Anda di halaman 1dari 23

SIAM REVIEW 

c 2000 Society for Industrial and Applied Mathematics


Vol. 42, No. 3, pp. 417439

Numerical Study of Flows of


Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Two Immiscible Liquids at Low


Reynolds Number
Jie Li
Yuriko Renardy

Abstract. We review our recent results on ngering, bamboo waves, and drop breakup in low Reynolds
number ows composed of two viscous liquids under shear.

Key words. volume-of-uid scheme, interface tracking, semi-implicit Stokes solver, two-layer ows,
low Reynolds number ows

AMS subject classications. 76D05, 65-02, 76E05

PII. S0036144599354604

1. Introduction. Many congurations are possible for the ow of two immisci-


ble uids: layers, ngers, encapsulated regimes, and drops [1, 2, 3]. We describe
our numerical simulations of these two-uid systems with the volume-of-uid (VOF)
scheme. In the case of the layered regime, we describe what happens when the two
liquids are moving past each other and form large amplitude waves, sheets, ngers,
and drops as a result of an interfacial instability driven primarily by the dierence in
viscosities. The jump in the viscosity from one uid to the other results in the jump
in the tangential velocity gradient across the interface [4] and can be thought of as
a viscous counterpart of the KelvinHelmholtz instability. In the case of a drop de-
forming under shear, we focus on the breakup regime, driven by shear acting against
the restoring force of interfacial tension.
2. Numerical Method. One of the greatest diculties in the study of two im-
miscible uid ows is that the domain of interest contains an unknown interface: the
interface moves from one location to another and may sometimes undergo severe de-
formations, including even breakup; the interface plays a major role in dening the
system and must be determined as part of the solution. In the numerical treatment
of the interface, we must answer three questions: (1) How do we represent the in-
terface on a mesh? (2) How will the interface evolve in time? and (3) How should
we apply boundary conditions on the interface? There are many interface tracking
methods, such as the moving grid method, the front tracking method, the level set
method, and the VOF method. The VOF method provides a simple way of treating
the topological changes of the interface, as well as ease of generalization to the three-
Received by the editors April 19, 1999; accepted for publication (in revised form) November
3, 1999; published electronically July 31, 2000. This work was supported by NSF-CTS 9612308,
NSF-INT9815106, and NCSA-CTS grants.
http://www.siam.org/journals/sirev/42-3/35460.html
Department of Mathematics, 460 McBryde Hall, Virginia Polytechnic Institute and State Uni-

versity, Blacksburg, VA 24061-0123 (jie@math.vt.edu, renardyy@math.vt.edu).


417
418 JIE LI AND YURIKO RENARDY

dimensional case. This approach was rst introduced by DeBar [5] in 1974, followed
by signicant advance from Youngss work [6] eight years later. More recent works
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

include [7, 8, 9, 10], and the review article of [11].


2.1. The Equations of Motion. We suppose that the two uids are immiscible
and that the density and the viscosity are constant in each uid, but we do allow
for the possibility of a jump across the interface. We use a concentration function C
to represent and track the interface:

1, uid 1,
(2.1) C(x) =
0, uid 2.

The concentration function is transported by the velocity eld u:


C
(2.2) + u C = 0.
t
The average values of the density and the viscosity are given by

(2.3) = C1 + (1 C)2 , = C1 + (1 C)2 ,

where subscripts refer to uids 1 and 2.


We suppose also that the ow is incompressible,

(2.4) .u = 0,

governed by the NavierStokes equation


 
u
(2.5) + u u = p + S + F,
t
where S is the viscous stress tensor
 
1 uj ui
Sij = + .
2 xi xj

The body force F includes the gravity and interfacial tension force [12, 7, 8, 13, 14, 15,
16]. The interfacial tension force is Fs = nS S , where is the interfacial tension,
is the mean curvature, and nS is the normal to the interface.
2.2. Temporal Discretization and Projection Method. The simultaneous solu-
tion of the large number of discrete equations arising from (2.4) and (2.5) is very
costly, especially in three spatial dimensions. An ecient approximation can be ob-
tained by decoupling the solution of the momentum equations from the solution of
the continuity equation by a projection method [17].
In the projection method, the momentum equations (2.5) are rst solved for an
approximate u without the pressure gradient, assuming that un is known:
u un 1
(2.6) = un un + ( (S) + F)n .
t
In general, the resulting ow eld u does not satisfy the continuity equation (2.4).
It is corrected by the pressure
un+1 u p
(2.7) =
t
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 419

z
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

x
y

zk

yj

x i

Fig. 2.1 Three-dimensional Cartesian mesh with variable cell sizes.

Wi,j,k-1/2

Pi,j
Vi,j-1/2,k
Ci,j

Ui-1/2,j,k

Fig. 2.2 Location of variables in a Mac mesh cell.

in order to yield a divergence-free velocity un+1 . Pressure is not known at this moment
but can be found from a Poisson equation,
 
p u
(2.8) = ,
t
which is obtained by taking the divergence of (2.7). In the problems we address
below, the boundary conditions for the velocity are periodicity and the Dirichlet
condition. Analogously, the boundary conditions for the pressure are periodicity and
the Neumann condition, respectively.
2.3. Spatial Discretization. An Eulerian mesh of rectangular cells is used, with
variable sizes: xi for the size of the ith mesh cell in the x-direction, yj for the size
of the jth mesh cell in the y-direction, and zk for the kth mesh cell in the z-direction
(Figure 2.1).
The momentum equations are nite-dierenced. As Figure 2.2 shows, the x-
component of velocity ui 12 ,j,k , the y-component of velocity vi,j 12 ,k , and the z
component of velocity wi,j,k 12 are centered at the right face, front face, and top face
of the cell, respectively, whereas the pressure pi,j,k is located at the center. This is
the so-called Mac method [18]. This mesh has the advantage that the pressure eld
420 JIE LI AND YURIKO RENARDY

does not allow spurious checkerboard oscillations. Another advantage of the Mac
method is that the Neumann condition for the pressure is automatically involved in
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

the numerical solution when the boundary condition for the velocity is the Dirichlet
condition [19]. No numerical boundary condition is needed for the pressure.
The convective terms appearing in (2.6) are treated with a nonconservative scheme,
DU L = (ui 12 ,j,k ui 23 ,j,k )/xi1 ,

