Three-Phase Hydraulic Conductances in

Angular Capillaries
Ahmed Al-Futaisi, SPE, Sultan Qaboos U., and Tad W. Patzek, SPE, U.C. Berkeley/Lawrence Berkeley
Natl. Laboratory

Summary
In this paper, we extend to three fluid phases a prior finite-element
study of hydraulic conductance of two-phase creeping flow in
angular capillaries. Previously, we obtained analytic expressions
for the hydraulic conductance of water in corner filaments. Here,
we present the results of a large numerical study with a highresolution finite-element method that solves the three-phase creeping flow approximation of the Navier-Stokes equation. Using the
projection-pursuit regression approach, we provide simple analytic
expressions for the hydraulic conductance of an intermediate layer
of oil sandwiched between water in the corners of the capillary and
gas in the center. Our correlations are derived for the oil layers
bounded by the concave or convex interfaces that are rigid or allow
perfect slip. Therefore, our correlations are applicable to drainage,
spontaneous imbibition, and forced imbibition with maximum feasible hysteresis of each contact angle, oil/water (O/W), and gas/oil
(G/O). These correlations should be useful in pore-network calculations of three-phase relative permeabilities of spreading oils.
Finally, we compare our results with the existing correlations by
Zhou et al.18 and Hui and Blunt,19 who assumed thin-film flow
with an effective film thickness proportional to the ratio of the
flow area to the length of the no-flow boundary. On average, our
correlations are two to four times closer to the numerical results
than the corresponding correlations by Zhou et al. and Hui and
Blunt.
Introduction
Because direct measurement of flow of three immiscible fluids is
very difficult, the pore-scale models of three-fluid systems15 have
blossomed. One of the more important advancements in such models was the approximation of single-pore geometries as angular
capillaries with square or arbitrary triangular cross sections. Although real pores are not exactly square or triangular, this approximation allows one to capture the flow of water in the pore corners
and the flow of oil and gas in the pore center. As illustrated in Fig.
1a, when three fluids are moving in a single angular capillary, the
most wetting fluid (water or Fluid 1) resides in the corner and the
most nonwetting fluid (gas or Fluid 3) fills the center. The third
fluid (oil or Fluid 2) forms an intermediate layer sandwiched between the other two fluids. In some cases of large contact angles
and positive spreading coefficients, we may find more than one
sandwiched layer (Fig. 1b). These intermediate layers are a few
micrometers thick and have been observed in micromodel experiments.69 It is drainage through these oil layers that is responsible
for the high oil recoveries seen experimentally.1012 Although it
was initially thought that only spreading oils could form such
layers in angular pores, it has been theoretically predicted and
experimentally verified that nonspreading oil can also form intermediate layers in the crevices of the pore space.6,9,13 Therefore,
the formation of sandwiched layers is not only related to the positive spreading coefficient, but also depends on the curvatures
of the O/W and G/O interfaces, the corner geometry, and the
contact angles.3

Copyright 2003 Society of Petroleum Engineers

This paper (SPE 86889) was revised for publication from paper SPE 75193, prepared for
presentation at the 2002 SPE/DOE Improved Oil Recovery Symposium, Tulsa, 1317 April.
Original manuscript received for review 30 April 2002. Revised manuscript received 30 May
2003. Manuscript peer approved 18 June 2003.

252

Creeping flow of oil in these intermediate layers is the subject

of this paper. In particular, we study the hydraulic conductances of
oil flow in stable fluid layers of different sizes and geometries. We
provide simple and accurate correlations for these conductances by
relating them to the interface geometry, fluid contact angles, and
pore geometry. The proposed correlations should be useful in pore
network calculations of three-phase relative permeabilities. A
similar approach has been successful in single-phase14 and twophase15 flow in angular capillaries. Assuming that equilibrium
O/W and G/O interfaces are stable in creeping flow,16,17 one can
solve the Navier-Stokes equation in the intermediate oil layer,
given its fixed boundaries. From the solution of the Navier-Stokes
equation, the average flow velocity of fluid i, vi, is calculated,
and its hydraulic conductance, gi, is estimated from a linear relation between the volumetric flow rate in the layer, Qi, and the
gradient of the total driving force per unit area, i:
Qi = viAi = gi i , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
where Aithe layer cross-sectional area.
Zhou et al.18 presented an approximate analytical solution for
oil flow along a layer sandwiched between water and air in angular
capillaries, and derived expressions for flow resistance and, therefore, hydraulic conductance with no-slip and perfect-slip conditions at the interfaces. Zhou et al. derived their expressions by
assuming thin-film flow with an effective film thickness proportional to the ratio of the flow area and the length of the no-flow
boundary. Their expressions were derived for the zero O/W contact angle, and are limited to the oil flow in the layers bounded by
concave menisci, as in drainage, but not in forced imbibition. For
nonzero O/W contact angles and convex interfaces, Hui and
Blunt19,20 proposed a modified version of the Zhou et al. expressions. Later in this paper, we will discuss in some detail relative
accuracy of all these expressions.
Our objectives are twofold: First, to develop a numerical algorithm that solves the velocity distribution and, therefore, hydraulic
conductance in three-phase flow with various geometries and interface boundary conditions; and, second, to provide simple and
accurate correlations for the hydraulic conductances of the intermediate layers of oil sandwiched between water and gas. The
paper is organized as follows. First, we calculate the various geometrical descriptors of the oil layer, such as its area, perimeter, and
shape factor. Second, we present the mathematical formulation of
the boundary value problems in creeping flow. Third, we describe
the finite-element approximations of these problems, and discuss
the numerical results. Finally, we correlate the layer hydraulic
conductance with the layer geometry, the pore corner geometry,
and the O/W and G/O contact angles.
Layer Geometry and Stability
Before simulating fluid flow in a sandwiched layer, we study its
geometry and stability. As shown in Fig. 2, the layer formation of
depends on five parameters: corner half-angle , O/W contact
angle 21, G/O contact angle 32, O/W meniscus-apex distance b1,
and G/O meniscus-apex distance b2. Zhou et al.18 and Hui and
Blunt19 define their hydraulic conductance expressions using the
radius of meniscus curvature instead of the meniscus-apex distance. The radius of curvature and the meniscus-apex distance are
related by Eq. 2, where r1 and r2 are the radii of curvature of the
O/W and G/O interfaces, respectively. We select the meniscusSeptember 2003 SPE Journal

