J. Math. Anal. Appl.

368 (2010) 658676

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and

Applications
www.elsevier.com/locate/jmaa

Asymptotic analysis of the differences between the StokesDarcy system

with different interface conditions and the StokesBrinkman system
Nan Chen a, , Max Gunzburger b , Xiaoming Wang c
a
School of Mathematical Sciences, Fudan University, Shanghai, China, 200433
b
Department of Scientic Computing, Florida State University, Tallahassee, FL 32306-4120, United States
c
Department of Mathematics, Florida State University, Tallahassee, FL 32306-4120, United States

a r t i c l e i n f o a b s t r a c t

Article history: We consider the coupling of the Stokes and Darcy systems with different choices for
Received 31 August 2009 the interface conditions. We show that, comparing results with those for the Stokes
Available online 18 February 2010 Brinkman equations, the solutions of StokesDarcy equations with the BeaversJoseph
Submitted by W. Layton
interface condition in the one-dimensional and quasi-two-dimensional (periodic) cases are
Keywords:
more accurate than are those obtained using the BeaversJosephSaffmanJones interface
StokesDarcy equations condition and that both of these are more accurate than solutions obtained using a zero
StokesBrinkman equations tangential velocity interface condition. The zero tangential velocity interface condition is in
BeaversJoseph condition turn more accurate than the free-slip interface boundary condition. We also prove that the
BeaversJosephSaffmanJones condition summation of the quasi-two-dimensional solutions converge so that the conclusions are
also valid for the two-dimensional case.
2010 Elsevier Inc. All rights reserved.

1. Introduction

Karst aquifers are among the most important type of groundwater systems. They are mostly made up of a porous
medium, referred to as the matrix, that contains a network of ssures and conduits that are the major underground high-
ways for water transport. The matrix holds water while in conduits, one has a free ow. Despite the fact that ssures and
conduits occupy less space compared to the matrix, they play an essential role in the transport of uid and contaminants
in karst aquifers. Neglecting or not properly accounting for the ow in conduits and ssures and especially the exchange of
uid between them and the matrix can lead to inaccuracies.
Considerable effort has been directed at modeling and simulating the interaction between the conned ow in the ma-
trices and the free ow in the conduit. The NavierStokes equations or their linearized counterpart, the Stokes equations, are
widely used to describe the free ow in the conduit whereas Darcys law is chosen to model the conned ow in the ma-
trix. For connecting the components of the coupled NavierStokesDarcy or StokesDarcy systems, two interface conditions
are well accepted: the continuity of the normal velocity across the interface which is a consequence of the conservation
of mass, and the balance of the stress force normal to the interface. Additional interface condition(s) is needed in order
to close the system; the BeaversJoseph interface condition [2] is regarded as perhaps providing the most faithful account-
ing of what happens at the matrix-conduit interface; there is abundant empirical evidence to support this claim. In the
BeaversJoseph interface condition, the tangential component of the stress force of the ow in the conduit at the interface
is proportional to the jump in the tangential velocity across the interface. Unfortunately, from a mathematical point of view,

Research supported in part by the U.S. National Science Foundation under grant number CMG DMS-0620035, the National Science Foundation of China
under the grant number NSFC-10871050, and the Chinese Scholarship Council.
*
Corresponding author.
E-mail addresses: chennan@fudan.edu.cn (N. Chen), gunzburg@fsu.edu (M. Gunzburger), wxm@math.fsu.edu (X. Wang).

0022-247X/$ see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmaa.2010.02.022
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 659

Fig. 1. The conduit (free ow) and matrix (porous media) domains f and p , respectively.

the BeaversJoseph interface condition poses some diculties because this condition makes an indenite contribution to
the total energy budget. Consequently, many simplied versions of this interface condition have emerged, among which the
BeaversJosephSaffmanJones interface condition [13,14,19] is widely used; in this condition, the contribution of the tan-
gential velocity in the porous media is neglected. As a result, the total energy budget is dissipative and hence analyses are
substantially facilitated.1 Despite the convenience for mathematical analysis, models using the BeaversJosephSaffman
Jones interface condition can lead to an inaccurate accounting of the exchange of uid between the matrix and conduit.
A third choice for the remaining interface conditions is provided in [7] (see also [8,12]); there, the tangential velocity of the
uid in the conduit is set to be zero at the interface. A fourth candidate is discussed in [6] where the free-slip condition
at the interface is proposed for the uid ow in the conduit. These are even greater simplications of the BeaversJoseph
interface condition and also further simplify mathematical and numerical analyses.
We need a reference solution to use to examine the differences resulting from the four choices of interface conditions
within the StokesDarcy model. For this purpose, we replace the Darcy system with the Brinkman system as the model
for the ow in the matrix. The Brinkman system is the extension of Darcys law when boundary layer regions cannot be
neglected. In fact, Neale and Nader showed in [18] the equivalency between DarcyStokes and BrinkmanStokes velocities
at the interface when the Brinkman viscosity is related to the BeaversJosephSaffman parameter for the simple case of
one-dimensional shear ow. Note that it is well known that, as the Darcy number goes to zero, the differential equations
of the Brinkman model reduce to those of the Darcy model [16]. Thus, the central question we address is the connection
between the interface conditions of the Brinkman model with those corresponding to the four choices for the StokesDarcy
model.
In this paper, we identify a non-dimensional parameter  which is given by the square root of the ratio of the perme-
ability to the porosity divided by a typical length scale in the porous media. We then perform an asymptotic analyses with
respect to  of the StokesDarcy model with four choices for the interface conditions. We use the StokesBrinkman model
as the reference model to effect comparisons.
We should mention that the BeaversJosephSaffmanJones interface condition has been rigorously validated in the sense
that, under appropriate assumptions, the solution of the StokesDarcy system with that interface condition is asymptotically
the leading order of the solution of the Stokes equations in both the conduit and pore regions at small Darcy number;
see [10,11]. Those results are complementary to our results because our work indicates that the BeaversJoseph interface
condition provides better approximations to the BrinkmanStokes model at small Darcy number than does the Beavers
JosephSaffmanJones model, but the correction is of lower order. Note, however, that the correction could be large in
absolute value for not too small values of the Darcy number, a case that may be of interest in some applications such as
metallic foams.
The paper is organized as follows. In Section 2, we provide the StokesBrinkman equations and also the StokesDarcy
equations along with the four choices for the interface conditions. Section 3 is devoted to the one-dimensional case in which
the tangential velocities only depend on horizontal variable and the normal velocities are identically zero. In Section 3, we
also consider the fourth choice of interface condition, i.e., free-slip interface condition. In Section 4, we discuss the quasi-
two-dimensional case in which the velocities depend on both the horizontal and vertical variables but are of special form.
This is followed by a convergence theorem in Section 5 that gives us full two-dimensional solutions. In Section 6, we
examine a separate issue by showing that the advective term is small for both the Brinkman or Darcy equations so that the
linearized models can be regarded as valid approximations. Finally, in Section 7, we provide some concluding remarks.

2. The StokesBrinkman and StokesDarcy models

We start with a full description of the two models we consider. The two-dimensional conduit domain f = [0, L ]
[h f , 0] is occupied by a free ow; the two-dimensional matrix p = [0, L ] [0, h p ] is occupied by a porous media,
where L is order h p . In this paper, we take L = h p . We consider functions that are periodic in the horizontal variables with
period h p . Because the conduit occupies a much smaller space relative to the matrix, h p  h f > 0; see Fig. 1. We also denote

1
Recently, in [3,4], the mathematical diculties have been overcome and analyses and numerical analyses of the StokesDarcy model with the Beavers
Joseph interface condition have been provided.
660 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

by u f , p f , f f and u p , p p , f p the uid velocity, kinematic pressure, and external body force in f and p , respectively;
denotes the kinematic viscosity, n the porosity, the permeability, and D(u ) = 12 ( u + ( u )T ) the deformation tensor.
The relationship between porosity and permeability is given by = 0 n3 /(1 n)2 , where 0 is the typical permeability;
see, e.g., [17]. We assume that n is a constant and that the ows in both the conduit and matrix are incompressible. The
stress tensor is denoted by T( u , p ) = p I + 2 D(
u ), where I denotes the identity tensor.
The original Brinkman equation is given by (see [5])
n
2
D(u p ) +  p + n p p = f p ,
u (1)

where  denotes the effective viscosity which can be different from . According to [18], the effective viscosity and the
viscosity are related through



