HOMOGENIZATION OF BIHARMONIC EQUATIONS IN DOMAINS PERFORATED WITH TINY HOLES

CHAOCHENG HUANG DEPARTMENT OF MATHEMATICS AND STATISTICS WRIGHT STATE UNIVERSITY DAYTON, OHIO 45435, USA

Abstract. This paper deals with the homogenization of the biharmonic equation

2u

= f in

a domain containing randomly distributed tiny holes, with the Dirichlet boundary conditions. The size of the holes is assumed to be much smaller compared to the average distance " between any two adjacent holes. We prove that as " the solution of
2u

! 0, the solutions the biharmonic equation converges to

+ u = f where depends on the shape of the holes and relative order of

with respect to ".

1. Introduction. Let be a bounded domain in RN . For any " > 0

" denotes

a perforated domain formed by randomly removing many tiny holes (depending on ") from . Consider the biharmonic equation (1.1)
2

u" = f in

"

with the Dirichlet boundary conditions (1.2) (1.3)

" "\ : u" = @u = 0 in @ @n " " u = g @u @n = h in @

We are concerned with the limiting behavior of the solution u" of problem (1.1){(1.3) as " ! 0. Many authors studied homogenization of the biharmonic equations (see 3], 10] and the references cited there). All the previous results were established for the case that the holes are distributed periodically and that the size of the holes is of the
1

same order as that of the period. It was shown (cf: 10]) that as " ! 0, the solution u" converges to 0 to order "4 the higher order asymptotic expansion was also studied in 10]. In this paper, we are interested in the case when the holes are distributed randomly and when the volume fraction of the holes approaches zero as the number increases to in nity. Our objective is to determine the order of the volume fraction of the holes in order to obtain a non-zero limit of the solution u" of problem (1.1){(1.3) as the number of the holes approaches 1. Let " be the \average" distance between two adjacent holes (the precise de nition will be stated in the next section), (depending on ") the size of the holes. We shall consider the case that " ! 0 as " ! 0: The main results are the follows. (i) If N 3 then the solution u" of (1.1){(1.3) always converges to 0, no matter how small the volume fraction of the holes is (ii) if N > 3 and the limit " ! exists and is nite, where
" "

= "4 ln 1 for N = 4

N ;4 "N

for N

then the solution u" converges to the solution of

2

u + u = f in

@u = @g in @ @n @n where 0 depends on the shape and distribution of the holes (iii) if N > 3 and = 1 g = 0 then the limit of the solution u" is 0. ;4 for N > 4 The results show that the critical order of the size of holes is = " NN 1 for N = 4: and = exp ; " 4 u=g
2

For the analogous problems of the second order equations with periodically distributed holes, the case when the size of the holes is much smaller than the period has been studied extensively in literatures. In 6], 11] and 8] the authors considered the problem for the Laplace equation. The homogenization for the wave equation in that case was studied in 5]. The problem for the Navier-Stokes equation was analyzed in 1], 2]. These papers showed that the critical order for the second order equations is 1 for N = 2: For homogenization of other ;2 for N > 2 and = exp ; = " NN "2 equations and general references, we refer to 1], 2], 3], 6], 7], 10], 11]. The proof of our results is based on the energy method. The main idea follows those in 1], 6] and 8]. Since the di erent natures between the fourth order equations and the second order equations, the treatment here technically is di erent from the papers mentioned above. In contrast with 1] and 6] where certain periodic structures are assumed, we consider the more general situation that holes are randomly distributed with a density function. Our paper is also the rst approach to explore the relationship between the orders of the equations and the critical orders of the holes that will produce non-trivial homogenized equations. The paper is organized as follows. In x2, we shall state the problem more precisely and introduce some notations. The case when N 3 is also studied in that section. In x3 we study the problem in a ball containing one hole and derive some estimates. Section 4 is devoted to the study of the homogenized equations and the results of convergence under various norms for N > 4. The case N = 4 is treated in x5.

2. Problem Setting and Preliminaries. Let and T be two bounded C 1 1

domains in RN locally lying in one side of their boundaries. We assume that (A1) B1 T B2
3

where Br = Br (o) the ball centered at origin with radius r. For any " > > 0 and a set of nite points fx" such that i : i = 1 : : : m" g " (A2) x" 2" for i 6= j: i ; xj We de ne the holes Ti" by (A3) Ti" = x" i + T: Set
"

= n

i2 "

Ti"

where " = fi : Ti" g: This means that we exclude the situation that the holes intersect with the boundary @ . We point out that this condition for " is set up only for simplicity. All the results can be extended to the case that the holes touching the boundary. We also assume that there exists a bounded density function P (x) > 0 such that m " N P (A4) " (x" ) ;! RR P (x) (x) dx i i=1 0 for any 2 C ( ) the set of all continuous functions with compact supports in . Note that under the above assumptions, the number of the holes m" is approximately c";N for some constants c. One example that satis es the above assumptions is the case that the holes are distributed periodically. In that case, we may take

x" i = ("k1 "k2

and P (x) is a constant. Consider the Dirichlet problem (2.1)
2

"kN ) kj 2 Z

u" = f in
4

"

(2.2) (2.3)

" u" = @u on @ " \ @n = 0 " @g on @ = u" = g @u @n @n

where n is the exterior normal to the boundary. To de ne a weak solution of (2.1){ (2.3), we need to introduce some function spaces. Let H m ( ) be the usual Sobolev space of all functions that have square-integrable derivatives up to mth order, and m ( ) the set of all functions that belong to H m ( 0 ) for any 0 such that 0 Hloc : m ( ) the closure of C 1 ( ), The norm in H m( ) is denoted by jj jjH m( ). Denote by H0 0 the space of all smooth functions with compact supports in . Let

H 2(
with the norm

) = fv 2 H 1( ) :

v 2 L2( )g

jjv jjH 2(

= jjvjjH 1( ) + jj vjjL2( ):

We know that H 2( ) is a Hilbert space and that the closure of C01( ) with the norm jj jjH 2( ) coincides with H02( ). For any boundary portion ; @ denote by H0m ( ;) the jj jjH 2( ) ;closure of all smooth functions that have the compact supports in ;. A weak solution of (2.1){(2.3) is de ned as a function u" 2 H02( " @ " \ ) such that u" ; g 2 H02( " @ ) and ZZ ZZ " (2.4) u vdx = fvdx
" "
2 ( " ). for any v 2 H0 Lemma 2.1. Under the assumption (A1)-(A4), if f 2 L2 ( ) and g 2 H 2 ( ), then there exists a unique weak solution u" of problem (2.1){(2.3). Furthermore, for any subdomain 0 the following estimates hold with a constant C ( 0 ) 0

independent of " (but depending on

)
)