Fig. 1Different configurations of (1) water, (2) oil, and (3) gas in a single triangular pore.

apex distances in order to avoid the infinite meniscus radius as the

interface becomes flat:
b1 = r1

cos21 +
cos32 +
, b2 = r2
. . . . . . . . . . . . . . . . . . (2)
sin
sin

The G/O interface is always concave (i.e., 32 + < /2). For 32

+ /2, the layer becomes unstable and cannot form. On the
other hand, the O/W interface can be concave (21 + < /2), flat
(21 + =/2), or convex (21 + > /2) (see Fig. 2). We shall
now define the dimensionless meniscus-apex distances b 1 and b 2
by scaling the spatial coordinates with the G/O meniscus-apex
distance, b2. As a result, the dimensionless distance b 2 is 1 and the
dimensionless distance b 1 is b1/b2. The three relevant descriptors
of the sandwiched layer geometry are its dimensionless area, circumference, and shape factor. First, however, we define the three
constants that will be used repeatedly in the calculations.
Eij0 =

ij
2

Eij1 =

cosij +
sin

Eij2 =

cosij +
cosij
sin

A L =

2
E32
1
32
E32
2 E0
2
E32
1

b1
b2

b1
b2

sin cos if 21 + = 2
21
E21
2 E0
2
E21
1

. . . . (4)
otherwise.

The layer dimensionless circumference, P L, is defined as

P L = 2 1

b1
+ L 21 + L 32 , . . . . . . . . . . . . . . . . . . . . . . . . . (5)
b2

where L 21the dimensionless length of the O/W interface calculated with

L 21 =

A L
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
P 2
L

The layer actual area is calculated as AL =(b2)2 A L, and the actual

perimeter is given by PL=b2 P L. The actual and normalized shape
factors are equal. Finally, the layer cannot exist, because the two
menisci touch, if

b1 E21
0
otherwise,
2
b2 E21
1

. . . . . . . . . . . . . . . . . . . (6)

and, similarly, the dimensionless length of the G/O interface, L 32,

is calculated by
E32
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
E32
1
, is defined as the ratio of the
The layer normalized shape factor, G
L
layer dimensionless area and square of the layer dimensionless
circumference as follows:
L 32 = 2

September 2003 SPE Journal

Mathematical Formulation
Assuming steady-state, creeping isothermal flow of Newtonian,
incompressible, and constant viscosity fluids, the combined continuity and Navier-Stokes equations that describe the flow reduce
to an elliptic Poisson equation are as follows:
i
x1, x2 i, . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
i

where i1,2,3 denotes water, oil, and gas, respectively; vi the

ith fluid velocity; i the ith fluid viscosity, x1 and x2 the
spatial coordinates across the capillary (Fig. 3), and the gradient of the total driving force per unit area defined as
i = pi + i f, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11)
wherepithe pressure gradient in fluid i, ithe ith fluid density, and fthe body force per unit mass. We also assume that the
interfacial tensions are constant, and the buoyancy forces are negligible in the capillary (i.e., both Bond numbers are much less
than one).
Using a scaling scheme similar to that implemented in Ref. 15,
we scale the spatial coordinates with the G/O meniscus-apex distance, b2, and the fluid velocities with the oil viscosity, 2, and the
respective gradient of the force, 2, driving oil flow:
x j =
v i =

b1
2 sin if 21 + = 2
b2

b1
1
cos cos sin32

1 32 +
if 21 + = 2
b2 cos2
E1
E32
1
. . . . (9)
E21
b1
1 cos32 sin
min 1, 32
otherwise
b2
E1 cos21 sin

