ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT

REACTION

NIKOLAOS S. PAPAGEORGIOU, DUSAN D. REPOVŠ, AND VICENŢIU D. RĂDULESCU

Abstract. We consider a Dirichlet problem driven by the anisotropic (p, q)-Laplacian

and a reaction with gradient dependence (convection). Using the frozen variable method
and eventually a fixed point theorem, we show that the problem has a positive smooth
solution.

1. Introduction
Let Ω ⊆ RN be a bounded domain with a C 2 -boundary ∂Ω. In this paper we study
the following anisotropic Dirichlet problem with a reaction which depends on the gradient
(convection)
−∆p(z) u(z) − ∆q(z) u(z) = r(z)|Du(z)|τ (z)−1 + f (z, u(z)) in Ω,
(
(1)
u = 0, u > 0.
∂Ω

For r ∈ C(Ω) we define

r− = min r and r+ = max r.
Ω Ω
We consider the set E1 = {r ∈ C(Ω) : 1 < r− }. For r ∈ E1 , by ∆r(z) we denote the
r-Laplace differential operator defined by
 
1,p(z)
∆r(z) = div |Du|r(z)−2 Du forall u ∈ W0 (Ω).

In problem (1) the differential operator is the sum of two such operators (anisotropic
(p, q)-equation or anisotropic double phase problem). The reaction (right hand side of (1))
depends also on the gradient of u (convection). This makes the problem nonvariational,
which means that eventually our proof should be topological, based on the fixed point
theory.
Recently there have been some existence results for elliptic equations driven by the (p, q)-
Laplacian (or even more general nonhomogeneous operators) and a gradient dependent
reaction. We mention works of Bai [2], Bai-Gasinski-Papageorgiou [3], Candito-Gasinski-
Papageorgiou [5], Faria-Miyagaki-Motreanu [11], Gasinski-Krech-Papageorgiou [13], Gasinski-
Winkert [16], Liu-Papageorgiou [18], Marano-Winkert [19], Papageorgiou-Rădulescu-Repovš
[22], Papageorgiou-Vetro-Vetro [24], Zeng-Liu-Migorski [30]. All the aforementioned papers
deal with isotropic equations. To the best of our knowledge, there are no works in the
literature dealing with anisotropic (p, q)-equations with convection. It appears that our
existence theorem here is the first such result.
We mention that equations driven by sum of two differential operators of different nature,
appear in mathematical models of physical processes. We refer to the works of Bahrouni-
Rădulescu-Repovš [1], Cencelj-Rădulescu-Repovš [6], Rădulescu [26], Zhikov [32] and the
references therein.
Key words and phrases. Anisotropic (p, q)-Laplacian, convection, regularity theory, maximum principle,
fixed point, minimal positive solution
2010 Mathematics Subject Classification. 35J75, 35J60, 35J20.
1
2 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

Our approach is based on the so-called ”frozen variable method”. According to this
method, in the reaction we fix (freeze) the gradient term and this way we have a variational
problem which can be treated using tools from the critical point theory. We need to find a
canonical way to choose a solution from the solution set of the ”frozen problem”. To this
end, we show that the ”frozen problem” has a smallest positive solution (minimal positive
solution). This way we can define the minimal solution map. Using an iterative process, we
show that this map is compact and then using the Leray-Schauder Alternative Principle, we
produce a fixed point for the minimal solution map. This fixed point is the desired positive
solution of (1).

2. Mathematical Background-Hypotheses
The analysis of problem (1) uses Lebesgue and Sobolev spaces with variable exponents.
A comprehensive treatment of such spaces can be found in the book of Diening-Harjulehto-
Hästo-Ruzicka [7].
Let M (Ω) be the space of all measurable functions u : Ω → R. As usual, we identify
two such functions which differ only on a Lebesgue null set. Let r ∈ E1 . Then the variable
exponent Lebesgue space Lr(z) (Ω) is defined as follows
 Z 
r(z) r(z)
L (Ω) = u ∈ M (Ω) : |u| dz < ∞ .
Ω
This space is equipped with the so called ”Luxemburg norm” defined by
( Z  r(z) )
|u|
kukr(z) = inf λ > 0 : dz 6 1 .
Ω λ
 
The space Lr(z) (Ω), k · kr(z) is separable and uniformly convex (thus reflexive by the
Milman-Pettis theorem, see Papageorgiou-Winkert [25], Theorem 3.4.28, p. 225). Let
r(z)
r0 ∈ E1 , be defined by r0 (z) = z ∈ Ω (the conjugate variable exponent to r(·)). We
r(z) − 1
0
have Lr(z) (Ω)∗ = Lr (z) (Ω) and also the following version of the Hölder inequality is true
Z  
1 1
|uv|dz 6 + 0 kukr(z) kvkr0 (z)
Ω r− r−
0
for all u ∈ Lr(z) (Ω), v ∈ Lr (z) (Ω).
Suppose that r1 , r2 ∈ E1 and we have r1 (z) 6 r2 (z) for all z ∈ Ω. Then
Lr2 (z) (Ω) ,→ Lr1 (z) (Ω) continuously.
Having the variable exponent Lebesgue spaces, we can define in the usual way the corre-
sponding variable exponent Sobolev spaces. So, given r ∈ E1 the variable exponent Sobolev
space W 1,r(z) (Ω) is defined by
n o
W 1,r(z) (Ω) = u ∈ Lr(z) (Ω) : |Du| ∈ Lr(z) (Ω) .

Here the gradient Du is understood in the weak sense. The space W 1,r(z) (Ω) is furnished
with the norm
kuk1,r(z) = kukr(z) + k|Du|kr(z) for all u ∈ W 1,r(z) (Ω).
For simplicity in the sequel we write kDukr(z) = k|Du|kr(z) .
Also, if r ∈ E1 is Lipschitz continuous (that is, r ∈ E1 ∩ C 0,1 (Ω)), then we define
1,r(z) k·k1,r(z)
W0 (Ω) = Cc∞ (Ω) .
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 3

1,r(z)
Both spaces W 1,r(z) (Ω) and W0 (Ω) are separable and uniformly convex (thus reflex-
1,r(z)
ive). For the space W0 (Ω) the Poicaré inequality is valid, namely there exists Ĉ > 0
such that
1,r(z)
kukr(z) 6 ĈkDukr(z) for all u ∈ W0 (Ω).
1,r(z)
This means that on W0 (Ω) we can consider the equivalent norm
1,r(z)
kuk1,r(z) = kDukr(z) for all u ∈ W0 (Ω).
Given r ∈ E1 , we introduce the critical variable exponent r∗ (·) corresponding to r(·),
defined by 
 N r(z)
∗ , if r(z) < N
r (z) = N − r(z) for all z ∈ Ω.
+∞, if N 6 r(z)


Consider r ∈ E1 ∩ C 0,1 (Ω), q ∈ E1 with q+ < N and assume that 1 < q(z) 6 r∗ (z)
(resp. 1 < q(z) < r∗ (z)) for all z ∈ Ω. Then we have the following embeddings (anisotropic
Sobolev embedding theorem)
1,r(z)
W0 (Ω) ,→ Lq(z) (Ω) continuously
 
