HOMOGENIZATION OF INTERFACES BETWEEN RAPIDLY

OSCILLATING FINE ELASTIC STRUCTURES AND FLUIDS

K.-H. HOFFMANN

, N. D. BOTKIN

, AND V. N. STAROVOITOV

SIAM J. APPL. MATH. c 2005 Society for Industrial and Applied Mathematics
Vol. 65, No. 3, pp. 9831005
Abstract. The paper studies the interaction of a periodic solid bristle structure with a uid.
Such problems arise, for example, when modelling biotechnological devices operating in liquids or
when simulating epithelium surfaces of blood vessels. The uid is described by the linearized Navier
Stokes equation whereas the solid part is governed by equations of linear elasticity. The interface
conditions are accounted. A homogenized model of the structure is derived by employing the two-
scale convergence technique. The model describes a new material which possesses some interesting
properties.
Key words. homogenization, uid-solid interface, biosensor, multi-layered structure
AMS subject classications. 35B27, 74F10, 74Q10
DOI. 10.1137/S0036139903421572
1. Introduction. We study a mechanical system consisting of a uid and a
rapidly oscillating elastic ne structure interacting with the uid. The goal is to
obtain averaged equations which eectively describe the behavior of the system.
This investigation is motivated by modelling a surface acoustic wave sensor based
on the generation and detection of horizontally polarized shear waves (see [3]). Acous-
tic shear waves are excited through an alternate voltage applied to electrodes deposited
on a quartz crystal substrate. The waves are transmitted into a thin isotropic guiding
layer covered by a thin gold lm that contacts a liquid containing a protein to be
detected. The protein adheres to a specic receptor (aptamer) placed on the surface
of the gold lm. The arising mass loading causes a phase shift in the electric signal
to be measured by an electronic circuit.
One can impress the aptamer-protein layer as a periodic bristle or pin structure
on the top of the gold lm contacting with the liquid (see Figure 1). The thickness of
the aptamer-protein layer is about 4 nm, and the number of bristles per surface unit
is enormous large. Therefore, the direct numerical modelling of such a structure using
uid-solid interface conditions is impossible. Proper models can be derived using the
homogenization technique from [12], [11], [1], [7], [8], and [5] along with the strict
treatment of the solid-uid interface (see, e.g., [6]).
Problems that are close to ours were studied in [13] and [2]. L. Baco and
C. Conca [2] considered the same geometry but the equations dier from ours.
J. Sanchez-Hubert [13] investigated almost the same problem. She used techniques
based on the Laplace transformation whereas we apply another approach which makes
it possible to obtain an explicit representation of solutions to the cell equation, which
allows us to investigate the limiting equations and to develop numerical algorithms.
2. Mathematical model. The coupled mechanical system under consideration
is shown in Figure 1. The solid part consists of a substrate and pins located on its top.

Received by the editors January 20, 2003; accepted for publication (in revised form) April 2,
2004; published electronically March 31, 2005.
http://www.siam.org/journals/siap/65-3/42157.html

Center of Advanced European Studies and Research, Ludwig-Erhard-Allee 2, 53175 Bonn, Ger-
many (phoman@caesar.de, botkin@caesar.de).

Lavrentyev Institute of Hydrodynamics, 630090 Novosibirsk, Russia, Center of Advanced Euro-

pean Studies and Research, Ludwig-Erhard-Allee 2, 53175 Bonn, Germany (starovoitov@caesar.de).
983
984 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
The pin structure is assumed to be periodic in the plane (x
1
, x
2
) and independent of x
3
.
The domain of the coupled system is denoted by R
3
. For simplicity, we suppose
that is the cube x R
3
[ x
k
(1; +1), k = 1, 2, 3. The domains occupied by
the uid and elastic continua are denoted by
F
and
S
, respectively; the boundary
separating the continua by . Thus, =
F

S
. Let ()
F
=
F
and
()
S
=
S
. Then the sets ()
F
and ()
S
are the boundaries of the
domains
F
and
S
, respectively.
2.1. Governing equations. We assume that the uid is weakly compressible,
which is physically correct because the operation frequency of the coupled structure
lies in the acoustic range and the displacements of the uid particles are small. This is
a typical acoustic approximation which additionally utilizes linearized NavierStokes
equations (see [9]).

Fig. 1. Coupled system: =

F

S
The solid part of the system will be described using the linear elasticity approach.
This linear setting is supplemented by the assumption that the domains
F
and
S
remain unchangeable. Therefore, the coupled mechanical system is described by the
following equations:

F
u
t
= p + div Pu
x
+
F
f in
F
, (2.1)
p
t
= div u in
F
, (2.2)

S
v
tt
= div Gv
x
+
S
f in
S
. (2.3)
Let n be the normal vector to the uid-solid interface . The no-slip and stress
equilibrium conditions on read
v
t
= u on , (2.4)
Gv
x
n =
_
p1 + Pu
x
_
n on . (2.5)
The boundary and initial conditions are prescribed:
u = 0 on ()
F
, (2.6)
v = 0 on ()
S
, (2.7)
u[
t=0
= u
0
, p[
t=0
= p
0
in
F
, (2.8)
v[
t=0
= v
0
, v
t
[
t=0
= v
0
in
S
. (2.9)
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 985
Here,
F
and
S
are the constant densities of the uid and of the solid parts, re-
spectively; u is the velocity eld of the uid, p is the pressure in the uid, v is the
displacement eld of the solid part, and f is an external force like the gravity. The
coecient characterizes the compressibility of the uid. The fourth-rank tensor
P = P
ijkl
is dened through the relation
Pu
x
= 1 div u + 2T(u). (2.10)
The unit tensor 1 has the components 1
ij
=
ij
, where
ij
is the Kronecker sym-
bol. The strain velocity tensor T(u) has, as is usual, the components T
ij
(u) =
1/2 (u
i
/x
j
+ u
j
/x
i
) . The symbols and denote positive balk and dynamic
viscosity coecients of the uid, respectively. As is usual, the summation over re-
peating indices is assumed. The components G
ijkl
of the elastic stiness tensor G
can be arbitrary up to base restrictions so that arbitrary anisotropic solids can be
considered.
The model (2.1)(2.9) was investigated in [10] where it was supposed to use the
velocity instead of the displacement in (2.3). Following this approach, we introduce
the integral operator

t
w =
_
t
0
w(s)ds
that enables us to rewrite (2.3) in the form

S
u
t
= div
_
G
t
u
x
_
+ div (
0
+
S
f, (2.11)
where u = v
t
, (
0
= Gv
0
x
in
S
. Similarly, the pressure p can be expressed from (2.3)
as follows:
p =
1
div
t
u + p
0
in
F
. (2.12)
Let be the characteristic function of the domain
F
. Then (2.1), (2.2), and (2.3)
can be written in the whole domain as one equation with discontinuous coecients
u
t
= div
_
M
t
u
x
_
+ div ^
0
+ f, (2.13)
where
M
t
= P +
_

1
1 1 + (1 )G
_

t
,
=
F
+
S
(1 ), ^
0
= p
0
1 + (1 )(
0
.
The interface condition (2.4) is equivalent to the continuity of u on but the
condition (2.5) now assumes the form
_
G
t
u
x
+(
0
_
n =
_

