Electronic Journal of Differential Equations, Vol. 2021 (2021), No. 26, pp. 1–28.

ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu

CONTINUOUS IMBEDDING IN MUSIELAK SPACES WITH AN

APPLICATION TO ANISOTROPIC NONLINEAR

NEUMANN PROBLEMS

AHMED YOUSSFI, MOHAMED MAHMOUD OULD KHATRI

Abstract. We prove a continuous embedding that allows us to obtain a

boundary trace imbedding result for anisotropic Musielak-Orlicz spaces, which

we then apply to obtain an existence result for Neumann problems with non-
linearities on the boundary associated to some anisotropic nonlinear elliptic

equations in Musielak-Orlicz spaces constructed from Musielak-Orlicz func-

tions on which and on their conjugates we do not assume the ∆2-condition.
The uniqueness of weak solutions is also studied.

1. Introduction

Let Ω be an open bounded subset of RN , (N ≥ 2). We denote by ~φ : Ω×R+ →
RN the vector function ~φ = (φ1, . . . , φN ) where for every i ∈ {1, . . . , N}, φi is a
Musielak-Orlicz function differentiable with respect to its second argument whose
complementary Musielak-Orlicz function is denoted by φ∗i (see preliminaries). We
consider the problem

−
N∑
i=1

∂xiai(x, ∂xiu) + b(x)ϕmax(x, |u(x)|) = f(x, u) in Ω,

u ≥ 0 in Ω,

N∑
i=1

ai(x, ∂xiu)νi = g(x, u) on ∂Ω,

(1.1)

where ∂xi = ∂
∂xi

and for every i = 1, . . . , N , we denote by νi the ith component

of the outer normal unit vector and ai : Ω × R → R is a Carathéodory function
such that there exist a locally integrable Musielak-Orlicz function (see definition 1.1
below) Pi : Ω× R+ → R+ with Pi � φi, a positive constant ci and a nonnegative
function di ∈ Eφ∗i (Ω) satisfying for all s, t ∈ R and for almost every x ∈ Ω the
following assumptions

|ai(x, s)| ≤ ci
(
di(x) + (φ∗i )

−1(x, Pi(x, s))
)
, (1.2)

φi(x, |s|) ≤ ai(x, s)s ≤ Ai(x, s), (1.3)

2010 Mathematics Subject Classification. 46E35, 35J20, 35J25, 35B38, 35D30.
Key words and phrases. Musielak-Orlicz space; imbedding; boundary trace imbedding;

weak solution; minimizer.
c©2021 Texas State University.

Submitted April 12, 2019. Published April 5, 2021.

1



2 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26(
ai(x, s)− ai(x, t)

)
·
(
s− t

)
> 0, for all s 6= t, (1.4)

the function Ai : Ω× R→ R is defined by

Ai(x, s) =

∫ s

0

ai(x, t)dt.

Here and in what follows, we define

φmin(x, s) = min
i=1,...,N

φi(x, s) and φmax(x, s) = max
i=1,...,N

φi(x, s).

Let ϕmax(x, y) = ∂φmax

∂y (x, y). We also assume that there exist a locally integrable

Musielak-Orlicz function R : Ω × R+ → R+ with R � φmax and a nonnegative
function D ∈ Eφ∗max

(Ω), such that for all s, t ∈ R and for almost every x ∈ Ω,

|ϕmax(x, s)| ≤ D(x) + (φ∗max)−1(x,R(x, s)), (1.5)

where φ∗max stands for the complementary function of φmax defined below in (2.1).
As regards the data, we suppose that f : Ω × R+ → R+ and g : ∂Ω × R+ → R+

are Carathéodory functions. We define the antiderivatives F : Ω × R → R and
G : ∂Ω× R→ R of f and g respectively by

F (x, s) =

∫ s

0

f(x, t)dt, G(x, s) =

∫ s

0

g(x, t)dt.

We say that a Musielak-Orlicz function φ satisfies the ∆2-condition, if there exists
a positive constant k > 0 and a nonnegative function h ∈ L1(Ω) such that

φ(x, 2t) ≤ kφ(x, t) + h(x).

Remark that the condition (∆2) is equivalent to the following condition: for all
α > 1 there exists a positive constant k > 0 and a nonnegative function h ∈ L1(Ω)
such that

φ(x, αt) ≤ kφ(x, t) + h(x).

We assume now that there exist two positive constants k1 and k2 and two locally
integrable Musielak-Orlicz functions M and H : Ω × R+ → R+ satisfying the ∆2-
condition and differentiable with respect to their second arguments with M � φ∗∗min,
H � φ∗∗min and H � ψmin, such that the functions f and g satisfy for all s ∈ R+

the following assumptions

|f(x, s)| ≤ k1m(x, s), for a.e. x ∈ Ω, (1.6)

|g(x, s)| ≤ k2h(x, s), for a.e. x ∈ ∂Ω, (1.7)

where

ψmin(x, t) =
(

(φ∗∗min)∗(x, t)
)N−1

N

, m(x, s) =
∂M(x, s)

∂s
,

h(x, s) =
∂H(x, s)

∂s
.

(1.8)

Finally, for the function b involved in (1.1), we assume that there exists a constant
b0 > 0 such that b satisfies the hypothesis

b ∈ L∞(Ω) and b(x) ≥ b0, or a.e. x ∈ Ω. (1.9)

Observe that (1.4) and the relation ai(x, ζ) = ∇ζAi(x, ζ) imply in particular that
for any i = 1, . . . , N , the function ζ → Ai(·, ζ) is convex.

Let us put ourselves in the particular case of ~φ = (φi)i∈{1,...,N} where for i ∈
{1, . . . , N}, φi(x, t) = |t|pi(x) with pi ∈ C+(Ω̄) = {h ∈ C(Ω̄) : infx∈Ω h(x) > 1}.



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 3

Defining pmax(x) = maxi∈{1,...,N} pi(x) and pmin(x) = mini∈{1,...,N} pi(x), one has

φmax(x, t) = |t|pM (x) and then ϕmax(x, t) = pM (x)|t|pM (x)−2t, where pM is pmax

or pmin according to whether |t| ≥ 1 or |t| ≤ 1 and then the space W 1L~φ(Ω) is

nothing but the anisotropic space with variable exponent W 1,~p(·)(Ω), where ~p(·) =
(p1(·), . . . , pN (·)) (see [7] for more details on this space). Therefore, the problem
(1.1) can be rewritten as

−
N∑
i=1

∂xiai(x, ∂xiu) + b1(x)|u|pM (x)−2u = f(x, u) in Ω,

u ≥ 0 in Ω,

N∑
i=1

ai(x, ∂xiu)νi = g(x, u) on ∂Ω,

(1.10)

where b1(x) = pM (x)b(x). Boureanu and Rǎdulescu [2] have proved the existence
and uniqueness of the weak solution of (1.10). They prove an imbedding and a
trace results which they use together with a classical minimization existence result
for functional reflexive framework (see [22, Theorem 1.2]). Problem (1.10) with
Dirichlet boundary condition and b1(x) = 0 was treated in [15]. The authors
proved that if f(·, u) = f(·) ∈ L∞(Ω) then (1.10) admits a unique solution by using
[22, Theorem 1.2]. The problem (1.10) with for all i = 1, . . . , N

ai(x, s) = a(x, s) = sp(x)−1,

with p ∈ C1(Ω) and b1 = g = 0 was treated in [12], where the authors proved the
three nontrivial smooth solutions, two of which have constant sign (one positive, the
other negative). In connection with Neumann problems, the authors [21] studied
the problem

−div a(∇u(z)) +
(
ζ(z) + λ

)
u(z)p−1 = f(z, u(z)) in Ω,

∂u

∂n
= 0 on ∂Ω

u > 0, λ > 0, 1 < p < +∞,

(1.11)

where the function a : RN → RN is strictly monotone, continuous and satisfies
certain other regularity and growth conditions. The function ζ involved in (1.11)
changes its sign and is such that ζ ∈ L∞(Ω). The reaction term f(z, x) is a
Carathéodory function. They proved the existence of a critical parameter value
λ∗ > 0 such that if λ > λ∗ problem (1.11) has at least two positive solutions, if
λ = λ∗ (1.11) has at least a positive solution and if λ ∈ (0, λ∗) problem (1.11) has
no positive solution.

Let us mention some related results in the framework of Orlicz-Sobolev spaces.
Le and Schmitt [17] proved an existence result for the boundary value problem

−div(A(|∇u|2)∇u) + F (x, u) = 0, in Ω,

u = 0 on ∂Ω,

in W 1
0Lφ(Ω) where φ(s) = A(|s|2)s and F is a Carathéodory function satisfying

some growth conditions. This result extends the one obtained in [11] with F (x, u) =
−λψ(u), where ψ is an odd increasing homeomorphism of R onto R. In [11, 17] the
authors assume that the N -function φ∗ complementary to the N -function φ satisfies



4 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

the ∆2 condition, which is used to prove that the functional u →
∫

Ω
Φ(|∇u|) dx is

coercive and of class C1, where Φ is the antiderivative of φ vanishing at the origin.
Here we are interested in proving the existence and uniqueness of the weak

solutions for problem (1.1) without any additional condition on the Musielak-Orlicz
function φi or its complementary φ∗i for i = 1, . . . , N . Therefore, the resulting
Musielak-Orlicz spaces Lφi(Ω) are neither reflexive nor separable and thus classical
existence results can not be applied.

The approach we use consists in proving first a continuous imbedding and a
trace result which we then apply to solve the problem (1.1). The results we prove
extend to the anisotropic Musielak-Orlicz-Sobolev spaces the continuous imbedding
obtained in [6] under some extra conditions and the trace result proved in [18]. The
imbedding result we obtain extends to Musielak spaces a part of the one obtained in
[19] in the anisotropic case and that of Fan [9] in the isotropic case (see Remark 3.2).
In the variable exponent Sobolev space W 1,p(x)(Ω) where 1 < p+ = supx∈Ω p(x) <
N , other imbedding results can be found for instance in [3, 4, 16] while the case
1 ≤ p− ≤ p+ ≤ N was investigated in [13].

To the best of our knowledge, the trace result we obtain here is new and does
not exist in the literature. The main difficulty we found when we deal with problem
(1.1) is the coercivity of the energy functional. We overcome this by using both
our continuous imbedding and trace results. Then we prove the boundedness of
a minimization sequence and by a compactness argument, we are led to obtain a
minimizer which is a weak solution of problem (1.1).

Definition 1.1. Let Ω be an open subset of RN , (N ≥ 2). We say that a Musielak-
Orlicz function φ is locally-integrable, if for every compact subset K of Ω and every
constant c > 0, we have ∫

K

φ(x, c) dx <∞.

The article is organized as follows: Section 2 contains some definitions. In Sec-
tion 3, we give and prove our main results, which we then apply in Section 4 to
solve problem (1.1). In the last section we give an appendix which contains some
important lemmas that are necessary for the accomplishment of the proofs of the
results.

