VOL. 81 1999 NO. 1
MULTIPLE SOLUTIONS FOR NONLINEAR DISCONTINUOUS ELLIPTIC PROBLEMS NEAR RESONANCE
BY
NIKOLAOS C. K O U R O G E N I S AND
NIKOLAOS S. P A P A G E O R G I O U (ATHENS)
Abstract. We consider a quasilinear elliptic eigenvalue problem with a discontinuous right hand side. To be able to have an existence theory, we pass to a multivalued problem (elliptic inclusion). Using a variational approach based on the critical point theory for locally Lipschitz functions, we show that we have at least three nontrivial solutions when λ → λ1 from the left, λ1 being the principal eigenvalue of the p-Laplacian with the Dirichlet boundary conditions.
1. Introduction. In a recent paper (see Kourogenis–Papageorgiou [10]), we examined quasilinear elliptic problems at resonance with discontinuous right hand side and we proved the existence of a nontrivial solution. In the present paper we examine quasilinear elliptic problems near resonance with discontinuities. Semilinear problems near resonance with a continuous right hand side were studied by Mawhin–Schmitt [12], [13], Chiappinelli–De Figueiredo [4] and Chiappinelli–Mawhin–Nugari [5]. In [13] the equation under consideration is an ordinary differential equation (i.e. N = 1) and the authors employ a sign condition to establish the existence of three non- trivial solutions. In [12] an analogous abstract result with the sign condition replaced by a Landesman–Lazer type hypothesis can be found. The authors obtain two solutions, one negative and the other positive. In [4], a similar multiplicity result is shown under the hypothesis of linear growth as x →-
∞. In all three papers the approximation of the first eigenvalue is from the left. In [5] the parameter λ is to the right of the first eigenvalue. Again the authors prove the existence of two solutions, one of them positive. All the aforementioned papers use bifurcation theory.
1991 Mathematics Subject Classification: Primary 35J20.
Key words and phrases: multiple solutions, discontinuous function, elliptic inclusion, first eigenvalue, p-Laplacian, Rayleigh quotient, nonsmooth Palais–Smale condition, coer- cive functional, Clarke subdifferential, critical point, generalized directional derivative.
The first named author supported by the General Secretariat of Research and Tech- nology of Greece.
[89]
Recently there appeared the interesting works of Ambrosetti–Garcia Azorero–Peral [1] and Ramos–Sanchez [15]. The authors of [1] study (Sec- tion 4) the existence of positive solutions for the eigenvalue problem −∆
px = λf (x), x|
Γ= 0, where f (x) ≃ x
p−1near 0 and infinity, and they prove a bifurcation result, both from zero and from infinity. Their approach is based on degree-theoretic arguments. The work of Ramos–Sanchez [15] is closer to ours. They study the semilinear version (i.e. p = 2) of our problem with the right hand side function f (z, x) continuous in both variables. In Section 2 of [15], they examine the case when λ approaches λ
1(the first eigenvalue of (−∆, H
01(Z))) from the left. In Theorem 2.6, they prove the existence of three nontrivial solutions (as we do here in Theorem 7). As we already said, in their problem f (z, x) is jointly continuous, they assume that F is bounded below on Z × R
+, they have a hypothesis similar to our hypothesis H(f )(iii), but they impose the asymptotic condition on f instead of F (the potential function corresponding to f ) as we do, and they also have hypoth- esis H(f )(iv). Their approach is different from ours and uses the theory of elliptic variational inequalities. It should be mentioned that in Section 4 of [15] they also study the case when λ approaches λ
1from the right.
Our approach here is variational and is based on the critical point theory for nonsmooth locally Lipschitz functionals, as developed by Chang [3]. For the convenience of the reader, in the next section we outline the basic aspects of this theory.
2. Preliminaries. Chang’s critical point theory for locally Lipschitz functionals is based on the subdifferential theory of Clarke [6] for such func- tionals. In the previous paper [10], in Section 2, we presented the basic definitions and facts from these theories that are needed in our analysis.
