Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

17
Nonlinear Analysis 43 (2001) 199 – 215 www.elsevier.nl/locate/na Existence and uniqueness of solutions for nonlinear obstacle problems with measure data Chiara Leone SISSA, Via Beirut 2-4, 34014 Trieste, Italy Received 29 October 1998; accepted 4 November 1998 Keywords: Nonlinear obstacle problems; Measure data 1. Introduction In this paper we consider the obstacle problem with measure data associated with a nonlinear elliptic dierential operator A of monotone type, mapping W 1;p 0 (); p¿ 1, into its dual W -1;p 0 (). The theory of variational inequalities (to which obstacle problems belong) has been widely studied in the classical context of data in W -1;p 0 (). For any datum F W -1;p 0 () the unilateral problem relative to A; , and the ob- stacle (denoted by VI (A; ; )) is the problem of nding a function u such that u W 1;p 0 (); u ; hA(u);v - ui≥hF; v - ui; v W 1;p 0 (); v : (1.1) This problem has a unique solution whenever there exists a function z in W 1;p 0 () such that z (by this we mean that the inequality is satised C p -quasi everywhere in ; see Section 2). Characterization 1: The solution u can be characterized (see, e.g., chapters II and III in [10]) as the smallest function in W 1;p 0 (), greater than or equal to , such that A(u) - F = in ; u =0 on @; (1.2) for some nonnegative element of W -1;p 0 (). E-mail address: [email protected] (C. Leone). 0362-546X/00/$ - see front matter ? 2000 Elsevier Science Ltd. All rights reserved. PII: S0362-546X(99)00190-X

Transcript of Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

Page 1: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

Nonlinear Analysis 43 (2001) 199–215www.elsevier.nl/locate/na

Existence and uniqueness of solutions fornonlinear obstacle problems with measure data

Chiara LeoneSISSA, Via Beirut 2-4, 34014 Trieste, Italy

Received 29 October 1998; accepted 4 November 1998

Keywords: Nonlinear obstacle problems; Measure data

1. Introduction

In this paper we consider the obstacle problem with measure data associated with anonlinear elliptic di�erential operator A of monotone type, mapping W 1;p

0 (); p¿ 1,into its dual W−1;p′

().The theory of variational inequalities (to which obstacle problems belong) has been

widely studied in the classical context of data in W−1;p′().

For any datum F ∈W−1;p′() the unilateral problem relative to A; �, and the ob-

stacle (denoted by VI(A; �; )) is the problem of �nding a function u such that

u∈W 1;p0 (); u ≥ ;

〈A(u); v− u〉 ≥ 〈F; v− u〉;∀v∈W 1;p

0 (); v ≥ :

(1.1)

This problem has a unique solution whenever there exists a function z in W 1;p0 () such

that z ≥ (by this we mean that the inequality is satis�ed Cp-quasi everywhere in ;see Section 2).Characterization 1: The solution u can be characterized (see, e.g., chapters II and

III in [10]) as the smallest function in W 1;p0 (), greater than or equal to , such that

A(u)− F = � in ;

u= 0 on @;(1.2)

for some nonnegative element � of W−1;p′().

E-mail address: [email protected] (C. Leone).

0362-546X/00/$ - see front matter ? 2000 Elsevier Science Ltd. All rights reserved.PII: S0362 -546X(99)00190 -X

Page 2: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

200 C. Leone / Nonlinear Analysis 43 (2001) 199–215

Characterization 2: Finally, when the obstacle is Cp-quasi upper semicontinu-ous (see Section 2) u is also characterized (see, e.g., Theorem 3:2 in [1]) by thecomplementarity system

u∈W 1;p0 (); u ≥ ;

A(u) = F + �;

�∈W−1;p′(); � ≥ 0;

�({u− ¿ 0}) = 0;

(1.3)

where the pointwise values of u are de�ned Cp-quasi everywhere (see Section 2).Since � is a nonnegative element of W−1;p′

(), by the Riesz Representation Theorem,it is a nonnegative Radon measure; this explains the meaning of the last line of (1.3),which can be written also as u= �-almost everywhere in .We want to extend these results to the case where the forcing term F belongs to a

suitable subspace of the space Mb() of Radon measures � on whose total variation|�| is bounded on . As usually, we identify Mb() with the dual of the Banach spaceC0() of continuous functions that are zero on the boundary; so that the duality is〈�; u〉= ∫

u d�, for every u in C0() and the norm is ‖�‖Mb() = |�|().When �∈Mb(), the duality 〈�; u〉 has not always a meaning if u∈W 1;q

0 (); q¡N .Hence the classical formulation of the variational inequality (1.1) fails, as well as theboundary value problem (1.2) cannot be thought in the usual variational sense, evenif � = 0. Thus, we need a weaker formulation of (1.2) which allows us to overcomethis di�culty.In the linear case the duality solutions obtained by Stampacchia in [15] satisfy a

formulation which ensures uniqueness, while in the nonlinear case we may use thenotion of entropy solution introduced in [2] for L1-data and extended in [5] to boundedRadon measures vanishing on all sets of p-capacity zero. When � is a measure of thistype, the problem

A(u) = � in ;

u= 0 on @;(1.4)

has a unique entropy solution (see Section 2).In this paper the set of these measures is denoted by M

pb;0(), while by M+

b ()and by M

p;+b;0 () we denote the positive cones of Mb() and M

pb;0(), respectively.

Using the notion of entropy solution we introduce a de�nition for unilateral problemswith measure data quite similar to Characterization 1.