1
div
t
uI p
0
I + Pu
x
_
n on (2.14)
accounting (2.12). The boundary and initial data are
u = 0 on ()
F
, (2.15)
u[
t=0
= u
0
in , (2.16)
986 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
where the uid initial condition u
0
is extended to
S
by setting u
0
(x) = v
0
(x) for
x
S
.
Remark 2.1. One can forget the initial distribution v
0
of the displacement when
considering (2.13). It is sucient to prescribe the initial velocity eld u
0
in , the
initial stress (
0
in
S
(this replaces the information about v
0
), and initial pressure
p
0
in
F
. The functions (
0
and p
0
yield the function ^
0
involved in (2.13).
Remark 2.2. For mechanical reasons, the tensors P
ijkl
and G
ijkl
have the follow-
ing properties:
Z
ijkl
= Z
ijlk
= Z
klij
= Z
jikl
, Z
ijkl
1
ij
1
kl
0,
Z
ijkl
1
ij
1
kl
= 0 if and only if 1
kl
+1
lk
= 0 for all k, l = 1, 2, 3.
Here, Z
ijkl
stands for P
ijkl
or G
ijkl
.
2.2. Renement of the structure. Let us dene the structure of the regions
,
F
, and
S
more precisely. The pin structure (see Figure 1) is supposed to be
(x
1
, x
2
)-periodic. Without loss of generality, we assume that the periodicity cell is a
square with the side length equal to , where is a positive number. After scaling
with the factor 1/, the cell becomes the unit square = [0, 1] [0, 1]. Let
S
be
the 1/-scaled projection of a solid pin to the (x
1
, x
2
)-plane. It is assumed to be a
smooth, simply connected domain in such that its boundary
S
does not meet
. Denote by
F
the domain
S
(see Figure 2).
Fig. 2. Structural cell = [0, 1] [0, 1].
Let x = (x
1
, x
2
) and ( x) be the -periodic extension of the characteristic func-
tion of the domain
F
to all R
2
. Then the function introduced in the previous
subsection can be represented as follows:
(x) = ( x, x
3
) =
_

_
1, x
3
> ,
(
x

), 0 x
3
,
0, x
3
< 0.
(2.17)
Remember that is the thickness of the pin layer. If 0, the pin structure becomes
ner in the (x
1
, x
2
)-plane, whereas its height remains constant. Thus, the problem
(2.13)(2.16) depends in fact on . For this reason, we call it Problem S

.
Definition 2.3. A function u is called a weak solution to Problem S

if
u L

(0, T; L
2
()) L
2
(0, T; H
1
(
F
)),
t
u L

(0, T; H
1
0
()) (2.18)
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 987
and the integral identity
_
T
0
_

_
u
t
+M
t
u
x
:
x
+^
0
:
x
f
_
dxdt =
_

u
0

0
dx (2.19)
holds for every smooth function such that [
t=T
= [

= 0.
In this denition and further, T is an arbitrary positive number; the colon denotes
the convolution of tensors so that | : 1 = |
ij
1
ij
for all second-rank tensors | and 1;
and the notation f
0
means f[
t=0
. Remark that the second inclusion of (2.18) prevents
jumps of u on .
2.3. Solvability of Problem S

. It is not dicult to prove existence of a weak

solution to Problem S

. This question was investigated in [10, section 9.1], and the

following result was established.
Theorem 2.4. Let u
0
L
2
(), ^
0
L
2
(), and f L
2
([0, T] ). Then
there exists a unique weak solution to Problem S

, and the following energy estimate

holds:
ess sup
t(0,T)
_
|u(t)|
2
L
2
()
+|T
_

t
u
_
|
2
L
2
(
S
)
_
+
_
T
0
|T(u(t))|
2
L
2
(
F
)
dt C,
(2.20)
where C is a constant which depends on |u
0
|
L
2
()
, |^
0
|
L
2
()
, and |f|
L
2
([0,T])
but does not depend on .
Corollary 2.5. Under the conditions of Theorem 2.4, there exists an indepen-
dent of constant C such that
ess sup
t(0,T)
|
t
u(t)|
H
1
()
C. (2.21)
Generally speaking, the estimates (2.20) and (2.21) are sucient to fulll the
homogenization of Problem S

due to Proposition 3.9 which will be given below.

However, some technical diculties must be overcome in this case. To avoid that, a
stronger estimate for u will be obtained under some compatibility conditions. The
next theorem states such a result.
Theorem 2.6. Let u
0
H
1
(), ^
0
L
2
(), f, f
t
L
2
([0, T] ), and
div
_
Pu
0
x
+^
0
_
L
2
(). (2.22)
Then the weak solution to Problem S

satises the estimate

ess sup
t(0,T)
_
|u
t
(t)|
L
2
()
+|u
x
(t)|
L
2
()
_
C, (2.23)
where C is an independent of constant.
Proof. Let us introduce a function w as a solution of the problem
w
t
= div
_
M
t
w
x
_
+ div
_

1
I div u
0
+ (1 )Gu
0
x
_
+ f
t
,
w[
t=0
= w
0
= div
_
Pu
0
x
+^
0
_
+ f
0
,
w[

= 0.
The energy estimate for this problem appears as follows:
ess sup
t(0,T)
_
|w(t)|
2
L
2
()
+|T
_

t
w
_
|
2
L
2
(
S
)
_
+
_
T
0
|T(w(t))|
2
L
2
(
F
)
dt C,
(2.24)
988 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
which yields
ess sup
t(0,T)
|
t
w|
H
1
()
C.
The assertion of the theorem is an immediate consequence of the last estimates because
the function dened as
u(x, t) =
_
t
0
w(x, s) ds +u
0
(x) =
t
w(x, t) +u
0
(x)
is the solution of Problem S

, and u
t
= w.
According to the denition of u
0
, the requirement u
0
H
1
() expresses the no-
slip condition on at the initial time instant t = 0. The requirement (2.22) expresses
the stress equilibrium condition on at t = 0. From the mechanical point of view,
such conditions hold for any time instant including the initial one. Therefore, the
requirements of the theorem are feasible.
3. Homogenization of the structure.
3.1. Two-scale convergence. Let us denote by u

the solution of Problem S

.
In order to emphasize the dependence of on , we denote it by

. Our goal is to
perform the passage to the limit in Problem S

as 0. To do this, we use the

two-scale convergence method introduced by G. Nguetseng and developed by other
mathematicians (see [12], [11], [1], [7]). Let us formulate the main results of this
approach adapted to our situation.
Theorem 3.7. Let w

be a bounded sequence in L
2
_
[0, T]
_
. There exists a
subsequence, still denoted by w

, and a function w(t, x,

) L
2
_
[0, T]
_
such
that
lim
0
_
T
0
_

(t, x)
_
t, x,
x

_
dx =
_
T
0
_

w(t, x,

)(t, x,

) d

dxdt
for every smooth function (t, x,

) which is -periodic in

. Such a sequence w

is
said to be two-scale convergent to w(t, x,

).
Recall the notation x = (x
1
, x
2
) and

= (
1
,
2
).
Theorem 3.8. Let a sequence w

converge weakly to w in L
2
_
0, T; H
1
()
_
.
Then w

two-scale converges to w and there exists a function w(t, x,

) in L
2
_
[0, T]
; H
1
#
()/R
_
such that w

two-scale converges to
x
w(t, x) +

w(t, x,

) up to a
subsequence.
Here H
1
#
() is the space of periodic functions which belong to the space H
1
().
Since all functions under consideration do not depend on
3
, the notation

