+ + < +
<
=
d u u
u u
u
L L x L H
L x L L x A H
L x H
x H
,
), / 2 sin(
0 ,
) (
0
0
0
(1)
where A and represent the amplitude and
wavelength of corrugation, respectively. The planar
geometry consists of three parts: an introductory
segment of constant gap H
0
and length L
u
, a
harmonically varying middle part of length , and
an exit segment of constant gap H
0
and length L
d
.
The minimum gap is H
c
. Details on the constriction
of each geometry are shown in Figure 3.
The equations governing the transient, two
dimensional, incompressible flow in the absence of
inertial effects (i.e., when Reynolds number << 1)
can be written as:
0 v = (2)
0 T v v
v
=


.

\

+
t
(3)
Figure 2. Flow domain and boundary conditions for gasliquid displacement in a constricted channel.
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
106
where v and p are the velocity and pressure fields,
respectively, and is the liquid density. The total
stress tensor for Newtonian liquids is
] v) ( v [ I T
T
+ + = p , where is the liquid
dynamic viscosity and the superscript T means
transpose. Because of the small dimensions of
the flow domain, body forces are usually neglected.
The flow domain and its boundaries are labeled
in Figure 2. The corresponding boundary conditions
are:
1. At the inflow plane, where the gas displaces the
liquid, there is a free surface; the velocity field
should satisfy a local force balance between the
traction in the liquid, the capillary pressure and the
gaseous pressure, given by Equation 4. It is
assumed that the gas viscosity is much smaller than
the liquid viscosity, leading to vanishing shear
stress along the gasliquid interface.
in fs
fs
fs
P
ds
d
n
t
T n = (4)
in Equation 4, is the liquid surface tension, t
fs
and
n
fs
are the local unit tangent and unit normal
vectors to the free surface (identified by the
subscript fs), dt
fs
/ds represents the curvature of
the meniscus and s identifies the arclength
coordinate along the free surface. P
in
is the
pressure applied at the free surface located at the
entrance of the channel. A step change in this value
is applied: at t = 0 s, the pressure is suddenly
increased to P
in
= 3.63 Pa and remains constant
until the end of the computations, when t
LAST
is
reached (final time). Also, there is no mass flow
rate across the gasliquid interface, and the
kinematic condition can be used considering the
mesh velocity , as in Equation 5 (Christodoulou
and Scriven, 1992).
0 ) x v ( n =
fs
(5)
2. Along the symmetry line, both the shear stress
and the vertical velocity vanish, which means:
0 ) t T ( n , 0 v n = = (6)
where n and t are the unit vector normal and
tangent to the boundary.
3. At the outflow plane:
out
P p = = 0, v) ( n (7)
that is, a fully developed parallel rectilinear flow is
observed and the pressure imposed is P
out
= 0 Pa.
The outflow plane was located at a distance equal
to 5H
0
downstream from the constriction,
H
0
= 1 mm. Changes on its location did not affect
the steady state solution.
Figure 3. Representation of the sinusoidal geometries of the constricted channel.
Horizontal and vertical axes in mm.
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
107
4. The noslip and nopenetration conditions
applies along the channel walls, as:
0 , 0 = = v u (8)
Finally, an initial condition is needed in order to
solve the transient flow problem. In this work, we
assume that initially there is no flow and the
pressure field is zero along all domain:
0 ) 0 ( , 0 ) 0 ( v = = = = t p t (9)
2.2 Geometries
Figure 3 illustrates the typical flow geometry
examined herein. The channel consists of three
parts: an introductory segment, a harmonically
varying middle part and an exit segment of lengths
L
u
, and L
d
respectively. The introductory and exit
segments are necessary in order to accommodate
the input and output boundary conditions. In order
to explore the effect of the geometry of the
constriction in the transient, twodimensional flow,
four different harmonically varying wall
configurations were considered. They are shown in
Figure 3 and are identified with numbers (1), (2),
(3) and (4), obtained with Equation 1.