2. Preliminaries

2.1. Anisotropic Musielak-Orlicz-Sobolev spaces. Let Ω be an open subset
of RN . A real function φ : Ω× R+ → R+ will be called a Musielak-Orlicz function
if it satisfies the following conditions

(i) φ(·, t) is a measurable function on Ω.
(ii) φ(x, ·) is an N -function, that is a convex nondecreasing function with

φ(x, t) = 0 if only if t = 0, φ(x, t) > 0 for all t > 0 and for almost ev-
ery x ∈ Ω,

lim
t→0+

φ(x, t)

t
= 0 and lim

t→+∞
inf
x∈Ω

φ(x, t)

t
= +∞.

We will extend these Musielak-Orlicz functions into even functions on all Ω × R.
The complementary function φ∗ of the Musilek-Orlicz function φ is defined by

φ∗(x, s) = sup
t≥0
{st− φ(x, t)}. (2.1)



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 5

It can be checked that φ∗ is also a Musielak-Orlicz function (see [20]). Moreover,
for every t, s ≥ 0 and a.e. x ∈ Ω we have the so-called Young inequality (see [20])

ts ≤ φ(x, t) + φ∗(x, s).

For any function h : R→ R the second complementary function h∗∗ = (h∗)∗ (cf.
(2.1)), is convex and satisfies

h∗∗(x) ≤ h(x), (2.2)

with equality when h is convex. Roughly speaking, h∗∗ is a convex envelope of h,
that is the biggest convex function smaller or equal to h.

Let φ and ψ be two Musielak-Orlicz functions. We say that ψ grows essentially
more slowly than φ, denote ψ � φ, if

lim
t→+∞

sup
x∈Ω

ψ(x, t)

φ(x, ct)
= 0,

for every constant c > 0 and for almost every x ∈ Ω. We point out that if ψ :
Ω × R+ → R+ is locally integrable then ψ � φ implies that for all c > 0 there
exists a nonnegative function h ∈ L1(Ω) such that

ψ(x, t) ≤ φ(x, ct) + h(x), for all t ∈ R and for a.e. x ∈ Ω.

The Musielak-Orlicz space Lφ(Ω) is defined by

Lφ(Ω) =
{
u : Ω→ R measurable :

∫
Ω

φ
(
x,
u(x)

λ

)
< +∞ for some λ > 0

}
.

Endowed with the so-called Luxemborg norm

‖u‖φ = inf
{
λ > 0 :

∫
Ω

φ
(
x,
u(x)

λ

)
dx ≤ 1

}
,

(Lφ(Ω), ‖ · ‖φ) is a Banach space. Observe that since limt→+∞ infx∈Ω
φ(x,t)
t = +∞

and if Ω has finite measure then we have the following continuous imbedding

Lφ(Ω) ↪→ L1(Ω). (2.3)

We will also use the space

Eφ(Ω) =
{
u : Ω→ R measurable :

∫
Ω

φ
(
x,
u(x)

λ

)
< +∞ for all λ > 0

}
.

Observe that for every u ∈ Lφ(Ω) the following inequality holds

‖u‖φ ≤
∫

Ω

φ(x, u(x)) dx+ 1. (2.4)

For two complementary Musielak-Orlicz functions φ and φ∗, Hölder’s inequality
(see [20]) ∫

Ω

|u(x)v(x)| dx ≤ 2‖u‖φ‖v‖φ∗ (2.5)

holds for every u ∈ Lφ(Ω) and v ∈ Lφ∗(Ω). Define φ∗−1 for every s ≥ 0 by

φ∗−1(x, s) = sup{τ ≥ 0 : φ∗(x, τ) ≤ s}.

Then, for almost every x ∈ Ω and for every s ∈ R we have

φ∗(x, φ∗−1(x, s)) ≤ s, (2.6)

s ≤ φ∗−1(x, s)φ−1(x, s) ≤ 2s, (2.7)



6 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

φ(x, s) ≤ s∂φ(x, s)

∂s
≤ φ(x, 2s). (2.8)

Definition 2.1. Let ~φ : Ω × R+ −→ RN be the vector function ~φ = (φ1, . . . , φN )
where for every i ∈ {1, . . . , N}, φi is a Musielak-Orlicz function. We define the
anisotropic Musielak-Orlicz-Sobolev space by

W 1L~φ(Ω) =
{
u ∈ Lφmax(Ω); ∂xiu ∈ Lφi(Ω) for all i = 1, · · ·, N

}
.

By the continuous imbedding (2.3), we obtain that W 1L~φ(Ω) is a Banach space

with respect to the following norm

‖u‖W 1L~φ(Ω) = ‖u‖φmax
+

N∑
i=1

‖∂xiu‖φi .

Moreover, we have the continuous embedding W 1L~φ(Ω) ↪→W 1,1(Ω).

3. Main results

In this section we prove an imbedding theorem and a trace result. Let us assume
the conditions∫ 1

0

(φ∗∗min)−1(x, t)

t1+ 1
N

dt < +∞ and

∫ +∞

1

(φ∗∗min)−1(x, t)

t1+ 1
N

dt = +∞, ∀x ∈ Ω.

(3.1)
Thus, we define the Sobolev conjugate (φ∗∗min)∗

(φ∗∗min)−1
∗ (x, s) =

∫ s

0

(φ∗∗min)−1(x, t)

t1+ 1
N

dt, for x ∈ Ω and s ∈ [0,+∞). (3.2)

It may readily be checked that (φ∗∗min)∗ is a Musielak-Orlicz function. We assume
that there exist two positive constants ν < 1

N and c0 > 0 such that∣∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣∣ ≤ c0[(φ∗∗min)∗(x, t) + ((φ∗∗min)∗(x, t))

1+ν
]
, (3.3)

for all t ∈ R and for almost every x ∈ Ω, provided that for every i = 1, . . . , N the

derivative
∂(φ∗∗min)∗
∂xi

(x, t) exists.

3.1. Imbedding theorem.

Theorem 3.1. Let Ω be an open bounded subset of RN , (N ≥ 2), with the cone
property. Assume that (3.1) and (3.3) are fulfilled, (φ∗∗min)∗(·, t) is Lipschitz contin-

uous on Ω and φmax is locally integrable. Then, there is a continuous embedding

W 1L~φ(Ω) ↪→ L(φ∗∗min)∗(Ω).

Some remarks about Theorem 3.1 are in order. We discuss how Theorem 3.1
include some previous results known in the literature when reducing to some par-
ticular Musielak-Orlicz functions.

Remark 3.2. (1) Let M(x, t) = tp(x) and m(x, t) = ∂M(x,t)
∂t = p(x)tp(x)−1, where

p(·) is Lipschitz continuous on Ω, with 1 < p− = infx∈Ω p(x) ≤ p(x) ≤ p+ =



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 7

supx∈Ω p(x) < N . Since M(·, t) and m(·, t) are continuous on Ω, we can use Lemma
5.8 (given in Appendix) to define the following Musielak-Orlicz function

φ(x, t) =


t
p(x)
1

tα1
tα if t ≤ t1,

tp(x) if t ≥ t1,

where t1 > 1 and α > 1 are two constants mentioned in the proof of Lemma 5.8.
Let us now consider the particular case where for all i = 1, . . . , N ,

φi(x, t) = φ(x, t) =


t
p(x)
1

tα1
tα if t ≤ t1,

tp(x) if t ≥ t1.
(3.4)

It is worth pointing out that since Ω is of finite Lebesgue measure, it can be seen
easily that W 1L~φ(Ω) = W 1Lφ(Ω) = W 1,p(·)(Ω). Thus, φ∗∗min(x, t) = φmin(x, t) =

φ(x, t) and

(φ∗∗min)∗(x, t) = (φmin)∗(x, t) = φ∗(x, t) =


( (N−α)t
Nαt1

) Nα
N−α t

Np(x)
N−α

1 if t ≤ t1,(
1

p∗(x) t
)p∗(x)

if t ≥ t1,

provided that α < N . Now we shall prove that (φ∗∗min)∗ satisfies (3.3) and our
imbedding result include some previous result known in the literature. For every
t ∈ R and for almost every x ∈ Ω we have

∂(φ∗∗min)∗
∂xi

(x, t) =

{
N

N−α
∂p(x)
∂xi

log(t1)(φ∗∗min)∗(x, t) if t ≤ t1,
∂p∗(x)
∂xi

log
(

t
ep∗(x)

)
(φ∗∗min)∗(x, t) if t ≥ t1.

• If t ≤ t1, then∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣ =

N

N − α
∣∣ ∂p
∂xi

(x)
∣∣ log(t1)(φ∗∗min)∗(x, t).

Since p(·) is Lipschitz continuous on Ω there exists a constant C1 > 0 satisfying∣∣ ∂p
∂xi

(x)
∣∣ ≤ C1 thus we obtain∣∣∂(φ∗∗min)∗

∂xi
(x, t)

∣∣ ≤ C1
N

N − α
log(t1)(φ∗∗min)∗(x, t). (3.5)

• If t ≥ t1, then∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣ =

∣∣∂p∗
∂xi

(x)
∣∣ ∣∣ log

( t

ep∗(x)

)∣∣(φ∗∗min)∗(x, t).

Since p(·) is Lipschitz continuous on Ω, it can be seen easily that p∗(·) is also Lip-

schitz continuous on Ω. Then, there exists a constant C2 > 0 satisfying
∣∣∣∂p∗∂xi

(x)
∣∣∣ ≤

C2. So that we have∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣ ≤ C2

∣∣ log
( t

ep∗(x)

)∣∣(φ∗∗min)∗(x, t).

Let 0 < ε < 1/N . For all t > 0 we can easily check that

log(t) ≤ 1

ε2Ne
tε. (3.6)



8 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

Now, since the Musielak-Orlicz function (φ∗∗min)∗ has a superlinear growth, we can
choose A > 0 for which there exists t0 > max{t1, e} (not depending on x) such that
At ≤ (φ∗∗min)∗(x, t) whenever t ≥ t0. Therefore,
• If t ≥ t0 then by (3.6) we obtain∣∣∂(φ∗∗min)∗

∂xi
(x, t)

∣∣
≤ C2

(
log
( t
e

)
+ log(

N2

N − p+
)
)

(φ∗∗min)∗(x, t)

≤ C2

ε2Ne1+ε
tε(φ∗∗min)∗(x, t) + C2 log(

N2

N − p+
)(φ∗∗min)∗(x, t)

≤ C2

ε2Ne1+εAε
((φ∗∗min)∗(x, t))

1+ε + C2 log(
N2

N − p+
)(φ∗∗min)∗(x, t).

(3.7)

• If t1 < t ≤ t0, then∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣ ≤ C2

(
log(t0) + log

( eN2

N − p+

))
(φ∗∗min)∗(x, t). (3.8)

Therefore, from (3.5), (3.7) and (3.8), we obtain that for every t ≥ 0 and for almost
every x ∈ Ω, there is a constant c0 > 0 such that∣∣∂(φ∗∗min)∗

∂xi
(x, t)

∣∣ ≤ c0((φ∗∗min)∗(x, t) + ((φ∗∗min)∗(x, t))
1+ε
)
.

Before we show that our imbedding result includes some previous known results in
the literature, we remark that the proof of Theorem 3.1 relies to the application
of Lemma 5.4 in Appendix for the function g(x, t) = ((φ∗∗min)∗(x, t))

α, α ∈ (0, 1),
where we have used the fact that Ω is bounded to ensure that maxx∈Ω g(x, t) <∞
for some t > 0. In the case of the variable exponent Sobolev space W 1,p(·)(Ω) built
upon the Musielak-Orlicz function given in (3.4), we do not need Ω to be bounded,
since

φ∗(x, t) ≤ max{t
Nα
N−α
1 , t

N2

N−p+ } <∞, for some t > 0.