=
(
1
,
2
, 0)

is used below.
As a simple application of the theorems stated above, we formulate (without
proof) the following result concerning the convergence of solutions of Problem S

.
Proposition 3.9. Let u

be the sequence of solutions to Problem S

. Then there
exist a subsequence (still denoted by u

) and a function u(t, x) such that

1. u

two-scale converges to u, and u

u weakly in L
2
_
[0, T]
_
;
2.
t
u

two-scale converges to
t
u, and
t
u

t
u in L
2
_
[0, T]
_
;
3.
t
u

two-scale converges to
x

t
u+

, where (t, x,

) is a function from
L
2
_
[0, T] ; H
1
#
()/R
_
.
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 989
3.2. Passage to the limit in Problem S

. Let the initial data of Problem S

satisfy the conditions of Theorem 2.6. A solution u

of Problem S

satises the
following integral identity,
(3.1)
_
T
0
_

t
+M
t
u
x
:
x
+^
0
:
x

f
_
dxdt =
_

u
0

0
dx,
where

, M
t
, and ^
0
are dened as in (2.13) but with replaced by

. Let us
take
(t, x) = (t, x) +
_
t, x,
x

_
,
where and are arbitrary functions that vanish for x and at t = T. The-
orem 3.8 enables the passage to the limit in (3.1) as 0. The limiting equations
look as follows:
(3.2)
_
T
0
_

_
u
t
+M
t
(u
x
+u

) :
x
+^
0
:
x
f
_
d

dxdt
=
_

u
0

0
d

dx,
_

_
M
t
(u
x
+u

) :

+^
0
:

_
d

= 0 in L
2
_
[0, T]
_
. (3.3)
These equations hold for all functions H
1
([0, T] ) and H
1
#
() such that
vanish on and at t = T. The coecients , M
t
, and ^
0
are dened as in (2.13)
with (x) replaced by (x,

). The function (x,

) is dened as in subsection 2.2:

(x,

) =
_

_
1, x
3
> ,
(

), 0 x
3
,
0, x
3
< 0.
Equation (3.3) is called a cell equation.
Equations (3.2) and (3.3) are coupled through the auxiliary function u. The next
step consists of nding u from the cell equation (3.3) and substituting the obtained
expression into (3.2).
4. Explicit solving of the cell equation.
4.1. Operator form of the cell equation in a Hilbert space. It is appro-
priate to rewrite (3.3) as an equation in the Hilbert space H = H
1
#
()/R with the
inner product
u, v) =
_

u
i

j
v
i

j
d

.
The norm in H is denoted by | |. Let us dene operators / and B as follows:
/u, v) =
_

P
ijkl
u
k

l
v
i

j
d

, Bu, v)=
_

ij

kl
+(1) G
ijkl
_
u
k

l
v
i

j
d

990 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV

for all functions u, v H. Due to the Riesz representation theorem, there exist n
0
,
a
kl
, and b
kl
, k, l = 1, 2, 3, such that
n
0
, v) =
_

^
0
: v

, a
kl
, v) =
_

P
ijkl
v
i

j
d

,
b
kl
, v) =
_

ij

kl
+ (1 ) G
ijkl
_
v
i

j
d

for all v H. Remark that /, B, a, b, and n

0
do not depend on t and depend on the
variable x just in the same way as the function (x,

). So we can consider x and t

in (3.3) as parameters.
Now, the problem (3.3) transforms to the following equation in the space H:
/u +B
t
u = g, (4.1)
where
g =
_
a
kl
+b
kl

t
_
u
k
x
l
n
0
and u(x, t) is from (3.2) and (3.3).
Since the operators / and B are trivial whenever x
3
, [0, ], we consider (4.1)
for x
3
[0, ], which corresponds to the treatment of the pin layer. In this case, the
operators / and B are degenerated. Therefore, some diculties appear when solving
(4.1).
The next section is devoted to the study of the data of (4.1) to prepare tools for
its explicit solving.
4.2. Properties of A, B, and g.
Proposition 4.10. The operator / has the following properties:
1. / is a bounded self-adjoint operator on H.
2. /u, u) 0 for all u H.
3. The null-space N(/) = u H : u is constant in
F
, and N(/)

u
H : u = 0 in
S
.
4. There exist positive constants c and C such that
c |u|
2
/u, u) C |u|
2
(4.2)
for all u N(/)

.
5. The range R(/) is closed in H, R(/) = N(/)

, and /
1
is dened and
bounded as an operator on R(/).
Proof. Assertions 1 and 2 are obvious (see Remark 2.2). The third assertion
consists of two parts. In order to prove the rst one we have only to establish that
N(/) u H : u is constant on
F

because the opposite inclusion is clearly true. Due to the positiveness of the operator
/, its null-space consists of functions u which satisfy the condition /u, u) = 0.
Thus, u N(/) implies
/u, u) =
_

P
ijkl
u
k

l
u
i

j
d

= 0.
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 991
Consequently, T(u) = 0 in
F
, and, hence, u is constant in
F
because of its peri-
odicity.
Let u N(/)

. By denition, this means that

_

u
k

l
v
k

l
d

=
_

S
u
k

l
v
k

l
d

= 0
for any function v C

() such that v is constant on

F
. Consequently, u is
harmonic in
S
, which proves the third assertion.
To validate assertion 3, we need only to prove the left inequality since the right
one is obvious. Due to the Korn inequality (see, e.g., [14]), there exists a positive
constant c
1
such that
_

F
[u

[
2
d

c
1
/u, u)
for every u H. If u N(/)

, then u is harmonic in
S
and there exist positive
constants c
2
and c
3
such that
c
2
_

S
[u

[
2
d

|u|
H
1/2
(
S
)
c
3
_

F
[u

[
2
d

.
That is, /u, u) c |u|
2
for some constant c.
When proving assertion 5, denote by /
R
the restriction of / to N(/)

. Due to
the estimate (4.2), R(/
R
) is closed in H. Since R(/) = R(/
R
), we conclude that
R(/) is also a closed subspace of H. This implies that N(/)

= R(/) = R(/), and

(4.2) is true for u R(/). Thus, /
1
exists and is bounded if / is considered being
restricted to R(/). The proposition is proved.
Proposition 4.11. The operator B has the following properties:
1. B is a bounded self-adjoint operator on H.
2. Bu, u) 0 for all u H.
3. The null-space N(B) = u H : T(u) = 0 in
S
and div u = 0 in
F
, and
N(B)

u H : u = q in
F
for some q L
2
().
4. There exist positive constants c and C such that
c |u|
2
Bu, u) C |u|
2
(4.3)
for all u N(B)

.
5. The range R(B) is closed in H, R(B) = N(B)

, and B
1
is dened and bounded
as an operator on R(B).
Proof. The rst two assertions are obvious. To prove the third one, note that
Bu, u) =
_

ij

kl
+ (1 )G
ijkl
_
u
i

j
u
k

l
d

=
1
_

F
(div u)
2
d

+
_

S
G
ijkl
u
i

j
u
k

l
d

for every u H. Therefore, Bu, u) = 0 if and only if div u = 0 in

F
and T(u) = 0
in
S
.
992 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
If u N(B)