DU R = (ui+ 12 ,j,k ui 12 ,j,k )/xi ,


 
u xi DU L + xi1 DU R
(2.9) u = ui,j,k ;
x i,j,k xi1 + xi

the derivatives are weighted by cell size such that the correct order of approximation
is maintained on a variable mesh. The pressure and viscous terms in the momentum
equation are calculated using second-order central nite dierences.
The numerical solution of (2.8) is the most time consuming part of our Navier
Stokes solver and, consequently, an ecient solution is crucial for the performance of
the whole method. Potentially, the multigrid method is the most ecient method.
The basic idea of the multigrid method [20] is to combine two complementary proce-
dures: one basic iterative method to reduce the high frequency error and one coarse
grid correction step to eliminate the low frequency error. We choose a two-color
GaussSeidel iterative method because it breaks the dependence between the variables
and therefore allows for parallelization of the scheme. We use a Galerkin method to
provide a good coarse grid correction.
2.4. Advection of the Interface. At the discrete level, the concentration function
is the volume fraction eld Cij : when a cell is lled by the uid 1, Cij = 1; when a
cell does not contain any of this uid, Cij = 0. The interfaces are in the cells with Cij
between 0 and 1. Given an interface, we can calculate a unique volume fraction eld.
On the other hand, when we represent the interface by a volume fraction eld, we lose
interface information and we cannot determine a unique interface; the interface needs
to be reconstructed. Piecewise linear interface calculation (PLIC) methods have been
developed in [7, 8] for two-dimensional and three-dimensional cases. The gist of these
methods is to calculate the approximate normal n to the interface in each cell, since
this determines one unique linear interface with the volume fraction of the cell. The
discrete gradient of the volume fraction eld yields
h C
(2.10) n= .
|h C|
A least-squares method [9] has also been implemented in this work. Numerical tests
suggest that this approximate interface is a second-order approximation.
The second step of the VOF method is to evolve the volume fraction eld C.
Because the interface evolution is governed by a transport equation, (2.2), the La-
grangian method is the natural choice [21, 7]. In this scheme, once the interface is
reconstructed, the velocity at the interface is interpolated linearly and then the new
interface position is calculated via xn+1 = xn + u(t). Figure 2.3 illustrates how
the Lagrangian method performs on an arbitrary two-dimensional mesh. In compari-
son with the Eulerian method, the Lagrangian method has the following advantages:
when the Courant condition (Max|u|)t/h < 1/2 is satised, this method is stable,
and the volume fraction always satises the physical constraint 0 C 1.
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 421
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 2.3 The Lagrangian method on an arbitrary two-dimensional mesh. The shaded polygon repre-
sents the part occupied by the uid in the central cell. The broken line shows the polygons
position after advection in the local velocity eld, represented by arrows. The uid is re-
distributed between neighboring cells, which the new polygon partially overlaps.

In the VOF method, the interfacial tension condition across the interface is ap-
plied not directly but rather as a body force over the cells that contain the interface.
Two such formulations have been implemented in this work. The rst is the continu-
ous surface force formulation [22], in which fs = nS and Fs = fs C. The second
is the continuous surface stress formulation [12], in which Fs = T = S nS and
T = [(1 nS nS )S ], which leads to a conservative scheme for the momentum
equation.
2.5. Semi-Implicit Stokes Solver. The above description completes the charac-
terization of our numerical method. However, one weakness is that it is an explicit
method and not suitable for simulations of low Reynolds number ows. For an explicit
method, the time step t should be less than the viscous time scale, T = h2 /,
where h denotes the mesh size. This stability limit is much more restrictive than the
CourantFriedrichsLevy (CFL) condition for simulations of low Reynolds number
ows. In order to carry out calculations for times of order 1, the implicit treatment
of the viscous terms is imperative.
Our approach is an original semi-implicit method. The time integration scheme
is constructed to be implicit for the Stokes operator and is otherwise explicit. In the
u component of the momentum equation (2.6), we treat only the terms related to u
(the terms with upper index ) implicitly and leave the other terms (the terms with
upper index n ) in the explicit part, that is,

u un 1 1 1 n
= (un )un + n F1n + n (2n ux ) + n ( uy + n vxn )
t x y
1 n
(2.11) + ( uz + n wxn ).
n z

This can be expressed as


       
t
(2.12) I 2 + + u = explicit terms.
x x y y z z

This procedure decouples the u component from the previous parabolic system. This
is still a rst-order method, but the eort needed to solve (2.12) is signicantly reduced
422 JIE LI AND YURIKO RENARDY

in comparison with the full implicit treatment. The same idea applies to the other
velocity components.
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

As far as the viscous terms are concerned, our semi-implicit scheme is uncondi-
tionally stable [23, 24]. In addition, the factorization technique of [25] is applied to
the left-hand side of (2.12):
           
t t t
I 2 I I
x x y y z z
u = explicit terms.

The error of this factorization is O(t3 ). The inversion of the left-hand side of the
above equation requires solving only tridiagonal matrices, which results in a signicant
reduction in computing and memory. In fact, the solution of these tridiagonal systems
can be done in only O(N ) operations (where N is the grid point number). In summary,
this scheme is rst-order accurate and unconditionally stable. The stability of this
scheme is crucial for simulation of low Reynolds number ow.
The direct simulation of two-uid ow is often limited by computing cost and
machine memory, especially in the three-dimensional case. Three issues have been of
the utmost importance in improving performance: accuracy, stability, and eciency.
The entire code (including the semi-implicit Stokes solver) is parallelized: on the
Origin2000 with 16 parallel processors, the eciency of our code is more than 80%.
3. Simulation of Two-Layer Couette Flow. Flows composed of two immiscible
liquids and undergoing shearing motions can form ngers as a result of an interfacial
instability due primarily to the viscosity jump. We attempt to capture the qualitative
features of ngering by simulating two-layer Couette ow, which is one of the simplest
of all the shearing ows of two uids one might consider. A base ow solution is shown
in Figure 3.1. The lower uid is uid 1 (with viscosity 1 ) and occupies 0 < z < l1 ;
uid 2, with viscosity 2 , occupies l1 < z < 1. There are four parameters: the
viscosity ratio m = 1 /2 ; the average depth of the lower liquid l1 ; the interfacial
tension parameter T = /(2 Ui ), where is the interfacial tension coecient and Ui
is the dimensional interfacial speed of the base ow; and a Reynolds number based on
the lower uid R1 = Ui l 1 /1 , where l is the dimensional plate separation. If the
densities are equal, then gravitational force is balanced by a pressure gradient and
can be neglected. If the densities are not equal, then two additional parameters are
the Froude number F , where F 2 = Ui2 /(gl ), and the density ratio r = 1 /2 .
3.1. Stability of Two-Layer Couette Flow. We explore the evolution of small
amplitude disturbances, the weakly nonlinear regime, saturation to spatially periodic
traveling waves, and ngering of large amplitude periodic disturbances. In order to
achieve the computational results, we have implemented the semi-implicit scheme of
section 2.5 to enable faster computations at low Reynolds numbers and a second-order
velocity interpolation for greater accuracy of the interface advection.
The stability of two-layer Couette ow pictured in Figure 3.1 for small perturba-
tions of wavenumber has been addressed in [26, 27, 1]. When the initial condition
is an eigenmode, the full numerical simulations yield the correct growth rates for the
interface, as well as for the L -norm and L2 -norm of the velocity eld. Mesh conver-
gence studies were conducted for small initial amplitudes and also with dierent time
steps. This comparison with linear theory provides a test for our code [23].
At low speeds, the ow can become unstable due to an interfacial instability that
arises from the jump in shear rates across the interface. This ow instability is a Hopf
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 423
z

z=1
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fluid 2
z=l
1
interface
Fluid 1
y
x
z=0

