Nonlinear Analysis: Modelling and Control, 2008, Vol. 13, No.

2, 145–158

Existence Theorems for Some Classes of Boundary Value

Problems Involving the P (X)-Laplacian
Ionică Andrei
Department of Mathematics
University of Craiova
200585 Craiova, Romania
[email protected]

Received: 08.12.2007 Revised: 06.03.2008 Published online: 02.06.2008

Abstract. We prove an alternative for a nonlinear eigenvalue problem involving the
p(x)-Laplacian and study a subcritical boundary value problem for the same operator.
The theoretical approach is the Mountain Pass Lemma and one of its variants, which is
very useful in the study of eigenvalue problems.
Keywords: p(x)-Laplacian, nonlinear eigenvalue problem, critical point theory.

1 Introduction
For any fixed real number p ∈ (1, +∞) the p-Laplacian is defined by

∆p u = div (|∇u|p−2 ∇u).

This operator appears in a variety of physical fields. For example, applications of ∆p have
been seen in Fluid Dynamics. The equation governing the motion of a fluid involves the
p-Laplacian. More exactly the shear stress ~τ and the velocity gradient ∇u of the fluid are
related in the manner that

~τ (x) = r(x)|∇u|p−2 ∇u,

where p = 2 (resp., p < 2 or p > 2) if the fluid is Newtonian (resp., pseudoplastic or

dilatant). Other applications of the p-Laplacian also appear in the study of flow through
porous media (p = 23 ), Nonlinear Elasticity (p ≥ 2), or Glaciology (1 < p ≤ 43 ).
This paper is motivated by recent advances in elastic mechanics and electrorheologi-
cal fluids (sometimes referred to as “smart fluids”) where some processes are modeled by
nonhomogeneous quasilinear operators.
We refer mainly to the p(x)-Laplace operator ∆p(x) u := div (|∇u|p(x)−2 ∇u), where
p is a continuous non-constant function. This differential operator is a natural generaliza-
tion of the p-Laplace operator ∆p u := div (|∇u|p−2 ∇u), where p > 1 is a real constant.

145
Ionică Andrei

However, the p(x)-Laplace operator possesses more complicated nonlinearities that the
p-Laplace operator, due to the fact that ∆p(x) is not homogeneous.
Throughout this paper, Ω stands for a bounded domain in RN . In the first section we
are concerned with the following nonlinear eigenvalue problem with Dirichlet boundary
condition and constraints on eigenvalues:

−∆p(x) u = λf (x, u), in Ω,

u = 0, on ∂Ω, (1)

0 < λ ≤ a,


where a > 0 is a given constant. The function f is supposed to satisfy

(H1 ) f is a Carathéodory function, i.e., measurable in x ∈ Ω and continuous in u ∈ R,

with f (x, 0) 6= 0 on a subset of Ω of positive measure;

(H2 ) |f (x, u)| ≤ C1 + C2 |u|q(x)−1 , for a.e. x ∈ Ω and all u ∈ R, with constants C1 ≥ 0,
C2 ≥ 0 and 1 < p(x) ≤ q(x) < p⋆ (x), where

 N p(x) , if p(x) < N,


⋆
p (x) = N − p(x)
+∞, if p(x) ≥ N ;


(H3 ) there are constants b1 ≥ 0, b2 ≥ 0, 1 ≤ γ < p(x) < ν such that, for a.e. x ∈ Ω
and every u ∈ R,
Zu
f (x, u)u − ν f (s, τ ) dτ ≥ −b1 − b2 |u|γ .
0

By the Sobolev embedding Theorem, there exists a constant C > 0 such that, for
1,p(x)
every u ∈ W0 (Ω),
q(x) q(x)
kukLq ≤ Ckuk 1,p(x) . (2)
W0

For p ∈ L∞ (Ω), let

p− (Ω) = ess inf p(x), p+ (Ω) = ess sup p(x).