2 = . (2)

Dividing both sides of Eq. (1) by 2 yields
n n 1 
2 D(
up) + p +
u pp = f p. (3)
2 2 2
Here we assume that the effective viscosity and the viscosity are the same, i.e., 2 = 1, as is usually done in practice
and analysis. The steady-state StokesBrinkman model for coupled conduit-matrix ows then takes the form

u f ) + p f = f f ,
2 D( u f = 0, in f ,
n (4)
2 D(
up) +  p + n p p = f p ,
u u p = 0, in p .

f
The original Brinkman model can be recovered easily by replacing n with n2 , and replacing f p by p2 .
At the interface between the conduit and matrix domains, two sets of interface conditions are widely used. One is the
standard continuity of the velocity and the stress force across the interface, i.e.,
   
 f = u p ,
u p f I + 2 D(  = np p I + 2 D(
uf) n .
up) n (5)

The other is continuity of the velocity, all velocity derivatives,2 and the pressure across the interface proposed by Le Bars
and Worster [15], i.e.,

 f = u p ,
u u f = u p , p f = pp. (6)

These two types of interface boundary conditions reduce to the same ones in the one-dimensional case whereas we use the
latter one when dealing with two-dimensional systems.
We introduce the non-dimensional variables utilizing typical reference quantities in the matrix:

x p u f
x = , p = , u = , f  = U . (7)
hp h p U U
0 0

With these notations in hand, the non-dimensional form of the Brinkman equation is

U nU n h p U U 
 u p +  p +
u  p p = f p,
h2p h p 0 0
which is, after dropping the primes,

0 n0
up +
  p + n p p = f p .
u (8)
h2p

We introduce a non-dimensional parameter, the Darcy number Da = 0 /h2p . When the Darcy number goes to zero, the
Brinkman equation (8) reduces to the Darcy equation

1
p =
u p p f p .
0 n

2
When considering the effective viscosity, this should be replaced by u f = 
u p .
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 661

Similarly, the non-dimensional form of Stokes equation is

0
u f + p p = f f .
 (9)
h2p
Although (9) also contain the Darcy number, the order of the term  u f is not O (1) since the non-dimensionalization is
based on reference quantities in the porous media. Therefore, this term cannot be dropped.
Collecting the above results and returning to the dimensional form of the equations, we have the StokesDarcy system

u f ) + p f = f f ,
2 D( uf = 0, in f ,

(10)
u p = p p 1 f p ,
up = 0, in p
n
for the coupled conduit-matrix ows. The StokesDarcy equations are supplemented by periodic boundary condition in the
horizontal direction and no-penetration and free-slip boundary condition at the top and bottom for simplicity3
u f 1
= u f 2 = 0, on y = h f , (11)
y
u p1
= u p2 = 0, on y = h p . (12)
y
The system is also augmented by the interface conditions

 f n p f = u p n p f ,
u (13a)
 

n p f T( n p f = g (h p y ),
u f , p f ) (13b)
 
p f T( n p f = p f (
u f , p f )  p ),
uf u (13c)

where h p = y + ( p p /( g )) denotes the hydraulic head, a constant, n p f a unit vector normal to the interface, and p f a
unit vector tangent to the interface.4
The interface conditions (13) for the StokesDarcy model are known as the BeaversJoseph conditions [2]. The rst two
interface conditions in (13) are quite natural; (13a) guarantees the conservation of mass and (13b) the continuity of the
normal stress5 across the interface . On the other hand, (13c) is a not a statement of continuity of the tangential stress or
the velocity derivative components across the interface .6 Near the interface , a boundary layer may form in the matrix;
this boundary layer is not resolved by the Darcy equations. Thus, (13c) models the jump in the tangential stress across that
boundary layer. In particular, it says that the tangential stress of the conduit ow at the interface is proportional to the
jump in the tangential velocities across the boundary layer, in the limit that the boundary layer thickness vanishes; see [2]
for details. The value of the parameter depends on the properties of the porous material as well as the geometrical setting
of the coupled problem; it also can be used as a model tuning parameter.
A widely accepted simplication of the BeaversJoseph conditions is the BeaversJosephSaffmanJones conditions [13,19]
 p on the right-hand side of (13c) is neglected so that equation is replaced by
in which the term p f u
 
p f T( f.
np f = p f u
u f , p f ) (10c )

A further simplication [7] of the BeaversJoseph conditions is to ignore the left-hand side, i.e., the tangential stress
force, in (10c ) so that, as a result, the tangential velocity of the uid in the conduit is set to zero, i.e., we have
p f u f = 0. (10c )
In the sequel, for simplicity, we refer to (10c ) as the zero tangential velocity interface condition, even though it only sets the
tangential velocity in the conduit to zero.
Yet another simplication [6] of the BeaversJoseph conditions is to ignore the right-hand side, i.e., setting = 0 in
(10c ) so that, as a result, velocity of the uid in the conduit satises the free-slip condition at the interface, i.e., we have
 
p f T(u f , p f )n p f = 0. (10c )
For simplicity, we refer to (10c ) as the free-slip interface condition in the sequel.

3
The free-slip condition at the bottom may be replaced by the no-slip condition. This is necessary for well posedness if we adopt the simplied free-
slip (10c ) at the interface between the conduit and matrix.
4
In the set up of Fig. 1, we can choose n p f and p f to be the unit vectors in the y and x directions, respectively.
5
The stress force (or force due to stress) acting on a surface in the ow having the unit normal vector n  is given by T(
u , p )
n so that the normal and
tangential stresses on that surface are given by n  (T( n) and  (T(
u , p ) u , p )
n), respectively.
6
Thus, (13c) is not obtainable through a direct reduction of (5) or (6).
662 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

In the karst aquifer setting, the non-dimensional parameter  = /n/h p is usually small. We investigate the asymptotic
behavior of the velocities with respect to  and then compare the solutions of the StokesBrinkman model (4)(6) with
those of the three StokesDarcy models, i.e., the model (10)(13) with the BeaversJoseph condition, the model (10), (13a),
(13b), and (10c ) with the BeaversJosephSaffmanJones condition, the model (10), (13a), (13b), and (10c ) with the zero
tangential velocity interface condition, and the model (10), (13a), (13b), and (10c ) with the free-slip interface condition.7

3. One-dimensional ows

In the one-dimensional case, we assume that the normal velocities are identically zero and that the tangential velocities
only depend on y so that we have the ansatz for the velocities, pressures, and body forces given by
      
 f = u f ( y ), 0 ,
u p f 0, f f = f f ( y ), 0 ,
    (14)
 p = u p ( y ), 0 ,
u p p 0, f p = f p ( y ), 0 .

3.1. Asymptotic solutions of the StokesBrinkman system

We rst focus on the asymptotic behavior of solutions of the StokesBrinkman system (4)(6).

Lemma 3.1. The exact solution of StokesBrinkman system (4)(6) in the one-dimensional case is given by

y t 0   n
1 y 2 A C 2 cosh( h p )
uf = f f (s) ds dt f f (s) ds +  ,
n sinh( n
hp)
0 0 h f
 
   y
 n n
n n
A cosh( y ) C 2 cosh ( ( y h p )) 1 
up =  + f p (s) e ( y s) e ( y s) ds, (15)
n sinh( n
hp) 2 n
0

where

h p 0
1  n n  1

A= f p (s) e (h p s) + e (h p s) ds, C2 = f f (s) ds.
2
0 h f

Proof. With the ansatz (14) in hand, the coupled StokesBrinkman equations (4) reduce to
n
u f = f f , y (h f , 0) and u p + up = f p, y (0, h p ),

the general solutions of which are given by

y t

1

uf = f f (s) ds dt + C 1 + C 2 y ,


0 0
 y (16)

 n n  n n

1

u p = 2 n f p (s) e ( y s) e ( y s) ds + C 3 e y + C 4 e y ,

0

respectively. The interface conditions at y = 0 reduce to u f (0) = u p (0) and u f (0) = u p (0) so that

n
C1 = C3 + C4 and C2 = (C 3 C 4 ). (17)

