Comput Geosci (2006) 10: 291–302
DOI 10.1007/s10596-006-9024-8
Homogenization of a Darcy–Stokes system modeling vuggy
porous media
Todd Arbogast · Heather L. Lehr
Received: 19 February 2005 / Revised: 27 March 2006 / Accepted: 5 April 2006 / Published online: 1 July 2006
© Springer Science+Business Media B.V. 2006
Abstract We derive a macroscopic model for single-
phase, incompressible, viscous fluid flow in a porous
medium with small cavities called vugs. We model the
vuggy medium on the microscopic scale using Stokes
equations within the vugular inclusions, Darcy’s law
within the porous rock, and a Beavers–Joseph–Saffman
boundary condition on the interface between the two
regions. We assume periodicity of the medium and
obtain uniform energy estimates independent of the
period. Through a two-scale homogenization limit as
the period tends to zero, we obtain a macroscopic
Darcy’s law governing the medium on larger scales.
We also develop some needed generalizations of the
two-scale convergence theory needed for our bimodal
medium, including a two-scale convergence result on
the Darcy–Stokes interface. The macroscopic Darcy
permeability is computable from the solution of a cell
problem. An analytic solution to this problem in a sim-
ple geometry suggests that: (1) flow along vug channels
is primarily Poiseuille with a small perturbation related
T. Arbogast (B)
Department of Mathematics, C1200,
The University of Texas at Austin,
Austin, TX 78712, USA
and Institute for Computational Engineering and Sciences,
C0200, The University of Texas at Austin,
Austin, TX 78712, USA
e-mail: arbogast@ices.utexas.edu
H. L. Lehr
Department of Mathematics, The Ohio State University,
231 W. 18th Avenue,
Columbus, OH 43210, USA
e-mail: heather@math.ohio-state.edu
to the Beavers–Joseph slip, and (2) flow that alternates
from vug to matrix behaves as if the vugs have infinite
permeability.
Keywords Beavers–Joseph boundary condition ·
Darcy–Stokes system · Homogenization · two-scale
convergence · vuggy porous media
1. Introduction
A vug is a cavity in a porous medium that is relatively
larger than the intergranular pore space. Vugular inclu-
sions are especially common in carbonate rocks and are
endemic to many of the world’s groundwater aquifers
and petroleum reservoirs. Although small, vugs can
significantly increase both the effective porosity and
permeability of the medium. We consider in this pa-
per a porous medium with many small vugs scattered
throughout its extent.
It is well established, both empirically and theoret-
ically, that Darcy’s law governs fluid flow in a porous
medium on scales above the pore diameter [6, 23, 27,
29]. Since the flow is expected to have a relatively low
Reynolds number, the Stokes equations should ade-
quately model fluid flow in the vugs.
In 1967, Beavers and Joseph [7] determined exper-
imentally that a free fluid in contact with a porous
medium flows faster than a fluid in contact with a com-
pletely solid surface. Although thin boundary layers
arise in both cases, the latter case is generally modeled
by assuming that all components of the velocity vanish
at the solid contact surface. In the former case, the
experiments of Beavers and Joseph demonstrate that
292 Comput Geosci (2006) 10: 291–302

the tangential velocity of the fluid cannot vanish. They [wherein q ∈ L2 () is an external source or sink satis-
proposed to account for this slippage by imposing a fying the compatibility condition that its average over

2
boundary condition of the form  vanishes, and f ∈ L2 () is a term related to body
∂Us α forces such as gravitation]:
= √ (Us − Ud ) ,
∂y K Vugular region (Stokes equations)
where ∂/∂ y is the normal derivative, Us is the tangential − 2µ 2 ∇ · Dus + ∇ ps = f in s , (1.1)
component of the Stokes velocity, Ud is the tangential
∇ · us = q in s , (1.2)
component of the Darcy velocity, K is the permeabil-
ity of the porous medium, and α is the dimensionless Rock matrix (Darcy equations)
Beavers–Joseph slippage coefficient. Saffman [24] jus-
µ(K )−1 ud + ∇ pd = f in d , (1.3)
tified this law theoretically and showed that the term
involving Ud could be dropped (see also [15, 16]). Jones ∇ · ud = q in d , (1.4)
[18] reinterpreted this law so that it applies to curved
Interface
boundaries and nontangential flows by formulating the
boundary condition in terms of the tangential compo- us · ηs = ud · ηs on 0  , (1.5)
nent of the fluid stress tensor [see equation (1.6) below]. α
We model the vuggy medium on the fine scale using 2ηs · Dus · τ = − √ us · τ on 0  , (1.6)
 K
Stokes equations in the vugs, Darcy’s law in the porous
rock, and the Beavers–Joseph–Saffman boundary con- 2µ 2 ηs · Dus · ηs = ps − pd on 0  , (1.7)
dition on the interface between the two. We assume pe- Outer boundary
riodicity of the medium and obtain as our homogenized
limit a macroscopic Darcy’s law governing the system us = 0 on ∂ ∩ ∂s , (1.8)
over large scales. To illustrate the ideas, we first derive ud ·η =0 on ∂ ∩ ∂d . (1.9)
this macroscopic model formally in section 2, and then
we derive it rigorously by the two-scale convergence The interface conditions represent continuity of mass
method [1, 4, 14, 22] in section 5. In section 3, we flux (1.5), the Beavers–Joseph–Saffman condition on
obtain the existence, uniqueness, and energy estimate the tangential stress (1.6), and the continuity of normal
results needed in the analysis. In section 4, we develop stress (1.7). Examples of direct numerical simulation of
the needed generalizations of the two-scale conver- these equations in a nonvuggy context (i.e., in a medium
gence theory needed for our bimodal medium. The with at most a few very large vugs) can be found in,
convergence of the homogenization is demonstrated in for example, [13, 19, 25]. In a vuggy context, see, for
section 5. The final section presents a simple analytical example, [3].
solution to illustrate the results. The homogenization problem is to determine the
Although our results would extend easily to R3 , for behavior of the system as  → 0. Note that in the
ease of presentation, we assume that the domain  is equations, we have scaled both the viscosity µ and
Lipschitz and bounded in R2 . We assume that the geo- the permeability K by  2 . This is the usual scaling
metric vug and pore structure of  is periodic of period for deriving Darcy’s law from Stokes flow (see [27]),
Y, where Y is a reference cell for the periodic tiling of since as  → 0, flow paths (in our case vugs) become
unit volume |Y|. The portion of the domain consisting constricted, so a corresponding decrease in viscosity is
of the vugs is denoted s , and that consisting of the required to maintain flow rates. Moreover, when ho-
porous rock is d . Let 0  be the interface between the mogenizing heterogeneity (see, e.g., [8, 26]), the ratio of
two regions. Let ηs be the outer unit normal to ∂s , and permeability to viscosity should be fixed, forcing a sim-
let τ be a unit tangent to 0  . ilar scaling of the permeability. These considerations
LetD be the symmetric gradient, i.e., D(ψ) is the ma- then imply the stated scaling of the Beavers–Joseph
1 ∂ψi ∂ψ j