The notation introduced there will also be used in this paper. So if X is a Banach space and f : X → R is locally Lipschitz we define the generalized directional derivative
f
0(x; h) = lim
x′→x λ↓0
f (x
′+ λh) − f (x
′) λ
and the generalized subdifferential
∂f (x) = {x
∗∈ X
∗: (x
∗, h) ≤ f
0(x; h) for all h ∈ X}.
A point x ∈ X is a critical point of f if 0 ∈ ∂f (x). We say that f
satisfies the nonsmooth Palais–Smale condition (nonsmooth PS-condition)
if any sequence {x
n}
n≥1along which {f (x
n)}
n≥1is bounded and m(x
n) =
inf{kx
∗k : x
∗∈ ∂f (x
n)} → 0 as n → ∞, has a strongly convergent sub-
sequence. This notion generalizes the classical one for C
1-functionals (see Rabinowitz [14]).
Consider the nonnegative p-Laplacian (2 ≤ p < ∞) differential operator
−∆
px = − div(kDxk
p−2Dx)
with Dirichlet boundary conditions; we use the notation (−∆
p, W
01,p(Z)).
The first (principal) eigenvalue λ
1of this operator is the least real number λ for which the nonlinear elliptic problem
(1)
− div(kDx(z)k
p−2Dx(z)) = λ|x(z)|
p−2x(z) a.e. on Z, x|
Γ= 0,
has a nontrivial solution. The first eigenvalue λ
1is positive, isolated and sim- ple (i.e. the associated eigenfunctions are constant multiples of each other).
Furthermore, we have the following variational characterization of λ
1> 0 (Rayleigh quotient):
λ
1= min[kDxk
pp/kxk
pp: x ∈ W
01,p(Z)].
This minimum is realized at the normalized eigenfunction u
1. Note that if u
1minimizes the Rayleigh quotient, then so does |u
1| and so we infer that the first eigenfuction u
1does not change sign on Z. In fact, we can show that u
16= 0 a.e. on Z and so we may assume that u
1(z) > 0 a.e. on Z.
Also, using nonlinear elliptic regularity theory (see Tolksdorf [16]), we can show that u
1∈ C
1,αfor some α > 0. For details we refer to Lindqvist [11]. The Lyusternik–Schnirelmann theory gives, in addition to λ
1, a whole strictly increasing sequence {λ
n}
n≥1of positive numbers for which there exist nontrivial solutions of the eigenvalue problem (1). In other words, the spectrum σ(−∆
p) of (−∆
p, W
01,p(Z)) contains at least these points.
However, in general, nothing is known about the possible existence of other points in σ(−∆
p) ⊆ [λ
1, ∞) ⊆ R
+. Nevertheless, we can define
µ = inf{λ > 0 : λ is an eigenvalue of (−∆
p, W
01,p(Z)), λ 6= λ
1}.
Because λ
1> 0 is isolated, we have µ > λ
1and if V is a topological com- plement of X = hu
1i = Ru
1, then
µ
V= inf[kDvk
pp/kvk
pp: v ∈ V ] > λ
1, µ = sup
V
µ
V.
The following theorem is due to Chang [6] and extends to a nonsmooth setting the well-known Mountain Pass Theorem due to Ambrosetti–Rabino- witz [2].
Theorem 1. If X is a reflexive Banach space, R : X → R is a locally Lipschitz functional which satisfies the PS-condition and x
0, x
1∈ X, β ∈ R and ̺ > 0 are such that
(i) kx
0− x
1k > ̺,
(ii) max[R(x
0), R(x
1)] < β ≤ inf[R(u) : ku − x
0k = ̺],
then R(·) has a critical point x ∈ X such that c = R(x) ≥ β. Moreover , c can be characterized by the following min-max principle :
c = inf
γ∈Γ
sup
t∈[0,1]
R(γ(t)) where Γ = {γ ∈ C([0, 1], X) : γ(0) = x
0, γ(1) = x
1}.
3. Auxiliary results. Let Z ⊆ R
Nbe a bounded domain with C
1- boundary Γ . We consider the following quasilinear elliptic problem:
(2)
− div(kDx(z)k
p−2Dx(z)) − λ|x(z)|
p−2x(z)
= f (z, x(z)) a.e. on Z, x|
Γ= 0, 2 ≤ p < ∞.
