Finite-dimensional approximation for a class of elliptic obstacle problems

Finite-dimensional approximation for a class of elliptic obstacle problems

Nonlinear Analysis 52 (2003) 1745 – 1754 www.elsevier.com/locate/na Finite-dimensional approximation for a class of elliptic obstacle problems Yish...

132KB Sizes 2 Downloads 86 Views

Nonlinear Analysis 52 (2003) 1745 – 1754

www.elsevier.com/locate/na

Finite-dimensional approximation for a class of elliptic obstacle problems Yisheng Huanga;∗ , Yuying Zhoub a Department

of Mathematics, Suzhou University, Suzhou, Jiangsu 215006, People’s Republic of China of Mathematics, Yunnan Normal University, Kunming, Yunnan 650092, People’s Republic of China

b Department

Received 27 November 2001; accepted 13 May 2002

Abstract In this paper we study the /nite-dimensional approximation for a class of elliptic obstacle problems involving unbounded domains by using the Galerkin method. The convergence of the approximation is deduced. ? 2003 Elsevier Science Ltd. All rights reserved. Keywords: Obstacle problem; Finite-dimensional approximation; Galerkin cone

1. Introduction and preliminaries p ∞ Let  be an arbitrary open set in RN , D1; 0 () the completion of C0 () in the p 1=p and 1 ¡ p ¡ N . Let P(x) be a sign-changing function with norm {  |∇u| d x} P = P + − P − (as usual, P ± (x) := max{± P(x); 0}). We assume that P satis/es the hypothesis:

< lim x→y |x−y|p P2 (x) = (P) P ∈ L1 (); P + =P1 +P2 = 0, P1 ∈ LN=p () for every y ∈ , 0 and lim |x|→∞ |x|p P2 (x) = 0.

x∈

x∈

Denote E the space {u ∈ D01; p (): u E ¡ ∞}, where u pE =  − by p P |u| d x. Let K be the cone {u ∈ E: u(x) ¿ 0 a.e. for x ∈ }. 

 

|∇u|p d x +



Supported by NNSF of China (10161010) and NSF of Yunnan Province (1999A0003R).



Corresponding author. Tel.: +86-512-625-27063; fax: +86-512-6522-2691. E-mail address: [email protected] (Y. Huang).

0362-546X/03/$ - see front matter ? 2003 Elsevier Science Ltd. All rights reserved. PII: S 0 3 6 2 - 5 4 6 X ( 0 2 ) 0 0 2 9 2 - 4

1746

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

In this paper we are concerned with the following obstacle problem: /nd u ∈ K such that    f(x; u)(v − u) (1) |∇u|p−2 ∇u · ∇(v − u) ¿  P(x)|u|p−2 u(v − u) + 





∀u ∈ K, where  is the real parameter and f(x; t) is the CarathEeodory function satisfying some conditions to be speci/ed later. The solvability of variational inequalities like (1), relating to an elliptic operator, has been studied extensively in recent years by various methods, such as bifurcation, variational, super-subsolution approaches and the topological /xed point method, see for example [7–9,11] and the references therein. In this paper we study the solvability of (1) in another way, which is the Galerkin method, i.e., the /nite-dimensional approximation of problem (1). This method has been used by Isac [3,4] to consider the solvability of a class of abstract nonlinear complementarity problems and in [5] to the solvability of equations involving the p-Laplacian. However, the arguments used in [5] cannot be applied to our situation even if K = E (in this case (1) is in fact an equation) since  may not be bounded. Motivated by Isac [3,4], we /rst study the /nite-dimensional approximation for an abstract variational inequality involving an operator satisfying the condition (S)+ (see the de/nition below), and deduce the convergence of the approximation, especially our results extend and improve corresponding ones in [3,4]; then we prove our main theorem by applying the preceding results. Before presenting our main result, we make some assumptions on f in problem (1). Suppose that f is a CarathEeodory function satisfying (f1 ) there exist R ¿ 0 and 0 : 0 6 0 ¡ 1 such that f(x; s)s 6 0 P(x)|s|p holds ∀x ∈  and ∀s: |s| ¿ R, where 1 will be given in the following; ∗  (f2 ) |f(x; s)| 6 (x) + (x)|s|p−1 for a.e. (x; s) ∈  × R, where 0 6 (x) ∈ L(p ) () with p∗ = Np=(N − p) (the critical Sobolev exponent), (p∗ ) = p∗ =(p∗ − 1), 0 6 (x) ∈ LN=p () ∩ L∞ loc (). It was shown in [12], under the above condition (P), that the eigenvalue problem u ∈ D01; p ()   has the least eigenvalue 1 := inf {  |∇u|p d x: u ∈ E;  P|u|p d x = 1} satisfying 1 ¿ 0, where p u := div(|∇u|p−2 ∇u) is the p-Laplacian.  Let {wn } be a sequence of linear independent elements of E such that n∈N Vn =E, where Vn = span{w1 ; w2 ; : : : ; wn } for each n ∈ N. Then the main result obtained in this paper is as follows. −p u = P(x)|u|p−2 u;