(2.5)

jju" jjH 2( " \

C ( 0) jjf jjL2( ):

When g = 0 (2.5) holds for 0 = : Proof: Existence and uniqueness are well-known. Assume rst that g = 0: Taking v = u" in (2.4), we obtain
jj u"jj2 L2 (
")

jjf jjL2 ( ) jju" jjL2 ( " ) :

Using the relation

ZZ
"

jj 5 u" jj2dx = ;

ZZ
"

u" u"dx

it follows that
jj 5 u" jj4 L2 (
")

jju" jj3 L2 ( " ) jjf jjL2 ( ) :

Inequality (2.5) then follows from the Poincare's inequality. For the general case that g 6 0, we choose a cut-o function such that = 0 in a neighborhood of @ and = 1 in 0: The desired estimates follow by choosing the test function v = 2u" in (2.4). Note that since the weak solution u" 2 H02( " @ ") the extension ( " (x) for x 2 " " u ^ (x) = u 0 (2.6) for x 2 Ti" is in H 2( ): By the Lp;estimates of the harmonic equations, we obtain Corollary 2.1. Let u" be the solution of problem (2.1){(2.3). Then for any
0 0

(2.7)

jju ^"jjH 2(

0)

= jju"jjH 2(

"\

0)

c ( 0) jjf jjL2( ):

If g = 0 then the inequality holds for 0 = : It follows that there exists a subsequence of fu ^"(x)g">0 that converges weakly in H 2 ( 0) to a function, say u. Another direct consequence of (2.7) is the following result. Corollary 2.2. Let N 3 and u ^" be the extended weak solution de ned in (2.6). Then as " ! 0 u ^" ! 0 uniformly in any 0 0 : If in addition, g = 0 then the convergence is uniform in : Proof: By (2.7) and the Sobolev embedding, it follows that u ^" is ; H older continuous for some > 0, and that its ; H older norm is uniformly bounded. Hence

(2.8)

ju ^"(x)j = ju ^"(x) ; u ^"(x" i )j c"

S" B (x"). Then " be the characteristic function of m for any x 2 B"(x" " i i ) \ 0 : Let i=1 (2.8) means that ju ^" "(x)j c" for any x 2 0: Since " ! P weakly in L2 we obtain u ^" " ! uP . Therefore u 0 since P > 0. In sequel, we shall consider the case N 4.

3. Local Problems. Let T be a domain satisfying (A1). For any

the following problem: (3.1) (3.2) (3.3)

3, consider

W = 0 in B n T W = @W @n = 0 on @T W = 1 @W @n = 0 on @B :
2

By Lemma 2.1, there exists a unique weak solution to the above problem. We extend this weak solution by 0 in T: Denote by W the extension of this solution. From (3.2),
7

it is easy to see that W 2 H 2(B ): Note that we are interested in W only for large : We begin with the estimates for H 2 ; norm. Lemma 3.1. Let W be the extension of the solution of (3.1)-(3.3). Then there exists a constant c > 0 independent of 3 such that the following estimates hold for W : (i) jj 52 W jjL2 (B ) c (ii) jj 5 W jjL2 (B ) c (iii) jjW ; 1jjL2 (B ) c 2 : Proof: For simplicity, we take 5: Let 2 C 1 RN such that = 0 in B2, = 1 in RN n B3 0 1: By taking v = W ; in (2.4), we obtain ZZ ZZ j W j2 dx = W dx:
B nT B nT

It follows that

ZZ
B nT

j W j2dx

ZZ

B nT

j j2 dx = c

ZZ

B3 nB2

j j2 dx

c:

By rescaling, we set V (x) = W ( x) for x in B1: The above estimate implies that ZZ ZZ j V j2dx = 4;N j W j2 dx c 4;N :
B1 B nT

Since V = 1 on @B1 by L2 ; Estimates for the second order elliptic equations, we obtain that
kV ; 1k2 H 2(B1 )

4; N

The assertions follow by rescaling back. We next study the behavior of W near @B : By the regularity theory for the higher order elliptic equations 9], we know that W is smooth up to the boundary
8

portion @B (this is also a consequence of (3.14) later in this section). We shall frequently use the following particular form of Rayleigh-Green's identity ZZ Z @v ds ; Z @ W vds (3.4) W vdx = W @n @n @B @B B nT

@v = 0 on @T . To show (3.4), we take a for any function v 2 H 2(B nT ) such that v = @n smooth function such that = 0 in B2, = 1 outside B3. Then v ; v 2 H02(B n T ). Hence by (2.4) we obtain ZZ ZZ ZZ W vdx = W (v )dx = W (v )dx:
B nT B nT B nB2

From the Rayleigh-Green's identity ZZ ZZ Z @v ~ ds ; Z @ w v (3.5) ~ ds + w v ~dx = w @n @n @ 1 @ 1 1 1

w v ~dx

that holds for any function w 2 H 4( 1) v ~ 2 H 2( 1) where 1 is any Lipschitz domain, the identity (3.4) follows by taking w = W v ~ = v and 1 = B n B2: To proceed studying the behavior of the function W near the boundary @B we need the Green's function GN for the biharmonic equation in B with the Dirichlet boundary condition. For N 5 the Green's function is 1 jx ; y j4;N ; ( ;1 jy jjx ; y j)4;N GN (x y) = 2(4 ; N )(2 ; N )!N !2;N (4 ; N ) 2 2 2 2 jy j (3.6) + 2! 2 (jxj ; )(jyj ; ) jx ; yj : N For N = 4 we have

(3.7)

G4(x y) = ; 8!1 2 (jxj2 ; 2)(jyj2 ; 2) jyj jx ; yj 4 ! 1 jy jjx ; yj + 4! ln jx ; yj : 4