Therefore, the embedding result in Theorem 3.1 can be seen as an extension to the
Musielak-Orlicz framework of the one obtained in [9, Theorem 1.1].

(2) Let us consider the particular case where, for i ∈ {1, . . . , N},

φi(x, t) =

 t
pi(x)
1

tα1
tα if t ≤ t1,

tpi(x) if t ≥ t1

where t1 > 1, 1 < α < N and ~φ = (φi)i∈{1,...,N} with

pi ∈ C+(Ω) = {h ∈ C(Ω) : inf
x∈Ω

h(x) > 1},

1 < pi(x) < N , N ≥ 3. We define p−i = infx∈Ω pi(x), pM (x) = maxi∈{1,...,N} pi(x),
pm(x) = mini∈{1,...,N} pi(x). Then

φ∗∗min(x, t) = φmin(x, t) =


t
pm(x)
1

tα1
tα if t ≤ t1,

tpm(x) if t ≥ t1,



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 9

whose Sobolev conjugate function is

(φ∗∗min)∗(x, t) =


( (N−α)t
Nαt1

) Nα
N−α t

Npm(x)
N−α

1 if t ≤ t1,(
1

(pm)∗(x) t
)(pm)∗(x)

if t ≥ t1.

Let us define p∗− = N∑N
i=1

1

p
−
i

−1
. Notice that p−i > p−m implies

p∗− >
Np−m
N − p−m

= (p−m)∗. (3.9)

Since Ω is of finite Lebesgue measure, it can be seen easily that W 1L~φ(Ω) =

W 1,~p(·)(Ω). So, by Theorem 3.1 we have W 1,~p(·)(Ω) ↪→ L(pm)∗(·)(Ω) and since

(pm)∗(x) ≥ (p−m)∗ for each x ∈ Ω, we deduce that W 1,~p(·)(Ω) ↪→ L(p−m)∗(Ω). There-
fore, by (3.9) the result we obtain can be found in [19, Theorem 1].

(3) Let us now consider the case where

φi(x, t) =

 t
pi(x)
1 log(t1+1)

tα1
tα if t ≤ t1,

tpi(x) log(t+ 1) if t ≥ t1,

where t1 > 1, 1 < α < N and for each i ∈ {1, . . . , N} the function pi(·) is Lip-
schitz continuous on Ω with 1 < infx∈Ω pi(x) ≤ pi(x) ≤ supx∈Ω pi(x) < N − 1.
Define pM (x) = maxi∈{1,...,N} pi(x), pm(x) = mini∈{1,...,N} pi(x) and φmin(x, t) =
mini∈{1,...,N} φi(x, t). Then

φmin(x, t) = φ∗∗min(x, t) =


t
pm(x)
1 log(t1+1)

tα1
tα if t ≤ t1,

tpm(x) log(t+ 1) if t ≥ t1.

Set A(x, t) = tpm(x) log(t+ 1). By [18, Example 2] there exist σ < 1
N , C0 > 0 and

t0 > 0 such that ∣∣∂A∗
∂xi

(x, t)
∣∣ ≤ C0(A∗(x, t))

1+σ,

for x ∈ Ω and t ≥ t0. Choosing this t0 > 0 in Lemma 5.8 given in Appendix, we
can take t1 > t0 + 1 obtaining∣∣∂A∗

∂xi
(x, t)

∣∣ ≤ C0(A∗(x, t))
1+σ, for all t ≥ t1. (3.10)

On the other hand, for t ≤ t1 we have

(φ∗∗min)∗(x, t) =
( (N − α)t

Nαt1

) Nα
N−α

( t
pm(x)
1

log(t1 + 1)

) N
N−α

.

Thus ∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣ =

N log(t1)

N − α
∣∣∂pm
∂xi

(x)
∣∣(φ∗∗min)∗(x, t).

Since pm(·) is Lipschitz continuous on Ω there exists a constant C3 > 0 satisfying∣∣∣∂pm∂xi (x)
∣∣∣ ≤ C3. So we have

∣∣∂(φ∗∗min)∗
∂xi

(x, t)
∣∣ ≤ C3N log(t1)

N − α
(φ∗∗min)∗(x, t). (3.11)



10 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

Therefore, by (3.10) and (3.11) the function (φ∗∗min)∗ satisfies the assertions of The-
orem 3.1 and then we obtain the continuous embedding

W 1L~φ(Ω) ↪→ L(φ∗∗min)∗(Ω).

Proof ofTheorem 3.1. Let u ∈ W 1L~φ(Ω). Assume first that the function u is

bounded and u 6= 0. Defining f(s) =
∫

Ω
(φ∗∗min)∗

(
x, |u(x)|

s

)
dx, for s > 0, one has

lims→0+ f(s) = +∞ and lims→∞ f(s) = 0. Since f is continuous on (0,+∞), there
exists λ > 0 such that f(λ) = 1. Then by the definition of the Luxemburg norm,
we obtain

‖u‖(φ∗∗min)∗ ≤ λ. (3.12)

On the other hand,

f(‖u‖(φ∗∗min)∗) =

∫
Ω

(φ∗∗min)∗

(
x,

u(x)

‖u‖(φ∗∗min)∗

)
dx ≤ 1 = f(λ)

and since f is decreasing,

λ ≤ ‖u‖(φ∗∗min)∗ . (3.13)

So that by (3.12) and (3.13), we obtain λ = ‖u‖(φ∗∗min)∗ and∫
Ω

(φ∗∗min)∗

(
x,
u(x)

λ

)
dx = 1. (3.14)

From (3.2) we can easily check that (φ∗∗min)∗ satisfies the differential equation

(φ∗∗min)−1(x, (φ∗∗min)∗(x, t))
∂(φ∗∗min)∗

∂t
(x, t) = ((φ∗∗min)∗(x, t))

N+1
N .

Hence, by (2.7) we obtain the inequality

∂(φ∗∗min)∗
∂t

(x, t) ≤ ((φ∗∗min)∗(x, t))
1
N (φ∗∗min)∗−1(x, (φ∗∗min)∗(x, t)), (3.15)

for a.e. x ∈ Ω. Let

h(x) =
[
(φ∗∗min)∗

(
x,
u(x)

λ

)]N−1
N

. (3.16)

Since (φ∗∗min)∗(·, t) is Lipschitz continuous on Ω and (φ∗∗min)∗(x, ·) is locally Lipschitz

continuous on R+, the function h is Lipschitz continuous on Ω. Hence, we can
compute using Lemma 5.6 (given in Appendix) for f = h, obtaining for a.e. x ∈ Ω,

∂h

∂xi
(x) =

N − 1

N

(
(φ∗∗min)∗

(
x,
u(x)

λ

))− 1
N
[∂(φ∗∗min)∗

∂t

(
x,
u(x)

λ

)∂xiu
λ

(x)

+
∂(φ∗∗min)∗
∂xi

(
x,
u(x)

λ

)]
,

where ∂xiu := ∂u
∂xi

. Therefore,

N∑
i=1

∣∣∣ ∂h
∂xi

(x)
∣∣∣ ≤ I1 + I2, for a.e. x ∈ Ω, (3.17)

where we have set

I1 =
N − 1

Nλ

(
(φ∗∗min)∗

(
x,
u(x)

λ

))−1
N ∂(φ∗∗min)∗

∂t

(
x,
u(x)

λ

) N∑
i=1

|∂xiu(x)|,



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 11

I2 =
N − 1

N

(
(φ∗∗min)∗

(
x,
u(x)

λ

))−1
N

N∑
i=1

∣∣∣∂(φ∗∗min)∗
∂xi

(
x,
u(x)

λ

)∣∣∣.
Now we estimate the two integrals

∫
Ω
I1(x) dx and

∫
Ω
I2(x) dx. By (3.15), we

can write

I1(x) ≤ N − 1

Nλ
(φ∗∗min)∗−1

(
x, (φ∗∗min)∗

(
x,
u(x)

λ

)) N∑
i=1

|∂xiu(x)|. (3.18)

By (2.6), we have∫
Ω

(φ∗∗min)∗
(
x, (φ∗∗min)∗−1

(
x, (φ∗∗min)∗

(
x,
u(x)

λ

)))
dx ≤

∫
Ω

(φ∗∗min)∗

(
x,
u(x)

λ

)
dx = 1.

Hence ∥∥(φ∗∗min)∗−1
(
·, (φ∗∗min)∗

(
·, u(·)

λ

))∥∥
(φ∗∗min)∗

≤ 1. (3.19)

From (2.5), (3.18) and (3.19) it follows that∫
Ω

I1(x) dx

≤ 2(N − 1)

Nλ

∥∥(φ∗∗min)∗−1
(
·, (φ∗∗min)∗

(
·, u(·)

λ

))∥∥
(φ∗∗min)∗

N∑
i=1

∥∥∂xiu∥∥φ∗∗min

≤ 2(N − 1)

Nλ

N∑
i=1

∥∥∂xiu∥∥φ∗∗min

≤ 2

λ

N∑
i=1

∥∥∂xiu∥∥φ∗∗min

.

(3.20)

Recalling the definition of φmin and (2.2), we obtain ‖∂xiu(x)‖φ∗∗min
≤ ‖∂xiu(x)‖φi ,

so that (3.20) implies ∫
Ω

I1(x) dx ≤ 2

λ

N∑
i=1

∥∥∂xiu(x)
∥∥
φi
. (3.21)

Using (3.3) we can write

I2(x) ≤ c1
[(

(φ∗∗min)∗

(
x,
u(x)

λ

))1− 1
N

+
(

(φ∗∗min)∗

(
x,
u(x)

λ

))1− 1
N +ν]

,

with c1 = c0(N−1). Since (φ∗∗min)∗(·, t) is continuous on Ω and ν < 1
N , we can apply

Lemma 5.4 (given in Appendix) with the functions g(x, t) =
((φ∗∗min)∗(x,t))

1− 1
N

+ν

t and

f(x, t) =
(φ∗∗min)∗(x,t)

t and ε = 1
8c1c∗

obtaining for t = |u(x)|
λ[

(φ∗∗min)∗

(
x,
u(x)

λ

)]1− 1
N +ν

≤ 1

8c1c∗
(φ∗∗min)∗

(
x,
u(x)

λ

)
+K0

|u(x)|
λ

. (3.22)

Using again Lemma 5.4 with the functions g(x, t) =
((φ∗∗min)∗(x,t))

1− 1
N

t and f(x, t) =
(φ∗∗min)∗(x,t)

t and ε = 1
8c1c∗

, we obtain by substituting t by |u(x)|
λ[

(φ∗∗min)∗

(
x,
u(x)

λ

)]1− 1
N ≤ 1

8c1c∗
(φ∗∗min)∗

(
x,
u(x)

λ

)
+K0

|u(x)|
λ

, (3.23)



12 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

where c∗ is the constant in the continuous embedding W 1,1(Ω) ↪→ L
N
N−1 (Ω), that

is

‖w‖
L

N
N−1 (Ω)