1,r(z)
resp. W0 (Ω) ,→ Lq(z) (Ω) compactly .
The following modular function is very useful in the study of the variable exponent spaces
Z
ρr (u) = |u|r(z) dz for all u ∈ Lr(z) (Ω) (r ∈ E1 ).
Ω
Again we write ρr (Du) = ρr (|Du|).
This function is closely related to the norm.
Proposition 1. If r ∈ E1 and {u, un }n∈N ⊆ Lr(z) (Ω), then
 
u
(a) kukr(z) = µ ⇔ ρr = 1;
µ
(b) kukr(z) < 1 (resp. = 1, > 1) ⇔ ρr (u) < 1 (resp. = 1, > 1);
r+ r−
(c) kukr(z) < 1 ⇒ kukr(z) 6 ρr (u) 6 kukr(z) ;
r− r+
(d) kukr(z) > 1 ⇒ kukr(z) 6 ρr (u) 6 kukr(z) ;
(e) kun kr(z) → 0 ⇔ ρr (un ) → 0;
(f) kun kr(z) → +∞ ⇔ ρr (un ) → +∞.
Given r ∈ E1 ∩ C 0,1 (Ω), we have that
1,r(z) 0
W0 (Ω)∗ = W −1,r (z) (Ω).
1,r(z) 0 1,r(z)
Then we introduce the nonlinear map Ar(z) : W0 (Ω) → W −1,r (z) (Ω) = W0 (Ω)∗
defined by Z
hAr(z) (u), hi = |Du|r(z)−2 (Du, Dh)RN dz
Ω
1,r(z)
for all u, h ∈
W0 (Ω).
This operator has the following properties (see Gasinski-Papageorgiou [15], Proposition
2.5 and Rădulescu-Repovš [27], p. 40).
1,r(z) 0
Proposition 2. The operator Ar(z) : W0 (Ω) → W −1,r (z) (Ω) is bounded (that is, maps
bounded sets to bounded sets), continuous, strictly monotone (hence maximal monotone too)
and of type (S)+ , that is,
w 1,r(z)
”un → u in W0 (Ω) and lim suphAr(z) (un ), un − ui 6 0
n→∞
4 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

⇓
1,r(z)
un → u in W0 (Ω)”.

We will also use the Banach space C01 (Ω) = {u ∈ C 1 (Ω) : u = 0}. This is an ordered
∂Ω
Banach space with positive cone C+ = u ∈ C01 (Ω) : u(z) > 0 for all z ∈ Ω . This cone has

a nonempty interior given by
 
∂u
intC+ = u ∈ C+ : u(z) > 0 for all z ∈ Ω, <0
∂n ∂Ω
with n(·) being the outward unit normal on ∂Ω.
Consider the following anisotropic eigenvalue problem
(2) −∆p(z) u(z) = λ̂|u(z)|p(z)−2 u(z) in Ω, u = 0,
∂Ω
 
1,r(z)
with p ∈ E1 . We say that (λ̂, û) ∈ R × W0 (Ω) \ {0} is an ”eigenpair” for problem
(2), if Z
1,r(z)
hAp(z) (û), hi = λ̂ |û(z)|p(z)−2 û(z)dz for all h ∈ W0 (Ω).
Ω
Then λ̂ is an ”eigenvalue” and û 6= 0 is a corresponding ”eigenfunction”. We let
L = {λ̂ ∈ R : λ̂ is an eigenvalue of (2)}.
For the anisotropic eigenvalue problem, in contrast to the isotropic one, we can have
inf L = 0 (see Fan-Zhang-Zhao [9], Theorem 3.1). If we can find η ∈ RN (N > 1) such that
for all z ∈ Ω, the function t 7→ ϑ(t) = p(z + tη) is monotone on Tz = {t ∈ R : z + tη ∈ Ω}
and p ∈ C 1 (Ω), then problem (2) has a principal eigenvalue λ̂1 > 0 with corresponding
Lp(z) -normalized positive eigenfunction û1 ∈ intC+ (see Fan-Zhang-Zhao [9], Theorem 3.3)
(see also Byun-Ko [4] and Fan [8]). We have
ρp (Dû1 ) ρp (Du) 1,r(z)
(3) 0 < λ̂1 = 6 for all µ ∈ W0 (Ω), u 6= 0.
ρp (û1 ) ρp (u)
As we already mentioned in the Introduction, our approach is eventually topological and
uses the ”Leray-Schauder Alternative Principle”.
Given a Banach space X, a map ξ : X → X is said to be ”compact” if it is continuous
and maps bounded sets into relatively compact sets. If X is reflexive and ξ : X → X is
w
completely continuous (that is, xn → x in X ⇒ ξ(xn ) → ξ(x)), then ξ(·) is compact (see
Gasinski-Papageorgiou [14], Proposition 3.1.7, p. 268). The Leray-Schauder Alternative
Principle” asserts the following:
Theorem 3. If X is a Banach space, ξ : X → X is compact and
D(ξ) = {u ∈ X : u = tξ(u), 0 < t < 1},
then one of the following statements is true
(a) D(ξ) is unbounded;
(b) ξ(·) has a fixed point.
Through this paper we will use the following notation. By k · k we denote the norm of
1,p(z)
the Sobolev space W0 (Ω). If p ∈ E1 ∩ C 0,1 (Ω), then by the Poincaré inequality, we have
1,p(z)
kuk = kDukp(z) for all u ∈ W0 (Ω).
1,p(z)
Given u ∈ W0 (Ω),u > 0, by [0, u] we denote the order interval
n o
1,p(z)
[0, u] = h ∈ W0 (Ω) : 0 6 h(z) 6 u(z) for a.a. z ∈ Ω .
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 5

Let g : Ω × R → R be a measurable function. By Ng (·) we denote the Nemytski (super-

position) operator defined by
Ng (u)(·) = g(·, u(·)) for all u : Ω → R measurable.
Evidently z 7→ Ng (u)(z) is measurable. Recall that if g : Ω × R → R is a Carathéodory
function (that is, for all x ∈ R z 7→ g(z, x) is measurable and for a.a. z ∈ Ω x 7→ g(z, x) is
continuous), then (z, x) 7→ g(z, x) is measurable (see Gasinski-Papageorgiou [14], p. 405).
1,p(z)
For every x ∈ R, we set x± = max{±x, 0} and then for u ∈ W0 (Ω) we define
± ±
u (·) = u(·) . We know that
1,p(z)
u± ∈ W0 (Ω), u = u+ − u− , |u| = u+ + u− .
1,p(z)
A set S ⊆ W0 (Ω) is said to be ”downward directed”, if for every pair u1 , u2 ∈ S, we
can find u ∈ S such that u 6 u1 , u 6 u2 .
Our hypotheses on the data of problem (1) are the following:
H0 : p ∈ C 1 (Ω), there exists a vector η̂ ∈ RN such that for all z ∈ Ω, the function t 7→ p(z +
tη̂) is monotone on Iz = {t ∈ R : z +tη̂ ∈ Ω}, q ∈ E1 ∩C 0,1 (Ω), τ ∈ E1 , τ+ < p− 6 p+ < p∗− ,
0 6 p+ − p− 6 1 and r ∈ L∞ (Ω), r(z) > 0 for a.a. z ∈ Ω, r 6= 0.
Remark 1. As we already mentioned earlier in this section, the hypotheses on the exponent
p(·), imply that the eigenvalue problem (2) has a principal eigenvalue λ̂1 > 0 and the
associated Lp(z) -normalized positive eigenfunction û1 ∈ intC+ (see Fan-Zhang-Zhao [9],
Fan [8] and Byun-Ko [4]).
The hypotheses on the perturbation f (z, x) are:
H1 : f : Ω × R → R is a Carathéodory function, f (z, 0) = 0 for a.a. z ∈ Ω and
(i) for every ρ > 0, there exists aρ ∈ L∞ (Ω) such that
0 6 f (z, x) 6 aρ (z) for a.a. z ∈ Ω, all 0 6 x 6 ρ;
(ii) there exists a function ϑ ∈ L∞ (Ω) such that
ϑ(z) 6 λ̂1 for a.a. z ∈ Ω, ϑ 6≡ λ̂1 ,
f (z, x) p+ F (z, x)
lim sup p(z)−1
6 ϑ(z) and lim sup 6 ϑ(z)
x→+∞ x x→+∞ xp(z)
Z x
uniformly for a.a. z ∈ Ω, with F (z, x) = f (z, s)ds;
0
(iii) there exist η0 > 0 and M > 0 such that
−η0 6 λ̂1 xp(z) − p+ F (z, x) for a.a. z ∈ Ω, all x > M ;
(iv) there exist µ ∈ E1 with µ+ < q− and δ > 0 such that
C0 xµ(z)−1 6 f (z, x) for a.a. z ∈ Ω, all 0 6 x 6 δ, some C0 > 0;
(v) for a.a. z ∈ Ω, all x ∈ R and all t ∈ (0, 1) we have
 