9

!;2

Here in (3.6) and (3.7), y = 2jyj;2y is the re ection of y with respect to B , and !N is the surface area of the unit sphere in RN : Note that the Green's function GN (x y) satis es the Dirichlet boundary condition
N (x y ) GN (x y) = @G @n(x) = 0 x 2 @B

for any y 2 B . By the Rayleigh-Green's identity (3.5), we can represent any smooth function w that has the compact support as the integral ZZ (3.8) w(x) = GN (x y) 2w(y)dy:
B

4 5 and W be the solution of (3.1)-(3.3). Then the following estimates hold for any x 2 B n (B4 B 2 ): (i) there exists a constant c independent of such that (3.9)
j W (x)j
N ;2

Lemma 3.2. Let N

(ii) there exists a constant c independent of and a positive function ( ) that decreases in such that

(3.10)
where

5( W )(x) +

(T ) x !N jxjN

() + c jxjN ;1 jxjN

(3.11)

(T ) =

ZZ

B nT

j W (y )j2dx:

1 such that = 0 in B2 = 1 in RN n B3: Set v = (W ; 1) : Then v 2 C 1(B ) \ H02, since W (x) is smooth for x 2 B n T , and v satis es (3.12)
2

Proof: Choose a cut-o function 0

v = F in B
10

where

F = 2 W
(3.13) +(W ; 1)

+ 4 5 ( W ) 5 + 4 5 W 5( )
2

+ 4 trace 52W 52 :

Since 5 (x) = 0 for x 2 = B3 n B2 it is easy to see that the function F is supported only in B3 n B2: By the interior estimates for the higher order elliptic equations, Lemma 3.1, and by the Sobolev embedding, we know that in B3 n B2 all the derivatives of W up to the third order are bounded by a constant independent of . Therefore, jF (x)j c for any x. For convenience, here and in what follows, c is always reserved as a universal constant. Using the Green's representation formula (3.8), we can write ZZ v(x) = GN (x y)F (y)dy:
B

In particular, for x 2 B n B3 the above equality reads as ZZ (3.14) W (x) = 1 + GN (x y)F (y)dy
B3 nB2

since supp F (3.15) (3.16)

B3 n B2. Hence, for x 2 B n B3 ZZ W (x) = xGN (x y )F (y )dy

B3 nB2

5( W )(x) =

ZZ

B3 nB2

5x( x GN (x y ))F (y )dy:

From the explicit formulas (3.6) and (3.7), we can compute GN directly. It follows that for N 4
x GN (x

jx ; y j2;N y) = ! (2 ; N ) + 1(x y)
N

11

where
1

4;N 2;N 2 2 2;N j y j ( j y j ; ) j y j j x ; y j + ! 2 jx ; y j;N (xi ; yi )yi = ; ! (2 ; N ) N N ! (4 ; N )(jy j2 ; 2 ) 1 jy j 2;N + 2! (2 ; N ) 2 jx ; y j2;N N

and the convention of summation of the repeated indices is applied. For 2 3 max( 2 4) jxj we can bound xGN by (3.17) since
jx ; y j = x ;
2

jy j

j x GN (x y )j

N ;2

jy j2

c 2:

The assertion (i) follows from (3.15) and (3.17). Next, by direct computation, we obtain
j 5x 1 (x y )j

for 2 jyj 3 max( 4) jxj 2 F , we deduce that

5( W )(x) =

: Therefore, from (3.16) and the boundedness of

(3.18)

1 ZZ x ; y F (y)dy + (x y) 2 !N B nB jx ; yjN 3 2 ZZ 1 x = ! jxjN F (y)dy + 2(x y) N B3 nB2 ! ZZ 1 x ;y x ; F (y)dy +! N B nB jx ; y jN jxjN

3 2

where
j 2(x y )j

ZZ
B3nB2

j 5 1(x y )F (y )jdy
12

for 2 jyj 3 max( 2 4) jxj : The third term in the last equality of (3.18) can be estimated by ! 1 ZZ x ; y ; x F (y)dy () !N B nB jx ; yjN jxjN jxjN
3 2

where we used the fact that W = 1 on @B : It follows from the Rayleigh-Green's W = 0 on @B identity (3.4) that, noting that @ @n ZZ ZZ F (y)dy = ; j W (y)j2dy = ; (T ):
B3 nB 2 B nT

where ( ) ! 0 as ! 1. Finally, from (3.12), for 2 jyj 3 max( 2 4) jxj we obtain ZZ ZZ Z @ v 2 F (y)dy = v(y)dy = ds @n B @B B3 nB2 Z @ W Z @ W = @n ds = @B @n W ds @B

The proof is complete. From Lemma 3.1, (T ) (de ned in (3.11)) is uniformly bounded. We next show that the limit of (T ) exists as ! 1. Lemma 3.3. Let N 5. Then (T ) ! (T ) as ! 1 where 0 < (T ) < 1 depends only on the domain T . Proof: Let W be the solution of (3.1)-(3.3) for large . By (ii) of Lemma 3.1, @ W = 0 by the Poincare inequality we jj 52 W jjL2 (B ) c: Since on @T W = @n obtain that for any 0 < r <
jjW jjH 2 (Br )

C (r)

where C (r) is independent of (but depends on r). By the diagonal argument, we 2 (RN n T ) such that may select a subsequence k ! 1 and a function W 2 Hloc
13

W k * W as

!1

where the convergence is interpreted in the sense that for any r > 0, W k weakly converges to W in H 2(Br n T ). Since W is biharmonic in B n T one can see that W = 0 on @T , and 52W is W is also biharmonic outside T . Furthermore, W = @ @n bounded in L2(RN n T ) since jj 52 W jjL2(B ) c. @ = 0 on @T = 1 outside B by 2 Now for any 2 Hloc (RN n T ) such that = @n r the de nition (2.4), we have, for k > r ZZ ZZ j W k (x)j2dx = Wk :
B k nT k B k nT

It follows that for large

k (T ) =

ZZ
B k nT

k!1 ; ;;;; !

Wk ZZ
Br nT

dx = W

ZZ
=

Br nT

Wk ZZ

RN nT

We now take, for any xed l ( (x) = W l1(x)

if x 2 B l n T if x 2 RN n B l :

2 : Hence Then by the \matching Lemma", 2 Hloc ZZ ( T ) ! 1 k k ; ;;;; ! N W W l dx R nT ZZ j W j2dx = (T ): l!1 ;;;;!