Ω Ω

For a later use we denote

+
−1)/q+ +
−1)/q+
a1 = c1 |Ω|(q and a2 = C c1 |Ω|(q + c2 (q + )−1 .


(3)

Our approach relies on the following version of the celebrated Mountain Pass Theo-
rem of Ambrosetti-Rabinowitz (see [1, 2]):

146
 Existence Theorems for Some Classes of Boundary Value Problems Involving the P (X)-Laplacian

Lemma 1 ( [3]). Let X be a Banach space and let F : X × R → R be a C 1 functional

verifying the hypotheses
(i) there exist constants ρ > 0 and α > 0 provided F (v, ρ) ≥ α, for every v ∈ X;
(ii) there is some r > ρ with F (0, 0) = F (0, r) = 0.
Then the number


c := inf max F g(τ ) ,
g∈Γ 0≤τ ≤1

where

Γ = {g ∈ C([0, 1], X × R); g(0) = (0, 0), g(1) = (0, r)},

is a critical value of F .
Let us now state our main result concerning the eigenvalue problem (1). We shall
keep the notations given in (2), (3) and, for simplicity, we use in the sequel k · k in place
of k · kW 1,p(x) .
0

Theorem 1. Assume that the function f : Ω × R → R satisfies conditions (H1 )–(H3 ).

Let β ∈ C 1 (R, R) be a function such that, for some constants 0 < ρ < r, σ > 0, the
following properties hold:
(β 1 ) β(0) = β(r) = 0;
q+
||u|| σ+1
(β 2 ) ρσ+1 ≥ q(x)a2 ||u|| q(x) and q(x) β(ρ) = a1 ;

(β 3 ) lim|t|→∞ β(t) = +∞;

(β 4 ) β ′ (t) < 0 if and only if t < 0 or ρ < t < r.
Then, for each a > 0, the following alternative holds:
either
(i) a > 0 is an eigenvalue in problem (1) with a corresponding eigenfunction u ∈
1,p(x)
W0 (Ω) located by

Z u(x)
1
Z
α≤− f (x, t) dt dx + kukp(x) ≤ a1 + α
ap(x)
Ω 0

or
(ii) one can find a positive number s with

ρ<s<r (4)

147
Ionică Andrei

1,p(x)
which determines an eigensolution (u, λ) ∈ W0 (Ω) × (0, a] of the problem (1) by the
relations

kuk = s−σ/q(x) (−β ′ (s))1/q(x) , (5)

(q(x)−p(x))/q(x)
λ−1 = a−1 + s(q(x)+σp(x))/q(x) − β ′ (s) , (6)

Z u(x)
sq(x)+1 σ+1 1
Z
α≤ kukq(x) + β(s) − f (x, t) dt dx + kukp(x)
q(x) q(x) ap(x) (7)
Ω 0
≤ a1 + α.

In the second section of this paper we consider another problem related to the
p(x)-Laplacian operator:

p(x)−2
−∆p(x) u = λ|u|
 u + |u|q(x)−2 u, in Ω,
u = 0, on ∂Ω , (8)

u 6≡ 0, in Ω.


Our result on this problem is

1,p(x)
Theorem 2. If λ < λ1 (−∆p(x) ) := inf{ Ω |∇u|p(x) ; u ∈ W0
R
(Ω), u 6= 0,
kukLp(x) = 1} and 1 < p(x) < q(x) < p⋆ (x), then the problem (8) has a weak solution.

The key argument in the proof is the Mountain-Pass Theorem in the following vari-
ant:

Ambrosetti-Rabinowitz Theorem. Let X be a real Banach space and F : X → R be

a C 1 -functional. Suppose that F satisfies the Palais-Smale condition and the following
geometric assumptions:

there exist positive constants R and c0 such that

(9)
F (u) ≥ c0 , for all u ∈ X with kuk = R;

F (0) < c0 and there exists v ∈ X such that

(10)
kvk > R and F (v) < c0 .