1 1
f z, x 6 p(z)−1 f (z, x).
t t
Remark 2. Hypothesis H1 (ii) implies that at +∞ we have nonuniform nonresonance with
respect to the principal eigenvalue λ̂1 > 0. Hypothesis H1 (iv) implies the presence of a
concave term near zero. Hypothesis H1 (v) is satisfied if for example for a.a. z ∈ Ω the
f (z, x) ◦
quotient function x 7→ p(z)−1 is nondecreasing on R+ = (0, +∞).
x
The following lemma will help us to exploit the nonuniform nonresonance condition in
hypothesis H1 (ii).
6 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

Lemma 4. If ϑ ∈ L∞ (Ω), ϑ(z) 6 λ̂1 for a.a. z ∈ Ω and ϑ 6≡ λ̂1 , then there exists C1 > 0
such that Z
C1 ρp (Du) 6 ρp (Du) − ϑ(z)|u|p(z) dz
Ω
1,p(z)
for all u ∈ W0 (Ω).
Proof. We argue by contradiction. So, suppose that the assertion of the lemma is not true.
1,p(z)
Then we can find {un }n∈N ⊆ W0 (Ω) such that
Z
1
(4) ρp (Dun ) − ϑ(z)|un |p(z) dz < ρp (Dun ) for all n ∈ N.
Ω n
Clearly without any loss of generality, we may assume that un > 0 for all n ∈ N. From
(4) we have
  Z
1 1
(5) 1− < ϑ(z)unp(z) dz for all n ∈ N.
n ρp (Dun ) Ω
Suppose that kun k → ∞ (hence ρp (Dun ) → +∞, see Proposition 1).
un 1,p(z)
Set vn = 1/p(z)
∈ W0 (Ω), n ∈ N. We have
ρp (Dun )
Dun 1 ln ρp (Dun )
(6) Dvn = 1/p(z)
− 2 1+p(z)
Dp,
ρp (Dun ) p(z)
ρp (Dun ) p(z)

|Dun |p(z)
⇒ |Dvn |p(z) 6 + Ĉn with Ĉn → 0+ as n → ∞,
ρp (Dun )
(7) ⇒ ρp (Dvn ) 6 1 + Ĉn for all n ∈ N.
By the Poincaré inequality we have that
1,p(z)
{vn }n∈N ⊆ W0 (Ω) is bounded.
So, by passing to a suitable subsequence if necessary, we may assume that
w 1,p(z)
(8) vn → v in W0 (Ω) and vn → v in Lp(z) (Ω).
The convexity of the modular function ρp (·) implies that it is sequentially weakly lower
semicontinous. So, from (8) we obtain
ρp (Dv) 6 lim inf ρp (Dvn ),
n→∞
(9) ⇒ ρp (Dv) 6 1 (see (6), and recall Ĉn → 0+ ).
On the other hand from (8) we have
Z Z
ϑ(z)vnp(z) dz → ϑ(z)v p(z) dz,
Ω Ω
Z
p(z)
⇒ 16 ϑ(z)v dz (see (5)),
Ω
Z
(10) ⇒ ρp (Dv) 6 ϑ(z)v p(z) dz 6 λ̂1 ρp (v)
Ω
(see (9) and hypothesis H1 (ii))
⇒ ρp (Dv) = λ̂1 ρp (v) (see (3)).
It follows that
(11) v = µû1 with µ > 0.
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 7

If µ = 0, then v = 0. Recall that

  Z
1
1− 6 ϑ(z)vnp(z) dz for all n ∈ N,
n Ω
⇒ 1 6 0, a contradiction.

Therefore µ > 0 and since û1 ∈ intC+ from (11) we have

(12) v(z) > 0 for all z ∈ Ω.

Returning to (10), from (12) and the condition on ϑ(·) (see hypothesis H1 (ii)), we infer
that
ρp (Dv) < λ̂1 ρp (v),
1,p(z)
a contradiction (see (3)). So, {un }n∈N ⊆ W0 (Ω) is bounded. We may assume that

w 1,p(z)
un → u in W0 (Ω) and un → u in Lp(z) (Ω).

Using (10) and reasoning as above, via inequality (3) we reach a contradiction. This
proves the lemma.


3. Analysis of the ”Frozen” Problem

As we already explained in the Introduction, we will fix (”freeze”) the gradient term in
the reaction. So, let v ∈ C01 (Ω) and consider the Carathéodory function gv (z, x) defined by

gv (z, x) = r(z)|Dv(z)|τ (z)−1 + f (z, x).

Using this function as forcing (source) term, we consider the following anisotropic Dirich-
let problem

(13) −∆p(z) u(z) − ∆q(z) u(z) = gv (z, u(z)) in Ω, u = 0, u > 0.

∂Ω
Z x
This problem is variational. Setting Gv (z, x) = gv (z, s)ds, we consider the C 1 -
0
1,p(z)
functional Ψv : W0 (Ω) → R defined by
Z Z Z
1 1
Ψv (u) = |Du|p(z) dz + |Du|q(z) dz − Gv (z, u+ )dz
Ω p(z) Ω q(z) Ω

1,p(z)
for all u ∈ W0 (Ω).

Proposition 5. If hypotheses H0 , H1 hold, then the functional Ψv (·) is coercive.

Proof. We proceed indirectly. So, suppose that Ψv (·) is not coercive. Then we can find
1,p(z)
{un }n∈N ⊆ W0 (Ω) such that

(14) kun k → ∞ as n → ∞ and Ψv (un ) 6 C2 for some C2 > 0, all n ∈ N.