De�nition 1.1. We say that u is the solution of the Obstacle Problem with datum�∈M

pb;0() and obstacle (denoted by OP(A; �; )) if

1. there exists a measure �∈Mp;+b;0 () such that u is the entropy solution of (1.4)

relative to � + �, and u ≥ Cp-quasi everywhere in ,

Page 3: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 201

2. for any �∈Mp;+b;0 () such that the entropy solution v of (1.4) relative to � + �

satis�es v ≥ Cp-quasi everywhere in , we have u ≤ v Cp-q.e. in :

By de�nition, it is clear that, if such a solution exists, it is unique.The nonnegative measure �, which is uniquely de�ned, will be called the obstacle

reaction relative to u, or the measure associated with it.By an approximation technique we will prove that for any datum � in M

pb;0() there

exists a solution in the sense of De�nition 1.1.As in the classical framework, in Section 5 we will show that the solution found

can be characterized by the complementarity system.More precisely, Theorem 2.9 shows that the solution u of OP(A; �; ) is the only

entropy solution of (1.4) relative to a measure of the form � + �; �∈Mp;+b;0 (), such

that u= ; �-almost everywhere in , and u ≥ Cp-quasi everywhere in .We also �nd a more technical characterization of the solution of OP(A; �; ), which

in the case of Cp-quasi upper bounded in turns out to be similar to (1.1). In thisframework we recover the de�nitions given by Boccardo and Gallou�et in [4] and byBoccardo and Cirmi in [3] when �=f∈L1(), and by Oppezzi and Rossi in [13,14]in a more general case.Finally, the approach to obstacle problems proposed in this paper allows us to obtain

a stability result with respect to strong convergence of data in Mb().

2. Assumptions, preliminaries, and main results

Let be a bounded, open subset of RN ; N ≥ 2. Let p and p′ be two real numbers,with p¿ 1; p′ ¿ 1 and (1=p) + (1=p′) = 1. Let a : ×RN 7→ RN be a Carath�eodoryfunction such that for almost every x in and for every �; � in RN (� 6= �):

a(x; �) · � ≥ �|�|p − h(x); (2.1)

|a(x; �)| ≤ �[k(x) + |�|p−1]; (2.2)

(a(x; �)− a(x; �)) · (�− �)¿ 0; (2.3)

a(x; 0) = 0; (2.4)

where � and � are two positive real constants, h is a nonnegative function in L1()and k is a nonnegative function in Lp′

().Thanks to hypotheses (2.1)–(2.3), the operator A : u 7→−div(a(x;∇u)) maps W 1;p

0 ()into its dual W−1;p′

(), and it is coercive, continuous, bounded, and monotone (see[12]).Let K be a compact subset of . The p-capacity of K with respect to is

de�ned as

Cp(K) = inf{∫

|∇u|p dx: u∈C∞

0 (); u ≥ �K

};

Page 4: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

202 C. Leone / Nonlinear Analysis 43 (2001) 199–215

where �K is the characteristic function of K and we use the convention inf ∅ = +∞.This de�nition can be extended to any open subset B of in the following way:

Cp(B) = sup{Cp(K): K compact; K ⊆}:Finally, it is possible to de�ne the p-capacity of any set A⊆ as

Cp(A) = inf{Cp(B): B open; A⊆B}:We say that a property P(x) holds Cp-quasi everywhere (Cp-q.e.) in a set E⊆, ifit holds for all x∈E except for a subset N of E with Cp(N ) = 0.A function u : 7→ �R is said to be Cp-quasi continuous (resp. Cp-quasi upper

semicontinuous) if for every �¿ 0 there exists a set E⊆, with Cp(E)¡�, such thatthe restriction of u to \ E is a continuous (resp. upper semicontinuous) functionwith values in �R. It is well known that every u∈W 1;p

0 () has a Cp-quasi continuousrepresentative, which is uniquely de�ned (and �nite) up to a set of p-capacity zero. Inthe sequel we shall always identify u with its Cp-quasi continuous representative, sothat the pointwise values of a function u∈W 1;p

0 () are de�ned Cp-quasi everywhere.A set E⊆ is said to be Cp-quasi open if for every �¿ 0 there exists an open set

U such that E⊆U ⊆ and Cp(U \ E) ≤ �.Let Mp

b;0() be the set of measures in Mb() which are “absolutely continuous”with respect to the p-capacity, that is a measure �∈Mb() belongs to M

pb;0() if and

only if �(A) = 0 for every Borel set A⊆ such that Cp(A) = 0.It is well known that, if � belongs to W−1;p′

() ∩Mb(), then � is in Mpb;0(),

every u in W 1;p0 () ∩ L∞() is summable with respect to � and

〈�; u〉=∫u d�;

where 〈·; ·〉 denotes the duality pairing between W−1;p′() and W 1;p

0 (), while in theright-hand side u denotes the Cp-quasi continuous representative and, consequently, thepointwise values of u are de�ned �-almost everywhere.A decomposition theorem for measures in M

pb;0() is known (see Theorem 2:1

in [5]): every measure that is zero on the sets of p-capacity zero can be split intothe sum of an element in W−1;p′

() and of a function in L1(), and, conversely,every measure in L1() +W−1;p′

() is zero on the sets of p-capacity zero. So, if ameasure � belongs to M

pb;0(), then every u∈W 1;p

0 () ∩ L∞() is summable withrespect to �.For every k ¿ 0 we de�ne the truncation function Tk : R 7→ R by

Tk(t) =

{t if |t| ≤ k;

k sign(t) if |t|¿k:

Let us consider the space T1;p0 () of all functions u : 7→ �R which are almost

everywhere �nite and such that Tk(u)∈W 1;p0 () for every k ¿ 0. It is easy to see that

every function u∈T1;p0 () has a Cp-quasi continuous representative with values in �R,

that will always be identi�ed with the function u. Moreover, for every u∈T1;p0 ()

Page 5: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 203

there exists a measurable function � : 7→ RN such that ∇Tk(u) = ��{|u|≤k} a.e. in (see Lemma 2:1 in [2]). This function �, which is unique up to almost everywhereequivalence, will be denoted by ∇u. Note that ∇u coincides with the distributionalgradient of u whenever u∈T

1;p0 () ∩ L1loc() and ∇u∈L1loc(;RN ).