boundary condition.
trix + . Denote by µ > 0 the fluid viscos-
2 ∂xj ∂ xi Below we will need to distinguish the geometry of
ity, K = K(x/) the Y-periodic, bounded, symmetric, the reference cell Y, so let Ys denote the Stokes region,
and uniformly positive-definite permeability tensor of Yd the Darcy region, and 0 the interface between the
the porous rock matrix, and α > 0 the Beavers–Joseph two. We assume that both Ys and Yd have positive
slip coefficient. The fluid velocity and pressure in the measure, and thus also the one-dimensional measure of
Stokes and Darcy regions are denoted us , ps and ud , pd , 0 is positive. As usual, x will represent a point in  and
respectively. These satisfy the following set of equations y a point in Y. In the sequel, let k · kω and (·, ·)ω denote
Comput Geosci (2006) 10: 291–302 293

the L2 (ω) norm and inner product, respectively, where Substituting the above expressions into our system
we omit the domain ω if it is . of equations (1)–(1.7), and recognizing that ∇ = ∇x +
 −1 ∇ y , we obtain the following equations. From the  −1
terms of (1.1) and (1.3), and the  0 terms of (1.7), we see
2. Formal homogenization that
∇ y ps0 = 0 in  × Ys , (2.1)
We proceed to formally homogenize our system of
equations in the usual manner [8, 14, 17, 26]. We make ∇ y pd0 = 0 in  × Yd , (2.2)
the ansatz that we can expand u` and p` (for ` = s, d)
ps0 − pd0 = 0 on  × 0. (2.3)
as
∞   ∞   It follows immediately that ps0 and pd0 are independent

X x 
X x of y and equal, so let
u` =  u`, j x,
j
and p` =  p`, j x,
j
,
 
j=0 j=0 p0 (x) = ps0 (x) = pd0 (x) on  .
where u`, j (x, y) and p`, j (x, y) are Y-periodic functions Now the  0 terms of (1.1), (1.3), and (1.5), the  −1
in y. terms of (1.2), (1.4), and (1.6), and the  1 terms of (1.7)
imply

−2µ∇ y · Dy u0s + ∇x p0 (x) + ∇ y ps1 (x, y) = f in  × Ys ,

∇ y · u0s =0 in  × Ys ,
µK(y)−1 u0d + ∇x p0 (x) + ∇ y pd1 (x, y) = f in  × Yd ,
∇ y · u0d =0 in  × Yd ,
u0s · ηs = · ηs
u0d on  × 0,
α
2ηs · Dy u0s · τ = − √ u0 · τ on  × 0,
K(y) s
2µηs · Dy u0s · ηs = ps1 − pd1 on  × 0.

With e j being the standard Cartesian basis vector in the so that

jth direction, let (ω j , 8 j ) be the periodic solution of the ū0 (x) = ū0s (x) + ū0d (x)
following auxiliary or cell problem 1X
  
= f j (x) − ∂ j p (x) ω̄sj + ω̄dj .
0
− 2∇ · Dωsj + ∇8sj = e j in Ys , (2.4) µ
j
∇ · ωsj = 0 in Ys , (2.5)
Now let the matrix K̃ be defined by
K−1 ωdj + ∇8dj = ej in Yd , (2.6) K̃i, j = ω̄sj,i + ω̄dj,i
∇ · ωdj = 0 in Yd , (2.7)
Z   Z   
1
= ωdj dy + ωsj dy . (2.13)
ωsj · ηs = ωdj · ηs on 0, (2.8) |Y| Yd i Ys i

α Then we see that

2ηs · Dωsj · τ = − √ ωsj · τ on 0, (2.9)
K µ K̃−1 ū0 + ∇ p0 = f in  . (2.14)
2ηs · Dωsj · ηs = 8sj − 8dj on 0. (2.10) Finally, using the  0 terms of (1.2) and (1.4), we
obtain
Then by linear algebra, we can express u0s and u0d as
∇x · u0` + ∇ y · u1` = q in  × Ys , ` = s, d ,
∂ p0
 
1X
u0` (x, y) = f j (x) − (x) ω`j (y), ` = s, d. (2.11) so that if we again average over Y and sum, we obtain
µ ∂xj
j Z Z
1 1 1
Define the averaging operator v̄` by averaging v` in ∇ · ū0 + ∇ y · us dy + ∇ y · u1d dy
|Y| Ys |Y| Yd
the following sense 1
Z
1
Z
= ∇ · ū0 + u1s · ηs dS + u1 · ηd dS
|Y| ∂Yd d
Z
1 |Y| ∂Ys
v̄` = v` (y) dy , (2.12)
|Y| Y` =q.
294 Comput Geosci (2006) 10: 291–302

By the periodicity of u1` in y, and the fact that u1s · ηs = Since λ is constant, we can take ψ = λ and conclude
−u1d · ηd on 0, we see that that λ = 0 and, further, that K̃ is positive definite. 