8 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

From (13) we have

Z Z
1 p(z)
|Dun | dz − F (z, u+
n )dz 6 C3 for some C3 > 0, all n ∈ N,
Ω p(z) Ω
Z Z
1
⇒ |Dun |p(z) dz − F (z, u+n )dz 6 C3 for all n ∈ N,
p+ Ω Ω
Z Z
1 1
⇒ + p(z)
λ̂1 (un ) dz − F (z, u+
n )dz + ρp (Du−
n ) 6 C3 for all n ∈ N(see (3)),
p+ Ω Ω p+
Z h
1 i 1
⇒ λ̂1 (u+
n ) p(z)
− p + F (z, u+
n ) dz + ρp (Du−
n ) 6 C3 for all n ∈ N,
p+ Ω p+
⇒ ρp (Du− n ) 6 C4 for some C4 > 0, all n ∈ N (see hypotesis H1 (iii)),
1,p(z)
(15) ⇒ {u−
n }n∈N ⊆ W0 (Ω) is bounded (see Proposition 1).
From (14) and (15) it follows that
ku+n k → ∞,
(16) ⇒ ρp (Du+n ) → ∞ (see Proposition 1).

u+
n 1,p(z)
We set yn = + 1/p(z) ∈ W0 (Ω). From proof of Lemma 4, we have that
ρp (Dun )
(17) {ρp (Dyn )}n∈N is bounded,
1,p(z)
⇒ {yn }n∈N ⊆ W0 (Ω) is bounded, yn > 0 for all n ∈ N.
We may assume that

w 1,p(z)
(18) yn → y in W0 (Ω) and yn → y in Lp(z) (Ω), y > 0.
From (14) and (15), we have
 Z 
1 + +
ρp (Dun ) − p+ F (z, un )dz 6 C5 for some C5 > 0, all n ∈ N,
p+ Ω
p+ F (z, u+
 Z 
1 n) C5
⇒ ρp (Dyn ) − Ĉn − + dz 6 for all n ∈ N
p+ Ω ρp (Dun ) ρp (Du+n)
(see (7) with vn replaced by yn ),
p+ F (z, u+
Z
1 C5 n)
(19) ⇒ ρp (Dyn ) 6 + + + dz + Ĉn for all n ∈ N.
p+ ρp (Dun ) Ω ρp (Dun )

From the sequential weak lower semicontinuity of the modular function ρp (·) (it is con-
tinuous and convex), we have
(20) ρp (Dy) 6 lim inf ρp (Dyn ) (see (17)).
n→∞

Also, from (16) we have

C5
(21) → 0 as n → ∞.
ρp (Du+n)

Moreover, note that

p(z)
p+ F (z, u+ p+ F (z, u+ u+

n) n) n
=
ρp (Du+ n) (u+n)
p(z) ρp (Du+
n)
1/p(z)
+
p+ F (z, un ) p(z)
= yn for all n ∈ N.
(u+n)
p(z)
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 9

From hypothesis H1 (ii), we have

p+ F (z, u+
n (z))
lim sup + p(z) 6 ϑ(z) for a.a. z ∈ {y > 0},
n→∞ (un )
p+ F (z, u+
Z Z
n ) p(z)
(22) ⇒ lim sup + p(z) yn dz 6 ϑ(z)y p(z) dz
n→∞ Ω (un ) Ω
(by Fatou’s lemma).
We return to (19), pass to the limit as n → ∞ and use (20), (21) and (22). We obtain
Z
ρp (Dy) − ϑ(z)y p(z) dz 6 0,
Ω
⇒ C1 ρp (Dy) 6 0 (see Lemma 4),
⇒ y = 0 (by Poincare’s inequality).
Then from (17) and (18), we obtain
(23) ρp (Dyn ) → 0.
But from (6) (with vn replaced by yn ) and using (16) we see that
(24) ρp (Dyn ) > C6 > 0 for all n > n0 .
Comparing (23) and (24) we have a contradiction.

Let Sv+ denote the set of positive solutions of problem (13) (the ”frozen problem”).
Proposition 6. If hypotheses H0 , H1 hold, then ∅ 6= Sv+ ⊆ intC+ .
Proof. From Proposition 5, we know that Ψv (·) is coercive. Also it is sequentially weakly
lower semicontinuous. Therefore by the Weierstrass-Tonelli theorem, we can find u0 ∈
1,p(z)
W0 (Ω) such that
h i
1,p(z)
(25) Ψv (u0 ) = min Ψv (u) : u ∈ W0 (Ω) .
Let u ∈ intC+ and let δ > 0 be as postulated by hypothesis H1 (iv). We choose t ∈ (0, 1)
small such that
0 6 tu(z) 6 δ for all z ∈ Ω.
We have
tq− tµ+
Ψt (tu) 6 [ρp (Du) + ρq (Du)] − ρµ (u)
q− µ+
(see hypothesis H1 (iv) and recall that t ∈ (0, 1)).
Since t ∈ (0, 1) and µ+ < q− < p− , by choosing t ∈ (0, 1) even smaller, we have
Ψv (tu) < 0,
⇒ Ψv (u0 ) < 0 = Ψv (0) (see (25)),
⇒ u0 6= 0.
From (25) we have
Ψ0v (u0 ) = 0,
Z
1,p(z)
(26) ⇒ hAp(z) (u0 ), hi + hAq(z) (u0 ), hi = gv (z, u+
0 )hdz for all h ∈ W0 (Ω).
Ω
1,p(z)
In (26) we choose h = −u−
0 ∈ W0 (Ω). We obtain
ρp (Du− −
0 ) + ρq (Du0 ) 6 0,
⇒ u0 > 0, u0 6= 0.
10 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

From (26) it follows that u0 ∈ Sv+ 6= ∅.

Then Theorem 4.1 of Fan-Zhao [10] (see also Gasinski-Papageorgiou [15], Proposition
3.1 and Papageorgiou-Rădulescu-Zhang [23], Proposition A1), we have that u0 ∈ L∞ (Ω).
Then Corollary 3.1 of Tan-Fang [29] implies that u0 ∈ C+ \ {0}. Finally anisotropic maxi-
mum principle of Papageorgiou-Rădulescu-Zhang [23], Proposition A2 (see also Zhang [31]),
implies that u0 ∈ intC+ .
We conclude that ∅ 6= Sv+ ⊆ intC+ .

Hypotheses H1 (i), (ii) imply that
h i
0 6 f (z, x) 6 C7 1 + xp(z)−1 for a.a. z ∈ Ω, all x > 0, some C7 > 0.

This growth condition combined with hypothesis H1 (iv), imply that given r ∈ (p+ , p∗− ),
we can find C8 = C8 (r) > 0 such that
f (z, x) > C1 xµ(z)−1 − C8 xr−1 for a.a. z ∈ Ω, all x > 0,
(27) ⇒ gv (z, x) > C1 xµ(z)−1 − C8 xr−1 for a.a. z ∈ Ω, all x > 0
(see hypotheses H0 ).
This unilateral growth condition on gv (z, ·) leads to the following auxiliary Dirichlet
problem
−∆p(z) u(z) − ∆q(z) u(z) = C1 u(z)µ(z)−1 − C8 u(z)r−1 in Ω,
(
(28)
u = 0, u > 0.
∂Ω

Proposition 7. If hypotheses H0 hold, then problem (28) admits a unique positive solution
u ∈ intC+ .
Proof. First we show the existence of a positive solution. To this end we consider the
1,p(z)
C 1 -functional τ : W0 (Ω) → R defined by
Z Z Z
1 p(z) 1 q(z) C8 + r C1 + µ(z)
τ (u) = |Du| dz + |Du| dz + ku kr − (u ) dz
Ω p(z) Ω q(z) r Ω µ(z)
1,p(z)
for all u ∈ W0 (Ω).
Since µ+ < q− , we see that τ (·) is coercive. Also it is sequentially weakly lower semicon-
1,p(z)
tinuous. So, we can find u ∈ W0 (Ω) such that
h i
1,p(z)
(29) τ (u) = min τ (u) : u ∈ W0 (Ω) < 0 = τ (0)
(recall µ+ < q− < p+ < r),
⇒ u 6= 0.
From (29) we have
τ 0 (u) = 0,
Z Z
+ µ(z)−1
(30) ⇒ hAp(z) (u), hi + hAq(z) (u), hi = C1 (u ) hdz − C8 (u+ )r+ dz
Ω Ω
1,p(z)
for all h ∈ W0 (Ω).
1,p(z)
In (30) we use h = −u− ∈ W0 (Ω) and obtain
ρp (Du− ) + ρq (Du− ) = 0,
⇒ u > 0, u 6= 0.
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 11