Since we do not assume that f (z, ·) is continuous, problem (2) need not have a solution. To develop a reasonable existence theory, we need to pass to a multivalued version of (2) by, roughly speaking, filling in the gaps at the discontinuity points of f (z, ·). For this purpose we introduce the following two functions:
f
1(z, x) = lim
x′→x
f (z, x
′) = lim
δ↓0
ess inf
|x′−x|<δ
f (z, x
′) f
2(z, x) = lim
x′→x
f (z, x
′) = lim
δ↓0
ess sup
|x′−x|<δ
f (z, x
′).
We introduce b f (z, x) = {y ∈ R : f
1(z, x) ≤ y ≤ f
2(z, x)} and instead of (2) we consider the following differential inclusion:
(3)
− div(kDx(z)k
p−2Dx(z)) − λ|x(z)|
p−2x(z)
∈ b f (z, x(z)) a.e. on Z, x|
Γ= 0, 2 ≤ p < ∞.
Problem (3) is a multivalued approximation of (2), which captures the dis- continuity features of f (z, ·) and permits the development of an existence theory. Of course, if f (z, ·) is continuous for almost all z ∈ Z, then the two problems coincide.
We introduce now our hypotheses on f :
H(f ): f : Z × R → R is a measurable function such that
(i) f
1, f
2are N-measurable functions (i.e. for every measurable function x : Z → R, the functions z → f
1(z, x(z)) and z → f
2(z, x(z)) are measurable; superpositional measurability);
(ii) for every r > 0, there exists a
r∈ L
∞(Z) such that |f (z, x)| ≤ a
r(z) for almost all z ∈ Z and all |x| ≤ r;
(iii) if F (z, x) =
Tx
0
f (z, r) dr then lim
|x|→∞pF (z, x)/|x|
p= 0 uni-
formly for almost all z ∈ Z;
(iv) lim
|t|→∞T
Z
F (z, tu
1(z)) dz = ∞;
(v) lim
x→0pF (z, x)/|x|
p< −λ
1and lim
x→0F (z, x)/|x|
p> −∞
uniformly for almost all z ∈ Z.
Remark . Hypotheses H(f ) are more general than the corresponding ones used by Kourogenis–Papageorgiou [10]. Indeed, while H(f )(i) is com- mon in both papers, H(f )(ii) is weaker than the one of [10], where it is assumed that |f (z, x)| ≤ a(z) for almost all z ∈ Z and all x ∈ R, with a ∈ L
∞(Z). Moreover, H(f )(iii) is weaker than the corresponding one in [10], since it does not imply that lim
x→±∞f (z, x) exist and are finite for almost all z ∈ Z. Finally, H(f )(v) is common in both papers and is needed in order to apply Theorem 1.
We introduce the energy functional R
λ: W
01,p(Z) → R defined by R
λ(x) = 1
p kDxk
pp− λ
p kxk
pp−
\
Z
F (z, x(z)) dz.
Since J : W
01,p(Z) → R defined by J(x) =
T
Z
F (z, x(z)) dz is locally Lipschitz (see Chang [3]) and
x → 1
p kDxk
pp→ λ p kxk
ppare continuous convex functions on W
01,p(Z), hence locally Lipschitz on W
01,p(Z), it follows that R
λ(·) is locally Lipschitz.
Proposition 2. If hypotheses H(f ) hold and λ < λ
1, then R
λ(·) is coercive.
P r o o f. Suppose not. Then we can find {x
n}
n≥1⊆ W
01,p(Z) such that kx
nk
1,p→ ∞ as n → ∞, and R
λ(x
n) ≤ M for all n ≥ 1. We have
1
p kDx
nk
pp− λ
p kx
nk
pp−
\
Z
F (z, x(z)) dz ≤ M.