3. SOLUTION METHOD
3.1 Formulation of the free boundary
problem
The primary difficulty in applying the finite
element method (FEM) to freesurface problems is
the need to specify a procedure for constructing a
mesh over the domain, the shape of which is a
priori unknown. For those FEM problems which
involve free surfaces that move during the
computation, an initial mesh can be updated
simply by using a continuous deformation
algorithm which maintains the same mesh
connectivity throughout the problem. By
preserving connectivity, we also preserve the
sparseness patterns of the various matrices that
arise in the discretization solution of the problem,
e.g. stiffness and mass matrices, Jacobians, among
others, with obvious efficiency advantages. The
shape of the interface needs to be determined as
part of the solution. Several schemes to obtain the
shape of the free surface together with the velocity
and pressure fields are available.
In our work, in order to solve this free boundary
problem by means of standard techniques for
boundary value problems, the set of differential
equations and boundary conditions posed in the
unknown domain had to be transformed to an
equivalent set defined in a known reference
domain. This transformation was made by a
mapping x = x() that connects the two domains.
The unknown physical domain was parameterized
by the position vector x, and the reference domain
by . The mapping used here was the one described
by De Santos (1991). The inverse of the mapping
that minimizes the function is governed by a pair of
elliptic differential equations identical to those
encountered in diffusional transport with variable
diffusion coefficients. The coordinate potentials
and of the reference domain satisfy the following
criteria:
0 ) ( , 0 ) ( = =
D D (10)
D
and D
are diffusionlike coefficients used to
control gradients in coordinate potentials and
hence the spacing between curves of constant , on
one hand, and of constant on the other hand,
that make up the sides of the elements that were
employed. These are quadrilateral elements.
Boundary conditions were needed in order to solve
the secondorder partial differential equations
(Equation 10). The solid walls and inlet and outlet
planes were described by functions of the
coordinates, along which stretching functions were
used to distribute the termini of the coordinate
curves selected to serve as element sides. The
required free boundary (gasliquid interface)
enlarged the system of governing equations with
the kinematic condition (Equation 5). The discrete
version of the mapping equations is generally
referred to as mesh generation equations.
In transient problems the frame of reference
lies across the spacetime domain for which the
physical grid points are constantly updated in time.
Therefore, time derivative at fixed Eulerian
locations in space t / needs to be
transformed to time derivative at fixed iso
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
108
parametric coordinates, denoted by
. As
described by Christodoulou and Scriven (1992),
the following transformation is employed:
x
t
(11)
where x is the mesh velocity, the time derivative
of the nodal position.
Replacing the local velocity time derivative
t v/ in Equation 3 with Equation 11, the
momentum equation becomes:
( ) 0 T v ) x  v ( v = + (12)
3.2 Spatial discretization with the
Galerkin / finite element method
The system of governing equations, Equations 2,
10 and 12, together with the appropriate boundary
conditions, was solved with the Galerkins method
with quadrilateral finite elements. The weighted
residual equations are obtained after multiplying
each governing equation by the weighting
functions
m
,
c
and
x
, then integrating them
over the unknown flow domain (bounded by
), applying the divergence theorem to the terms
with divergences and mapping the integrals onto
the known reference domain (bounded by ).
In the Galerkins method, the weighting and the
basis functions are the same. Superscripts m, c, and
x denote momentum, continuity and mesh
residuals. Each independent variable v, p and x is
approximated with a linear combination of a finite
number of known basis functions ) , (
x , , v
p
j
,
resulting the expressions below:
) , ( ) ( x
~
x
) , ( ) (
~
) , ( ) ( v
~
v
x
v
j
M
j
j
p
j
N
j
j
j
M
j
j
t
t p p
t
=
=
=
(13)
where ) ( v
~
t
j
, ) (
~
t p
j
and ) ( x
~
t
j
are the basis
functions coefficients that represent the unknowns
of the discretized problem. They are arranged in a
vector of basis function coefficients ) ( u t :
T
j j j
t t p t t )] ( x
~
), (
~
), ( v
~
[ ) ( u = (14)
The velocity and node positions were
represented in terms of Langrangian biquadratic
basis functions [ ) , (
v
j
, ) , (
x
j
] with the
condition ) , ( ) , (
x v
j j
= , and the pressure in
terms of linear discontinuous basis functions
) , (
p
j
. This is the socalled mixed formulation,
corresponding to polynomials needed to
interpolate the velocity field that have higher order
than the pressure. This combination is appropriate
to avoid instabilities during the process solution for
Newtonian liquids, also know as the BabuskaBrezzi
condition (Fortin, 1981).