∇ · ū0 = q on  (2.15)
3. Existence and a priori energy estimates
Thus, we conclude from the formal analysis that ū0
should satisfy a Darcy’s law on all of  (2.14) and (2.15),
In this section, we prove a theorem that gives existence
with effective permeability matrix K̃, independent of
of solutions u and p to (1.1)–(1.9) for each  and
the fluid viscosity and given by (2.13).
energy estimates for u and p independent of . Let

V  = V  () = v ∈ H(div, ) | vs = v|s ∈ H1 (s ) ,


Lemma 2.1. The tensor K̃, as defined by (2.13) and
(2.4)–(2.10), is symmetric and positive definite.
V0 = V0 () = v ∈ V  () | v · η = 0 on


∂ ∩ ∂d and v = 0 on ∂ ∩ ∂s ,

Proof. The existence and uniqueness of a weak so-
lution to (2.4)–(2.10) follow from an analysis simi- where η is the outward unit normal to , and let W =
lar to that given below for Theorem 3.1. Note that L2 ()/R .
(2.4)–(2.10) are equivalent to the variational equation We first recast the original problems (1.1)–(1.9) into

α
 a variational problem. Combine equations (1.2) and
2 (Dωis , Dψ)Ys + √ ωis · τ, ψ · τ (1.4), multiply by a test function w ∈ W, and integrate
K 0
over . Then combine and multiply equations (1.1) and
(1.3) by a test function v ∈ V  , integrate, integrate by
 
+ K ωi , ψ
−1 d
= (ei , ψ)Y , (2.16)
Yd parts, and manipulate the boundary terms to obtain the
for ψ an infinitely differentiable and periodic vector variational form of the system for u ∈ V0 and p ∈ W
function in Y such that ∇ · ψ = 0. With ψ = ωsj on Ys satisfying
and ψ = ωdj on Yd (actually a sequence approaching the µα 
 

2
Dus , Dv  + √ us · τ, vs · τ


same), we obtain that 2µ
s K 0
 
− p , ∇ · v  + µ (K )−1 ud , v 
   
|Y| K̃i, j = ei , ωsj + ei , ωdj


Ys  Yd d
α s
  
= 2 Dωi , Dω j + √ ωi · τ, ω j · τ = ( f, v) , v ∈ V0 ,
s s s
(3.1)
Ys K 0
∇ · u , w  = (q, w) , w ∈ W,
 

+ K ωi , ω j
−1 d d , (3.2)
Yd
where ud = u|d and us · ηs = ud · ηs on 0  is implicit
and symmetry follows immediately. from u ∈ V0 .
To show that K̃ is positive definite, take any λ ∈ R2
and define
Theorem 3.1. For each , there exists (u , p ) ∈ V0 × W
ξ ` (y) =
X
λi ωi` for y ∈ Y` . satisfying (1.1)–(1.9) weakly, i.e., (3.1) and (3.2), such
i
that
√
Then, from (2.16), we conclude that k∇us ks + kus · τ k0  + ku k
+ k∇ · u k + k p k ≤ C (k f k + kqk) (3.3)
α s
 
|Y| λ K̃λ = 2(Dξ , Dξ )Ys + √ ξ · τ, ξ · τ
T s s s
K 0 with C independent of .
+ K ξ ,ξ Y ,
−1 d d


d
In order to prove this result, we first prove a lemma
and that K̃ is positive semi-definite. To see definiteness, related to Korn’s inequality.
suppose that λT K̃λ = 0 and conclude that each inte-
grand above vanishes. But now (2.16) implies that Lemma 3.2. There exists C independent of  such that
for all v ∈ V0 (),
α
 
0 = 2(Dξ s , Dψ)Ys + √ ξ s · τ, ψ · τ kvs ks + k∇vs ks ≤ C kDvs ks
K 0
√
+ kvs · τ k0  + kvd kd + k∇ · vk .


+ K ξ , ψ Y = (λ, ψ)Y .
−1 d


d
Comput Geosci (2006) 10: 291–302 295

If d is not empty, the above inequality holds for v ∈ v((y + nEi )) ∈ V(Y ). By our remark above, if ∂ inter-
V  (). sects the interior of a cell (i.e.,  is not tiled exactly by
scaled translates of Y ), the inequality (3.4) still holds on
Proof. First we show that a similar result holds for v̂ ∈ that truncated cell since v ∈ V0 . Thus, inequality (3.4)
V(Y ) = {v ∈ H(div, Y ) | vs = v|Ys ∈ H1 (Ys )}, and then holds on each Y i , so
we use a translation and scaling argument to pass to all
of . Suppose it is not true that there exists Ĉ such that kv̂si kYsi + k∇ v̂si kYsi ≤ Ĉ kDv̂si kYsi
kv̂s kYs + k∇ v̂s kYs ≤ Ĉ kDv̂s kYs + kv̂si · τ k0i + kv̂di kY i + k∇ · v̂ i kY i ,


d
+ kv̂s · τ k0 + kv̂d kYd + k∇ · v̂kY .


(3.4)
and if we sum over all i and make the change of var-
Then there exists a sequence {v̂n }∞
n=1 ∈ V(Y ) such that iables x = (y + nEi ) on each Y i , we obtain
kv̂n,s kYs + k∇ v̂n,s kYs = 1 (3.5)
 −1 kvs ks + k∇vs ks ≤ C kDvs ks
and
+  −1/2 kvs · τ k0  +  −1 kvd kd + k∇ · vk


kDv̂n,s kYs + kv̂n,s · τ k0
1
+ kv̂n,d kYd + k∇ · v̂n kY ≤. (3.6) which gives the desired result v ∈ V0 .
n For v ∈ V  , as long as d is not empty, we can adjoin
The last two terms on the left-hand side above tell any truncated cell to an entire cell lying beside it (at
us that v̂n,d → 0 in H(div, Yd ), and (3.5) implies that least for  small). Since there will be a portion of 0
v̂n,s * v̂s weakly in H1 (Ys ) for some v̂s . Let v̂ be the within this new composite cell, we have the inequal-
extension by zero of v̂s to Y. Since v̂n is bounded in ity (3.4) on the composite cell regardless of the outer
H(div, Y ), it converges weakly, and we conclude that boundary conditions. Hence, if d is not empty, the
in fact theorem holds also for all v ∈ V  . 
v̂n * v̂ in H(div, Y ).
Proof. Proof of Theorem 3.1. The theorem follows
Since from the inf–sup theory of saddle point problems [5,
kv̂n,d · ηk(H1/2 (0))∗ ≤ Ckv̂n,d k H(div,Yd ) → 0 , 10–12]. For u, v ∈ V0 () and w ∈ W, we have the bi-
00 linear forms
where the norm on the left-hand side is the norm of the
1/2 µα
 