1,p(z)
It follows that u ∈ W0 (Ω) is positive solution of (28). As before (see the proof of
Proposition 6), the anisotropic regularity theory and the maximum principle imply that
u ∈ intC+ .
Next we show the uniqueness of this positive solution of problem (28). To this end, we
introduce the integral functional j : L1 (Ω) → R = R ∪ {+∞} defined by
 Z Z
1 1/µ+ 1 1,p(z)
Du dz + Du1/µ+ dz, if u > 0, u1/µ+ ∈ W0 (Ω),

j(u) = Ω p(z) Ω q(z)
 +∞, otherwise.
We set domj = u ∈ L1 (Ω) : j(u) < +∞ (the effective domain of j(·)). Theorem 2.2 of


Takač-Giacomoni [28], implies that j(·) is convex.

1,p(z)
Suppose that v ∈ W0 (Ω) is another positive solution of problem (28). Again we have
v ∈ intC+ . Then using Proposition 4.1.22, p. 274, of Papageorgiou-Rădulescu-Repovš [21],
we have
u v
, ∈ L∞ (Ω).
v u
1,p(z)
Let h = uµ+ − v µ+ ∈ W0 (Ω). For |t| < 1 small, we have
uµ+ + th ∈ domj, v µ+ + th ∈ domj.
Then the convexity of j(·) implies the Gateaux differentiability of j(·) at uµ+ and at v µ+
in the direction. Moreover, using Green’s identity we obtain
−∆p(z) u − ∆q(z) u
Z
0 µ+ 1
j (u )(h) = hdz
µ+ Ω uµ+ −1
Z  
1 C1 r−µ+
= − C8 u hdz
µ+ Ω uµ+ −µ(z)
−∆p(z) v − ∆q(z) v
Z
0 µ+ 1
j (v )(h) = hdz
µ+ Ω v µ+ −1
Z  
1 C1 r−µ+
= − C8 v hdz.
µ+ Ω v µ+ −µ(z)
The convexity of j(·) implies the monotonicity of j 0 (·). So, we have
Z  
1 1
06 µ+ −µ(z)
− µ −µ(z) (uµ+ − v µ+ )dz
Ω u v +
Z
+ C8 v r−µ+ − ur−µ+ (uµ+ − v µ+ )dz 6 0,
 
Ω
⇒ u = v.
Therefore u ∈ intC+ is the unique positive solution of problem (28).

Proposition 8. If hypotheses H0 , H1 hold, then u 6 u for all u ∈ Sv+ .
Proof. Let u ∈ Sv+ . We introduce the Carathéodory function k(z, x) defined by
C1 (x+ )µ(z)−1 − C8 (x+ )r−1 , if x 6 u(z)

(31) k(z, x) =
C1 u(z)µ(z)−1 − C8 u(z)r−1 , if u(z) < x.
Z x
1,p(z)
We set K(z, x) = k(z, s)ds and consider the C 1 -functional τ̂ : W0 (Ω) → R defined
0
by Z Z Z
1 1
τ̂ (u) = |Du|p(z) dz + |Du|q(z) dz − K(z, u)dz
Ω p(z) Ω q(z) Ω
12 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

1,p(z)
for all u ∈ W0 (Ω).
From (31) and Poincaré’s inequality, we see that τ̂ (·) is coercive. Also, it is sequentially
1,p(z)
weakly lower seimicontinuous. So, we can find û ∈ W0 (Ω) such that
h i
1,p(z)
(32) τ̂ (û) = min τ̂ (u) : u ∈ W0 (Ω) < 0 = τ̂ (0)
(as before since µ+ < q− < p+ < r),
⇒ û 6= 0.
From (32) we have
τ̂ 0 (û) = 0,
Z
1,p(z)
(33) ⇒ hAp(z) (û), hi + hAq(z) (û), hi = k(z, û)hdz for all h ∈ W0 (Ω).
Ω
− 1,p(z)
In (33) first we choose h = −û ∈ W0 (Ω) and obtain û > 0, û 6= 0. Next in (33) we
1,p(z)
use h = (û − u)+ ∈ W0 (Ω). We have
hAp(z) (û), (û − u)+ i + hAq(z) (û), (û − u)+ i
Z h i
= C1 uµ(z)−1 − C8 ur−1 (û − u)+ dz (see (31))
ZΩ
6 gv (z, u)(û − u)+ dz (see (27))
Ω
= hAp(z) (u), (û − u)+ i + hAq(z) , (û − u)+ i (since u ∈ Sv+ ),
⇒ û 6 u.
So, we have proved that
(34) û ∈ [0, u], û 6= 0.
From (34), (31), (33) and Proposition 7 we infer that
û = u ∈ intC+ ,
⇒ u 6 u for all u ∈ Sv+ (see (34)).

Using this lower bound, we can show that Sv+ has a smallest element (minimal positive
solution). So, we have a canonical way to choose an element from the solution set Sv+ as v
varies (a selection of the solution multifunction v 7→ Sv+ ).
Proposition 9. If hypotheses H0 , H1 hold, then there exists ũv ∈ Sv+ such that ũv 6 u for
all u ∈ Sv+ .
Proof. From Papageorgiou-Rădulescu-Repovš [20] (see proof of Proposition 7), we know
that Sv+ is downward directed. So using Lemma 3.10, p. 178, of Hu-Papageorgiou [17], we
can find a decreasing sequence {un }n∈N ⊆ Sv+ such that
inf un = inf Sv+ .
n∈N
We have
Z
1,p(z)
(35) hAp(z) (un ), hi + hAq(z) (un ), hi = gv (z, un )hdz for all h ∈ W0 (Ω), all n ∈ N,
Ω

(36) u 6 un 6 u1 for all n ∈ N (see Proposition 8).

1,p(z)
In (35) we choose h = un ∈ W0 (Ω). Using (36) and hypothesis H1 (i) we infer that
1,p(z)
{un }n∈N ⊆ W0 (Ω) is bounded.
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 13

So, by passing to a suitable subsequence if necessary, we may assume that

w 1,p(z)
(37) un → ũv in W0 (Ω) and un → ũv in Lp(z) (Ω).
1,p(z)
In (35) we choose the test function h = un − ũn ∈ W0 (Ω), pass to the limit as n → ∞
and use (37). Then
 
lim hAp(z) (un ), un − ũv i + hAq(z) (un ), un − ũv i = 0,
n→∞
 
⇒ lim sup hAp(z) (un ), un − ũv i + hAq(z) (un ), un − ũv 6 0
n→∞
(since Aq(z) (·) is monotone),
⇒ lim suphAp(z) (un ), un − ũv i 6 0 (see (37)),
n→∞
1,p(z)
(38) ⇒ un → ũv in W0 (Ω) (see Proposition 2).
In (35) we pass to the limit as n → ∞ and use (38). Then
Z
1,p(z)
hAp(z) (ũv ), hi + hAq(z) (ũv ), hi = gv (z, ũv )hdz for all h ∈ W0 (Ω),
Ω
u 6 ũv .
In follows that ũv ∈ Sv+ and ũv = inf Sv+ .