We are now in position to recall the notion of entropy solution introduced in [2]for L1 data and extended to measures in M

pb;0() in [5], which ensures us that, when

�∈Mpb;0(), the equation

A(u) = � in

u= 0 on @;(2.5)

has a unique entropy solution in the sense of the following de�nition.

De�nition 2.1. Let �∈Mpb;0(). A function u is an entropy solution of problem (2.5)

if u belongs to T1;p0 (), and∫

a(x;∇u)∇Tk(u− ’) dx ≤

∫Tk(u− ’) d� (2.6)

for every ’ in W 1;p0 () ∩ L∞() and every k ¿ 0.

Remark 2.2. Actually, it is possible to prove that equality holds in (2.6) (see [11]).

Observe that, if u is the entropy solution of (2.5), then (the Cp-quasi continuousrepresentative of) u is �nite up to a set of p-capacity zero.We recall also the following stability result (see Theorem 1:2 in [11]):

Theorem 2.3. Let {fn}⊂L1() and {Fn}⊂W−1;p′() be such that

fn * f weakly in L1();

Fn → F strongly in W−1;p′();

let un be the entropy solutions of (2:5) relative to � = fn + Fn; and let u be theentropy solution of (2:5) relative to � = f + F . Then

limn→∞ Tk(un) = Tk(u) strongly in W 1;p

0 ();

for every k ¿ 0.

Remark 2.4. Actually, in [2,5,11] the assumptions on the operator A are slightly di�er-ent. Indeed, in these papers, the operator A satis�es (2.1) with h=0 (this fact impliescondition (2.4)). On the other hand, nothing essential changes in the proofs of the ex-istence, uniqueness and stability results, if we assume (2.1) with a general h∈L1(),and we add hypothesis (2.4).Finally, we consider a function : 7→ �R such that

≤ u� Cp-q:e: in ; (2.7)

Page 6: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

204 C. Leone / Nonlinear Analysis 43 (2001) 199–215

where � is an element of W−1;p′() ∩Mb() and u� is the variational solution of

A(u) = � in ;

u= 0 on @:

Moreover, we de�ne the set

K () := {z ∈T1;p0 (): z ≥ Cp-q:e: in }:

Without loss of generality, we may suppose that is Cp-quasi upper semicontinuousthanks to the following Proposition (see Proposition 1:5 in [8]).

Proposition 2.5. Let : 7→ �R satisfying∃ z ∈W 1;p

0 () such that z ≥ Cp-q:e: in : (2.8)

Then there exists a Cp-quasi upper semicontinuous function ̂ : 7→ �R such that

1. ̂ ≥ Cp-q.e. in ;2. if ’ : 7→ �R is Cp-quasi upper semicontinuous and ’ ≥ Cp-q.e. in ; then

’ ≥ ̂ Cp-q.e. in .

Thus, in particular, K () = K ̂ ().Observe that assumption (2.7) is satis�ed if (2.8) holds, and there exists a compact

J ⊂ such that ≤ 0 in \ J . Indeed, we take as � the obstacle reaction correspond-ing to the solution of VI(A; 0; ). Then, by (1.3) supp�⊂ J , and �∈W−1;p′

() ∩Mb(). On the other hand, Example 5:3 in [6] shows that, in general, (2.8) does notimply (2.7).In De�nition 1.1 we speci�ed the formulation of obstacle problems we will adopt

in this paper. Let us note, however, that in this de�nition, since u and v are Cp-quasicontinuous, it is enough to prove u ≤ v a.e. in to obtain also the inequality Cp-q.e.in .In Section 4 we will prove the following existence theorem.

Theorem 2.6. Let satisfy (2:7) and let �∈Mpb;0(). Then there exists a unique

solution of OP(A; �; ). Moreover the corresponding obstacle reaction � satis�es

‖�‖Mb() ≤ ‖(� − �)−‖Mb(): (2.9)

In Section 5 we will prove also the following theorem about continuous dependenceon data.

Theorem 2.7. Let satisfy hypothesis (2:7); let �n; �∈Mpb;0() and �n and � the

measures associated to the solution un and u of OP(A; �n; ) and OP(A; �; ); respec-tively. If �n → � strongly in Mb(); then �n → � strongly in Mb(). Moreover; forevery k ¿ 0; Tk(un) converges to Tk(u) strongly in W 1;p

0 ().

Remark 2.8. Under slightly stronger hypotheses on the obstacle, when the data areL1() functions, the solutions considered in this paper coincide, for uniqueness reasons,

Page 7: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 205

with those given by Boccardo and Gallou�et in [4] and by Boccardo and Cirmi in[3]. Indeed, these solutions are obtained as limit of solutions of variational obstacleproblems, whose data are smooth and converge strongly in Mb(). By Theorem 2.7these solution satisfy De�nition 1.1.Finally, we will show that the solution found can be characterized by the comple-

mentarity system.