, then the equalities

0 = u, v) =
_

=
_

T(u) : T(v) d

=
_

F
T(u) : T(v) d

(4.4)
hold for every v N(B). Let u
k
N(B)

be a sequence of smooth functions that

converges to u in H. Such a sequence exists because C

() is dense in N(B)

.
Relation (4.4) is also valid for all u
k
. If v is an arbitrary smooth function such that
div v = 0 and supp v
F
, then v N(B), and
0 =
_

F
T(u
k
) : T(v) d

=
_

F
div
_
T(u
k
)
_
v d

.
Consequently, there exist functions q
k
L
2
() such that divT(u
k
) = q
k
for all k.
Passing to the limit yields divT(u) = q. That is, u = q, where q = q divu.
This proves the third assertion.
The right inequality of the fourth assertion is obvious. Let us prove the left
one. According to the classical theory of the Stokes equations (see [4, Chap. 4]), the
following estimate holds for all u N(B)

:
_

F
[u

[
2
d

c
1
_
|div u|
2
L2(
F
)
+|u

|
2
H
1/2
(
S
)/R
_
,
where u

is the trace of u on
S
. On the other hand,
|u

|
2
H
1/2
(
S
)/R
c
2
_

S
[u

[
2
d

.
Thus, there exists a positive constant c
3
such that
|u|
2
c
3
_
_

S
[u

[
2
d

+|div u|
2
L2(
F
)
_
(4.5)
for every u N(B)

. In order to obtain (4.3), it is sucient to prove that there

exists a positive constant c
4
such that
_

S
[u

[
2
d

c
4
Bu, u) (4.6)
for u N(B)

. This can be done using standard contradiction arguments. As-

sume the converse, i.e., there exists a sequence u
n
N(B)

, n N, such that
_

S
[u
n

[
2
d

= 1 and Bu
n
, u
n
) 0 as n 0. The estimate (4.5) implies that the
sequence u
n
is bounded in H too. Thus, there exists its subsequence (still denoted
by u
n
) that converges weakly in H and H
1
(
S
)/R but strongly in L
2
() to a func-
tion u. Note that u N(B)

since N(B)

is weakly closed in H. Using the Korn

inequality yields
_

S
[u
n

[
2
d

C
_
B(u
n
u), u
n
u) +|u
n
u|
2
L
2
()
_
.
The passage to the limit in this inequality implies that Bu, u) = 0 and u
n

u in H. This means that u N(B)

N(B) and u = 0 in H. On the other

HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 993
hand,
_

S
[u

[
2
d

= lim
n
_

S
[u
n

[
2
d

= 1. This contradiction proves (4.6) and,

consequently, (4.3).
The proof of the fth assertion is the same as for the operator / in Proposi-
tion 4.10.
Proposition 4.12. The following is true:
a
kl
, b
kl
, n
0
R(/) R(B), k, l = 1, 2, 3.
Consequently, g R(/) R(B) for almost all t and x, where g is the right-hand side
of the cell equation (4.1).
Proof. Due to Propositions 4.10 and 4.11, w R(/) R(B) if and only if
w, v) = 0 for all v N(/) N(B). Let us verify this condition for a
kl
. The
functions b
kl
and n
0
can be treated in the same way. Let v be an arbitrary function
from N(/). That is, v is a constant in
F
because of Proposition 4.10. Thus,
a
kl
, v) =
_

F
P
ijkl
v
i

j
d

= 0.
If v N(B) then T(v) = 0 in
S
according to Proposition 4.11, and
a
kl
, v) =
_

F
P
ijkl
v
i

j
d

=
_

P
ijkl
v
i

j
d

S
P
ijkl
v
i

j
d

=
_

S
P
ijkl
T
ij
(v) d

= 0.
Here, we used the periodicity of v in and the symmetry of the tensor P (see
Remark 2.2). This proves the proposition.
Proposition 4.13.
N(/) N(B) = 0.
Proof. If u N(/) N(B), then T(u) = 0 in due to Propositions 4.10 and
4.11. That is, u is constant in because of its periodicity. This means that u = 0 in
H.
The result of Proposition 4.13 implies that the operator /+B is invertible for
every > 0. Besides that, it is not dicult to see that the operator (/ + B)
1
is
bounded in H: Let us introduce the following closed subspaces of H:
E
A
= (/+B)
1
R(/),
E
B
= (/+B)
1
R(B),
E = E
A
E
B
= (/+B)
1
_
R(/) R(B)
_
.
Note that the spaces E, E
A
, and E
B
do not depend on . More precisely, if
E

A
= (/ + B)
1
R(/) then E

A
= E

A
for all > 0 and > 0. This follows from
simple arguments like those. If x E

A
, then (/ + B)x R(/) and Bx R(/).
Consequently, (/+B)x R(/) and x E

A
. That is, E

A
E

A
. In the same way
we can obtain that E

A
E

A
.
Lemma 4.14. The operator / maps the space E
B
into R(B), and the operator B
maps the space E
A
into R(/).
994 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
Proof. The rst part is true due to the following implications:
x E
B
= (/+B)x R(B) = /x R(B).
The second part is being proved analogously.
Lemma 4.15. If X is a closed subspace of H then /(X) and B(X) are closed
in H.
Proof. Let us verify this assertion for the operator / by taking an arbitrary
sequence u
n
/(X) which converges to a function u in H. There exists a corre-
sponding sequence v
n
R(/) X such that u
n
= /(v
n
). Due to Proposition 4.10,
the operator /
1
is bounded on R(/). This implies that the sequence v
n
con-
verges in H to a function v which is in X because X is closed. In the limit, we have
u = /(v). That is, u /(X), which proves the lemma.
Proposition 4.16.
BE
A
= /E
B
= R(/) R(B).
That is, for every R(/) R(B), there exist
B
E
B
and
A
E
A
such that
= /
B
= B
A
.
Proof. Let us prove the rst claim. Due to Lemma 4.14, BE
A
R(/) R(B).
Besides that, Lemma 4.15 implies that BE
A
is a closed subspace in H. Suppose that
BE
A
,= R(/) R(B). Then there exists x R(/) R(B) such that x, y) = 0 for
every y BE
A
. That is, x, B(/+B)
1
/z) = 0 for all z H, and
/(/+B)
1
Bx, z) = 0 for all z H.
Consequently, (/ + B)
1
Bx N(/) and, hence, Bx (/ + B)N(/) = BN(/).
That is, there exists y N(/) such that Bx = By and, therefore, B(xy) = 0. This
implies that w = xy N(B). Thus, x = y +w, where y N(/), and w N(B).
That is, x N(/) N(B). Consequently, x = 0 because
_
N(/) N(B)
_