dual space of H00 (0) (see [20]), we conclude that v̂d · a  (u, v) = 2µ 2 (Dus , Dv)s + √ us · τ, vs · τ
η = v̂s · η = 0 on 0. On the other hand, kv̂n,s · τ k0 → K 0
0, so that v̂s · τ = 0 and further that v̂s = 0 on 0. Now +(µ(K )−1 ud , v)d ,
Korn’s inequality can be applied on Ys to show that b(v, w) = (w, ∇ · v) .
kv̂n,s kYs + k∇ v̂n,s kYs ≤ CkDv̂n,s kYs .
Then (3.1) and (3.2) can be rewritten as follows: find
But the left-hand side is 1 by (3.5), and the right-hand (u , p ) ∈ V0 × W such that
side tends to 0 by (3.6), contradicting the assumption
that the inequality (3.4) fails to hold. Now we have a  (u , v) − b(v, p ) = ( f, v), v ∈ V0 , (3.7)
inequality (3.4) for any v̂ ∈ V(Y ), wherein Ĉ does not
depend on . b(u , w) = (q, w), w ∈ W. (3.8)
We remark that we also have the inequality for
functions defined only on an open subset Yp ⊂ Y, as We endow V0 () with the norm
long as Yp ∩ Yd is not empty. When Yp ⊂ Ys , we can
obtain the same inequality provided that we have the
|||u||| = (kuk2 + k∇ · uk2 +  2 k∇us k2s )1/2 ,
boundary condition v̂ = 0 on some positive measure
subset of ∂Yp .
[ for which it is complete. We claim that
By the structure of , we can write it as  = Y i ,
i∈I µα
 
Y i= (Y + nEi ) ∩ , nEi is some vector whose √ us · τ, vs · τ ≤ Ckus · τ k0  kvs · τ k0 


where
components are integers, and I is some appropriate K 0

index set. Let v ∈ V0 () and define for y ∈ Y, v̂ i (y) = ≤ C|||u||| |||v||| , (3.9)
296 Comput Geosci (2006) 10: 291–302

so that both a  and b are bounded (i.e., continuous) with differentiable functions in  × Y that have compact
constants independent of . To see the claim, compute support in  and are periodic in Y, and we recall the
X following definition.
kus · τ k20  =  kus · τ k20i
Definition 4.1. If {u } ⊂ L2 () and u0 (x, y) ∈ L2 ( ×
i∈I

Y ) are such that

X
= kûs · τ̂ k20i
i∈I Z
X lim u (x) φ(x, x/) dx
≤ 2 ĈkûkYsi kûk H1 (Ysi ) →0 
i∈I
Z Z
1
X = u0 (x, y) φ(x, y) dy dx
≤ Ĉ kukYsi kuk H1 (Ysi ) |Y|  Y
i∈I for any function φ ∈ D(; C#∞ (Y )), then {u } is said to
Ĉ X  two-scale converge in  × Y to u0 (x, y), and we write
kuk2Y i +  2 kuk2H1 (Y i )

≤ this as
2 s s
i∈I
u *
* u0 in  × Y as  → 0 .
Ĉ 
kuk2s +  2 kuk2H1 ( )

=
2 s
Lemma 4.1. Let ` = s or d and χ` be the character-
≤ C|||u|||2 . (3.10) istic function on ` . If u is such that ku k ≤ C for
some constant C independent of , then a subsequence
Moreover, a  is coercive on V0 ∩ {v ∈ V0 : ∇ · v = 0} by
of χ` u two-scale converges to some ψ0` ∈ L2 ( × Y)
Lemma 3.2, with bound independent of .
such that supp(ψ0` ) ⊂  × Ȳ` . Moreover, if u two-scale
It remains to show the inf–sup condition, but this
converges to u0 ∈ L2 ( × Y ), then χ` u ** u0 |×Y` in
follows from the corresponding condition known for the
 × Y` .
Stokes system on ; that is,
(∇ · v, w) (∇ · v, w) Proof. Because kχ` u k ≤ ku k ≤ C, a subsequence of
inf sup ≥ inf sup
w∈W v∈V 
0
|||v||| kwk w∈W v∈(H1 )2 |||v||| kwk
0
χ` u two-scale converges to some ψ0` ∈ L2 ( × Y ) [1].
Take a test function φ ∈ D(; C#∞ (Y )) supported in
(∇ · v, w)
≥ inf sup  × Yk where k 6= `. Then φ  (x) = φ(x, x/) is sup-
w∈W
v∈(H10 )2 2kvk(H1 ())2 kwk ported in k and
≥γ >0,
Z
0 = lim χ` u φ  dx
→0 
for some γ independent of , since (H01 )2 ⊂ V0 and Z Z
2kvk(H1 ())2 ≥ |||v||| . 1
= ψ ` φ dy dx
Now the inf–sup theory provides the existence and |Y|  Y 0
uniqueness of a solution to our systems (3.7) and (3.8) 1
Z Z
[12]. Moreover, = ψ ` φ dy dx.
|Y|  Yk 0
|||u ||| + k p k ≤ C(k f k + kqk), (3.11)
This holds for all such φ, so ψ0` = 0 on  × Yk.
where C depends on γ , the coercivity bound for a  , Now, take a test function φ ∈ D(; C#∞ (Y )) with
and the continuity bounds for a  and b, each of which support in  × Y` . Then
is independent of . Finally, (3.10) and (3.11) imply Z Z
  
(3.3).  χ` u φ dx = u φ  dx ,
 

so, taking the limit as  → 0,

4. Two-scale convergence results for bimodal media 1
Z Z
1
Z Z
`
ψ φ dy dx = u0 φ dy dx .
|Y|  Y` 0 |Y|  Y
In this section, we make note of some slight extensions
of the two-scale convergence results of Allaire [1, 14]. An application of Lusin’s theorem completes the
Lemmas 4.1 and 4.4 can be deduced easily from the lemma. 
proof of Theorem 2.7 in [1]. We include the following
statements and proofs for clarity and completeness. We Lemma 4.1 allows us to make the following definition
first recall that D(; C#∞ (Y )) is the set of infinitely and gives the following corollary. Note that a function
Comput Geosci (2006) 10: 291–302
in D(; C#∞ (Y` )) is considered to be only Y-periodic in
y, with no condition imposed on 0.
Definition 4.2. If ` = s or d and {u` } ⊂ L2 (` ) is such
that, for any function φ(x, y) in D(; C#∞ (Y` )),
Z
lim u` (x) φ(x, x/) dx
→0 
`
Z Z
1
= u0 (x, y) φ(x, y) dy dx
|Y|  Y`
for some u0 (x, y) in L2 ( × Y` ), then {u` } is said to
two-scale converge in  × Y` to u0 (x, y) as  → 0.
Corollary 4.2. If u` ∈ L2 (` ) and there exists C > 0
such that ku` k` ≤ C for all  > 0, then there exists
a subsequence that two-scale converges in  × Y` to
u0 ∈ L2 ( × Y` ).
The following lemma is immediate and illuminates
the connection between weak and two-scale conver-
gence.
Lemma 4.3. If {u` } two-scale converges to u0 (x, y) in
 × Y` , and û` denotes the extension Zof u` by zero to
1 `
, then û` converges weakly to u0 (x, y) dy in
|Y| Y`
L2 ().
The next results will be needed to prove our homoge-
nization result. As usual, H#1 (Y ) denotes the Y-periodic
functions in H1 (Y ).
Lemma 4.4. Fix ` = s or d.
(a) If ku` k` ≤ C and k∇u` k` ≤ C for some con-

