Monatsh Math (2016) 181:485–500

DOI 10.1007/s00605-016-0944-7

A minimization problem with variable growth

on Nehari manifold

Xia Zhang1

Received: 26 May 2013 / Accepted: 10 June 2016 / Published online: 15 June 2016
© Springer-Verlag Wien 2016

Abstract In this paper, based on the theory of variable exponent space, we study a
class of minimizing problem on Nehari manifold via concentration compactness prin-
ciple. Under suitable assumptions, by showing a relative compactness of minimizing
sequences, we prove the existence of minimizers.

Keywords Variable exponent Sobolev space · Concentration compactness principle ·

Nehari manifold

Mathematics Subject Classification 46E35 · 47J30

1 Introduction

In this paper, we study the following minimizing problem

I0  inf{I (u) : u ∈ W 1, p(x) (R N )\{0}, I  (u), u = 0}, (1.1)

where the energy functional I : W 1, p(x) (R N ) → R N is defined by

  
|∇u| p(x) + |u| p(x) |u|α(x)
I (u) = − d x.
RN p(x) α(x)

Communicated by J. Escher.

B Xia Zhang
[email protected]

1 Department of Mathematics, Harbin Institute of Technology, Harbin 150001, China

123
486 X. Zhang

We will use variable exponent Sobolev spaces W 1, p(x) (R N ), the definitions of which
will be given in Sect. 2. Assume that p, α ∈ C(R N ) satisfy

1 < p− ≤ p(x) ≤ p+ < N (1.2)

and

p(x) α(x) p ∗ (x).

We could verify that I ∈ C 1 (W 1, p(x) (R N ), R). Then, a necessary condition for u ∈

W 1, p(x) (R N ) to be a critical point of I is that I  (u), u = 0, which defines the Nehari
manifold

N  {u ∈ W 1, p(x) (R N )\{0} : I  (u), u = 0}.

The minimizing problem (1.1) is often referred to as the minimizing problem with
artificial constraint. A critical point u 0 = 0 of I is a ground state or a least energy
critical point if

I (u 0 ) = inf I (u).
u∈N

In [7], the author gave a minimax characterization of inf u∈N I (u) and proved that
there exists a least energy critical point of I . As it is simpler to prove that the minimax
value is a critical value of I . In this paper, we will study the existence of minimizer for
minimizing problem (1.1) by showing a relative compactness of minimizing sequences
of (1.1).
A main difficulty is due to the loss of the compactness for the embedding
W 1, p(x) (R N ) → L α(x) (R N ). In order to overcome this difficulty, we assume that
p(x), α(x) are periodic (see the following definition) and use the concentration com-
pactness principle established by Lions in [13,14].

Definition 1.1 Let e1 , e2 , . . . , e N be the standard basis of R N . A function f : R N →

R is called T -periodic, if for any x ∈ R N , f (x + Ti ei ) = f (x), i = 1, 2, . . . , N ,
where T = (T1 , T2 , . . . , TN ) ∈ R N .

Then, we obtain the following main result.

Theorem 1.1 If p(x), α(x) are T -periodic and α− > p+ , then there exists u 0 ∈ N
such that I (u 0 ) = inf u∈N I (u).

2 Preliminaries

In the studies of a class of nonlinear problems with variable exponential growth, see
for example [1–4,9–11,15,16,18], variable exponent spaces play an important role.
Since they were thoroughly studied by Kovác̆ik and Rákosník [12], variable exponent
spaces have been used to model various phenomena. In [17], Růz̆ic̆ka presented the

123
A minimization problem with variable growth. . . 487

mathematical theory for the application of variable exponent Sobolev spaces in electro-
rheological fluids. As another application, Chen, Levine and Rao [5] suggested a model
for image restoration based on a variable exponent Laplacian.
For the convenience of the reader, we recall some definitions and basic properties
of variable exponent spaces L p(x) () and W 1, p(x) (), where  ⊂ R N is a domain.
For a deeper treatment on these spaces, we refer to [6].
Let P() be the set of all Lebesgue measurable functions p :  → [1, ∞], we
denote

ρ p(x) (u) = |u| p(x) d x + sup |u(x)|,
\∞ x∈ ∞

where ∞ = {x ∈  : p(x) = ∞}.

