Mathematics in Engineering, 5(4): 1–33.

DOI:10.3934/mine.2023073
Received: 14 March 2022
Revised: 29 November 2022
Accepted: 04 January 2023
http://www.aimspress.com/journal/mine Published: 31 January 2023

Research article
Anisotropic elliptic equations with gradient-dependent lower order terms
and L1 data†

Barbara Brandolini1 and Florica C. Cı̂rstea2,∗

1
Dipartimento di Matematica e Informatica, Università degli Studi di Palermo, via Archirafi 34,
90123 Palermo, Italy
2
School of Mathematics and Statistics, The University of Sydney, NSW 2006, Australia
†
This contribution is part of the Special Issue: Nonlinear PDEs and geometric analysis
Guest Editors: Julie Clutterbuck; Jiakun Liu
Link: www.aimspress.com/mine/article/6186/special-articles

* Correspondence: Email: florica.cirstea@sydney.edu.au; Tel: +61293512965;

Fax: +61293514534.
Abstract: We prove the existence of a weak solution for a general class of Dirichlet anisotropic
elliptic problems such as Au + Φ(x, u, ∇u) = Bu + f in Ω, where Ω is a bounded open subset of RN and
f ∈ L1 (Ω) is arbitrary. The principal part is a divergence-form nonlinear anisotropic operator A, the
prototype of which is Au = − Nj=1 ∂ j (|∂ j u| p j −2 ∂ j u) with p j > 1 for all 1 ≤ j ≤ N and Nj=1 (1/p j ) > 1.
P P
As a novelty in this paper, our lower order terms involve a new class of operators B such that A − B is
→
−
bounded, coercive and pseudo-monotone from W01, p (Ω) into its dual, as well as a gradient-dependent
nonlinearity Φ with an “anisotropic natural growth” in the gradient and a good sign condition.
Keywords: nonlinear anisotropic elliptic equations; Leray–Lions operators; pseudo-monotone
operators; lower order terms; summable data

To Professor Neil S. Trudinger, on the occasion of his 80th birthday, with admiration and gratitude.

1. Introduction and main results

1.1. Setting of the problem

In their famous book [34], Gilbarg and Trudinger captured the astonishing achievements in the
theory of nonlinear elliptic second order partial differential equations. For recent developments of
fully nonlinear equations and their applications to optimal transportation and conformal geometry, see
e.g., Trudinger [41, 42].
 2

A quasilinear operator is not always the differential of a functional of the Calculus of Variations.
What makes it possible to go further than the Calculus of Variations in the convex case is the abstract
concept of monotone operator and, more generally, of pseudo-monotone operator. Several papers
[10, 11, 13, 15, 16] deal with nonlinear elliptic problems in a bounded open subset Ω of RN involving
coercive, bounded, continuous and pseudo-monotone Leray–Lions type operators from W01,p (Ω) into
its dual W −1,p (Ω), where 1 < p < ∞ and p0 = p/(p − 1) is the conjugate exponent of p. The prototype
0

model of such an operator is the p-Laplacian ∆ p u = div (|∇u| p−2 ∇u). The techniques developed in these
papers accommodate for a lower-order term g(x, u, ∇u) with a “natural growth” in the gradient |∇u| and
0
without any restriction of its growth in |u|. Either f ∈ L1 (Ω) or h ∈ W −1,p (Ω) could be included because
of the “sign-condition” on g (that is, g(x, t, ξ) t ≥ 0 for a.e. x ∈ Ω and all (t, ξ) ∈ R × RN ). For related
works, we refer to [3, 4, 12, 29, 30, 35].
In this paper, we expand the above research program into the anisotropic arena by providing a
suitable general framework under which for every f ∈ L1 (Ω) we prove the existence of a weak solution
to Dirichlet anisotropic elliptic problems such as

Au + Φ(x, u, ∇u) + Θ(x, u, ∇u) = Bu + f in Ω,




(1.1)
u ∈ W 1,→
 −p
Φ(x, u, ∇u) ∈ L1 (Ω).
0 (Ω),


Here, and henceforth, Ω is a bounded, open subset of RN (N ≥ 2). We impose no smoothness condition
on ∂Ω. Without loss of generality, we assume throughout that

1 < p j ≤ p j+1 < ∞ for every 1 ≤ j ≤ N − 1 and p < N, (1.2)

where p = N/ is the harmonic mean of p1 , . . . , pN . Let ∇u = (∂1 u, . . . , ∂N u) be the gradient

PN
j=1 (1/p j )
1,→
−p
of u. Let W0 (Ω) be the closure of Cc∞ (Ω) (the set of smooth functions with compact support in Ω)
→−0 →−
with respect to the norm kukW 1,→−p (Ω) = Nj=1 k∂ j ukL p j (Ω) . We use W −1, p (Ω) to denote the dual of W01, p (Ω)
P
0
→
−
and h·, ·i for the duality between W −1, p (Ω) and W 1, p (Ω). The prototype for A is the anisotropic → −p -
→
−0
0
Laplacian, namely,
N
X
Au = − ∂ j (|∂ j u| p j −2 ∂ j u), (1.3)
j=1

(see (1.7) and (1.8)). The model for Φ in (1.1) is as follows

 N  N
X  X
Φ(u, ∇u) =  |∂ j u| + 1 |u| u +
 p j  m−2
b j |∂ j u|q j |u|θ j −2 u, (1.4)
j=1 j=1

where b j ≥ 0 and 0 ≤ q j < p j , while m, θ j > 1 for all 1 ≤ j ≤ N are arbitrary (see (1.9) and (1.10)).
We assume throughout that Θ(x, t, ξ), Ω × R × RN → R is a Carathéodory function (that is, measurable
on Ω for every (t, ξ) ∈ R × RN and continuous in t, ξ for a.e. x ∈ Ω) and there exists a constant CΘ > 0
such that
|Θ(x, t, ξ)| ≤ CΘ for a.e. x ∈ Ω and for all (t, ξ) ∈ R × RN . (1.5)
Furthermore, our problem (1.1) features a new class of operators B as follows.

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 3

Definition 1.1. Let (1.7) and (1.8) hold. By BC we denote the class of all bounded operators B from
→− →−0
W01, p (Ω) into W −1, p (Ω) satisfying the following two properties:
→
− →
−0
(P1 ) The operator A − B from W01, p (Ω) into W −1, p (Ω) is coercive (see Definition 2.2).
→
−
(P2 ) If u` * u and v` * v (weakly) in W01, p (Ω) as ` → ∞, then

lim hBu` , v` i = hBu, vi.

`→∞

Let BC+ be the class of operators in BC satisfying the extra condition

(P3 ) For ν0 > 0 in the coercivity condition of (1.8) and each k > 0, it holds

N
X p
ν0 k∂ j ukL pj j (Ω) − hBu, T k ui → ∞ as kukW 1,→−p (Ω) → ∞. (1.6)
0
j=1

We use T k for the truncation at height k, see (1.16). Unlike A, the operator −B is not coercive
in general. Our assumption (P2 ) is reminiscent of (iii) in the Hypothesis (II) of Theorem 1 in the
celebrated paper [37] by Leray and Lions. Every operator satisfying (P2 ) is strongly continuous (see
Lemma 2.6) and pseudo-monotone (cf. [44, p. 586]).
→
−
In Example 1.2 we use that p∗ = N p/(N − p) is the critical exponent for the embedding W01, p (Ω) ,→
Lr (Ω) (see Remark A.2 in the Appendix). For any r > 1, let r0 = r/(r − 1).
→
−0
h ∈ W −1, p (Ω) be arbitrary. Let ρ, αk ∈ R for 0 ≤ k ≤ 4. For
∗ 0
Example 1.2. Let F ∈ L(p ) (Ω) and h, e
→−
every u ∈ W01, p (Ω), we define
(1) Bu = h;
ϑ−2
(2) Bu = F  + ρ |u| ub with 1 < ϑ < p if  ρ > 0 and 1 < ϑ < p if ρ < 0;
∗

(3) Bu = α0 + α1 kukL1r (Ω) + α2 |heh, ui|b2 (α3 h + α4 F), where r ∈ [1, p∗ ); we take b1 ∈ (0, p/p01 ) and
b2 ∈ (0, p1 − 1) if α3 ,0; b1 ∈ (0, p − 1) and b2 ∈ (0, pp10 /p ) if α3 = 0;
0

(4) Bu = − j=1 ∂ j β j (x) + |u|σ j −1 u , where β j ∈ L j (Ω) and 0 < σ j < p/p0j for every 1 ≤ j ≤ N.
PN

In each of these situations, B belongs to the class BC+ .