Fig. 3.1 Flow schematics for two-layer Couette ow.

bifurcation. The weakly nonlinear theory of [27, 13] yields a StuartLandau equation
for the amplitude function Z(t) of the primary mode, dZ/dt Z = |Z|2 Z, where
denotes the Landau coecient, and the critical eigenvalue from the linearized
stability analysis. The dynamics just above the onset of instability is determined by
the interaction of the primary mode with the mean ow mode and the second har-
monic and by a cubic self-interaction. The traveling wave solution Z(t) = exp(ict)Z0
is predicted to saturate when the real part of the Landau coecient is negative.
The saturation amplitude and wave shape at saturation are also predicted by the
theory.
3.2. Long-Time Saturation. In the laboratory experiments [28], the long-time
saturation predicted by Hopf bifurcation theory is reached after a time on the order
of 1000 s. The lower uid is a water-glycerine (32/68%) mixture with viscosity 0.0191
Pas, density 1169 kg/m3 ; the upper uid is mineral oil with viscosity 0.0297 Pas and
density 846 kg/m3 . Interfacial tension is 0.03 Pa m. The onset condition is modeled
in [29], with upper plate velocity 0.44 m/s, channel depth 20mm, wavelength 6.8 cm,
and depth of lower uid 12.74 mm. Our parameters for this condition [27] are shown
in the caption of Figure 3.2. Qualitative features of this ow are successfully captured
on a 256 256 mesh: Figure 3.2(a) illustrates the evolution of the interface amplitude
vs. time, showing saturation after a time of order 1000 s. Figure 3.2(b) shows the
simulated wave shape at 2500 s with at crests and sharp troughs, in agreement with
the weakly nonlinear theory and the experimental observations.
3.3. Two-Dimensional Fingers or Sheets. Large amplitude perturbations typi-
cally lead to the formation of ngers.
3.3.1. Moderate Reynolds Number. Figure 3.3 shows the case of viscosity ratio
m = 0.5, equal density, T = 0.01, Reynolds number R1 = 500, and wavenumber
= /2. The initial amplitude is A(0) = 0.05. Interface proles are plotted for
t = 0, 5, 10, 12, 13, 14, 15, 20. The interface moves vertically under the viscosity-jump
instability and is elongated horizontally by the base shear ow. The two uids do
not penetrate into each other in the same manner. Note that the upper uid is more
viscous than the lower one. The qualitative features of this ow are retained if the
amplitude is increased, but the sequence occurs faster for larger amplitudes.
3.3.2. Low Reynolds Number. Figure 3.4 shows the case of a low Reynolds
number, R1 = 40, with wavenumber = 6.3. The initial amplitude of the interface
is A(0) = 0.05. Interface proles for time t = 0, 2, 5, 8, and 10 are plotted. One
424 JIE LI AND YURIKO RENARDY
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

A(t)

(a) time

(b) Waveform

Fig. 3.2 Simulation of saturation for the experiments in [28], R1 = 394, T = 3.14, F 2 = 0.528,
density ratio r = 1.4, viscosity ratio m = 0.65, depth ratio l1 = l1 /l = 0.637, dimension-
less wavenumber = 1.9. (a) Maximum of interface amplitude against time. (b) Wave
shape at time t = 2500.

1 1
t=0 t=13

0.5 0.5

0 0
0 1 2 3 4 0 1 2 3 4
1 1
t=5 t=14

0.5 0.5

0 0
0 1 2 3 4 0 1 2 3 4
1 1
t=10 t=15

0.5 0.5

0 0
0 1 2 3 4 0 1 2 3 4
1 1
t=12 t=20

0.5 0.5

0 0
0 1 2 3 4 0 1 2 3 4

Fig. 3.3 Sequence of interface positions for A(0) = 0.05, = /2, Re1 = 500, m = 0.5, l1 = 0.372,
T = 0.01, equal densities, and zero gravity. The calculation was carried out on a 512 512
mesh. The ow domain is 4 1 with periodicity in the x-direction.

striking fact compared to the ow with Reynolds number R1 = 500 is that the vertical
growth of the ngers is small and the horizontal elongation is large. This is because
there are two competing mechanisms governing the shape of the interface. First, the
growth rate of the instability originating from the viscosity stratication contributes
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 425
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

t=0

t=2

t=5

t=8

t = 10

Fig. 3.4 Sequence of interface positions for A(0) = 0.05, = 6.3, Re1 = 40, m = 0.5, l1 = 0.372,
T = 0.01, equal densities, and zero gravity. t = 0, 5, 10, 15, and 20. The calculation was
carried out on a 160 160 mesh. The vertical axis is magnied from 0.25 to 0.50 to show
the details; the ow domain is 1 1 with periodicity in the x-direction.

to the vertical growth of the interface; this tends to zero as the Reynolds number
tends to zero. Second, the base shear ow of Figure 3.1 convects the large, initially
sinusoidal interface, as illustrated in Figure 3.5; the crest of the wave moves forward
faster than the trough. At low Reynolds numbers, the second mechanism is dominant
and initiates the ngering that is observed in Figure 3.4. Elongated two-dimensional
ngers have also been recorded [30] for creeping ow.
3.3.3. Three-Dimensional Fingers. The response to periodic perturbations in
three dimensions is analyzed by using a horizontal undulation of the phase in the
initial condition. The initial interface is

(3.1) z = l1 + Ax (0) cos(x x + (y)), = Ay (0) cos(y y),

where Ax (0) is the two-dimensional perturbation amplitude, Ay (0) the spanwise per-
turbation amplitude, x the x-direction wavenumber, and y the y-direction wavenum-
ber. The perturbation formulation is inspired by the experiment of [31] on shear
layers.
The two-dimensional ngering study in the previous section provides useful in-
formation for the subsequent three-dimensional investigation. As we have seen, small
426 JIE LI AND YURIKO RENARDY

moving wall
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

x
wall at rest

Fig. 3.5 Large amplitude disturbance in creeping ow leads to ngering because the trough is left
behind while the crest hurries on.