2v1 =

i = 1, 2, 3, j i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3)
The layer dimensionless cross-sectional area, A L, is defined
as follows:
32
E32
2 E0

=
G
L

xj
b2
vi 2
b22 2

j = 1, 2, 3
j = 1, 2, 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12)

After scaling, the dimensionless form of Eq. 10 is

i2v i = 1
1 =

1
3
, = 1, 3 = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (13)
2 2
2

Although Eqs. 10 and 11 are applicable to any of the three fluids,

we focus here on the oil in the sandwiched layer. Our formulation
is incomplete without specifying the boundary conditions for this
layer. As shown in Fig. 3, we need to impose boundary conditions
along the duct walls, s, the O/W interface, 21, and the G/O
interface, 32.
253

Fig. 2Sandwiched layers with concave, flat, or convex O/W interface.

We impose a no-slip boundary condition along the duct walls

as follows:
v i = 0 on s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (14)
We consider two different boundary conditions along the O/W and
G/O interfaces. The first boundary condition requires infinite surface shear viscosity of the interface, which becomes a surfactantladen rigid wall with a no-slip condition,
v i = 0 on 21 or 32. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (15)
The second boundary condition describes a perfect-slip condition
with zero surface shear viscosity of the interface. This condition
translates mathematically to
v i ni = 0 on 21 or 32, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (16)
where nithe unit outward normal vector along the interface.
These two interface boundary conditions result in four possible
configurations of boundary conditions:
1. BC-1: No-slip O/W interface and no-slip G/O interface.
2. BC-2: No-slip O/W interface and perfect-slip G/O interface.
3. BC-3: Perfect-slip O/W interface and no-slip G/O interface.
4. BC-4: Perfect-slip O/W interface and perfect-slip G/O interface.
Another possible interface boundary condition assumes continuity
of velocity and shear stress along the interface:
v 2 n2 = v 1 n1 on 21
v 2 n2 = 3v 3 n3 on 32. . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
This last boundary condition is not considered here. Finally, there
is a no-flow boundary condition along the lines of symmetry (see
the following discussion).

Numerical Approximation
We solve the boundary value problem (Eqs. 13 through 16) numerically using the finite-element method (FEM). The FEM solution was obtained using the MATLAB Partial Differential Equation (PDE) Toolbox (MathWorks, Natick, Massachusetts).21 The
Toolbox provides a powerful and flexible environment for the
study and solution of partial differential equations in two space
dimensions and time. We now present examples of mesh generation, solution visualization, and convergence.
Mesh Generation. Accuracy of the FEM solution depends on how
well the computational domain is discretized. A good numerical
solution requires a mesh that follows the shape of the duct walls
and the curvatures of the interfaces. The PDE toolbox has many
powerful built-in functions that enable the user to generate the
finite-element meshes with the required properties. In our simulations, we have used the built-in MATLAB function initmesh to
initiate the first discretization of the domain. Then, the initial mesh
is refined five more times using the function refinemesh. In each
refinement, each triangle in the mesh is divided into four new
triangles. Because of the layer symmetry with respect to the corner-angle bisector, the FEM analysis is performed on 12 of the
domain. Depending on the layer geometry, the maximum number
of elements generated for a layer (half-domain) in our simulations
was 109,824, whereas the minimum number of elements was
5,376. The average number of elements of all the cases considered
in this analysis was 20,446. In Fig. 4, we present examples of layer
meshes for two geometrical configurations. The number of elements shown in the plot is just for illustration; the meshes used in
the numerical analysis were much finer.
Dimensionless Velocity Profile. Once we have generated a mesh
and specified appropriate boundary conditions, we solve Eq. 13 in
order to calculate the profile of the dimensionless velocity in the
layer. In Fig. 5, we show the dimensionless velocity profiles in a
specific layer for the four boundary conditions: BC-1, BC-2, BC-3,
and BC-4. The layer dimensionless area, A L , the circumference,
, are 0.1901, 3.7913, and 0.0132,
P L , and the shape factor, G
L
respectively. Benefitting from the symmetry of the layer with respect to the corner-angle bisector, the FEM analysis is performed
using 16,896 elements on 12 of the layer; the second half mirrors
the first one. From Fig. 5, one can see that the interface boundary
conditions influence significantly the velocity profile and, therefore, the average velocity and hydraulic conductance of the layer.
The average dimensionless velocity in the layer, <v L>, is calculated as follows:
N
v A
k=1 k k
, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)
v L =
N
A k

k=1

Fig. 3Spatial coordinates of the sandwiched layer: The flow

is along the duct in the x3-direction and the velocity distribution
is calculated in the x1 and x2 directions. s denotes the duct
wall, whereas 21 and 32 denote the O/W and G/O interfaces,
respectively.
254

where Nthe total number of elements in the layer, A kthe element k dimensionless area, and v kthe element k dimensionless
September 2003 SPE Journal

Fig. 4Layer mesh generation for two geometrical configurations: (a) =24, 21=0, 32=0, and b1/b2=0.25; and (b) =60, 21=80,
32=10, and b1/b2=0.3.

velocity calculated at the center of the element. The average dimensionless velocities in the layers shown in Fig. 5 are 0.0016,
0.0053, 0.0027, and 0.0163, respectively, with the boundary conditions BC-1, BC-2, BC-3, and BC-4.
The dimensionless hydraulic conductance, g L, of the layer is
N

g L = 2

v A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (19)
k k

k=1

The factor of 2 compensates for the calculations on 12 of the layer.