The variable exponent Lebesgue space L p(x) () is the class of all functions u such
that ρ p(x) (tu) < ∞, for some t > 0. L p(x) () is a Banach space equipped with the
norm
||u|| p(x) = inf{λ > 0 : ρ p(x) (λu) ≤ 1}.
For any p ∈ P(), we define the conjugate function p  (x) as
⎧
⎪
⎨∞, x ∈ 1 = {x ∈  : p(x) = 1},
p  (x) = 1, x ∈ ∞ ,
⎪
⎩ p(x) , x ∈ \( ∪  ).
p(x)−1 1 ∞

We denote

p+ = sup p(x), p− = inf p(x)

x∈  x∈ 

and denote by p1 p2 the fact that inf x∈ ( p2 (x) − p1 (x)) > 0.

Theorem 2.1 Let p ∈ P(). For any u ∈ L p(x) () and v ∈ L p (x) (),

|uv| d x ≤ 2||u|| p(x) ||v|| p (x) .


Theorem 2.2 Let p ∈ P() with p+ < ∞. For any u ∈ L p(x) (), we have
p− p+
(1) if ||u|| p(x) ≥ 1, then ||u|| p(x) ≤  |u| p(x) d x ≤ ||u|| p(x) ;
p+ p−
(2) if ||u|| p(x) < 1, then ||u|| p(x) ≤  |u| p(x) d x ≤ ||u|| p(x) .

The variable exponent Sobolev space W 1, p(x) () is the class of all functions u ∈
L p(x) () such that |∇u| ∈ L p(x) (). W 1, p(x) () is a Banach space equipped with
the norm
||u||1, p(x) = ||u|| p(x) + ||∇u|| p(x) .
1, p(x)
By W0 () we denote the subspace of W 1, p(x) () which is the closure of
∞
C0 () with respect to the norm || · ||1, p(x) . Under the condition 1 ≤ p− ≤ p(x) ≤

123
488 X. Zhang

1, p(x)
p+ < ∞, W 1, p(x) () and W0 () are reflexive. And we denote the dual space of
1, p(x) −1,  (x)
W0 () by W p ().
For u ∈ W 1, p(x) (), if we define

 
|u| p(x) + |∇u| p(x)
|||u||| = inf t > 0 : dx ≤ 1 ,
 t p(x)

then ||| · ||| and || · ||1, p(x) are equivalent norms on W 1, p(x) (). In fact, we have

1
||u||1, p(x) ≤ |||u||| ≤ 2||u||1, p(x) .
2

Theorem 2.3 For any u ∈ W 1, p(x) (), we have

(1) if |||u||| ≥ 1, then |||u||| p− ≤  (|∇u| p(x) + |u| p(x) ) d x ≤ |||u||| p+ ;

(2) if |||u||| < 1, then |||u||| p+ ≤  (|∇u| p(x) + |u| p(x) ) d x ≤ |||u||| p− .

¯
Theorem 2.4 Let  be a bounded domain with the cone property. If p ∈ C()
satisfying (1.2) and q is a measurable function defined on  with

N p(x)
p(x) ≤ q(x) p ∗ (x)  a.e. x ∈ ,
N − p(x)

then there is a compact embedding W 1, p(x) () → L q(x) ().

Theorem 2.5 Let  be a domain with the cone property. If p is uniformly continuous
on  and satisfies (1.2), q is a measurable function defined on  with

p(x) ≤ q(x) p ∗ (x) a.e. x ∈ ,

then there is a continuous embedding W 1, p(x) () → L q(x) ().

To prove Theorem 1.1, we recall the following lemma of Lions type (see Lemma
3.1 in [8]).

Lemma 2.1 Assume that p : R N → R is a uniformly continuous function satisfying

condition (1.2). If {u n } is bounded in W 1, p(x) (R N ) and

sup |u n |q(x) d x → 0, as n → ∞,
y∈R N B(y,r )

for some r > 0 and some q(x) satisfying p(x) ≤ q(x) p ∗ (x), then u n → 0 in
L α(x) (R N ) for any α(x) with p(x) α(x) p ∗ (x), where B(y, r ) is an open ball
with center y and radius r .

123
A minimization problem with variable growth. . . 489

3 Main results

In this section, using concentration compactness principle, we show that the mini-
mizing sequence of (1.1) is relative compactness. Then, we obtain the existence of
solution for (1.1), which is also a least energy critical point of I .
In the following, we assume that p(x), α(x) are T -periodic and α− > p+ . Firstly,
several technical results will be established. For any a > 0, define

a
I−a = inf{I (u) + : u ∈ W 1, p(x) (R N ), I  (u), u = −a}.
p+

Theorem 3.1 For any a ≥ 0, I−a satisfies the following properties:

(1) I0 > 0;
(2) for any b > a ≥ 0, I−b > I−a ;
(3) I−a , as a function of a, is continuous in a ∈ (0, ∞).