So, we define the minimal solution map β : C01 (Ω) → C01 (Ω) by
β(v) = ũv ∈ intC+ for all v ∈ C01 (Ω).
Clearly a fixed point of this map will be the positive solution of problem (1). To produce
a fixed point of β(·), we will use the Leray-Schauder Alternative Principle (see Theorem 3).
This theorem requires that the minimal solution map β(·) is compact. We prove this in the
next section.

4. The Minimal Solution Map

In this section we show that the minimal solution map β : C01 (Ω) → C01 (Ω) is compact.
To this end the following proposition is helpful.
Proposition 10. If hypotheses H0 , H1 hold, vn → v in C01 (Ω) and u ∈ Sv+ , then we can
find un ∈ Sv+n n ∈ N such that un → u in C01 (Ω).
Proof. First we consider the following anisotropic Dirichlet problem
(39) −∆p(z) y(z) − ∆q(z) y(z) = gvn (z, u(z)) in Ω, y = 0, y > 0, n ∈ N.
∂Ω
Hypotheses H0 , H1 (i) imply that gvn (·, u(·)) ∈ L∞ (Ω). We consider the operator V :
1,p(z) 0 1,p(z)
W0 (Ω) → W −1,p (z) (Ω) = W0 (Ω)∗ defined by
1,p(z)
V (u) = Ap(z) (u) + Aq(z) (u) for all u ∈ W0 (Ω).
By Proposition 2, V (·) is continuous, strictly monotone (thus, maximal monotone too)
and also we have
1,p(z)
hV (u), ui = ρp (Du) + ρq (Du) for all u ∈ W0 (Ω),
⇒ V (·) is coercive (see Proposition 1).
Therefore V (·) is surjective (see Papageorgiou-Rădulescu-Repovš [21], Corollary 2.8.7, p.
1,p(z)
135). So, we can find yn ∈ W0 (Ω), yn > 0, yn 6= 0 such that
(40) V (yn ) = Ngvn (u) for all n ∈ N.
14 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

The strict monotonicity of V (·) implies that this solution of problem (39) is unique. On
1,p(z) 1,p(z)
(40) we act with yn ∈ W0 (Ω) and obtain that {yn }n∈N ⊆ W0 (Ω) is bounded. Then
the anisotropic regularity theory (see [10], [23]) implies that

(41) yn ∈ L∞ (Ω), kyn k∞ 6 C9 for some C9 > 0, all n ∈ N.

Then invoking Lemma 3.3 of Fukagai-Narakuwa [12], we can find α ∈ (0, 1) and C10 > 0
such that
yn ∈ C01,α (Ω), kyn kC 1,α (Ω) 6 C10 for all n ∈ N.
0

We know that C01,α (Ω) ,→ C01 (Ω) compactly. Therefore by passing to a subsequence if
necessary, we can have

(42) yn → ũ in C01 (Ω) as n → ∞.

Passing to the limit as n → ∞ in (40) and using (42), we obtain

V (ũ) = Ngv (u).

⇒ ũ = u (from the uniqueness of the solution).

Therefore for the original sequence we have

yn → u in C01 (Ω).

Next we consider the following anisotropic Dirichlet problem

−∆p(z) w(z) − ∆q(z) w(z) = gvn (z, yn (z)) in Ω, w = 0, w > 0.

∂Ω

Reasoning as above, we infer that this problem has a unique positive solution wn1 ∈ intC+
and
wn1 → u in C01 (Ω) as n → ∞.

Setting wn0 = yn and continuing this way, we generate a sequence {wnk }n∈N0 ⊆ intC+ such
that
1,p(z)
(43) V (wnk ) = Ngvn (wnk−1 ) in W0 (Ω)∗ for all k, n ∈ N,
(44) wnk → u in C01 (Ω) as n → ∞ for every k ∈ N.
1,p(z)
Claim: For every n ∈ N, the sequence {wnk }k∈N ⊆ W0 (Ω) is bounded.
To prove the Claim, we argue by contradiction. So, suppose that at least for a subse-
quence, we have
kwnk k → ∞ as k → +∞.
Then, we can say that
n o
(45) ρp (Dwnk ) → +∞ as k → +∞, ρp (Dwnk ) is nondecreasing.
k∈N

wnk 1,p(z) 1,p(z)

We set x̂k = ∈ W0 (Ω), k ∈ N0 . On (43) we act with wnk ∈ W0 (Ω)
ρp (Dwnk )1/p(z)
and obtain
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 15

Z
ρp (Dwnk ) + ρq (Dwnk ) = gvn (z, wnk−1 )wnk dz
Ω
" #q(z)
|Dwnk |
Z
1
⇒ ρp (Dx̂k ) − Ĉk + p(z)−q(z) 1 dz
Ω
ρp (Dwnk ) p(z) ρp (Dwnk ) p(z)
(see (7) with vn replaced by x̂k )
gvn (z, wnk−1 )
Z
= 1 x̂k dz
Ω ρp (Dw k )1− p(z)
n
gvn (z, wnk−1 )
Z
6 1 x̂k dz (see (45))
Ω ρp (Dwn k−1 1− p(z)
)
" #p(z)−1
gvn (z, wnk−1 ) wnk−1
Z
= k−1 p(z)−1 1 x̂k dz
Ω (wn ) ρp (Dwnk−1 ) p(z)
gvn (z, wnk−1 ) p(z)−1
Z
= x̂
k−1 p(z)−1 k−1
x̂k dz,
Ω (wn )
gvn (z, wnk−1 ) p(z)−1
Z
(46) ⇒ ρ(Dxˆk ) 6 x̂
k−1 p(z)−1 k−1
x̂k dz + Ĉk for all k ∈ N.
Ω (wn )

From the proof of Lemma 4, we know that

(47) 0 < C11 6 kx̂k k 6 C12 for all k > k0 , some 0 < C11 6 C12 .
So we may assume that
w 1,p(z)
(48) x̂k → x̂ in W0 (Ω) and x̂k → x̂ in Lp(z) (Ω) as k → ∞.
We return to (46), pass to the limit as k → ∞ and use (48), the fact that ρp (·) is
sequentially weakly lower semicontinuous (being continuous convex) and hypothesis H1 (ii).
We obtain
Z
ρp (Dx̂) 6 ϑ(z)x̂p(z) dz,
Ω
⇒ C1 ρp (Dx̂) 6 0 (see Lemma 4),
⇒ x̂ = 0.
So, from (46) we have that
ρp (Dx̂k ) → 0 as k → ∞,
1,p(z)
⇒ x̂k → 0 in W0 (Ω) (see Proposition 1),
which contradicts (47).
This proves the Claim.
As before using the Claim and the anisotropic regularity theory, we can find α ∈ (0, 1)
and C13 > 0 such that
wnk ∈ C01,α (Ω), kwnk kC 1,α (Ω) 6 C13 for all k ∈ N.
0

Since C01,α (Ω) ,→ C01 (Ω) compactly, at least for a subsequence we have

wnk → un in C01 (Ω) as k → ∞, un 6= 0 for all n ∈ N.

