ASYMPTOTIC BEHAVIOUR OF SOLUTIONS

OF FOURTH ORDER DIRICHLET PROBLEMS

arXiv:math/0006009v1 [math.FA] 2 Jun 2000

Paolo DALL’AGLIO

Abstract

The behaviour of solutions to fourth order problems is studied through the decomposition
into a system of second order ones, which leads to relaxed formulations with the intro-
duction of measure terms. This allows to solve a shape optimization problem for a simply
supported thin plate.

Ref. S.I.S.S.A. 32/97/M (March 97)

 Asymptotic behaviour of solutions of fourth order Dirichlet problems 1

1. Introduction

In this paper we study the asymptotic behaviour of solutions of fourth order elliptic
problems on varying domains.
This has been widely studied in the past, in the case of second order elliptic oper-
ators (see for instance [6], [9], [7]). We will use such results decomposing fourth order
differential equations into a system of second order ones.
Given a bounded open set U in IRn , n ≥ 2 and a function f ∈ H -1(Ω) , the fourth
order equation
 ∆2 u = f in H -1(Ω)


∆u ∈ H10 (Ω) (1.1)

u ∈ H10 (Ω)


is linked to the model for the vertical displacement u of an thin plate, occupying a region
U , simply supported on ∂U , subjected to a load f . Simply supported means that the
boundary is fixed, but that the plate is free to rotate around the tangent to ∂U . For the
general treatment of plate theory we refer to [15], [11], [4], [14], [5].
In particular we want to study the asymptotic behaviour of solutions when the
domain varies. To this aim we will show (Proposition 3.1) that problem (1.1) is equivalent
to the system of second order equations

−∆u = v in H -1(U )



u ∈ H10 (U )

(1.2)

 −∆v = f in H -1(U )
v ∈ H10 (U ).


This problem can be handled with the theorems valid in the second order case (Theorems
2.1 and 2.2), and it will be proved that if un are the solutions of problems like (1.2) on
a sequence of subdomains Un of a given bounded domain Ω , then a subsequence of un
converges weakly in H10 (Ω) to a function uµ solving

 −∆uµ +
 µuµ = v
 1
uµ ∈ H0 (Ω) ∩ L2µ (Ω)
(1.3)
 −∆v + µv = f
 v ∈ H1 (Ω) ∩ L2 (Ω).

0 µ

where µ is a measure.
The study will be carried on for general fourth order elliptic operators with constant
coefficients and no lower order terms, that can be splitted into two second order ones.
2 P. DALL’AGLIO

A motivation for the study of the asymptotic behaviour of solutions of Dirichlet

problems in varying domains without geometric assumptions on the domains Un are the
so-called shape optimization problems: given a function j : Ω × IR → IR we consider the
following problem Z
min j(x, uU (x)) dx, (1.4)
U∈U(Ω)
Ω

where U(Ω) is the family of all open subset of Ω and uU is the solution of the problem
of type (1.1) in the set U .
In section 6 it will be shown that, in general, problem (1.4) does not have a solution.
Hence a relaxed optimization problem will be introduced, where the set over which we
minimize is the set of functions uµ , where µ is a measure and uµ solves the relaxed
problem (1.3). This set is the closure of {uU : U ∈ U(Ω)} in L2 (Ω) . This problem will
always have solution and its minimum will coincide with the infimum of integral in (1.4).
This shape optimization problem, in the second order case, was studied in [1], [2],
[3].

2. Notations and preliminary results

Given an open subset U of IRn , H10 (U ) is the usual Sobolev Space, H -1(U ) its dual,
and h·, ·i the duality pairing. If U ⊂ Ω and u ∈ H10 (U ) , then the function

u in U
ũ :=
0 in Ω \ U

is in H10 (Ω) . From now on we will always denote, with the same symbol u , a function
and its extension ũ .
In this paper we will deal with elliptic operators L : D ′ (Ω) → D ′ (Ω) of the form
X
Lu = cα ∂ α u,
|α|=m

where α and β are multiindeces, and m is the order of the operator that will be 2 or 4.
In any case they will be without lower order term and with constant coefficients.
The operators will be assumed to be elliptic in the sense that
X
cα ξ α ≥ γ|ξ|m , ∀ ξ ∈ IRn ,
|α|=m
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 3

where γ is a real positive constant. This, in our case, is the same as

P (ξ) 6= 0 ∀ ξ ∈ IRn \ {0}, (2.1)

X
where P is the polynomial cα ξ α associated to the operator L .
|α|=m
In this work, differential problems are always meant to be solved in the usual weak
sense. This means, for instance, that, for u ∈ H10 (U ) the expression

−∆u = f in H -1(U )

is an equality of linear functionals

XZ
h−∆u, vi = ∂i u∂i v dx = hf, vi,
i U

for any v ∈ H10 (U ) .