Theorem 2.9. Let � be in Mpb;0() and satisfy (2:7); then the following statements

are equivalent:(1) u is the solution of OP(A; �; ) and � is the associated obstacle reaction;(2) u∈K (); �∈M

p;+b;0 (); u is the entropy solution of (2:5) relative to �+ �; and∫

Tk(u− ’) d� ≤

∫Tk(v− ’) d�;

∀’∈W 1;p0 () ∩ L∞(); ∀v∈K ();

(2.10)

(3) u∈K (); �∈Mp;+b;0 (); u is the entropy solution of (2:5) relative to �+ �; and

u= �-a:e: in : (2.11)

Remark 2.10. Observe that if is Cp-q.e. upper bounded, we can consider in (2.10)’∈W 1;p

0 ()∩L∞(); ’ ≥ Cp-q.e. in and v=’, so that, taking into account thatu is the entropy solution of (2.5) relative to � + �; u satis�es∫

a(x;∇u)∇Tk(u− ’) dx ≤

∫Tk(u− ’) d�; (2.12)

which is quite similar to the usual variational formulation (1.1). Formula (2.12) wasjust obtained in [3] when the datum � is a function in L1(). This is an alternativeproof of the fact that a solution of OP(A; �; ) coincides with that given by Boccardoand Gallou�et in [4], and by Boccardo and Cirmi in [3].

At the end of Section 5 we will show that a solution of OP(A; �; ) is also arenormalized solution of the obstacle problem, according to the de�nition of Oppezziand Rossi [13,14].

3. Preparatory results

In Section 2 we mentioned the decomposition result for Radon measures that do notcharge the sets of p-capacity zero (see Theorem 2:1 in [5]). Starting from this we canobtain the following theorem.

Theorem 3.1. Let �∈Mpb;0(); then for every �¿ 0 there exist f� ∈L1() and

F� ∈W−1;p′() ∩Mb() such that � = f� + F� and

‖f�‖L1() ≤ ‖�‖Mb(); ‖F�‖W−1;p′ () ≤ �:

Page 8: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

206 C. Leone / Nonlinear Analysis 43 (2001) 199–215

Proof. The proof of Theorem 2:1 in [5] shows that if � is an element of Mpb;0(),

then � can be written as � = f + F , where f∈L1(); F ∈W−1;p′(), and

‖f‖L1() ≤ ‖�‖Mb(); ‖F‖W−1;p′ () ≤ 1:If we write � as � (�=�) and we apply the same result to �=�, we easily conclude.

Corollary 3.2. Let �n; �∈Mpb;0() be such that �n converges to � in the strong

topology of Mb(). Then there exist fn; f∈L1() and Fn; F ∈W−1;p′()∩Mb()

such that

fn → f strongly in L1();

Fn → F strongly in W−1;p′();

where �n = fn + Fn and � = f + F .

Proof. Thanks to the Theorem 3.1 the sequence �n−� can be decomposed as �n−�=gn + Gn, where

‖gn‖L1() ≤ ‖�n − �‖Mb(); ‖Gn‖W−1;p′ () ≤ �n;

with �n tending to zero as n tends to in�nity. Hence, if � = f + F , where f∈L1()and F ∈W−1;p′

(), we can de�ne fn :=f + gn and Fn :=F + Gn, and we con-clude.

Remark 3.3. By the previous corollary we obtain that the strong convergence inMb()is su�cient to apply Theorem 2.3 about continuous dependence of entropy solutions.

As concerning the approximation of measures in Mpb;0(), it is useful to state a

lemma, which is quite simple, but is proved here for the sake of completeness.

Lemma 3.4. Let �i ∈Mpb;0() (i = 1; 2) be such that �1 ≤ �2; then there exists

�ni ∈W−1;p′

() ∩ Mb() (i = 1; 2) such that �ni converges to �i strongly in Mb()

and �n1 ≤ �n

2.

Proof. It su�ces to consider the decomposition of �2 − �1 in the sense of Theorem2:2 in [7]:

0 ≤ �2 − �1 = h ;

where ∈W−1;p′() ∩ M+

b () and h∈L1(; ); h ≥ 0. On the other hand, �1 de-composes as �1 = h1 1, with 1 ∈W−1;p′

() ∩ M+b () and h1 ∈L1(; ). We take

�n1 :=Tn(h1) 1 and �n

2 :=Tn(h) + �n1, and we can conclude.

Remark 3.5. The proof of the previous lemma shows that, if �∈Mp;+b;0 (), then there

exists a nondecreasing sequence �n ∈W−1;p′() ∩ M+

b () such that �n converges to� strongly in Mb().

Page 9: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 207

Now we state a comparison principle concerning the entropy solutions.

Theorem 3.6. For i=1; 2; let ui be the entropy solution of (2:5) relative to �i ∈Mpb;0().

Suppose that �1 ≤ �2; then u1 ≤ u2 almost everywhere in .

Proof. The result is well known when �i ∈W−1;p′()∩Mb() (using the monotonicity

of A). Using Lemma 3.4 together with Theorem 2.3 we can conclude.

Now we give a result concerning the solutions of obstacle problems in the variationalframework.

Theorem 3.7. Let satisfy hypothesis (2:7) and let � in W−1;p′() ∩ Mb(). Let

u be the solution of VI(A; �; ) and � be the obstacle reaction relative to u. Then �satis�es (2:9).

