Standard Galerkin method SG Petrov–Galerkin method PG
where M
ik
¼ Z
Q
W
i
N
k
dQ The diagonalized lumped matrix coefficients are then
M
lump ii
¼ Z
Q
W
i
dQ ¼ : V
i
13 Reformulating the discretized equations using eqn 10 and
eqn 12 gives fV
i
Dt ˆS
n þ 1 n
, i
¹ ˆ S
n n
, i
¼ X
j[h
i
Z
Q
ˆp
n þ 1 w
, j
¹ ˆp
n þ 1 w
, i
ÿ ¹ r
w
g ˆ D
j
¹ ˆ D
i
ÿ 3 l
n þ 1 w
, ij
=W
i
·K·=n
j
dQ ¹ q
n þ 1 w
, i
V
i
¹ m
n þ 1 w
, i
¹ fV
i
Dt ˆS
n þ 1 n
, i
¹ ˆ S
n n
, i
¼ X
j[h
i
h ˆp
n þ 1 w
, j
¹ ˆp
n þ 1 w
, i
ÿ þ ˆp
n þ 1 c
, j
¹ ˆp
n þ 1 c
, i
ÿ ¹
r
n
g ˆ D
j
¹ ˆ D
i
ÿ i
l
n þ 1 n
, ij
=W
i
·K·=N
j
dQ ¹
q
n þ 1 n
, i
V
i
¹ m
n þ 1 n
, i
ð 14Þ
where q
n þ 1 a
, i
V
i
¼ Z
Q
W
i
q
n þ 1 a
dQ In the dp
c
dS
w
=S
w
formulation, the difference of the nodal capillary pressure values ˆp
n þ 1 c
, j
¹ ˆp
n þ 1 c
, i
in the integral is replaced by
¹ dp
c
=dS
w
ˆS
n þ 1 n
, j
¹ ˆ S
n þ 1 n
, i
. Since the node values of pressure and gravity in the integral terms above
are constants, the equations are reduced to fV
i
Dt ˆS
n þ 1 n
, i
¹ ˆ S
n n
, i
¼ X
j[h
i
ˆp
n þ 1 w
, j
¹ ˆp
n þ 1 w
, i
ÿ ¹ r
w
g ˆ D
j
¹ ˆ D
i
ÿ 3
Z
Q
l
n þ 1 w
, ij
=W
i
·K·=n
j
dQ ¹ q
n þ 1 w
, i
V
i
¹ m
n þ 1 w
, i
¹ fV
i
Dt ˆS
n þ 1 n
, i
¹ ˆ S
n n
, i
¼ X
j[h
i
ˆp
n þ 1 w
, j
¹ ˆp
n þ 1 w
, i
ÿ þ ˆp
n þ 1 c
, j
¹ ˆp
n þ 1 c
, i
ÿ ¹r
n
g ˆ D
j
¹ ˆ D
i
ÿ 3
Z
Q
l
n þ 1 n
, ij
=W
i
·K·=N
j
dQ ¹ q
n þ 1 n
, i
V
i
¹ m
n þ 1 n
, i
ð 15Þ
For homogeneous media, the permeability tensor K is
constant over the entire domain. At interfaces between zones of different rock properties, K is harmonically
weighted. We apply now the influence coefficient tech- nique
19,6
to all three discretization methods, characterized by the following approximation
where l
A a
, ij
¼ l
¯S ,
¯S being the saturation at some specific point in the intersection of the support of W
a,i
and N
j
. A and B are introduced as general superscripts indicating the
choice of discretization for the mobility term l
a
and trans- missivity integral transmissibility g
a
, respectively. Sub- stitution of Eq. 16 into eqn 15 yields the following
generalized Galerkin finite element formulation ¹
1
d
aw
f ˆS
n þ 1 n
, i
¹ ˆ S
n n
, i
V
i
Dt ¼
X
j[h
i
l
An þ 1 a
, ij
g
B a
, ij
w
n þ 1 a
, j
¹ w
n þ 1 a
, i
ÿ þ
q
n þ 1 a
, i
V
i
þ m
n þ 1 a
, i
a [ {w ,
n} ð
17Þ where
w
n þ 1 a
, i
¼ ˆp
n þ 1 w
, i
þ d
an
ˆ P
n þ 1 c
, i
¹ r
a
g ˆ D
i
18 In contrast to the generalized Galerkin finite element repre-
sentation in eqn 17 for the =p
c
formulation, the integral terms of the dp
c
dS
w
=S
w
formulation are evaluated using a four-point Gauss quadrature rule.
For the different discretization methods we will demon- strate that, in general, upstream weighting improves the
front approximation for homogeneous media. However, not all choices of upstream weighting guarantee the correct
physical behaviour of the numerical solution.
The different discretizations of the mobility term and transmissivity integral are now presented, illustrated,
and discussed.