Then the functional F possesses at least a critical point.

We refer to [4] and [5] for related bifurcation results in the semi-linear case and to
the works [6–9], and [10] for recent qualitative results both in the semi-linear and in the
quasi-linear case.

148
 Existence Theorems for Some Classes of Boundary Value Problems Involving the P (X)-Laplacian

2 Proof of Theorem 1
In order to set problem (1) in terms of Lemma 1 we introduce the functional F ∈
1,p(x)
C 1 (W0 (Ω) × R) by

Z v(x)
|t|σ+1 σ+1 1
Z
F (v, t) = kvkq(x) + β(t) − f (x, t) dt dx + kvkp(x) . (11)
q(x) q(x) ap(x)
Ω 0

From (β1 ) and (11) we derive that condition (ii) of Lemma 1 is valid.
1,p(x)
From (H2 ), (2) and (3) we see that, for every v ∈ W0 (Ω),

Z v(x)
Z
+
f (x, t) dt dx ≤ c1 kvkL1 + c2 (q + )−1 kvkqLq+
Ω 0 +
−1)/q+
+
≤ c1 |Ω|(q kvkLq+ + c2 (q + )−1 kvkqLq+
+ + (12)
−1)/q+ +
−1)/q+
≤ c1 |Ω|(q + c1 |Ω|(q + c2 (q + )−1 kvkqLq+

+
−1)/q+ + + +
≤ c1 |Ω|(q + C c1 |Ω|(q −1)/q + c2 (q + )−1 kvkq


+
= a1 + a2 kvkq

Relations (11), (12) and (β2 ) yield

+
ρσ+1 ||v||q
 
σ+1
F (v, ρ) ≥ − a2 kvkq(x) + β(ρ) − a1 ≥ α,
q(x) ||v||q(x) q(x)
1,p(x)
for every v ∈ W0 (Ω). This shows that the requirement (i) of Lemma 1 is fulfilled.
We check now that F verifies the Palais-Smale condition. To this end, let (vn , tn ) be
1,p(x)
a sequence in W0 (Ω) × R such that F (vn , tn ) is bounded and
′
F ′ (vn , tn ) = Fv (vn , tn ), Ft (vn , tn ) → 0, in W −1,p (x) (Ω) × R,

p(x)
where p′ (x) = p(x)−1 . Therefore

|F (vn , tn )| ≤ M (13)
σ+1 q(x)−p(x)
−Fv (vn , tn ) = |tn | kvn k ∆p(x) vn
′
+ f (·, vn ) + a−1 ∆p(x) vn → 0 in W −1,p (x) (Ω) (14)
σ q(x) ′
Ft (vn , tn ) = |tn | (sgn tn ) kvn k + β (tn ) → 0, in R. (15)

From (11), (12) and (13) we infer that

+
|tn |σ+1 ||v||q
 
σ+1
M≥ − a 2 + β(tn ) − a1 .
q+ ||v||q(x) q(x)

149
Ionică Andrei

But, by condition (β3 ), this shows that (tn ) is bounded in R.

Without loss of generality we may assume that (vn ) is bounded away from 0. We
treat separately two cases.
Firstly, assume that along a subsequence one has tn → 0. Then, by (β4 ), it follows
that β ′ (tn ) → β ′ (0) = 0. So, by (15),

|tn |σ kvn kq(x) → 0, as n → ∞. (16)

From (11), (13) and (16) we see that

vZ
n (x)
1
Z
f (x, τ ) dτ dx − kvn kp(x) is bounded in R. (17)
ap(x)
Ω 0

Since tn → 0 and (vn ) is bounded away from zero it is clear from (16) that

|tn |σ+1 kvn kq(x)−p(x) k k∆p(x) vn kW −1,p′ (x)

0

= |tn | |tn |σ kvn kq(x)−p(x) kvn kp(x)−1

= |tn | |tn |σ kvn kq(x) kvn k−1 → 0, as n → ∞.