2
stant C, then there exists u0,` ∈ L2 (; H#1 (Y` ))
such that some subsequence of {u` } two-scale
converges in  × Y` to u0,` and {∇u` } two-scale
converges in  × Y` to ∇ y u0,` .
(b) If ku` k` ≤ C and k∇ · u` k` ≤ C for some con-
stant C, then there exists u0,` ∈ L2 (; H (div,
Y` )))2 such that some subsequence of {u` } two-
scale converges in  × Y` to u0,` and {∇ · u` } ,
Z to  by zero, converges weakly in L ()
extended 2
1 Lemma 4.4
to ∇x · u0,` dy. Moreover, ∇ y · u0,` = 0.
|Y| Y`
Proof. For result (a), by Corollary 4.2, we have for
some subsequence both
u` *
* u0,` in  × Y` ,
∇u` *
* ψ0,` in  × Y` .
297

2
Let φ ∈ D(; C#∞ (Y` )) be such that φ|×0 = 0, and
let φ  (x) = φ(x, x/). Compute
(∇u` , φ  )` = −(u` , ∇x · φ  + ∇ y · φ  )` ,
so that as  → 0,
(ψ0,` , φ)×Y` = −(u0,` , ∇ y · φ)×Y`
= (∇ y u0,` , φ)×Y` − (u0,` · η, φ)×(∂Y` \0) .
With φ|×∂Y` = 0, we conclude that ψ0,` = ∇ y u0,` .
Then we further conclude that u0,` is periodic in y ∈ Y` ,

2
i.e., that u0,` ∈ L2 (; H#1 (Y` )) .
For (b), Corollary 4.2 gives us two-scale convergence
of uZ` to u0,` , and then weak convergence of ∇ · u` to
1
∇x · u0,` dy follows easily. To obtain ∇ y · u0,` =
|Y| Y`
0, note that for φ ∈ D(; C0∞ (Y )),
Z Z  
∇ · u` φ  dx = − u` · ∇x φ  +  −1 ∇ y φ  dx ,
` `
By Corollary 4.2, the left-hand side and the first term
on the right-hand side both converge as  → 0. Thus,
we obtain
Z
lim u` · ∇ y φ dx = 0 ,
→0 
which implies that ∇ y · u0,` = 0. 
Lemma 4.5. If us is such that kus ks and k∇us ks
are bounded independent of  and 0  is a smooth
submanifold of  , then for φ ∈ (D(; C#∞ (Ys )))2 ,
lim (us · τ, φ · τ )0  = |Y|−1 (u0,s · τ, φ · τ )×0 ,
→0
where u0,s is the two-scale limit of us in  × Ys .
Moreover,
lim (us · ηs , φ · ηs )0  = |Y|−1 (u0,s · ηs , φ · ηs )×0 .
→0
Proof. Let 8 ∈ (D(; C#∞ (Ys )))2×2 and 8 (x) = 8 (x,
x/). Then
(∇us , 8 )s = −(us , ∇x · 8 )s
−(us , ∇ y · 8 )s + (us , 8 · ηs )0  .
Taking the limit of both sides as  → 0, we obtain from
|Y|−1 (∇ y u0,s , 8)×Ys = −|Y|−1 (u0,s , ∇ y · 8)×Ys
+ lim (us , 8 · ηs )0  .
→0
This implies that
|Y|−1 (u0,s , 8 · ηs )×0 = lim (us , 8 · ηs )0  . (4.1)
→0
298 Comput Geosci (2006) 10: 291–302

Since 0 is smooth, the tubular neighborhood the- L2 ( × Y) and u0 (x, y) ∈ L2 ( × Y) such that the fol-
orem from topology allows us to extend the normal lowing two-scale convergences hold:
vector field ηs on 0 to a smooth vector field N̂(y) on
p *
* p0 in  × Y, (5.4)
Y. If we do this locally, and patch the results together
using a partition of unity argument, we can obtain u *
* u0 in  × Y, (5.5)
smooth vector fields after periodic extension. We can
u` *
* u0,` in  × Y` , (5.6)
then define N and T  on all of  by setting N (x) =
N(x/) = N̂(y) and T  (x) = T(x/) = T̂(y). ∇us *
* ∇ y u0,s in  × Ys . (5.7)
Now take 8 = T  (φ · T  )(N )T in (4.1), where φ ∈
2
(D(; C#∞ (Ys )))2 . This yields the first result. Replacing Moreover, u0,s ∈ L2 (; H#1 (Ys )) ,
T  with N gives the second result.  ∇ · u = q , (5.8)

In fact, the following more general definition makes and

sense and was previously stated in [2, 9, 21]. ∇ y · u0 (x, y) = 0 . (5.9)
Let 9(x, y) ∈ C0∞ ( × Y) be Y-periodic. Take v(x) =
Definition 4.3. If ` = s or d and {u` } ⊂ L2 (0  ) is such
9(x, x/) in the variational problem (3.1), so that
that, for any function φ(x, y) in D(; C#∞ (0)) which is
p , ∇ y · 9 + O() = 0 .


Y-periodic in y,
As  → 0, we obtain
Z
lim u` (x) φ(x, x/) dx
→0 0 
p0 , ∇ y · 9 ×Y = 0 .


Z Z
1
= u0 (x, y) φ(x, y) dS dx This implies that ∇ y p0 = 0, so that p0 (x, y) = p0 (x)
|Y|  0
only.
for some u0 (x, y) in L2 ( × 0), then {u` } is said to two- Next, take 9 ∈ (D(; C#∞ (Y )))2 with ∇ y · 9 = 0, and
scale converge on  × 0 to u0 (x, y) as  → 0. let v(x) = 9(x, x/) in the variational equation (3.1), so
that
µα
 
2µ 2 Dus , D9   + √ us · τ, 9  · τ


5. Proof of the homogenization result s K 0
 
+ µ(K )−1 ud , 9   − p , ∇x · 9  = ( f, 9  ).