For the layers in Fig. 5, the dimensionless hydraulic conductances
are 0.000302, 0.0010, 0.000513, and 0.0031, respectively, for BC1, BC-2, BC-3, and BC-4.
Noticing that AL(b2)2 A L, and using Eqs. 1 and 12, we can
relate the layer actual hydraulic conductance, gL, to the dimensionless conductance, g L, as follows:
g L =

b24
g . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)
2 L

Convergence. To test the convergence of our FEM solution, we

calculate the ratio of the hydraulic conductances, g LN1 and g LN2,
calculated with the number of finite elements N1 and N24N1.
This ratio approaches 1 when the solution is converging. In Fig. 6,
we show an example of this calculation for BC-2: no-slip O/W
interface and perfect-slip G/O interface. Similar results were obtained for the other three boundary conditions. From Fig. 5, one
can see that with a few thousands of elements, about 6,000, the
numerical solution converges. Similar observation holds for simulations with the other boundary conditions.
Results
To derive a universal curve, which approximates the hydraulic
conductance of a sandwiched layer, intensive simulations of different layer geometries are required, set by the different combinations of , 21, 32, and b1/b2. In this analysis we have generated
17,167 stable layer geometries with arbitrary values of 5 to
85, 210 to 170, 320 to 80, and b1/b20.1 to 0.9. For each
generated layer, the dimensionless hydraulic conductance, g L, is
calculated for the four boundary conditions shown in Fig. 5.
In this section, we investigate the e.ects of the layer dimensionless parameters , 21, 32, and b1/b2 on its dimensionless
hydraulic conductance, g L. All four boundary conditions presented

in Fig. 5 exhibit similar behavior; therefore, we restrict our description to a single boundary configuration, BC-2: The no-slip
O/W interface and the perfect slip G/O interface. The same conclusions can be applied to the other three boundary conditions.
Fig. 7 relates the logarithm of the dimensionless hydraulic
conductance to the corner half-angle , the O/W contact angle 21,
and the G/O contact angle 32 for a fixed ratio b1/b2. For small
corner half-angles, < 10, the variation of the hydraulic conductance with 21 and 32 is relatively minor. However, as the corner
half angle increases, < 10, the hydraulic conductance varies
considerably with 21 and 32. For a fixed , increasing 21 decreases the layer hydraulic conductance, whereas increasing 32
increases it. Also, both contact angles, 21 and 32, affect the
hydraulic conductance considerably. The effect of the ratio b1/b2
on the layer hydraulic conductance is shown in Fig. 8. One may
notice that increasing b1/b2 results in a large variation of the hydraulic conductance. A smaller variation is observed when b1/b2 is
less than 0.4.
As mentioned in the mathematical formulation of the problem,
we are not considering here the momentum transfer boundary condition that assumes continuity of velocity and shear stress along
the interfaces (Eq. 17). However, because this boundary condition
is somewhere between the no-slip and perfect-slip ones, it is
worthwhile to compare the hydraulic conductances obtained with
the latter two boundary conditions by fixing some parameters: We
set the boundary condition and the contact angle at the G/O interface to no-slip and 0, respectively. With the same layer geometry,
the problem is solved twice, first with the no-slip O/W interface
(BC-1), and second with the perfect-slip O/W interface (BC-3).
The dimensionless hydraulic conductances for the two cases are
plotted in Fig. 9. The type of BC has a strong influence on the
hydraulic conductance. For small values of b1/b2 (i.e., for the two
interfaces of the sandwiched layer far from each other), the values
of the two conductances are similar, and their ratio is close to 1.
However, as the corner half-angle increases, this ratio becomes
smaller, reflecting the clear difference between the two BCs. For
the extreme cases of large b1/b2 (the interfaces very close to each
other) and large , the conductance ratio can be less than 0.3. This
simple comparison illustrates the importance of considering the
third boundary condition (Eq. 17), because the real flow is perhaps
neither perfect-slip nor no-slip.
Figs. 7 through 9 demonstrate that the different combinations of
, 21, 32, and b1/b2, and the boundary conditions may result in a

Fig. 5Velocity distribution in a sandwiched layer (=60, 21=90, 32=10, and b1/b2=0.25) with different interface boundary
conditions: BC-1, BC-2, BC-3, and BC-4, respectively.
September 2003 SPE Journal