Thus, (14) implies

f (·, vn ) + a−1 ∆p(x) vn → 0, as n → ∞. (18)

From (17) and (18) we find that, for some constant M > 0 and with ν > 2 in (H3 ),

Z vZn (x)
−1 1 p(x)
M +ν kvn k ≥ kvn k − f (x, τ ) dτ dx
ap(x)
Ω 0
Z 
1
Z
+ f (x, vn )vn dx + a−1 (∆p(x) vn )vn dx
ν
Ω Ω
  vZ
n (x) !
1 1 1 1
Z
p(x)
= − kvn k + f (x, vn )vn − ν f (x, τ ) dτ dx,
a p(x) ν ν
Ω 0

if n is sufficiently large. Then hypothesis (H3 ) and inequality (2) ensure us that some new
constants d1 ≥ 0 and d2 ≥ 0 exist such that
 
1 1 1 1
M + ν −1 kvn k ≥ kvn kp(x) − b1 |Ω| + b2 kvn kγLγ


−
a p(x) ν ν
 
1 1 1
≥ − kvn kp(x) − d1 − d2 kvn kγ .
a p(x) ν

150
 Existence Theorems for Some Classes of Boundary Value Problems Involving the P (X)-Laplacian

Recalling that 1 ≤ γ < p(x) < ν, the last estimate shows that (vn ) is bounded in
1,p(x)
W0 (Ω). On the other hand, the growth condition in (H2 ) ensures that the restriction
1,p(x)
of Nemytskii’s operator to W0 (Ω), namely,
1,p(x) ′
(Ω) 7−→ f ·, v(·) ∈ W −1,p (x) (Ω),


v ∈ W0
is a compact mapping, in the sense that it maps any bounded set onto a relatively compact
one (see, for details, de Figueiredo [11] or Rabinowitz [2]). Thus, passing eventually to a
subsequence,

−1,p′ (x)
f ·, vn (·) converges in W0 (Ω). (19)
1,p(x)
By (18) and (19) we conclude that (vn ) possesses a convergent subsequence in W0 (Ω).
Assume now that (tn ) is bounded away from 0. Then, by (15), we see that (vn ) is
1,p(x)
bounded in W0 (Ω). Hence (19) holds. From (14) it follows that
′
1 + a|tn |σ+1 kvn kq(x)−p(x) ∆p(x) vn is convergent in W −1,p (x) (Ω),

′
which shows that (∆p(x) vn ) converges in W −1,p (x) (Ω). Finally, we obtain that, up to
1,p(x)
a subsequence, (vn ) converges in W0 (Ω). This concludes the verification of the
Palais-Smale condition for the functional F .
The hypotheses of Lemma 1 are now verified. Thus, there exists a point (u, s) ∈
1,p(x)
W0 (Ω) × R satisfying
1
−∆p(x) u = f (·, u); (20)
a−1 + |s|σ+1 kukq(x)−p(x)

|s|σ (sgn s) kukq(x) + β ′ (s) = 0; (21)

u(x)
|s|σ+1 σ+1 1
Z Z
kukq(x) + β(s) − f (x, t) dt dx + kukp(x) ≥ α. (22)
q(x) q(x) ap(x)
Ω 0
From (21) we observe that
sβ ′ (s) ≤ 0. (23)
There are two cases:
Case 1. s = 0. Then the assertion (i) in the alternative of Theorem 1 is deduced from (20)
and (22). The last inequality of (i) is obtained from the definition of c and Γ in Lemma 1,
making use of the path g ∈ Γ given by g(t) = (0, tr), for 0 ≤ t ≤ 1.
Case 2. s 6= 0. We argue by contradiction. If s < 0 then, by (β4 ), it follows that
β ′ (s) < 0, which contradicts (23). So, the only possibility is s > 0. Using (β4 ) again it
turns out
ρ ≤ t ≤ r. (24)