RN nT

It remains to show that (T ) > 0 and that the whole sequence (T ) ! (T ). To this end, it su ces to show that the weak limit W is unique.
14

We observe that the Green's functions (3.6), (3.7) and their gradients satisfy the following estimates for x 2 = B4 y 2 B3 n B2:
jGN (x y )j j5xGN (x y )j jxjN ;4

jxjN ;3

+ Nc;4

+ Nc;3

for N

5: It follows from the expressions (3.14) that for x 2 = B3

jW (x) ; 1j j5W (x)j jxjN ;4 jxjN ;3

+ Nc;4 + c
N ;3

for N 5: Since W k ! W strongly in H 1(Br n T ) for any r it follows that for almost all x 2 = B3 (3.19) (3.20)
jW (x) ; 1j j 5 W (x)j jxjN ;4

c : jxjN ;3

We now claim that W is the only biharmonic function in RN n T that satis es ^ W = @W @n = 0 on @T and (3.19), (3.20). Otherwise, suppose that W is another ^ satis es biharmonic function having the above properties. Then V = W ; W (3.21)
jV (x)j jxjN ;4

j 5 V (x)j

jxjN ;3

for large x: Then for any smooth function that has the compact support in RN ZZ V (V 4) = 0:
RN nT

It follows that

ZZ
RN nT

j Vj

2 4

ZZ

RN nT

(jV

j + j 5 V 5 4 j)j V jdx:

15

Therefore, by the Cauchy's inequality, ZZ ZZ j V j2 2 dx c (jV jj 52 j + jV jj 5 j2 + j 5 V jj 5 jj j)2 dx:

RN nT RN nT

We choose such that 0

j5 j

1 = 1 in Br = 0 in RN n B2r c 52 c: r r2

From (3.21), it follows that ZZ

Br nT

j V j2dx

ZZ
RN nT

j V j2 2 dx

rN ;4 :

Letting r ! 1 we obtain V = 0 in RN n T: Consequently, V = 0 since V = 0 on the boundary. Hence the whole sequence fW g converges to W (in the sense we explained before). It immediately follows that (T ) ! (T ): Finally, if (T ) = 0 then W = 0. Since W = @W = 0 on @T we can extend it by 0 into T . The @n extended function is harmonic in RN , and is bounded by (3.19). Hence W constant. A contradiction to (3.19) and the fact W = 0 on @T:

4. Homogenized Problem I: N 5. Let u" be the solution of (2.1)-(2.3), and

u be a weak limit of a subsequence of u ^", the extension of u" by 0 in the holes Ti": In this section we shall study the homogenized problem for N 5. Set
"

= "N : 5 g 2 C 2, and that the limit

N ;4

N ;4

Theorem 4.1. Assume (A1)-(A4), N

(4.1)

= lim "!0 "N

16

exists. If < 1 then the whole sequence u ^" converges weakly to u in H 2( ): Furthermore, u satis es

(4.2) (4.3)

u + (T )Pu = f in @u = @g on @ u = g @n @n

If = 1 then for any subdomain 0 the whole sequence u ^" converges to 0 0 weakly in H 2( 0): To prove the theorem, we need to construct a test function for any " > 0. Let

W be the solution of (3.1)-(3.3). Under the assumption (A2), the balls fB" (x" i )g are disjoint. De ne a function w" in by 8 " > " > W x ; xi for x 2 B" (x" < i ) n Ti " (4.4) w (x) = > 0 for x 2 Ti" > : 1 otherwise, where = " ! 1 as " ! 1. One can see easily that w" 2 H 2( ).
Lemma 4.1. Let N

5,

1 and w" be the function de ned in (4.4). Then kw" ; 1k2 L2 k5w" k2 L2 52w"
2

(4.5) (4.6) (4.7)

c ""4 c ""2
L2

c ": c";N
i=1 " "4

Consequently, w" ! 1 weakly in H 2( ). Proof: By Lemma 3.1, we obtain, noting that jm"j
kw" ; 1k2 L2
m" ZZ X i=1 B"(x")nT " i i N cm" kW

m" 2 " X W x ; xi ; 1 dx + jTi"j

(4.8)

; 1k2 L2 (B ) + cm"
17

k5w" k2 L2

m" 1 X i=1

ZZ

(4.9) and
k w" k2 L2

= m"

" B"(x" i )nTi N ;2 k5W

5W k2 L2 (B )

x ; x" i c " "2

dx

m" 1 X i=1

ZZ

(4.10)

= m"

" B" (x" i )nTi N ;4 k W

" W x ; xi

dx

k2 L2 (B )

c ":

The assertions (4.5) and (4.6) follow immediately. Again from (4.8)-(4.10), we have the estimate kw" ; 1k2 c " : The assertion (4.7) follows from the equivalence 2( H0 ) 2 of two norms k kH02( ) and k kH02( ) since (w" ; 1) 2 H0 ( ): Proof of Theorem 4.1: Consider rst the case that < 1. In this case we claim that (as claimed in Lemma 2.1 in the case g = 0) ku ^"kH 2( ) is uniformly bounded, and (4.11)
ku" kH 2 (
")

c kf kL2( ) + kgkc2 :

Indeed, we choose the function v = u" ; w" g: It is easy to verify that v is eligible as a test function in (2.4). It follows from (2.4) that ZZ ZZ ZZ 2 " " " j u j dx = u (w g) dx + f (u" ; w"g) dx:
" " "

By Lemma 4.1, we obtain (4.12)

k u"k2 L2 (
")

c kf kL2( ) + kgkc2 ku"kL2( ") :

Next, from the identity

ZZ
"

v vdx = ;
18

ZZ
"

jrv j2 dx

and using (4.12), we have (4.13)

kru"k4 L2 (
")
4 c kf kL2( ) + kgkc2 ku"k3 L2 ( " ) + c kg kc2 :

Since u ^" ; g 2 H02( ) by applying Poincare's inequality to u ^" ; g we derive from (4.12) and (4.13) that
kru"kL2 (
") + k

u"kL2(

")

c kf kL2( ) + kgkc2 :

(4.11) follows from the equivalence between H02 ( )-norm and H02 ( )-norm for u ^" ; g: By (4.11), we know that there exists a subsequence of u ^" that weakly converges to a function u: It now su ces to show that any weak limit u satis es (4.2). By (2.4), for any 2 C01( ), we have ZZ ZZ u ^" (w" )dx = fw" dx: This equality can be rewritten as ZZ ZZ w" u ^" dx + (^ u" ) (4.14) where =

w"dx =

ZZ

fw" dx +

ZZ

(^ u w"