Proof (1) In fact, for any u ∈ N , we have

 
(|∇u| p(x)
+ |u| p(x)
) dx = |u|α(x) d x,
RN RN

which implies
 
1 1
I (u) ≥ (|∇u| p(x)
+ |u| p(x)
) dx − |u|α(x) d x
p+ R N α− R N
 
1 1
= − (|∇u| p(x) + |u| p(x) ) d x ≥ 0. (3.1)
p+ α− RN

If I0 = 0, there exists {u n } ⊂ N such that I (u n ) → 0, as n → ∞. Then


(|∇u n | p(x) + |u n | p(x) ) d x → 0.
RN

It follows from Theorem 2.3 that u n → 0 in W 1, p(x) (R N ), as n → ∞. By

Theorem 2.5, we have u n → 0 in L α(x) (R N ). Using Theorems 2.2 and 2.3, we
derive
 
0 = I  (u n ), u n  = (|∇u n | p(x) + |u n | p(x) ) d x − |u n |α(x) d x
RN RN
α−
≥ |||u n ||| p+
− ||u n ||α(x)
≥ |||u n ||| p+ − C α− |||u n |||α−
1
≥ |||u n ||| p+ ,
2

as n is sufficiently large. That is a contradiction. So, we get I0 > 0.

123
490 X. Zhang

(2) For any b > a ≥ 0, I−b > I−a . Note that for any t > 0,
 
I  (tu), tu = t p(x) (|∇u| p(x) + |u| p(x) ) d x − t α(x) |u|α(x) d x.
RN RN

Take u = 0. For any 0 < t < 1, we obtain

 
I  (tu), tu ≥ t p+ (|∇u| p(x) + |u| p(x) ) d x − t α− |u|α(x) d x.
RN RN

As p+ < α− , there exists 0 < t (u) < 1 such that I  (t (u)u), t (u)u > 0.
Then, for any I  (u), u = −b, there exists tu ∈ (t (u), 1) such that
I  (tu u), tu u = −a. We have


p(x) |∇u|
b a p(x) + |u| p(x)
I (u) + − I (tu u) − = (1 − tu ) dx
p+ p+ RN p(x)

|u|α(x) b a
− (1 − tuα(x) ) dx + − .
R N α(x) p + p +

Denote
 
|∇u| p(x) + |u| p(x) |u|α(x)
g(t) = (1 − t p(x) ) dx − (1 − t α(x) ) d x.
RN p(x) RN α(x)

For any 0 < t < 1,

 

g (t) = − t p(x)−1
(|∇u| p(x)
+ |u| t α(x)−1 |u|α(x) d x
p(x)
) dx +
 RN  RN

< −t p+ −1 (|∇u| p(x) + |u| p(x) ) d x + t α− −1 |u|α(x) d x

R R
N N

p+ −1 α(x) α− −1 α(x)
= −t ( |u| d x − b) + t |u| dx
RN RN
p+ −1
< bt .

Thus g  (t) − bt p+ −1 < 0, which implies g(t) − b p+

p+ t is decreasing on (0, 1).
Then

b p+ b b
g(tu ) − tu > g(1) − =− ,
p+ p+ p+

b p+
i.e., I (u) + b
p+ − I (tu u) − a
p+ > p + tu − p+ .
a

123
A minimization problem with variable growth. . . 491

In addition, we have

I  (u), u − I  (tu u), tu u

 
(1 − tuα(x) )|u|α(x) d x
p(x)
= (1 − tu )(|∇u| p(x) + |u| p(x) ) d x −
RN RN
 
α−
|u|α(x) d x
p+
< (1 − tu ) (|∇u| p(x)
+ |u| p(x)
) d x + (tu − 1)
RN RN
  
α(x) α−
|u|α(x) d x
p+
= (1 − tu ) |u| d x − b + (tu − 1)
RN RN
p
< −b(1 − tu + ), (3.2)

p p
which implies −b + a < −b(1 − tu + ). Thus btu + > a.
By the definition of I−a , we obtain

b a b p+ a b p+ a
I (u) + > I (tu u) + + tu − ≥ I−a + tu − . (3.3)
p+ p+ p+ p+ p+ p+

Take {u n } ⊂ W 1, p(x) (R N ) such that I  (u n ), u n  = −b and I (u n ) + b

p+ → I−b ,
as n → ∞. We have

b
I (u n ) +
p+
 
1 1 b
≥ (|∇u n | p(x) + |u n | p(x) ) d x − |u n |α(x) d x +
p+ R N α− R N p+

1 1
= (|∇u n | p(x) + |u n | p(x) ) d x −
p+ R N α−
 
b
× (|∇u n | p(x)
+ |u n | p(x)
) dx + b +
p+
 R
N

1 1
≥ − (|∇u n | p(x) + |u n | p(x) ) d x.
p+ α− RN

Then, {u n } is bounded in W 1, p(x) (R N ).