Proof. By an approximation argument we may suppose that �+ and �− (the positiveand the negative part of �, respectively) belong to W−1;p′

() (as in Theorem 2:2 ofDall’Aglio and Leone [6]).Moreover we observe that we can consider, without loss of generality, the case of a

nonpositive obstacle. Suppose that we have proved inequality (2.9) in this case (with� = 0). Now, if is a general obstacle satisfying hypothesis (2.7), we consider thenew obstacle − u�, which is nonpositive, and the operator B(v) = −div(a(x;∇v +∇u�) − a(x;∇u�)), which is of the same type of A. The measure � associated withthe solution v of VI(B; � − �; − u�) satis�es the inequality (2.9). Now it is easy tocheck that the function u= v + u� is the solution of VI(A; �; ) and � is the measureassociated with u.The proof of (2.9) for a nonpositive obstacle (�= 0) consists of two steps.Step 1. Suppose that there exists a positive number � such that ≤ −� Cp-q.e.

in .Observe that the function u�+ , solution of

A(u�+) = �+ in

u�+ = 0 on @;

is nonnegative and hence greater than or equal to , belongs to W 1;p0 (), and

A(u�+)− � = �− in :

So u ≤ u�+ (by Characterization 1). Let us consider the truncated function T�(u�+−u)= (u�+ − u) ∧ �; it is nonnegative and less than or equal to �.Let us compute 〈�; T�(u�+ − u)〉 in the duality between W−1;p′

() and W 1;p0 ():

〈�; T�(u�+ − u)〉= 〈A(u); T�(u�+ − u)〉 − 〈�; T�(u�+ − u)〉= 〈A(u)− A(u�+); T�(u�+ − u)〉+ 〈�−; T�(u�+ − u)〉≤ 〈�−; T�(u�+−u)〉;

Page 10: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

208 C. Leone / Nonlinear Analysis 43 (2001) 199–215

where in the last inequality we used the monotonicity of A. Then, using the Cp-quasi-continuous representatives of functions in W 1;p

0 (), we can write∫T�(u�+ − u) d� ≤

∫T�(u�+ − u) d�−:

By the positivity of the measures � and �− and the properties of the truncated function,we deduce that

�({u�+ − u¿�}) ≤ �−():

When � tends to 0, the previous formula becomes

�({u�+ − u¿ 0}) ≤ �−():

Now, if we prove that �({u�+ = u}) = 0 we get the result. It su�ces to observe that�({u�+ = u}) ≤ �({u ≥ 0}) ≤ �({u¿ }), which is zero thanks to (1.3).Step 2: Suppose only that ≤ 0 Cp-q.e. in .Let us de�ne, for all �¿ 0, the sequence of functions � := − �, and consider

the solutions u� of VI(A; �; �). If we call �� the measures associated with u�, by theprevious step we get ��() ≤ �−().As � tends to zero the solution u� tends to u strongly in W 1;p

0 (), hence �� tendsto � strongly in W−1;p′

(), and this implies �() ≤ �−().

We compare now the two problems VI(A; �; ) and OP(A; �; ) when the forcingterm � is an element of W−1;p′

() ∩Mb().

Proposition 3.8. Let � be an element of W−1;p′()∩Mb() and let satisfy (2:7);

then the solution of VI(A; �; ) is the solution of OP(A; �; ).

Proof. Let u be the solution of VI(A; �; ) and � be the corresponding obstacle reaction,which is a nonnegative element of W−1;p′

(). Owing to Theorem 3.7 it belongs toMb(), hence to M

p;+b;0 (). Thus u is the entropy solution of (2.5) relative to � + �.

Take v∈K () the entropy solution of (2.5) corresponding to �+ �, where � belongsto M

p;+b;0 (). We want to prove that u ≤ v almost everywhere in . Consider �k an

approximation of � (see Lemma 3.4 and Remark 3.5):

�k ∈W−1;p′() ∩Mb(); �k ≥ 0; �k ↗ � strongly in Mb()

and vk the solution of (2.5) corresponding to � + �k . Thanks to Theorem 3.6 the se-quence vk is nondecreasing and tends to v in the sense of Theorem 2.3 (see Remark3.3). Thus, in particular, vk tends to v; Cp-q.e. in . Let k= ∧vk and denote the solu-tions of VI(A; �; k) by uk . Naturally, from the minimality of uk (see Characterization1), we deduce

uk ≤ vk a:e: in :

Since uk ≤ uk+1 Cp-q.e. in ; uk converges to a function u∗ Cp-q.e. in . Thus, u∗ ≥ Cp-q.e. in . It is easy to check that uk is bounded in W 1;p

0 (); thanks to Lemma1:2 in [9], u∗ is (the Cp-quasi continuous representative of) a function of W 1;p

0 ()

Page 11: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 209

and uk converges to u∗ weakly in W 1;p0 (). Moreover, it can be easily proved that

u∗ satis�es the variational formulation (1.1) and, consequently, u∗ coincides with u.Hence, passing to the limit, we conclude that u ≤ v a.e. in .

Remaining in the variational framework we state a result that generalizes Lemma7:3 in [6] to the nonlinear operator A. We omit the proof, because we may use thesame tools as in the linear case. In Section 4 we will extend it to general measuresin M

pb;0().

Lemma 3.9. Let satisfy (2:7) and let �1; �2 ∈W−1;p′()∩Mb(). Let �1 and �2 be

the reactions of the obstacle corresponding to the solutions u1 and u2 of VI(A; �1; )and VI(A; �2; ); respectively. If �1 ≤ �2; then �1 ≥ �2.

4. Proof of the existence theorem

In this section we will prove Theorem 2.6; to simplify the exposition, it is convenientto divide the proof into various lemma.

Lemma 4.1. Let satisfy (2:7) and let {�n}⊂Mpb;0() be a nondecreasing sequence

of measures converging to � strongly in Mb(); suppose that; for every n; problemOP(A; �n; ) has a solution un and let �n be the measure associated with it. If �n

converges to � strongly in Mb(); then there exists a solution u of OP(A; �; ); and� is the obstacle reaction relative to u.