≤ c∗‖w‖W 1.1(Ω), for all w ∈W 1,1(Ω). (3.24)

By (3.22) and (3.23), we obtain∫
Ω

I2(x) dx ≤ 1

4c∗
+

2K0c1
λ
‖u‖L1(Ω). (3.25)

Putting together (3.21) and (3.25) in (3.17) we obtain

N∑
i=1

‖∂xih‖L1(Ω) ≤
1

4c∗
+

2

λ

N∑
i=1

‖∂xiu(x)‖φi +
2K0c1
λ
‖u‖L1(Ω)

≤ 1

4c∗
+

2

λ

N∑
i=1

‖∂xiu(x)‖φi +
2K0c1c2

λ
‖u‖φmax

,

where c2 is the constant in the continuous embedding (2.3). Then it follows that

N∑
i=1

‖∂xih‖L1(Ω) ≤
1

4c∗
+
c3
λ
‖u‖W 1L~φ(Ω), (3.26)

with c3 = max{2, 2K0c1c2}. Now, using again Lemma 5.4 (in Appendix) with the

functions g(x, t) =
[
(φ∗∗min)∗(x, t)

]1− 1
N /t and f(x, t) = (φ∗∗min)∗(x, t)/t and ε = 1

4c∗
,

for t = |u(x)|/λ, we obtain

h(x) ≤ 1

4c∗
(φ∗∗min)∗

(
x,
u(x)

λ

)
+K0

|u(x)|
λ

,

From (2.3), we obtain

‖h‖L1(Ω) ≤
1

4c∗
+
K0c2
λ
‖u‖Lφmax (Ω). (3.27)

Thus, by (3.26) and (3.27) we obtain

‖h‖W 1,1(Ω) ≤
1

2c∗
+
c4
λ
‖u‖W 1L~φ(Ω),

where c4 = c3 + K0c2, which shows that h ∈ W 1,1(Ω) and which together with
(3.24) yield

‖h‖
L

N
N−1 (Ω)

≤ 1

2
+
c4c∗
λ
‖u‖W 1L~φ(Ω).

Having in mind (3.14), we obtain∫
Ω

[h(x)]
N
N−1 dx =

∫
Ω

(φ∗∗min)∗

(
x,
u(x)

λ

)
dx = 1.

So that one has

‖u‖(φ∗∗min)∗ = λ ≤ 2c4c∗‖u‖W 1L~Φ(Ω). (3.28)

We now extend the estimate (3.28) to an arbitrary u ∈W 1L~φ(Ω). Let Tn, n > 0, be

the truncation function at levels ±n defined on R by Tn(s) = min{n,max{s,−n}}.
Since φmax is locally integrable, by [1, Lemma 8.34.] one has Tn(u) ∈ W 1L~φ(Ω).

So that in view of (3.28)

‖Tn(u)‖(φ∗∗min)∗ ≤ 2c4c∗‖Tn(u)‖W 1L~φ(Ω)
≤ 2c4c∗‖u‖W 1L~φ(Ω)

. (3.29)



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 13

Let kn = ‖Tn(u)‖(φ∗∗min)∗ . Thanks to (3.29), the sequence {kn}∞n=1 is nondecreasing
and converges. If we denote k = limn→∞ kn, by Fatou’s lemma we have∫

Ω

(φ∗∗min)∗

(
x,
|u(x)|
k

)
dx ≤ lim inf

∫
Ω

(φ∗∗min)∗

(
x,
|Tn(u)|
kn

)
dx ≤ 1.

This implies that u ∈ L(φ∗∗min)∗(Ω) and

‖u‖(φ∗∗min)∗ ≤ k = lim
n→∞

‖Tn(u)‖(φ∗∗min)∗ ≤ 2c4c∗‖u‖W 1L~φ(Ω)
.

Inequality (3.28) trivially holds if u = 0. Then proof is complete. �

Corollary 3.3. Let Ω be an open bounded subset of RN , N ≥ 2, with the cone
property. Assume that (3.1), (3.3) are fulfilled, (φ∗∗min)∗(·, t) is Lipschitz continuous

on Ω and φmax is locally integrable. Let A be a Musielak-Orlicz function where
the function A(·, t) is continuous on Ω and A � (φ∗∗min)∗. Then, the embedding
W 1L~φ(Ω) ↪→ LA(Ω) is compact.

Proof. Let {un} is a bounded sequence in W 1L~φ(Ω). By Theorem 3.1, {un} is

bounded in L(φ∗∗min)∗(Ω). Since the embedding W 1L~φ(Ω) ↪→ W 1,1(Ω) is continuous

and the imbedding W 1,1(Ω) ↪→ L1(Ω) is compact, we deduce that there exists a
subsequence of {un} still denoted by {un} which converges in measure in Ω. Since
A � (φ∗∗min)∗, by Lemma 5.5 (in Appendix) the sequence {un} converges in norm
in LA(Ω). �

3.2. Trace result. We prove here a trace result which is a useful tool to prove the

coercivity of some energy functionals. Recall that ψmin(x, t) = [(φ∗∗min)∗(x, t)]
N−1
N is

a Musielak-Orlicz function. Indeed, we have

∂

∂t
(ψmin)−1(x, t) =

∂

∂t
(φ∗∗min)−1

∗
(
x, t

N
N−1

)
.

By (3.2), we obtain

∂

∂t
(ψmin)−1(x, t) =

N

N − 1
t

1
N−1

(φ∗∗min)−1
(
x, t

N
N−1

)
t
N
N−1 + 1

N−1

=
N

N − 1

(φ∗∗min)−1
(
x, t

N
N−1

)
t
N
N−1

.

Being the inverse of a Musielak-Orlicz function, it is clear that (φ∗∗min)−1 satisfies

lim
τ→+∞

(φ∗∗min)−1(x, τ)

τ
= 0 and lim

τ→0+

(φ∗∗min)−1(x, τ)

τ
= +∞.

Moreover, (φ∗∗min)−1(x, ·) is concave so that if 0 < τ < σ then we obtain

(φ∗∗min)−1(x, τ)

(φ∗∗min)−1(x, σ)
≥ τ

σ
.

Hence, if 0 < s1 < s2, then

∂
∂t (ψmin)−1(x, s1)
∂
∂t (ψmin)−1(x, s2)

=
(φ∗∗min)−1

(
x, s

N
N−1

1

)
(φ∗∗min)−1

(
x, s

N
N−1

2

) s
N
N−1

2

s
N
N−1

1

≥ s
N
N−1

1

s
N
N−1

2

s
N
N−1

2

s
N
N−1

1

= 1.

It follows that ∂
∂t (ψmin)−1(x, t) is positive and decreases monotonically from +∞

to 0 as t increases from 0 to +∞ and thus ψmin is a Musielak-Orlicz function.



14 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

Theorem 3.4. Let Ω be an open bounded subset of RN , N ≥ 2, with the cone
property. Assume that (3.1), (3.3) are fulfilled, (φ∗∗min)∗(·, t) is Lipschitz continuous

on Ω and φmax is locally integrable. Let ψmin the Musielak-Orlicz function defined
in (1.8). Then, the following boundary trace embedding W 1L~φ(Ω) ↪→ Lψmin(∂Ω) is
continuous.

Remark 3.5. In the case where for all i = 1, . . . , N ,

φi(x, t) = φ(x, t) =


t
p(x)
1

tα1
tα if t ≤ t1,

tp(x) if t ≥ t1,

for some t1 > 0, with p ∈ L∞(Ω), 1 ≤ infx∈Ω p(x) ≤ supx∈Ω p(x) < N , |∇p| ∈
Lγ(·)(Ω), where γ ∈ L∞(Ω) and infx∈Ω γ(x) > N . It is worth pointing out that since
Ω is of finite Lebesgue measure, it can be seen easily that W 1L~φ(Ω) = W 1Lφ(Ω) =

W 1,p(·)(Ω). Then φ∗∗min(x, s) = φmin(x, t) = φ(x, t) and so

(φ∗∗min)∗(x, t) = (φmin)∗(x, t) = φ∗(x, t) =


( (N−α)t
Nαt1

) Nα
N−α t

Np(x)
N−α

1 if t ≤ t1,(
1

p∗(x) t
)p∗(x)

if t ≥ t1.

As above we can prove that (φ∗∗min)∗ satisfies the conditions of Theorem 3.4 and then
our trace result is an extension to Musielak-Orlicz framework of the one proved by
Fan in [8].

Proof of Theorem 3.4. Let u ∈ W 1L~φ(Ω). Because of the continuous embedding

W 1L~φ(Ω) ↪→ L(φ∗∗min)∗(Ω), the function u belongs to L(φ∗∗min)∗(Ω) and then u belongs

to Lψmin
(Ω). Clearly W 1L~φ(Ω) ↪→ W 1,1(Ω) and by the Gagliardo trace theorem

(see [10]) we have the embedding W 1,1(Ω) ↪→ L1(∂Ω). Hence, we conclude that for
all u ∈ W 1L~φ(Ω) there holds u|∂Ω ∈ L1(∂Ω). Therefore, for every u ∈ W 1L~φ(Ω)

the trace u|∂Ω is well defined. Assume first that u is bounded and u 6= 0. Since
(φ∗∗min)∗(·, t) is continuous on ∂Ω, the function u belongs to Lψmin(∂Ω). Let

k = ‖u‖Lψmin
(∂Ω) = inf

{
λ > 0 :

∫
∂Ω

ψmin

(
x,
u(x)

λ

)
dx ≤ 1

}
.

We distinguish the two cases: k ≥ ‖u‖L(φ∗∗
min

)∗ (Ω) and k < ‖u‖L(φ∗∗
min

)∗ (Ω).

Case 1: Assume that

k ≥ ‖u‖L(φ∗∗
min

)∗ (Ω). (3.30)

Going back to (3.16) we can repeat exactly the same lines with l(x) = ψmin

(
x, u(x)

k

)
instead of the function h, obtaining

‖l‖W 1,1(Ω) ≤
( 1

4c
+
c3
k
‖u‖W 1L~φ(Ω) + ‖l‖L1(Ω)

)
, (3.31)

where c is the constant in the imbedding W 1,1(Ω) ↪→ L1(∂Ω), that is

‖w‖L1(∂Ω) ≤ c‖w‖W 1,1(Ω), for all w ∈W 1,1(Ω). (3.32)

Since (φ∗∗min)∗(·, t) is continuous on Ω, using Lemma 5.4 (in Appendix) with the

functions f(x, t) =
(φ∗∗min)∗(x,t)

t and g(x, t) = l(x)
t and ε = 1

4c , we obtain for t = |u(x)|
k

l(x) ≤ 1

4c
(φ∗∗min)∗

(
x,
u(x)

k

)
+K0

|u(x)|
k

. (3.33)



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 15

By (3.30), we have∫
Ω

(φ∗∗min)∗
(
x,
u(x)

k

)
dx ≤

∫
Ω

(φ∗∗min)∗

(
x,

u(x)

‖u‖(φ∗∗min)∗

)
dx ≤ 1.