Theorem 1.1. Let Kn := K ∩ Vn ∀n ∈ N. Suppose that f and P satisfy conditions (f1 ) and (f2 ) and condition (P), respectively. Then, (1) ∀: 0 ¡  ¡ 1 − 0 and ∀n ∈ N, problem (1) has a solution un ∈ Kn ; (2) there exists a subsequence {unj } of {un } such that unj → u0 ∈ K as j → ∞; moreover u0 is a solution of problem (1) for each : 0 ¡  ¡ 1 − 0 .

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

1747

2. Preliminaries In this section we will present some existence results for the abstract variational inequalities, which we shall use to prove our main theorem. Hence we give some notation /rst. Let X be a real Banach space and X∗ its dual space. Denote by · ; · the duality pairing between X and X∗ , and K is a cone of X (i.e., K is a closed convex subset of X such that K ⊂ K for all  ∈ R+ and K ∩ (−K) = {0}). Given a nonlinear mapping g : K → X∗ , the variational inequality problem VIP(g; K) is /nd x ∈ K; such that g(x); y − x ¿ 0 holds for all y ∈ K: If g(x) = x − T1 (x) + T2 (x), where T1 is a linear mapping and T2 is a nonlinear mapping, then the VIP(g; K) contains as a particular case the mathematical model used in the study of the post-critical equilibrium state of a thin plate subjected to unilateral conditions, see for example [3,6,10] and references therein. Denition 2.1. A Galerkin approximation of the cone K is a countable family of cones {Kn }n∈N satisfying the following properties: (1) Kn ⊂ K∀n ∈ N; (2) dim(Kn ) ¡ + ∞ ∀n ∈ N; (3) limn→∞ [Proj|Kn (x)] = x; ∀x ∈ K. We call a cone K to be a Galerkin cone if K has a Galerkin approximation. Denition 2.2. Let K be a cone of the Banach space X. A mapping f: K → X∗ is called to satisfy the weak Karamandian condition ((WK) for short) if, there exists a bounded closed convex subset A ⊂ K, such that for each x ∈ K \ A there exists z ∈ A satisfying f(x); x − z ¿ 0: Denition 2.3. Let {Kn }n∈N be a Galerkin approximation of a cone K ⊂ X, a mapping f : K → X∗ is called to satisfy the generalized Karamardian condition ((GK) for short) if, for any n ∈ N, there exists a bounded closed convex set Dn ⊂ Kn , such that for each x ∈ Kn \ Dn there exists y ∈ Dn satisfying f(x); x − y ¿ 0: Remark 2.1. De/nitions 2.1–2.3 can be seen in [3] if X is a Hilbert space or in [13] for X being a Banach space. Let X and K be as in De/nition 2.2. Recall that a mapping f : X → X∗ is called to satisfy the Karamardian condition if there exists a compact convex set C ⊂ K such that for each x ∈ K \ C there exists z ∈ C satisfying f(x); x − z ¿ 0, see [3] for more details. The following de/nition can be seen in [4].

1748

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

Denition 2.4. A mapping f : X → X∗ is called to satisfy the condition (S)+ if, {un }n∈N ⊂ X satis/es un * u and lim sup f(un ), un − u 6 0. Then there exists a subsequence {unk } such that unk → u. Lemma 2.1. Suppose that M is a >nite-dimensional Banach space, K0 is a cone of M, operator G: K0 → M is continuous satisfying the condition (WK). Then VIP(G; K0 ) has a solution. Proof. Let {u1 ; u2 ; : : : ; um } ⊂ K0 . Denote by B the convex hull of A∪{u1 ; u2 ; : : : ; um }, where A ⊂ K0 is the set in the de/nition of the condition (WK). Clearly, B is a bounded closed convex subset of K0 . Consequently, there exists v0 ∈ B such that G(v0 ); u − v0  ¿ 0

∀u ∈ B

(2)

(cf. [1]). We claim that v0 ∈ A. In fact, if v0 ∈ B \ A, by the assumption that the operator G satis/es the condition (WK), there exists z ∈ A ⊂ B such that G(v0 ); v0 − z ¿ 0, which contradicts (2). Hence v0 ∈ A. De/ne a multivalued mapping H : K0 → 2A as ∀u ∈ K0 ;