255

Fig. 6Convergence of the FEM solution for eight different layers with BC-2: No-slip O/W interface and perfect-slip G/O interface.

huge variation of the layer hydraulic conductance. They also foreshadow the diffculty of creating an easy-to-use relationship among
these parameters and the layer hydraulic conductance. As our
analysis shows, there is no straightforward relation between the
layer conductance and its geometrical features, such as area, perimeter, or shape factor. Therefore, a rigorous statistical procedure
is required to derive best correlations for the layer hydraulic
conductance.
Universal Curves for the Layer
Hydraulic Conductance
Here we attempt to derive the universal curves that approximate
the hydraulic conductance of the sandwiched layer. As shown
above, regular statistical procedures may not describe adequately
the nonlinear interactions among the dimensionless dependent

Fig. 7The logarithm of the dimensionless hydraulic conductance vs. corner half-angles and contact angles 21 and 32 for
b1/b2=0.8, and with BC-2.
256

variable Y(g L) and any of the layer dimensionless predictor variables:

.
x = ,21,32,b1 b2,P L,A L,L 21,L 32,G
L
Therefore, we choose projection-pursuit regression to obtain the
universal curves for the layer hydraulic conductances. Projectionpursuit regression22,23 is a computer-intensive procedure that applies an additive model to the projected variables:
M

Y = 0 +

f X + , . . . . . . . . . . . . . . . . . . . . . . . . . (21)
j j

j=1

where (Y)any transformation of the dependent variable Y,

0the mean of (Y), j and k are the constants calculated by the

Fig. 8The logarithm of the dimensionless hydraulic conductance vs. the b1/b2 and contact angles 21 and 32 for =15, and
with BC-2.
September 2003 SPE Journal

Fig. 9The ratio of the dimensionless hydraulic conductance of BC-1 to BC-3 for different layer geometries.

regression model, fj are the functions fitted by the model, Ma

dimension chosen by the user, and the deviation of the fitted
values from the corresponding true ones. Thus projection-pursuit
regression uses an additive model on the predictor variables, which
are formed by projecting matrix X in M carefully chosen directions. The pursuit part indicates that an optimization technique is
used to find good direction vectors, k. The statistical language
S-PLUS24,25 is used to implement the projection-pursuit regression. Using projection-pursuit regression, the dimensionless hydraulic conductance in the layer is given by

BC-3: Perfect-Slip O/W Interface and No-Slip G/O

Interface.

0 = 7.5325
1 = 1.6435

z = 0.1367ln + 0.4339lnG
L
+ 0.7444lnA L 0.174032 0.4568

b1
b2

lng L = 0 + 1f1z, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (22)

f1z = 1.5380z + 4.7172. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (25)

where 0, 1, z, and f1(z) depend on the interfaces boundary conditions as follows.

BC-4: Perfect-Slip O/W Interface and Perfect-Slip

G/O Interface.

BC-1: No-Slip O/W Interface and No-Slip G/O

Interface.

0 = 7.9998

+ 0.3587lnA L 0.7446L 21 0.3169L 32

z = 0.0797ln + 0.5540lnG
L

f1z = 0.0409z5 + 0.4377z4 + 1.7096z3.

b1
b2

f1z = 1.3062z + 4.9465. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23)

BC-2: No-Slip O/W Interface and Perfect-Slip G/O
Interface.

0 = 7.2153

z = 0.3227ln + 0.5948lnG
L
b1
b2

f1z = 0.0008541z2 + 4.9495. . . . . . . . . . . . . . . . . . . . . . . . . . . (24)

September 2003 SPE Journal

+ 3.0028z2 + 4.1682z + 3.8056. . . . . . . . . . . . . . . . . . . . . . . . (26)

In Figs. 10, 12, 14, and 16, we plot the deviations of the hydraulic
conductances calculated with the projection-pursuit regression
from the corresponding FEM solutions, for the four boundary conditions, respectively. The relative error is calculated as follows:
Relative Error =

1 = 1.8026

+ 0.7141lnA L 0.151632 0.0959

1 = 1.7558

z = 0.4405ln + 0.14981lnG
L

1 = 1.7474

+ 0.7698lnA L 0.149432 0.2679

0 = 6.4543

g L g LFE
g LFE

100. . . . . . . . . . . . . . . . . . . . . . . (27)

where g LFEthe hydraulic conductance calculated by the finiteelement model. The mean absolute relative errors of these approximations are 9.5, 9.8, 13.9, and 18.0% for BC-1, BC-2, BC-3,
and BC-4, respectively. The actual conductance is calculated with
Eq. 20.
The most widely used expressions for the hydraulic conductance of the sandwiched layers were derived by Zhou et al.18
257

Fig. 10Relative deviation of projection-pursuit regression

flow conductance from the corresponding FEM solution for
BC-1 (the mean absolute error is 9.5%).