151
Ionică Andrei

If t = ρ or t = r, relation (21) and assumption (β4 ) imply u = 0. This leads to a

contradiction between (20) and our hypothesis (H1 ). We proved that (24) reduces to (4).
1,p(x)
Since s > 0, (21) gives rise to (5). From (20) it is clear that (u, λ) ∈ W0 (Ω) × R is
an eigensolution of (1), where
1
λ= . (25)
a−1 + sσ+1 kukq(x)−p(x)
Substituting kuk as determined by (5) in (25) we arrive at (6). The first inequality of (7)
is just (22). The second inequality of (7) follows from Lemma 1, by choosing the path
g(t) = (0, tr), 0 ≤ t ≤ 1.
Corollary 1. Assume that the function f : Ω×R → R satisfies hypotheses (H1 )–(H3 ) and
let a > 0 be a number which is not an eigenvalue of the problem (1). Then there exists a
1,p(x)
sequence (un , λn ) ∈ W0 (Ω) × (0, a) of eigensolutions of (1) with the properties
1,p(x)
un → 0 in W0 (Ω), λn → 0 and λ−1
n kun k
p(x)
→ 0, as n → ∞.
Proof. For every ε > 0 one can find βε ∈ C 1 (R, R) satisfying (β1 )–(β4 ) with ρ = ρε <
r = rε , which depends on ε, and σ > 0, α > 0 independent of ε such that
|βε′ (t)| ≤ εq(x) t−1 , for every t ≥ (q(x)a2 )1/(σ+1) . (26)
Applying Theorem 1, one obtains the number s = sε ∈ (ρε , rε ) that describes an
eigensolution (uε , λε ) of (1) by equalities (5) and (6) with u = uε and λ = λε . Clearly,
we can assume
sε → +∞, as ε → 0. (27)
Hence, by (5), (26) and (27), we infer that

1/q(x)
kuε k = s−σ/q(x)
ε − β ′ (sε ) ≤ εs−(σ+1)/q(x)
ε → 0, as ε → 0. (28)
We know that the following equality holds
1
− ∆p(x) uε = f (x, uε ).
λε
1,p(x)
Letting ε → 0 we notice that, in view of (H1 ) and uε → 0 in W0 (Ω), it follows that
λε → 0 as ε → 0. In addition, we get from (6) that

q(x)−p(x)
(λ−1
ε −a
−1 q(x)
) = sq(x)+σp(x)
ε − β ′ (sε ) ≤ εq(x)(q(x)−p(x)) s(σ+1)p(x)
ε . (29)
By (28) and (29) we observe that
kuε kp(x) λ−1 −1
≤ εq(x) ,


ε −a

which implies, taking into account (28), that

λ−1
ε kuε k
p(x)
→ 0, as ε → 0.
This completes our proof.

152
 Existence Theorems for Some Classes of Boundary Value Problems Involving the P (X)-Laplacian

Corollary 2. Under the hypotheses of Corollary 1, for every function β ∈ C 1 (R, R) sa-
tisfying conditions (β1 )–(β4 ) with fixed constants ρ, r, σ, α, there is a one-to-one mapping
from [1, +∞) into the set of eigensolutions (u, λ) of the problem (1). In particular, there
exist uncountable many solutions (u, λ) of (1).
Proof. Notice that if β ∈ C 1 (R, R) satisfies the requirements (β1 )–(β4 ) for given num-
bers ρ, r, σ, α, then this is true for each function δβ, with an arbitrary number δ ≥ 1.
We may suppose that there is some a > 0 which is not an eigenvalue of (1). Applying
Theorem 1 with δβ, for δ ≥ 1, in place of δ, one finds an eigensolution (uδ , λδ ) ∈
1,p(x)
W0 (Ω) × (0, a) and a number sδ ∈ (ρ, r) such that
−σ/q(x)
1/q(x) 1/q(x)
kuδ k = sδ − β ′ (sδ ) δ (30)

and, by (25),

λ−1
δ = a
−1
+ sσ+1
δ kuδ kq(x)−p(x) . (31)