_
R(/)
R(B)
_
= 0. The proposition is proved.
Let us introduce the restrictions /
E
and B
E
of the operators / and B to the
space E.
Theorem 4.17.
1. The operators /
E
and B
E
map E onto R(/) R(B).
2. The operators /
E
, B
E
: E R(/) R(B) are one to one.
3. There exist bounded operators /
1
E
, B
1
E
: R(/) R(B) E.
Proof. Let us prove these assertions for the operator /
E
only. The operator B
E
can be treated in the same way.
1. Since E E
B
, Proposition 4.16 and Lemma 4.15 imply that /E R(/)
R(B), and /E is a closed subspace in H. Suppose that /E ,= R(/) R(B). This
means that there exists x R(/) R(B) such that x, y) = 0 for every y /E.
That is,
(/+B)
1
/x, z) = x, /(/+B)
1
z) = 0
for all z R(/) R(B). Thus, due to Proposition 4.16,
(/+B)
1
/x, Bz) = 0 for all z E
A
. (4.7)
Since (/+B)
1
/x E
A
, we can take z = (/+B)
1
/x. Then the relation (4.7)
implies that (/ + B)
1
/x N(B), that is, /x /N(B). Consequently (see the
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 995
end of the proof of Proposition 4.16), x = 0, which proves the rst assertion of the
theorem.
2. We have to prove that N(/) E = 0. Let x E and /x = 0. Then
Bx = (/+B)x R(/) R(B), that is, Bx R(/). But x N(/) = R(/)

and,
consequently, Bx, x) = 0. Since B is a positive operator, the last relation implies
that x N(B). Thus, x N(/) N(B) = 0, which proves the second assertion of
the theorem.
3. This assertion is the consequence of parts 1 and 2. The theorem is
proved.
4.3. Solving the cell equation. Now we are in position to nd an explicit
representation of solutions to the cell equation (4.1). With a new unknown function
=
t
u, the problem (4.1) assumes the form
/
t
+B = g, (0) = 0. (4.8)
As it follows from Theorem 4.17, the operator /
E
(/ restricted to E) is invertible,
the operator /
1
E
B
E
bounded, and /
1
E
g E. Therefore, the problem

t
+/
1
E
B
E
= /
1
E
g, (0) = 0 (4.9)
is uniquely solvable on the subspace E, and the solution is of the form
(t) =
_
t
0
e
(ts)A
1
E
B
E
/
1
E
g(s) ds. (4.10)
Theorem 4.18. Equations (4.8) and (4.9) are equivalent.
Proof. Obviously, if is a solution to (4.9), then satises (4.8). If is a solution
to (4.8), then the function = e
t
solves the problem
/
t
+ (/+B) = e
t
g, (0) = 0. (4.11)
Since the operator / + B is nondegenerate for any > 0, we can rewrite (4.11) as
follows:
(/+B)
1
/
t
+ = e
t
(/+B)
1
g, (0) = 0. (4.12)
Due to Proposition 4.12, g R(/), and, hence (t) must belong to E
A
for all t.
Therefore, (t) E
A
for all t. On the other hand, (4.8) can be rewritten as follows:
(/+B)
t
B
t
+ B = g.
That is,

t
= (/+B)
1
B(
t
) + (/+B)
1
g.
Taking into account that (t) and
t
(t) E
A
for all t, we establish, using Proposi-
tion 4.16, that
t
(t) E for all t. Since (0) = 0, we conclude that (t) E for all t.
Therefore, is a solution of (4.9). The theorem is proved.
Thus, the unique solution of the problem (4.8) is given by (4.10) and the unique
solution u of the problem (4.1) reads as
u(t) =
t
(t) = /
1
E
g(t) /
1
E
B
E
_
t
0
e
(ts)A
1
E
B
E
/
1
E
g(s) ds. (4.13)
996 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
5. Homogenized structure.
5.1. Limiting equations. Substitution of the expression for g into (4.13) gives
u(t) = e
tA
1
E
B
E
/
1
E
n
0
/
1
E
a
kl
u
k
(t)
x
l

_
t
0
m
kl
(t s)
u
k
(s)
x
l
ds,
(5.1)

t
u(t) =
_
e
tA
1
E
B
E
1
_
B
1
E
n
0
B
1
E
b
kl
_
t
0
u
k
(s)
x
l
ds
_
t
0
m
kl
(t s)
u
k
(s)
x
l
ds,
(5.2)
where
m
kl
(t) = /
1
E
B
E
e
tA
1
E
B
E
_
/
1
E
a
kl
B
1
E
b
kl
_
E,
m
kl
(t) = e
tA
1
E
B
E
_
/
1
E
a
kl
B
1
E
b
kl
_
E.
The integration by parts and the formula
d
ds
e
(ts)A
1
E
B
E
= /
1
E
B
E
e
(ts)A
1
E
B
E
are applied when deriving (5.1) and (5.2). Now we are in position to compute the
principal term
_

M
t
ijkl
u
k

l
d

= a
ij
, u) +b
ij
,
t
u)
appearing in the limiting (homogenized) equation (3.2). Utilizing (5.1) and (5.2) and
computing other terms in (3.2), we obtain the following limiting equation:
(5.3)
_
T
0
_

u
i

i
t
+
_
P
ijkl

ijkl
_
u
k
x
l

i
x
j
+
_
t
0
_

ij

kl
+ (1 )G
ijkl

ijkl
+
ijkl
(t s)
_
u
k
x
l
ds

i
x
j
_
dxdt
=
_
T
0
_

f
i

i

_

ij
p
0

ij
+ (1 )(
0
ij
_

i
x
j
_
dxdt +
_

u
0

0
dx,
where
(x) =
_

=
F
+ (1 )
S
,

ij
= a
ij
, e
tA
1
E
B
E
/
1
E
n
0
) +b
ij
,
_
e
tA
1
E
B
E
1
_
B
1
E
n
0
),

ijkl
= a
ij
, /
1
E
a
kl
),

ijkl
= b
ij
, B
1
E
b
kl
),

ijkl
(t) = a
ij
, m
kl
) b
ij
, m
kl
).
Let us denote by P, G, and o
0
the tensors with components
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 997
P
ijkl
= P
ijkl

ijkl
, G
ijkl
=
1

ij

kl
+ (1 )G
ijkl

ijkl
,
o
0
ij
=
ij
p
0

ij
+ (1 )(
0
ij
.
Let us divide the domain into three parts:

f
= x [ x
3
> ,
s
= x [ x
3
< 0,
h
= x [ 0 < x
3
< .
Let
+

be the boundary between

f
and
h
,

the boundary between

s
and
h
.
That is, =
f

+

s
. Note that (x) = 1 if x
f
, (x) = 0 if
x
s
, and is a constant from the interval (0, 1) for x
h
. As for
ijkl
,
ijkl
,
ij
,
and
ijkl
, they are constants for x
h
and equal to zero if x
f

s
, so that the
integral identity (5.3) delivers the following equations which should be understood in
the distributional sense:

F
u
t
div Pu
x

1
div
t
u = p
0
+
F
f, x
f
,
(5.4)