Proof. First we observe that � and � belong to Mpb;0(), by the strong convergence of

�n and �n in Mb(). We can use Theorem 2.3 about continuous dependence of entropysolutions (see Remark 3.3). For every k ¿ 0, the sequence Tk(un) converges to Tk(u) inthe strong topology of W 1;p

0 (), u being the entropy solution of (2.5) relative to �+�.The strong convergence in W 1;p

0 () implies that, up to a subsequence, still denoted byun; Tk(un) converges to Tk(u); Cp-q.e. in ; thus Tk(u) ≥ Tk( ); Cp-q.e. in , and,letting k tend to in�nity, u ≥ ; Cp-q.e. in . Let us take now the entropy solutionv of (2.5) relative to � + �, with � in M

p;+b;0 (), and assume that v ≥ ; Cp-q.e.

in . Since � ≥ �n we can write � + � = �n + �n, with �n in Mp;+b;0 (). As un is

the solution of OP(A; �n; ), we obtain that un ≤ v a.e. in , and, passing to thelimit, u ≤ v a.e. in . In conclusion, u is the solution of OP(A; �; ), according toDe�nition 1.1.

Lemma 4.2. Let satisfy (2:7) and let {�n}⊂Mpb;0() be a nonincreasing sequence

of measures converging to � strongly in Mb(); suppose that; for every n; problemOP(A; �n; ) has a solution un and let �n be the measure associated with it. If �n

converges to � strongly in Mb(); then there exists a solution u of OP(A; �; ); and� is the obstacle reaction relative to u.

Page 12: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

210 C. Leone / Nonlinear Analysis 43 (2001) 199–215

Proof. As in the previous lemma, for every k ¿ 0; Tk(un) converges to Tk(u) in thestrong topology of W 1;p

0 (); u being the entropy solution of (2.5) relative to � + �;moreover u ≥ ; Cp-q.e. in . Let us take the entropy solution v of (2.5) relative to�+ �, with � in M

p;+b;0 (), and assume that v ≥ ; Cp-q.e. in . De�ne the sequence

vn as the entropy solution of (2.5) relative to �n+ �. Since �n ≥ �, by the comparisonprinciple of entropy solutions (see Theorem 3.6), we obtain vn ≥ v ≥ , and, thanks toTheorem 2.3 and Remark 3.3, vn converges to v a.e in . By the de�nition of ObstacleProblems (De�nition 1.1) we have that un ≤ vn a.e. in , and, passing to the limit,u ≤ v a.e. in . Thus u is the solution of OP(A; �; ).

Lemma 4.3. Let satisfy (2:7) and let �∈Mpb;0() with �− ∈W−1;p′

() ∩Mb().Then there exists a solution of OP(A; �; ); and the corresponding obstacle reactionsatis�es (2:9).

Proof. Since �+ is a nonnegative measure in Mpb;0(), thanks to Remark 3.5, there ex-

ists a nondecreasing sequence �+n ∈W−1;p′()∩M+

b (), converging to �+ in the strongtopology of Mb(). Let us de�ne �n := �+n −�−, so that �n ∈W−1;p′

()∩Mb() and�n is a nondecreasing sequence converging to � strongly in Mb(). By Proposition 3.8the solution un of VI(A; �n; ) is a solution of OP(A; �n; ). Thanks to Theorem 3.7the corresponding obstacle reaction �n satis�es

‖�n‖Mb() ≤ ‖(�n − �)−‖Mb(): (4.1)

As �n ≤ �n+1, by Lemma 3.9 �n ≥ �n+1. Hence, if we de�ne

�(B) := limn→∞ �n(B) for every B Borel set in ;

we know from measure theory that � is a nonnegative Borel measure, it is bounded be-cause �n is nonincreasing, and it is in M

pb;0(), since all �n are. Besides, �n converges

to � strongly in Mb(). By Lemma 4.1, there exists u solution of OP(A; �; ), and� is the measure associated with it. Moreover, passing to the limit in (4.1), weobtain (2.9).

Lemma 4.4. Let satisfy (2:7) and let �1; �2 ∈Mpb;0(); with �−

1 ; �−2 ∈W−1;p′

()∩Mb(). Let �1 and �2 be the reactions of the obstacles corresponding to the solutionsu1 and u2 of OP(A; �1; ) and OP(A; �2; ); respectively. If �1 ≤ �2; then �1 ≥ �2.

Proof. Since �+1 and �2− �1 are nonnegative measures in Mpb;0(), owing to Remark

3.5 there exist two nondecreasing sequences �+1; n; �n ∈W−1; p′()∩M+

b (), convergingstrongly in Mb() to �+1 and �2−�1, respectively. Let us de�ne �1; n :=�+1; n−�−

1 and�2; n :=�n + �1; n; thus, for i = 1; 2, �i; n belongs to W−1;p′

() ∩ Mb(), and �1; n isa nondecreasing sequence converging to �1 strongly in Mb(). For i = 1; 2, considerthe solution ui; n of VI(A; �i; n; ) and the corresponding obstacle reaction �i; n. Since�1; n ≤ �2; n, by Lemma 3.9 �1; n ≥ �2; n. Using the same arguments of the proof ofLemma 4.3, we obtain that, for i= 1; 2; �i; n converges to �i strongly in Mb(). Thuswe get the result.

Page 13: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 211

Proof of Theorem 2.6. Consider �− and approximate it in the strong topology ofMb() with a nondecreasing sequence �−

n ∈W−1;p′() ∩ M+

b () (see Remark 3.5).De�ning �n :=�+−�−

n , by Lemma 4.3 we can consider the solutions un of OP(A; �n; )and the measures �n associated with them. By Lemma 4.3 we know that �n satis�es

‖�n‖Mb() ≤ ‖(�n − �)−‖Mb(): (4.2)

Since �n is nonincreasing �n is nondecreasing, by Lemma 4.4. Thus, de�ning

�(B) := limn→∞ �n(B) for every B Borel set in ;

we know from measure theory that � is a nonnegative Borel measure, it is boundedbecause �n satis�es (4.2), and it is in M

pb;0(), since all �n are. Moreover, �n converges

to � strongly in Mb(). By Lemma 4.2, there exists u solution of OP(A; �; ), and� is the corresponding obstacle reaction. Moreover, passing to the limit in (4.2) weobtain (2.9).