Once all variables are represented in terms of
the basis functions, the system of partial
differential equations reduces to a set of ordinary
differential and algebraic equations that describe
the evolution of the coefficients with time:
0 ) f , u u, ( R = (15)
R is the set of weighted residual equations, u is the
vector of basis functions coefficients, u is their
time derivatives, and f is a vector that contain all
the input parameters (physical properties,
geometry of the flow and boundary condition
information). Additional details about this
derivation are reported by Romero and Carvalho
(2008), who undertook a numerical study of
coating flows under periodic disturbances.
3.3 Temporal discretization of the
transient equations
The temporal discretization of the set of
ordinary differentialalgebraic equations, Equation
15, followed a predictorcorrector algorithm. The
predictor step consists of a forward Euler method
and the corrector step consists of a firstorder fully
implicit Euler method, which is unconditionally
stable. The approximate form of the velocity time
derivatives in the momentum conservation
equation, Equation 12, using backward finite
difference is:
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
109
) , v (
v v
v
1 1
1
1
+ +
+
+
=
=
n n
n n
n n
t f
t t
(16)
the indices n+1 and n indicate current and previous
point of time, respectively. The difference t
n+1
t
n
defines the current time step t
n+1
x
. A similar
expression is used for the mesh velocity .
The system of ordinary differentialalgebraic
equations is reduced to simultaneous algebraic
nonlinear equations for the coefficients of the basis
functions of all fields. At each time step, the
resulting set of nonlinear algebraic equations was
solved simultaneously by the Newtons method
with an initial estimate resulting from an explicit
Euler step from the solution at the previous two
time steps:
) u , R(u u ) J(u ) M(u
1
1
1
1 1 n
(k)
n
) (k (k)
n
(k)
n
t
+
+
+ +
=


.

\

+
(17)
) (k (k)
n
) (k
n
1
1
1
1
u u u
+
+
+
+
+ = (18)
the indices k+1 and k indicate the current and
previous Newtons steps. The iteration is repeated
until convergence. Here, u / R J is the
Jacobian matrix which measures the sensitivity of
the residuals to the unknowns of the problem. The
entries of this matrix are calculated analytically.
The term u / R M is the mass matrix which
measures the sensitivity of the residuals to changes
in the time derivatives of the unknowns.
Because the finite element basis functions used
are different from zero only over a very small
portion of the domain, the Jacobian matrix is
sparse and was stored in compressed sparse
formats. In each Newtons iteration, the linearized
form of Equation 17 was factorized into a lower L
and upper U triangular matrix by the frontal solver
(Almeida, 1996). The tolerance on the L2norm of
the residual vector and on the last Newton update
of each declared solution was set to 10
6
. The
sensitivity of the computed solution to the size of
the time step used was tested by comparing the
transient response obtained with different time
steps. The solution varied by less than 2 % if the
time step t was lower than 1.5. A mesh
convergence analysis was done by comparing the
free surface configuration at capillary number Ca =
0.2, with three different meshes. A mesh with 680
elements (13356 degrees of freedom) was
considered satisfactory and was used to obtain the
solutions reported here. Further refining of the
mesh led to changes on the bubble tip position by
less than 1 %.
4. RESULTS AND DISCUSION
The Newtonian liquid initially occupies the
entire domain. At the beginning of the
computations (t = 0), the gasliquid interface is
considered flat. A step change in the pressure P
in
is
applied on the gas side of the gasliquid interface at
the entrance of the channel by means of Equation
4, forcing the transient penetration of the gas
bubble into the varying slot crated by the neck in
the parallel plates. Because of that, the steady
state is not reached and the flow domain changes
in shape and size, demanding several
reconstructions of the mesh. Differently from other
approaches, where the bubble is static and the wall
moves, in our work the wall is fixed and the bubble
invades the channel. Because the prescribed
pressure at the exit of the domain is P
out
out in
P P
= 0, the
applied pressure difference remains
constant. However, once the bubble moves along
the slot in direction to the output boundary, the
volume of the liquid in the domain decreases,
thereby reducing the resistance to the flow. This
results in acceleration of the flow field.