As said above, the limit of a sequence of solutions of Dirichlet problems is not, in
general, the solution of a problem of the same kind, but is the solution of a problem
where a measure term appears. To deal with these problems we need to recall some
notions.
For the notion of capacity of a set E ⊂ Ω , which we will indicate by cap(E) , we
refer to textbooks as [12] or [13]. We shall always identify a function u ∈ H10 (Ω) with its
quasi-continuous representative.
Now, let M0 (Ω) be the set of Borel measures which are zero on the sets of zero
capacity.
For one such measure µ ∈ M0 (Ω) , L2µ (U ) will be the space of functions such that
Z
|u|2 dµ < +∞.
U
X
Given a second order operator A = aij ∂i ∂j , with constant coefficients, for a function
ij
u ∈ H10 (U ) ∩ L2µ (U ) to solve the equation

Au + µu = f,

will mean that

n Z
X Z Z
aij ∂j u∂i v dx + uv dµ = f v dx, (2.2)
i=1 U U U
4 P. DALL’AGLIO

for all test functions in v ∈ H10 (U ) ∩ L2µ (U ) .

It can be easily proved that the space H10 (U ) ∩ L2µ (U ) is a Hilbert space whenever
µ is in M0 (Ω) , and hence, by Lax-Milgram Lemma, we have existence and uniqueness
of solutions for a problem of the form

Au + µu = f
u ∈ H10 (U ) ∩ L2µ (U ),

for any linear elliptic second order operator A .

The decomposition of a fourth order problem in a system of two second order equa-
tions allows us to study the asymptotic behaviour applying well known theorems for the
second order case separately to each equation. The following results, that can be found,
for instance in [9], [7], are the key points of the theory.

Theorem 2.1. Let A be a second order elliptic operator, as described above. For every
sequence {µn } in M0 (Ω) there exists a subsequence µnk such that, for every sequence
{g } in H -1(Ω) , strongly convergent to g ∈ H -1(Ω) , we have
n

znk ⇀ z weakly in H10 (Ω),

where znk and z solve

Aznk + µnk znk = gnk

 
Az + µz = g
znk ∈ H10 (Ω) ∩ L2µn (Ω), u ∈ H10 (Ω) ∩ L2µ (Ω),
k

respectively.

Theorem 2.2. For any measure µ ∈ M0 (Ω) there exists a sequence Un of open subsets
of Ω such that, for any sequence {gn } in H -1(Ω) , strongly convergent to g ∈ H -1(Ω) ,
the solutions zn of the problems

Azn = gn in H -1(Un )


zn ∈ H10 (Un )

converge weakly in H10 (Ω) to the solution z of


Az + µz = g
u ∈ H10 (Ω) ∩ L2µ (Ω).
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 5

3. Decomposition of fourth order operators

Fourth order elliptic problems are studied mainly with two different kinds of bound-
ary conditions. In the model case of the bi-laplacian, they correspond to two different
physical problems regarding, as said above, the displacement of a thin plate. Problems
of the type
∆ u = f in H -1(Ω)
 2

u ∈ H20 (Ω)
∂u
correspond to having u = = 0 on ∂Ω , that is to say that the plate is clamped along
∂n
its boundary. The asymptotic behaviour of the solutions of such problems has been
studied in [10].
In this work we deal with the second kind of boundary conditions, as in (1.1), and
we do it decomposing the problem in a system of two second order equations.
The decomposability of the fourth order operator
X
Lu = cα ∂ α u
|α|=4

can be seen through the associated polynomial

X
P (ξ) = cα ξ α .
|α|=4

It is a simple algebraic fact that, if the polynomial can be splitted into two second
degree polynomials
P (ξ) = Q(ξ)R(ξ),

then other decompositions can be obtained only by exchanging the order or multiplying
and dividing by constants. Observe that, according to (2.1), if P is elliptic, then so are
R and Q .
We remark here that such a decomposition can always be done in the two dimensional
case. In higher dimensions this is not always possible.
So assume that
 