We now prove rigorously the homogenization results d
obtained formally in section 2. In the first theorem,
Using Lemma 4.5, passing to the two-scale limit gives
we obtain only weak convergence in H(div, ) × W
µα
 
to the solution of the homogenized problem. In the 2µ (Dy u0,s , Dy 9)×Ys + √ u0,s · τ, 9 · τ
second theorem, we show that, in fact, we have strong K ×0
convergence of (u , p ) in L2 () × W.
+(µK−1 u0,d , 9)×Yd

Theorem 5.1. There exists (u, p) ∈ H(div, ) × W such −( p0 , ∇x · 9)×Y = ( f, 9)×Y .

that the velocity u converges weakly to u in H(div, ), Integrating by parts and collecting terms, we obtain
p converges weakly to p in W, and (u, p) is the unique
solution to the homogenized Darcy problem (−2µ∇ y · Dy u0,s + ∇x p0 − f, 9)×Ys
+(µK−1 u0,d + ∇x p0 − f, 9)×Yd
µ K̃−1 u + ∇ p = f in , (5.1)
µα
 
∇ ·u=q in , (5.2) + √ u0,s · τ + 2µτ · Dy u0,s · ηs , 9 · τ
K ×0

u·η =0 on ∂, (5.3) +(2µηs · Dy u0,s · ηs , 9 · ηs )×0 = 0 . (5.10)

where the tensor K̃ is defined by (2.4)–(2.10) and (2.13).
Proof. By our energy estimates in Theorem 3.1 and the
two-scale convergence results of [1, 14], Corollary 4.2,
and Lemma 4.4, for ` = s, d, there exists p0 (x, y) ∈
It is a well-known result that if (8, 9)×Y = 0, with
∇ y · 9 = 0 and 9 ∈ D(; C0∞ (Y ))2 , then 8 = ∇ y ψ for
some ψ ∈ L2 (; H1 (Y )) [28]. Restrict to 9 ∈ (D(;
C0∞ (Ys )))2 to obtain
(−2µ∇ y · Dy u0,s + ∇x p0 − f, 9)×Ys = 0 .
Comput Geosci (2006) 10: 291–302 299

Thus, there exists p1,s (x, y) ∈ L2 (; H1 (Ys )) such that so integrating by parts yields
in  × Ys ,
− 2µ∇ y · Dy u0,s + ∇x p0 − f = −∇ y p1,s .
µα
 
√ u0,s · τ + 2µτ · Dy u0,s · ηs , 9 · τ
Likewise, we obtain p1,d (x, y) ∈ L2 (; H1 (Yd )) such K ×0
that in  × Yd ,
+ 2µηs · Dy u0,s · ηs − p1,s
K−1 u0,d + ∇x p0 − f = −∇ y p1,d .
+ p1,d , 9 · ηs ×0 = 0 .


(5.11)
Thus, for all 9 ∈ D(; C#∞ (Y ))2 satisfying the diver-
gence constraint,
(−∇ y p1,s , 9)×Ys + (−∇ y p1,d , 9)×Yd By noting, for example, that there exists a weak so-
lution (see [28] or the proof of Theorem 3.1) 9 ∈
µα
 
+ √ u0,s · τ + 2µτ · Dy u0,s · ηs , 9 · τ L2 (; (H#1 (Y ))2 ), w ∈ L2 (; L2 (Y )/R) to
K ×0
+(2µηs · Dy u0,s · ηs , 9 · ηs )×0 = 0 ,

−1 y 9 + ∇ y w =0 on  × Ys ,
∇y · 9 =0 on  × Ys ,
9 · ηs =0 on  × 0,
9 ·τ = √µα u0,s · τ + 2µτ · Dy u0,s · ηs on  × 0,
K
9 =0 on  × (∂Ys \ 0),

and by the fact that D(; C#∞ (Y)) is dense in the two-scale variational equations (5.9) and (5.10) are
L2 (; H#1 (Y)), we obtain that each individual term in equivalent to
equation (5.11) vanishes. We then finally obtain that

− 2µ∇ y · Dy u0,s + ∇x p0 (x) + ∇ y p1,s (x, y) = f in  × Ys , (5.12)

∇ y · u0,s = 0 in  × Ys , (5.13)

µK−1 u0,d + ∇x p0 (x) + ∇ y p1,d (x, y) = f in  × Yd , (5.14)

∇ y · u0,d = 0 in  × Yd , (5.15)

2µ ηs · Dy u0,s · ηs = p1,s − p1,d on  × 0, (5.16)

α
2τ · Dy u0,s · ηs = − √ u0,s · τ on  × 0. (5.17)
K

Let (ωsj , 8sj ) and (ωdj , 8dj ) be Y-periodic solutions to

the auxiliary problem on Ys and Yd given in (2.4)–(2.10).
N  
Then, because the above problem has a unique solution, 1 X
= f j (x) − ∂x j p0 (x)
it is clear that we can express u0,s and u0,d as in (2.11): µ|Y|
j=1
N 
∂ p0

1 Z Z 
(x) ω`j (y) ,
X
u0,` (x, y) = f j (x) − ωsj (y) + ωdj (y)
µ ∂xj Ys Yd
j=1

where ` = s or d. Averaging over Ys and Yd as in (2.12), K̃

= ( f − ∇ p0 ) , (5.18)
we get µ

ū0 = ū0,s + ū0,d , where K̃ is as in equation (2.13).

300 Comput Geosci (2006) 10: 291–302

By Lemma 4.3 and equation (5.8), we obtain the since ū0 satisfies the system of equations (5.12)–(5.17).
weak convergence results in L2 () Finally, by the symmetry of a  (·, ·), we have
Z
 1
p * p0 (x, y) dy = p0 (x) , (5.19)
|Y| Y
lim a  (u − u0 , u − u0 ) = 0 ,
→0
Z
 1
u * u0 (x, y) dy = ū0 , (5.20)
|Y| Y
q = ∇ · u * ∇ · ū0 = q . (5.21) and by Lemma 3.2, we see that lim ku − u0 k L2 ( ) +

→0
k∇(us − u0,s )k L2 (s ) = 0.

Setting p = p0 and u = ū0 , equations (5.18)–(5.21) give
the theorem, since convergence on the boundary of the Now we show the strong convergence of the pressure
domain is trivial.  p to p0 in W. By the inf–sup condition for Stokes,

Theorem 5.2. Let (u0 , p0 ) ∈ L2 ( × Y) × W be the two-

scale limit of (u , p ) as before. Then u − u0 converges (w, ∇ · v)
to 0 strongly in L2 (), where u0 = u(x, x/), and p sup > γ kwk0 ,
v∈H01 kvk1
converges strongly to p0 (x) in W.