1.2. Our assumptions

Let (1.2) and (1.5) hold. The anisotropic →
−p -Laplacian in (1.3) is the prototype for a coercive,
→− →
−0
bounded, continuous and pseudo-monotone operator A : W01, p (Ω) → W −1, p (Ω) in divergence form
Au = − Nj=1 ∂ j (A j (x, u, ∇u)), that is,
P

N Z
X →
−
hAu, vi = A j (x, u, ∇u) ∂ j v dx for every u, v ∈ W01, p (Ω). (1.7)
j=1 Ω

• For each 1 ≤ j ≤ N, let A j (x, t, ξ) : Ω × R × RN → R be a Carathéodory function and assume that

0
there exist constants ν0 , ν > 0 and a nonnegative function η j ∈ L p j (Ω) such that for a.e. x ∈ Ω, for all

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 4

(t, ξ) ∈ R × RN and every b

ξ ∈ RN , we have
N N

X X 
Ai (x, t, ξ) ξi ≥ ν0 |ξi | pi

[coercivity],






i=1 i=1 



N



X  
Ai (x, t, ξ) − Ai (x, t, ξ) ξi − ξi > 0 if ξ , ξ [monotonicity],


(1.8)
b b b 



i=1 


 1/p0  

N
   j 

/p
∗ 0
X 
|A j (x, t, ξ)| ≤ ν η j (x) + |t| j +  |ξi | i  
p

 p

 
 
[growth condition]. 
    




i=1

We note that in the growth condition in (1.8), we take the greatest exponent for |t| from the
viewpoint of the anisotropic Sobolev inequalities. This requires modifying the standard proof of
pseudo-monotonicity of A (see Lemma 2.7).
• Suppose that Φ(x, t, ξ) : Ω × R × RN → R is a Carathéodory function and there exist a nonnegative
function c ∈ L1 (Ω) and a continuous nondecreasing function φ : R → R+ such that for a.e. x ∈ Ω and
for all (t, ξ) ∈ R × RN ,
 N 
X 
Φ(x, t, ξ) t ≥ 0 [sign-condition], |Φ(x, t, ξ)| ≤ φ(|t|)  |ξ j | p j + c(x) . (1.9)
j=1

For Theorem 1.3 (ii), we further assume that there exist constants τ, γ > 0 such that
N
X
|Φ(x, t, ξ)| ≥ γ |ξ j | p j for all |t| ≥ τ, a.e. x ∈ Ω and all ξ ∈ RN . (1.10)
j=1

1.3. Main results

→
−
By a solution of (1.1) we mean any function u ∈ W01, p (Ω) such that Φ(x, u, ∇u) ∈ L1 (Ω) and, for
→
−
every v ∈ W01, p (Ω) ∩ L∞ (Ω),
Z Z Z
hAu, vi + Φ(x, u, ∇u) v dx + Θ(x, u, ∇u) v dx = hBu, vi + f v dx. (1.11)
Ω Ω Ω

Under the assumptions in Section 1.2, the main advance in this paper is the following.
Theorem 1.3. Let (1.2), (1.5), (1.8) and (1.9) hold.
(i) If f = 0 in (1.1), then (1.1) has a solution U for every B in the class BC. Moreover,
Φ(x, U, ∇U) U ∈ L1 (Ω) and (1.11) holds with u = v = U.
(ii) If (1.10) is satisfied, then (1.1) has at least a solution for every f ∈ L1 (Ω) and B in the class BC+ .
Theorem 1.3 is new even when transposed to isotropic Leray–Lions type operators A from W01,p (Ω)
0
into W −1,p (Ω). This is due to the introduction of B in (1.1), which adds extra difficulties. Were a
→−
solution of (1.1) to exist in W01, p (Ω), then we would expect it to be unbounded. This was observed by
0
Bensoussan, Boccardo and Murat [11] for isotropic nonlinear elliptic equations involving h ∈ W −1,p (Ω)

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 5

and Leray–Lions type operators of the p-Laplacian type. Hence, the regularizing effect that otherwise
→−
Φ would bring to the solutions in W01, p (Ω) is countered by the presence of B in our class BC.
→
−
We stress that, without the term Φ, one cannot expect to find solutions of (1.1) in W01, p (Ω) for every
f ∈ L1 (Ω). In the isotropic case, this observation was made by Boccardo and Gallouët [16]. We
leverage the gradient-dependent lower-order term Φ(x, u, ∇u) with an “anisotropic natural growth” in
the gradient and a good sign condition (see (1.9) and (1.10)) to prove the a priori estimates contained
→
−
in Lemmata 3.1 and 4.1, and hence obtain the existence of solutions of (1.1) in W01, p (Ω).
For Theorem 1.3 (ii) we encounter two obstacles: a low summability for f and, on the other hand,
the unrestricted growth of Φ with respect to |u|. Previously mentioned works in the isotropic case
provide ways to surmount one problem at a time. The function f ∈ L1 (Ω) can surely be approximated
by L∞ (Ω)-functions fε in the sense that | fε | ≤ | f | and fε → f a.e. in Ω as ε → 0. Also Φ could be
replaced by a “nice” function Φε , preserving the properties of Φ, but gaining boundedness, see (3.1).
However, as it was pointed out by Bensoussan and Boccardo [10] in the isotropic case, one cannot deal
with both approximations for f and Φ simultaneously. For the approximate problems involving both
→
−
Φε and fε , we would not be able to obtain that the solutions uε are uniformly bounded in W01, p (Ω) with
respect to ε. For the above reason, we need to consider f = 0 first and prove Theorem 1.3 (i), which is
a crucial step in establishing the second assertion of Theorem 1.3, but at the same time of independent
interest.
The techniques and results we obtain here provide the means to address other types of lower order
terms than f ∈ L1 (Ω), yet maintaining the class BC of operators B. We briefly mention possible
developments. It is natural to ask what happens when b j in (1.4) is negative for 1 ≤ j ≤ N. Then,
the sign-condition on Φ in (1.9) breaks down. Since we impose no restriction on the growth of Φ with
respect to |u|, the current paper lays the foundation for dealing with potentially singular lower order
terms with no sign restriction. The approximation of such terms is afforded by our inclusion in (1.1)
of the term Θ. The approximate problems become of the type (1.1) for which we gain existence of
solutions via our Theorem 1.3 (i). It is essential that we can take the solution itself as a test function.
This fact can be exploited to obtain a priori estimates for the solutions and pass to the limit. Such an
analysis goes beyond the scope of this paper and will be carried out elsewhere (see [19]).
Our work is also motivated by the various applications of anisotropic elliptic and parabolic partial
differential equations to the mathematical modelling of physical and mechanical processes. Such
equations provide, for instance, the mathematical models for the dynamics of fluids in anisotropic
media when the conductivities of the media are distinct in different directions. The term B allows us
to model a large class of situations, including the ones when the diffusion/absorption balance is subject
to suitable conditions (see [6, Chapter 1]). Anisotropic equations also appear in biology as a model for
the propagation of epidemic diseases in heterogeneous domains [9].
With a rapidly growing literature on anisotropic problems, several questions have been resolved on
the existence, uniqueness and regularity of weak solutions (see, for instance, [1, 2, 5, 7, 8, 14, 22–25, 28,
31–33, 38]). Many difficulties arise in passing from the isotropic setting to the anisotropic one since
some fundamental tools available for the former (such as the strong maximum principle, see [43])
cannot be extended to the latter.
We end this brief overview by observing that, in the case of variable exponents, that is when
p j = p j (x), there are many applications to electrorheological fluids, thermorheological fluids, elastic
materials, and image restoration (see, for example, [17, 18, 26, 27] and the references therein).