Integrating (3.33) over Ω, we obtain

‖l‖L1(Ω) ≤
1

4c
+
K0c2
k
‖u(x)‖Lφmax (Ω) ≤

1

4c
+
K0c2
k
‖u‖W 1L~φ(Ω), (3.34)

where c2 is the constant of the imbedding (2.3). Thus, by (3.31) and (3.34) we
obtain

‖l‖W 1,1(Ω) ≤
1

2c
+
C4

k
‖u‖W 1L~φ(Ω),

where C4 = c2K0 + c3. This implies that l ∈W 1,1(Ω) and by (3.32) we arrive at

‖l‖L1(∂Ω) ≤
1

2
+
cC4

k
‖u‖W 1L~φ(Ω).

As

‖l‖L1(∂Ω) =

∫
∂Ω

|l(x)| dx =

∫
∂Ω

ψmin

(
x,
u(x)

k

)
dx = 1,

we obtain ‖u‖Lψmin
(∂Ω) = k ≤ 2cC4‖u‖W 1L~φ(Ω).

Case 2: Assume that k < ‖u‖(φ∗∗min)∗ . By Theorem 3.1, there is a constant c > 0
such that

‖u‖Lψmin
(∂Ω) = k < ‖u‖(φ∗∗min)∗ ≤ c‖u‖W 1L~φ(Ω).

Finally, in both cases there exists a constant c > 0 such that

‖u‖Lψmin
(∂Ω) ≤ c‖u‖W 1L~φ(Ω).

For an arbitrary u ∈ W 1L~φ(Ω), we proceed as in the proof of Theorem 3.1 by

truncating the function u. �

4. Application to anisotropic elliptic equations

In this section, we apply the above results to obtain the existence and uniqueness
of the weak solution for the problem (1.1).

4.1. Properties of the energy functional.

Definition 4.1. Let Ω be an open bounded subset of RN , N ≥ 2. By a weak
solution of problem (1.1), we mean a function u ∈ W 1L~φ(Ω) satisfying for all

v ∈ C∞(Ω) the identity∫
Ω

N∑
i=1

ai(x, ∂xiu)∂xiv dx+

∫
Ω

b(x)ϕmax(x, u)v dx

−
∫

Ω

f(x, u)v dx−
∫
∂Ω

g(x, u)v ds = 0.

(4.1)

We note that all the terms in (4.1) make sense. Indeed, for the first term in the
right hand side in (4.1), we can write by using (2.8)∫

Ω

φ∗i (x, φ
∗−1
i (x, Pi(x, ∂xiu(x)))) dx ≤

∫
Ω

Pi(x, ∂xiu(x)) dx

≤
∫

Ω

pi(x, ∂xiu(x))∂xiu(x) dx,



16 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

where Pi is the Musielak-Orlicz function given in (1.2) and pi(x, s) = ∂Pi
∂s (x, s).

Since Pi is locally integrable and Pi � φi, we can use Lemma 5.7 (see the Appendix)
obtaining pi(·, ∂xiu(·)) ∈ LP∗i (Ω). So that by Hölder’s inequality (2.5), we obtain∫

Ω

φ∗i (x, φ
∗−1
i (x, Pi(x, ∂xiu(x)))) dx ≤ 2‖pi(·, ∂xiu(·))‖P∗i ‖∂xiu‖Pi <∞.

Thus, φ∗−1
i (·, Pi(·, ∂xiu(·))) ∈ Lφ∗i (Ω). Since v ∈ C∞(Ω) and φmax is locally inte-

grable, then v ∈ W 1L~φ(Ω). So we can use the growth condition (1.2) and again

the Hölder inequality (2.5) to write∫
Ω

ai(x, ∂xiu)∂xiv dx

≤ 2ci‖di(·)‖φ∗i ‖∂xiv‖φi + 2ci‖φ∗−1
i (·, Pi(·, ∂xiu(·)))‖φ∗i ‖∂xiv‖φi <∞.

(4.2)

For the second term, the inequality (2.8) enables us to write∫
Ω

φ∗max(x, φ∗−1
max(x,R(x, u(x)))) dx ≤

∫
Ω

R(x, u(x)) dx ≤
∫

Ω

r(x, u(x))u(x) dx,

where R is the Musielak-Orlicz function given in (1.5) and r(x, s) = ∂R(x,s)
∂s . Since

R is locally integrable and R� φmax, Lemma 5.7 (in Appendix) gives∫
Ω

φ∗max(x, φ∗−1
max(x,R(x, ∂iu))) dx ≤ 2‖r(·, u(·))‖R∗‖u‖R <∞,

which shows that ϕmax(·, u(·)) ∈ Lφ∗max
(Ω). Thus,∫

Ω

b(x)ϕmax(x, u)v dx ≤ 2‖b‖∞‖ϕmax(·, u(·))‖φ∗max
‖v‖φmax

<∞. (4.3)

We now turn to the third term in the right hand side in (4.1). By using (1.6) and
the Hölder inequality (2.5), one has∣∣ ∫

Ω

f(x, u)v dx
∣∣ ≤ k1‖m(·, u(·))‖LM∗ (Ω)‖v‖LM (Ω). (4.4)

Since M is locally integrable and M � φ∗∗min, then M � φmax and Lemma 5.7
ensures that |

∫
Ω
f(x, u)v dx| <∞. For the last term in the right hand side in (4.1),

using (1.7) to have ∣∣ ∫
∂Ω

g(x, u)v ds
∣∣ ≤ k2

∫
∂Ω

|h(x, u)v| ds.

Since the primitive H of h is a locally integrable function satisfying H � φ∗∗min we
can use a similar way as in Lemma 5.7 to obtain h(·, u) ∈ LH∗(∂Ω) and since ∂Ω
is a bounded set, the imbedding (2.3) gives h(x, u) ∈ L1(∂Ω). On the other hand,
since v ∈ C∞(Ω) one has v ∈ L∞(∂Ω). Therefore,

|
∫
∂Ω

g(x, u)v ds| <∞.

Define the functional I : W 1L~φ(Ω)→ R by

I(u) =

∫
Ω

N∑
i=1

Ai(x, ∂iu) dx+

∫
Ω

b(x)φmax(x, u) dx

−
∫

Ω

F (x, u) dx−
∫
∂Ω

G(x, u)ds.

(4.5)



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 17

Some basic properties of I are established in the following lemma.

Lemma 4.2. Let Ω be an open bounded subset of RN , N ≥ 2. Then

(i) The functional I is well defined on W 1L~φ(Ω).

(ii) The functional I has a Gâteaux derivative I ′(u) for every u ∈ W 1L~φ(Ω).

Moreover, for every v ∈W 1L~φ(Ω)

〈I ′(u), v〉 =

∫
Ω

N∑
i=1

ai(x, ∂iu)∂iv dx+

∫
Ω

b(x)ϕmax(x, u)v dx

−
∫

Ω

f(x, u)v dx−
∫
∂Ω

g(x, u)v ds.

So that, the critical points of I are weak solutions to the problem (1.1).

Proof. (i) For almost every x ∈ Ω and for every ζ ∈ R, we can write

Ai(x, ζ) =

∫ 1

0

d

dt
Ai(x, tζ)dt =

∫ 1

0

ai(x, tζ)ζdt.

Then, by (1.2) we obtain

|Ai(x, ζ)| ≤ cidi(x)|ζ|+
∫ 1

0

φ∗−1
i (x, Pi(x, tζ))|ζ|dt ≤ cidi(x)|ζ|+φ∗−1

i (x, Pi(x, ζ))|ζ|.

In a similar manner as in (4.2), we arrive at∣∣ ∫
Ω

Ai(x, ∂iu(x)) dx
∣∣ <∞.

Hence, the first term in the right hand side in (4.5) is well defined. For the second
term, we using (2.8), the Hölder inequality (2.5) and (4.3) obtaining∣∣ ∫

Ω

b(x)φmax(x, u(x)) dx
∣∣ ≤ 2‖b‖∞‖ϕmax(·, u(·))‖φ∗max

‖u‖φmax
<∞,

while for the third term we can estimate it using (1.6) and (2.8) as follows∫
Ω

∣∣F (x, u(x))
∣∣ dx ≤ k1

∫
Ω

|m(x, u)u| dx.

So that by Hölder’s inequality (2.5) we obtain∫
Ω

∣∣F (x, u(x))
∣∣ dx ≤ 2k1‖m(·, u(·))‖M∗‖u‖M <∞.

Regarding the last term in the right hand side in (4.5), we can use (1.7) and (2.8)
to have ∫

∂Ω

|G(x, u)| ds ≤ k2

∫
∂Ω

h(x, u)uds.

Since the function H is locally integrable and satisfies H � φ∗∗min, it follows that
H � φmax and in a similar manner as in Lemma 5.7 (in Appendix) we obtain
h(·, u) ∈ LH∗(∂Ω). Applying the Hölder inequality (2.5) one has∫

∂Ω

|G(x, u)| ds <∞.

(ii) For i = 1, . . . , N we define the functional Λi : W 1L~φ(Ω)→ R by

Λi(u) =

∫
Ω

Ai(x, ∂iu(x)) dx.



18 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

We denote by B, L1, L2 : W 1L~φ(Ω)→ R the functionals

B(u) =

∫
Ω

b(x)φmax(x, u(x)) dx, L1(u) =

∫
Ω

F (x, u(x)) dx,

L2(u) =

∫
∂Ω

G(x, u(x))ds.

Observe that for u ∈W 1L~φ(Ω), v ∈ C∞(Ω), and r > 0, we have

Λi(u+ rv)− Λi(u)

r
=

∫
Ω

1

r

(
Ai

(
x,
∂u

∂xi
(x) + r

∂v

∂xi
(x)
)
−Ai

(
x,
∂u

∂xi
(x)
))

dx,

1

r

(
Ai

(
x,
∂u

∂xi
(x) + r

∂v

∂xi
(x)
)
−Ai

(
x,
∂u

∂xi
(x)
))
−→ ai

(
x,
∂u

∂xi
(x)
) ∂v
∂xi

(x),

as r → 0 for almost every x ∈ Ω. On the other hand, by the mean value theorem
there exists ν ∈ [0, 1] such that

1

r

∣∣Ai(x, ∂u
∂xi

(x) + r
∂v

∂xi
(x)
)
−Ai

(
x,
∂u

∂xi
(x)
)∣∣

=
∣∣∣ai(x, ∂u

∂xi
(x) + νr

∂v

∂xi
(x)
)∣∣∣| ∂v

∂xi
(x)|.

Hence, by using this equality and (1.2) we obtain

1

r

∣∣∣Ai(x, ∂u
∂xi

(x) + r
∂v

∂xi
(x)
)
−Ai

(
x,
∂u

∂xi
(x)
)∣∣∣

≤ ci
(
di(x) + φ∗−1

i

(
x, φi

(
x,
∂u

∂xi
(x) + νr

∂v

∂xi
(x)
)))∣∣∣ ∂v

∂xi
(x)
∣∣∣.

Next, by Hölder’s inequality (2.5) we obtain

ci

(
di(x) + φ∗−1

i

(
x, φi

(
x,

∂

∂xi
u(x) + νr

∂

∂xi
v(x)

)))∣∣∣ ∂
∂xi

v(x)
∣∣∣ ∈ L1(Ω).