In the following, we will verify that there exists t0 ∈ (0, 1] such that tu n → t0 ,
as n → ∞. Indeed, if the conclusion is not satisfied, we may assume that there
exists a subsequence, still denoted by {tu n }, such that tu n → 0. Thus, tu n u n → 0
in W 1, p(x) (R N ). Similarly to the discussion in (1), we get a contradiction. Then
p
bt0 + > a.
Using (3.3) we obtain

b b p+ a
I (u n ) + ≥ I−a + tu − .
p+ p+ n p+

123
492 X. Zhang

Let n → ∞, we have

b p+ a
I−b ≥ I−a + t0 − > I−a .
p+ p+

(3) For any a > 0, it suffices to verify that

I−a = lim I−(a+ 1 ) = lim I−(a− 1 ) .

n→∞ n n→∞ n

Using (2) we get I−(a+ 1 ) > I−a > I−(a− 1 ) , where n ∈ N.

n n
Firstly, we will verify that I−a = limn→∞ I−(a+ 1 ) .
n
Take {u n } ⊂ W 1, p(x) (R N ) such that I  (u n ), u n  = −a and I (u n ) + a
p+ → I−a ,
as n → ∞. Then {u n } is bounded in W 1, p(x) (R N )
and there exists tn ∈ (1, +∞)
such that I  (tn u n ), tn u n  = −(a + n1 ).
Denote vn = tn u n and sn = t1n . We obtain sn ∈ (0, 1), I  (sn vn ), sn vn  = −a,
I (sn vn ) + pa+ → I−a and I  (vn ), vn  = −(a + n1 ).
p
Similarly to (3.2), we get (a + n1 )sn + > a, thus sn → 1. Then tn → 1 and {vn }
a+ 1
is bounded in W 1, p(x) (R N ). We obtain I (vn ) + pa+ → I−a . As I (vn ) + p+n ≥
I−(a+ 1 ) , we get I−a ≥ lim supn→∞ I−(a+ 1 ) , thus I−a = limn→∞ I−(a+ 1 ) .
n n n
Next, we will verify that I−a = limn→∞ I−(a− 1 ) .
n
For any ε > 0 and n ∈ N, there exists u n such that I  (u n ), u n  = −(a − n1 ) and
a− 1
I−(a− 1 ) + ε > I (u n ) + p+n . Then, {u n } is bounded in W 1, p(x) (R N ) and there
n
exists tn > 1 such that I  (tn u n ), tn u n  = −a.
Denote vn = tn u n , sn = t1n . Thus sn ∈ (0, 1), I  (sn vn ), sn vn  = −(a − n1 ) and
I  (vn ), vn  = −a. We obtain asn + > a − n1 , which implies sn → 1. Then tn → 1
p

and {vn } is bounded in W 1, p(x) (R N ). Note that

1 a
I−(a− 1 ) + ε > I (sn vn ) − − I (vn ) + I (vn ) +
n np+ p+
1
> I (sn vn ) − − I (vn ) + I−a .
np+

It follows that I−(a− 1 ) + ε > I−a − ε as n is sufficiently large. Thus we have

n
limn→∞ I−(a− 1 ) = I−a .
n
Combining with the above discussion, we obtain the result. 


Next, based on the following concentration compactness lemma, we will prove the
main result.

Lemma 3.1 Let {ρn } be a sequence in L 1 (R N ) satisfying ρn ≥ 0 and R N ρn (x) d x →
λ > 0, as n → ∞. Then there exists a subsequence, still denoted by {ρn }, satisfying
one of the three following possibilities:

123
A minimization problem with variable growth. . . 493

(1) (Compactness) There exists a sequence {xn } in R N such that {ρn } is tight, that is,
for any ε > 0, there exists R > 0 such that

ρn (x) d x ≥ λ − ε.
B(xn ,R)

(2) (Vanishing) For any R > 0,


lim sup ρn (x) d x = 0.
n→∞
y∈R N B(y,R)

(3) (Dichotomy) There exists β ∈ (0, λ) such that for any ε > 0, there exist R > 0,
{yn } ⊂ R N and Rn → ∞ satisfying: for n sufficiently large,
 
 
 ρn (x) d x − β  < ε

B(yn ,R)

and
 
 
 ρn (x) d x − (λ − β) < ε.