+ 2 w" 5 u ^ 5 ;2 u ^" 5 w" 5 ) dx:

Since u ^" is uniformly bounded in H 2( ) and w" ! 1 weakly in H 2( ) we know that ! 0 as " ! 0, and that ZZ ZZ " " (4.15) w u ^ dx ;! u dx (4.16)

ZZ

fw"

dx ;!
19

ZZ

f dx:

It remains to study the limit of the second term on the left-hand side of (4.14). We now apply the identity (3.4) with v = u ^" . It follows that ZZ (^ u" ) w"dx
" B"(x" i )nTi

(4.17)

For any x 2 @B (x" i ) by Lemma 3.2 we have 1 W x ; x" i (4.18) j w" (x)j =
2

@B"(x" i)

u w" @ (^

@n

"

) ds ; Z

@B"(x" i)

@ w" u " ds: ^ @n c " "2

where = " . Next we rescaling u ^" by setting v(x) = (^ u" )(x" i + "x) for x 2 B1 : By applying the trace inequality Z ZZ 2 2 j5v j ds c j5v j2 + 52v dx we derive that (4.19)
@B1 B1

Combining (4.18) and (4.19), the rst term on the right-hand side of (4.17) can be estimated by Z u" ) ds c"2 " Z @ (^ u" ) ds w" @ (^ @n @n @B"(x" @B"(x" i) i) 1 0 12 2 Z " @ (^ u ) dsC c" N2+3 " B (4.20) @ A @n
1 0 12 ZZ 2 c" N2+2 " B j5(^ u" )j2 + "2 52(^ u" ) dxC @ A " 1 0 B"(xi ) ZZ 2 j5(^ u" )j2 + "2 52(^ u" ) dxC c" " B A: @"N +

@B"(x" i)

j5(^ u" )j2 ds

c ZZ j5(^ " )j2 + "2 52(^ " ) 2 dx: u u " B"(x")

i

@B"(x" i)

B" (x" i)

20

We now evaluate the second integral on the right-hand side of (4.17). By (ii) of Lemma 3.2, for x 2 @B"(x" i)
r( w")(x) =

1 5 ( W ) x ; x" i 3 " ( T ) N ; 4 x ; xi = ; ! N + jx ; x" N ij

where
j 1j

() 1 N ;1 + N

c"

"

! 1 ( )+ :

It follows that (4.21) where

Z
@B"(x" i)

Z @ w" u ( T ) " " ^" ds + 2(i) @n ^ ds = ; !N "@B"(x") u

i

j 2(i)j

Z
@B"(x" i)

j 1u ^" j ds

(4.22)

""

( )+ 1

! Z
@B"(x" i)

ju ^" j ds:

Combining (4.17), (4.20)-(4.22), we derive that ZZ ZZ m" X (^ u" ) w"dx = (4.23) where
j 3j

i=1 B"(x")nT " i i m" Z (T ) " " X " u ^ !N i=1 @B"(x")

i

(^ u" )

w" dx ds +
3

0 1 ZZ m" X 2 "B j5(^ u" )j2 + "2 52(^ u" ) dxC @ "N + A + 2(i)
B" (x" i)

(4.24)

c" " ku ^" kH 2 +

m" X i=1

i=1

(i):
21

By assumption (A4), for any smooth function , Z m" Z m" m" X X X " ds = !N "N (x" ) + " ( (x) ; (x" i i )) ds i=1@B"(x") i=1 i=1 @B"(x") i i ZZ ;! !N P (x) (x)dx (as " ;! 0) : Approximating u ^" by smooth functions, we thus obtain ZZ m" Z X " (4.25) u ^ ds ;! !N Pu dx (as " ;! 0) : "
i=1 @B"(x" ) i

From (4.22), (4.25) also implies that (4.26)

m" X i=1

! 1 c ( )+ : 2 (i)

It follows from (4.23)-(4.26) and Lemma 3.3 that ZZ ZZ (4.27) (^ u" ) w" dx ;! (T ) Pu dx: Letting " ! 0 in (4.14), we obtain from (4.15), (4.16), and (4.27) that ZZ ZZ ZZ u dx + (T ) Pu dx = f for any 2 C01( ): This is the weak formulation of (4.2), (4.3). It remains to prove the case that = 1: In this case, we de ne

w ~" (x) = 1" w"(x)

where w" is de ned in (4.4). It follows from (4.8)-(4.10) that
kw ~"kL2 + k5w ~"kL2 ;! 0 (as " ;! 0) 52w ~"
L2

22

c:

We note that when g 6 0 u ^" in general is not uniformly bounded in H 2 ( ) : By Lemma 2.1, however, it is locally uniformly bounded. Let 0 is any 0 subdomain of and 2 C01 ( 0) : By replacing w" with w ~" in (4.14), it follows that ZZ (^ u" ) w ~" dx ;! 0: Since u ^" is uniformly bounded in H 2 ( 0) there exists a subsequence that converges weakly (in H 2 ( 0)) to a function u 2 H 2 ( 0) : Moreover, the estimates (4.17)-(4.26) remain true except that the constant c must be replaced by c ( 0) which depends on 0 : Therefore we can evaluate the above integral by the same methods as in deriving (4.27). By modifying (4.17)-(4.26), we obtain ZZ ZZ " " (^ u ) w ~ dx ;! (T ) Pu dx: Hence

ZZ

Pu dx = 0

for any 2 C01( 0): It follows that u = 0 since P > 0 and 0 can be chosen arbitrarily: The proof is complete. From the above results, we know that u ^" ; u ! 0 weakly in H 2 if is nite. One may ask if the convergence is also strong. In general, it cannot be strongly convergent. But if we add a corrector (1 ; w" )u the convergence is strong. Theorem 4.2. Let N 5 g 2 C 2, and u be the solution of (4.2), (4.3). Suppose that < 1, where is de ned in (4.1). Then ZZ p (4.28) 52 (^ u" ; w" u) dx ;! 0
where p =