By H(f )(iii), given ε > 0 we can find M
1= M
1(ε) > 0 such that F (z, x) ≤ (ε/p)|x|
pfor almost all z ∈ Z and all |x| > M
1. Also, for al- most all z ∈ Z and all |x| ≤ M
1we have |F (z, x)| ≤ a
1(z) with a
1∈ L
∞(Z) (see hypothesis H(f )(ii)). So finally, we can say that for almost all z ∈ Z and all x ∈ R we have F (z, x) ≤ a
1(z) + (ε/p)|x|
p. Hence
1
p kDx
nk
pp− λ
p kx
nk
pp− ka
1k
1− ε
p kx
nk
pp≤ M, that is,
1
p kDx
nk
pp≤ M + ka
1k
1+ λ + ε
p kx
nk
pp.
Since kx
nk
1,p→ ∞, we have kDx
nk
p→ ∞ and so from the last inequality it follows that kx
nk
p→ ∞. Let y
n= x
n/kx
nk
p, n ≥ 1. Dividing by kx
nk
ppwe obtain
(4) 1
p kDy
nk
pp≤ M kx
nk
pp+ ka
1k
1kx
nk
pp+ 1
p (λ + ε),
hence {y
n}
n≥1⊆ W
01,p(Z) is bounded (by Poincar´e’s inequality). Thus by passing to a subsequence if necessary, we may assume that y
n w→ y in W
01,p(Z), y
n→ y in L
p(Z), y
n(z) → y(z) a.e. on Z as n → ∞ and
|y
n(z)| ≤ h
1(z) a.e. on Z with h
1∈ L
p(Z).
Passing to the limit in (4) and using the fact that kDyk
p≤ lim kDy
nk
p(weak lower semicontinuity of the norm functional), we obtain 1
p kDyk
pp≤ λ + ε p .
Let ε > 0 be such that λ + ε < λ
1(recall that by hypothesis λ < λ
1).
Also, since ky
nk
p= 1, n ≥ 1, we have kyk
p= 1, i.e. y 6= 0. So we can write kDyk
pp< λ
1kyk
pp,
which contradicts the variational characterization of λ
1(Rayleigh quotient, see Section 2). This proves the coercivity of R
λ(·).
Let X = hu
1i = Ru
1and V a topological complement. Then W
01,p(Z) = X ⊕ V .
Proposition 3. If hypotheses H(f ) hold, then there exists β < 0 such that R
λ(v) ≥ β for all v ∈ V and all 0 < λ < λ
1.
P r o o f. From Section 2 we know that there exists µ > λ
1such that for all v ∈ V we have
kDvk
pp≥ µkvk
pp. Also, since 0 < λ < λ
1, we have
R
λ(v) ≥ 1
p kDvk
pp− λ
1p kvk
pp−
\
Z
F (z, v(z)) dz.
From the proof of Proposition 2 we know that given ε > 0 we can find a
1∈ L
∞(Z) (depending on ε > 0) such that for almost all z ∈ Z and all x ∈ R we have
F (z, x) ≤ a
1(z) + ε p |x|
p. So we can write
R
λ(v) ≥ 1
p kDvk
pp− λ
1p kvk
pp−ka
1k
1− ε
p kvk
pp≥ 1 p
1− λ
1+ ε µ
kDvk
pp−ka
1k
1.
Choose ε > 0 so that λ
1+ ε < µ. From the last inequality it follows that R
λ(·) is coercive on V , uniformly in 0 < λ < λ
1. Thus we can find β < 0 such that R
λ(v) ≥ β for all v ∈ V and all 0 < λ < λ
1.
Proposition 4. If hypotheses H(f ) hold, then there exists b t > 0 such that for every |t| > b t there exists δ
t> 0 such that R
λ(±tu
1) < β for all λ ∈ (λ
1− δ, λ
1).
P r o o f. We have R
λ(tu
1) = |t|
pp kDu
1k
pp− λ|t|
pp ku
1k
pp−
\
Z
F (z, tu
1(z)) dz
= |t|
pp kDu
1k
pp− λ|t|
pλ
1p kDu
1k
pp−
\
Z
F (z, tu
1(z)) dz.
By H(f )(iv), we can find b t > 0 so large that for every |t| > b t,
−(β + 1) <
\
Z
F (z, ±tu
1(z)) dz.