R N \B(yn ,R)

The Proof of Theorem 1.1. Take {u n } ⊂ W 1, p(x) (R N )\{0} be a minimizing sequence

of (1.1), i.e.,

I  (u n ), u n  = 0

and

I (u n ) → I0 , as n → ∞.

It follows from (3.1) that {u n } is bounded in W 1, p(x) (R N ). We assume that

R N (|∇u n | + |u n | p(x) ) d x → λ. Then, λ > 0.
p(x)

Next, we will apply the concentration compactness method with

ρn = |∇u n | p(x) + |u n | p(x) .

In the following, we show that for such {ρn }, the cases “vanishing” and “dichotomy”
do not hold.
(1) By Lemma 3.1, if vanishing takes place, then for any R > 0,

lim sup (|∇u n | p(x) + |u n | p(x) ) d x = 0.
n→∞
y∈R N B(y,R)

123
494 X. Zhang

Thus limn→∞ sup y∈R N B(y,R) |u n | p(x) d x = 0. As p(x) α(x) p ∗ (x), it
follows from Lemma 2.1 that u n → 0 in L α(x) (R N ). Note that
 
(|∇u n | p(x)
+ |u n | p(x)
) dx = |u n |α(x) d x,
RN RN

we get u n → 0 in W 1, p(x) (R N ). That is a contradiction.

(2) In the case of dichotomy, using Lemma 3.1, there exists β ∈ (0, ρ) such that for
any ε > 0, there exist R > 0, yn ∈ R N and Rn → ∞, then
 
 
 ρn (x) d x − β  < ε

B(yn ,R)

and
 
 
 ρn (x) d x − (ρ − β) < ε,

R N \B(yn ,R)

for n sufficiently large. Since Rn → ∞, we assume that Rn > 6R. Note that

ρn (x) d x
B(yn ,Rn )\B(yn ,R)
  
= ρn (x) d x − ρn (x) d x − ρn (x) d x,
RN R N \B(yn ,Rn ) B(yn ,R)

thus 
(|∇u n | p(x) + |u n | p(x) ) d x < 3ε. (3.4)
B(yn ,Rn )\B(yn ,R)

Take ξ, η ∈ C ∞ (R, R) such that ξ(t) = 1, |t| ≤ 1; ξ(t) = 0, |t| ≥ 2 and η(t) = 1,
|t| ≥ 1; η(t) = 0, |t| ≤ 21 . Define
   
|x − yn | |x − yn |
u (1)
n (x) = ξ u n (x), u (2)
n (x) = η u n (x).
R Rn

We have

(|∇u (1)
n |
p(x)
+ |u (1)
n |
p(x)
) dx
R N
    

= (|∇u n | p(x) + |u n | p(x) ) d x + ∇ξ |x − yn | · u n
 R
B(yn ,R) B(yn ,2R)\B(yn ,R)
   p(x) 
|x − yn | 
+ξ ∇u n  + |u (1)
n |
p(x)
d x.
R

Note that B(yn , 2R)\B(yn , R) ⊂ B(yn , Rn )\B(yn , R), it follows from (3.4) that
|x−yn | |x−yn | p(x) + |u (1) | p(x) ) d x → 0.
B(yn ,2R)\B(yn ,R) (|∇ξ( R ) · u n + ξ( R )∇u n | n

123
A minimization problem with variable growth. . . 495

Then,

(|∇u (1)
n |
p(x)
+ |u (1)
n |
p(x)
) d x → β,
RN

as n → ∞. Similarly, we get

(|∇u (2)
n |
p(x)
+ |u (2)
n |
p(x)
) d x → λ − β, (3.5)
RN

  (1)  (2)
|∇u n | p(x) |∇u n | p(x) |∇u n | p(x)
dx − dx − dx → 0 (3.6)
RN p(x) RN p(x) RN p(x)
and
  
|u n | p(x) |u (1)
n |
p(x) |u (2)
n |
p(x)
dx − dx − d x → 0. (3.7)
R N p(x) RN p(x) RN p(x)

(1) (2)
Denote vn = u n − u n − u n . We obtain supp vn ⊂ B(yn , Rn )\B(yn , R). Note
that

(|∇vn | p(x) + |vn | p(x) ) d x
R N

= (|∇u n − ∇u (1) (2) p(x)
n − ∇u n | + |u n − u (1) (2) p(x)
n − un | ) dx
B(yn ,Rn )\B(yn ,R)

≤ 3 p+ (|∇u n | p(x) + |∇u (1)
n |
p(x)
+ |∇u (2)
n |
p(x)
B(yn ,Rn )\B(yn ,R)
+ |u n | p(x) + |u (1)
n |
p(x)
+ |u (2)
n |
p(x)
) d x.