N : N ;2

23

Proof: For any 2 C 1 ( ) such that = g on @ we write, by using the weak

formulation (2.4) for u" ZZ j (^ u" ; w" )j2 dx ZZ ZZ " " " (4.29) = u ^ (^ u ; w ) dx ; (w" ) (^ u" ; w" ) dx ZZ ZZ = f (^ u" ; w" ) dx ; (w" (^ u" ; w" )) dx ZZ ; w" (u (^ u" ; w" )) dx + where = (4.30)

ZZ

((^ u" ; w" ) w" + 2 w" 5 ZZ + (w" ; 1) (^ u" ; w" ) dx ZZ ; 2 (5w" 5 ) (^ u" ; w" ) dx:

5 (^ u" ; w" )) dx

Since w" ! 0, it is easy to see that ! 0 as " ! 0: By repeating the process of deriving (4.27), we obtain ZZ ZZ " " " (4.31) w ( (^ u ; w )) dx ;! (T ) P (u ; ) dx since (^ u" ; w" ) ! (u ; ) weakly in H 2: It follows from (4.29)-(4.31) that ZZ ZZ ZZ j (^ u" ; w" )j2 dx ;! f (u ; ) dx ; (u ; ) dx ZZ (4.32) ; (T ) P (u ; ) dx ZZ ZZ = j (u ; )j2 dx + (T ) P (u ; )2 dx: From the Sobolev imbedding, for any 2 C 1
k kLp1

c k kH 2
24

p1 = N2N ;4

k5 kLp2

c k kH 2 p2 = N2N : ;2 c

Therefore,

ZZ
+c

(4.33)

ZZ

j (w" (u ; ))jp dx

ZZ

j w" jp j ; ujp dx

(j5w"jp j5 ( ; u)jp + jw"jp j ( ; u)jp) dx

p p c k ; ukp Lpq + c k5 ( ; u)kLpq1 + c k ( ; u)kLpq2

where

N ; 2) q = 2 q = 2 (N ; 2) : q = 2(( 2 N ; 4) 1 N
We thus obtain, since pq = p1 pq1 = p2 pq2 = 2 ZZ ZZ ZZ p p " " " " j (^ u ; w u)j dx j (^ u ; w )j dx + j (w" (u ; ))jp dx
p 0ZZ 12 c @ j (^ u" ; w" )j2 dxA + c k ; ukp H2 :

Letting " ! 0 and using (4.32), we derive that ZZ lim j (^ u" ; w" u)jp c k ; ukp H2 : "!0 The assertion follows by letting ! u and the Lp ; Estimates for the second order elliptic equations. In the special case that u is a C 2 function (this is true, for example, if T f are regular enough), the Theorem 4.2 can be strengthened such that (4.28) holds for p = 2:
25

Theorem 4.3. Let N

< 1 g 2 C 2, and u be the C 2 solution of (4.2), c ku ^" ; w" ukH 1( ) + c"2 ;! 0:

= + (T ) ZZ

(4.3). Then
ku ^" ; w" uk2 H 2(
)

Proof: Taking = u in (4.29) and using the weak formulation for u we have
(4.34)

ZZ

(^ u" ; w" u)j2 dx

ZZ

uP (^ u" ; w"u) dx
(u (^ u" ; w"u)) dx + :

w"

It is easy to see that (de ned in (4.30)) can be estimated by 1 " 2 ^" ; w"ukH 1 + c kw" ; 1k2 (4.35) j j c ku u" ; w" u)k2 L2 + c k5w kL2 + 4 k (^ L2 : We now estimate the second integral in the right-hand side of (4.34). By (3.4), " in each B"(x" i )nTi we have ZZ (4.36) w" (u (^ u" ; w" u)) dx
" B"(x" i )nTi

By (4.18) and the trace inequality analog to (4.19), we obtain, similar to (4.20), Z @ (u (^ w" @n u" ; w"u)) ds @B"(x" i) Z c ""2 j5 (u (^ u" ; w"u))j ds (4.37)
1 0 12 Z c "" N2+3 B j5 (u (^ u" ; w" u))j2 dsC @ A

@B"(x" i)

Z @ w" @ (u (^ u" ; w" u)) ds ; (u (^ u" ; w"u)) ds: w" @n @n @B (x" )

" i

@B"(x" i)

11 0 2 ZZ N +2 C B 2 j5 (u (^ u" ; w"u))j dsA c "" 2 @

B"(x" i)

@B"(x" i)

26

0 11 2 ZZ 2 N +4 B C 2 " " " 5 (u (^ u ; w u)) dsA : +c " 2 @

By the estimates analog to (4.21) and (4.22), the second integral in the right-hand side of (4.36) can be bounded by Z @ w" Z " " " " ; w" u)j ds ( u (^ u ; w u)) ds c " ju (^ u @n @B"(x" @B"(x" i) i) 1 0 12 B ZZ j(u (^ 2 @ u" ; w" u))j2 dxC (4.38) c "" N A
B"(x" i)

0 11 2 ZZ N +2 B C 2 " " " +c " 2 @ j5 (u (^ u ; w u))j dxA

where we also used the trace formula like (4.19) and the fact that ( ) is bounded. Substituting (4.38) and (4.38) into (4.36), we derive that ZZ ZZ m" X " " " (4.39) w (u (^ u ; w u)) dx w" (u (^ u" ; w" u)) dx
B"(x" i)

B"(x" i)

"

ku (^ u" ; w" u)kH 1( ) + "2

i=1 B"(x")nT " i i " 2 u ; w" u)) 5 (u (^

L2 (

where we have used the inequality m" !2 m" X X (4.40) ai m" a2 i: Note that ku (^ u" ; w" u)kH 2( ) is uniformly bounded. Combining (4.34), (4.35) and (4.40), we nd
k (u (^ u" ; w" u))k2 L2 (
) 2 2 c ku ^" ; w" ukH 1( ) + c kw" ; 1kH 1 ( ) + c" : 2 ) is isometric to H0 ( ), it follows that

i=1

i=1

2 By (4.5), (4.6) and the fact that H0 (

ku (^ u" ; w"u)k2 H 2(

27

c ku ^" ; w" ukH 1( ) + c"2:

1 ln ; 1 as ! 1. Hence In this case, the Green's function (3.7) behaves like 4! 8!4 4 we need di erent approaches to treat the limiting equation. We rst look for a biharmonic function V in B n B2 (for > 5) such that