H (u) := {w ∈ A: G(w); u − w ¿ 0}:

It follows, from (2) and the fact that v0 ∈ A, that for all u ∈ B, v0 ∈ H (u). Therefore, v0 ∈∩m i=1 H (ui ), where {u1 ; : : : ; um } is the subset of K0 which appears in the beginning of the proof of this lemma. So {H (u): u ∈ K0 } has the /nite intersection property. Consequently, by the compactness of A, ∩u∈K0 H (u) = ∅, i.e., there exists w0 ∈ A such that w0 ∈ H (u) for all u ∈ K0 . We see that there exists {wn } ⊂ H (u) satisfying wn → w0 as n → ∞. It follows by the de/nition of H (u) that G(wn ); u − wn  ¿ 0 and then G(w0 ); u − w0  ¿ 0

∀u ∈ K0 ;

which completes the proof of this lemma. Theorem 2.1. Let X be a re@exive Banach space, X∗ the dual space. Suppose that K is a Galerkin cone of X, {Kn }n∈N is a Galerkin approximation of K, G; F : K → X∗ are continuous operators with G satisfying the condition (S)+ , F being compact. Assume that G − F satis>es the condition (GK) for an equibounded family {Dn }n∈N . Then (1) for each n ∈ N, VIP(G − F; Kn ) has a solution un ∈ Dn ; (2) if {G(un )} is bounded, then there exists a subsequence {unj } of {un } such that unj → u∗ ∈ K, and u∗ is a solution of VIP(G − F; K). Proof. (1) It follows by Lemma 2.1 that VIP(F − G; Kn ) has a solution for each n ∈ N, that is, for each n ∈ N there exists un ∈ Kn such that G(un ) − F(un ); z − un  ¿ 0

∀z ∈ Kn :

(3)

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

1749

Noting that G − F satis/es the condition (GK) for {Dn }, we see that for each n ∈ N, un ∈ Dn . (2) It implies from the equiboundedness of {Dn } that there exists a subsequence {unj } of {un } such that unj * u∗ ∈ K as j → ∞ (since K is weakly closed). Set uˆ n = Proj|Kn u∗

∀j ∈ N;

then, uˆ n ∈ Kn and uˆ n → u∗ . Inequality (3) yields that G(un ) − F(un ); uˆ n − un  ¿ 0; hence, G(unj ); unj − uˆ nj  6 F(unj ); unj − uˆ nj : Since unj * u∗ and F is compact, we get that {F(unj )} has a subsequence, without loss of the generality, we still denote by {F(unj )}, which converges to F(u∗ ) as j goes to in/nity. It implies from the above inequality that lim supG(unj ); unj − uˆ nj  6 0:

(4)

j→∞

By the boundedness of {G(unj )}, we may assume that {G(unj )} weakly converges to some l ∈ X∗ as j goes to in/nity. Hence, lim G(unj ); uˆ nj − u∗  = 0:

(5)

j→∞

We observe that G(unj ); unj − u∗  6 G(unj ); unj − uˆ nj  + G(unj ); uˆ nj − u∗ : It follows from inequalities (4) and (5) that lim supG(unj ); unj − u∗  6 0: j→∞

Note that G satis/es the condition (S)+ , hence there exists a subsequence of {unj }, we may denote by {unj }, such that unj → u∗ . We complete the proof by showing that u∗ is a solution of VIP(G − F; K). In fact, let z ∈ K be an arbitrary element, we denote by zn = Proj|Kn z. It implies from (3) that G(unj ) − F(unj ); znj − unj  ¿ 0

∀n ∈ N:

Letting j → ∞ in the above inequality and noting the simple fact that {znj − unj } tends to z − u∗ as j tends to ∞, we have G(u∗ ) − F(u∗ ); z − u∗  ¿ 0; that is, u∗ solves the VIP(G − F; K). Let r → ’(r) be a strictly increasing continuous function that maps R+ into R+ and satis/es limr→+ ∞ ’(r) = + ∞. If · ∗ denotes the dual norm of the given norm ·