Zhous layer dimensionless hydraulic conductance for 32 + <

/2 and 210 is given by
gL =

= Ac2321 sin2
2cos32 1 cot1 3R2o3

2 + f11 cot1 f23Ro2, . . . . . . . . . . . . . . . . . . . (28)

where the constants 1, 2, 3, A c, R o are as follows:

32
2

tan,
2

sin
A c =
cos32 +

cos32 cotcos32 sin32

12A c sin21 3 1 + f13 1 f23

A w =

A3o321 sin2tan

A w
A

. . . . . (32)

where Aw and Ao are calculated by

2 = cotcos32 sin32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (29)

3 =

The quantity f is used to indicate the boundary condition at the

Fluid/Fluid interface. A value of f1 represents a no-slip boundary, while a value of f0 is a perfect-slip boundary. f1 is the
boundary condition at G/O interface and f2 is the boundary condition at the O/W interface. For nonzero O/W contact angles, or for
convex interfaces, Hui and Blunt19,20 proposed the following
modification to Eq. 28:
g L =

= 12sin21 322cos32 12

1 =

Fig. 11Relative deviation of Zhou et al.18 and Hui and Blunt19

flow conductance from the corresponding FEM solution for BC1 (the mean absolute error is 22.2%).

b1
sin
b2 cos21 +

cos21 cotcos21 sin21

+ 21 +

A o = A c A w. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (33)
, . . . . (30)

b1 cos32 +
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (31)
b2 cos21 +

Deviations of the hydraulic conductances calculated with Eqs. 28

and 32 from the corresponding FEM solutions with the four
boundary conditions in this analysis are shown in Figs. 11, 13, 15,
and 17. The points with light colors were generated with the Zhou
et al. approximation (Eq. 28) for the zero O/W contact angle,

Fig. 12Relative deviation of projection-pursuit regression

flow conductance from the corresponding FEM solution for
BC-2 (the mean absolute error is 9.8%).

Fig. 13Relative deviation of Zhou et al.18 and Hui and Blunt19

flow conductance from the corresponding FEM solution for
BC-2 (the mean absolute error is 24.1%).

R o =

258

+ 32 +

September 2003 SPE Journal

Fig. 14Relative deviation of projection-pursuit regression

flow conductance from the corresponding FEM solution for
BC-3 (the mean absolute error is 13.9%).

Fig. 15Relative deviation of Zhou et al.18 and Hui and Blunt19

flow conductance from the corresponding FEM solution for
BC-3 G/O interface (the mean absolute error is 28.0%).

whereas the dark-color points belong to the Hui and Blunt approximation (Eq. 32) for nonzero O/W contact angles. The mean absolute relative errors of these approximations are 22.2, 24.1, 28.0,
and 64.9% for BC-1, BC-2, BC-3, and BC-4, respectively. Thus,
the projection-pursuit regression method gives results better than
the expressions proposed by Zhou et al. and Hui and Blunt.

dependence on the system and displacement processes is discussed elsewhere.26,27

Discussion
The hydraulic conductance of an intermediate layer of oil (see Fig.
1), sandwiched between the corner water and the central water or
gas, has a rather complex dependance on the combination of the
boundary conditions at both interfaces, the contact angles, 21 and
32, and the layer thickness at the wall, b2b1, that in turn depends
on the respective capillary pressures, the corner half-angle, , and
the contact angles. When a combination of the corner half-angle
and the contact angles causes the layer surfaces to be either concave or flat and concave (see Fig. 2), the layer hydraulic conductance is relatively high. When the inner surface of the layer becomes convex, 21 + > /2, the layer conductance rapidly decreases. The diminishing layer thickness at the wall, b2b1 0,
which corresponds to b2/b1 1, causes the layer conductance to
decrease exponentially, especially when both contact angles go to
zero and the corner half-angle increases (see Fig. 7), or when the
inner contact angle 12 goes to 180 (see Fig. 8). The components
of the independent variable z in Eq. 22 and their scaling followed
those in Ref. 15. In particular, the inverse of the layer thickness at
the wall b1/b2 enters z with the negative sign: The thinner the layer,
the more negative the logarithm of the hydraulic conductance.
The oil hydraulic conductance at low oil saturations is a complex function of the system geometry and rock wettability. Its

Conclusions
A numerical approach has been developed to obtain the hydraulic
conductance of an intermediate fluid layer sandwiched between
two other fluids. This approach has led to the simple and accurate correlations of the layer hydraulic conductance with the various layer geometry descriptors, and with four different boundary conditions:
1. Expressions for calculating the layer perimeter, area, shape
factor, and stability have been derived from the corner half-angle
, the interface contact angles 21 and 32, and the meniscus-apex
distances b1 and b2.
2. A standard finite-element method has been used to solve
numerically Poissons equation for creeping, isothermal flow in an
intermediate fluid layer formed in a corner of a polygonal capillary. Each flow domain was discretized with a mesh that followed
the shape of the capillary walls and the curvatures of the interfaces.
3. The layer hydraulic conductance has been calculated for four
sets of boundary conditions: BC-1 (no-slip water/oil interface and
no-slip oil/gas interface), BC-2 (no-slip water/oil interface and
perfect-slip oil/gas interface), BC-3 (perfect-slip water/oil interface and no-slip oil/gas interface), and BC-4 (perfect-slip water/oil
interface and perfect-slip oil/gas interface).
4. Simple comparison between the hydraulic conductance values for perfect-slip and no-slip boundary conditions indicates that
it might be essential to consider the momentum transfer boundary
condition that assumes continuity of velocity and shear stress
along the interfaces.