Let δ = δ
t= λ
1p/(t
pkDu
1k
pp) > 0. Then for λ
1− δ < λ < λ
1we have R
λ(±tu
1) < t
pδ
λ
1p kDu
1k
pp+ β − 1 = β.
The next proposition shows that R
λ(·) satisfies a sort of nonsmooth Palais–Smale condition over closed and convex subsets of W
01,p(Z).
Proposition 5. If hypotheses H(f ) hold, K ⊆ W
01,p(Z) is a nonempty, closed and convex set and {x
n}
n≥1⊆ K and ε
n> 0, ε
n↓ 0 satisfy
|R
λ(x
n)| ≤ M, 0 ≤ R
λ0(x
n; y − x
n) + ε
nky − x
nk for all y ∈ K, then {x
n}
n≥1⊆ W
01,p(Z) has a strongly convergent subsequence.
P r o o f. Since by hypothesis {R
λ(x
n)}
n≥1is bounded and because R
λ(·) is coercive (see Proposition 2) we infer that {x
n}
n≥1⊆ W
01,p(Z) is bounded.
So by passing to a subsequence if necessary, we may assume that x
n w→ x in W
01,p(Z) and x
n→ x in L
p(Z) as n → ∞. By h·, ·i we denote the duality brackets of the pair (W
01,p(Z), W
−1,q(Z)). We know that R
0λ(x
n; x − x
n) = sup{hx
∗, x − x
ni : x
∗∈ ∂R
λ(x
n)}, n ≥ 1. Because ∂R
λ(x
n) ⊆ W
−1,q(Z) is weakly compact, we can find x
∗n∈ ∂R
λ(x
n) such that R
0λ(x
n; x − x
n) = hx
∗n, x − x
ni, n ≥ 1. We have
x
∗n= A(x
n) − λk(x
n) − v
n, n ≥ 1, k(x) = |x|
p−2x, where A : W
01,p(Z) → W
−1,q(Z) is defined by
hA(x), yi =
\
Z
kDx(z)k
p−2(Dx(z), Dy(z))
RNdz
for all y ∈ W
01,p(Z) and v
n∈ ∂J(x
n) with J(x
n) =
T
Z
F (z, x
n(z)) dz. It is easy to see that A is monotone, demicontinuous, hence maximal monotone (see Hu–Papageorgiou [8]) and f
1(z, x
n(z)) ≤ v
n(z) ≤ f
2(z, x
n(z)) a.e. on Z for all n ≥ 1 (see [10]). Then {v
n}
n≥1⊆ L
q(Z) is bounded and
0 ≤ hx
∗n, x − x
ni + ε
nkx − x
nk
= hA(x
n), x − x
ni −
\
Z
v
n(z)(x − x
n)(z) dz
− λ(k(x
n), x − x
n)
pq+ ε
nkx − x
nk, hence limhA(x
n), x
n− xi ≤ 0 since
\
Z
v
n(z)(x − x
n)(z) dz
n→∞−−→ 0, (k(x
n), x − x
n)
pq n→∞−−→ 0, ε
nkx − x
nk −−→ 0.
n→∞But A, being maximal monotone, is generalized pseudomonotone (see for example [8], Remark III.6.3, p. 365). So we have
hA(x
n), x
ni −−→ hA(x), xi,
n→∞hence kDx
nk
pn→∞−−→ kDxk
p. Recall that Dx
n w→ Dx in L
p(Z, R
N) and that L
p(Z, R
N) is uniformly con- vex. So it has the Kadec–Klee property (see [8], Definition I.1.72, p. 28).
Thus Dx
n→ Dx in L
p(Z, R
N), from which we conclude that x
n→ x in W
01,p(Z) as n → ∞.
Because ∂R
λ(x) ⊆ W
−1,q(Z) is weakly compact, we can find x
∗∈
∂R
λ(x) such that kx
∗k = m(x) = inf{ky
∗k : y
∗∈ ∂R
λ(x)}. Then the same proof as for Proposition 5 gives us the following result:
Proposition 6. If hypotheses H(f ) hold, then R
λ(·) satisfies the non- smooth (PS )-condition.