16 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

From (43) in the limit as k → ∞, we obtain

1,p(z)
V (un ) = Ngvn (un ) in W0 (Ω)∗ for all n ∈ N,
(49) ⇒ un ∈ Sv+n for all n ∈ N.
As we did in the proof of the Claim, via a contradiction argument, we also show that
1,p(z)
{un }n∈N ⊆ W0 (Ω) is bounded and from this we infer that {un }n∈N ⊆ C01 (Ω) is relatively
compact. So, we can say that
un → ũ in C01 (Ω) as n → ∞.

The double limit lemma (see Gasinski-Papageorgiou [14], Proposition A.2.35, p. 906),
implies that ũ = u and so finally we have
un → u in C01 (Ω) and un ∈ Sv+ for all n ∈ N (see (49)) .

Using this proposition, we can show the compactness of the minimal solution map β(·).
Proposition 11. If hypotheses H0 , H1 hold, then the minimal map β(·) : C01 (Ω) → C01 (Ω)
is compact.
Proof. First we show that β(·) maps bounded sets in C01 (Ω) onto relatively compact subsets
of C01 (Ω).
So, let B ∈ C01 (Ω) be bounded. Hypotheses H1 (i), (ii) imply that given ε > 0, we can
find Cε > 0 such that
(50) 0 6 f (z, x) 6 [ϑ(z) + ε]xp(z)−1 + Cε for a.a. z ∈ Ω, all x > 0.
For v ∈ B, we write β(v) = ũv ∈ intC+ . We have
hAp(z) (ũv ), hi + hAq(z) (ũv ), hi
Z h i
= r(z)|Dv|τ (z)−1 + f (z, ũv ) hdz
ZΩ h i
(51) 6 r(z)|Dv|τ (z)−1 + (ϑ(z) + ε)ũτv (z)−1 + Cε hdz
Ω
1,p(z)
for all h ∈ W0 (Ω), h > 0 see (50)
1,p(z)
In (51) we choose h = ũv ∈ W0 (Ω), h > 0. We obtain
Z h i
ρp (Dũv ) − ϑ(z)ũvp(z) dz − ερp (ũv ) 6 C14 1 + ρp (Dũv )1/p−
Ω
for some C14 > 0,
 
ε h i
(52) ⇒ C1 − ρp (Dũv ) 6 C14 1 + ρp (Dũv )1/p−
λ̂1
(see (3) and Lemma 4).
 
Choosing ε ∈ 0, λ̂1 C1 , since p− > 1, from (52) we infer that

{ρp (Dũv )}v∈B ⊆ R+ is bounded,

1,p(z)
⇒ {ũv }v∈B ⊆ W0 (Ω) is bounded (see Proposition 1).
From this as before we obtain that
{ũv = β(v)}v∈B ⊆ C01 (Ω) is relatively compact.
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 17

Next we show that β(·) is continuous. Suppose that vn → v in C01 (Ω). According to
Proposition 10, we can find un ∈ Sv+n ⊆ intC+ n ∈ N such that

(53) un → β(v) = ũv in C01 (Ω) as n → ∞.

From the first part of the proof, we have that

{β(vn )}n∈N ⊆ C01 (Ω) is relatively compact.

So, for at least a subsequence we have

(54) β(vn ) → ũ∗ in C01 (Ω) as n → ∞.

Recall that u 6 β(vn ) for all n ∈ N (see Proposition 8). Hence u 6 ũ∗ and so using (54)
we conclude that ũ∗ ∈ Sv+ ⊆ intC+ . Then

(55) β(v) 6 ũ∗ .

On the other hand, we have

β(vn ) 6 un for all n ∈ N,

⇒ ũ∗ 6 β(v) = ũv (see (54) and (53))
⇒ ũ∗ = β(v) (see (55))
⇒ β(vn ) → β(v) in C01 (Ω) (see (54)),
⇒ β(·) is compact.

5. Positive solution
In this section using the Leray-Schauder Alternative Principle (see Theorem 3) on the
minimal solution map β(·), we produce a fixed point which is a positive solution of problem
(1).
We introduce the set

D = {u ∈ C01 (Ω) : u = tβ(u), 0 < t < 1}.

Proposition 12. If hypotheses H0 , H1 hold, then D ⊆ C01 (Ω) is bounded.

Proof. Let u ∈ D. Then u ∈ intC+ and

1
u = β(u) with 0 < t < 1.
t
So, we have
   
1 1
hAp(z) u , hi + hAq(z) u , hi
t t
Z Z  
1
(56) = r(z)|Du|τ (z)−1 hdz + f z, u, u hdz
Ω Ω t
1,p(z)
for all h ∈ W0 (Ω).
18 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

1,p(z)
In (56) we choose h = u ∈ W0 (Ω). Then
Z Z  
1 τ (z)−1 1
ρp (Du) 6 r(z)|Du| udz + f z, u udz
tp− Ω Ω t
Z Z
1
6 r(z)|Du|τ (z)−1 udz + p+ −1 f (z, u)udz
Ω t Ω
(see hypothesis H1 (v)),
Z Z
τ (z)−1 1−(p+ −p− )
⇒ ρp (Du) 6 r(z)|Du| udz + t f (z, u)udz
ZΩ Z Ω

6 r(z)|Du|τ (z)−1 udz + f (z, u)udz

Ω Ω
(since by hypotheses H0 0 6 p+ − p− 6 1 and t ∈ (0, 1)),
Z Z
p(z)
⇒ ρp (Du) − ϑ(z)u dz − ερp (u) 6 r(z)|Du|τ (z)−1 udz + C15
Ω Ω
for some C15 = C15 (ε) > 0 (see (50)).
Using Lemma 4 and choosing ε ∈ (0, λ̂1 C1 ) (see (3)), we obtain
 Z 
(57) ρp (Du) 6 C16 1 + r(z)|Du|τ (z)−1 udz for some C16 > 0.
Ω

Using Hölder’s inequality and the Sobolev embedding theorem (see Section 2), we have
Z
(58) r(z)|Du|τ (z)−1 udz 6 C17 k|Du|τ (z)−1 kτ 0 (z) kuk
Ω
for some C17 > 0.
We may assume that
(59) k|Du|τ (z)−1 kτ 0 (z) > 1, kuk > 1.
Then from (58) it follows that
Z
r(z)|Du|τ (z)−1 udz
Ω
  1
0
τ (z)−1 τ+
6 C18 ρτ 0 |Du| kuk for some C18 > 0.
(see (59) and Proposition 1)
  10 1
6 C19 ρτ 0 |Du|τ (z)−1 + ρp (Du) p+
τ
(60)
(using the Poincaré inequality, (59) and Proposition 1).
We return to (57) and use (60). We obtain
1− p1
h i
τ+ −1
ρp (Du) + 6 C20 1 + kDukτ+
for some C20 > 0 (see Proposition 1),
p−
(p+ −1)
6 C21 1 + kukτ+ −1
 
(61) ⇒ kuk p+

for some C21 > 0 (see (59) and Proposition 1).

p−
But [p+ − 1] > p− − 1 > τ+ − 1 (see hypotheses H0 ). So, from (61) it follows that
p+
1,p(z)
D ⊆ W0 ()Ω is bounded.
 ANISOTROPIC (p, q)-EQUATIONS WITH GRADIENT DEPENDENT REACTION 19

As before (see the proof of Proposition 6), using the anisotropic regularity theory, we
infer that
D ⊆ C01 (Ω) is relatively compact, thus bounded.