The free-slip boundary conditions at y = h p and y = h f reduce to u f (h f ) = 0 and u p (h p ) = 0, respectively, so that,
together with (16), we have

7
In this last case, the free-slip boundary condition at the bottom boundary y = h f should be replaced by the no-slip condition in order to ensure well
posedness.
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 663

0

1

f f (s) ds + C 2 = 0,



h f

h p 


1  n (h s) n

 n n n 

f p (s) e p
+e (h p s )
ds + C 3 e h p C 4 e h p = 0.
2
0

Together with (17), we then have

 n n 0
2
A C 2 (e hp + e hp ) 1
C1 = n n , C2 = f f (s) ds,
n e hp e hp
h f
 n  n
A C 2 e h p 
A C 2e hp
C3 = , C4 = .
n e n h p e n h p n e n h p e n h p

Inserting these into the general solutions (16) completes the proof. 2

Lemma 3.2. Let  = /n/h p . The asymptotic solution of the StokesBrinkman system (4)(6) in the one-dimensional case is given
by

 y t 0 0 
1 2 2 3 3  4 4 
uf f f (s) ds dt y f f (s) ds +  h p f f (s) ds +  h p f p (0) +  h p f p (0) +  h p f p (0) ,

0 0 h f h f
 0 
1 1 y 1  1 1 (1 hy )
up  3 h3p f p (0) +  h p f f (s) ds e hp
+  2 h2p f p ( y ) +  4 h4p f p ( y )  3 h3p f  (h p )e p . (18)

h f

1
Proof. Note that the rst two terms in the equation for u f in (15) do not depend on  . In light of the fact that   1, by
2 1
dividing both the numerator and denominator of the third term by e  and dropping the exponentially small term e  , we
obtain
 n n
2
A C 2 (e hp + e hp )  
 h p 2
1
n n Ae  C 2 .
n e hp e hp

Expanding 
1
Ae  with respect to  yields

h p
1  1 (2 hs ) 1 s 

1
Ae  = f p (s) e p +e hp
ds
2
0
 h p h p 
1 1 s
2
1 s
= f p (s)e hp
ds + e fp (s)e  h p ds
2
0 0
1  1 
 h p f p (0) +  2 h2p f p (0) +  3 h3p f p (0) 2 2 h2p f p (h p )e  . (19)
2
Thus, we obtain the rst relation in (18).
In the same spirit, the rst term of u p in (15) can be reduced to
 
  n n
A cosh( y ) C 2 cosh( ( y h p )) 1  1 y 1 y 
1 y
 h p
Ae  e  h p + e  h p C 2 e  h p
n sinh( n
hp)

by dropping the exponentially small factors. With (19) in hand, it is easy to show that
664 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

y
 1 y 1 y   h p 1 hyp 1 s
h p
Ae 1
e hp
+ e  hp e f p (s)e hp
2
0
 h p hp  y
h p 1 s
2
1 s  1 y 1 y   h p 1 hyp 1 s
= f p (s)e hp
ds + e f p (s)e  hp
ds e hp
+ e  hp e f p (s)e hp
2 2
0 0 0

h p h p h p
 h p 1 hyp 1 s  h p 1 hyp 1 s  h p 2 1 hyp 1 s
e f p (s)e hp
ds + e f p (s)e hp
ds + e e f p (s)e hp
ds
2 2 2
y 0 0

 2 h2p 1 (1 hy )    2 h2p  
e p f p (h p ) +  h p f p (h p ) + f p ( y ) +  h p f p ( y ) +  2 h2p f p ( y )
2 2
 2 h2p 1 hy    2 2
h 1
(1 hy )  
f p (0) +  h p f p (0) + f p (h p )  h p f p (h p )
p
+ e p e  p
2 2
 3 h3p 1 (1 hy )   2 h2p    2 h2p 1 hy  
e p f (h ) +
p p f p ( y ) +  h p f p ( y ) +  2 h2p f p ( y ) + e p f p (0) +  h p f p (0) . (20)
2 2
For the second term of u p in (15), we repeat the same procedure as for (19) to obtain

y
 h p 1 hyp 1 s 1    1 y
e f p (s)e  hp
ds  2 h2p f p (0)  3 h3p f p (0) e  h p
2 2
0
 
+  2 h2p f p ( y )  3 h3p f p ( y ) +  4 h4p f p ( y ) . (21)

Combining with (20) yields

 0 
1 3 3  1 y 1  1 1 (1 hyp )
up  h p f p (0) + h p f f (s) ds e hp
+  2 h2p f p ( y ) +  4 h4p f p ( y )  3 h3p f  (h p )e

h f

which leads to the second relation in (18). 2

3.2. Asymptotic solutions of the StokesDarcy system

Next, we focus on deriving the asymptotic solution of one-dimensional StokesDarcy system (10), (13a), and (13b) with
the BeaversJoseph interface condition (13c).

Lemma 3.3. Let  = /n/h p . The asymptotic solution of the one-dimensional StokesDarcy equations with the BeaversJoseph
interface condition (13c) is given by


2 2

u 0p ,BJ = h f p ( y ),

p
y t 0 0 (22)
1 y n 2 2
u f ,BJ =

0
f f (s) ds dt f f (s) ds + hp f f (s) ds + h f p (0).

p
0 0 h f h f

Proof. In the one-dimensional case, the StokesDarcy equations (10) reduce to

u f = f f , y (h f , 0) and up = f p, y (0, h p ), (23)
n
the general solutions of which are given by

y t
1
u 0f ,BJ = f f (s) ds dt + C 1 + C 2 y and u 0p ,BJ = f p ( y ). (24)
n
0 0
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 665

Clearly, the rst two interface boundary conditions (13a) and (13b) are satised automatically, while the BeaversJoseph
condition (13c) reduces to
 
u 0f ,BJ   
 = u 0f ,BJ u 0p ,BJ  .
y  y =0 y =0

To determine the coecients, we impose this condition and the free-slip boundary condition at y = h f on the general
solution (23) to obtain
0
A A
C1 = + f p (0) and C2 = , A= f f (s) ds.
n
h f

Accordingly, wearrive at the asymptotic solution (22) of StokesDarcy equations with BeaversJoseph interface conditions
by setting  = /n/h p . 2

Using the same arguments, one obtains the asymptotic solution of the StokesDarcy system (10), (13a), and (13b) with
the BeaversJosephSaffmanJones interface condition (10c ) and the zero tangential velocity interface condition (10c ).

Lemma 3.4. Let  = /n/h p . The asymptotic solution of the one-dimensional StokesDarcy system with the BeaversJoseph
SaffmanJones interface condition (10c ) is given by


2 2

u 0p ,BJSJ = h f p ( y ),

p
y t 0 0 (25)
1 y n
u f ,BJSJ =

0
f f (s) ds dt f f (s) ds + hp f f (s) ds.


0 0 h f h f

Lemma 3.5. Let  = /n/h p . The asymptotic solution of the one-dimensional StokesDarcy system with zero tangential velocity
interface condition (10c ) is given by


2 2
0
u p , Q = h p f p ( y ),


y t 0 (26)
1 y

u 0
= f ( s ) ds dt f f (s) ds.

f ,Q
f

0 0 h f

Remark. In the one-dimensional case, one observes the following from (18), (22), (25), and (26).

1. The optimal choice8 of is = n.
2. Solutions with both BeaversJoseph and BeaversJosephSaffmanJones interface conditions have low sensitivity on
for near this optimal value or larger. This can be seen via a direct differentiation together with the KozenyCarman
formula relating permeability and porosity [1]. This low sensitivity is observed in numerical simulations [9].
3. The leading-order term for the velocities for the StokesBrinkman and StokesDarcy systems are both of O (1/ ) in the
conduit and are both of O ( 2 / ) in the matrix.
4. The velocity for the StokesBrinkman system contains boundary layers in the matrix near both y = 0 and y = h p , the
order being O ( / ) and O ( 3 / ), respectively.