Proof. Since (ū0 , p0 ) satisfies (5.1) and (5.2) and µ

and K̃ are smooth, if f ∈ H1 () and q ∈ L2 , then p0 ∈ there exists a sequence v  such that kv  k1 = 1 and
H2 (). It follows that ū0 ∈ H1 () and ∇x u0 (x, y) ∈
L2 (). We also know that ∇ y u0 (x, y) = 0 ∈ L2 ( × Ys ),
so u0 (x, x/) ∈ V  . Since D() is dense in V  , we γ 
p − p0 , ∇ · v  > k p − p0 k0 .



(5.24)
can use u0 as a two-scale test function. By two-scale 2
convergence

lim a  (u , u0 ) = ā(u0 , u0 ) , By Theorem 3.5 in Appendix A of [14], there ex-
→0
ists v0 (x) ∈ H1 () and v1 (x, y) ∈ L2 (; H#1 (Y)/R) such
where that (possibly after passing to a subsequence)
ā(v, z) = 2µ Dy v, Dy z ×Y


s

µα v * ∇v  *
 
* v0 and * ∇x v0 (x) + ∇ y v1 (x, y) .
 
+ √ vs · τ, zs · τ + µK−1 v, z .
K ×0 ×Yd

Similarly,
Note that we can assume v1 is such that its average over
lim a  (u0 , u0 ) = ā(u0 , u0 ) Y is zero. It follows that
→0

by two-scale convergence. Subtracting, we obtain lim a 

→0
(u − u0 , u0 ) = 0.
 Z 

lim p0 , ∇ · v p0 , ∇ · v0 +



= ∇v1 dy
We can also show that lim a  (u , u ) = ā(u0 , u0 ). →0 Y 
→0
Since = ( p0 , ∇ · v0 ) ,

a  (u , u ) − b(u , p ) = ( f, u ), (5.22)

b(u , p ) = (q, p ), (5.23) by the periodicity of v1 . Now,

we know that a  (u , u ) = ( f, u ) + (q, p ) . Taking

the limit of both sides as  → 0, by our weak con-
lim ( p , ∇ · v  ) = lim ( p , ∇ · (v  − v0 ))
vergence results in Theorem 5.1, the right-hand side →0 →0
converges to ( f, ū0 ) + (q, p0 ) . By (5.2), q = ∇ · ū0 , so + ( p0 , ∇ · v0 )
integrating by parts gives that = lim a  (u , v  − v0 )
→0
lim a  (u , u ) = ā(u0 , u0 ) , + ( p0 , ∇ · v0 ) .
→0
Comput Geosci (2006) 10: 291–302 301

Finally, we can show which has flow in the y1 -direction only. It follows from
(2.13) that K̃21 = 0 (so K̃ is diagonal), and
lim a  (u , v  − v0 ) √
→0  
 1 1 3 K 2
= lim 2µ  Du ,  D(v  − v0 ) 

K̃11 = h + h + K(` − h) . (6.4)
→0 s
` 12 2α
 q 
  √ 
+ µα K us · τ, (v − v0 )s · τ This should be contrasted to the situation in which
 0 the porous matrix is replaced by an impermeable

+ µK u , v − v0 
−1  medium. Then ωsj = 0 on 0, and we have the well-


d
 known problem of Poiseuille flow in a pipe. The solu-

= lim 2µ  D(u − u0 ),  D(v  − v0 ) 



tion is
→0 q s

   √  1
+ µα K (u − u0 )s · τ, (v − v0 )s · τ ω̌1s (y) = − y2 (h − y2 )e1 , (6.5)
0 2

+ µK (u − u0 ), v − v0 
−1 


d
8̌1 (y) = 0. (6.6)
+2µ  Du0 ,  D(v  − v0 ) 


 q s  In this case, we would compute the effective permeabil-
  √  ity for a unit pressure drop in the y1 -direction (which
+ µα K u0,s · τ, (v − v0 )s · τ
 0 corresponds to the forcing function e1 ) as

+(µK u0 , v − v0 )d
−1 
h3
=0, K̃11,Poiseuille = , (6.7)
12`

where the first three terms above converge to zero be-
cause lim ku − u0 k0 = 0, lim k∇(u − u0 )k0 = 0, and
→0 →0
because v  − v0 is bounded in H1 . The final three terms
above converge to zero by the two-scale convergence
results for v  (and because one term has an extra factor
of ). Putting everything together, we have that
lim p − p0 , ∇ · v  = 0


→0
as  → 0. By (5.24), p → p0 as  → 0 strongly in W.
 K̃11 = K̃11,Poiseuille + K̃11,Beavers-Joseph + K̃11,Darcy ,(6.9)
which is the first term on the right side of (6.4).
If instead we assume the vugular region is imperme-
able, the bulk flow would be reduced from the porous
medium case by the geometric factor (` − h)/`:
`−h
K̃11,Darcy = K. (6.8)
`
This is the last term on the right side of (6.4). Thus,
when considering flow in the direction of the vugular
channel, we have the representation

where
√
6. A simple analytical solution of the auxiliary problem K 2
K̃11,Beavers-Joseph = h (6.10)
2α`
It is not so easy to construct analytical solutions to
represents the Beavers–Joseph interface effect of fluid
the auxiliary problems (2.4)–(2.10), except at least
slippage. This term increases the Darcy–Stokes flow
in the following case. Let Y = (0, `) × (0, `) be a
from the arithmetic average of the pure Stokes “pipe
square of side length ` > 0. With Ys = (0, `) × (0, h)
flow” and the pure Darcy flow. Note that we generally
and Yd = (0, `) × (h, `) repeated periodically, we have
have K  h2 and α = O(1). Thus, K̃11,Poiseuille is the
a horizontally layered medium. Note that 0 con-
leading term, and K̃11,Beavers-Joseph is the next order
sists of two segments, y2 = h and, by periodicity,
term in the expansion.
y2 = 0 ⇐⇒ y2 = `.
When j = 2 in the auxiliary problems (2.4)–(2.10), it
When j = 1, it is easy to verify that the solution is
is easy to verify that the solution is
√
`
 