X Xn n
X
Lu = cα ∂ α = bij ∂j ∂i  akl ∂k ∂l  u = BAu.
|α|=4 i,j=1 k,l=1
6 P. DALL’AGLIO

n
X n
X
Proposition 3.1. Let U be a subset of Ω , and A = aij ∂j ∂i , B = bij ∂j ∂i be
i,j=1 i,j
second order elliptic operators with constant coefficients. Let f ∈ H -1(U ) . The following
three problems are equivalent:

Au = v in H -1(U )

1 2
u ∈ H (U ) : Au ∈ L (U )

-

1 0
 BAu = f in H (U )  

u ∈ H10 (U )
Z 
(i) Au ∈ H10 (U ) (ii) AuBϕ dx = hf, ϕi (iii)
Bv = f in H -1(U )
u ∈ H10 (U )  U
  

1 2
∀ϕ ∈ H0 (U ) : Bϕ ∈ L (U ) v ∈ H10 (U )

(3.1)

Proof. (i) ⇒ (ii) . The equality hBAu, ϕi = hf, ϕi holds in particular for any ϕ in
H10 (U ) such that Bϕ ∈ L2 (U ) . Since B is symmetric and Au ∈ H10 (U ) we get
Z
AuBϕ dx = hBϕ, Aui = hBAu, ϕi = hf, ϕi
U

(ii) ⇒ (iii) . Let v be solution of

v ∈ H10 (U ), hBv, ϕi = hf, ϕi ∀ϕ ∈ H10 (U ).

Then
hBϕ, vi = hf, ϕi

and if, in particular, ϕ is such that Bϕ ∈ L2 (U ) , we get

Z
vBϕ dx = hf, ϕi,
U

which, subtracted to the equation in (ii) gives

Z
(Au − v)Bϕ dx = 0.
U

Observe now that, thanks to Lax-Milgram theorem, every function in L2 (U ) can be

written as Bϕ, with ϕ ∈ H10 (U ) , so we obtain
Z
(Au − v)z dx = 0, ∀z ∈ L2 (U ).
U
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 7

Taking z = Au − v , we get that kAu − vk2L2 (U) = 0 , hence Au = v , and u solves problem
(iii) .
(iii) ⇒ (ii) . If u is a solution of (iii) then

hBv, ϕi = hf, ϕi ∀ϕ ∈ H10 (U ).

and if, in particular, Bϕ ∈ L2 (U ) we have

Z
vBϕ dx = hf, ϕi.
U

Since now Au = v we have Z

AuBϕ dx = hf, ϕi,
U

for every ϕ ∈ H10 (U ) with Bϕ ∈ L2 (U ) , hence u solves (ii) .

(ii) ⇒ (i) . We have already proved that from (ii) it follows that Au ∈ H10 (U ) ,
hence BAu ∈ H -1(U ) , and so
Z
hf, ϕi = AuBϕ dx = hBϕ, Aui = hBAu, ϕi;
U

that is, BAu = f in H -1(U )

Remark 3.2. The equivalence af these problems gives for free existence and uniqueness,
since this is true for problem (iii) thanks to the observation made in section 2. This
is not obvious for problem (ii) because the spaces of solution and of test functions are
different.

Remark 3.3. Notice that if the boundary of U is regular, then by regularity theorems,
Au ∈ L2 (U ) implies that u belongs to H2 (U ) . The same is true for the test functions
ϕ. So the problem(3.1) (ii) becomes:

u ∈ H10 (U ) ∩ H2 (U )


Z
AuBϕ dx = hf, ϕi
 U

∀ϕ ∈ H10 (U ) ∩ H2 (U ).
8 P. DALL’AGLIO

Remark 3.4. It is important to remark that the boundary conditions in the equivalent
problems (3.1) (i)(ii)(iii) depend on the choice of the decomposition L = BA .
If we have two decompositions, in the sense that

Lϕ = B1 A1 ϕ = B2 A2 ϕ, ∀ϕ ∈ D ′ (Ω),

then the boundary conditions in (3.1) (i) are different. Hence also the other problems
(ii) and (iii) differ. Had we chosen the Dirichlet boundary conditions, that is seeking
u ∈ H20 (U ) , all decompositions would have given the same solution. But in this case
what makes the difference is the boundary condition, as can be seen with the following
example.