Fig. 3.6 Simulation of two-layer Couette ow for Reynolds number R1 = 500, = /2, m = 0.5,
l1 = 0.372, T = 0.01, equal densities, and zero gravity. The initial interface height is
z = 0.372 + 0.05 cos{6.3x + 0.1 cos[y/(4)]}. The interface position is shown at t = 12. A
nger forms in the low viscosity uid. The three-dimensional nger is in the process of
breaking up into a series of drops of the low viscosity uid.

perturbations to low Reynolds number ow involves small interface structures, and


a ne mesh is needed to capture them. In addition, a three-dimensional simulation
is limited by machine memory and computation time. Figure 3.6 shows the case of
Reynolds number R1 = 500. The spanwise wavenumber is selected to be y = 4;
therefore, the computational domain is a 40.51 box, and the mesh is 12832128.
As in the previous section, the initial two-dimensional amplitude is Ax (0) = 0.05 for
a ow with wavenumber x = , R1 = 500, viscosity ratio m = 0.5, undisturbed
interface height l1 = 0.372, and T = 0.01. We use a spanwise amplitude Ay (0) = 0.1.
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 427
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 4.1 (a) Schematic of core-annular ow. (b) Axisymmetric waves.

In Figure 3.6, we have inverted our three-dimensional visualization box vertically to


provide a better view. The nger is in the lower uid; the case t = 12 shows this
nger to be breaking up, yielding a series of drops of the low viscosity uid.
A droplet will pinch o in a VOF scheme when the neck has a thickness on the
order of a grid cell. It is not clear, then, that a numerically observed droplet breakup
corresponds to a physical one, unless a renement study is conducted. In [32] it
was observed that at low resolution for a two-dimensional computation of two-layer
Couette ow, drops were observed to form, while at high resolution, these droplets
were replaced by an elongated nger or lament. The conclusion here is that when
drops break o of laments and are of the size of the mesh, then a renement needs
to be done in order to ascertain whether the drops are numerical or physical. This
problem is inherent in the VOF approach. There is a dierence between the breakup
into drops in the two-dimensional case and the three-dimensional Couette ow. The
breakup recorded in [23] for two-dimensional Couette ow is not physical because
there is no breakup mechanism in two dimensions. Hence, the numerical renement
in [32] resolved the issue as thin ngers. Drop breakup in three dimensions, however,
is physical for two reasons. First, there is clearly an interfacial tension driven breakup
mechanism, and the VOF results are similar to the more accurate boundary integral
scheme for neck formation [33]. Second, a numerical resolution study for drop breakup
is shown in section 5.3 in reference to Figure 5.3, which compares satisfactorily with
its renement in Figure 5.4.

4. Simulation of Core-Annular Flow. Core-annular ow is a pressure-driven


ow through a pipe of one uid at the core and another uid in the annulus (Figure
4.1). This arrangement arises naturally for uids with markedly dierent viscosities,
because higher viscosity material tends to become encapsulated by lower viscosity
material. An industrial application is the lubricated pipelining of crude oil with the
addition of water [2, 3]. The purpose is to eciently transport a very viscous liquid,
which on its own would require costly work. However, when the viscous uid just along
the wall is replaced by a much less viscous immiscible one, in this case water, then
the work required for transportation is signicantly lowered. The ideal arrangement
has a perfectly cylindrical interface (Figure 4.1(a)) and is called perfect core-annular
ow (PCAF), but a wavy interface is also viable (Figure 4.1(b)).
428 JIE LI AND YURIKO RENARDY

4.1. Parameters. Consider the PCAF of Figure 4.1(a). The dimensionless pipe
radius is a = R2 /R1 , where R2 is the dimensional pipe radius and R1 is the radius of
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

the undisturbed interface. The pressure gradient in the axial direction is a constant:
dP /dx = f . The viscosity ratio is denoted m = 2 /1 , the radius ratio is a, the
density ratio is = 2 /1 , and the ratio of driving forces in the core and annulus is
K = (f + 1 g)/(f + 2 g). The base velocity eld is in the axial direction and is
a function of the radial variable. An interfacial tension parameter is J = R1 1 /21 ,
where denotes the interfacial tension. Reynolds numbers Rei are dened by Rei =
i V0 R1 /i , i = 1, 2, where V0 denotes the centerline axial velocity. In summary,
PCAF is characterized by six dimensionless parameters: m, a, , J, K, and Re1 = Re.
PCAF can lose stability to a variety of regimes. Several qualitatively dierent
regimes of ow have been documented in the experiments of [34], which were con-
ducted for a pipe diameter of 3/8 inch, m = 1/601, = 1.10, R2 = 3/16 inch,
= 8.54 dyn/cm, and aJ = 0.102. The oil at the core is lighter than the water in the
annulus, so that buoyancy and the pressure gradient act in the same sense in up-ow,
where the core oil is observed to produce bamboo waves, and in the opposite sense in
down-ow, where the core is compressed and typically buckles into corkscrew waves
[35]. The volume ow rates of water and oil are denoted by Qw and Qo , respectively.
The supercial water velocity is dened by Vw = Qw /A and the supercial oil velocity
by Vo = Qo /A. Here, A = R22 is the cross-sectional area of the pipe. Conservation
of volume implies that the parameter a will remain constant as the PCAF evolves
into the nonlinear regime. The evolution is, however, more complicated for supercial
velocities Vo and Vw .
The hold-up ratio h, dened as the ratio of the input oil/water ow rate ratio
to the in situ oil/water volume ratio, is an important practical parameter for core-
annular ow. Experiments show that h is constant in up-ow and fast ows. The
radius ratio a is related to the ow rates via the hold-up ratio h. If the ow is perfectly
core-annular, then it can be shown that
 
(4.1) a = 1 + hQw /Qo = 1 + hVw /Vo .

