straightforward fashion,
by differentiating (A 1).

REFERENCES
BLAKE,J. R. 1971 A note on the image system for a stokeslet in a no-slip boundary. Proc. C u d .
Phil. SOC.70,303-310.
CAPONI,E. A., FORNBERQ, B., KNIGHT,D. D., MCLEAN,J. W., SAFFMAN, P. G. & YUEN,H. C.
1982 Calculations of laminar viscous flow over a moving wavy surface. J. Fluid Mech. 124,
341-362.
DEIBER,J. A. & SCHOWALTER, W.R. 1979 Flow through tubes with sinusoidal axial variations
in diameter. AIChE J. 25, 638-645.
FRAENKEL, L. E. 1962 Laminar flow in symmetrical channels with slightly curved walls. I. On
the Jeffery-Hamel solutions for flow between plane walls. Proc. Roy. Soc. Lond. A267,
119-138.
HALL,P. 1974 Unsteady viscous flow in a pipe of slowly varying cross-section. J. Fluid Mech. 64,
209-226.
HAPPEL,J. & BRENNER,H. 1965 Low Reyno2de Number Hydrodynamics. Noordhoff.
HASEGAWA, E. & IZUCHI,H. 1983 On steady flow through a channel consisting of an uneven and
a plane wall. Part 1. Case of no relative motion in two walls. Bull. JSME 26(214), 514-520.
 514 C. Pozrikidis
HASEQAWA, E. & IZUCHI, H. 1984 On steady flow through a chanqel consisting of an uneven and
a plane wall. Part 2. Case of walls with a relative velocity. Bull. JSME 27 (230), 1631-1636.
HASIMOTO, H. & SANO,0 1980 Stokeslets and eddies in creeping flow. Ann. Rev. Fluid Mech. 12,
335-363.
HIQDON, J. J. L. 1985 Stokes flow in arbitrary two-dimensional domains: shear flow over ridges
and cavities. J . Fluid Meeh. 159, 195-226.
JEFFREY, D. J. & SHERWOOD, J. D. 1980 Streamline patterns and eddies in low-Reynolds-number
flow. J . Fluid Mech. 96,315-334.
LADYZHENSKAYA, 0. A. 1969 T h Mathematical Theory of Viseous Incompressible Flow. Gordon &
Breach.
LANQLOIS, W. E. 1964 Slow Viscous Flow. Macmillan.
LEE,S. H. & LEAL,L. G. 1986 Low-Reynolds-number flow past cylindrical bodies of arbitrary
cross-sectional shape. J . Fluid Mech. 164, 401-427.
MOFFATT,H. K. 1964 Viscous and resistive eddies near a sharp corner. J . FZuid Mech. 18, 1-18.
MIJNSON,B. R., RANQWALLA, A. A. & MANN,J. A. 1985 Low Reynolds number circular couette
flow past a wavy wall. Phys. Fluids 28, 2679-26.
PAN,F. & ACRIVOS, A. 1967 Steady flow in rectangular cavities. J. Fluid Mech. 28, 643-655.
POZRIKIDIS, C. 1987 A study of peristaltic flows. J . Fluid Mech. 180, 515-527.
SOBEY,I. J. 1980 On flow through furrowed channels. Part 1. Calculated flow patterns. J . Fluid
Mech. 96,1-26.
SOBEY,I. J. 1983 The occurrence of separation in oscillatory flow. J . Fluid Mech. 134, 247-257.
TAYLOR, G. I. 1971 A model for the boundary condition of a porous material. Part 1. J . Fluid
Mech. 49, 319-326.
WANQ,C. Y. 1978 Drag due to a striated boundary in slow Couette flow. Phys. Fluids 21,697-698.
https://doi.org/10.1017/S0022112087001927 Published online by Cambridge University Press