S
u
t
div
t
Gu
x
= div (
0
+
S
f, x
s
,
(5.5)

u
t
div Pu
x
div
t
Gu
x
div
_
t
0
(t s)u
x
(s) ds + div o
0
=

f, x
h
.
(5.6)
The natural interfacial boundary conditions at
+

and

can be derived from the

integral identity (5.3). Equations (5.4) and (5.5) coincide with (2.1) and (2.11), re-
spectively. That is, the governing equations for the pure fractions do not change after
the homogenization, which have been expected. What is new is an integral-dierential
equation (5.6) which cannot be reduced to a pure dierential equation by dierenti-
ating or by a substitution like w =
t
u. The operators involved in the equation have
to be investigated to conrm the parabolic type of its principal part.
5.2. Investigation of P and G. It is not dicult to verify that the ten-
sors P and G have the symmetry properties mentioned in Remark 2.2. Therefore,
P
ijkl
Z
ij
Z
kl
= 0 and G
ijkl
Z
ij
Z
kl
= 0 for every skew symmetric matrix Z. The main
objective of this subsection is to prove the strong positiveness of the tensor P and the
nonnegativeness of G on the space of symmetric matrices. The null-space of G will
be also described.
Proposition 5.19. For every second-rank tensor Z, the following is valid:
P
ijkl
Z
ij
Z
kl
0, G
ijkl
Z
ij
Z
kl
0.
Proof. Let us prove the assertion for P. Denote z = a
ij
Z
ij
. Due to Proposi-
tion 4.12, z R(/)R(B) and, as it follows from Theorem 4.17, there exists a unique
y E such that /
E
y = z. This means that
_

P
ijkl
y
i

j
v
k

l
d

= z, v) =
_

P
ijkl
Z
ij
v
k

l
d

(5.7)
for all v H. On the other hand, the denition yields

ijkl
Z
ij
Z
kl
= a
ij
Z
ij
, /
1
E
a
kl
Z
kl
) = z, /
1
E
z) = /
E
y, y).
998 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
From the last relation and (5.7) with v = y, we obtain
(5.8)
P
ijkl
Z
ij
Z
kl
= P
ijkl
Z
ij
Z
kl
/
E
y, y) =
_

_
P
ijkl
Z
ij
Z
kl
P
ijkl
y
i

j
y
k

l
_
d

=
_

P
ijkl
_
Z
ij

y
i

j
__
Z
kl

y
k

l
_
d

.
The right-hand side of the last relation is clearly positive and the required assertion
is proved for the tensor P. Positiveness of the tensor G can be veried in the same
way.
The next theorem states the strong positiveness of the tensor P.
Theorem 5.20. There exists a positive constant C such that
P
ijkl
Z
ij
Z
kl
C [Z[
2
for every symmetric second-rank tensor Z. Here, [Z[
2
= Z
ij
Z
ij
.
Proof. Assume that the assertion of the theorem is false. Then there exists a
sequence Z
n
such that [Z
n
[ = 1 and P
ijkl
Z
n
ij
Z
n
kl
0 as n . The sequence
Z
n
is compact in R
3
R
3
and, therefore, it has a subsequence denoted again
by Z
n
, which converges to a matrix Z
0
such that [Z
0
[ = 1. This means that
the corresponding sequences z
n
and y
n
, dened as z
n
= a
ij
Z
n
ij
and y
n
= /
1
E
z
n
,
converge in H to z
0
and y
0
, respectively. We use here the notations introduced in
the proof of the previous proposition. Thus, the relation
P
ijkl
Z
0
ij
Z
0
kl
= P
ijkl
Z
0
ij
Z
0
kl
/
E
y
0
, y
0
) =
_

P
ijkl
_
Z
0
ij

y
0
i

j
__
Z
0
kl

y
0
k

l
_
d

= 0.
holds due to (5.8). That is,
P
ijkl
_
Z
0
ij

y
0
i

j
__
Z
0
kl

y
0
k

l
_
= 0 in ,
and, consequently, T(y
0
) = Z
0
in
F
. This implies that T(y
0
Z
0
) = 0 in
F
.
Therefore, y
0
() is a linear function of for
F
. The only linear function satisfying
the periodicity boundary conditions on is a constant, which implies that Z
0
= 0.
This is impossible because [Z
0
[ = 1. This contradiction proves the theorem.
Remark that the arguments like those in the proof of Theorem 5.20 do not lead to
a contradiction in the case of the tensor G. The next theorem shows that the tensor
G is degenerated and describes its null-space.
Theorem 5.21. The tensor G is degenerate, and G
ijkl
Z
ij
Z
kl
= 0 for a symmet-
ric matrix Z if and only if Z
11
+Z
22
= 0 and Z
33
= 0.
Proof. Let us denote z = b
ij
Z
ij
. Due to Proposition 4.12 and Theorem 4.17,
z R(/) R(B), and there exist unique elements y
E
E and y
R
R(B) such that
By
E
= z, By
R
= z. (5.9)
It follows that y
N
= y
E
y
R
N(B). Besides that, B
1
E
By
E
= y
E
. Therefore,
z, B
1
E
z) = By
E
, y
E
) = By
R
, y
R
+y
N
) = By
R
, y
R
).
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 999
The second equation in (5.9) implies that
_

K
ijkl
()
y
R
i

j
v
k

l
d

=
_

K
ijkl
()Z
ij
v
k

l
d

, (5.10)
for all v H, where
K
ijkl
() =
1

ij

kl
+ (1 )G
ijkl
.
As a consequence of this equation, we nd
G
ijkl
Z
ij
Z
kl
= K
ijkl
()Z
ij
Z
kl
z, B
1
E
z) = K
ijkl
()Z
ij
Z
kl
By
R
, y
R
)
=
_

K
ijkl
()
_
Z
ij

y
R
i

j
__
Z
kl

y
R
k

l
_
d

=
_

1
(trZ div y
R
)
2
d

+
_

S
G
ijkl
_
Z
ij
T
ij
(y
R
)
__
Z
kl
T
kl
(y
R
)
_
d

.
Notice that (5.10) is the EulerLagrange equation for the functional
F
z
(y) =
_

K
ijkl
()
_
Z
ij

y
i

j
__
Z
kl

y
k

l
_
d

.
Due to Proposition 4.11 (assertion 4), this functional is strictly convex on R(B) and
y
R
is its unique minimizer there. That is,
G
ijkl
Z
ij
Z
kl
= F
z
(y
R
) = min
yR(B)
F
z
(y).
Thus, G
ijkl
Z
ij
Z
kl
= 0 if and only if there exists y
R
R(B) such that F
z
(y
R
) =
0. It is not dicult to see that F
z
(y) = F
z
(y + w) for every w N(B). Since
R(B) N(B) = H, the existence of y
R
R(B) with F
z
(y
R
) = 0 is equivalent to the
existence of a function y H which satises the condition F
z
(y) = 0. Due to the
positiveness of the functional F
z
, we can conclude that G
ijkl
Z
ij
Z
kl
= 0 if and only if
there exists y H such that
div y = tr Z as


F
, (5.11)
T(y) = Z as


S
. (5.12)
Suppose that both of the last conditions are satised. Since functions from H do
not depend on
3
, (5.12) implies that Z
33
= 0. Moreover, due to (5.12), div y = tr Z
in
S
. That is, div y = tr Z in . Integrating this equality over we nd that
tr Z = 0 because y is periodic. Thus, we have proved the assertion of the theorem in
one direction (the necessity).
Let us suppose that Z
11
+Z
22
= 0 and Z
33
= 0. In order to complete the proof
of the theorem, we have to prove that there exists a function y H satisfying (5.11)
and (5.12). Equation (5.12) is easy to solve. Namely, its solution appears as follows:
y() = Z +Q +y
0
,


S
,
where Q is a skew-symmetric matrix and y
0
is a constant which can be dropped
because functions from the space H are dened up to a constant. Let us denote T =
1000 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
Z +Q. Since functions from H do not depend on
3
, we nd that T
i3
= 0 (i = 1, 2, 3)
and y
3
= T
31