Fig. 16Relative deviation of projection-pursuit regression

flow conductance from the corresponding FEM solution for
BC-4 (the mean absolute error is 18.0%).

Fig. 17Relative deviation of Zhou et al.18 and Hui and Blunt19

flow conductance from the corresponding FEM solution for
BC-4 (the mean absolute error is 64.9%).

September 2003 SPE Journal

259

5. Projection-pursuit regression has been used to obtain the

universal correlations of the layer hydraulic conductance with the
relevant descriptors of the layer geometry. These expressions have
been listed for each boundary condition described in Conclusion 3.
The hydraulic conductance expressions proposed here have been
compared with the expressions proposed by Zhou et al.18 and Hui
and Blunt.19 We have compared the relative deviation of the conductance calculated with each method from the corresponding finite-element solution. We have shown that the mean absolute relative errors of the projection-pursuit regression expressions are 9.5,
9.8, 13.9, and 18.0% for the four boundary conditions, respectively. The corresponding mean absolute errors of the Zhou et al.
and Hui and Blunt expressions are 22.2, 24.1, 28.0, and 64.9%.
Thus, the expressions proposed here are better than those proposed
by Zhou et al. and Hui and Blunt by a factor of 2 to 4.
6. Overall, our correlations of the layer hydraulic conductances
are simple and can be used with confidence in the computationally
intensive two-phase and three-phase pore-network simulations of
drainage and imbibition with contact-line pinning and contactangle hysteresis.
Nomenclature
A cross-sectional area, L2
A dimensionless area
b1 oil/water meniscus-apex distance, L
b2 gas/oil meniscus-apex distance, L
Ei geometrical constant
f body force per unit mass, FM-1
fj jth statistical approximation function
g hydraulic conductance, L6F-1T-1
g dimensionless hydraulic conductance
G shape factor
normalized shape factor
G
L dimensionless length of O/W interface
21
L 32 dimensionless length of G/O interface
n unit outward normal vector
P circumference, L
P dimensionless circumference
r1 radius of curvature of O/W interface, L
r2 radius of curvature of O/W interface, L
Q volumetric flow rate L3T-1
<v> average flow velocity, LT-1
velocity, LT-1
X vector of independent variables
Y dependent variable
z statistical fit variable
statistical fit constant
corner half-angle
i ith boundary of flow domain
approximation error
21 O/W contact angle
32 G/O contact angle
transformation of dependent variable
statistical fit constant
viscosity, FL-2T
gradient of driving force per unit area, FL-3
density, ML-3
pressure gradient, FL-3
Acknowledgments
The first author was sponsored by a fellowship from the Sultan
Qaboos U., the Sultanate of Oman. Partial support was provided by
the U.S. DOE ORT Partnership under Contract DE-ACO376FS0098 to the Lawrence Berkeley Natl. Laboratory. Partial support was also provided by gifts from ChevronTexaco and Phillips
Petroleum to UC Oil, Berkeley. Extensive help with the projection-pursuit method has been provided by Dr. David Brillinger,
Statistics Dept., U.C. Berkeley.
260