Computations are performed until t = t
LAST
has
been reached. The last time, t
LAST
, is different for
each geometry because the mesh deformation
limits are influenced by the type of constriction
present in the channel.
A gas bubble moving at constant pressure
through a constricted noncircular capillary is
illustrated in Figure 4. As the bubble invades the
constriction, the mean curvature of the gasliquid
interface increases, so that the bubble front may
squeeze into the constriction. After the front has
passed the tightest part, or neck, of the
constriction, the interfacial curvature begins to
decrease. As commented by Ransohoff et al.
(1987), this drop in curvature, if sufficiently large,
initiates a wettingliquid pressure gradient directed
from the front of the bubble to the neck of the
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
110
constriction. Liquid flows into the constriction, and
eventually enough liquid may accumulate at the
constriction neck to block the pore and snapoff a
gas bubble. Snapoff occurs when wetting fluid
(liquid) accumulates by capillary action in a pore
throat occupied by nonwetting phase (gas) and
then bridges across the throat to block flow.
Capillary pressure must be high enough for gas to
invade the throat, but then fall by about a factor of
two for liquid to reinvade the throat, which means
that if the pore body radius is at least about twice
the throat radius, snapoff occurs, in agreement
with the basic Roofs criteria (Rossen, 2000).
A comparison of bubble position at the last time
of computation for the different channel
undulation is shown in Figure 5. In the four cases,
the mesh is strongly deformed around the
constricted section and when it passes through it.
Convergence of the solution process is lost at time
instants which are functions of the intensity of the
channel variation. The remaining film has a non
uniform thickness distribution being thicker before
contractions.
Figure 6 shows the time evolution of the bubble
tip position for the four channel configurations.
The bubble tip moves toward the channel exit
monotonically. The evolution of the tip exhibits
smallamplitude oscillations. The time required for
the bubble tip to reach a certain position in the
tube increases with the intensity of the undulation
variation.
Figure 4. Mesh bubble evolution at different points of time as a function of time step t: (a) 12t, (b) 48t, (c)
71t and (d) 86t. The geometry is the number (1) showed in Figure 3. Horizontal and vertical axes in mm.
Figure 5. Comparison of the mesh deformation at the last time of computing: (a) 86t, (b) 115t, (c) 146t and
(d) 150t; for geometries (1), (2), (3) and (4) respectively. Horizontal and vertical axes in mm.
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
111
A strong decline in the pressure of the bubble
tip is observed at the beginning of the process for
the four channel configurations, as shown in Figure
7, after which it declines linearly until reaching the
constricted section, where the pressure decrease is
more pronounced, since more
energy is consumed there. In addition, there is a
time delay in the response of the pressure at the
bubble tip. A minimum is detected after the bubble
front has passed through the channel constriction.
For example, in the geometry number (1), the
bubble reaches the constriction at t = 71t, and the
pressure is minimized at t = 75t. The response of
the bubble tip velocity is also presented in Figure 7.
The bubble tip velocity is almost constant before
the constriction, and after the bubble passes the
constriction it immediately expands and
accelerates.
5. FINAL REMARKS
Our intention in this work has been to present
the numerical technique and the first results
obtained when applying the transient gasliquid
displacement approach in constricted channels.
Certainly, in further investigations we will attempt
to improve the treatment of these results,
especially using dimensionless parameters, by
selecting a more refined mesh and a smaller time
intervals, and running each case in a more
powerful computer.
ACKNOWLEDGMENTS
I would like to acknowledge a reviewer of this
article for his valuable suggestions.
Figure 7. Pressure and velocity of the bubble tip as a function of time.
Pressure in Pa, velocity in mm/s and time in s.