Example 3.5. The operator

Lu := uxxxx + 2uyyyy + 3uxxyy

can be written, for instance, as the product of

Au = ∆u and Bu = ∆u + uyy .

But the integral in equation (3.1) (ii) will be different according to which operator we
will apply first:
Z Z
AuBv dx = (uxx vxx + 2uyy vyy + 2uxx vyy + uyy vxx ) dx ,
Z Z
BuAv dx = (uxx vxx + 2uyy vyy + uxx vyy + 2uyy vxx ) dx .

Computations show that these two integrals differ by a term on the boundary, which
would vanish if functions where in H20 (U ) .

4. The asymptotic behaviour

We come now to examine problem (3.1) (i) when we have a sequence of domains Un
all contained in Ω . Let f be in H -1(Ω) (this implies it is also in H -1(Un ) for any Un )
and consider always functions of H10 (Un ) trivially extended to the whole of Ω . As proved
in Proposition 3.1, we can study directly
Au = v in H -1(Un )

 n 1n


un ∈ H0 (Un )
(4.1)
Bv = f in H -1(Un )
 n 1


vn ∈ H0 (Un ).
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 9

We first consider the second equation

Bvn = f in H -1(Un )


vn ∈ H10 (Un ).

From Theorem 2.1 we know that there exist a subsequence, which we still call Un , and
a measure µB , depending on the sequence Un , on the operator B , but not on f , such
that
vn ⇀ v weakly in H10 (Ω)

and v solves 
Bv + µB v = f
(4.2)
v ∈ H10 (Ω) ∩ L2µB (Ω),

in the sense specified in (2.2).

We can now apply Theorem 2.1 to the problem in u

Aun = vn in H -1(Un )


un ∈ H10 (Un ),

taking as Un only those in the subsequence obtained for B . Again there exists a subse-
quence, which we still call Un , and a measure µA , depending on the sequence Un , on
the operator A , but not on the sequence vn , such that

un ⇀ u weakly in H10 (Ω)

and u solves 
Au + µA u = v
(4.3)
u ∈ H10 (Ω) ∩ L2µA (Ω).

This allows us to conclude that, if un are the solutions of system (4.1) then, up to a
subsequence,
un ⇀ u weakly in H10 (Ω)

and u is the solution of 

 Au + µA u = v
 u ∈ H1 (Ω) ∩ L2 (Ω)

0 µA
(4.4)
 Bv + µ B v = f
 v ∈ H1 (Ω) ∩ L2 (Ω).

0 µB
10 P. DALL’AGLIO

5. The single equation formulation

The goal of this section is to write problem (4.4) as a single equation of the form
Z
(Au + µA u)(Bϕ + µB ϕ) dx = hf, ϕi.
Ω

Of course Au + µA u alone doesn’t make sense, because it has to be under-

stood in the sense explained in chapter 2. What we will do now, is hence to define
suitable function spaces, which will play the role of z ∈ H10 (Ω) : Az ∈ L2 (Ω) and


w ∈ H10 (Ω) : Bw ∈ L2 (Ω) and operators ΛµA and ΛµB which give meaning to the


integral Z
ΛµA u ΛµB ϕ dx.
Ω

The construction of such operators will be done only relatively to operator A , being the
one relative to B perfectly analogous.
Consider the operator

RA : H -1(Ω) −→ H10 (Ω) ∩ L2µA (Ω)

z 7−→ w

where w is the solution of 

Aw + µA w = z
(5.1)
w ∈ H10 (Ω) ∩ L2µA (Ω);
so that we have u = RA v (and v = RB f ).
Define now the space

VA (Ω) := w ∈ H10 (Ω) ∩ L2µA (Ω) : ∃z ∈ H10 (Ω) ∩ L2µA (Ω) s.t. w = RA z .


Let us prove that the function z is unique, for each w .

Assume, by contradiction, that there exist two z1 , z2 ∈ H10 (Ω) ∩ L2µA (Ω) such that
w = RA z1 = RA z2 . From the equation we have
Z
(z1 − z2 )ψ dx = 0, ∀ψ ∈ H10 (Ω) ∩ L2µA (Ω).
Ω

Since z1 − z2 ∈ H10 (Ω) ∩ L2µA (Ω) we get

Z
(z1 − z2 )2 dx = kz1 − z2 k2L2 (Ω) = 0
Ω
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 11

and we conclude that z1 = z2 .