We have also deduced the asymptotic behavior of solutions of the StokesDarcy system with the free-slip interface
condition (10c ); however, for well posedness, in this case we have to replace the free-slip boundary condition at the
bottom boundary y = h f by the physical no-slip boundary condition. In the case of the no-slip boundary condition at
y = h f , we have also deduced the asymptotic behavior of solutions of the StokesDarcy system with the three other
interface conditions, i.e., with (13c) or (10c ) or (10c ), and also for the StokesBrinkman system; in all cases, the leading
order behavior does not change from that for the free-slip boundary condition at y = h f , so we do not report on these
results here. The calculations for the free-slip interface condition (10c ) are very much the same as those for the other
interface conditions, so that we also do not report on them here. We merely include the implications resulting from the use
of the interface condition (10c ) in the comparisons made in Section 3.3.

8
When considering the effective viscosity in the Brinkman system, the corresponding optimal choice of is = n = n
/ .
666 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

3.3. Comparison of solutions of the StokesBrinkman and StokesDarcy systems

With (18), (22), (25), and (26) in hand, we can compare solutions of the StokesDarcysystem with different interface
conditions with solutions of the StokesBrinkman system under the optimal choice of = n.

Proposition 3.6. For the one-dimensional case, the difference between the velocity in the conduit for the StokesBrinkman sys-
tem and the StokesDarcy system with the BeaversJoseph interface condition is of O ( 3 / ). That difference is of O ( 2 / ) for the
BeaversJosephSaffmanJones interface condition and the difference is of O ( / ) for the zero tangential velocity interface condi-
tion.

Proposition 3.7. For the one-dimensional case, the difference between the velocity in the matrix for the StokesBrinkman system
and the StokesDarcy equations with either the BeaversJoseph, the BeaversJosephSaffmanJones, or the zero tangential velocity
interface conditions are all of O ( 4 / ).

From Proposition 3.6, we see that, when comparing with solutions of the StokesBrinkman system, the solution of the
StokesDarcy system with the BeaversJoseph interface condition ts better than does the solution obtained using the
BeaversJosephSaffmanJones interface condition and both are better ts that are solutions obtained using the tangential
velocity interface condition. Note that the special case of f f and f p being constants is essentially the same as that studied
in [19].
The free-slip interface condition (10c ) formally corresponds to the case = 0 in the BeaversJoseph interface con-
dition (13c). Comparisons of asymptotic solutions with those for the StokesBrinkman system are given in the following
proposition.

Proposition 3.8. For the one-dimensional case, the difference between the velocity in the conduit for the StokesBrinkman system and
the StokesDarcy system with the free-slip interface condition is of O (1/ ). The difference between the velocities in the matrix is of
O ( 4 / ) as it is for the other interface conditions.

From Propositions 3.6 and 3.8, we see that, when comparing with solutions of the StokesBrinkman system, the solution
of the StokesDarcy system with the BeaversJoseph interface condition ts better than does the solution obtained using the
BeaversJosephSaffmanJones interface condition and both are better ts than are solutions obtained using the tangential
velocity interface condition and all three are better ts than are solution obtained using the zero-slip interface condition.9
Note that the special case of f f and f p being constants is essentially the same as that studied in [19].

4. Quasi-two-dimensional ows

We now consider solutions and body forces that depend on both x and y, assuming periodicity in the horizontal direc-
tion, i.e., we invoke the ansatz


K 
K
  i2 kx
f =
u  f ,k =
u u f ,1,k ( y ), u f ,2,k ( y ) e hp
,
k= K k= K


K 
K
  i2 kx
f f = f f ,k = f f ,1,k ( y ), f f ,2,k ( y ) e hp
,
k= K k= K


K 
K
  i2 kx
p =
u  p ,k =
u u p ,1,k ( y ), u p ,2,k ( y ) e hp
,
k= K k= K


K 
K
  i2 kx
f p = f p ,k = f p ,1,k ( y ), f p ,2,k ( y ) e hp
,
k= K k= K

where the integer k denotes the wave number. Here, we also make the assumption that the Fourier decomposition only
contains a nite number of modes. Because solutions and data are real functions, we only need to consider k  0. To
simplify notation, in the sequel we set 
k = 2h k .
p

9
Due to its poor performance as an approximation to the BeaversJoseph interface condition and to save space, we do not consider the zero-slip interface
condition any further.
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 667

4.1. Solutions of the StokesBrinkman system

We start by deriving the general solution of the StokesBrinkman system (4) for each xed k.

Lemma 4.1. For the StokesBrinkman system (4) and for each xed k, the normal velocity in the conduit takes the form
       
u f ,2,k e ikx = C 1 ek y + C 2 e k y + C 3 yek y + C 4 ye k y + u f s,k e ikx , (27)
where the coecients C i , i = 1, 2, 3, 4, are to be determined and the particular solution u f s,k is given by
 y y
 1 + 
1 
ky ks 1 + ks
 
ky  ks
u f s,k = e e F (s) ds + e eks F (s) ds
4 
k3 
k3
0 0
y y 

ky ks 1
 
ky ks 1

ye e F (s) ds ye e F (s) ds , (28)

k2 
k2
0 0

where

i 

F (s) = k 2
f f ,2,k (s) + f f ,1,k (s) . (29)

k


 f ,k is solenoidal, there exists a stream function such that u f ,k = (
Proof. Because u
1
y , x ). Let = i u ( y )e ikx .
k f ,2,k
Then,
  1 ikx
 f ,k = u f ,2,k ( y ), i
u ku f ,2,k ( y ) e . (30)
i
k
The pressure in the Stokes equation for the conduit can be eliminated by taking the curl of that equation, resulting in
  
2 = f f ,k = i
k f f ,2,k ( y ) f f ,1,k ( y ) e ikx

which, together with (30), leads to the ordinary differential equation

 (4)  
k 2 u f ,2,k + 
u f ,2,k 2 k4 u f ,2,k e ikx = f f ,k . (31)

ik
Let

 i 
k f f ,k e ikx = 
F := i k 2 f f ,2,k ( y ) + i
k f f ,1,k ( y ) = 
k2 f f ,2,k ( y ) + f f ,1,k ( y ) .

k
Then, (31) becomes
1
k 2 u f ,2,k + 
u f ,2,k 2
(4 )
k4 u f ,2,k = F (32)

for which we have the solution (27)(28). 2


Using the same argument, we select = 1 u p ,2,k ( y )e ikx such that
ik

  1 ikx
p =
u , = u p ,2,k ( y ), i
ku p ,2,k ( y ) e (33)
y x i
k
and dene

i 
G ( y ) := 
k2 f p ,2,k ( y ) + f p ,1,k ( y ) . (34)

k
Then, we have the following result.

Lemma 4.2. For the StokesBrinkman system (4) and for each xed 
k, the normal velocity in the matrix takes the form
 
i
   
k2 + n
 n
k2 +  
u p ,2,k e kx
= C 5e ky
+ C6e ky
+ C7e y + C 8e
y
+ u ps,k e ikx , (35)
where the coecients C i , i = 5, 6, 7, 8, are to be determined and the particular solution u ps,k is given by
668 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

 y y  y 
1  1   1  
k2 + n 1  n
k2 +
u ps,k = e ky
e ks
G (s) ds e ky ks
e G (s) ds e y  e
s
G (s) ds
2 
kn 
kn  n n
0
0
0
k2 +
 y  

k2 + n
y
1 
k2 + n
sG
+e  e (s) ds .

k2 + n n
0

It remains to determine the coecients C i , i = 1, 2, . . . , 8. We set, as in the one-dimensional case,  = /n/h p and let

E= 
k2 + (nh2p /)/h p .