4. Multiplicity theorem. In this section we state and prove the main result of this paper, which shows that problem (3) has at least three non- trivial solutions for all λ ∈ (λ
1− δ, λ
1) with δ > 0 small enough (i.e. for equations near resonance).
Theorem 7. If hypotheses H(f ) hold, then there exists δ > 0 such that for all λ ∈ (λ
1− δ, λ
1) problem (3) has at least three nontrivial solutions.
P r o o f. From Proposition 3 of [10], we know that we can find β
1, β
2> 0 such that for all 0 < λ ≤ λ
1and all x ∈ W
01,p(Z), we have R
λ(x) ≥ β
1kxk
p1,p− β
2kxk
θ1,pwith θ > p. So we can find ̺ > 0 small enough so that inf[R
λ(x) : kxk
1,p= ̺] > 0 for all λ ∈ (0, λ
1]. Let β ∈ R be as in Proposition 3 and let b t > 0 be as in Proposition 4. Choose t
1> max{b t, ̺} and let δ = δ
t1be as in Proposition 4. Then t
1u
1∈ B
̺(0) and R
λ(t
1u
1) < β < 0.
So we can apply Theorem 1 and obtain y
06= 0 such that R
λ(y
0) > 0 > β
and 0 ∈ ∂R
λ(y
0).
Let U
±= {x ∈ W
01,p(Z) : x = ±tu
1+ v, t > 0, v ∈ V }. We show that R
λ(·) attains its infimum on both subsets U
+and U
−. To this end let m
+= inf[R
λ(x) : x ∈ U
+] = inf[R
λ(x) : x ∈ U
+] (since R
λ(·) is locally Lipschitz). Let
R
λ(x) =
R
λ(x) if x ∈ U
+, +∞ otherwise.
Evidently, R
λ(·) is a lower semicontinuous function on the Banach space W
01,p(Z) which is bounded below (see Proposition 2). By Ekeland’s vari- ational principle (strong form, see for example De Figueiredo [7] or Hu–
Papageorgiou [8]), we can find {x
n}
n≥1⊆ U
+such that R
λ(x
n) ↓ m
+as n → ∞ and
R
λ(x
n) ≤ R
λ(y) + ε
nky − x
nk for all y ∈ W
01,p(Z), hence
R
λ(x
n) ≤ R
λ(y) + ε
nky − x
nk for all y ∈ U
+.
Because U
+is convex, for every t ∈ (0, 1) and every w ∈ U
+, we have y
n= (1 − t)x
n+ tw ∈ U
+for all n ≥ 1. So we have
−ε
nkw − x
nk ≤ R
λ(x
n+ t(w − x
n)) − R
λ(x
n) t
and therefore
0 ≤ R
0λ(x
n; w − x
n) + ε
nkw − x
nk for all w ∈ U
+.
Proposition 5 says that by passing to a subsequence if necessary, we may assume that x
n→ y
1as n → ∞ in W
01,p(Z). If y
1∈ ∂U
+= V , then because of Proposition 3, we have lim R
λ(x
n) = R
λ(y
1) = m
+> β. On the other hand, Proposition 4 implies that there exists δ > 0 such that m
+< β for λ ∈ (λ
1− δ, λ
1). So we have a contradiction from which it follows that y
1∈ U
+, y
16= 0. Thus y
1is a local minimum of R
λ(·) and so 0 ∈ ∂R
λ(y
1) (see Section 2). Working similarly on the other open set U
−, we obtain y
2∈ U
−, y
16= y
26= y
0such that 0 ∈ ∂R
λ(y
2) for all λ ∈ (λ
1− δ, λ
1).