1
+ T
32

2
for


S
. We extend y
3
to the whole domain in such a
way that it would be a periodic function (assuming equal values on the opposite edges
of ).
In order to determine y
1
and y
2
in
F
, we have to solve the problem
y
1

1
+
y
2

2
= 0,


F
,
y(

) = T ,


S
,
y
1
and y
2
are periodic in .
This problem is clearly solvable, and the theorem is completely proved.
As one can see from (5.6), the tensor G describes elastic stresses in the homoge-
nized continuum. Theorem 5.21 says that the homogenized material has rather strange
properties. Namely, it does not resist to the deformation, if the rst invariant and the
component (3,3) of the corresponding strain tensor are equal to zero. In other words,
such deformations do not produce any stresses. The described class of deformations
is suciently large. It contains all deformations which do not change volume. The
following assertion is a simple consequence of Theorem 5.21.
Corollary 5.22. If i ,= j and k ,= l, then G
ijkl
= 0.
This property of the tensor G yields an interesting conclusion about the passage to
the limit as 0. If we set = 0 formally, the elastic structure will occupy the whole
layer
h
. Therefore, it can seem that the limiting material must be the same as the
original elastic one so that lim
0
G = G. Nevertheless, it is wrong in general because
the properties of the tensor G stated in Theorem 5.21 and in Corollary 5.22 do not
depend on . Thus, if, for instance, the tensor G is not degenerate or G
1212
,= 0, then
lim
0
G ,= G. The physical reason is that the elastic structure consists of separate
bristles for each > 0, which diers from the bulk material corresponding to = 0.
6. Numerical procedures. The formulas for the coecients P, G, and con-
tain the functions a
kl
, b
kl
, n
0
, the operators /
E
, B
E
, and their inverse dened in
H = H
1
#
()/R. From the mathematical point of view, all these functions and oper-
ators are well dened and completely described. However, numerical implementation
of these formulas requires some eort. The computation of the functions a
kl
, b
kl
, and
n
0
is not dicult if one uses the nite element method. The situation with the oper-
ators /, B is not so trivial, because they must be restricted to the subspace E, which
creates additional problems when using nite elements. Below, we propose numerical
procedures that can be implemented using conventional nite element software.
6.1. Calculation of a
kl
, b
kl
, and n
0
. Let us introduce functions
k
H,
k = 1, 2, 3, as solutions of the following problems:
_

i
v

i
d

=
_

k
d

for all v H.
These problems can be easily solved applying the nite element method. Note that

3
= 0 because functions from the space H do not depend on
3
. It is not dicult to
see that
a
i
kl
= P
kli1

1
+ P
kli2

2
,
b
i
kl
=
1

kl

i
G
kli1

1
G
kli2

2
,
n
i
0
= p
0

i
(
0
i1

1
(
0
i2

2
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 1001
for i, k, l 1, 2, 3. Here, the superscript i denotes the components of the vectors
a
kl
, b
kl
, and n
0
.
6.2. Calculation of A
1
E
and B
1
E
. The problem can be formulated as follows:
for every w R(/) R(B), nd u, v E such that /u = w and Bv = w. It is
enough to solve this problem for the operator /. The operator B can be treated
similarly. Let us consider the equation
(/+ B) u

= w. (6.1)
As follows from Proposition 4.13, the operator / + B is invertible in H for every
> 0. Thus, there exists a unique u

H that satises (6.1). Moreover, u

E
for every > 0 by denition of the subspace E. Equation (6.1) can be easily solved
numerically with nite elements. Let us show that u

is an approximation of a function
u E that satises the equation /u = w. Since u

E, we can rewrite (6.1) as

(/
E
+ B
E
) u

= w. Consequently,
u

= /
1
E
(w B
E
u

). (6.2)
Due to Theorem 4.17, the operator /
1
E
: R(/) R(B) E is bounded. Therefore,
there exists an independent of constant C such that
|u

| C(1 + |u

|).
This means that the sequence u

is weakly compact in H. That is, there exist

u H and a subsequence u

such that u

u weakly in H as 0. Since E
is weakly closed, u E. The passage to the limit in (6.1) yields the desired relation
/u = w. Moreover, the whole sequence u

converges to u in H. In reality,
u

u = /
1
E
B
E
u

, which implies
|u

u| C. (6.3)
Thus, the order of the approximation is obtained.
6.3. Calculation of e
tA
1
E
B
E
A
1
E
B
E
. Problem: For every w E and all
t > 0, nd u(t) = e
tA
1
E
B
E
/
1
E
B
E
w. Let us consider the following equation:
u
t
+/
1
E
B
E
u = 0, u(0) = /
1
E
B
E
w. (6.4)
It is obvious that u(t) = e
tA
1
E
B
E
/
1
E
B
E
w E is a unique solution of this equation
(see section 4.3). Let us construct an approximate solution to problem (6.4) using the
semidiscretization method. Fix t and introduce = t/N, N N. Dene functions
u
n
, n = 1, 2, . . . , N, as solutions of the following problem:
(/+ B)u
n
= /u
n1
, u
0
= /
1
E
B
E
w. (6.5)
That is,
u
n
= (/+ B)
1
/u
n1
.
Note that u
n
E for all n 0 because u
0
E and u
n1
E implies u
n
E for
any n > 0. Therefore, the operators / and B can be replaced by /
E
and B
E
in (6.5).
1002 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
To estimate u
N
u(t) we prove rst that there exists a constant C such that
|u
n
| C for all n. Indeed,
u
n
= u
0
/
1
E
B
E
n

k=1
u
k
,
which implies due to the boundness of /
1
E
and B
1
E
(see Theorem 4.17), the existence
of a constant C such that
|u
n
| |u
0
| + C
n

k=1
|u
k
|.
The Gronwall inequality now implies the required estimate. The boundness of |u
n
|
and (6.5) provide the following estimate:
|u
n
u
n1
| C |/
1
E
B
E
||u
n
| C. (6.6)
Let us introduce two time interpolations of u
n
,
u

(s) = u
n
_
1 n +
s

_
+u
n1
_
n
s

_
as s [(n 1), n],
u

(s) = u
n
as s ((n 1), n].
Due to (6.6),
_
t
0
| u

(s) u

(s)| ds =
N

n=1
|u
n
u
n1
|
_
n
(n1)
_
n
s

_
ds
=

2
N

n=1
|u
n
u
n1
| C.
Moreover, we have
u

t
+/
1
E
B
E
u

= 0.
Therefore,
|u
N
u(t)| = | u

(t) u(t)| |/
1
E
B
E
|
_
t
0
| u

(s) u(s)| ds
C
_
t
0
_
| u

(s) u(s)| +| u

(s) u

(s)|
_
ds C
_
+
_
t
0
| u

(s) u(s)| ds
_
.
Finally, the Gronwall inequality yields
|u
N
u(t)| C.
6.4. Numerics. In this section we give some examples which demonstrate prop-
erties of the homogenized continuum for various values of . We consider the system
consisting of the water and an isotropic elastic material (polymer) with the following
properties:
Pu
x
=
f
1 div u + 2
f
T(u), Gu
x
=
s
1 div u + 2
s
T(u),

f
= 1.e3,
f
= 1.e3,

s
= 2.777778e + 9,
s
= 4.166667e + 9.
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 1003
Fig. 3. Structural cell = [0, 1] [0, 1].
0.8
0.6
0.4
0.2
0.2 0.4
a
b
0.6 0.8 1
0
0
1

Fig. 4. Dependence of |P P|/|P| (curve a) and |P|/|P| (curve b) on .