The dominated convergence theorem can be applied to obtain

lim
r→0

Λi(u+ rv)− Λi(u)

r
= 〈Λ′i(u), v〉 :=

∫
Ω

ai

(
x,
∂u

∂xi
(x)
) ∂v
∂xi

(x) dx,

for i = 1, . . . , N . By a similar calculus as in above, we can show that

〈B′(u), v〉 =

∫
Ω

b(x)ϕmax(x, u)v dx, 〈L′1(u), v〉 =

∫
Ω

f(x, u)v dx,

〈L′2(u), v〉 =

∫
Ω

g(x, u)v dx.

�

4.2. Existence of solutions. Our main existence result reads as follows.

Theorem 4.3. Let Ω be an open bounded subset of RN , N ≥ 2, with the cone
property. Assume that (1.2), (1.3), (1.4), (1.6), (1.7), (1.9), (3.1) and (3.3) are
fulfilled and suppose that φmax and φ∗min are locally integrable and (φ∗∗min)∗(·, t) is

Lipschitz continuous on Ω. Then, problem (1.1) admits at least a weak solution in
W 1L~φ(Ω).



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 19

Proof. We divide the proof into three steps.
Step 1: Weak∗ lower semicontinuity property of I. We define the functional J :
W 1L~φ(Ω)→ R by

J(u) =

∫
Ω

N∑
i=1

Ai(x, ∂iu) dx+

∫
Ω

b(x)φmax(x, u) dx,

so that I(u) = J(u)− L1(u)− L2(u), where

L1(u) =

∫
Ω

F (x, u) dx, L2(u) =

∫
∂Ω

G(x, u)ds.

First, we claim that J is sequentially weakly lower semicontinuous. Indeed, since
u 7→ φmax(x, u) is continuous, it is sufficient to show that the functional

u 7→ K(u) =

∫
Ω

N∑
i=1

Ai(x, ∂iu) dx,

is sequentially weakly−∗ lower semicontinuous. To do this, let un
∗
⇀ u in W 1L~φ(Ω)

in the sense that ∫
Ω

unϕdx→
∫

Ω

uϕdx, for all ϕ ∈ Eφ∗max
, (4.6)∫

Ω

∂iunϕdx→
∫

Ω

∂iuϕdx, for all ϕ ∈ Eφ∗i . (4.7)

By the definitions of φmin and φmax, (4.6) and (4.7) hold for every ϕ ∈ Eφ∗min
(Ω).

Being φ∗min locally integrable, one has L∞(Ω) ⊂ Eφ∗min
(Ω). Therefore, for every

i ∈ {1, . . . , N}
∂iun ⇀ ∂iu and un ⇀ u in L1(Ω), (4.8)

for the weak topology σ(L1, L∞). Since the embedding W 1L~φ(Ω) ↪→ W 1,1(Ω) is

continuous and the embedding W 1,1(Ω) ↪→ L1(Ω) is compact, we conclude that the
sequence {un} is relatively compact in L1(Ω). Therefore, there exist a subsequence
still indexed by n and a function v ∈ L1(Ω) such that un → v strongly in L1(Ω).
In view of (4.8), we have v = u almost everywhere on Ω and un → u in L1(Ω).
Passing once more to a subsequence, we can have un → u almost everywhere on Ω.
Recall that ζ → Ai(x, ζ) is a convex function, so by (1.3) we can use [5, Theorem
2.1, Chapter 8] obtaining

K(u) =

∫
Ω

N∑
i=1

Ai(x, ∂iu) dx ≤ lim inf

∫
Ω

N∑
i=1

Ai(x, ∂iun) dx = lim inf K(un).

We shall now prove that L1 and L2 are continuous. Since M � φ∗∗min, it follows
that M � (φ∗∗min)∗, then by Corollary 3.3 we have un → u in LM (Ω). Thus, there
exists n0 such that for every n ≥ n0, ‖un − u‖M < 1

2 . By (1.6), we obtain∫
Ω

|F (x, un(x))| dx ≤ k1

∫
Ω

M(x, un(x)) dx.

Let θn = ‖un − u‖M . By the convexity of M , we can write

M(x, un(x)) = M
(
x, θn

(un(x)− u(x)

θn

)
+ (1− θn)

u(x)

1− θn

)



20 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

≤ θnM
(
x,
un(x)− u(x)

θn

)
+ (1− θn)M

(
x,

u(x)

1− θn

)
.

Hence, ∫
Ω

M(x, un(x)) dx ≤ θn + (1− θn)

∫
Ω

M
(
x,

u(x)

1− θn

)
dx. (4.9)

Moreover,

M
(
x,

u(x)

1− θn

)
≤M(x, 2u(x)).

Since M is locally integrable and M � φmax, there exists a nonnegative function
h ∈ L1(Ω) such that∫

Ω

M(x, 2|u(x)|) dx ≤
∫

Ω

φmax

(
x,
|u(x)|
‖u‖φmax

)
dx+

∫
Ω

h(x) dx <∞.

Thus, the Lebesgue dominated convergence theorem yields

lim
n→∞

∫
Ω

M
(
x,

u(x)

1− θn

)
dx =

∫
Ω

M(x, u(x)) dx

and therefore by (4.9) we have

lim sup
n→∞

∫
Ω

M(x, un(x)) dx ≤
∫

Ω

M(x, u(x)) dx.

In addition, by Fatou’s Lemma we obtain∫
Ω

M(x, u(x)) dx ≤ lim inf
n→∞

∫
Ω

M(x, un(x)) dx.

Gathering the two inequalities above, we obtain

lim
n→+∞

∫
Ω

M(x, un(x)) dx =

∫
Ω

M(x, u(x)) dx.

Applying [14, Theorem 13.47] we obtain M(x, un(x)) → M(x, u(x)) strongly in
L1(Ω) which in turn implies that M(x, un(x)) is equi-integrable and then so is
F (x, un(x)). Since F (x, un)→ F (x, u) almost everywhere on Ω by Vitali’s theorem
we have L1(un)→ L1(u). Similarly, we can show that L2(un)→ L2(u). That is to
say that L1 and L2 are continuous. Since J is weakly−∗ lower semicontinuous, we
conclude that I is weakly-∗ lower semicontinuous.

Step 2: Coercivity of the functional I. By (1.3), (1.9) and (2.4) we can write

I(u) ≥
∫

Ω

N∑
i=1

φi(x, ∂iu) dx+ b0

∫
Ω

φmax(x, u) dx−
∫

Ω

F (x, u) dx−
∫
∂Ω

G(x, u)ds

≥
N∑
i=1

‖∂iu‖φi + b0‖u‖φmax
−N − b0 −

∫
Ω

F (x, u) dx−
∫
∂Ω

G(x, u)ds

≥ min{1, b0}‖u‖W 1L~φ(Ω) −
∫

Ω

F (x, u) dx−
∫
∂Ω

G(x, u)ds−N − b0.

By (1.6) and (1.7) we obtain

I(u) ≥ min{1, b0}‖u‖W 1L~φ(Ω) − k1

∫
Ω

M(x, u) dx− k2

∫
∂Ω

H(x, u)ds−N − b0.

As M � (φ∗∗min)∗ and H � ψmin, by Theorems 3.1 and 3.4 there exist two positive
constant C1 > 0 and C2 > 0 such that ‖u‖LM (Ω) ≤ C1‖u‖W 1L~φ(Ω) and ‖u‖LH(∂Ω) ≤



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 21

C2‖u‖W 1L~φ(Ω). Since M and H satisfy the ∆2-condition, there exist two positive

constants r1 > 0 and r2 > 0 and two nonnegative functions h1 ∈ L1(Ω) and
h2 ∈ L1(∂Ω) such that

I(u) ≥ min{1, b0}‖u‖W 1L~φ(Ω) − k1r1

∫
Ω

M
(
x,

|u(x)|
C1‖u‖W 1L~φ(Ω)

)
dx

− k2r2

∫
∂Ω

H
(
x,

|u(x)|
C2‖u‖W 1L~φ(Ω)

)
ds−

∫
Ω

h1(x) dx−
∫
∂Ω

h2(x)ds−N − b0

≥ min{1, b0}‖u‖W 1L~φ(Ω) − k1r1

∫
Ω

M
(
x,
|u(x)|
‖u‖LM (Ω)

)
dx

− k2r2

∫
∂Ω

H
(
x,

|u(x)|
‖u‖LH(∂Ω)

)
ds−

∫
Ω

h1(x) dx−
∫
∂Ω

h2(x)ds−N − b0

≥ min{1, b0}‖u‖W 1L~φ(Ω) −
∫

Ω

h1(x) dx−
∫
∂Ω

h2(x)ds− k1r1 − k2r2 −N − b0,

which implies I(u)→∞ as ‖u‖W 1L~φ(Ω) →∞.

Step 3: Existence of a weak solution. Let λ > 0 be arbitrary. Since I is coercive
there exists R > 0 such that

‖u‖W 1L~φ(Ω) > R⇒ I(u) > λ.

We define Eλ = {u ∈W 1L~φ(Ω) : I(u) ≤ λ} and denote by BR(0) the closed ball in

W 1L~φ(Ω) of radiusR centered at the origin. We claim that α = infv∈W 1L~φ(Ω) I(v) >

−∞. If not, for all n > 0 there is a sequence un ∈ Eλ such that I(un) < −n. As
Eλ ⊂ BR(0), by the Banach-Alaoglu-Bourbaki theorem there exists u ∈ BR(0)
such that, passing to a subsequence if necessary, we can assume that un ⇀ u weak∗

in W 1L~φ(Ω). So that the weak-∗ lower semicontinuity of I implies I(u) = −∞
which contradicts the fact that I is well defined on W 1L~φ(Ω). Therefore, for every

n > 0 there exists a sequence un ∈ Eλ such that I(un) ≤ α + 1
n . Thus, there

exists u ∈ BR(0) such that for a subsequence still indexed by n, un ⇀ u weak−∗
in W 1L~φ(Ω). Since I is weakly−∗ lower semicontinuous we obtain

I(u) = J(u)−L1(u)−L2(u) ≤ lim inf
n→∞

(
J(un)−L1(un)−L2(un)

)
= lim inf

n→∞
I(un) ≤ α.

Note that u belongs also to Eλ, which yields I(u) = α ≤ λ. This shows that
I(u) = min{I(v) : v ∈ W 1L~φ(Ω)}. Moreover, inserting v = −u− as test function

in (4.1) and then using (2.8), we obtain u ≥ 0. This ends the proof of Theorem
4.3. �

4.3. Uniqueness. To prove the uniqueness of the weak solution we need the fol-
lowing monotonicity assumptions:(

f(x, s)− f(x, t)
)(
s− t

)
< 0 for a.e. x ∈ Ω, ∀s, t ∈ R with s 6= t, (4.10)(

g(x, s)− g(x, t)
)(
s− t

)
< 0 for a.e. x ∈ Ω, ∀s, t ∈ R with s 6= t, (4.11)(

ϕmax(x, s)− ϕmax(x, t)
)(
s− t

)
> 0 for a.e. x ∈ Ω, ∀s, t ∈ R with s 6= t. (4.12)

Theorem 4.4. Let u be the weak solution of (1.1) given by Theorem 4.3. If in
addition (1.4), (4.10), (4.11) and (4.12) are fulfilled, then the weak solution u is
unique.