Let δ1 , δ2 ≥ 1 with δ1 6= δ2 . Then (31) shows that sδ1 = sδ2 . Thus (30) yields δ1 = δ2 .
This contradiction completes the proof.
In some situations the qualitative informations provided by Theorem 1 and Corol-
laries 1 and 2 can be improved by direct methods in studying the eigenvalue problem
(1).
Example. Assume that the Carathéodory function f : Ω × R → R satisfies (H1 ) and the
growth condition
Zt
f (x, τ ) dτ ≤ C1 + C2 |t|p(x) , for a.e. x ∈ Ω and all t ∈ R, (32)
0

with constants C1 ≥ 0 and C2 ≥ 0. Using the constant C > 0 entering in (2), with
q(x) = p(x), we check that every number λ > 0 which satisfies
1
λ < λ⋆ := (33)
pCC2
is an eigenvalue of the boundary value problem
(
−∆p(x) u = λf (x, u), in Ω,
u = 0, on ∂Ω.

In order to justify this, corresponding to each λ in (33) we introduce the functional

1,p(x)
Iλ : W0 (Ω) → R by

Z v(x)
1
Z
Iλ (v) = − f (x, t) dt dx + kvkp(x) .
λp(x)
Ω 0

153
Ionică Andrei

The assumption (32) allows us to write

1 p(x)
Iλ (v) ≥ kvkp(x) − C1 |Ω| − C2 kvkLp(x)
λp(x)
  (34)
1
≥ − CC2 kvkp(x) − C1 |Ω|,
λp(x)
1,p(x)
for every v ∈ W0 (Ω). From (33) and (34) it follows that the functional Iλ is bounded
1,p(x)
from below, coercive and (sequentially) weakly lower semicontinuous on W0 (Ω).
1,p(x)
Therefore the infimum of Iλ is achieved at some u ∈ W0 (Ω) which solves the above
boundary value problem corresponding to any λ in (33).

3 Proof of Theorem 2
Our hypothesis
|∇u|p(x) dx
R
λ < λ1 (−∆p(x) ) := inf RΩ
1,p(x)
W0 (Ω)\{0} Ω
|u|p(x) dx
1,p(x)
implies the existence of some C0 > 0 such that, for every v ∈ W0 (Ω),
Z Z
|∇v|p(x) − λ|v|p(x) dx ≥ C0 |∇v|p(x) dx.


(35)
Ω Ω

Set
(
uq(x)−1 , if u ≥ 0,
g(u) =
0, if u < 0
Ru
and G(u) = 0 g(t)dt. Denote
1
Z Z
|∇u|p(x) − λ|u|p(x) dx −


F (u) = G(u) dx.
p(x)
Ω Ω

Observe that
|G(u)| ≤ C |u|q(x)
1,p(x)
and, by our hypothesis 1 < p(x) < q(x) < p⋆ (x), W0 (Ω) ⊂ Lq(x) (Ω), which
1,p(x)
implies that F is well defined on W0 (Ω).
A straightforward computation shows that F is a C 1 function and, for every v ∈
1,p(x)
W0 (Ω),
Z Z
′ p(x)−2 p(x)−2