Thus vn → 0 in W 1, p(x) (R N ). Using Theorem 2.5 we get vn → 0 in L α(x) (R N ),

as n → ∞. Then
  (1)  (2)
|u n |α(x) |u n |α(x) |u n |α(x)
dx − dx − dx
R N α(x) RN α(x) RN α(x)
  (1)  (1)
|u n |α(x) |u n |α(x) |u n |α(x)
= dx − dx − dx
R N α(x) B(yn ,R) α(x) B(yn ,2R)\B(yn ,R) α(x)
 (2)  (2)
|u n |α(x) |u n |α(x)
− dx − dx
R N \B(yn ,Rn ) α(x) B(yn ,Rn )\B(yn , 12 Rn ) α(x)
 
|u n |α(x) |u (1)
n |
α(x)
= dx − dx
B(yn ,Rn )\B(yn ,R) α(x) B(yn ,2R)\B(yn ,R) α(x)
 (2)
|u n |α(x)
− dx
B(yn ,Rn )\B(yn , 21 Rn ) α(x)

123
496 X. Zhang

 (2)  (1)
|u n + vn |α(x) |u n + vn |α(x)
= dx + dx
B(yn ,Rn )\B(yn ,2R) α(x) B(yn ,2R)\B(yn ,R) α(x)
 
|u (1)
n |
α(x) |u (2)
n |
α(x)
− dx − d x → 0.
B(yn ,2R)\B(yn ,R) α(x) B(yn ,Rn )\B(yn , 21 Rn ) α(x)

Using (3.6) and (3.7) we get

I (u n ) − I (u (1) (2)
n ) − I (u n ) → 0.

Similarly, we obtain

I  (u n ), u n  − I  (u (1) (1)  (2) (2)

n ), u n  − I (u n ), u n  → 0.

(1) (1) (1) (2) (2)

If I  (u n ), u n  = 0, I (u n ) ≥ I0 . We derive I  (u n ), u n  → 0 and
(2)
I (u n ) → 0, as n → ∞. It follows from (3.5) that
 
1 1 (2) α(x)
I (u (2)
n ) ≥ (|∇u (2)
n | p(x)
+ |u (2) p(x)
n | ) d x − |u n | dx
N p+ R N α−
R 
1 1 1  (2) (2)
= − (|∇u (2)
n |
p(x)
+ |u (2)
n |
p(x)
) dx + I (u n ), u n 
p+ α− α−
 R
N

1 1
→ − (λ − β) = 0,
p+ α−

which is a contradiction.
(1) (1) (1)
In the following, we assume that I  (u n ), u n  = −an < 0. Then I (u n )+ pa+n ≥
(1) (1) (2) (2)
I−an > I0 . If an → 0, I  (u n ), u n  → 0, Thus I  (u n ), u n  → 0 and
(2)
I (u n ) → 0. Similarly to the above discussion, we get a contradiciton. Note that
(1)
{u n } is bounded in W 1, p(x) (R N ), we assume that an → a > 0.
Define


an
Ian = inf I (u) − : u ∈ W 1, p(x) (R N ), I  (u), u = an .
p+

It is easy to verify that Ian ≥ 0. Then

I (u (1) (2)
n ) + I (u n ) ≥ I−an + Ian ≥ I−an ,

which implies I0 ≥ limn→∞ I−an = I−a > I0 . That is a contradiction.

Therefore, by Lemma 3.1, the case “compactness” holds. There exists {yn } ⊂ R N
such that for any ε > 0, there exists R > 0, then

(|∇u n | p(x) + |u n | p(x) ) d x ≥ λ − ε.
B(yn ,R)

123
A minimization problem with variable growth. . . 497

(n) (n) (n)

There are integers l1 , l2 , . . . , l N such that

(n) (n)
xn = yn − (l1 T1 , . . . , l N TN ) ∈ D,

where D = {x = (ξ1 , . . . , ξ N ) ∈ R N : − T2i ≤ ξi < T2i , i = 1, 2, . . . , N }.