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 6

1.4. Sketch of the main ideas in the proof of Theorem 1.3

We remark that because of Φ, even when f = 0, we cannot directly apply the theory of pseudo-
monotone operators to prove the existence claim in Theorem 1.3 (i). To overcome this difficulty, we
consider the approximate problem

Auε + Φε (x, uε , ∇uε ) + Θ(x, uε , ∇uε ) = Buε in Ω,




(1.12)
uε ∈ W 1,→

 −p
(Ω)
0

for which we obtain the existence of a solution uε as a consequence of our Theorem 2.1 in Section 2.
Indeed, Φε + Θ satisfies the same type of assumption as Θ in (1.5), that is, there exists a constant Cε > 0
such that |(Φε + Θ)(x, t, ξ)| ≤ Cε for a.e. x ∈ Ω and all (t, ξ) ∈ R × RN . Thus, by Theorem 2.1, for every
→
−
ε > 0, the approximate problem (1.12) has a solution uε ∈ W01, p (Ω). In Lemma 3.1 we prove a priori
→
−
estimates in W01, p (Ω) for the solutions uε , which (up to a subsequence) converge weakly to some U in
→
−
W01, p (Ω) and a.e. in Ω as ε → 0.
We point out that in Section 5, we will be able to show that, up to a subsequence,
→
−
uε → U (strongly) in W01, p (Ω) as ε → 0. (1.13)

We achieve this by combining and extending techniques from the isotropic case in [10] and [13] to
establish in Lemma 3.2 that, up to a subsequence of uε , we have
→
−
∇uε → ∇U a.e. in Ω and T k (uε ) → T k (U) (strongly) in W01, p (Ω) as ε → 0 (1.14)

for every integer k ≥ 1, where T k (·) is given in (1.16). Then, we can pass to the limit as ε → 0
in the weak formulation of the solution uε and obtain that U is a solution of (1.1) with f = 0 (see
Subsection 3.2).
Generally speaking, the proof of Theorem 1.3 (ii), which we give in Section 4, follows a similar
course with that of Theorem 1.3 (i) in Section 3. But there are some modifications that we outline
below. We approximate f ∈ L1 (Ω) by L∞ (Ω)-functions fε and we apply Theorem 1.3 (i) to obtain a
solution Uε for the problem

AUε + Φ(x, Uε , ∇Uε ) + Θ(x, Uε , ∇Uε ) = BUε + fε in Ω,




(1.15)
Uε ∈ W 1,→
 −p
Φ(x, Uε , ∇Uε ) ∈ L1 (Ω).
0 (Ω),


We emphasize that unlike in (1.12), we have Φ (and not Φε ) in (1.15). Because of this reason, coupled
with the introduction of fε , we need the extra assumption (1.10) and to choose B in the class BC+
→−
to obtain that {Uε }ε is uniformly bounded in W01, p (Ω) with respect to ε (see Lemma 4.1 for details).
→
−
Then, extracting a subsequence, Uε tends to some U0 weakly in W01, p (Ω) and a.e. in Ω. With an almost
identical argument, we gain the counterpart of (1.14), namely, up to a subsequence, ∇Uε → ∇U0 a.e.
→
−
in Ω and T k (Uε ) → T k (U0 ) (strongly) in W01, p (Ω) as ε → 0 for every integer k ≥ 1. To conclude the
proof of Theorem 1.3 (ii), it remains to pass to the limit in the weak formulation of Uε . The change
appearing here compared with the corresponding argument in Subsection 3.2 is the strong convergence
of Φ(x, Uε , ∇Uε ) to Φ(x, U0 , ∇U0 ) in L1 (Ω). For the latter, we adapt an argument from [13]. For details,
we refer to Lemma 4.3 in Subsection 4.3.

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 7

Structure of this paper. In Section 2 we prove an existence result (Theorem 2.1), which gives the
existence of a solution uε of (1.12) for every ε > 0. We dedicate Sections 3 and 4 to the proof
of Theorem 1.3 (i) and Theorem 1.3 (ii), respectively. In Section 5 we make further comments on
Theorem 1.3 (i) by proving the strong convergence in (1.13). In the Appendix we include some facts
used in the paper and, for completeness, prove the anisotropic counterparts of well-known isotropic
convergence results, see Lemmata A.4 and A.5. These will be used in the proof of Lemmata 2.7 and
3.2, respectively.
Notation. For k > 0, we let T k : R → R stand for the truncation at height k, that is,
s
T k (s) = s if |s| ≤ k, T k (s) = k if |s| > k. (1.16)
|s|

Moreover, we define Gk : R → R by

Gk (s) = s − T k (s) for every s ∈ R. (1.17)

In particular, we have Gk = 0 on [−k, k] and t Gk (t) ≥ 0 for every t ∈ R.

→−
For every u ∈ W01, p (Ω) and for a.e. x ∈ Ω, we define

Φ(u)(x),
b = Φ(x, u(x), ∇u(x)), Θ(u)(x),
b = Θ(x, u(x), ∇u(x)),
bj (u)(x) = A j (x, u(x), ∇u(x))
A for every 1 ≤ j ≤ N.

We set →
−p = (p , p , . . . , p ) and →
1 2 N
−p 0 = (p0 , p0 , . . . , p0 ).
1 2 N
As usual, χω stands for the characteristic function of a set ω ⊂ RN .

2. An existence result

Throughout this section, we assume (1.2), (1.5), and (1.8), besides B belonging to the class BC.
Here, our aim is to prove the existence of a solution to the following problem

Au + Θ(x, u, ∇u) = Bu in Ω,




(2.1)
u ∈ W 1,→
 −p
0 (Ω).


→
−
By a solution of (2.1), we mean a function u ∈ W01, p (Ω) such that
Z
→
−
hAu, vi + Θ(x, u, ∇u) v dx − hBu, vi = 0 for every v ∈ W01, p (Ω). (2.2)
Ω

Theorem 2.1. Problem (2.1) admits at least a solution.

We establish Theorem 2.1 via the theory of pseudo-monotone operators. Before giving the proof of
Theorem 2.1 in Subsection 2.1, we recall a few concepts that we need in the sequel (see, for example,
[20] and [44, p. 586]).

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 8
→
− →
−0
Definition 2.2. An operator P : W01, p (Ω) → W −1, p (Ω) is called
→
−
(a1 ) monotone (strictly monotone) if hPu − Pv, u − vi ≥ 0 for every u, v ∈ W01, p (Ω) (with equality if
and only if u = v);
→
−
(a2 ) pseudo-monotone if whenever u` * u (weakly) in W01, p (Ω) as ` → ∞ and lim sup`→∞ hPu` , u` −
→
−
ui ≤ 0, we get that hPu, u − wi ≤ lim inf `→∞ hPu` , u` − wi for all w ∈ W01, p (Ω);
→
−
(a3 ) strongly continuous∗ if u` * u (weakly) in W01, p (Ω) as ` → ∞ implies that Pu` → Pu in
→−0
W −1, p (Ω) as ` → ∞;
(a4 ) coercive if hPu, ui/kukW 1,→−p (Ω) → ∞ as kukW 1,→−p (Ω) → ∞;
0 0
→−
(a5 ) of M type if u` * u (weakly) in
†
as ` → ∞, together with Pu` * g (weakly) in
W01, p (Ω)
−1,→
−p 0
W (Ω) as ` → ∞ and lim sup`→∞ hPu` , u` i ≤ hg, ui, imply that g = Pu and hPu` , u` i → hg, ui as ` →
∞.
→
− →
−0
Proposition 2.3. Every strongly continuous operator P : W01, p (Ω) → W −1, p (Ω) is pseudo-monotone.
→
− →
−0
Every bounded operator P : W01, p (Ω) → W −1, p (Ω) of M type is pseudo-monotone. The sum of two
pseudo-monotone operators is pseudo-monotone.