Lemma 4.3. For the StokesBrinkman system (4)(6) and for each xed 
k, the coecients C i , i = 1, 2, . . . , 8, are the solution of the
linear system

1 1 0 0 1 1 1 1
k k 1 1 k 
k E E C1 f1
C2 f2

k2 
k2 2k 2k 
k2 
k2 E2 E2
C3 f3
kh f   
e ekh f h f e kh f h f ekh f 0 0 0 0 C4 f4
= ,
0 0 0 0 e

kh p 
e kh p e Eh p
e Eh p C5 f5

0 0 2 2 1 1
0 0 C6 f6
k 
k
2 kh 2 kh   C7 f7
k e f k e f (
k 2 h f + 2
k)e kh f (
k 2 h f 2
k)ekh f 0 0 0 0
  C8 f8
0 0 0 0 
k 2 ekh p 
k2 e kh p E 2 e Eh p E 2 e Eh p

where

f 1 = f 2 = f 3 = 0, f 4 = u f s,k (h f ), f 5 = u ps,k (h p ),

i 1
f6 = f f ,1,k (0) f p ,1,k (0) , f 7 = u f s,k (h f ), f 8 = u ps,k (h p ).

k n

Proof. Using (30) and (33), the rst two interface conditions in (6) for the StokesBrinkman system reduce to

u f ,2,k (0) = u p ,2,k (0), u f ,2,k (0) = u p ,2,k (0), u f ,2,k (0) = u p ,2,k (0)

which imply that

C1 + C2 = C5 + C6 + C7 + C8, (C1)

kC 1 
 kC 2 + C 3 + C 4 = 
kC 5 
kC 6 + EC 7 EC 8 , (C2)

k 2C1 +
 k 2 C 2 + 2 kC 4 = 
kC 3 2 k 2C5 +
k 2C6 + E 2C7 + E 2C8. (C3)

Next, the StokesBrinkman equations (4) imply that the pressure on the two sides of the interface is given by
  2  
p f ,k  i
kx u f ,1,k 1 2 u f ,1,k 2 u f ,2,k 

 = f f ,1 ,k (0 )e + 2 2
+ 2
+ 
x y =0 x 2 y x y y =0

   1 ikx
= f f ,1,k (0)e ikx + 2 k 2 C 3 2k 2C4 e ,
i
k
   
p p ,k  1 i
kx 2 2 u p ,1,k 1 2 u p ,1,k 2 u p ,2,k 

 = f p ,1 ,k (0 )e + 2
+ 2
+ u p ,1 ,k 
x y =0 n n x 2 y x y y =0
1  2   1 ikx  1 
= f p ,1,k (0)e ikx + k ( EC 7 EC 8 ) E 3 C 7 E 3 C 8 e + (kC 5 
kC 6 + EC 7 EC 8 ) e ikx .
n n 
ik i
k
Let p f ,k |x=0, y =0 = p f ,k (0, 0) and p p ,k |x=0, y =0 = p p ,k (0, 0). Integrating p f and p p from 0 to x yields

1  i    
p f ,k = f f ,1,k (0) e kx 1 + 2 (C 3 + C 4 ) e ikx 1 + p f ,k (0, 0) (36)
i
k
and
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 669

1  i  E (
k 2 E 2)   
p p ,k = f p ,1,k (0) e kx 1 (C 7 C 8 ) e ikx 1

nik n k 2
   
(kC 5 
kC 6 + EC 7 EC 8 ) e ikx 1 + p p ,k (0, 0)

k2

1  i  1 1   
= f p ,1,k (0) e kx 1 + C5 + C6 e ikx 1 + p p ,k (0, 0). (37)
ni
k 
k 
k
With (36) and (37) in hand, we set x = 0 in the third interface condition in (6) (which holds for all x) to obtain p p ,k (0, 0)
p f ,k (0, 0) = 0. Then,

1 1 i 1
2C 3 2C 4 C5 + C6 = f f ,1,k (0) f p ,1,k (0) . (C4)

k 
k 
k n
Besides the interface boundary conditions, we impose free-slip boundary conditions at h f and h p :

u f ,1,k (h f ) = 0, u p ,1,k (h p ) = 0, u f ,2,k (h f ) = 0, u p ,2,k (h p ) = 0

so that
      
 k 2 ekh f C 2 + 2
k 2 e kh f C 1 +  k 2 h f e kh f C 3 2
k  k +
k 2 h f ekh f C 4 = u f s,k (h f ), (C5)

  
k 2 ekh p C 5 + 
k2 e kh p C 6 + E 2 e Eh p + E 2 e Eh p = u ps,k (h p ), (C6)


kh f 
kh f 
kh f 
kh f
e C1 + e C2 h f e C3 + h f e C 3 = u f s,k (h f ), (C7)
 
e kh p
C5 + e kh p
C 6 + e Eh p C 7 + e Eh p C 8 = u ps,k (h p ). (C8)
Consequently, combining (C1)(C8) completes the proof. 2

4.2. Solutions of the StokesDarcy systems

We next obtain the solutions of the StokesDarcy system (10) in the quasi-two-dimensional case. In the same spirit as
for the StokesBrinkman system, we invoke the ansatz


K 
K
  
 0f =
u  0f ,k =
u u 0f ,1,k ( y ), u 0f ,2,k ( y ) e ikx ,
k= K k= K


K 
K
  
 0p =
u  0p ,k =
u u 0p ,1,k ( y ), u 0p ,2,k ( y ) e ikx
k= K k= K
 
for solutions of the StokesDarcy system. By selecting the streamfunctions = 1 u 0f ,2,k ( y )e ikx and = 1 u 0p ,2,k ( y )e ikx , the
ik ik
velocities can be written as
  1 ikx   1 ikx
 0f ,k = u 0f , 2,k ( y ), i
u ku 0f ,2,k ( y ) e ,  0p ,k = u 0p ,2,k ( y ), i
u ku 0p ,2,k ( y ) e .
i
k i
k
Using the same argument as for the StokesBrinkman system, we obtain the following result.

Lemma 4.4. For the StokesDarcy system (10) and for each xed 
k, the normal velocity in the conduit and matrix are given by
       
u 0f ,2,k e ikx = C 10 ek y + C 20 e k y + C 30 yek y + C 40 ye k y + u 0f s,k e ikx (38)

and
     
u 0p ,2,k e ikx = C 50 ek y + C 60 e k y + u 0ps,k e ikx , (39)

respectively, where the coecients C i0 , i = 1, 2, . . . , 6, are to be determined and the particular solutions u 0f s,k and u 0ps,k are given by
 y y
 1 + 
1  ks 1 + ks
   ks
u 0f s,k = eky
e F (s) ds + e ky
eks F (s) ds
4 
k3 
k3
0 0
y y 

ky ks 1
 
ky ks 1

ye e F (s) ds ye e F (s) ds

k2 
k2
0 0
670 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

and

y y
1  1  1  1 
u 0ps,k = ek y e ks G (s) ds + e ky eks G (s) ds,
2 n 
k 2 n 
k
0 0

respectively, where F and G are dened in (29) and (34), respectively.

Lemma 4.5. For the StokesDarcy system (10), (13a), and (13b) and for each xed k, the coecients C i0 , i = 1, 2, . . . , 6, are the solution
of the linear system
0 0
1 1 0 0 1 1 C1 g1
2 k2 2
k2 0 0 1/ 1/ C 0 g0
2 2
2 kh f 2 kh f  
k e k e (2 2
k h f k )e kh f (2
k h f
k 2 )ekh f 0 0 C 0 g0
3 = 3
e kh f 
kh f
h f e 
kh f
h f e

kh f
0 C 40 g 40
e 0
 
0 0 0 0 ekh p
e kh p C 50 g 50
   
2
k 2 +  hk 2
k2  hk 2
k +  h1 2
k 2 +  h1  hk p k
h p C 60 g 60
p p p p

for the BeaversJoseph interface condition (13c),

0 0
1 1 0 0 1 1 C1 g1
2 k2 2
k2 0 0 1/ 1/ C 0 g0
2 2
2 kh f    
k e k 2 ekh f (2
k h f
k 2 )e kh f (2
k h f
k 2 )ekh f 0 0 C 0 g0
3 = 3
e kh f 
kh f 
h f e kh f h f e

kh f
0 C 40 g 40
e 0
 
0 0 0 0 ekh p e kh p C 50 g 50
 
2
k 2 +  hk 2
k 2  hk 2
k +  h1 2
k 2 +  h1 0 0 C 60 g 60
p p p p

for the BeaversJosephSaffmanJones interface condition, and (10c )

0 0
1 1 0 0 1 1 C1 g1
2k2 2k2 0 0
1/ 1/ C 2 g 20
0

2 kh f 2 kh f  
k e k e (2
k h f
k 2 )e kh f (2
k h f
k 2 )ekh f 0 0 C 30 g 30
=
kh f    0 0
e ekh f h f e kh f h f ekh f 0 0 C 4 g4
 
0 0 0 0 ekh p e kh p C 50 g 50

k k 1 1 0 0 C 60 g 60
for the zero tangential velocity interface condition (10c ), where

i 1
g 1 = g 6 = 0, g2 = f f ,1,k (0) f p ,1,k (0) ,
n
g3 = u 0f s,k (h f ), g4 = u 0f s,k (h f ), g 5 = u 0ps,k (h p ).