5. Homogenized Problem II: N = 4. In this section we study the case N = 4:

V = 0 in B n B2 V = @V @n = 0 on @B2 V = 1 @V @n = 0 on @B :
2

Existence and uniqueness follows from Lemma 2.1. In fact, such a solution can be constructed explicitly. One may verify that for jxj = r 2 < r < ! 2 r 4 1 2 2 2 (5.1) r + ( ; 4) ; 2 + 4 ln 2 ; r2 V (x) = where = ( 2 ; 4) ; We observe that for large c1 2 ln By direct computation, we obtain (5.2)
j j
2

+ 4 ln 2 :

c2 2 ln for some constant c2 > c1 > 0:

2 4 V (x) = 2 2 ; r+ 2

j5V (x)j j V (x)j

c 1+ + r r3 ! c 1+ 2 + 2 : r2 r4

We now, for convenience, extend the function V into B by setting V = 0 in B2: The extended function is still in H 2 (B ) : One can see that for large , the following
28

estimates hold: (5.3) (5.4)

k5V kL2 (B ) k V kL2 (B )

c ln p 2 c 1 + ln
3

pc :

ln

We next de ne a test function w" as in (4.4) by 8 x ; x" i < V for x 2 B" (x" i) (5.5) w" (x) = : 1 otherwise in where = " : Set (5.6)
Lemma 5.1. Let =
"
4

" ! 0. Then
)

1 : = ; "4 ln
;1 ln " c" 4 1 ; ln ;1 ln " " : c 4 1 ; ln
6

(5.7) (5.8)

k5w" k2 L2 ( k w" k2 L2 (

"

" 2 Consequently, if " 4 is uniformly bounded, then w ! 1 weakly in H ( ) : Proof: The proof is based on the estimates similar to (4.9) and (4.10). It follows from (5.3) and (5.4) that

k5w"k2 L2 ( 52w"
2

L2 (

cm" 2 k5V (x)k2 L2 (B ) 2 c 2 c 2 = 2 4 " jln j " jln " ; ln j2 2 cm" 52V (x) L2 (B ) "4 jln "c; ln j :

The assertion follows immediately. We now in the position to prove the main result in the case N = 4:
29

Theorem 5.1. Let N = 4 u" be the solution of (2.1)-(2.3) with g 2 C 2 and u ^"

be the extension of u" by 0 in Ti" : If

= lim "!0

"

<1

then the whole sequence u ^" ! u weakly in H 2 ( ) as " ! 0: Moreover, u satis es the following homogenized problem:

(5.9)

u + 4 !4 Pu = f in u = @u @n = 0 on @ :

If = 1 then u = 0. Proof: Consider rst the case that < 1: As in the proof of Theorem 4.1, we can show that u ^" is uniformly bounded in H 2: Since the rest of the proof is local, we
1 ( ) we take w" may assume, without loss of generality, that g = 0: For any 2 C0 " B2 (x" ) and that the function w" is as a test function. Note that B (x" i ) Ti i biharmonic only in B2 (x" i ) : Hence in this time we need to handle some extra terms. As in (4.14), we can write the variational formulation (2.4) as ZZ ZZ ZZ " " " " w u (5.10) ^ dx + (^ u ) w dx = fw" dx +

where =

ZZ

(^ u w"

+ 2 w" 5 u ^ 5 ;2 u ^" 5 w" 5 ) dx:

Consider rst the case that < 1. Then by Lemma 5.1, ! 0 as ! 0: Also (4.15) and (4.16) hold. By (3.5), note that w" = 0 in B2 (x" i ) we have ZZ ZZ " " (^ u" ) w" dx (^ u ) w dx = " B"(x" B"(x" i )nB2 (xi ) i)
30

(5.11)

@ w" u ^" ds @n @B"(x" @B"(x" i) i) Z " u ) ds ; Z @ w" u + w" @ (^ ^" ds @n @n @B2 (x" @B2 (x" i) i) 1 2 3 4 = Ii + Ii + Ii + Ii :
=
"

u ) ds ; w" @ (^ @n

We now proceed to estimate each Iik: By (5.2), for any x 2 @B" (x" i) ! 2+4 x ; x" 2 1 i " = 2 2 ; 2 def = w (x) = 2 V where (5.12) It follows that
m" X i=1

j j
m" Z X

2 2

ln

2 c " 4" ln " 1 ; ln

2 c " 4" :

Ii1 =
=

m" ZZ @ (^ u" ) ds = X " ) dx: (^ u @n i=1 B x" i=1 @B x" "( i ) "( i ) Combining with (5.12), we obtain ZZ m" X 1 " 2 (5.13) Ii c 4" j (^ u" )j dx c"2 ku ^" kH 2( ) : i=1

i=1 @B x" "( i ) m " X Z

u" ) ds w" @ (^ @n

To estimate Ii3 we observe that for x 2 @B2 (x" i ) from (5.2), ! 2+4 1 x ; x" 2 i " w (x) = 2 V = 2 2 ; 4 def = while (5.14)
31
2

c : ln

Therefore,
m" X i=1

Ii =
3

m" X i=1 @B
2

(x" i) i i=1 It follows from the Cauchy inequality and (5.14) that ZZ m" X c Ii3 j (^ u" )j dx 2 ln Sm" B2 (x") i=1 i i=1 1 12 1 0 c m" B (x" ) 2 @ ZZ j (^ u" )j2 dxA 2 i 2 ln i=1
2 1

(x" i) ZZ m" X
i=1 B

u" w" @ (^
(^ u"

@n

m" ) ds = X

Z
2

i=1 @B

) dx =

ZZ

(x" i)

@ (^ u" ) ds @n

Sm" B2 (x")

(^ u" ) dx:

c (m" 4) 2 ku 2 (5.15) ^" kH 2( ) c " ^" kH 2 ( ) : 4 " ku 2 ln We next estimate Ii4: Again from (5.2), for x 2 @B2 (x" i) ! 2 2 @ w" = @ 1 V x ; x" 2 @ ( + 4) i = 2 @n 2 ; 2 @n @n 2 jx ; x" ij ;2 2 ( 2 + 4) 2 def ^ = 2 = 8 3
while (5.16) It follows that m" m" X X (5.17) Ii4 = ;
i=1

c : ln

i=1 @B x" 2 ( i)

m" @ w" u " ds = ; ^ X ^ @n i=1

Z
@B2 (x" i)

u ^" ds:

Consider the following Neumann problem

2

= q in B1 = 0 @@n = 1 on @B1
32

!4 = 4: Existence of a smooth solution for general Neumann problem where q = jB 1j is followed by variation inequality. But in this particular case, one can easily verify that the function 4 3r2 (x) = r ; r = jxj 48 provide a smooth solution. We now set, for x 2 B2 x : " (x) = (2 )3 2 Then " satis es 2 " = 2q in B2 (5.18) " "=0 @ (5.19) @n = 1 on @B2 :
Using the identity (3.5) with w = " v = u ^" we obtain Z ZZ ZZ " u ^" ds = 2q u ^" dx ; (^ u" ) dx: @B2 (x" B2 (x" B2 (x" i) i) i) Substituting this into (5.17), we derive m" ZZ m" ZZ m" ^ X X X q " " 4 ^ u ^ dx + (^ u" ) dx Ii = ; 2 i=1 B x" i=1 B x" i=1 2 ( i) 2 ( i) (5.20) = J1 + J2: x by (5.16), we obtain Since " (x) = 2 2 1 0 11 0 12 2 ZZ ZZ m" C B C ^ XB B B jJ2 j j " j2 dx C j (^ u" )j2 dx C @ A @ A i=1 B x" " B x 2 ( i) 2 ( i) 1 0 12 ZZ m " B c X 3k k " )j2 dx C B C (5.21) j (^ u 2 L ( B ) @ A 3 ln 1 i=1 " B2 (xi ) c 1 k (^ u" )kL2(B1) ln "2
33

where we used (4.40). To estimate J1 we observe that for any function 2 H 2 (B2 ) such that = 0 in B , by rescaling the following Poincare's inequality holds: ZZ ZZ 2 dx c 2 j5 j2 dx: B2 (x" B 2 (x " i) i) Using this inequality twice we deduce that ZZ ZZ ju ^" j2 dx c 2 j5 (^ u" )j2 dx c B2 (x" B2 (x" i) i) Therefore
jJ1 j

ZZ

B2 (x" i)

52 (^ u" ) dx:

(5.22)

m" ZZ c X ju ^" j dx 4 ln i=1 B x" 2 ( i) 1 0 12 ZZ m " C c X 2B B ju ^" j2 dxC @ A 4 ln i=1 B2 (x" ) i 1 0 12 m" B ZZ c X c ku " ) 2 dxC B C 52 (^ u ^" kH 2 ( ) : @ A 2 ln i=1 " ln B2 (x" i) m" X i=1

Combining (5.13), (5.15), (5.21) and (5.22), we nd (5.23)

Ii1 + Ii3 + Ii4

c ku " " ln ^ kH 2( ) ;! 0:
2

Finally, we evaluate Ii2: For x 2 @B" (x" i ) we have, by (5.2),

2 @ w" = 1 @ V x ; x" i = 4 ( + 1) : 2 @n @n "3

Hence (5.24)
m" X i=1

Ii2 = ;

m" Z X

m" Z 2 @ w" u " ds = ; 4 ( + 1) X u ^" ds: ^ 3 @n " i=1@B x" i=1@B x" "( i ) "( i )

34

Using the arguments as we did in showing (4.25), we can easily obtain ZZ m" Z X " u ^" ds ;! !4 Pu dx as " ;! 0:
i=1@B"(x" )
i

Note that
;

4 ( 2 + 1) = ; 4 ( 2 + 1) "4 ( 2 ; 4) ; ( 2 + 4) ln

Therefore, it follows from (5.24) that ZZ m" X Ii2 ;! 4 !4 Pu dx:

i=1

"4 2

;! 4 :

Combing this with (5.10) and (5.23), we obtain from (5.11) ZZ ZZ (^ u" ) w"dx ;! 4 !4 Pu dx: The assertion for < 1 follows. The case = 1 can be handled in the way similar to the proof of Theorem " 4.1 (by taking w ~" = w" as the test function in (5.10), instead of w" ). We skip the details. We conclude this paper by the corrector results for N = 4: Theorem 5.2. Let N = 4: Then the assertion (4.28) in Theorem 4.2 remains true for any p < 2: Furthermore, if u 2 C 2 then (4.28) holds for p = 2:

Acknowledgments
The author would like to express his deep appreciation to Professor Avner Friedman for many valuable discussion.
35

REFERENCES 1] G. Allaire, Homogenization of the Navier-Stokes equations in open sets perforated with tiny holes I: Abstract framework, a volume distribution of holes, Arch. Rat. Mech. Anal., 113 (1991), 209-259. 2] G. Allaire, Homogenization of the Navier-Stokes equations in open sets perforated with tiny holes II: Noncritical sizes of the holes for a volume distribution and surface distribution of holes, Arch. Rat. Mech. Anal., 113 (1991), 261-298. 3] A. Bensoussan, J. L. Lions and G. Papanicolaou, Asymptotic Analysis for Periodic Structures, North-Holland, 1978. 4] S. Bergman, M. Schi er, Kernel functions and elliptic di erential equations in mathematical physics, Acad. Press, 1953. 5] D. Cioranescu, P. Donato, F. Murat and E. Zuazua, Homogenization and corrector for the wave equation in domains with small holes, Asymptotic Anal., Vol. 8 (1993), 252-293 6] D. Cioranescu and F. Murat, Un terme etrange venu d'ailleurs, Nonlinear Partial Di erential Equations and their Applications, College de France Seminar: Vol. 2, 98-138, vol. 3, 154178, Pittman, London, 1982. 7] D. Cioranescu and J. S. J. Paulin, Homogenization in open sets with holes, J. Math. Anal. Appl., 71 (1979), 590-607. 8] A. Friedman and C. Huang, E ective permeability of the boundary of a domain, to appear in Comm. PDE. 9] J. L. Lions and E. Magenes, Non-Homogeneous Boundary Value Problems and Applications, Springer, 1972. 10] O. A. Oleinik, A. S. Shamaev and G.A. Yosi an, Mathematical Problems in Elasticity and Homogenization, North-Holland, 1992. 11] G. C. Papanicolaou and S. R. S. Varadhan, Di usion in region with many small holes, in Lecture Notes in Control and Information, vol. 75, Springer, 1980, 190-206.

36