2.1. Proof of Theorem 2.1

→
− →
−0
We immediately observe from (1.5) that the operator PΘ : W01, p (Ω) → W −1, p (Ω) is bounded, where
we define Z
→
−
hPΘ (u), vi := Θ(u)
b v dx for every u, v ∈ W 1, p (Ω).
0 (2.3)
Ω

In view of (2.2), the existence of a solution to (2.1) follows whenever the operator A + PΘ − B :
→− →−0 →−
W01, p (Ω) → W −1, p (Ω) is surjective. Since W01, p (Ω) is a real, reflexive, and separable Banach space, it
→
− →
−0
is known that A + PΘ − B : W01, p (Ω) → W −1, p (Ω) is surjective whenever it is bounded, coercive and
pseudo-monotone (see, for instance, [44, p. 589]). In Lemma 2.5, we establish the boundedness and
coercivity of A + PΘ − B, whereas its pseudo-monotonicity is concluded in Corollary 2.8.
For the reader’s convenience and to make our presentation self-contained, we give all the details
→
− →−0
about the pseudo-monotonicity of A + PΘ − B : W01, p (Ω) → W −1, p (Ω). These computations could
be of interest also in the corresponding isotropic case, when, to our best knowledge, only very special
instances of B have been considered and the details are usually scattered in the literature.
→− →−0
The property (P2 ) ensures that ±B : W01, p (Ω) → W −1, p (Ω) is strongly continuous (see Lemma 2.6)
and, hence, pseudo-monotone by Proposition 2.3. As the sum of two pseudo-monotone operators is
→
− →−0
pseudo-monotone, to prove that A + PΘ − B : W01, p (Ω) → W −1, p (Ω) is pseudo-monotone, it suffices
→
− →
−0
to show that A + PΘ : W01, p (Ω) → W −1, p (Ω) is pseudo-monotone. The proof of the latter is more
involved, see Lemma 2.7. In view of Proposition 2.3 and Lemma 2.4, it is enough to show that A + PΘ
is an operator of M type. We proceed with the details.
→
− →
−0
Lemma 2.4. The operator A + PΘ : W01, p (Ω) → W −1, p (Ω) is bounded, coercive and continuous.
∗
Strongly continuous operators are also referred to as completely continuous (see, for instance, Showalter [39, p. 36]).
†
Some authors (see, for example, Le Dret [36, p. 232]) use the terminology sense 1 pseudomonotone instead of M type.

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 9
→
− →
−0
Proof. The boundedness of the operator A+PΘ : W01, p (Ω) → W −1, p (Ω) is a consequence of the growth
condition of A j in (1.8), coupled with (1.5). The coercivity of A+PΘ follows readily from (1.5) and the
coercivity assumption in (1.8). Moreover, by Hölder’s inequality and the continuity of the embedding
→− →−
W01, p (Ω) ,→ L p (Ω), we find a positive constant C such that, for every u1 , u2 ∈ W01, p (Ω),
∗

k(A + PΘ )(u1 ) − (A + PΘ )(u2 )kW −1,→−p 0 (Ω)

X N Z Z 
≤ sup |A bj (u2 )||∂ j v| dx +
bj (u1 ) − A |Θ(u
b 1 ) − Θ(u
b 2 )||v| dx
1,→
−p
v∈W0 (Ω), j=1 Ω Ω
kvk 1,→
−p ≤1
W (Ω)
0
N
X
≤ kA bj (u2 )k p0 + C ||Θ(u
bj (u1 ) − A b 1 ) − Θ(u
b 2 )||L(p∗ )0 (Ω) .
L j (Ω)
j=1

→
− →
−0
We get the continuity of A + PΘ : W01, p (Ω) → W −1, p (Ω) by showing the following.
→
− →−
Claim: The mappings Θ bj : W 1, p (Ω) → L p0j (Ω) are continuous for each
b : W 1, p (Ω) → L(p∗ )0 (Ω) and A
0 0
1 ≤ j ≤ N.
Proof of the Claim. Let 1 ≤ j ≤ N be arbitrary. By the growth condition of A j in (1.8), there exist a
0
constant C > 0 and a nonnegative function η j ∈ L p j (Ω) such that
N
 
p0j
 p0j ∗
X
bj (u)| ≤ C η + |u| p +

|A |∂i u| pi  ∈ L1 (Ω) (2.4)
j
i=1

→
− →
−
for all u ∈ W01, p (Ω). Since the embeddings W01, p (Ω) ,→ L p (Ω) and L∞ (Ω) ,→ L(p ) (Ω) are continuous,
∗ ∗ 0

→− →
−
from (2.4) and (1.5), we infer that A bj : W 1, p (Ω) → L p0j (Ω) and Θ
b : W 1, p (Ω) → L(p∗ )0 (Ω) are well-
0 0
→
−
defined. To prove the continuity of these mappings, we let un → u (strongly) in W01, p (Ω) as n → ∞.
Hence, un → u (strongly) in L p (Ω) and ∂i un → ∂i u (strongly) in L pi (Ω) as n → ∞ for every 1 ≤ i ≤ N.
∗

Now, using (2.4) with un instead of u, we obtain that {|A bj (un )| p0j }n≥1 is uniformly integrable over Ω.
By passing to a subsequence {unk }k≥1 of {un }, we have unk → u and ∇unk → ∇u a.e. in Ω as k → ∞.
Since A j and Θ are Carathéodory functions, we have Θ(u b nk ) → Θ(u) b and A bj (unk ) → Abj (u) a.e. in Ω
as k → ∞. Then, by (1.5) and the Dominated Convergence Theorem, Θ(u b nk ) → Θ(u) ∗ 0
b in L(p ) (Ω).
By Vitali’s Theorem, we see that A bj (u) in L p0j (Ω) as k → ∞. Since the limits Θ(u)
bj (unk ) → A b and
A j (u) are independent of the subsequence {unk }k≥1 , we conclude that Θ(un ) → Θ(u) in L (Ω) and (p )0
∗
b b b
bj (un ) → A
A bj (u) in L p0j (Ω) as n → ∞.
This completes the proof of the Claim and of Lemma 2.4. 
→
− →
−0
Lemma 2.5. The operator A + PΘ − B : W01, p (Ω) → W −1, p (Ω) is bounded and coercive.
Proof. Using Lemma 2.4 and Definition 1.1, we find that A + PΘ − B is a bounded operator from
→− →−0
W01, p (Ω) into W −1, p (Ω). We now show that it is also coercive, namely,

hAu + PΘ (u) − Bu, ui

→ ∞ as kukW 1,→−p (Ω) → ∞. (2.5)
kukW 1,→−p (Ω) 0
0

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 10
→
−
Using (1.5) and the continuity of the embedding W01, p (Ω) ,→ L1 (Ω), we find a constant C > 0 such
→
−
that hPΘ (u), ui ≥ −CkukW 1,→−p (Ω) for every u ∈ W01, p (Ω). Then, by the coercivity property of A − B, we
0
readily conclude (2.5). 
→
− →
−0
Lemma 2.6. Every operator B : W01, p (Ω) → W −1, p (Ω) satisfying (P2 ) is strongly continuous.
→
− →
−0
Proof. Let u` * u (weakly) in W01, p (Ω) as ` → ∞. We show that Bu` → Bu in W −1, p (Ω) as ` → ∞.
Assume by contradiction that there exist ε0 > 0 and a subsequence of {u` } (relabeled {u` }) such that

sup |hBu` − Bu, vi| > ε0 for every ` ≥ 1.