Proof. Condition (13a) results in u 0f ,2,k (0) = u 0p ,2,k (0) so that

C 10 + C 20 = C 50 + C 60 . (D1)

Normalized with = 1 and written in component form, (13b) takes the form

 u 0f ,1,k 0
1 u f ,1,k u 0f ,2,k 
p f ,k 0 ( + ) 0
(0, 1) + 2
x 2 y x
0
= p p ,k
0 p f ,k 1 u f ,1,k u 0f ,2,k u 0f ,2,k 1
2
( y + x ) y
which implies

u 0f ,2,k
p f ,k 2 = p p ,k . (40)
y
The Stokes equation in (10) implies
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 671

 2 u0  2 0 2 u 0f ,2,k
p f ,k f ,1,k 1 u f ,1,k
= f f ,1,k (0) + 2 + + .
x x2 2 y2 x y
Let p f ,k |x=0, y =0 = p f ,k (0, 0) and p p ,k |x=0, y =0 = p p ,k (0, 0). Integrating from 0 to x yields

1  i     
p f ,k = f f ,1,k (0) e kx 1 + 2 C 30 + C 40 e ikx 1 + p f ,k (0, 0). (41)

ik
On the other hand, the Darcy equation in (10) implies
p p ,k 1 1   0  0  1 ikx
= f p ,1,k (0) u 0p ,1,k = f p ,1,k e ikx + kC 5 kC 6 e .
x n n i
k
Integrating from 0 to x results in
1    1   
p p ,k = f p ,1,k (0) e ikx 1 C 50 C 60 e ikx 1 + p p ,k (0, 0). (42)
i
kn 
k
Furthermore, it is obvious that

u 0f ,2,k  0  
2 kC 1 
= 2  kC 20 + C 30 + C 40 e ikx . (43)
y
Inserting (41)(43) into (40) and setting x = 0 yields p p ,k (0, 0) p f ,k (0, 0) = 2 (
kC 10 
kC 20 + C 30 + C 40 ). Elementary calcula-
tions show that (40) can be reduced to

1 0 1 0 i 1
2
k 2 C 10 + 2
k 2 C 20 + C5 C6 = f f ,1,k (0) f p ,1,k (0) . (D2)
n
Written in component form, the BeaversJoseph interface condition (13c) takes the form

 u 0f ,1,k 0
1 u f ,1,k
0
u 
p f ,k 0 ( y + f x,2,k ) 0
(1, 0) + 2 0
x
u0
2
u 0f ,2,k

0 p f ,k 1 u f ,1,k 1
2
( y + f x,2,k ) y
 
u 0f ,1,k u 0p ,1,k
= (1, 0) ,
u 0f ,2,k u 0p ,2,k

i.e.,
 u0 u 0f ,2,k
f ,1,k  
+ = u 0f ,1,k u 0p ,1,k .
y x

We then have, after setting = n,
         
n n n n n n
2
k2 +  0 2
k C 1 + 2k  0 
k C 2 + 2k + 0 
C 3 + 2k + 0
C4  0
kC + 
kC 0 = 0. (D3)
5 6
Using the same argument for the BeaversJosephSaffmanJones interface boundary condition (10c ) leads to
       
n n n n
2
k2 +  0 2
k C 1 + 2k  0 
k C 2 + 2k + 0 
C 3 + 2k + C 40 = 0. (D3 )

Similarly, for the zero tangential velocity interface condition (10c ), we have

kC 0 
 kC 0 + C 0 + C 0 = 0. (D3 )
1 2 3 4

We also impose the free-slip boundary conditions u 0f , 1,k (h f ) = 0 and u 0f ,2,k (h f ) = 0 at h f so that
      
k 2 e kh f C 10 + 
 k 
k 2 ekh f C 20 + 2 k +
k 2 h f e kh f C 30 2 k 2 h f ekh f C 40 = u 0f s,k (h f ), (D4)
   
e kh f C 10 + ekh f C 20 h f e kh f C 30 h f ekh f C 40 = u 0f s,k (h f ) (D5)

along with the no-ow condition u 0p ,2,k (h p ) = 0 across the boundary at h p so that
 
ekh p C 50 + e kh p C 60 = u 0ps,k (h p ). (D6)

Combining (D1)(D6), (D3 ) and (D3 ) completes the proof. 2

672 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

4.3. Comparison of asymptotic solutions of the StokesBrinkman and StokesDarcy systems

With the coecients solved using MATLAB, we have the following results.

Proposition 4.6. The asymptotic solution of the normal velocity for the quasi-two-dimensional StokesBrinkman system is given by

 1    
2  
u f ,2,k C k,1 e ikx + C k,2 e ikx
+ C k,3 e ikx + O  3 / ,

 
2
 
2
 
4    
2
 
4
u p ,2,k P k,1 e ikx + P k,2 e ikx + P k,4 e ikx + O  5 / + P k,5 e ikx e E y + P k,6 e ikx e E ( y h p ) , (44)

where the coecients are listed in Appendix A.

Proposition 4.7. The asymptotic solution of the normal velocity for the quasi-two-dimensional StokesDarcy system with the Beavers
Joseph interface condition is given by

 1    
2  
u 0f ,2,k,BJ C k,1 e ikx + C k,2 e ikx
+ C k,3 e ikx + O  3 / ,

  2
  3
 4  
u 0p ,2,k,BJ P k,1 e ikx + P k,2 e ikx + P k,3 e ikx + O  6 / , (45)

with the BeaversJosephSaffmanJones interface condition by

 1    
2  
u 0f ,2,k,BJSJ C k,1 e ikx + C k,2 e ikx
+ C k,4 e ikx + O  3 / ,

 
2
 
3
 
4
 
2  
u 0p ,2,k,BJSJ P k,1 e ikx + P k,2 e ikx + P k,3 e ikx + P k,7 e ikx + O  6 / , (46)

and with the zero tangential velocity interface condition by

 1  
u 0f ,2,k, Q C k,1 e ikx + O  3 / ,

kx 
i
2
 3  
4  
u 0p ,2,k, Q P k,1 e + P k,2 e ikx + P k,8 e ikx + O  5 / , (47)

where the coecients are listed in Appendix A.

Remark. In the quasi-two-dimensional case, one observes the following from (44)(47).

1. The leading-order terms for the normal velocities for the StokesBrinkman and the StokesDarcy systems are both of
O (1/ ) in the conduit and are both of O ( 2 / ) in the matrix.
2. The normal velocity for the StokesBrinkman system contains a boundary layer in the matrix near both y = 0 and
y = h p , the order being O ( 2 / ) and O ( 4 / ), respectively.
3. The tangential velocities and normal velocities are not independent. Indeed, they are associated with each other via
the streamfunctions. Elementary calculations show that the leading order of the tangential velocities in the conduit
and matrix are the same as the normal velocities. However, the order of the tangential velocity in the boundary layer
changes to O ( / ) and O ( 3 / ) at y = 0 and y = h p , respectively.
4. The normal velocities of StokesDarcy equations with BeaversJoseph interface condition at the interface are differ-
ent from that with BeaversJosephSaffmanJones interface condition. The difference is of order O (  2 / ), which is
approximately of order O ( 5 / ).

With (44)(47) in hand, we can compare solutions of the StokesDarcy system with different interface conditions with
solutions of the StokesBrinkman system.

Proposition 4.8. For the quasi-two-dimensional case, the difference between the normal velocity in the conduit for the Stokes
Brinkman system and the StokesDarcy system with the BeaversJoseph interface condition is of O ( 3 / ). That difference is of order
O ( 2 / ) for the BeaversJosephSaffmanJones interface condition and is of order O ( / ) for the zero tangential velocity interface
condition.

Proposition 4.9. In quasi-two-dimensional case, the difference between the normal velocity in the matrix for the StokesBrinkman
system and the StokesDarcy system with the BeaversJoseph, the BeaversJosephSaffmanJones, and the zero tangential velocity
interface conditions are all of order O ( 4 / ).
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 673

Thus, as for the one-dimensional case, we see from Proposition 4.8 that, when comparing with solutions of the Stokes
Brinkman system, the solutions of the StokesDarcy system with the BeaversJoseph interface condition ts better than
does the solution obtained using the BeaversJosephSaffmanJones and both of these t better than the solution obtained
using the zero tangential velocity interface condition.