So we have defined
VA (Ω) = RA H10 (Ω) ∩ L2µA (Ω)

and on its image RA is one-to-one. So, on the space VA (Ω) , the inverse operator can be
defined:
ΛµC : VA (Ω) −→ H10 (Ω) ∩ L2µA (Ω).
w 7−→ z s.t. w = RA z
Observe now that the operator RA is symmetric, that is:

hh, RA gi = hg, RAhi,

for any h ∈ H -1(Ω) and g ∈ H10 (Ω) .

We are now able to write system (4.4) as a single equation. We have v = RB f
and u = RA v , so that u ∈ VA (Ω) . For any ϕ ∈ VB (Ω) there exists a unique z ∈
H10 (Ω) ∩ L2µB (Ω) such that ϕ = RB z . Hence,

hf, ϕi = hf, RB zi = hz, RB f i = hz, vi

Z
= hΛµB ϕ, ΛµA ui = ΛµB ϕ ΛµA u dx.
Ω

Last equality holds because both ΛµB ϕ and ΛµA u are in L2 (Ω) .
Hence we can conclude that the function u , solution of (4.4), also solves

u ∈ VA (Ω)


Z
ΛµB ϕ ΛµA u dx = hf, ϕi
 Ω

∀ϕ ∈ VB (Ω).

It is easily seen that also the converse holds. Let v be a solution of


Bv + µB v = f
v ∈ H10 (Ω) ∩ L2µB (Ω);

then, by the definition, v = RB f . Hence, taken any z ∈ H10 (Ω) ∩ L2µB (Ω) , we set
ϕ := RB z , and we have:

hv, zi =hRB f, zi = hf, RB zi = hf, ϕi

=hΛµA u, ΛµB ϕi = hΛµA u, zi.
12 P. DALL’AGLIO

Taking then z = v − ΛµA u , we obtain

kv − ΛµA uk2L2 = 0,

so
ΛµA u = v,

that is, by the definition of ΛµA ,

u = RA v,

hence u solves (4.4).

All this can be summarized in the following

Theorem 5.1. A function u ∈ H10 (Ω) ∩ L2µA (Ω) is a solution of

u ∈ VA (Ω)


Z
ΛµB ϕ ΛµA u dx = hf, ϕi
 Ω

∀ϕ ∈ VB (Ω),

if and only if it solves 

 Au + µA u = v
 u ∈ H1 (Ω) ∩ L2 (Ω)

0 µA
 Bv + µ B v = f
 v ∈ H1 (Ω) ∩ L2 (Ω).

0 µB

It is importat to remark that the road back to the equivalence with a fourth order
problem stops here. This is because, in general, the image of ΛµA is not contained in
the domain of ΛµB , even if A = B and the two measures coincide.

6. The optimization problem

Let us consider now the optimization problem (1.4).

Let f be a function in H -1(Ω) and let j : Ω × IR → IR satisfy the standard
Carathéodory conditions and be such that

|j(x, s)| ≤ b(x) + β|s|p , for a.e. x ∈ Ω and ∀s ∈ IR, (6.1)

2n
with b ∈ L1 (Ω) , β ∈ IR and 1 ≤ p < 2∗ , where 2∗ = is the Sobolev exponent.
n−2
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 13

We want to study Z
min j(x, uU (x)) dx, (6.2)
U∈U(Ω)
Ω

where U(Ω) is the family of all open subsets of Ω and uU is the solution, trivially
extended in Ω \ U , of the problem

 ∆2 u = f in H -1(U )


∆u ∈ H10 (U )
u ∈ H10 (U ).


In general problem (6.2) does not have solution, in the sense that the infimum is not
attained on the set {uU : U ∈ U(Ω)}. This can be seen with the following example.

Example 6.1. Consider the case of Ω ⊂ IRn with smooth boundary. Consider a function

w ∈ C ∞ (Ω), s.t. w > 0 in Ω, w = 0 and ∆w = 0 on ∂Ω

and set
z := −∆w + w, f := −∆z + z. (6.3)

Consider now
2
j(x, s) = (s − w(x)) ,

so that the integral Z

2
(u(x) − w(x)) dx (6.4)
Ω

takes the value zero if and only if u = w a.e. It will turn out from the theory (see
equation (6.7)) that zero is in fact the infimum of the functional over all solutionds of
Dirichlet problems on subsets of Ω .
We prove now that, with this choice of f , w can not be the solution of the problem

 ∆2 uU = f in H -1(U )


∆uU ∈ H10 (U )
uU ∈ H10 (U ),


for any U ⊂ Ω , or, equivalently (thanks to Proposition 3.1), of system

−∆uU = v in H -1(U )



uU ∈ H10 (U )


 −∆v = f in H -1(U )
v ∈ H10 (U ).