1 K
ω1 (y) =
s 2
− y2 + hy2 + h e1 , (6.1) ω2 (y) = Ke2 , (6.11)
2 α `−h

ω1d (y) = Ke1 , (6.2) 8s2 (y) = y2 , (6.12)

h
81 (y) = 0, (6.3) 8d2 (y) = (` − y2 ). (6.13)
`−h
302
Again, K̃12 = K̃21 = 0 and
`
K̃22 = K. (6.14)
`−h
In this case, the flow is entirely in the y2 -direction.
It is well known and easily verified that one-
dimensional flow across a porous medium of permeabil-
ity k1 for distance h and k2 for distance ` − h results
in a flow rate that is the same as that in a uniform
`k1 k2
medium of permeability k = , which is
hk2 + (` − h)k1
the harmonic average permeability. If we apply this
result to our case, assuming that the vugular channel has
infinite permeability, we obtain exactly (6.14).
In conclusion, this example suggests that the effective
permeability represents an average of two extremes.
When the vugs are interconnected in some direction,
we have primarily Poiseuille flow behavior, with a low
order correction term for the effective permeability
related to the Beavers–Joseph slip (and an even lower
order correction related to flow entirely in the porous
matrix). When the flow is along paths that alternate
from vug to matrix, the fluid behaves as if it were flow-
ing in a porous medium with the vugs having infinite
permeability.
Acknowledgement We thank the anonymous referee for sug-
gesting Theorem 5.2, and the US National Science Foundation
for supporting this work (under grants DMS-0074310 and DMS-
0417431).
References
1. Allaire, G.: Homogenization and two-scale convergence.
SIAM J. Math. Anal. 23, 1482–1518 (1992)
2. Allaire, G., Damlamain, A., Hornung, U.: Two-scale conver-
gence on periodic surfaces and applications. In: Bourgeat,
A. et al. (eds.) Proceedings of the International Confer-
ence on Mathematical Modelling of Flow Through Porous
Media (May 1995), pp. 15–25. World Scientific Pub.,
Singapore (1996)
3. Arbogast, T., Brunson, D.S.: A computational method for
approximating a Darcy–Stokes system governing a vuggy
porous medium (in press).
4. Arbogast, T., Douglas Jr., J., Hornung, U.: Derivation of the
double porosity model of single phase flow via homogeniza-
tion theory. SIAM J. Math. Anal. 21, 823–836 (1990)
5. Babuška, I.: The finite element method with Lagrangian mul-
tipliers. Numer. Math. 20, 179–192 (1973)
6. Bear, J.: Dynamics of Fluids in Porous Media. Dover, New
York (1972)
7. Beavers, G.S., Joseph, D.D.: Boundary conditions at a natu-
rally permeable wall. J. Fluid Mech. 30, 197–207 (1967)
Comput Geosci (2006) 10: 291–302
8. Bensoussan, A., Lions, J.L., Papanicolaou, G.: Asymptotic
Analysis for Periodic Structure. North Holland, Amsterdam
(1978)
9. Bouchitté, G., Fragala, I.: Homogenization of thin structures
by the two-scale method with respect to measures. SIAM J.
Math. Anal. 32(6), 1198–1226 (2001)
10. Brenner, S.C., Scott, L.R.: The Mathematical Theory of Finite
Element Methods. Springer-Verlag, New York (1994)
11. Brezzi, F.: On the existence, uniqueness and approximation
of saddle-point problems arising from Lagrangian multipliers.
RAIRO 8, 129–151 (1974)
12. Brezzi, F., Fortin, M.: Mixed and Hybrid Finite Element
Methods. Springer-Verlag, New York (1991)
13. Gartling, D.K., Hickox, C.E., Givler, R.C.: Simulation of cou-
pled viscous and porous flow problems. Comp. Fluid Dyn. 7,
23–48 (1996)
14. Hornung, U. (ed.): Homogenization and Porous Media, In-
terdisciplinary Applied Mathematics Series. Springer-Verlag,
New York (1997)
15. Jäger, W., Mikelić, A.: On the boundary conditions at the con-
tact interface between a porous medium and a free fluid. Ann.
Sc. Norm. Super. Pisa, Classe Fis. Mat. Ser. IV 23, 403–465
(1996)
16. Jäger, W., Mikelić, A.: On the interface boundary condition
of Beavers, Joseph, and Saffman. SIAM J. Appl. Math. 60,
1111–1127 (2000)
17. Jikov, V.V., Kozlov, S.M., Oleinik, O.A.: Homogenization
of Differential Operators and Integral Functions. Springer-
Verlag, New York (1994)
18. Jones, I.P.: Low Reynolds number flow past a porous spheri-
cal shell. Proc. Camb. Philol. Soc. 73, 231–238 (1973)
19. Layton, W.J., Schieweck, F., Yotov, I.: Coupling fluid flow
with porous media flow. SIAM J. Numer. Anal. 40, 2195–2218
(2003)
20. Lions, J.L., Magenes, E.: Non-Homogeneous Boundary
Value Problems and Applications 1. Springer-Verlag, Berlin
(1970)
21. Neuss-Radu, M.: Some extensions of two-scale convergence.
C. R. Acad. Sci., Sér. 1 Math. 322, 899–904 (1996)
22. Nguetseng, G.: A general convergence result for a functional
related to the theory of homogenization. SIAM J. Math. Anal.
20, 608–623 (1989)
23. Peaceman, D.W.: Fundamentals of Numerical Reservoir
Simulation. Elsevier, Amsterdam (1977)
24. Saffman, P.G.: On the boundary condition at the interface of
a porous medium. Stud. Appl. Math. 1, 93–101 (1971)
25. Salinger, A.G., Aris, R., Derby, J.J.: Finite element for-
mulations for large-scale, coupled flows in adjacent porous
and open fluid domains. Int. J. Numer. Methods Fluids 18,
1185–1209 (1994)
26. Sanchez-Palencia, E.: Non-homogeneous Media and Vibra-
tion Theory, no. 127 in Lecture Notes in Physics. Springer-
Verlag, New York (1980)
27. Tartar, L.: Incompressible fluid flow in a porous
medium–convergence of the homogenization process.
In: Non-homogeneous Media and Vibration Theory, E.
Sanchez-Palencia, Lecture Notes in Physics 127, pp. 368–377.
Springer-Verlag, Berlin (1980)
28. Temam, R.: Navier–Stokes Equations, Theory and Numerical
Analysis, 2nd ed. North-Holland, Amsterdam (1979)
29. Whitaker, S.: Flow in porous media I: A theoretical derivation
of Darcy’s law. Transp. Porous Media 1, 3–25 (1986)