4.2. Bamboo Waves. Our motivation for studying bamboo waves in vertical
core-annular ow is that these structures are well documented in the experiments
of [34]. In the bamboo wave regime, trains of Stokes-like waves with sharp crests
are connected by long laments. The waves are axisymmetric and occur in a very
robust regime of up-ow. The main dierence between up-ow and down-ow is that
in down-ow, the driving pressure gradient and gravity act in the same direction,
making the heavier uid, water, fall while the buoyancy holds the oil back; in up-ow,
gravity hinders the water and the oil is encouraged to ow upward. Naturally, if
the driving pressure gradient is suciently strong and dominant, then the dierence
between up-ow and down-ow vanishes.
The ow regime denoted #1 in the ow chart of [34] has the supercial oil velocity
Vo = 10 cm/s and a = 1.28. This is shown on the right of Figure 4.2. The other
parameters for this ow are given in the gure caption. The experimental snapshot
shows the coexistence of waves with wavenumbers roughly between 1.5 and 2.0. Since
the observed wavespeeds and wavelengths of the bamboo wave regimes are close to
the linear theory for stability of PCAF, the initial condition for our numerical simu-
lation is seeded with an eigenmode [34, 36, 2, 35, 24]. When perturbations grow to
larger amplitudes, saturation to the fully nonlinear bamboo waves is achieved. Figure
4.2 shows a direct comparison of the saturated waves from our simulations with a
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 429
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 4.2 Up-ow for Re = 0.93, a = 1.28, m = 0.00166, = 1.1, J = 0.0795, and K = 0.4552. The
photograph on the right was experimentally obtained in [34]. Our numerical simulations
are on the left, for dimensionless wavenumbers = 1.5, 1.75, 2.0.

photograph from experiments. Previously, a steady solution was calculated in [37]


under the assumptions of solid core and density matching; their interface shape is too
rounded and asymmetric compared to the experimental snapshot, which shows an
almost symmetric form of the crest, with a pointed peak. The crest is slightly sharper
on the front and less sharp on its back. These details are successfully reproduced in
our results.
4.3. The Hold-Up Ratio. The parameters for Figure 4.2 can be questioned for
accuracy regarding the Reynolds number and K. This is because the experimental
value for the pressure gradient is not known; rather, the supercial velocities are
given in [34, 37]. The reconstruction of the rest of the parameters leaves some room
for guesswork. In fact, our investigation of the hold-up ratio at saturation shows that
the experiments are probably closer to a larger Reynolds number: Figure 4.3 shows
the base velocity eld at Re = 3.0, K = 0.9993. This is fully up-ow, in both uids.
We performed a calculation for this ow from an initial amplitude A(0) = 0.005 on
a 256 256 mesh. The wavenumber is 2. The evolution of the hold-up ratio against
time is shown in the lower plot. The hold-up ratio decreases dramatically between
t = 20 and t = 40. This period corresponds to the transition from the linear regime,
in which the eigenmode grows according to the growth rate predicted by linearized
stability theory for PCAF, to a nonlinear regime. In the fully nonlinear bamboo wave
regime, the hold-up ratio stabilizes around 1.46, which is a reasonable approximation
of the experimental value 1.39.
The interface position after saturation is plotted in Figure 4.4(a) and is very
similar to the one in Figure 4.2 for = 2, Re = 0.93, with the crest a little more
430 JIE LI AND YURIKO RENARDY
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 4.3 The upper plot shows the velocity prole of PCAF for Re = 3, a = 1.28, m = 0.00166,
= 1.1, J = 0.0795, K = 0.9993, and wavenumber = 2. The lower plot shows the
calculated hold-up ratio h against time. At t = 0, the PCAF hold-up ratio h = 2.61. As
PCAF evolves, h decreases rst and stabilizes around 1.46.

pointed. The streamlines and contours of the pressure eld are shown in Figure
4.4(b). For an incompressible uid, the axisymmetric stream function is dened
by u = (1/r)(/x), and v = (1/r)(/r). This is calculated numerically by a
standard central dierence scheme. In the left half of Figure 4.4(b), the contours of
the pressure eld are plotted. In the water, the pressure eld reaches its maximum
value above the crest, at location A, and its minimum value below the crest, at B.
The force that arises from this pressure eld in the water and the buoyancy in the
oil work together in up-ow, where they lead to stretching of the core into bamboo
waves (cf. the cartoon of Figure 15.5 in [2]). The pressure eld in the oil core is also
shown in the gure. In the right half of Figure 4.4(b), the streamlines are plotted in
the frame of reference moving with the oil core. C denotes the stagnant line that is
at rest in the reference frame. This stagnant line distinguishes the set of streamlines
into two categories. One category of streamlines is inside the stagnant line and forms
a recirculation zone. Another category of streamlines is outside the stagnant line.
While some of them are completely in the water, others enter into the oil on the
upper side of the crest and return to the water on the down side of the crest. The
behavior of these streamlines is due to the fact that the waves move slowly compared
to the oil core. The wavespeed is 10% smaller than the core velocity. Although the
waveform is stationary, the velocity and pressure elds are not.
4.4. Effect of Radius Ratio and Temporally Periodic Flow. In the previous ex-
ample, the radius ratio is a = 1.28, so that the oil core is relatively close to the
pipe wall and the interaction between them is strong, leading to the vortices found
near the wave crests. For the experimental data point #1 in the ow chart of [34],
a = 1.61, J = .063354, K = 2.0303, and Re1 = 3.73754. Linear theory shows that
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 431
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

A
B

(a) (b)

Fig. 4.4 Up-ow for Re = 3.0, a = 1.28, m = 0.00166, = 1.1, J = 0.0795, and K = 0.9993. (a)
Interface shape at t = 100. (b) Pressure contours on the left half. Pressure eld reaches
the maximum value on the upper side of the crest, at location A and the minimum value on
the down-side of the crest, at location B. Streamlines in the frame of reference moving with
the oil core are shown on the right half. The broken line represents the axis of symmetry.

the most dangerous mode for this ow has dimensionless wavenumber = 2.4. We
set the initial amplitude of perturbation A(0) = 0.01. The evolution of the maximum
amplitude A(t) against the time is plotted, on a log10 -linear scale, in the upper plot of
Figure 4.5. The calculation is carried out on a 256 256 mesh over one spatial period
on a domain [0, 1.61] [0, 2.618]. The initial growth of the perturbation compares well
with the predicted linear theory. After roughly 40 seconds, a fully nonlinear evolution
begins to take shape. Linear theory yields a wavespeed 0.9542, and the computed
wavespeed is 0.8068 at time t = 200 (middle plot of Figure 4.5).
The most surprising feature of this ow is that it demonstrates temporal peri-
odicity, as evident from the evolution of the hold-up ratio vs. time (lower plot of
Figure 4.5). As the PCAF evolves into the nonlinear bamboo wave regime, the hold-
up ratio rst decreases and then oscillates around 2.15, with a well-dened temporal
period. The period is about 8.
4.5. Effect of Reynolds Number and Temporally Periodic Flow. As the speed
is increased, the bamboo waves shorten and peaks become more pointed. This is illus-
trated in Figure 4.6. Based on the linear stability theory for PCAF, the wavenumber
for the fastest growing mode decreases as the Reynolds number increases. Figure 4.6
shows that the slow and long waves (low Reynolds number ow) have asymmetrical
crest shape, at on the upper side and steeper on the down side. This asymmetry
is due to the eect of the buoyancy. In fast ow (large Reynolds number ow), the
asymmetric eect of buoyancy is less important, as evidenced by the almost symmetric
shape of wave crests for Re = 3.74.
432 JIE LI AND YURIKO RENARDY
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 4.5 Up-ow for = 2.4, Re1 = 3.73754, a = 1.61, m = 0.00166, = 1.1, J = .063354, and
K = 2.030303. For the interfacial mode, theoretical linear growth rate is 0.066 and the
wave speed is 0.954. The upper plot shows maximum amplitude A(t) vs. time on a log10 -
linear scale. The middle plot shows wave crest position vs. time. The lower plot shows
hold-up ratio vs. time. The hold-up ratio oscillates around 2.15, with period approximately
8. The solid line represents the theoretical linearized growth rate, and circles represent the
calculation carried out on a 256 256 mesh.