The constant which characterizes compressibility of the water is = 4.597696e 10.
We take the structural cell of the form shown in Figure 3.
First, we investigate properties of the tensor P. The graphics in Figure 4 present
the dependence of [P P[/[P[ and [P[/[P[ on , where [P[ =
_
ijkl
P
ijkl
P
ijkl
_
1/2
.
As one can see, lim
1
P = P and lim
0
P = 0.
0.7
0.5
0.3
0.1
0
0.9
a
b
c
0.2 0.4 0.6 0.8 1

Fig. 5. Dependence of |GG|/|G| (curve a), |GQ|/|G| (curve b), and |GR|/|G| (curve
c) on .
The dependence of the tensor G on is more complex. Let us introduce two
tensors R and Q having the following components:
R
ijkl
=
1

ij

kl
, Q
ijkl
=
_
G
ijkl
if i = j and k = l,
0 otherwise.
The curves in Figure 5 show the dependence of [G G[/[G[, [G R[/[G[ and [G
Q[/[G[ on . One can see that G does not tend to G as 0 (see curve a). This
fact was already noted in section 5.2 after Corollary 5.22. It is not surprising that
1004 K.-H. HOFFMANN, N. D. BOTKIN, AND V. N. STAROVOITOV
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0
0 2e-011 4e-011 6e-011 8e-011 1e-010

Fig. 6. Graphics of the function |(t)|/|G| for various values of . The time unit is 1 second.
lim
1
G ,= 0. It can be explained by the presence of the tensor R in the denition
of G. Really, curve c in Figure 5 shows that lim
1
(GR) = 0. Curve b shows that
lim
0
G = Q. This means that, in the limit case as 0 (the elastic continuum
occupies the whole domain
h
), the tensor G can be obtained from the tensor G by
vanishing all of the nondiagonal components.
Thus, the limits of (5.6) as 0 and 1 look as follows:

S
u
t
div
t
Qu
x
div
_
t
0
(t s)u
x
(s) ds =
S
f, ( 0), (6.7)

F
u
t
div Pu
x

1
div u div
_
t
0
(t s)u
x
(s) ds =
F
f, ( 1). (6.8)
We take here the initial data being equal to zero.
In Figure 6, the graphs of the function [(t)[/[G[ are presented for several values
of . The function [(t)[ decreases very rapidly. In fact, [(t)[/[G[ vanishes practically
at the time t 10
10
s. Thus, the memory term in (5.6) is very small and can be
dropped in applications, if high frequency oscillations are not present.
7. Conclusions. Homogenization of a ne elastic structure immersed in a vis-
cous weakly compressible uid yields a continuum that possesses very interesting and
rather unexpected properties. Equation (5.6) describing the behavior of the resulting
continuum includes three basic terms. Two of them containing the tensors P and
G are related to stresses. The third integral-term represents a memory eect that
is responsible for viscoelastic properties of the resulting material. The presence of
such a memory is not surprising because similar results were already obtained by
other authors (see, for instance, [13]). More interesting from the mathematical and
mechanical viewpoints is the investigation of the above mentioned stress terms. The
term containing the tensor P describes a viscous damping and originates from the
uid part of the structure. Theorem 5.20 states the strict positiveness of P, which
implies the ellipticity of the corresponding dierential operator. The term containing
the tensor G represents elastic stresses. The tensor G is degenerate, its kernel is de-
scribed in Theorem 5.21. The theorem implies that volume conserving deformations
(shear deformations in particular) do not produce elastic stresses.
All of the coecients involved in the homogenized equation are found in an explicit
form. Although expressions representing them are rather complex, the coecients can
be computed numerically using the algorithms given in section 6. The numerical treat-
ment delivers another interesting properties of the homogenized model. It is stated
HOMOGENIZATION OF FLUID/ELASTIC INTERFACES 1005
numerically that the memory eect is very weak. The system forgets the current
history in a very short time. Therefore, the memory term can be dropped in most
of the applications. Another interesting question is the dependence of properties of
the homogenized continuum on the parameter which represents the volume fracture
of the uid so that the pure uid corresponds to = 1. As was expected, the ho-
mogenized equations coincide in the limit ( 1) with the ones being used for the
description of the original uid. In the opposite limiting case ( 0) the homoge-
nized equations dier from the model of the original elastic continuum. In particular,
the limiting elastic continuum can be nonisotropic even though the original material
is isotropic. Thus, the limiting continuum inherits certain geometric properties of
the ne elastic structure even if the uid vanishes and the solid occupies the whole
volume.
REFERENCES
[1] G. Allaire, Homogenization and two-scale convergence, SIAM J. Math. Anal., 23 (1992),
pp. 14821518.
[2] L. Baffico, C. Conca, A mixing procedure of two viscous uids using some homogenization
tools, Comput. Methods Appl. Mech. Engrg. 190 (2001), pp. 42454257.
[3] N. Botkin, M. Schlensog, M. Tewes, and V. Turova, A mathematical model of a biosen-
sor, in Technical Proceedings of the Fourth International Conference on Modeling and
Simulation of Microsystems, Hilton Head Island, SC, 2001, 231234.
[4] G. P. Galdi, An Introduction to the Mathematical Theory of the Navier-Stokes equations,
Vol. 1. Linearized Steady Problems, Springer Tracts in Natural Philosophy 38, Springer-
Verlag, New York, 1994.
[5] K.-H. Hoffmann and N. D. Botkin, Homogenization of von K arm an plates excited by piezo-
electric patches, ZAMM Z. Angew. Math. Mech., 80 (2000), pp. 579590.
[6] K.-H. Hoffmann and V. N. Starovoitov, On a motion of a solid body in a viscous uid.
Two-dimensional case, Adv. Math. Sci. Appl., 9 (1999), pp. 633648.
[7] U. Hornung, Homogenization and Porous Media, Springer-Verlag, New York, 1997.
[8] W. J ager and A. Mikeli c, On the roughness-induced eective boundary conditions for an
incompressible viscous ow, J. Dierential Equations, 170 (2001), pp. 96122.
[9] L. Landau and E. Lifshits., Lehrbuch der theoretischen Physik, Band VI: Hydrodynamik, 5th
ed., Akademie-Verlag, Berlin, 1991.
[10] J.-L. Lions, Quelques methodes de resolution des problemes aux limites non lineaires, Etudes
mathematiques, Gauthier-Villars, Paris, 1969.
[11] D. Lukkassen, G. Nguetseng, and P. Wall, Twoscale convergence, Int. J. Pure Appl.
Math., 2 (2002), pp. 3586.
[12] G. Nguetseng, A general convergence result for a functional related to the theory of homoge-
nization, SIAM J. Math. Anal., 20 (1989), pp. 608623.
[13] J. Sanchez-Hubert, Asymptotic study of the macroscopic behaviour of a solid-uid mixture,
Math. Methods Appl. Sci., 2 (1980), pp. 111.
[14] R. Temam, Probl`emes mathematiques en plasticite, Methodes Mathematiques de lInformatique
12, Gauthier-Villars, Montrouge, 1983.