22 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

Proof. Suppose that there exists another weak solution w of problem (1.1). We
choose v = u− w as a test function in (4.1) obtaining∫

Ω

N∑
i=1

ai(x, ∂xiu)∂xi(u− w) dx+

∫
Ω

b(x)ϕmax(x, u)(u− w) dx

−
∫

Ω

f(x, u)(u− w) dx−
∫
∂Ω

g(x, u)(u− w) ds = 0.

Then choosing v = w − u as a test function in the weak formulation of solution
(4.1) solved by w, we obtain∫

Ω

N∑
i=1

ai(x, ∂xiw)∂xi(w − u) dx+

∫
Ω

b(x)ϕmax(x,w)(w − u) dx

−
∫

Ω

f(x,w)(w − u) dx−
∫
∂Ω

g(x,w)(w − u) ds = 0.

Combining the previous two equalities, we obtain∫
Ω

N∑
i=1

(
ai(x, ∂xiu)− ai(x, ∂xiw)

)
(∂xiu− ∂xiw) dx

+

∫
Ω

b(x)
(
ϕmax(x, u)− ϕmax(x,w)

)
(u− w) dx

−
∫

Ω

(
f(x, u)− f(x,w)

)
(u− w) dx−

∫
∂Ω

(
g(x, u)− g(x,w)

)
(u− w)ds = 0.

In view of (1.4), (4.10), (4.11) and (4.12) we obtain u = w a.e. in Ω. �

5. Appendix

We present some important results that are necessary for the accomplishment of
the proofs of the above results.

Lemma 5.1 ([23, Theorem B.1]). Let φ be a locally integrable Musielak-Orlicz
function. The space Eφ is separable.

Lemma 5.2 ([23, Lemma B.4]). Let φ and φ∗ be two complementary Musielak-
Orlicz functions. For every η ∈ Lφ∗(Ω), the linear functional Fη defined for every
ζ ∈ Eφ(Ω) by

Fη(ζ) =

∫
Ω

ζ(x)η(x) dx (5.1)

belongs to the dual space of Eφ(Ω), denoted Eφ(Ω)∗, and its norm ‖Fη‖ satisfies

‖Fη‖ ≤ 2‖η‖φ∗ , (5.2)

where ‖Fη‖ = sup{|Fη(u)| : ‖u‖LM (Ω) ≤ 1}.

Lemma 5.3. Let φ be a locally integrable Musielak-Orlicz function. The dual space
of Eφ(Ω) can be identified to the Musielak-Orlicz space Lφ∗(Ω).

Proof. According to Lemma 5.2 any element η ∈ Lφ∗(Ω) defines a bounded linear
functional Fη on Lφ(Ω) and also on Eφ(Ω) which is given by (5.1). It remains to
show that every bounded linear functional on Eφ(Ω) is of the form Fη for some
η ∈ Lφ∗(Ω). The proof of this claim is done in [23]. For the convenience of the



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 23

reader we give it here. Given F ∈ Eφ(Ω)∗ we define the complex measure λ by
setting

λ(E) = F (χE),

where E is a measurable subset of Ω having a finite measure and χE stands for
the characteristic function of E. Due to the fact that φ is locally integrable, the
measurable function φ

(
·, φ−1(x0,

1
2|E| )χE(·)

)
belongs to L1(Ω) for any x0 ∈ Ω.

Hence, there is an ε > 0 such that for any measurable subset Ω′ of Ω, one has

|Ω′| < ε⇒
∫

Ω′
φ
(
x, φ−1

(
x0,

1

2|E|
)
χE(x)

)
dx ≤ 1

2
.

As φ(·, s) is measurable on E, Luzin’s theorem implies that for ε > 0 there exists
a closed set Kε ⊂ E, with |E \Kε| < ε, such that the restriction of φ(·, s) to Kε is
continuous. Let k be the point where the maximum of φ(·, s) is reached in the set
Kε. ∫

E

φ
(
x, φ−1

(
k,

1

2|E|
))
dx

=

∫
Kε

φ
(
x, φ−1

(
k,

1

2|E|
))
dx+

∫
E\Kε

φ
(
x, φ−1

(
k,

1

2|E|
))
dx.

For the first term in the right-hand side of the equality, we can write∫
Kε

φ
(
x, φ−1

(
k,

1

2|E|
))
dx ≤

∫
E

φ
(
k, φ−1

(
k,

1

2|E|
))
dx ≤ 1

2
.

Since |E \Kε| < ε, the second term can be estimated as∫
E\Kε

φ
(
x, φ−1

(
k,

1

2|E|
))
≤ 1

2
.

Thus, we obtain ∫
Ω

φ
(
x, φ−1

(
k,

1

2|E|
)
χE(x)

)
dx ≤ 1.

It follows that

|λ(E)| ≤ ‖F‖‖χE‖φ ≤
‖F‖

φ−1
(
k, 1

2|E|
) .

As the right-hand side tends to zero when |E| converges to zero, the measure λ
is absolutely continuous with respect to the Lebesgue measure and so by Radon-
Nikodym’s Theorem (see for instance [1, Theorem 1.52]), it can be expressed in the
form

λ(E) =

∫
E

η(x) dx,

for some nonnegative function η ∈ L1(Ω) unique up to sets of Lebesgue measure
zero. Thus,

F (ζ) =

∫
Ω

ζ(x)η(x) dx

holds for every measurable simple function ζ. Note first that since φ is locally
integrable any measurable simple function lies in Eφ(Ω) and the set of measurable
simple functions is dense in (Eφ(Ω), ‖ · ‖φ). Indeed, for nonnegative ζ ∈ Eφ(Ω),
there exists a sequence of increasing measurable simple functions ζj converging
almost everywhere to ζ and |ζj(x)| ≤ |ζ(x)| on Ω. By the theorem of dominated



24 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

convergence one has ζj → ζ in Eφ(Ω). For an arbitrary ζ ∈ Eφ(Ω), we obtain the
same result splitting ζ into positive and negative parts.

Let ζ ∈ Eφ(Ω) and let ζj be a sequence of measurable simple functions converging
to ζ in Eφ(Ω). By Fatou’s Lemma and the inequality (5.2) we can write∣∣ ∫

Ω

ζ(x)η(x) dx
∣∣ ≤ lim inf

j→+∞

∫
Ω

|ζj(x)η(x)| dx

= lim inf
j→+∞

F (|ζj | sgn η)

≤ ‖F‖ lim inf
j→+∞

‖ζj‖φ ≤ ‖F‖‖ζ‖φ,

which implies that η ∈ Lφ∗(Ω). Thus, the linear functional Fη(ζ) =
∫

Ω
ζ(x)η(x) dx

and F are both defined on Eφ(Ω) and have the same values on the set of measurable
simple functions, so by a density argument they agree on Eφ(Ω). �

Lemma 5.4. Let Ω be an open bounded subset of RN . Let f, g : Ω × (0,+∞) →
(0,+∞) be continuous nondecreasing functions with respect to there second argu-

ment and g(·, t) is continuous on Ω with limt→∞
f(x,t)
g(x,t) = +∞, then for all ε > 0,

there exists a positive constant K0 such that

g(x, t) ≤ εf(x, t) +K0, for all t > 0.

Proof. Let ε > 0 be arbitrary. There exists t0 > 0 such that t ≥ t0 implies
g(x, t) ≤ εf(x, t). Then, for all t ≥ 0,

g(x, t) ≤ εf(x, t) +K(x),

where K(x) = supt∈(0,t0) g(x, t). Being g(·, t) continuous on Ω, one has g(x, t) ≤
εf(x, t) +K0 with K0 = maxx∈ΩK(x). �

Lemma 5.5. Let Ω be an open bounded subset of RN . Let A, B be two Musielak-
Orlicz functions such that B � A, with B(·, t) is continuous on Ω. If a sequence
{un} is bounded in LA(Ω) and converges in measure in Ω then it converges in norm
in LB(Ω).

Proof. Let us fix ε > 0. Defining vj,k(x) =
uj(x)−uk(x)

ε , we shall prove that {uj} is a
Cauchy sequence in the Banach space LB(Ω). Clearly {vj,k} is bounded in LA(Ω),
say ‖vj,k‖A ≤ K for all j and k and for some positive constant K. Since B � A
there exists a positive number t0 such that for t ≥ t0 one has

B(x, t) ≤ 1

4
A
(
x,

t

K

)
.

On the other hand, since B(·, t) is continuous on Ω we denote x0 the point where
the maximum of B(·, t) is reached in Ω. Let δ = 1

4B(x0,t0) and set

Ωj,k =
{
x ∈ Ω : |vj,k| ≥ B−1

(
x0,

1

2|Ω|

)}
.

Since {uj} converges in measure, there exists an integer N0 such that |Ωj,k| ≤ δ
whenever j, k ≥ N0. Defining

Ω′j,k = {x ∈ Ωj,k : |vj,k| ≥ t0} and Ω′′j,k = Ωj,k \ Ω′j,k,



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 25

one has∫
Ω

B(x, |vj,k(x)|) dx =

∫
Ω\Ωj,k

B(x, |vj,k(x)|) dx+

∫
Ω′j,k

B(x, |vj,k(x)|) dx

+

∫
Ω′′j,k

B(x, |vj,k(x)|) dx.
(5.3)

For the first term in the right-hand side of (5.3), we can write∫
Ω\Ωj,k

B(x, |vj,k(x)|) dx ≤
∫

Ω\Ωj,k
B
(
x,B−1

(
x0,

1

2|Ω|

))
dx

≤
∫

Ω

B
(
x0, B

−1
(
x0,

1

2|Ω|

))
dx ≤ 1

2
.

Since B � A, the second term in the right hand side of (5.3) can be estimated as
follows ∫

Ω′j,k

B(x, |vj,k(x)|) dx ≤ 1

4

∫
Ω

A
(
x,
|vj,k|
K

)
dx ≤ 1

4
,

while for the third term in the right hand side of (5.3), we obtain∫
Ω′′j,k

B(x, |vj,k(x)|) dx ≤
∫

Ωj,k

B(x, t0) dx ≤ δB(x0, t0) =
1

4
.

Finally, putting all the above estimates in (5.3), we arrive at∫
Ω

B(x, |vj,k(x)|) dx ≤ 1, for every j, k ≥ N0,

which yields ‖uj−uk‖B ≤ ε. Thus, {uj} converges in the Banach space LB(Ω). �

Lemma 5.6. Let u ∈ W 1.1
loc (Ω) and let F : Ω × R+ → R+ be a Lipschitz con-

tinuous function. If f(x) = F (x, u(x)) then f ∈ W 1.1
loc (Ω). Moreover, for every

j = 1, . . . , N , the weak derivative ∂xjf of f is such that

∂xjf(x) =
∂F (x, u(x))

∂xj
+
∂F (x, u(x))

∂t
∂xju(x), for a.e. x ∈ Ω.