14 P. DALL’AGLIO

First consider the case when cap(Ω \ U ) = 0 . Then the spaces H10 (Ω) and H10 (U )
coincide (see for instance Theorem 4.5 in [13]), and the problem in U is the same as the
problem in Ω . We show that w 6= uΩ . From (6.3) we have

∆2 w − ∆w = −∆z
=f −z
= f + ∆w − w.
w
So ∆2 w = f + 2∆w − w , and hence w can be such that ∆2 w = f only if −∆w = −
2
but this is impossible because, by the maximum principle, w should be negative, while
we have choosen it strictly positive.
The second case is when cap(Ω \ U ) > 0 . If this happens uU has to be zero in Ω \ U ,
so that uU 6= w on a set of non-zero capacity. But for functions in H10 (Ω) to be equal
Lebesgue-a.e. is the same as cap-a.e., hence uU can not be a minimizer for (6.4).

The example shows then, that to be able to solve always our optimization problem,
it is convenient to seek our minima in the larger set M := {uµ : µ ∈ M0 (Ω)} , which is
the closure of N := {uU : U ∈ U(Ω)} in the weak topology of H10 (Ω) . In other words
we introduce the relaxed optimization problem
Z
min j(x, uµ (x)) dx, (6.5)
µ∈M0 (Ω)
Ω

where uµ denotes the solution fo the relaxed problem


 −∆uµ + uµ µ = vµ
 u ∈ H1 (Ω) ∩ L2 (Ω)

µ 0 µ
(6.6)
 −∆v µ + vµ µ = f
 v ∈ H1 (Ω) ∩ L2 (Ω)

µ 0 µ

and each equation is meant in the sense of formula (2.2).

In order to see that solving the new problem (6.5) gives the complete solution to
problem (6.2), we have to show two things: that (6.5) has a solution and that
Z Z
min j(x, uµ (x)) dx = inf j(x, uU (x)) dx. (6.7)
µ∈M0 (Ω) U∈U(Ω)
Ω Ω

As for the first thing, we need to show that the set M is closed in the weak topology
of H10 (Ω) . So consider minimizing sequence µn and let uµn be the solutions of the
 Asymptotic behaviour of solutions of fourth order Dirichlet problems 15

systems 
 −∆uµn + uµn µn = vµn
 u ∈ H1 (Ω) ∩ L2 (Ω)

µn 0 µn
 −∆vµn + vµn µn = f
 v ∈ H1 (Ω) ∩ L2 (Ω).

µn 0 µn

Applying Theorem 2.1 to the second equation we have that there exist a subsequence,
called for simplicity µn , and a measure µ ∈ M0 (Ω) such that vµn ⇀ vµ weakly in H10 (Ω) ,

−∆vµ + vµ µ = f
vµ ∈ H10 (Ω) ∩ L2µ (Ω).

Applying again Theorem 2.1 to the first equation (only those corresponding to the
subsequence), we obtain that

uµn ⇀ uµ weakly in H10 (Ω).

and uµ is the solution of (6.6).

Hence uµ is admissible and, by the minimizing property of the sequence {µn } ,
Z Z
j(x, uµ (x)) dx = min j(x, uν (x)) dx.
ν∈M0 (Ω)
Ω Ω

On the other hand observe that every problem of the form

−∆z = w

z ∈ H10 (U )



 −∆w =

1
g
w ∈ H0 (U )

can be viewed as a relaxed problem with the measure


U 0 if B \ U has capacity zero
µ (B) :=
+∞ otherwise

since this holds for second order problems as can be seen, for instance, in [2].
This ensures us that problem (6.5) is indeed an extension of (6.2), in the sense that
N ⊆ M , so that
Z Z
min j(x, uµ (x)) dx ≤ inf j(x, uU (x)) dx.
µ∈M0 (Ω) U∈U(Ω)
Ω Ω