5. The two-dimensional ows

In Section 4, we obtained, for each wave number k  0, the quasi-two-dimensional solutions u  f ,k = (u f ,1,k , u f ,2,k )
and u p ,k = (u p ,1,k , u p ,2,k ) of the StokesBrinkman system in the conduit and matrix, respectively. Likewise, we obtained
the corresponding solutions u  0f ,k,BJ = (u 0f ,1,k,BJ , u 0f ,2,k,BJ ) and u 0p ,k,BJ = (u 0p ,1,k,BJ , u 0p ,2,k,BJ ) of the StokesDarcy system with
 0f ,k,BJSJ = (u 0f ,1,k,BJSJ , u 0f ,2,k,BJSJ ) and u 0p ,k,BJSJ = (u 0p ,1,k,BJSJ , u 0p ,2,k,BJSJ ) of the Stokes
the BeaversJoseph interface condition, u
Darcy system with the BeaversJosephSaffmanJones interface condition, and u  0f ,k, Q = (u 0f ,1,k, Q , u 0f ,2,k, Q ) and u 0p ,k, Q =
(u 0p ,1,k, Q , u 0p ,2,k, Q ) of the StokesDarcy system with the zero tangential velocity interface condition. Summation of the quasi-
two-dimensional solutions lead to the two-dimensional solutions


K 
K
f =
u  f ,k ,
u p =
u  p ,k ,
u
k= K k= K


K 
K
 0f ,BJ =
u  0f ,k,BJ ,
u  0p ,BJ =
u  0p ,k,BJ ,
u
k= K k= K


K 
K
 0f ,BJSJ =
u  0f ,k,BJSJ ,
u  0p ,BJSJ =
u  0p ,k,BJSJ ,
u
k= K k= K


K 
K
 0f , Q =
u  0f ,k, Q ,
u  0p , Q =
u  0p ,k, Q .
u
k= K k= K

We have the result


K  3
   
u f ,k u 0  u f ,k u 0  O  ,
f ,k,BJ f ,k,BJ

k= K


K  2
   
u f ,k u 0  u f ,k u 0  O  ,
f ,k,BJSJ f ,k,BJSJ

k= K


K 
   
u f ,k u 0  u f ,k u 0  O  ,
f ,k, Q f ,k, Q

k= K

which leads to the following conclusion.

Theorem 5.1. For the two-dimensional case, the difference between the normal velocity in the conduit for the StokesBrinkman system
and the StokesDarcy system with the BeaversJoseph interface condition is of order O ( 3 / ). The difference is of order O ( 2 / ) for
the BeaversJosephSaffmanJones interface condition and is of order O ( / ) for the zero tangential velocity interface condition.
The difference between the normal velocity in the matrix for the StokesBrinkman system and StokesDarcy system with the
BeaversJoseph, the BeaversJosephSaffmanJones and the zero tangential velocity interface conditions are all of order O ( 4 / ).

Thus, again, the BeaversJoseph interface condition ts better than the BeaversJosephSaffmanJones interface condition
and both of these t better than the zero tangential velocity interface condition, when comparing solutions with that of the
StokesBrinkman system.

6. The convection term in the Brinkman and Darcy equations

The asymptotic analyses of the previous sections can be put to other uses. For example, it can be used to justify neglect-
ing the convection term in the matrix in the Brinkman and Darcy equations.
The steady-state StokesBrinkman and StokesDarcy equations with convection in the matrix are given by
674 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

u f + p f = f f ,
 div uf = 0, in f ,
n (48)
(
u p )
u p 
up +  p + n p p f p = 0,
u div up = 0, in p

and

u f + p f = f f ,
 div uf = 0, in f ,
n (49)
(
u p )
up +  p + n p p f p = 0,
u div up = 0, in p ,

respectively. We have shown, in the previous sections, that away from the boundary layer, u p O ( 2 / ), which is a small
 p that the derivatives of the velocities are of
quantity in a typical karst aquifer. It is obvious from the expression for u
 p 2  O (1), f p O (1), and
2
O ( 2 / ) as well so that the advective term is of O ( 4 / 2 ). On the other hand, n u
n p p O (1). In light of these results, we conclude that the advective term is smaller than the others terms and therefore
it is justied to neglect it in both the Brinkman and Darcy equation.

7. Conclusion and remarks


1
We have derived asymptotic solutions with respect to the non-dimensional parameter = n hp
for the time-
independent StokesDarcy system with the BeaversJoseph, BeaversJosephSaffmanJones, zero tangential velocity, and
2
free-slip interface conditions. The leading order of the velocity is of O ( 1 ) in the conduit whereas it is of O (  ) in

the matrix. It is observed that the optimal choice of the BeaversJoseph constant is n for both the Beavers
Joseph and BeaversJosephSaffmanJones interface conditions. We also notice that the solutions with the BeaversJoseph

and BeaversJosephSaffmanJones interface conditions show low sensitivity with respect to for [ n, ). Com-
pared with asymptotic solutions of StokesBrinkman system, which we also derived, the solution using the Beavers
Joseph interface condition ts better in the conduit compared to that for the BeaversJosephSaffmanJones inter-
face condition and both t better than that obtained using the zero tangential velocity condition; in the matrix, the
three choices of conditions yield the same asymptotic behavior. We have also investigated the use of the free-slip in-
terface condition and determined that this is the least accurate among all interface boundary conditions considered
here.
In this paper, we only considered the steady-state case and neglected the convection term in the Brinkman and Darcy
systems. It would be interesting to examine the time-dependent case. Also, investigating the results when the porosity and
permeability are no longer constants would be a challenging and meaningful work.

Appendix A. The normal velocities for the quasi-two-dimensional case

The velocities in the conduit are

2 
 1     3
u 0f ,2,k,BJ C k,1 e ikx + C k,2 e ikx + C k,3 e ikx + O ,

2 
 1     3
u 0f ,2,k,BJSJ C k,1 e ikx + C k,2 e ikx + C k,4 e ikx + O ,

2 
 1     3
u f ,2,k C k,1 e ikx + C k,2 e ikx + C k,3 e ikx + O ,


i 1 3
u 0f ,2,k, Q C k,1 e kx
+O ,

where

D2       
C k,1 = D 3 + h f 1 e 2kh f e k(h f y ) e k(h f + y )
  2 4
2k(1 4kh f e kh f
e kh f
)
   k(h y )      
2kh f 1 + e 2kh f ye f 1 e 2kh f 2 kh f e 2kh f ye k(h f + y )
D1      
+ (2 +  kh f ) + (2 
kh f )e 2kh f e k(h f + y ) e k(h f y )
 2 4
2(1 4kh f e kh f
e kh f
)
        
+ 3 + 2 kh f + e 2kh f  k ye k(h f y ) + 1 + (3 2 kh f )e 2kh f k ye k(h f + y ) ,
 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676 675

D 2h p        
C k,2 =  
h f 1 + e 2kh f (1 + y ) e k(h f y ) + e k(h f + y )
1 4
kh f e 2kh f e 4kh f
D 1
kh p        
+  
(2 + 
kh f ) (2 
kh f )e 2kh f + y
k 2 1 e 2kh f e k(h f + y ) e k(h f y ) ,
1 4
kh f e 2kh f e 4kh f
i f p ,1,k (0)h2p         
C k,3 =  
1 + e 2kh f ek y + 1 + 4
kh f + e 2kh f e k(2h f + y )
1 4
kh f e 2kh f e 4kh f
  
k f p ,2,k (0) + i f 

   
(0)
+ 2
k y ek y + e k(2h f + y ) + h2p 1 + (1 2
kh f )e 2kh f ek y
p ,1,k
 
k(1 4kh f e e 4kh f
) 2
kh f

   k(2h + y )    k y  
1 + 2
kh f + e 2kh f e f 
ky 1 + e 2kh f e + e k(2h f + y ) ,
i f p ,1,k (0)h2p         
C k,4 = 1 + (1 2
kh f )e 2kh f ek y + 1 + 2
kh f + e 2kh f e k(2h f + y )
1 4
kh f e 2
kh f
e 4
kh f

 
k f p ,2,k (0) + i f 
     (0)
+
k y 1 + e 2kh f ek y + e k(2h f + y ) + h2p
p ,1,k
  
k(1 4
kh f e 2kh f e 4kh f )
             