F (u)(v) = |∇u| ∇u · ∇v − λ |u| uv dx − g(u)v dx.
Ω Ω

154
 Existence Theorems for Some Classes of Boundary Value Problems Involving the P (X)-Laplacian

We prove in what follows that F satisfied the hypotheses of the Mountain-Pass

Theorem.
Verification of (9): We may write, for every u ∈ R,
|g(u)| ≤ |u|q(x)−1 .
Thus, for every u ∈ R,
1
|G(u)| ≤ |u|q(x) . (36)
q(x)
Now, by (36) and the Sobolev embedding Theorem,
1
F (u) ≥ C0 kukp(x) − kukq(x) , (37)
q(x)
1,p(x)
for every u ∈ W0 (Ω).
For ε > 0 and R > 0 small enough, we deduce by (36) that, for every u ∈
1,p(x)
W0 (Ω) with kuk = R,
F (u) ≥ c0 > 0.
1,p(x)
Verification of (10): Choose u0 ∈ W0 (Ω), u0 > 0 in Ω. Then, by 1 < p(x) <
q(x) < p⋆ (x), it follows that if t > 0 is large enough,
Z p(x) Z
t q(x)
|∇u0 |p(x) − λ|u0 |p(x) dx − tq(x) u0 dx < 0.


F (tu0 ) =
p(x)
Ω Ω

1,p(x)
Verification of the Palais-Smale condition: Let (un ) be a sequence in W0 (Ω)
such that
sup |F (un )| < +∞, (38)
n
kF ′ (un )kW −1,p′ (x) → 0, as n → ∞. (39)
1,p(x)
We prove firstly that (un ) is bounded in W0 (Ω). Remark that (39) implies that, for
1,p(x)
every v ∈ W0 (Ω),
Z
|∇un |p(x)−2 ∇un · ∇v − λ|un |p(x)−2 un v dx

Ω
Z (40)
= g(un )v dx + o(1)kvk, as n → ∞.
Ω

Choosing v = un in (40) we find

Z Z
p(x) p(x)


|∇un | − λ |un | dx = g(un )un dx + o(1)kun k. (41)
Ω Ω

155
Ionică Andrei

Remark that (38) means that there exists M > 0 such that, for any n ≥ 1,
1
Z Z
|∇un |p(x) − λ |un |p(x) dx − G(un ) dx ≤ M.


(42)
p(x)
Ω Ω

But a simple computation yields

Z Z
g(un )un dx = q(x) G(un ) dx. (43)
Ω Ω

Combining (41), (42) and (43) we find

Z
α G(un ) dx = O(1) + o(1)kun k, (44)
Ω

where α = q(x) − p(x) > 0. Thus, by (41) and (44),

kun kp(x) ≤ O(1) + o(1) kun k,

which means that kun k is bounded.

It remains to prove that (un ) is relatively compact. We consider the case p(x) < N .
First of all we remark that (40) may be written
Z Z
|∇un |p(x)−2 ∇un · ∇v dx = h(un )v dx + o(1)kvk, (45)
Ω Ω

1,p(x)
for every v ∈ W0 (Ω), where

h(u) = g(u) + λ |u|p(x)−2 u.

Obviously, h is continuous and there exists C > 0 such that

|h(u)| ≤ C 1 + |u|(N p(x)−N +p(x))/(N −p(x)) .

(46)

Moreover

h(u) = o |u|N p(x)/(N −p(x)) ,

as |u| → ∞. (47)
′ 1,p(x)
Observing that (−∆p(x) )−1 : W −1,p (x) (Ω) → W0 (Ω) is a continuous operator, it
′
follows by (45) that it suffices to show that h(un ) is relatively compact in W −1,p (x) (Ω).
By Sobolev’s Theorem, this will be achieved by proving that a subsequence of h(un ) is
convergent in (L(N p(x))/(N −p(x)) (Ω) )⋆ = L(N p(x))/(N p(x)−N +p(x))(Ω).
1,p(x)
Since (un ) is bounded in W0 (Ω) ⊂ L(N p(x))/(N −p(x))(Ω) we can suppose that,
up to a subsequence,

un → u ∈ LN p(x)/(N −p(x)) (Ω), a.e. in Ω.

156
 Existence Theorems for Some Classes of Boundary Value Problems Involving the P (X)-Laplacian

Moreover, by Egorov’s Theorem, for each δ > 0, there exists a subset A of Ω with |A| < δ
and such that

un → u, uniformly in Ω \ A.