To prove the inverse inequality we need the following

16 P. DALL’AGLIO

Lemma 6.2. Let µ ∈ M0 (Ω) . Then there exists a sequence Un of subsets of Ω such
that
uUn ⇀ uµ weakly in H10 (Ω)

where uUn and uµ solve respectively

−∆uUn = vn in H -1(Un )
 
  −∆uµ + uµ µ = vµ
  u ∈ H1 (Ω) ∩ L2 (Ω)

uUn ∈ H10 (Un )

µ 0 µ
and
 −∆vn = f in H -1(Un )  −∆vµ + vµ µ = f
  v ∈ H1 (Ω) ∩ L2 (Ω).

vn ∈ H10 (Un )

µ 0 µ

Proof. We know, from Theorem 2.2, that given µ, we have a sequence of sets Un such
that, for every g ∈ H -1(Ω) , if

−∆zn = g in H -1(Un )


zn ∈ H10 (Un ).

then the sequence zn tends weakly in H10 (Ω) to zµ , solution of


−∆zµ + vµ µ = g
zµ ∈ H10 (Ω) ∩ L2µ (Ω).

Since the sequence Un depends only on the operator and not on the right hand side the
same sequence of sets suits for the problem in u . Hence uUn converge to uµ solution of
the second system. Z
Since, by the hypothesis (6.1), the operator u 7→ j(x, u) dx is continuous in the
strong topology of L2 (Ω) (see [2] Theorem 4.1), the density result enables us to conclude
that Z Z
min j(x, uµ (x)) dx = inf j(x, uA (x)) dx.
µ∈M0 (Ω) U∈U(Ω)
Ω Ω

This ensures us also that every solution of a relaxed system can be approximated by
solutions of fourth order problems.

Remark 6.3. We remark here that the model case of the bi-laplacian can be extended
without changes in the proofs to the case of a fourth order elliptic operator with constat
coefficients L such that
Lu = C 2 u,

where C is a second order elliptic operator with constant coefficients.

 Asymptotic behaviour of solutions of fourth order Dirichlet problems 17

References

[1] ATTOUCH H.: Variational converence for functions and operators. Pitman, Boston, 1984
[2] BUTTAZZO G., DAL MASO G.: Shape optimization for Dirichlet Problems: Relaxed For-
mulation and Optimality Conditions. Appl. Math. Optim. 23 (1991), 17-49.
[3] CHIPOT M., DAL MASO G.: Relaxed shape optimization: thecase of nonnegative data
for the Dirichlet problem. Adv. Math. Sci. Appl. 1 (1992), 47-81.
[4] CIARLET P.G.: The finite element method for elliptic problems. Studies in mathematics
and its applications. North Holland, New York, 1978
[5] CIARLET P.G.: Plates and junctions in elastic multi-structures. An asymptotic analysis.
Masson, paris, 1990
[6] CIORANESCU D., MURAT F.: Un terme étrange venu d’ailleurs I and II. In Nonlinear
partial differential equations and their applications. Pitman research notes in mathematics;
n. 60, 98-138; n. 70, 154-178.
[7] DAL MASO G., GARRONI A.: New results on the asymptotic behaviour of Dirichlet
problems in perforated domains. Math. Mod. Meth. Appl. Sci. 3 (1994), 373-407.
[8] DAL MASO G., MALUSA A.: Approximation of relaxed Dirichlet problems by boundary
value problems in perforated domains. Proc. Roy. Soc. Edinburgh 125A (1995), 99-114.
[9] DAL MASO G., MOSCO U.: Wiener’s criterion and Γ -convergence. Appl. Math. Optim.
15 (1987), 15-63.
[10] DAL MASO G., PADERNI G.: Variational inequalities for the biharmonic operator with
variable coefficients. Ann. Mat. Pura Appl. 103 (1988), 203-227.
[11] DUVAUT G., LIONS J.-L.: Inequalities in mechanics and physics. Springer-Verlag, Berlin,
1976
[12] EVANS L.C., GARIEPY R.F.: Measure theory and fine properties of functions. CRC Press,
Boca Raton, 1992
[13] HEINONEN J., KILPELÄINEN T., MARTIO O.: Nonlinear potential theory of degenerate
elliptic equations. Clarendon Press, Oxford, 1993
[14] LAGNESE J.E., LIONS J.-L.: Modelling analysis and control of thin plates. Masson, Paris,
1988
[15] LANDAU L., LIFSHITZ E.: Theory of elasticity. Course of theoretical physics. Pergamon
Press, 1975