1 + (1 2 kh f + e 2kh f e k(2h f + y ) 
kh f )e 2kh f ek y 1 + 2 k y 1 + e 2kh f ek y + e k(2h f + y ) ,

h f
h f

1  
     
D1 = e kh f
e ks
(1 + 
ks) k f f ,2,k (s) + i f f ,1,k (s) ds + ekh f eks (1 + 
ks) 
k f f ,2,k (s) + i f f ,1,k (s) ds
4
k2
0 0
h f
h f
!

kh f 

ks  
   
ks  

+hfe e k k f f ,2,k (s) + i f f ,1,k (s) ds + h f ekh f e k k f f ,2,k (s) + i f f ,1,k (s) ds ,
0 0
h f
h f

1  
     
D2 = e kh f
e ks
(1 
ks) k f f ,2,k (s) + i f f ,1,k (s) ds + ekh f eks (1 + 
ks) 
k f f ,2,k (s) + i f f ,1,k (s) ds
4
0 0
h f
h f
!

kh f 

ks  
   
ks  

+hfe e k k f f ,2,k (s) + i f f ,1,k (s) ds + h f ekh f e k k f f ,2,k (s) + i f f ,1,k (s) ds ,
0 0

y y
1  
     
D3 = e ky
e ks
(1 + 
ks) k f f ,2,k (s) + i f f ,1,k (s) ds + e k y eks (1 + 
ks) 
k f f ,2,k (s) + i f f ,1,k (s) ds
4
k2
0 0

y y !
 

ks 
   
ks 

+hfe ky
e k k f f ,2,k (s) + i f f ,1,k (s) ds + h f e k y
 
e k k f f ,2,k (s) + i f f ,1,k (s) ds .
0 0

The velocities in the porous media are


2
  
3
 
4
6
u 0p ,2,k,BJ P k,1 e ikx + P k,2 e ikx + P k,3 e ikx + O ,

2 3 4 2  6
0 kx 
i kx 
i kx 
i kx 
i 
u p ,2,k,BJSJ P k,1 e + P k,2 e + P k,3 e + P k,7 e +O ,

2 3 4 
      5  
2
 
4
u p ,2,k P k,1 e ikx + P k,2 e ikx + P k,4 e ikx + O + P k,5 e ikx e E y + P k,6 e ikx e E ( y h p ) ,

2 3 4  6
      
u 0p ,2,k, Q P k,1 e ikx + P k,2 e ikx + P k,8 e ikx + O ,

where
676 N. Chen et al. / J. Math. Anal. Appl. 368 (2010) 658676

h2p   k( y h ) h2p   

P k,1 = 
k f p ,2,k (h p ) + i f  (h p e
) p
+ 
k f p ,2,k ( y ) + i f p ,1,k ( y ) + i f p ,1,k (0)e k y h2p ,
k
p ,1 ,k 
k
     
P k,2 = 2
ki f p ,1,k (0)e k y h3p i f f ,1,k (0)e k y h2p + 4 k3 h3p (2 +  kh f ) (2  kh f )e 2kh f D 1 e k( y +h f )
   
4k 2 h3p h f 1 + e 2kh f D 2 e k( y +h f ) ,
    
P k,3 = 2k 2 i f f ,1,k (0)e k y h3p + 4 k4 h4p (2 +  kh f ) (2  kh f )e 2kh f D 1 e k( y +h f )
   
4k3 h4p h f 1 + e 2kh f D 2 e k( y +h f ) ,
2 h f 4
8h f 
k3 i f p ,1,k (0)e 2k h     
2
ki f f ,1,k (0)e k y h3p + 4
k4 h4p 1 e 2kh f D 1 e k( y +h f )
p
P k,4 =  
1 4
kh f e 2kh f e 4kh f
   
4
k 2 h4p 1 + e 2kh f D 2 e k( y +h f ) ,
2
k 2 h2p    
P k,5 = (1 + 
kh f ) (1 
kh f )e 2kh f D 1 e kh f
 2 4
1 4kh f e kh f
e kh f

2h2p    
+ (1 
kh f ) (1 + 
kh f )e 2kh f D 2 e kh f ,
 2 4
1 4kh f e kh f
e kh f
 
 4 
P k,6 = kh p k f p ,2,k ( y ) + i f p ,1,k ( y ) ,

8
k4 h5p h f i f f ,1,k (0)e 2kh f
P k,7 =  
,
1 4
kh f e 2kh f e 4kh f

P k,8 = 2
k 2 i f f ,1,k (0)e k y h3p ,
where D 1 and D 2 are given as above.

References

[1] J. Bear, Dynamics of Fluids in Porous Media, Dover, 1988.

[2] G. Beavers, D. Joseph, Boundary conditions at a naturally permeable wall, J. Fluid Mech. 30 (1967) 197207.
[3] Y. Cao, M. Gunzburger, F. Hua, X. Wang, Coupled StokesDarcy model with BeaversJoseph interface boundary condition, Commun. Math. Sci. 8 (1)
(March 2010) 125.
[4] Y. Cao, M. Gunzburger, X. Hu, F. Hua, X. Wang, W. Zhao, Finite element approximations for StokesDarcy ow with BeaversJoseph interface conditions,
SIAM J. Numer. Anal. 47 (2010) 42394256.
[5] H. Brinkman, A calculation of the viscous force exerted by a owing uid on a dense swarm of particles, Appl. Sci. Res. A 1 (1947) 2734.
[6] M. Discacciati, E. Miglio, A. Quarteroni, Mathematical and numerical models for coupling surface and groundwater ows, Appl. Numer. Math. 1 (2002)
5774.
[7] M. Discacciati, A. Quarteroni, Analysis of a domain decomposition method for the coupling of the Stokes and Darcy equations, in: F. Brezzi, et al. (Eds.),
Numerical Mathematics and Advanced Applications. Proceedings of ENUMATH 2001, Springer, Milan, 2003, pp. 320.
[8] M. Discacciati, A. Quarteroni, Convergence analysis of a subdomain iterative method for the nite element approximation of the coupling of Stokes
and Darcy equations, Comput. Vis. Sci. 6 (2004) 93103.
[9] F. Hua, Modeling, analysis and simulation of StokesDarcy system with BeaversJoseph interface condition, PhD thesis, Florida State University, 2009.
[10] W. Jger, A. Mikelic, On the boundary condition at the contact interface between a porous medium and a free uid, Ann. Sc. Norm. Super. Pisa Cl.
Sci. (4) 23 (1996) 4031465.
[11] W. Jger, A. Mikelic, On the interface boundary condition of Beavers, Joseph, and Saffman, SIAM J. Appl. Math. 60 (2000) 11111127.
[12] B. Jiang, A parallel domain decomposition method for coupling of surface and groundwater ows, Comput. Methods Appl. Mech. Engrg. 198 (2009)
947957.
[13] I. Jones, Reynolds number ow past a porous spherical shell, Proc. Cambridge Philos. Soc. 73 (1973) 231238.
[14] W. Layton, F. Schieweck, I. Yotov, Coupling uid ow with porous media ow, SIAM J. Numer. Anal. 40 (2003) 21952218.
[15] G. Le Bars, M. Worster, Interfacial conditions between a pure uid and a porous medium implications for binary alloy solidication, J. Fluid Mech. 550
(2006) 149173.
[16] J.-L. Lions, Perturbations singulires dans les problmes aux limites et en contrle optimal, Springer-Verlag, Berlin, New York, 1973.
[17] W. McCabe, J. Smith, P. Harriot, Unit Operations of Chemical Engineering, seventh ed., McGrawHill, New York, 2005.
[18] G. Neale, W. Nader, Practical signicance of Brinkmans extension of Darcys law: coupled parallel ows within a channel and a bounding porous,
Canad. J. Chem. Eng. 52 (1974) 475478.
[19] P. Saffman, On the boundary condition at the interface of a porous medium, Stud. Appl. Math. 1 (1971) 7784.