1,→
−p
v∈W0 (Ω),
kvk 1,→
−p ≤1
W (Ω)
0

→
−
Hence, there also exists {v` } in W01, p (Ω) with kv` kW 1,→−p (Ω) ≤ 1 such that
0

|hBu` − Bu, v` i| > ε0 for all ` ≥ 1. (2.6)

→
− →
−
By the boundedness of {v` } in W01, p (Ω), up to a subsequence, v` * v (weakly) in W01, p (Ω) as ` →
→
−0
∞. Since Bu ∈ W −1, p (Ω), we have hBu, v` i → hBu, vi as ` → ∞. Hence, from (P2 ) we find that
|hBu` − Bu, v` i| → 0 as ` → ∞, which is in contradiction with (2.6). Thus, B is strongly continuous,
completing the proof. 
→
− →
−0
Lemma 2.7. The operator A + PΘ : W01, p (Ω) → W −1, p (Ω) is pseudo-monotone.
Proof. Since the operator A + PΘ is bounded, it is enough to show that it is of M type (see
→− →−0
Proposition 2.3). To this end, suppose that there exist u, {u` }`≥1 in W01, p (Ω) and g ∈ W −1, p (Ω) such that
→
−
u` * u (weakly) in W01, p (Ω) as ` → ∞, (2.7)
−1,→
−p 0
(A + PΘ )(u` ) * g (weakly) in W (Ω) as ` → ∞, (2.8)
lim sup h(A + PΘ )(u` ), u` i ≤ hg, ui. (2.9)
`→∞

We prove that

g = (A + PΘ )(u), (2.10)
h(A + PΘ )(u` ), u` i → hg, ui as ` → ∞. (2.11)
→
−
We first show that (2.11) holds. From (2.7) and the compactness of the embedding W01, p (Ω) ,→ L p (Ω)
(see Remark A.2), we obtain that, up to a subsequence,

u` → u strongly in L p (Ω) and a.e. in Ω. (2.12)

Moreover, using (2.4) with u replaced by u` , we get that A bj (u` ) is bounded in L p0j (Ω) for every 1 ≤ j ≤
0
N. Hence, in view of (1.5), there exist µ ∈ L p (Ω) and g j ∈ L p j (Ω) for 1 ≤ j ≤ N so that, up to a further
0

subsequence of {u` } (denoted by {u` }), we have

0
b ` ) * µ (weakly) in L p0 (Ω)
Θ(u and bj (u` ) * g j
A (weakly) in L p j (Ω) (2.13)

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 11
→
−
as ` → ∞ for every 1 ≤ j ≤ N. Thus, by the reflexivity of W01, p (Ω) and (2.8), we get
N Z
X Z
hg, vi = lim h(A + PΘ )(u` ), vi = g j ∂ j v dx + µ v dx (2.14)
`→∞ Ω Ω
j=1

→
−
for every v ∈ W01, p (Ω). From (2.12) and (2.13), we infer that
Z Z
lim Θ(u` ) u` dx =
b µ u dx. (2.15)
`→∞ Ω Ω

From (2.9), (2.14) and (2.15), we obtain that

 N Z Z 
X 
lim sup h(A + PΘ )(u` ), u` i = lim sup  bj (u` ) ∂ j u` dx +
A Θ(u
b ` ) u` dx
`→∞ `→∞ Ω Ω

j=1
N Z Z (2.16)
X
≤ hg, ui = g j ∂ j u dx + µ u dx,
j=1 Ω Ω

that is,
N Z
X N Z
X
lim sup bj (u` ) ∂ j u` dx ≤
A g j ∂ j u dx. (2.17)
`→∞ j=1 Ω j=1 Ω

In light of (2.15)–(2.17), we conclude (2.11) by showing that

N Z
X N Z
X
lim inf bj (u` ) ∂ j u` dx ≥
A g j ∂ j u dx. (2.18)
`→∞ Ω Ω
j=1 j=1

The proof of (2.18) is a bit different from the classical one in the isotropic case since in our growth
condition on A j in (1.8), we have taken the greatest exponent for |t| from the viewpoint of the anisotropic
Sobolev inequalities. Let us emphasize what is new compared with the classical proof. Let 1 ≤ j ≤ N
be arbitrary. Since u` → u a.e. in Ω and A j is a Carathéodory function, we see that

A j (x, u` , ∇u) → A j (x, u, ∇u) a.e. in Ω. (2.19)

0
The growth condition in (1.8) gives a constant C > 0 and a nonnegative function η j ∈ L p j (Ω) such that
N
 
p0j
 p0j ∗
X
|A j (x, u` , ∇u)| ≤ C η + |u` | p +

j |∂i u| pi  (2.20)
i=1

for every ` ≥ 1. Because the power of |u` | in the right-hand side of (2.20) is p∗ , the critical exponent, the
→
−
compactness of the embedding W01, p (Ω) ,→ L p (Ω) fails, in general. Hence, we cannot claim anymore
∗

0
that {|A j (x, u` , ∇u)| p j }`≥1 is uniformly integrable over Ω. Thus, we cannot apply Vitali’s theorem to
0
deduce the strong convergence of A j (x, u` , ∇u) to A j (x, u, ∇u) in L p j (Ω) as ` → ∞. However, if we fix
k ≥ 1, then by the growth condition in (1.8), we infer that
0
{|A j (x, u` , ∇u)| p j χ{|u` |≤k} }`≥1 is uniformly integrable over Ω.

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 12

Then, since χ{|u` |≤k} → χ{|u|≤k} as ` → ∞, from (2.19) and Vitali’s theorem, we get
0
A j (x, u` , ∇u) χ{|u` |≤k} → A j (x, u, ∇u) χ{|u|≤k} strongly in L p j (Ω) as ` → ∞. (2.21)

We return to the proof of (2.18) with modifications suggested by (2.21). By the Dominated
Convergence Theorem, we obtain (2.18) by showing that for every integer k ≥ 1,
N Z
X N Z
X
lim inf bj (u` ) ∂ j u` dx ≥
A g j (∂ j u) χ{|u|≤k} dx. (2.22)
`→∞ Ω Ω
j=1 j=1

Proof of (2.22). Fix an integer k ≥ 1. The coercivity condition in (1.8) yields that
N
X N
X
bj (u` ) ∂ j u` ≥
A bj (u` ) (∂ j u` ) χ{|u` |≤k} .
A (2.23)
j=1 j=1

For the right-hand side of (2.23), we use the monotonicity condition in (1.8), that is,
N
X N
X
bj (u` ) (∂ j u` ) χ{|u` |≤k} ≥
A bj (u` ) (∂ j u) χ{|u` |≤k}
A
j=1 j=1
N
(2.24)
X
+ A j (x, u` , ∇u) (∂ j u` − ∂ j u) χ{|u` |≤k} .
j=1

Let 1 ≤ j ≤ N be arbitrary. By the Dominated Convergence Theorem, we have (∂ j u) χ{|u` |≤k} →

bj (u` ) * g j (weakly) in L p0j (Ω) as
(∂ j u) χ{|u|≤k} strongly in L p j (Ω) as ` → ∞. Recall from (2.13) that A
` → ∞. Hence we have

bj (u` ) (∂ j u) χ{|u` |≤k} → g j (∂ j u)χ{|u|≤k} strongly in L1 (Ω) as ` → ∞.