Corollary 4.5. Let satisfy (2:7) and let �1; �2 ∈Mpb;0(). Let �1 and �2 be the

reactions of the obstacle corresponding to the solutions u1 and u2 of OP(A; �1; )and OP(A; �2; ); respectively. If �1 ≤ �2 then �1 ≥ �2.

Proof. It su�ces to consider �−1 and approximate it in the strong topology of Mb()

with a nondecreasing sequence �−1; n ∈W−1;p′

()∩M+b () (see Remark 3.5). De�ning

�1; n :=�+1 −�−1; n and �2; n :=�1; n+�2−�1, for i=1; 2 we have that �−

i; n ∈W−1;p′()∩

M+b (), and �i; n is a nonincreasing sequence converging to �i strongly in Mb().

Moreover, �1; n ≤ �2; n. We consider the solutions ui; n of OP(A; �i; n; ) and the corre-sponding obstacle reactions �i; n, which are nondecreasing in n and satisfy �1; n ≥ �2; n(by Lemma 4.4). Using the same tools of the proof of Theorem 2.6, we obtain that�i; n converges to �i strongly in Mb(). In conclusion �1 ≥ �2.

5. Proof of the stability result and of the complementarity conditions

In this section we will prove Theorems 2.7 and 2.9.

Proof of Theorem 2.7. Since �n converges to � strongly in Mb(), there exists a sub-sequence �nj of �n such that

+∞∑j=1

‖�nj − �‖Mb()¡+∞:

Let us de�ne �′j :=� −∑+∞

i=j (�ni − �)− and �′′j :=� +

∑+∞i=j (�ni − �)+; it is easy to

check that

�′j ↗ �; �′′

j ↘ � strongly in Mb()

and

�′j ≤ �nj ≤ �′′

j :

Page 14: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

212 C. Leone / Nonlinear Analysis 43 (2001) 199–215

For every j≥ 1, let u′j and u′′j be the solutions of OP(A; �′j; ) and OP(A; �′′

j ; ),respectively, and let �′j and �′′j be the corresponding associated measures. Reasoningas in the proof of Theorem 2.6, we obtain that u′j and u′′j converge to u (in the senseof Theorem 2.3), while �′j and �′′j converge to � in the strong topology of Mb(). Onthe other hand, thanks to Corollary 4.5, we have �′′j ≤ �nj ≤ �′j, so that �nj convergesto � strongly in Mb(). Finally, since the result does not depend on the subsequenceall �n converges to � strongly in Mb().By Remark 3.3 we can apply Theorem 2.3 to obtain the convergence of Tk(un) to

Tk(u) in the strong topology of W1;p0 (), for every k ¿ 0.

Proof of Theorem 2.9. We will divide the proof in three steps.Step 1: (1) ⇒ (2). It su�ces to prove (2.10). We proceed by an approximation

argument. Let �n ∈W−1;p′() ∩Mb() be such that �n converges to � in the strong

topology of Mb() (see Lemma 3.4). If we consider the solution un of VI(A; �n; )and the measure �n associated to it, thanks to (1.1) we have

〈�n; w − un〉 ≥ 0;∀w∈K () ∩W 1;p

0 (): (5.1)

We choose as test function in (5.1) w=un−Tk(un−’)+Tk(v−’), where ’∈W 1;p0 ()∩

L∞() and v∈K (), and we obtain

〈�n; Tk(un − ’)〉 ≤ 〈�n; Tk(v− ’)〉:Recalling that �∈M

pb;0(), thanks to Theorem 2.7 we can pass to the limit and we

get the result.Step 2: (2) ⇒ (3). Let t be a positive real number. Observe that the set {u −

¿ t} is Cp-quasi open, because u is Cp-quasi continuous and is Cp-quasi uppersemicontinuous. Thanks to Lemma 1:5 in [7] there exists an increasing sequence vnof nonnegative functions in W 1;p

0 () which converges to �{u− ¿t} Cp-q.e. in . Thefunction u− tvn belongs to K (), thus we can apply (2.10) with v=u− tvn; observingthat u− tvn ≤ u and � is a nonnegative measure, we get∫

Tk(u− ’) d�=

∫Tk(u− tvn − ’) d�

for every ’ in W 1;p0 () ∩ L∞().

Passing to the limit as n goes to in�nity, we get∫{u− ¿t}

Tk(u− ’) d�=∫{u− ¿t}

Tk(u− t − ’) d�: (5.2)

Now we choose ’= Th(u) in (5.2). Let us estimate the left-hand side of (5.2):∣∣∣∣∣∫{u− ¿t}

Tk(u− Th(u)) d�

∣∣∣∣∣ ≤ k�({|u|¿h});

Page 15: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 213

which tends to zero if h tends to in�nity (recall that u is �nite up to a set of p-capacityzero and � vanishes on the sets of p-capacity zero).In the same way we can split the integral in the right-hand side into two parts:

∫{u− ¿t}∩{|u|≤h}

Tk(−t) d�+∫{u− ¿t}∩{|u|¿h}

Tk(u− t − Th(u)) d�

as before, the second integral tends to zero if h goes to in�nity.In conclusion, we obtain Tk(t) �({u− ¿ t}) = 0, from which �({u− ¿ t}) = 0,

for every t ¿ 0. Letting t tend to zero, we get the result.Step 3: (3)⇒ (1). We have to prove that, for any � inMp;+

b;0 () such that the entropysolution v relative to �+ � belongs to K (), we have u ≤ v almost everywhere in .The procedure used in [2] allows to obtain that u satis�es (in particular)

∫a(x;∇u)∇Tk(u− ’)+ dx −

∫Tk(u− ’)+ d� =

∫Tk(u− ’)+ d�; (5.3)

for every ’ in W 1;p0 () ∩ L∞().