(n,1) (n)
Denote vn (x) = u n (x + h 1n ), where h n = (l1 T1 , · · · , l N TN ). As p(x), α(x)
are T -periodic, we get

(|∇vn | p(x) + |vn | p(x) ) d x → λ,
RN

I (vn ) → I0

and

I  (vn ), vn  = 0.

Thus, {vn } is bounded in W 1, p(x) (R N ). We assume that vn → v weakly in

W 1, p(x) (R N ) and vn → v a.e. in R N .
Note that

(|∇vn | p(x) + |vn | p(x) ) d x
B(xn ,R)

= (|∇u n (x + h n )| p(x) + |u n (x + h n )| p(x) ) d x
B(xn ,R))

= (|∇u n | p(x) + |u n | p(x) ) d x ≥ λ − ε.
B(yn ,R)

Denote by L the diameter of D, we have B(xn , R) ⊂ B(0, R + L). Thus


(|∇vn | p(x) + |vn | p(x) ) d x ≥ λ − ε,
B(0,R+L)

B(0,R+L) (|∇vn | + |vn | p(x) ) d x → λ. Then,
which implies p(x)


(|∇vn | p(x) + |vn | p(x) ) d x → 0.
R N \B(0,R+L)

Using Theorem 2.5, we get R N \B(0,R+L) |vn |α(x) d x → 0. Thus
α(x) d x = 0. It follows from Theorem 2.4 that
α(x)
R N \B(0,R+L) |v| B(0,R+L) |vn |

d x → B(0,R+L) |v|α(x) d x. Then
 
α(x)
|vn | dx → |v|α(x) d x.
RN RN

123
498 X. Zhang

Note that |vn − v|α(x) ≤ 2α+ (|vn |α(x) + |v|α(x) ), then by Fatou Lemma, we have

2α+ +1 |v|α(x) d x
RN

= lim inf (2α+ |vn |α(x) + 2α+ |v|α(x) − |vn − v|α(x) ) d x
R N n→∞

≤ lim inf (2α+ |vn |α(x) + 2α+ |u|α(x) − |vn − v|α(x) ) d x
n→∞ R N
 
= 2α+ +1 |v|α(x) d x − lim sup |vn − v|α(x) d x.
RN n→∞ RN

Thus

|vn − v|α(x) d x → 0,
RN

i.e. vn → v in L α(x) (R N ). Then I0 ≥ I (v).

In the following, we will verify that
 
|v|α(x) d x = (|∇v| p(x) + |v| p(x) ) d x.
RN RN

As R N (|∇vn | p(x) + |vn | p(x) ) d x = R N |vn |α(x) d x, we have

 
α(x)
|v| dx ≥ (|∇v| p(x) + |v| p(x) ) d x.
RN RN

If R N |v|α(x) d x = R N (|∇v| p(x) + |v| p(x) ) d x, we get the result.

If R N |v|α(x) d x > R N (|∇v| p(x) + |v| p(x) ) d x, denote

 
(|∇v| p(x)
+ |v| p(x)
) dx − |v|α(x) d x = −a.
RN RN

We obtain
 
lim (|∇vn | p(x) + |vn | p(x) ) d x = (|∇v| p(x) + |v| p(x) ) d x + a. (3.8)
n→∞ R N RN

Denote
 
|∇vn | p(x) + |vn | p(x) |∇v| p(x) + |v| p(x)
b = lim dx − d x.
n→∞ RN p(x) RN p(x)

123
A minimization problem with variable growth. . . 499

Note that

p+ − p(x)
(|∇vn | p(x) + |vn | p(x) ) d x
RN p(x)

p+ − p(x)
− (|∇v| p(x) + |v| p(x) ) d x
RN p(x)

≥ ( p+ − p(x))(|∇v| p(x)−2 ∇v(∇vn − ∇v) + |v| p(x)−2 v(vn − v)) d x.
RN

As vn → v weakly in W 1, p(x) (R N ), we obtain


p+ − p(x)
lim (|∇vn | p(x) + |vn | p(x) ) d x
n→∞ R N p(x)

p+ − p(x)
≥ (|∇v| p(x) + |v| p(x) ) d x
RN p(x)

Using (3.8) we get


p(x) − p+ p+
lim ( (|∇vn | p(x) + |vn | p(x) ) + (|∇vn | p(x) + |vn | p(x) )) d x
n→∞ R N p(x) p(x)

p(x) − p+ p+
= ( (|∇v| p(x) + |v| p(x) ) + (|∇v| p(x) + |v| p(x) )) d x + a,
RN p(x) p(x)

thus

p+ − p(x)
bp+ − a = lim |∇vn | p(x) + |vn | p(x) d x
n→∞ R N p(x)

p+ − p(x)
− |∇v| p(x) + |v| p(x) d x ≥ 0.
RN p(x)