Proof. Let ϕ ∈ D(Ω) and let {ej}Nj=1 be the standard basis in RN . We can write

−
∫

Ω

F (x, u(x))∂xjϕ(x) dx

= − lim
h→0

∫
Ω

F (x, u(x))
ϕ(x)− ϕ(x− hej)

h
dx

= lim
h→0

∫
Ω

F (x+ hej , u(x+ hej))− F (x, u(x))

h
ϕ(x) dx

= lim
h→0

∫
Ω

F (x+ hej , u(x+ hej))− F (x, u(x+ hej))

h
ϕ(x) dx

+ lim
h→0

∫
Ω

F (x, u(x+ hej))− F (x, u(x))

h
ϕ(x) dx

= lim
h→0

∫
Ω

Q1(x, h)ϕ(x) dx+ lim
h→0

∫
Ω

Q2(x, h)
u(x+ hej)− u(x)

h
ϕ(x) dx,



26 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

where

Q1(x, h) =

{F (x+hej ,u(x+hej))−F (x,u(x+hej))
h if h 6= 0,

∂F (x,u(x))
∂xj

if h = 0

and

Q2(x, h) =

{
F (x,u(x+hej))−F (x,u(x))

u(x+hej)−u(x) if u(x+ hej) 6= u(x),

∂F (x,u(x))
∂t otherwise.

Since F (·, ·) is Lipschitz continuous, there exist two positive constants k1 and k2

not depending on h, such that ‖Q1(·, h)‖∞ ≤ k1 and ‖Q2(·, h)‖∞ ≤ k2. Thus,

for some sequence of values of h tending to zero, Q1(·, h) converges to ∂F (x,u(x))
∂xj

and Q2(·, h) converges to ∂F (x,u(x))
∂t both in L∞(Ω) for the weak-star topology

σ∗(L∞(Ω), L1(Ω)). On the other hand, since u ∈W 1,1(supp(ϕ)) we have

lim
h→0

∫
supp(ϕ)

u(x+ hej)− u(x)

h
ϕ(x) dx =

∫
supp(ϕ)

∂ju(x)ϕ(x) dx.

It follows that

−
∫

Ω

F (x, u(x))∂xjϕ(x) dx

=

∫
Ω

∂F (x, u(x))

∂xj
ϕ(x) dx+

∫
Ω

∂F (x, u(x))

∂t
∂xju(x)ϕ(x) dx,

which completes the proof. �

Lemma 5.7. Let Ω be an open bounded subset of RN . Let A and φ be two Musielak-
Orlicz functions. We assume that φ : Ω× R+ → R+ is locally integrable, differen-
tiable with respect to its second argument and φ � A. Then ϕ(·, s) ∈ Lφ∗(Ω) for

every s ∈ LA(Ω) where ϕ(x, s) = ∂φ(x,s)
∂s .

Proof. Let s ∈ LA(Ω). By (2.8) we can write∫
Ω

φ∗(x, ϕ(x, s)) dx =

∫
Ω

∫ ϕ(x,s)

0

ϕ−1(x, τ)dτ dx

≤
∫

Ω

|s|ϕ(x, |s|) dx ≤
∫

Ω

φ(x, 2|s|) dx.

It is obvious that if s = 0 then ϕ(·, s) ∈ Lφ∗(Ω). Assume that s 6= 0. Since φ is

locally integrable and φ � A, there exists a nonnegative function h ∈ L1(Ω) such

that φ(x, 2|s|) ≤ A
(
x, |s|‖s‖A

)
+ h(x) for a.e. x ∈ Ω. Thus,∫

Ω

φ(x, 2|s|) dx ≤
∫

Ω

A
(
x,
|s|
‖s‖A

)
dx+

∫
Ω

h(x) dx <∞.

Hence, ϕ(·, s) ∈ Lφ∗(Ω). �

Let φ : Ω× R+ → R+ be a real function such that the partial function φ(x, ·) is
convex. The function φ is called the principal part of the Musielak-Orlicz function
M if M(x, t) = φ(x, t) for large values of the argument t.

Lemma 5.8. Let t0 > 0 be arbitrary and let φ : Ω×[t0,+∞[→ R+ be a real function

where the partial function φ(x, ·) is convex. We define the function ϕ(x, t) = ∂φ(x,t)
∂t .

If φ(·, t) and ϕ(·, t) are continuous on Ω and limt→+∞ infx∈Ω ϕ(x, t) = +∞. Then
φ(x, t) is the principal part of a Musielak-Orlicz function M(x, t).



EJDE-2021/26 CONTINUOUS IMBEDDING IN MUSIELAK SPACES 27

Proof. Since limt→+∞ infx∈Ω ϕ(x, t) = +∞, then there exists t1 > t0 + 1 (not
depending on x) such that supx∈Ω ϕ(x, t0 + 1) + supx∈Ω φ(x, t0) ≤ ϕ(x, t1). Thus,
we have

inf
x∈Ω

φ(x, t1) ≤ φ(x, t1) =

∫ t0+1

t0

ϕ(x, τ)dτ +

∫ t1

t0+1

ϕ(x, τ)dτ + φ(x, t0)

≤ sup
x∈Ω

ϕ(x, t0 + 1) + sup
x∈Ω

φ(x, t0) + (t1 − t0 − 1)ϕ(x, t1)

≤ (t1 − t0)ϕ(x, t1)

< t1ϕ(x, t1)

≤ t1 sup
x∈Ω

ϕ(x, t1),

from which it follows that

α =
t1 supx∈Ω ϕ(x, t1)

infx∈Ω φ(x, t1)
> 1.

We define the function

M(x, t) =

{
φ(x,t1)
tα1

tα if t ≤ t1,

φ(x, t) if t ≥ t1.

The function M(x, t) is a Musielak-Olicz function inasmuch as its derivative,

∂M(x, t)

∂t
=

{
αφ(x,t1)

tα1
tα−1 if t ≤ t1,

ϕ(x, t) if t ≥ t1,
is a function which is positive for t > 0, right-continuous for t ≥ 0 non-decreasing,

and limt→+∞
∂M(x,t)

∂t = +∞. �

Acknowledgments. The authors would like to thank the anonymous referees for
their comments and suggestions that have contributed to the improvement of this
article.

References

[1] R. A. Adams, J. J. F. Fournier; Sobolev Spaces, 2nd ed., Pure Appl. Math. (Amst.), 140,

Elsevier/Academic Press, Amsterdam, 2003.
[2] M.-M. Boureanu, V. Rǎdulescu; Anisotropic Neumann problems in Sobolev spaces with vari-

able exponent, Nonlinear Anal., 75 (2012), 4471–4482.

[3] L. Diening, P. Harjulehto, P. Hästö, M. Ružička; Lebesgue and Sobolev Spaces with Variable
Exponents, in: Lecture Notes in Math., vol. 2017, Springer-Verlag, Berlin, 2011.

[4] D.E. Edmunds, J. Rácosnik; Sobolev embedding with variable exponent, Studia Math., 143

(2000) 267–293.
[5] I. Ekeland, R. Temmam; Analyse convexe et problèmes variationnels, (French) Collection

Études Mathématiques. Dunod, Paris, 1974.
[6] X. L. Fan; An imbedding theorem for Musielak-Sobolev spaces, Nonlinear Anal., 75 (2012),

1959–1971.
[7] X. L. Fan; Anisotropic variable exponent Sobolev spaces and ~p(·)−Laplacian equations, Com-

plex Var. Elliptic Equ., 56 (2011), 623–642.

[8] X.L. Fan; Boundary trace embedding theorem for variable exponent Sobolev spaces, J. Math.
Anal. Appl., 339 (2008), 1395–1412.

[9] X.L. Fan, J.S. Shen, D. Zhao; Sobolev embedding theorems for spaces Wk,p(·)(Ω), J. Math.
Anal. Appl., 262 (2001), no. 2, 749–760.

[10] E. Gagliardo; Caratterizzazioni delle tracce sulla frontiera relative ad alcune classi di funzioni

in n variabili, Rend. Sem. Mat. Univ. Padova. (in Italian) Tome 27 (1957), 284–305.



28 A. YOUSSFI, M. M. OULD KHATRI EJDE-2021/26

[11] M. Garcia-Huidobro, V. Le, R. Manasevich, K. Schmitt; On principal eigenvalues for quasi-

linear elliptic differential operators: an Orlicz-Sobolev space setting, Nonlinear Differ. Equ.

Appl. NoDEA, 6 (1999), no. 2, 207–225.
[12] L. Gasiński, N. S. Papageorgiou, Anisotropic nonlinear Neumann problems, Calc. Var., 42

(2011), 323–354.

[13] P. Harjulehto, P. Hästö; Sobolev inequalities for variable exponents attaining the value 1 and
n, Publ. Math., 52 (2008), 347–363.

[14] E. Hewitt, K. Stromberg; Real and Abstract Analysis, Springer Varlag, Berlin Heidelberg,

New York, 1965.
[15] B. Kone, S. Ouaro, S. Traore; Weak solutions for anisotropic nonlinear elliptic equations

with variable exponents, Electron. J. Differential Equations, 2009 (2009) 144, 1–11.

[16] O. Kováčik, J. Rácosnik; On spaces Lp(x) and W 1,p(x), Czechoslovak Math. J., 41 (1991),
592–618.

[17] V. Le, K. Schmitt; Quasilinear elliptic equations and inequalities with rapidly growing coef-

ficients, J. Lond. Math. Soc. (2) 62 (2000), no. 3, 852–872.
[18] D. Liu, B. Wang and P. Zhao; On the trace regularity results of Musielak-Orlicz-Sobolev

spaces in a bounded domain, Commun. Pure Appl. Anal., 15 (2016),Issue 1, 1643–1659.

[19] M. Mihailescu, P. Pucci, V. Radulescu; Eigenvalue problems for anisotropic quasilinear el-
liptic equations with variable exponent, J. Math. Anal. Appl., 340 (2008), no. 1, 687–698.

[20] J. Musielak; Orlicz Spaces and Modular Spaces, in: Lecture Notes in Math., Vol 1034,
Springer-Varlag, Berlin, 1983.

[21] N. S. Papageorgiou, C. Vetro, F. Vetro; Positive and nodal solutions for nonlinear nonhomo-

geneous parametric Neumann problems, Electron. J. Differential Equations, 2020 Vol. 2020
(2020), no. 12, 1–20.

[22] M. Struwe; Variational Methods: Applications to Nonlinear Partial Differential Equations

and Hamiltonian Systems, Springer-Verlag, Berlin, 1996.
[23] A. Youssfi, Y. Ahmida; Some Approximation Results in Musielak-Orlicz Spaces, Czech Math

J., 70 (2020), 453–471 .

Ahmed Youssfi

University Sidi Mohamed Ben Abdellah, Laboratory of Mathematical Analysis and Ap-
plications, National School of Applied Sciences, P.O. Box 72 Fès-Principale, Fez, Mo-

rocco
Email address: ahmed.youssfi@usmba.ac.ma, ahmed.youssfi@gmail.com

Mohamed Mahmoud Ould Khatri

University Sidi Mohamed Ben Abdellah, National School of Applied Sciences, P.O. Box
72 Fès-Principale, Fez, Morocco

Email address: mahmoud.ouldkhatri@usmba.ac.ma


	1. Introduction
	2. Preliminaries
	2.1. Anisotropic Musielak-Orlicz-Sobolev spaces

	3. Main results
	3.1. Imbedding theorem
	3.2. Trace result

	4. Application to anisotropic elliptic equations
	4.1. Properties of the energy functional
	4.2. Existence of solutions
	4.3. Uniqueness

	5. Appendix
	Acknowledgments

	References