Similarly, v satis�es

−∫a(x;∇v)∇Tk(v− ’)− dx +

∫Tk(v− ’)− d� =−

∫Tk(v− ’)− d�: (5.4)

We choose ’= Th(v) in (5.3) and ’= Th(u) in (5.4), and we add the two equations.For the left-hand side we can use the same tools of the proof of uniqueness of

entropy solutions (see Theorem 3:3 in [5]); thus we obtain

∫{0¡u−v≤k; |u|≤h; |v|≤h}

(a(x;∇u)− a(x;∇v))∇(u− v) dx

≤ !k(h) +∫Tk(u− Th(v))+ d�−

∫Tk(v− Th(u))− d�; (5.5)

where !k(h) tends to zero if h goes to in�nity.Since � is a nonnegative measure we can rewrite (5.5) as

∫{0¡u−v≤k; |u|≤h; |v|≤h}

(a(x;∇u)− a(x;∇v))∇(u− v) dx

≤ !k(h) +∫Tk(u− Th(v))+ d�;

now we let h tend to in�nity, so that

∫{0¡u−v≤k}

(a(x;∇u)− a(x;∇v))∇(u− v) dx ≤∫Tk(u− v)+ d�:

Page 16: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

214 C. Leone / Nonlinear Analysis 43 (2001) 199–215

Observing that {u− v¿ 0}⊆{u− ¿ 0}, since v∈K (), the term in the right-handside is zero, by hypothesis (2.11). In conclusion

∫{0¡u−v≤k}

(a(x;∇u)− a(x;∇v))∇(u− v) dx ≤ 0;

for every k ¿ 0; by (2.3) this implies u ≤ v almost everywhere in .

Remark 5.1. If is Cp-q.e. upper bounded, we point out that the solution u ofOP(A; �; ) satis�es

∫a(x;∇u)∇(h(u)(u− ’)) dx ≤

∫h(u)(u− ’) d� (5.6)

for every ’∈W 1;p0 () ∩ L∞(), ’ ≥ Cp-q.e. in , and for every h∈C1c (R), h ≥ 0

in R. Indeed, working as in the proof of the �rst step of Theorem 2.9, we choose in(5.1) w=un−�h(un)(un−’), with h and ’ as before and � a positive constant such that�‖h‖∞ ≤ 1. Passing to the limit we obtain (5.6). Thus u is also a renormalized solutionof the obstacle problem according to the de�nition of Oppezzi and Rossi [13,14].

References

[1] H. Attouch, C. Picard, Problemes variationnels et theorie du potentiel non lineaire, Ann. Fac. Sci.To�ulouse 1 (1979) 89–136.

[2] L. Benilan, L. Boccardo, T. Gallou�et, R. Gariepy, M. Pierre, J.L. Vazquez, An L1 theory of existenceand uniqueness of nonlinear elliptic equations, Ann. Scuola Norm. Sup. Pisa 22 (1995) 240–273.

[3] L. Boccardo, G.R. Cirmi, Existence and uniqueness of solutions of unilateral problems with L1-data, inpreparation.

[4] L. Boccardo, T. Gallou�et, Probl�emes unilat�eraux avec donn�ees dans L1, C.R. Acad. Sci. Paris, S�erie I311 (1990) 617–619.

[5] L. Boccardo, T. Gallou�et, L. Orsina, Existence and uniqueness of entropy solutions for nonlinear ellipticequations with measure data, Ann. Inst. Henri Poincar�e 13 (1996) 539–551.

[6] P. Dall’Aglio, C. Leone, Obstacle problems with measure data, submitted for publication.

[7] G. Dal Maso, On the integral representation of certain local functionals, Ricerche Mat. 22 (1983)85–113.

[8] G. Dal Maso, Some necessary and su�cient conditions for the convergence of sequences of unilateralconvex sets, J. Funct. Anal. 62 (1985) 119–159.

[9] G. Dal Maso, A. Garroni, The capacity method for asymptotic Dirichlet problems, Asymptotic Anal.15 (1997) 299–324.

[10] D. Kinderlehrer, G. Stampacchia, An introduction to variational inequalities and their applications,Academic Press, New York, 1980.

[11] C. Leone, A. Porretta, Entropy solutions for nonlinear elliptic equations in L1, Nonlinear Anal. 32(1998) 325–334.

[12] J.L. Lions, Quelques M�ethodes de R�esolution des Probl�emes aux Limites Non lin�eaires, Dunod,Gauthier-Villars, Paris, 1969.

Page 17: Existence and uniqueness of solutions for nonlinear obstacle problems with measure data

C. Leone / Nonlinear Analysis 43 (2001) 199–215 215

[13] P. Oppezzi, A.M. Rossi, Renormalized solutions for divergence problems with L1 data, Atti Sem. Mat.Univ. Modena, in press.

[14] P. Oppezzi, A.M. Rossi, Unilateral problems with measure data, submitted for publication.[15] G. Stampacchia, Le probl�eme de Dirichlet pour les �equations elliptiques du second ordre �a coe�cients

discontinus, Ann. Inst. Fourier (Grenoble) 15 (1965) 189–258.