At Re = 3, a temporally periodic streamline pattern with period of about 10 is


found. While the periodicity appears already in low Reynolds number ow (Re = 1.0,
1.5, and 2.0), it becomes quite marked in the Re = 3 ow. When the Reynolds number
is increased, it is not surprising that a steady solution would lose stability to time-
dependent and eventually chaotic solutions as the ow transitions. However, it is sur-
prising that the time-dependent bamboo waves would appear to the eye to be steady.

5. Simulation of Drop Breakup in Three Dimensions. The deformation and


breakup of a drop in shear ow lies at the foundation of dispersion science and mul-
tiphase ow and has invoked a great deal of interest in the scientic community since
the time of G. I. Taylor [38, 39, 40, 41, 42, 43, 44, 45, 46]. More recently, experimen-
tal observations of the sheared breakup have been recorded in [43]: a strong shear
is applied to a single drop, which elongates and undergoes end pinching via a pro-
cess termed elongative end pinching as opposed to retractive end pinching, studied
in [47]. These processes, which yield daughter drops, are paradigms of theoretical
investigations into emulsication and mixing [46, 42, 48, 49].
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 433
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Re = 1.0 Re = 1.5 Re = 2.0 Re = 3.0 Re = 3.7

Fig. 4.6 Up-ow for a = 1.61, m = 0.00166, = 1.1, J = 0.063354, and K = 2.030303.

The experimental work of [43] focuses on a viscous drop suspended in a second


immiscible liquid (the matrix liquid) in a cylindrical Couette device. The dierence
in density between the two liquids is a minor eect, and the ow is suciently slow,
so that centrifugal eects in the cylindrical device are not important. A theoretical
model for this is simply three-dimensional Couette ow with zero gravity, as shown
in Figure 5.1. The liquid drop has an undeformed radius a and viscosity d , and the
matrix liquid has viscosity m . The matrix liquid is undergoing a simple shear ow
between two parallel plates, placed a distance d apart. The undisturbed velocity eld
is u = zi, where is the imposed shear rate. Additional parameters for our numerical
simulations are the interfacial tension and the spatial periodicities x and y in the
x- and y-directions, respectively. There are six dimensionless parameters: a capillary
number Ca = am /, where an average shear rate is dened as = U /d ; the
viscosity ratio = d /m ; the Reynolds number Re = m a2 /m ; the dimensionless
plate separation d = d /a; and dimensionless spatial periodicities x = x /a and
y = y /a. The shear stress induced by the ow competes with the interfacial tension
to deform and rupture the drop. The capillary number denotes the ratio between these
two competing eects and provides a useful measure of eciency of the shear ow to
deform the drop.
Computational studies [33] have elucidated regimes where the drop deforms to
the point of breaking, but results on the motion past breakup are limited. We present
434 JIE LI AND YURIKO RENARDY

z
plate
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

fluid 2 fluid 1

plate x

Fig. 5.1 A drop in simple shear ow, bounded at the top and bottom by walls. The y-axis points
into the paper.

L-B
D=
L+B L
x
a
B
-z

Fig. 5.2 Scalar measures of deformation and orientation.

a numerical exploration of the stages in the formation of daughter drops under shear.
The simulations were conducted as three-dimensional initial value problems [16].
5.1. Drop Breakup in Simple Shear. In the past, all numerical studies of a vis-
cous drop in shear ow have been performed with the boundary integral method,
combined with a front tracking method. An advantage of the front tracking method
is the use of marker particles to track the interface explicitly. However, the implemen-
tation of boundary integral methods poses a major obstacle, because it is dicult to
handle merging and folding interfaces. The boundary integral method incorporates a
simple shear ow out to innity. In [16], the limitation of this assumption has been
investigated by changing the plate separation.
5.2. Evolution to Steady Drop Shape. Two dimensionless parameters, the cap-
illary number Ca and the viscosity ratio , characterize the behavior of the suspended
drop, provided the Reynolds number Re = m a2 /m is small. Previous studies have
shown that when is less than 4, there is a critical capillary number Cac , above
which the drop disintegrates. Below the critical capillary number, the drop evolves to
a steady shape. The critical capillary number in shear ow is lowest for 0.5; for
= 1, Cac 0.41 is critical [50]. In the case where the drop evolves to a steady shape,
two parameters have been used to measure the deformation attained by the drop in
its nal stage. The rst is the Taylor deformation parameter, D = (L B)/(L + B),
where L and B are the half-length and half-breadth of the drop, respectively. The
second parameter is the angle of orientation of the drop with the axis of shear strain
(Figure 5.2).
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 435
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 5.3 Evolution of drop shape for Ca = 0.42 in domain 3 1 2. Viscosities and densities are
equal, Re = 0.0.

The initial condition is shown in Figure 5.1: the drop is a sphere. The no-slip
condition is imposed on the top and bottom plates and periodic conditions in the x-
and y-directions. Constant velocities are imposed on the top and bottom plates such
that the shear rate is constant during the entire computation. The steady parameters
D and of the previous works [44, 51] were retrieved for the computational domain
2 1 4. When the plate separation is smaller, e.g., 2 1 1 domain, the drop is
found to break up at lower capillary numbers.