A (2.25)

Since ∂ j u` * ∂ j u (weakly) in L p j (Ω) as ` → ∞, using (2.21), we gain the following

A j (x, u` , ∇u) (∂ j u` − ∂ j u) χ{|u` |≤k} → 0 strongly in L1 (Ω). (2.26)

In light of (2.25) and (2.26), we see that

N Z
X N Z
X
bj (u` ) (∂ j u) χ{|u` |≤k} +
A A j (x, u` , ∇u) (∂ j u` − ∂ j u) χ{|u` |≤k}
j=1 Ω j=1 Ω

converges as ` → ∞ to the right-hand side of (2.22). Using this convergence, jointly with the
inequalities in (2.23) and (2.24), we conclude the proof of (2.22).
As mentioned above, from (2.22) we obtain (2.18). Inequalities (2.17) and (2.18) ensure that
N Z
X N Z
X
lim bj (u` ) ∂ j u` dx =
A g j ∂ j u dx. (2.27)
`→∞ Ω Ω
j=1 j=1

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 13

It remains to establish (2.10). From (2.25)–(2.27), we get

N Z h
X i
A j (x, u` , ∇u` ) − A j (x, u` , ∇u) (∂ j u` − ∂ j u) χ{|u` |≤k} dx → 0 as ` → ∞. (2.28)
j=1 Ω

By (2.28) and the monotonicity condition in (1.8), we infer that

N h
X i
A j (x, u` , ∇u` ) − A j (x, u` , ∇u) (∂ j u` − ∂ j u) → 0 a.e in {|u` | ≤ k} as ` → ∞. (2.29)
j=1

→
−
For z, v, w ∈ W01, p (Ω) and a.e. x ∈ Ω, we define Dz (v, w)(x) as follows
N h
X i
Dz (v, w)(x) = A j (x, z(x), ∇v(x)) − A j (x, z(x), ∇w(x)) ∂ j (v − w)(x) (2.30)
j=1

(see Subsection A.1 in the Appendix). By a standard diagonal argument, we can find a subsequence of
{u` } (still denoted by {u` }) such that the convergence in (2.29) holds for every k ≥ 1. This implies that
N h
X i
Du` (u` , u) = A j (x, u` , ∇u` ) − A j (x, u` , ∇u) (∂ j u` − ∂ j u) → 0 a.e. in Ω as ` → ∞.
j=1

Thus, by Lemma A.4 in the Appendix, up to a subsequence, ∇u` → ∇u a.e. in Ω as ` → ∞. Since Φ

and A j (with 1 ≤ j ≤ N) are Carathéodory functions, we find that Θ(u
b ` ) → Θ(u) bj (u` ) → A
b and A bj (u)
a.e. in Ω as ` → ∞. Using this fact, jointly with (2.13), we obtain that µ = Θ(u)
b and g j = A bj (u) for
every 1 ≤ j ≤ N. From (2.14) we conclude that
N Z
X Z
hg, vi = bj (u) ∂ j v dx +
A Θ(u)
b v dx = hAu, vi + hPΘ (u), vi
j=1 Ω Ω

→
−
for every v ∈ W01, p (Ω). This proves that g = (A + PΘ ) u, namely, (2.10) holds.
In conclusion, by satisfying the M type condition in Definition 2.2, the operator A + PΘ turns out
to be pseudo-monotone. 
→
− →
−0
Corollary 2.8. The operator A + PΘ − B : W01, p (Ω) → W −1, p (Ω) is pseudo-monotone.
Proof. The claim follows from Lemmata 2.6 and 2.7, jointly with Proposition 2.3. 

3. Proof of the first assertion in Theorem 1.3

Here, we assume (1.2), (1.5), (1.8) and (1.9), whereas B belongs to the class BC. For every ε > 0,
we define Φε (x, t, ξ) : Ω × R × RN → R as follows
Φ(x, t, ξ)
Φε (x, t, ξ) = (3.1)
1 + ε |Φ(x, t, ξ)|

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 14

for a.e. x ∈ Ω and all (t, ξ) ∈ R × RN . For ε > 0 fixed, Φε satisfies the same properties as Φ, that is, the
sign-condition and the growth condition in (1.9). Moreover, Φε becomes a bounded function, namely,
for a.e. x ∈ Ω and every (t, ξ) ∈ R × RN ,

Φε (x, t, ξ) t ≥ 0, |Φε (x, t, ξ)| ≤ min {|Φ(x, t, ξ)|, 1/ε}. (3.2)

We consider approximate problems to (1.1) with f = 0 and Φ replaced by Φε , that is,

Auε + Φε (x, uε , ∇uε ) + Θ(x, uε , ∇uε ) = Buε in Ω,




(3.3)
uε ∈ W 1,→
 −p
0 (Ω).


→
−
As in Theorem 2.1, by a solution of (3.3), we mean a function uε ∈ W01, p (Ω) such that
N Z
X Z Z
bj (uε ) ∂ j v dx +
A Φ
bε (uε ) v dx + Θ(u
b ε ) v dx = hBuε , vi (3.4)
j=1 Ω Ω Ω

→
−
for every v ∈ W01, p (Ω), where for convenience we define

Φ
bε (uε )(x) = Φε (x, uε (x), ∇uε (x)) for a.e. x ∈ Ω.

Lemma 3.1. For every ε > 0, there exists a solution uε to (3.3). Moreover, we have:
(a) For a positive constant C, independent of ε, it holds
Z
kuε kW 1,→−p (Ω) + Φ
bε (uε ) uε dx ≤ C. (3.5)
0 Ω

→
−
(b) There exists U ∈ W01, p (Ω) such that, up to a subsequence of {uε },
→
−
uε * U (weakly) in W01, p (Ω) and uε → U a.e. in Ω as ε → 0. (3.6)

Proof. Let ε > 0 be arbitrary. From (3.2), we see that Φε + Θ satisfies the same assumptions as Θ in
Section 2. So, Theorem 2.1 applies with PΘ replaced by PΘ,ε , where
Z   →
−
hPΘ,ε (u), vi := Θ(u)
b +Φ bε (u) v dx for every u, v ∈ W 1, p (Ω).
0
Ω

→
−
This means that (3.3) admits at least a solution uε ∈ W01, p (Ω) for every ε > 0.
(a) By taking v = uε in (3.4), we derive that
Z
hAuε + PΘ (uε ) − Buε , uε i + Φ
bε (uε ) uε dx = 0. (3.7)
Ω

→
−
Moreover, since B is a bounded operator from W01, p (Ω) into its dual, it follows that for some constant
C0 > 0, we have
kBuε kW −1,→−p 0 (Ω) ≤ C0 for every ε > 0.

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 15

Using (1.5), the coercivity condition in (1.8) and Young’s inequality, we infer that for every δ > 0,
there exists a constant Cδ > 0 such that
N
X p
hAuε + PΘ (uε ) − Buε , uε i ≥ ν0 k∂ j uε kL pj j (Ω) − (C0 + CΘ ) kuε kW 1,→−p (Ω)
0
j=1
N
(3.8)
X p
≥ (ν0 − δ) k∂ j uε kL pj j (Ω) − Cδ
j=1

for every ε > 0. Thus, using (3.7) and (3.8), jointly with (3.2), we arrive at
N
X N
X Z
p p
(ν0 − δ) k∂ j uε kL pj j (Ω) ≤ (ν0 − δ) k∂ j uε kL pj j (Ω) + Φ
bε (uε ) uε dx ≤ Cδ .
j=1 j=1 Ω

By choosing δ ∈ (0, ν0 ), we readily conclude the assertion of (3.5).

→
−
(b) From (3.5) and the reflexivity of W01, p (Ω), we infer that, up to a subsequence, uε converges
→
−
weakly to some U in W01, p (Ω). Then, we conclude (3.6) by using Remark A.2 in the Appendix, which
implies that, up to a subsequence, uε → U (strongly) in Lq (Ω) if q ∈ [1, p∗ ) and uε → U a.e. in Ω as
ε → 0. 
For the remainder of this section, uε and U have the same meaning as in Lemma 3.1.