1750

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

on X, the duality mapping J between X and X∗ with respect to ’ is given by J (0) = 0; J (u) = {f ∈ X∗ : f ∗ = ’( u ) and f; x = f ∗ u ; u = 0}: If ’ is the identity mapping of R+ , then J is the normalized duality mapping. In particular, if X is a Hilbert space, the normalized duality mapping is the identity mapping of X. If X is a reOexive Banach space, by the results due to Asplund, it is true that X can be renormed so that X and X∗ are both locally uniformly convex (LUC in short); in this case, J is single valued. Hence, we may assume in what follows that X and X∗ are LUC. From [3], J has the following properties: (P1 ) J (x)−J (y); x−y ¿ (’( x )−’( y ))( x − y ); x; y ∈ X. In particular, J (x)− J (y); x − y ¿ 0, i.e., J is monotone; (P2 ) J is continuous; (P3 ) J satis/es the condition (S)+ . By Theorem 2.1 we can easily get Corollary 2.1. Let K be a Galerkin of X, {Kn}n∈N a Galerkin approximation of K, J : X → X∗ the duality mapping given above, F : K → X∗ a mapping. Suppose that (i) F is strongly continuous; (ii) J − F satis>es the condition (GK) with an equibounded family {Dn }n∈N . Then VIP(J − F; K) has a solution. Corollary 2.2. Let K; J be as in Corollary 2.1. Suppose that S; T : K → X∗ with S being bounded and T being compact. Furthermore, assume that (i) J − S satis>es the condition (S)+ ; (ii) J − S − T satis>es the condition (GK) with an equibounded family {Dn }n∈N . Then VIP(J − S − T; K) has a solution. Remark 2.2. It is clear that if X is a Hilbert space, then the duality mapping J given above is the identity mapping of X; moreover, notice that in Corollary 2.1 we do not need a condition that “the function x → x; (J − F)(x) from K into R is lower semi-continuous”, therefore, Corollary 2.1 extends Theorem 1 in [3]. If X is a Hilbert space, J; S; T satisfy all conditions of Corollary 2.2, and (S + T )(K) ⊂ K, then by Corollary 2.2 we get that S + T has a /xed point in K. Hence, Corollary 2.2 improves Theorem 5 in [4].

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

1751

3. Proof of the main theorem Proof of Theorem 1.1. Clearly, K = ∪n∈N Kn . De/ne operators L; G1 ; G2 ; F: K → E∗ as  L(u); v = |∇u|p−2 ∇u∇v d x 

 G1 (u); v = P − (x)|u|p−2 uv d x 

∀u; v ∈ K:

 G2 (u); v = P + (x)|u|p−2 uv d x 

 F(u); v = f(x; u)v d x 

Easily, we see that L; G1 ; G2 ; F are well de/ned; moreover G2 ; F are completely continuous (cf. [2,12]). First we shall prove that for each  ¿ 0, L + G1 satis/es the condition (S)+ . To do this, we assume that un * u in K and lim supn→∞ L(un ) + G1 (un ); un − u 6 0. Note that limn→∞ L(u) + G1 (u); un − u = 0, hence lim sup (L + G1 )(un ) − (L + G1 )(u); un − u 6 0: n→∞

It follows by the HSolder inequality that for each  ¿ 0 (L + G1 )(un ) − (L + G1 )(u); un − u  = (|∇un |p−2 ∇un − |∇u|p−2 ∇u)(∇un − ∇u) 

 +



P − (|un |p−2 un − |u|p−2 u)(un − u)

 ¿



p

|∇un |

 ×

+

¿ 0;



p

p

P |un |







 −

1=p 

|∇u|



1=p

(p−1)=p 

|∇u|

p

(p−1)=p

P |un |



p

 −







1=p

|∇un |



 −

p



×

(p−1)=p









p

(p−1)=p 

P |u|

p

P |u|

1=p 

(6)

1752

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

which and (6) yield that for each  ¿ 0,     |∇un |p d x → |∇u|p d x; P − |un |p d x → P − |u|p d x; 







and then un E → u E . Therefore, un → u in K, which shows that J + G1 satis/es the condition (S)+ for each  ¿ 0. Next we claim that there exists a real number . ¿ 0 such that for each : 0 ¡  ¡ 1 − 0 L(u) − Q(u) − F(u); u ¿ 0

∀ u ∈ K \ D. ;

(7)

where the operator Q := G2 − G1 and D. := {u ∈ K: u E 6 .}. By contradiction, we may assume that there exists {un } ⊂ K with un E → ∞ such that L(un ) − Q(un ) − F(un ); un  6 0; that is,  

|∇un |p d x 6 

 

P(x)|un |p d x +

 

f(x; un )un d x:

Let vn := un = un E , then {vn } is bounded in K. We may assume that vn * v0 ∈ K and vn (x) → v0 (x) a.e. in . By the preceding inequality:    f(x; un ) p p |∇vn | d x 6  P(x)|vn | d x + v d x: (8) p−1 n   un E  For each n ∈ N, we de/ne n; R := {x ∈ : |un (x)| 6 R}, where R is the positive real number given in (f1 ). We split the second integral on the right-hand side of (8) into integrals over  \ n; R and n; R , then by (f1 ), for each n ∈ N we have   f(x; un ) f(x; un ) v d x = p un d x p−1 n u \n; R \n; R un E n E  6 0 



P(x)|vn |p d x − 0

 n; R

P(x) p u d x: un pE n

Since P ∈ L1 (), n; R (P(x)= un pE )unp d x → 0 as n → ∞. Using the argument in the proof of [2, Lemma 4.2], we denote n; R (K) := n; R ∩ B(0; K), where B(0; K) is the ball centered at 0 and having radius K, i.e., B(0; K)={x ∈ RN : |x| ¡ K}. By (f2 ), the HSolder inequality and the Sobolev inequality,   |f(x; un )| (x)|un | + (x)|un |p |v | d x 6 dx n p−1 un pE n; R un E n; R

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

 6

6

(x)

n; R



|v | d x + p−1 n

un E

c1  L(p∗ ) () un p−1 E

n; R \n; R (K)

p

(x)|vn | d x +

+ c2  LN=p (n; R \n; R (K)) +

 n; R (K)

1753

(x)|un |p dx un pE

c3 (K) un pE

for some positive constants c1 , c2 and c3 (K) (dependent of K). Now ∀2 ¿ 0 we can choose K so that the second term on the right-hand side of the above inequality is 6 2=2, and then N0 large enough such that the sum of the /rst and the last terms on the right-hand side of the above inequality is 6 2=2 if n ¿ N0 . Note that P = P + − P − . We  see, by (8), the de/nition of the /rst eigenvalue 1 , the fact that the mapping u →  P + (x)|u|p d x is weakly continuous (cf. [12]), the weakly lower semicontinuity of the norm, that      + 0 |∇v0 |p d x 6 ( + 0 ) P(x)|v0 |p d x 6 |∇v0 |p d x ¡ |∇v0 |p d x; 1     a contradiction. Accordingly, we have shown that for each  ¡ 1 − 0 , Inequality (7) holds. Therefore, L − Q − F satis/es the condition (WK) for each  ¡ 1 − 0 , and so L − Q − F satis/es the condition (GK) for each  ¡ 1 − 0 . Finally, by applying Theorem 2.1, we complete the proof. Acknowledgements The authors would like to thank the referee for his helpful comments. References [1] P. Hartman, G. Stampacchia, On some nonlinear elliptic diTerential functional equations, Acta Math. 115 (1966) 271–310. [2] Y.S. Huang, Positive solutions of quasilinear elliptic equations, Topol. Meth. Nonlinear Anal. 12 (1998) 91–107. [3] G. Isac, Nonlinear complementarity problem and Galerkin method, J. Math. Anal. Appl. 108 (1985) 563–574. [4] G. Isac, On an Altman type /xed point theorem on convex cones, Rocky Mountain J. Math. 25 (1995) 701–714. [5] P. Jebelean, Finite-dimensional approximation and coerciveness in a problem with p-Laplacian, Nonlinear Anal. TMA 33 (1998) 253–259. [6] R.S. Kubrusly, J.T. Oden, Nonlinear eigenvalue problems characterized by variational inequalities with applications to the postbuckling analysis of unilaterally-supported plates, Nonlinear Anal. TMA 5 (1981) 1265–1284. [7] V.K. Le, On global bifurcation of variational inequalities and applications, J. DiTerential Equations 141 (1997) 254–294. [8] V.K. Le, Subsolution-supersolution method in variational inequalities, Nonlinear Anal. 45 (2001) 775–800. [9] V.K. Le, K. Schmitt, Minimization problems for noncoercive functionals subject to constraints, Tran. Am. Math. Soc. 347 (1995) 4485–4513.

1754

Y. Huang, Y. Zhou / Nonlinear Analysis 52 (2003) 1745 – 1754

[10] R.C. Riddel, Eigenvalue problems for nonlinear elliptic variational inequalities on a cone, J. funct. Anal. 26 (1977) 333–355. [11] A. Szulkin, Positive solutions of variational inequalities: a degree theoretic approach, J. DiTerential Equations 57 (1985) 90–111. [12] A. Szulkin, M. Willem, Eigenvalue problems with inde/nite weight, Studia. Math. 135 (1999) 191–201. [13] M. ThEera, Existence results for the nonlinear complementarity problem and application to nonlinear analysis, J. Math. Anal. Appl. 154 (1991) 572–586.