5.3. Rupture of a Drop in Shear. When the shear rate is increased past a critical
value, the drop ruptures. The critical capillary number is roughly 0.41. Indeed, our
computation predicts an unsteady solution for Ca = 0.42 in the domain 3 1 2.
Figure 5.3 shows the evolution of the drop shape on a 96 32 64 mesh grid. The
drop continues to deform and eventually breaks up. The competition between the
externally imposed shear ow and the surface tension driven ow is clearly evident
in the gures. Initially, the most noticeable motion is the elongation of the drop,
stretched by the viscous shear stress of the external ow (time T = 0.0, 10.0, and
20.0). To time T = 30, we see clearly that a waist is formed near the center of the
drop, and the drop continually thins. The drop is beginning to lengthen slowly, and a
visible neck is formed near the bulbous end. This neck will eventually lead to the ends
pinching o [52], and the remaining liquid thread in the middle will form some small
satellite droplets. The number of these satellite droplets and their volume cannot be
computed precisely at this time, because the mesh grid is not suciently ne, but we
see here that the VOF method provides the possibility for this kind of investigation
in the future.
436 JIE LI AND YURIKO RENARDY

1.3

1.2
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

1.1
T = 36.0 1

0.9
0.8
1.4 2 2.7

1.3
1.2

1.1
T = 38.0 1

0.9
0.8
1.4 2 2.7

1.3

1.2
1.1
T = 38.5 1
0.9
0.8
1.4 2 2.7

1.3

1.2
1.1
T = 39.0 1
0.9
0.8
1.4 2 2.7

Fig. 5.4 Velocity elds during the breakup of a drop in the simple shear ow for capillary number
Ca = 0.42. This shows a cross-sectional cut.

To examine the breakup procedure more carefully, we have done the calculation
on a 19264128 mesh grid. We present the velocity eld on the cross-sectional cut in
the x-z plane in Figure 5.4; the ow pattern is symmetrical, and we need to show only
the right half-eld. The precise role of the surface tensiondriven ow during breakup
can be examined from this gure. At time T = 36, the result of the competition
between the external ow and the surface tension force is a vortical motion inside the
bulbous end of the drop, except near the neck; the surface tension force drives a fast
ow motion toward the bulbous end, while in the waist near the center, the ow is
much weaker. The consequence is that the neck quickly and continually narrows (the
neck has the same size as the mesh grid at this time), while the width of the central
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 437
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Fig. 5.5 Reproduction from Marks [43], showing a typical breakup. Matrix viscosity 7.0 Pas, drop
viscosity 4.3 Pas, interfacial tension 10.7 mN/m, initial drop radius 0.048 cm, shear rate
2.17/s, and equal densities. Reprinted with permission.

Fig. 5.6 Interface evolution as viewed from the top of the computational box 12 1 1 during
breakup for Ca = 0.55, = 0.77, and equal densities. The calculation was performed on a
256 64 64 grid.

waist remains almost unchanged at time T = 38. At T = 38.5, the drop breaks up
at the neck and produces a main drop and a middle liquid thread. More resolution is
required to investigate the breakup of the middle liquid thread.
The deformation and breakup of a drop in shear ow has recently been investi-
gated experimentally by Marks [43]. Figure 5.5 shows one of Markss observations,
and Figure 5.6 shows our simulation for parameters close to the experimental case.
Each breakup experiment of [43] began with the largest daughter drops being formed
by the elongative end pinching, which is shown in both of these gures. Subsequently,
there is a sequence of small, then large drops, as both these gures show. The reader
is referred to [16].
6. Summary. Our investigation of two-layer Couette ow has shown the devel-
opment of elongated ngers, which allow the migration of the more viscous into the
less viscous liquid, and vice versa, depending on the uid parameters. Our numerical
simulations of axisymmetric core-annular ow show that bamboo waves arise from lin-
early unstable small amplitude perturbations on core-annular ow; the disturbances
grow and saturate at fully nonlinear waveforms. In order to proceed further with the
drop breakup problem and explore the details of the breakup regime and daughter
438 JIE LI AND YURIKO RENARDY

drops, we need to enhance the eciency of our scheme, for example, with an adaptive
mesh technique [53].
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

REFERENCES

[1] D. Joseph and Y. Renardy, Fundamentals of Two-Fluid Dynamics, Part I, Springer-Verlag,


New York, 1993.
[2] D. Joseph and Y. Renardy, Fundamentals of Two-Fluid Dynamics, Part II, Springer-Verlag,
New York, 1993.
[3] D. D. Joseph, R. Bai, K. Chen, and Y. Y. Renardy, Core-annular ows, Ann. Rev. Fluid
Mech., 29 (1997), pp. 6590.
[4] C. S. Yih, Instability due to viscosity stratication, J. Fluid Mech., 26 (1967), p. 337.
[5] R. DeBar, Fundamentals of the KRAKEN Code, Technical Report UCIR-760, Lawrence Liv-
ermore National Laboratory, Livermore, CA, 1974.
[6] D. L. Youngs, Time-dependent multi-material ow with large uid distortion, in Numerical
Methods for Fluid Dynamics, Academic Press, New York, 1982, pp. 273285.
[7] J. Li, Calcul dinterface ane par morceaux (Piecewise Linear Interface Calculation), C. R.
Acad. Sci. Paris Ser. II, 320 (1995), pp. 391396.
[8] J. Li, Resolution numerique de lequation de Navier-Stokes avec reconnexion dinterfaces.
Methode de suivi de volume et application a latomisation, Ph.D. thesis, Universite Pierre
et Marie Curie, Paris, 1996.
[9] E. J. Puckett, A. S. Almgren, J. B. Bell, D. L. Marcus, and W. J. Rider, A higher order
projection method for tracking uid interfaces in variable density incompressible ow, J.
Comput. Phys., 130 (1997), pp. 269282.
[10] W. J. Rider and D. B. Kothe, Reconstructing volume tracking, J. Comput. Phys., 141 (1998),
pp. 112152.
[11] R. Scardovelli and S. Zaleski, Direct numerical simulation of free surface and interfacial
ow, Ann. Rev. Fluid Mech., 31 (1999), pp. 567604.
[12] B. Lafaurie, C. Nardone, R. Scardovelli, S. Zaleski, and G. Zanetti, Modelling merging
and fragmentation in multiphase ows with SURFER, J. Comput. Phys., 113 (1994), pp.
134147.
[13] A. V. Coward, Y. Renardy, M. Renardy, and J. R. Richards, Temporal evolution of
periodic disturbances in two-layer Couette ow, J. Comput. Phys., 132 (1997), pp. 346
361.
[14] D. Gueyffier, J. Li, A. Nadim, R. Scardovell, and S. Zaleski, Volume of uid interface
tracking and smoothed surface stress methods applied to multiphase ow and pendant drop
pinching, J. Comput. Phys., 152 (1999), pp. 423456.
[15] S. Popinet and S. Zaleski, A front-tracking algorithm for the accurate representation of
surface tension, Internat. J. Numer. Methods Fluids, 30 (1999), pp. 775793.
[16] J. Li, Y. Renardy, and M. Renardy, Numerical simulation of breakup of a viscous drop in
simple shear ow with a volume-of-uid method, Phys. Fluids, 12 (2000), pp. 269282.
[17] A. J. Chorin, A numerical method for solving incompressible viscous ow problems, J. Comput.
Phys., 2 (1967), pp. 1226.
[18] F. H. Harlow and J. E. Welsh, Numerical calculation of time-dependent viscous incompress-
ible ow with free surface, Phys. Fluids, 8 (1965), pp. 21822189.
[19] R. Peyret and T. D. Taylor, Computational Methods for Fluid Flow, Springer-Verlag, New
York, 1990.
[20] A. Brandt, Multi-level adaptive solutions to boundary-value problems, Math. Comput., 31
(1977), pp. 333390.
[21] A. J. Chorin, Flame advection and propagation algorithms, J. Comput. Phys., 35 (1980), pp.
111.
[22] J. U. Brackbill, D. B. Kothe, and C. Zemach, A continuum method for modeling surface
tension, J. Comput. Phys., 100 (1992), pp. 335354.
[23] J. Li, Y. Renardy, and M. Renardy, A numerical study of periodic disturbances on two-layer
Couette ow, Phys. Fluids, 10 (1998), pp. 30563071.
[24] J. Li and Y. Renardy, Direct simulation of unsteady axisymmetric core-annular ow with
high viscosity ratio, J. Fluid Mech., 391 (1999), pp. 123149.
[25] Y. Zang, R. L. Street, and J. R. Koseff, A non-staggered grid, fractional step method
for time-dependent incompressible Navier-Stokes equations in curvilinear coordinates, J.
Comput. Phys., 114 (1994), pp. 1833.
TWO-LIQUID FLOW AT LOW REYNOLDS NUMBER 439