3.1. Strong convergence of T k (uε )

We recall the notation introduced in (2.30), and for simplicity, instead of Duε (T k (uε ), T k (U))(x), we
write Dε,k (x), that is,
N h
X i
Dε,k (x) = A j (x, uε , ∇T k (uε )) − A j (x, uε , ∇T k (U)) ∂ j (T k (uε ) − T k (U)). (3.9)
j=1

Lemma 3.2. There exists a subsequence of {uε }, relabeled {uε }, such that
→
−
∇uε → ∇U a.e. in Ω and T k (uε ) → T k (U) (strongly) in W01, p (Ω) as ε → 0 (3.10)

for every integer k ≥ 1.

Proof. Recall that {uε } satisfies (3.6) in Lemma 3.1. By a standard diagonal argument, it suffices to
show that for every integer k ≥ 1, there exists a subsequence {uε } (depending on k and relabeled {uε })
satisfying
→
−
∇T k (uε ) → ∇T k (U) a.e. in Ω and T k (uε ) → T k (U) (strongly) in W01, p (Ω). (3.11)

Moreover, in light of Lemma A.5 in the Appendix, we conclude (3.11) by showing that, for every
integer k ≥ 1, there exists a subsequence of {uε } (depending on k and relabeled {uε }) such that

Dε,k → 0 in L1 (Ω) as ε → 0. (3.12)

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 16

Let k ≥ 1 be fixed. Clearly, the monotonicity assumption in (1.8) yields that Dε,k ≥ 0 a.e. in Ω.
Hence, to prove (3.12), it suffices to show that (up to a subsequence of {uε }),
Z
lim sup Dε,k (x) dx ≤ 0. (3.13)
ε→0 Ω

We define zε,k as follows

zε,k := T k (uε ) − T k (U).
We observe that
∂ j zε,k χ{|uε |≥k} = −∂ j T k (U) χ{|uε |≥k} = −∂ j U χ{|uε |≥k} χ{|U|<k} .
Moreover, we see that
χ{|uε |≥k} χ{|U|<k} → 0 a.e. in Ω as ε → 0. (3.14)
By the Dominated Convergence Theorem, for every 1 ≤ j ≤ N, we have

∂ j U χ{|uε |≥k} χ{|U|<k} → 0 (strongly) in L p j (Ω) as ε → 0. (3.15)

On the other hand, from the growth condition on A j in (1.8) and the a priori estimates in Lemma 3.1,
0
we infer that {A j (x, uε , ∇T k (uε ))}ε and {A j (x, uε , ∇T k (U))}ε are bounded in L p j (Ω) and, hence, up to a
0
subsequence of {uε }, they converge weakly in L p j (Ω) for each 1 ≤ j ≤ N. This, jointly with (3.15),
gives that h i
Ξ j,ε,k (x) = A j (x, uε , ∇T k (uε )) − A j (x, uε , ∇T k (U)) ∂ j Uχ{|uε |≥k} χ{|U|<k}
converges to 0 in L1 (Ω) as ε → 0 for every 1 ≤ j ≤ N. It follows that
Z N Z
X
Dε,k (x) χ{|uε |≥k} dx = − Ξ j,ε,k (x) dx → 0 as ε → 0.
Ω j=1 Ω

Thus, to conclude (3.13), it remains to show that

Z
lim sup Dε,k (x) χ{|uε |<k} dx ≤ 0. (3.16)
ε→0 Ω

Proof of (3.16). We define ϕλ : R → R as follows

ϕλ (t) = t exp (λt2 ) for every t ∈ R.

We choose λ = λ(k) > 0 large such that 4ν02 λ > φ2 (k), where φ appears in the growth assumption on
Φ, see (1.9). This choice of λ ensures that for every t ∈ R
φ(k) 1 φ(k) 1
λt2 − |t| + > 0 and, hence, ϕ0λ (t) − |ϕλ (t)| > . (3.17)
2ν0 4 ν0 2
→
−
For v ∈ W01, p (Ω), we define
N Z
φ(k)
X " #
Eε,k (v) = A j (x, uε , ∇v)∂ j zε,k ϕ0λ (zε,k ) − |ϕλ (zε,k )| χ{|uε |<k} dx.
j=1 Ω ν0

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 17

Returning to the definition of Dε,k in (3.9) and using (3.17), we arrive at

Z
1
Dε,k (x) χ{|uε |<k} dx ≤ Eε,k (T k (uε )) − Eε,k (T k (U)). (3.18)
2 Ω

Since T k (uε ) = uε on the set {|uε | < k}, in light of (3.18), we complete the proof of (3.16) by showing
that

lim Eε,k (T k (U)) = 0, (3.19)

ε→0
lim sup Eε,k (uε ) ≤ 0. (3.20)
ε→0

Proof of (3.19). For each 1 ≤ j ≤ N, the growth condition in (1.8) gives a nonnegative function
0
F j ∈ L p j (Ω) such that on the set {|uε | < k}, we have |A j (x, uε , ∇T k (U))| ≤ F j for every ε > 0. Since
|zε,k | ≤ 2k, we can find a constant Ck > 0 such that

φ(k)
ϕ0λ (zε,k ) − |ϕλ (zε,k )| ≤ Ck .
ν0
On the other hand, for each 1 ≤ j ≤ N, we have

∂ j zε,k χ{|uε |<k} = ∂ j zε,k + ∂ j U χ{|U|<k} χ{|uε |≥k} .

This, together with (3.15) and the weak convergence of ∂ j zε,k to 0 in L p j (Ω) as ε → 0, implies that
∂ j zε,k χ{|uε |<k} converges weakly to 0 in L p j (Ω) as ε → 0. Hence, we have
N Z
X
|Eε,k (T k (U))| ≤ Ck F j |∂ j zε,k | χ{|uε |<k} dx → 0 as ε → 0,
j=1 Ω

which proves (3.19).

Proof of (3.20). From (3.6), we have
→
−
zε,k → 0 a.e. in Ω and zε,k * 0 (weakly) in W01, p (Ω) as ε → 0.
→
−
Since |zε,k | ≤ 2k a.e. in Ω, we get ϕλ (zε,k ) ∈ W01, p (Ω) ∩ L∞ (Ω). Moreover,
→
−
ϕλ (zε,k ) → 0 a.e. in Ω and ϕλ (zε,k ) * 0 (weakly) in W01, p (Ω) as ε → 0. (3.21)

Observe that uε zε,k ≥ 0 on the set {|uε | ≥ k}, which gives that

Φ
bε (uε ) ϕλ (zε,k ) χ{|uε |≥k} ≥ 0.

Thus, by testing (3.4) with v = ϕλ (zε,k ), we obtain that

Z Z
hAuε , ϕλ (zε,k )i + Φε (uε ) ϕλ (zε,k ) χ{|uε |<k} dx ≤ hBuε , ϕλ (zε,k )i −
b Θ(u
b ε ) ϕλ (zε,k ) dx. (3.22)
Ω Ω

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 18

To simplify exposition, we now introduce some notation:

Z  X 
 1 N 
Xk (ε) = φ(k) Abj (uε ) ∂ j (T k U) + c(x) |ϕλ (zε,k )| χ{|uε |<k} dx,
Ω ν0 j=1
 

XN Z
Yk (ε) = bj (uε ) ∂ j U ϕ0λ (zε,k ) χ{|U|<k} χ{|uε |≥k} dx.
A
j=1 Ω

We rewrite the first term in the left-hand side of (3.22) as follows

N Z
X
hAuε , ϕλ (zε,k )i = bj (uε )∂ j zε,k ϕ0λ (zε,k ) χ{|uε |<k} dx − Yk (ε).
A (3.23)
j=1 Ω

The coercivity condition in (1.8) and the growth condition of Φ in (1.9) imply that
 N