Therefore we have proved that

• the minimal solution map β(·) is compact (see Proposition 11);
• D ⊆ C01 (Ω) is bounded (see Proposition 12).
So, we can apply Theorem 3 (the Leray-Schauder Alternative Principle) and have the
following existence theorem for problem (1).
Theorem 13. If hypotheses H0 , H1 hold, then problem (1) has a positive solution û ∈
intC+ , u 6 û.

References
[1] A. Bahrouni-V.D. Rădulescu-D.D. Repovš: ”Double phase transonic flow problems with variable
growth: nonlinear patterns and stationary waves” Nonlinearity 32 (2019), 2481-2495.
[2] Y. Bai: ”Existence of solutions to nonhomogeneous Dirichlet problem with dependence on the
gradient” Electr. Jour. Diff. Equ 2018: 101, pp 1-18.
[3] Y. Bai-L. Gasinski-N.S Papageorgiou: ”Nonlinear Dirichlet problems with the combined effects
of singular and convection terms” Electr. Jour. Diff. Equ 2019: 57, pp. 1-13.
[4] S.S. Byun-E. Ko: ”Global C 1,α regularity and existence of multiple solutions for singular p(x)-
Laplacian equations” Calc. Var. PDEs 56(2017), No. 76, 29 pp.
[5] P. Candito-L. Gasinski-N.S. Papageorgiou: ”Nonnilnear nonhomogeneous Robin problems with
convection” Annales Acad. Scient. Fennicae: Mathematica 44 (2019), 755-767.
[6] M. Cencelj-V.D. Rădulescu-D.D. Repovš: ”Double phase problems with variable growth” Non-
linear Anal. 177 (2018), 270-287.
[7] L. Diening-P. Harjulehto-P. Hästo-M. Ruzička: ”Lebesgue and Sobolev Spaces with Variable
Exponents” Lecture Notes in Math. Vol. 2017, Springer, Heildberg (2011).
[8] X. Fan: ”Remarks on eigenvalue problems involving the p(x)-Laplacian” J. Math. Anal. Appl.
352(2009), 85-98.
[9] X. Fan-D. Zhang-D. Zhao: ”Eigenvalues of p(x)-Laplacian Dirichlet problem” J. Math. Anal.
Appl. 302(2005), 306-317.
[10] X. Fan-D. Zhao: ”A class of De Giorgi type Hölder continuity” Nonlinear Anal. 36 (1999),
295-318.
[11] L.F.O. Faria-O.H. Miyagaki-D. Motreanu: ”Comparison and positive solutions for problems
with (p, q)-Laplacian and convection term” Proc. Edinh. Math. Soc. 57(2014), 687-698.
[12] N. Fukagai-K. Narukawa: ”On the existence of multiple positive solutions of quasilinear elliptic
eigenvalue problems” Annali Mat. Pura Appl. 186 (2007), 539-564.
[13] L. Gasinski-I. Krech- N.S. Papageorgiou: ”Nonlinear nonhomogeneous Robin problems with
gradient dependent reaction” Nonlinear Anal-RWA 55(2020), 103-135.
[14] L. Gasinski-N.S. Papageorgiou: ”Nonlinear Analysis” Chapman & Hall/CRC, Boca Raton, Fl.
2006.
[15] L. Gasinski-N.S. Papageorgiou: ”Anisotropic nonlinear Neumann problems” Calc. Var. 42
(2011), 323-354.
[16] L. Gasinski-P. Winkert: ”Existence and uniqueness results for double phase problems with
convection term” J. Differential Equ., doi:1011016/j.jde.2019.10022.
[17] S. Hu-N.S. Papageorgiou: ”Handbook of Multivalued Analysis. Volume I: Theory” Kluwer
Academic Publishers, Dodrecht, The Netherlands, 1997.
[18] Z. Liu-N.S. Papageorgiou: ”Positive solutions for resonant (p, q)-equations with convection”
Adv. Nonlin. Anal. 10(2021), 217-232.
[19] S. Marano-P. Winkert: ”On a quasilinear elliptic problem with convection term and a nonlinear
boundary condition” Nonlinear Anal. 187(2019), 159-169.
20 N.S. PAPAGEORGIOU, D. D. REPOVŠ, AND V.D. RĂDULESCU

[20] N.S. Papageorgiou-V.D. Rădulescu-D.D. Repovš: ”Positive solutions for perturbations of the
Robin eigenvalue problem plus an indefinite potential” Discr. Cont. Dyn. Syst. 37 (2017), 2589-
2618.
[21] N.S. Papageorgiou-V.D. Rădulescu-D.D. Repovš: ”Nonlinear Analysis-Theory and Methods”
Springer,Cham 2019.
[22] N.S. Papageorgiou-V.D. Rădulescu-D.D. Repovš: ”Positive solutions for nonlinear Neumann
problems with singular terms and convection” J. Math. Pures Appl. 136(2020), 1-21.
[23] N.S. Papageorgiou-V.D. Rădulescu-Y. Zhang: ”Anisotropic singular double phase Dirichlet
problems” submitted.
[24] N.S. Papageorgiou-C. Vetro-F. Vetro: ”Nonlinear Robin problems with unilateral constraints
and dependence on the gradient” Electr. Jour. Diff. Equ. 2018: 182, pp. 1-14.
[25] N.S. Papageorgiou-P. Winkert: ”Applied Nonlinear Functional Analysis” De Gruyter, Berlin,
2018.
[26] V.D. Rădulescu: ”Isoptropic and anisotropic double phase problems: old and new” Opuscula
Math. 39(2019), 259-280.
[27] V.D. Rădulescu-D.D. Repovš: ”Partial Differential Equations with Variable Exponents: Varia-
tional Methods and Qualitative Analysis” CRC Press, Boca Raton, FL. 2015.
[28] P. Takač-J. Giacomoni: ”A p(x)-Laplacian extension of the Diaz-Saa inequality and some ap-
plications” Proc. Royal Soc. Edinburgh, DOI:10.1007|prm.2018.91
[29] Z. Tan-F. Fang: ”Orlicz-Sobolev versus Hölder local minimizer and multiplicity results for
quasilinear elliptic equations” J. Math. Anal. Appl. 402(2013), 348-370.
[30] S. Zeng-Z. Liu-S. Migorski: ”Positive solutions to nonlinear nonhomogeneous inclusion problems
with dependence on the gradient” J. Math. Anal. Appl. 463(2018), 432-448.
[31] Q. Zhang: ”A strong maximum principle for differential equations with nonstandard p(x)-
growth conditions” J. Math. Anal. Appl. 312 (2005), 24-32.
[32] V.V. Zhikov: ”On variational problems and nonlinear elliptic equations with nonstandard
growth conditions” J. Math. Sci. 173(2011), 463-570.

(N.S. Papageorgiou) National Technical University, Department of Mathematics, Zografou

Campus, Athens 15780, Greece
Email address: [email protected]

(D.D. Repovš) University of Ljubljana, Faculty of Education and Faculty of Mathematics

and Physics, Kardeljeva ploscad 16, SI-1000 Ljubljana, SLOVENIA
Email address: [email protected]

(V.D. Rădulescu) Faculty of Applied Mathematics, AGH University of Science and Technol-
ogy, al. Mickiewicza 30, 30-059 Kraków, Poland & Department of Mathematics, University
of Craiova, 200585 Craiova, Romania
Email address: [email protected]