[26] Y. Renardy, Instability at the interface between two shearing uids in a channel, Phys. Fluids,
28 (1985), pp. 34413443.
[27] Y. Renardy, Weakly nonlinear behavior of periodic disturbances in two-layer Couette-
Downloaded 07/18/12 to 128.118.88.243. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

Poiseuille ow, Phys. Fluids A, 1 (1989), pp. 16661676.


[28] P. Barthelet and F. Charru, Interfacial long waves in two-layer shear ows: Model equa-
tions in the light of experiments, in Advances in Multi-Fluid Flows, Y. Renardy et al., eds.,
SIAM, Philadelphia, 1996, pp. 1332.
[29] P. Boomkamp, Weakly Nonlinear Stability of Parallel Two-Phase Flow: Comparison with
Experiments, preprint, 1998.
[30] C. Pozrikidis, Instability of two-layer creeping ow in a channel with parallel sided walls, J.
Fluid Mech., 351 (1997), pp. 139165.
[31] J. C. Lasheras and H. Choi, Three-dimensional instability of a plane free shear layer: An
experimental study of the formation and evolution of streamwise vortices, J. Fluid Mech.,
189 (1988), pp. 5386.
[32] Y. Renardy and J. Li, Comment on A numerical study of periodic disturbances on two-layer
Couette ow, Phys. Fluids, 11 (1999), pp. 31893190.
[33] V. Cristini, J. Blawzdziewicz, and M. Loewenberg, Drop breakup in three-dimensional
viscous ows, Phys. Fluids, 10 (1998), pp. 17811783.
[34] R. Bai, K. Chen, and D. D. Joseph, Lubricated pipelining: Stability of core-annular ow.
Part 5. experiments and comparison with theory, J. Fluid Mech., 240 (1992), pp. 97142.
[35] Y. Renardy, Snakes and corkscrews in core-annular down-ow of two uids, J. Fluid Mech.,
340 (1997), pp. 297317.
[36] H. Hu and N. Patankar, Non-axisymmetric instability of core-annular ow, J. Fluid Mech.,
290 (1995), pp. 213234.
[37] R. Bai, K. Kelkar, and D. D. Joseph, Direct simulation of interfacial waves in a high-
viscosity-ratio and axisymmetric core-annular ow, J. Fluid Mech., 327 (1996), pp. 134.
[38] G. I. Taylor, The viscosity of a uid containing small drops of another uid, Proc. Roy. Soc.
London Ser. A, 138 (1932), pp. 4148.
[39] G. I. Taylor, The formation of emulsions in denable elds of ow, Proc. Roy. Soc. London
Ser. A, 146 (1934), pp. 501523.
[40] E. J. Hinch and A. Acrivos, Steady long slender droplets in two-dimensional streaming mo-
tion, J. Fluid Mech., 91 (1979), pp. 401414.
[41] H. P. Grace, Dispersion phenomena in high viscosity immiscible uid systems and application
of static mixers as dispersion devices in such systems, Chem. Engrg. Commun., 14 (1982),
pp. 225277.
[42] T. G. Mason and J. Bibette, Shear rupturing of droplets in complex uids, Langmuir, 13
(1997), pp. 46004613.
[43] C. R. Marks, Drop Breakup and Deformation in Sudden Onset Strong Flows, Ph.D. thesis,
University of Maryland at College Park, 1998.
[44] S. Kwak and C. Pozrikidis, Adaptive triangulation of evolving closed, or open surfaces by
the advancing-front method, J. Comput. Phys., 145 (1998), pp. 6188.
[45] J. M. Rallison, The deformation of small viscous drops and bubbles in the shear ows, Ann.
Rev. Fluid Mech., 16 (1984), pp. 4566.
[46] H. A. Stone, Dynamics of drop deformation and breakup in viscous uids, Ann. Rev. Fluid
Mech., 26 (1994), pp. 65102.
[47] B. J. Bentley and L. G. Leal, An experimental investigation of drop deformation and breakup
in steady, two-dimensional linear ows, J. Fluid Mech., 167 (1986), pp. 241283.
[48] M. Siegel, Inuence of surfactant on rounded and pointed bubbles in two-dimensional Stokes
ow, SIAM J. Appl. Math., 59 (1999), pp. 19982027.
[49] M. Loewenberg and E. J. Hinch, Numerical simulation of a concentrated emulsion in shear
ow, J. Fluid Mech., 321 (1996), pp. 395419.
[50] J. M. Rallison, A numerical study of the deformation and burst of a drop in general shear
ows, J. Fluid Mech., 109 (1981), pp. 465482.
[51] M. R. Kennedy, C. Pozrikidis, and R. Skalak, Motion and deformation of liquid drops and
the rheology of dilute emulsions in simple ow, Comput. Fluids, 23 (1994), p. 251.
[52] H. A. Stone, B. J. Bentley, and L. G. Leal, An experimental study of transient eects in
the breakup of viscous drops, J. Fluid Mech., 173 (1986), pp. 131158.
[53] A. M. Khokhlov, Fully threaded tree algorithms for adaptive renement uid dynamics sim-
ulations, J. Comput. Phys., 143 (1998), pp. 519543.