Note that

a
I (vn ) − I (v) −
p+

|∇vn | p(x) + |vn | p(x)
= (
RN p(x)
α(x) 
|vn | |∇v| p(x) + |v| p(x) |v|α(x) a
− ) dx − ( − ) dx − ,
α(x) R N p(x) α(x) p +

which implies I0 − I (v) − pa+ ≥ b − pa+ ≥ 0. Thus I0 ≥ I (v) + a

p+ ≥ I−a > I0 .
That is contradiction. Now, we complete the proof.




123
500 X. Zhang

References
1. Adamowicz, T., Hästö, P.: Harnack’s inequality and the strong p(x)-Laplacian. J. Differ. Equ. 250(3),
1631–1649 (2011)
2. Alves, C.O., Souto, M.A.S.: Existence of solutions for a class of problems in R N involving p(x)-
Laplacian. Prog. Nonlinear Differ. Equ. Appl. 66, 17–32 (2005)
3. Antontsev, S., Chipot, M., Xie, Y.: Uniquenesss results for equation of the p(x)-Laplacian type. Adv.
Math. Sc. Appl. 17(1), 287–304 (2007)
4. Chabrowski, J., Fu, Y.: Existence of solutions for p(x)-Laplacian problems on a bounded domain. J.
Math. Anal. Appl. 306(2005)604–618. Erratum in: J. Math. Anal. Appl. 323(2006)1483
5. Chen, Y., Levine, S., Rao, M.: Variable exponent, linear growth functionals in image restoration. SIAM
J. Appl. Math. 66, 1383–1406 (2006)
6. Diening, L., Harjulehto, P., Hästö, P., Růžička, M.: Legesgue and Sobolev spaces with variable expo-
nents. Lecture Notes in Mathematics, vol. 2017. Springer-Verlag, Heidelberg (2011)
7. Fan, X.L.: p(x)-Laplacian equations in R N with periodic data and nonperiodic perturbations. J. Math.
Anal. Appl. 341, 103–119 (2008)
8. Fan, X.L., Zhao, Y.Z., Zhao, D.: Compact imbedding theorems with symmetry of Strauss-Lions type
for the space W 1, p(x) (). J. Math. Anal. Appl. 255, 333–348 (2001)
9. Fan, X.L., Han, X.: Existence and multiplicity of solutions for p(x)-Laplacian equations in R N .
Nonlinear Anal. 59, 173–188 (2004)
10. Fu, Y.Q., Zhang, X.: Multiple solutions for a class of p(x)-Laplacian equations in R N involving the
critical exponent. Proc. R. Soc. A 466, 1667–1686 (2010)
11. Ge, B., Xue, X.P., Zhou, Q.M.: Existence of at least five solutions for a differential inclusion problem
involving the p(x)-Laplacian. Nonlinear Anal. Real World Appl. 12(4), 2304–2318 (2011)
12. Kováčik, O., Rákosník, J.: On spaces L p(x) and W k, p(x) . Czechoslovak Math. J. 41, 592–618 (1991)
13. Lions, P.L.: The concentration-compactness principle in the calculus of variations. The locally compact
case. Part 1, Ann. Inst. H. Poincaré Anal. Non Linéaire 1, 109–145 (1984)
14. Lions, P.L.: The concentration-compactness principle in the calculus of variations. The locally compact
case. Part 2, Ann. Inst. H. Poincaré Anal. Non Linéaire 1, 223–283 (1984)
15. Mihbreveailescu, M., Rbreveadulescu, v: A multiplicity result for a nonlinear degenerate problem
arising in the theory of electrorheological fluids. Proc. R. Soc. A A(462), 2625–2641 (2006)
16. Mihbreveailescu, M., Rbreveadulescu, V.: On a nonhomogeneous quasilinear eigenvalue problem in
Sobolev spaces with variable exponent. Proc. Amer. Math. Soc. 135, 2929–2937 (2007)
17. Růz̆ic̆ka, M.: Electro-rheological fluids: modeling and mathematical theory. Springer-Verlag, Berlin
(2000)
18. Zhang, C., Zhou, S.L.: Renormalized and entropy solutions for nonlinear parabolic equations with
variable exponents and L 1 data. J. Differ. Equ. 248(6), 1376–1400 (2010)

123