So, it is sufficient to show that

Z
|h(un ) − h(u)|N p(x)/(N p(x)−N +p(x)) dx ≤ η,
A

for any fixed η > 0. But, by (46),

Z Z
|h(u)|N p(x)/(N p(x)−N +p(x)) dx ≤ C 1 + |u|N p(x)/(N −p(x)) dx,

A A

which can be made arbitrarily small if we choose a sufficiently small δ > 0.

We have, by (47),
Z Z
N p(x)/(N p(x)−N +p(x))
|h(un )−h(u)| dx ≤ ε |un −u|N p(x)/(N −p(x)) dx+Cε |A|,
A A

which can be also made arbitrarily small, by Sobolev’s Theorem and by the boundedness
1,p(x)
of (un ) in W0 (Ω).
Hence, F satisfies Palais-Smale Condition and, by Ambrosetti-Rabinowitz Theorem,
the problem (8) has a weak solution.
Remark. We are not able to decide at this stage what happens if λ > λ1 (−∆p(x) ). The
main difficulty consists in the impossibility of defining in a suitable manner the orthogonal
1,p(x)
of a set, so to split the Banach space W0 (Ω), p 6= 2, as a direct sum of its first
eigenspace and the corresponding orthogonal. A more general version of Theorem 2 can
be obtained by replacing the term |u|q(x)−2 u in (8) by a function f(x,u) whose behaviour
at u = 0 and for |u| → +∞ is similar to the one of |u|q(x)−2 u. The final part of
the proof of Theorem 2, that is, the deduction of the relative compactness of un from its
boundedness, can also be derived using the continuity of Nemytskii’s operator u 7−→ h(u)
⋆
on Lp (x) (Ω).

References
1. A. Ambrosetti, P. H. Rabinowitz, Dual variational methods in critical point theory and
applications, J. Funct. Anal., 14, pp. 349–381, 1973.
2. P. H. Rabinowitz, Minimax Methods in Critical Point Theory with Applications to Differential
Equations, CBMS Reg. Conf. Ser. Math., Vol. 65, Amer. Math. Soc., Providence, R.I., 1986.
3. D. Motreanu, A saddle point approach to nonlinear eigenvalue problems, Math. Slovaca, 47,
pp. 463–477, 1997.

157
Ionică Andrei

4. P. Mironescu, V. Rădulescu, A bifurcation problem associated to a convex, asymptotically

linear function, C.R. Acad. Sci. Paris I, 316, pp. 667–672, 1993.

5. P. Mironescu, V. Rădulescu, The study of a bifurcation problem associated to an asymptotically

linear function, Nonlinear Anal., T.M.A., 26, pp. 857–875, 1996.
6. M. Ghergu, V. Radulescu, Singular Elliptic Problems. Bifurcation and Asymptotic Analysis,
Oxford Lecture Series in Mathematics and Its Applications, Vol. 37, Oxford University Press,
2008.
7. A. Kristály, H. Lisei, Cs. Varga, Multiple solutions for p-Laplacian type Equations, Nonlinear
Anal. TMA (in press, doi:10.1016/j.na.2006.12.031).
8. A. Kristály, Cs. Varga, Multiple solutions for elliptic problems with singular and sublinear
potentials, Proc. Amer. Math. Soc., 135, pp. 2121–2126, 2007.

9. D. Motreanu, V. Rădulescu, Existence theorems for some classes of boundary value problems
involving the p-Laplacian, PanAmerican Mathematical Journal, 7, pp. 53–66, 1997.

10. V. Rădulescu, Qualitative Analysis of Nonlinear Elliptic Partial Differential Equations,

Contemporary Mathematics and Its Applications, Hindawi Publ. Corp. (in press).
11. D. G. de Figueiredo, Lectures on the Ekeland Variational Principle with Applications and
Detours, TATA Institute of Fundamental Research, Bombay, 1989.

158