Figure 6. Evolution of the bubble tip position. Bubble position in mm and time in s.
BRAZILIAN JOURNAL OF PETROLEUM AND GAS  v. 3 n. 3  p. 103112  2009  ISSN 19820593
112
6. REFERENCES
Bretherton, F. P.; The motion of long bubbles in
tubes. J. Fluid Mech., vol. 10, p. 166188, 1961.
Christodoulou, K. N. and Scriven, L. E.;
Discretization of free surface flows and other
moving boundary problems. Journal of
Computational Physics, vol. 99, p. 3955, 1992.
Cox, B. G.; On driving a viscous fluid out of a tube.
J. Fluid Mech., vol. 14, p. 8196, 1962.
Dussan V., E. B.; On the spreading of liquids on
solid surfaces: static and dynamic contact lines.
Ann. Rev. Fluid Mech., vol. 11, p. 371400,
1979.
De Santos, J. M.; Twophase cocurrent downflow
through constricted passages. PhD thesis,
University of Minnesota, MN, USA, 1991.
Fairbrother, F. and Stubbs, A.; Studies in
electroendosmosis, Part vi: The bubbletube
methods of measurement. J. Chem. Soc., vol. 1,
p. 527529, 1935.
Gauglitz, P. A., Laurent, C. M. St. and Radke, C. J.;
An experimental investigation of gasbubble
breakup in constricted square capillaries.
Journal of Petroleum Technology, p. 11371146,
1987.
Kamisli, F. and Ryan, M. E.; Gasassisted
displacement of a Newtonian fluid confined in a
HelleShaw cell. Chemical Engineering Journal,
vol. 84, p. 193200, 2001.
Kang, Q., Zhang, D. and Chen, S.; Immiscible
displacement in a channel: simulations of
fingering in two dimensions. Advances in Water
Resources, vol. 27, p. 1322, 2004.
Kolb, W. B. and Cerro, R. L.; The motion of long
bubbles in tubes of square cross section. Phys.
Fluids A, 5 (7), p. 15491557, 1993.
Lenormand, R.; Liquids in porous media. J. Phys.:
Condens. Matter 2, p. SA79SA88, 1990.
Morrow, N. and Mason, G.; Recovery of oil by
spontaneous imbibition. Current Opinion in
Colloids & Interface Science, vol. 6, p. 321337,
2001.
Olbricht, W. L. and Kung, M. D.; The deformation
and breakup of liquid drops in low Reynolds
number flow through a capillary. Phys. Fluids A
4, vol. 7, p. 13471354, 1992.
Olbricht, W. L.; Porescale prototypes of
multiphase flow in porous media. Annu. Rev.
Fluid Mech., vol. 28, p. 187213, 1996.
Ransohoff, T. C., Gauglitz, P. C. and Radke, C. J.;
Snapoff of gas bubbles in smoothly constricted
noncircular capillaries. AIChE J., vol. 33, n. 5, p.
753765, 1987.
Ratulowski, J. and Chang, H. C.; Transport of gas
bubbles in capillaries. Phys. Fluids A 1, vol. 10, p.
16421655, 1989.
Romero, O. J., Scriven, L. E. and Carvalho, M. S.;
Effect of curvature of coating die edges on the
pinning of contact line. AIChE J., vol. 52, p. 447
455, 2006.
Romero, O. J. and Carvalho, M. S.; Response of slot
coating flows to periodic disturbances. Chemical
Engineering Science, vol. 63, n. 8, p. 21612173,
2008.
Rossen, W. R.; Snapoff in constricted tubes and
porous media. Colloids and Surfaces A, vol. 166,
p. 101107, 2000.
Slattery, J. C.; Interfacial effects in the entrapment
and the displacement of residual oil. AIChE J.,
vol. 20, p. 11451154, 1974.
Stark, J. and Manga, M.; The motion of long
bubbles in a network of tubes. Transport in
Porous Media, vol. 40, p. 201218, 2000.
Taylor, G.; Deposition of a viscous fluid on the wall
of a tube. J. Fluid Mech., vol. 10, p. 161165,
1961.