References
1. ren, P.E. and Pinczewski, W.V.: Fluid distribution and pore-scale
displacement mechanisms in drainage dominated three-phase flow,
Transport in Porous Media (1995) 20, 105.
2. Mani, V. and Mohanty, K.K.: Pore-Level Network Modeling of
Three-Phase Capillary Pressure and Relative Permeability Curves,
SPEJ (September 1998) 238.
3. Fenwick, D.H. and Blunt, M.J.: Three-dimensional modeling of threephase imbibition and drainage, Adv. Water Resour. (1998) 21, No. 2,
121.
4. Lerdahl, T.R., ren, P.E., and Bakke, S.: A Predictive Network Model
for Three-Phase Flow in Porous Media, paper SPE 59311 presented at
the 2000 SPE/DOE Symposium on Improved Oil Recovery, Tulsa, 35
April.
5. Van Dijke, M.I.J., Sorbie, K.S., and McDougall, R.S.: A ProcessBased Approach for Three-Phase Capillary Pressure and Relative Permeability Relationships in Mixed-Wet Systems, paper SPE 59310
presented at the 2000 SPE/DOE Symposium on Improved Oil Recovery, Tulsa, 35 April.
6. Dong, M., Dullien, F.A., and Chatzis, I.: Imbibition of oil in film form
over water present in edges of capillaries with an angular crosssection, J. Colloid and Interface Science (1995) 172, 278.
7. ren, P.E., Billiote, J., and Pinczewski, W.V.: Mobilization of Waterflood Residual Oil by Gas Injection for Water-Wet Conditions,
SPEFE (1992) 70; Trans., AIME, 293.
8. Soll, W.E., Celia, M.A., and Wilson, J.L.: Micromodel studies of
three-fluid porous media systems: Pore-scale processes relating to capillary pressure-saturation relationships, Water Resour. Res. (1993) 29,
No. 9, 2963.
9. Keller, A.A., Blunt, M.J., and Roberts, P.V.: Micromodel observation
of the role of oil layers in three phase flow, Transport in Porous
Media (1997) 20, 105.
10. Dumor, J. M. and Schols, R.S.: Drainage Capillary Functions and the
Influence of Connate Water, SPEJ (October 1974) 437.
11. Kantzas, A., Chatzis, I., and Dullien, F.A.L.: Enhanced Oil Recovery
by Inert Gas Injection, paper SPE 13264 presented at the 1988 SPE/
DOE Improved Oil Recovery Symposium, Tulsa, 1720 April.
12. Zhou, D. and Blunt, M.J.: Effect of spreading coefficient on the distribution of light non-aqueous phase liquid in the subsurface, Journal
of Contaminant Hydrology (1997) 25, 1.
13. Blunt, M.J., Zhou, D., and Fenwick, D.H.: Three-phase flow and
gravity drainage in porous media, Transport in Porous Media (1995)
20, 77.
14. Patzek, T.W. and Silin, D.B.: Shape factor and hydraulic conductance
in noncircular capillaries: I. One-phase creeping flow, J. Colloid and
Interface Science (2001) 236, No. 2, 295.
15. Patzek, T.W. and Kristensen, J.D.: Shape factor and hydraulic conductance in noncircular capillaries: II. Two-phase creeping flow, J.
Colloid and Interface Science (2001) 236, No. 2, 305.
16. Ransohoff, T.C. and Radke, C.J.: Laminar flow of a wetting liquid
along the corners of a predominantly gas-occupied noncircular pore, J.
Colloid and Interface Science (1988) 121, 392.
17. Blunt, M. and Scher, H.: Pore-level modeling of wetting, Phys. Rev.
E (1995) 52, No. 6, 9387.
18. Zhou, D., Blunt, M.J., and Orr, F.M.: Hydrocarbon drainage along
corners of noncircular capillaries, J. Colloid and Interface Science
(1987) 187, 11.
19. Hui, M.H. and Blunt, M.J.: Effects of wettability on threephase flow
in porous media, J. Phys. Chem. B (2000) 104, 3833.
20. Hui, M.H. and Blunt, M.J.: Pore-Scale Modeling of Three-Phase Flow
and the Effects of Wettability, paper SPE 59309 presented at the 2000
SPE/DOE Improved Oil Recovery Symposium, Tulsa, 35 April.
21. Mathworks: Partial Differential Equation Toolbox Users Guide, The
Mathworks Inc., Seattle (1998).
22. Friedman, J.H. and Stuetzle, W.: Projection pursuit regression, Journal of the American Statistical Association (1981) 76, 183.
September 2003 SPE Journal

23. Friedman, J.H.: Exploratory projection pursuit, Journal of the American Statistical Association (1987) 82, 249.
24. MathSoft: S-PLUS 4 Guide to Statistics Users Guide, The Mathsoft
Inc., Data Analysis Products Division, MathSoft, Seattle (1997).
25. Venables, W.N. and Ripley, B.D.: Modern Applied Statistics with SPlus, Springer-Verlag Inc., New York City (1998).
26. Al-Futaisi, A. and Patzek, T.W.: Impact of wettability on two-phase
flow characteristics of sedimentary rock: Quasistatic model, Water
Resour. Res. (2003) 39, No. 2, 1042.
27. Al-Futaisi, A. and Patzek, T.W.: Impact of oil-water drainage and
imbibition on subsequent gas injection: A three phase pore network
model, Physical Review E, in press (March 2003).

September 2003 SPE Journal

Ahmed Al-Futaisi is currently an assistant professor of civil and

architectural engineering at the Sultan Qaboos U. in Oman.
e-mail: futaisi@uclink4.berkeley.edu. His research interests involve pore-level modeling of multiphase flow in porous media,
and groundwater management. He holds a PhD degree in
civil and environmental engineering from U.C. Berkeley. Tad W.
Patzek is a professor of geoengineering at the Civil and Environmental Engineering Dept., U.C. Berkeley, and is Mr. AlFutaisis thesis advisor. e-mail: patzek@patzek.berkeley.edu. His
research interests involve multiphase flow in porous media at a
variety of scales, evolution of rock microstructure, mechanical
properties of fragile rocks, and reservoir management and
control. He is a co-author of more than 100 papers.

261