Finally, let y = y
k, k ∈ {1, 2, 0}. Since 0 ∈ ∂R
λ(y) (where λ
1− δ
1<
λ < λ
1, δ
1= δ
t1), we have
A(y) − λ|y|
p−2y = v in W
01,p(Z)
with f
1(z, y(z)) ≤ v(z) ≤ f
2(z, y(z)) a.e. on Z. Thus for θ ∈ C
0∞(Z) we have
hA(y), θi − λ
\
Z
|y(z)|
p−2y(z)θ(z) dz =
\
Z
v(z)θ(z) dz, hence
\
Z
kDy(z)k
p−2(Dy(z), Dθ(z))
RNdz =
\
Z
(v(z) + λ|y(z)|
p−2y(z))θ(z) dz
and consequently
h− div(kDyk
p−2Dy), θi =
\
Z
(v(z) + λ|y(z)|
p−2y(z))θ(z) dz
(by Green’s formula; see for example Kenmochi [9]). Since C
0∞(Z) is dense in W
01,p(Z), we conclude that y ∈ W
01,p(Z) solves problem (3).
This proves that y
1, y
2, y
0are three dinstinct nontrivial solutions of (3).
Remark . It will be very interesting to know if a similar multiplicity theorem is also valid for the resonant problem.
Acknowledgments. The authors wish to express their gratitude to a very knowledgeable referee for his corrections, remarks, constructive criti- cism and for furnishing additional relevant references.
REFERENCES
[1] A. A m b r o s e t t i, J. G a r c i a A z o r e r o and I. P e r a l, Multiplicity results for some nonlinear elliptic equations, J. Funct. Anal. 137 (1996), 219–242.
[2] A. A m b r o s e t t i and P. R a b i n o w i t z, Dual variational methods in critical point theory and applications, ibid. 14 (1973), 349–381.
[3] K. C. C h a n g, Variational methods for nondifferentiable functionals and their ap- plications to partial differential equations, J. Math. Anal. Appl. 80 (1981), 102–129.
[4] R. C h i a p p i n e l l i and D. D e F i g u e i r e d o, Bifurcation from infinity and multiple solutions for an elliptic system, Differential Integral Equations 6 (1993), 757–771.
[5] R. C h i a p p i n e l l i, J. M a w h i n and R. N u g a r i, Bifurcation from infinity and mul- tiple solutions for some Dirichlet problems with unbounded nonlinearities, Nonlinear Anal. 18 (1992), 1099–1112.
[6] F. H. C l a r k e, Optimization and Nonsmooth Analysis, Wiley, New York, 1983.
[7] D. D e F i g u e i r e d o, The Ekeland Variational Principle with Applications and Detours, Springer, Berlin, 1989.
[8] S. H u and N. S. P a p a g e o r g i o u, Handbook of Multivalued Analysis, Volume I : Theory, Kluwer, Dordrecht, 1997.
[9] N. K e n m o c h i, Pseudomonotone operators and nonlinear elliptic boundary value problems, J. Math. Soc. Japan 27 (1975), 121–149.
[10] N. C. K o u r o g e n i s and N. S. P a p a g e o r g i o u, Discontinuous quasilinear elliptic problems at resonance, Colloq. Math. 78 (1998), 213–223.
[11] P. L i n d q v i s t, On the equation div(kDxkp−2Dx) + λ|x|p−2x = 0, Proc. Amer.
Math. Soc. 109 (1991), 157–164.
[12] J. M a w h i n and K. S c h m i t t, Landesman–Lazer type problems at an eigenvalue of odd multiplicity, Results Math. 14 (1988), 138–146.
[13] —, —, Nonlinear eigenvalue problems with a parameter near resonance, Ann. Polon.
Math. 51 (1990), 241–248.
[14] R. R a b i n o w i t z, Minimax Methods in Critical Point Theory with Applications to Differential Equations, CBMS Regional Conf. Ser. in Math. 65, Amer. Math. Soc., Providence, RI, 1986.
[15] M. R a m o s and L. S a n c h e z, A variational approach to multiplicity in elliptic prob- lems near resonance, Proc. Roy. Soc. Edinburgh Sect. A 127 (1997), 385–394.
[16] P. T o l k s d o r f, Regularity for a more general class of quasilinear elliptic equations, J. Differential Equations 51 (1984), 126–150.
Department of Mathematics National Technical University Zografou Campus
Athens 15780, Greece E-mail: npapg@math.ntua.gr
Received 16 September 1998;
revised 28 January 1999