 1 X 
|Φ
bε (uε )| χ{|uε |<k} ≤ φ(k) 
 ν0
bj (uε )∂ j uε + c(x) χ{|uε |<k} .
A  (3.24)
j=1

In the right-hand side of (3.24) we replace ∂ j uε by ∂ j zε,k + ∂ j T k (U), then we multiply the inequality
by |ϕλ (zε,k )| and integrate over Ω with respect to x. It follows that the second term in the left-hand side
of (3.22) is at least
N Z
φ(k) X bj (uε )∂ j zε,k |ϕλ (zε,k )| χ{|uε |<k} dx − Xk (ε).
− A
ν0 j=1 Ω
Using this fact, as well as (3.23), in (3.22), we see that Eε,k (uε ) satisfies the estimate
Z
Eε,k (uε ) ≤ Xk (ε) + Yk (ε) + hBuε , ϕλ (zε,k )i − Θ(u
b ε ) ϕλ (zε,k ) dx. (3.25)
Ω

To conclude the proof of (3.20), it suffices to show that each term in the right-hand side of (3.25)
→−
converges to 0 as ε → 0. Recall that ϕλ (zε,k ) ∈ W01, p (Ω) ∩ L∞ (Ω) satisfies (3.21). Thus, using (1.5) and
the property (P2 ) of B, we get that the third, as well as the fourth, term in the right-hand side of (3.25)
converges to zero as ε → 0.
We next look at Xk (ε). In view of the pointwise convergence in (3.21) and c ∈ L1 (Ω), we infer from
the Dominated Convergence Theorem that

c(x)|ϕλ (zε,k )| χ{|uε |<k} → 0 in L1 (Ω) as ε → 0. (3.26)

Next, up to a subsequence of {uε }, we find that A bj (uε ) converges weakly in L p0j (Ω) as ε → 0 for every
→
−
1 ≤ j ≤ N using the boundedness of A bj : W 1, p (Ω) → L p0j (Ω) (see Lemma 2.4). Hence, PNj=1 A bj (uε ) ∂ j U
0
converges in L1 (Ω) as ε → 0. Then, there exists a nonnegative function F ∈ L1 (Ω) (independent of ε)
such that, up to a subsequence of {uε }, we have
N
X
bj (uε ) ∂ j U ≤ F
A a.e. in Ω for every ε > 0. (3.27)
j=1

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 19

We can now again use the Dominated Convergence Theorem to conclude that
N
X
bj (uε ) ∂ j T k (U) |ϕλ (zε,k )| χ{|uε |<k} → 0 in L1 (Ω) as ε → 0.
A (3.28)
j=1

From (3.26) and (3.28), we find that limε→0 Xk (ε) = 0. Since |ϕ0λ (zε,k )| is bounded above by a constant
independent of ε (but dependent on k), we can use a similar argument, based on (3.14) and (3.27), to
obtain that, up to a subsequence of {uε }, limε→0 Yk (ε) = 0. This ends the proof of the convergence to
zero of the right-hand side of (3.25) as ε → 0. Consequently, the proof of (3.20), and thus of (3.16), is
complete. 

3.2. Passing to the limit

From now on, the meaning of {uε }ε is given by Lemma 3.2. Using Lemma 3.1, we prove in
Lemma 3.4 that U is a solution of (1.1) with f = 0 and, moreover, U satisfies all the properties
stated in Theorem 1.3 (i). Besides (3.10), the other fundamental property that allows us to pass to the
→−
limit as ε → 0 in (3.4) for every v ∈ W01, p (Ω) ∩ L∞ (Ω) is the following convergence

Φ
bε (uε ) → Φ(U)
b (strongly) in L1 (Ω) as ε → 0. (3.29)

The proof of (3.29) is the main objective of our next result.

Lemma 3.3. We have Φ(U),

b Φ(U)
b U ∈ L1 (Ω) and (3.29) holds.

Proof. From the pointwise convergence uε → U and ∇uε → ∇U a.e. in Ω as ε → 0, jointly with
the fact that Φ(x, t, ξ) : Ω × R × RN → R is a Carathéodory function, we infer that Φ(ub ε ) → Φ(U)
b
and Φbε (uε ) uε → Φ(U)
b U a.e. in Ω as ε → 0. Using this fact and that {Φ
bε (uε ) uε }ε is a sequence
of nonnegative functions that is uniformly bounded in L (Ω) with respect to ε (from Lemma 3.1), by
1

Fatou’s Lemma we conclude that

Φ(U)
b U ∈ L1 (Ω).
This and the growth condition in (1.9)yield that Φ(U)
b ∈ L1 (Ω). Indeed, for any M > 0, on the set
Ω ∩ {|U| ≤ M}, we have |Φ(U)| ≤ φ(M) Nj=1 |∂ j U| p j + c(x) ∈ L1 (Ω). In turn, on the set Ω ∩ {|U| > M},
b P

it holds |Φ(U)|
b ≤ M −1 Φ(U)
b U ∈ L1 (Ω).
To finish the proof of Lemma 3.3, it remains to establish (3.29).
Proof of (3.29). Since Φ bε (uε ) → Φ(U)
b a.e. in Ω as ε → 0 and Φ(U)
b ∈ L1 (Ω), by Vitali’s Theorem,
it suffices to show that {Φ
bε (uε )}ε is uniformly integrable over Ω. We next check this fact. For every
M > 0, we define
Dε,M := {|uε | ≤ M} and Eε,M := {|uε | > M}.
For every x ∈ Dε,M , using the growth condition of Φ in (1.9), we find that
 N 
X 
|Φ
bε (uε )(x)| ≤ |Φ(u
b ε )(x)| ≤ φ(M)  |∂ j T M (uε )| j + c(x) ,

p

j=1

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 20

with c ∈ L1 (Ω). Let ω be any measurable subset of Ω. It follows that

Z  N Z 
X p

|Φ
bε (uε )| dx ≤ φ(M)  k∂ j (T M uε )k pj j +
L (ω)
c(x) dx  .
ω∩Dε,M ω
 
j=1

On the other hand, using (3.5) in Lemma 3.1, we see that

Z Z
1 bε (uε ) uε dx ≤ C ,
|Φε (uε )| dx ≤
b Φ
ω∩Eε,M M ω∩Eε,M M
where C > 0 is a constant independent of ε and ω. Consequently, we find that
Z  N Z 
X p
 C
|Φ
bε (uε )| dx ≤ φ(M)  k∂ j (T M uε )k p j +
j
L (ω)
c(x) dx  + .
 M (3.30)
ω ω

j=1

Lemma 3.2 yields that ∂ j T M (uε ) → ∂ j T M (U) (strongly) in L p j (Ω) as ε → 0 for every 1 ≤ j ≤ N. Since
c ∈ L1 (Ω), from (3.30) we get the uniform integrability of {Φ bε (uε )}ε over Ω. We end the proof of (3.29)
by Vitali’s Theorem. 
By Lemma 3.3, to finish the proof of Theorem 1.3 (i), we need to show the following.
Lemma 3.4. The function U is a solution to (1.1) with f = 0 and, moreover, (1.11) holds for v = u = U.
→
−
Proof. Fix v ∈ W01, p (Ω) ∩ L∞ (Ω) arbitrary. Since uε is a solution of (3.3), we have
N Z
X Z Z
bj (uε ) ∂ j v dx +
A Φ
bε (uε ) v dx + Θ(u
b ε ) v dx = hBuε , vi. (3.31)
j=1 Ω Ω Ω

By Lemma 3.3, the second term in the left-hand side of (3.31) converges to Ω Φ(U) v as ε → 0,
R
b
whereas the right-hand side of (3.31) converges to hBU, vi based on the weak convergence of uε to U
→−
in W01, p (Ω) as ε → 0. Using (3.6) and (3.10), we find that

Θ(u
b ε ) → Θ(U)
b and A bj (U) a.e. in Ω for 1 ≤ j ≤ N.
bj (uε ) → A (3.32)
Thus, in light of (1.5), and the Dominated Convergence Theorem, we obtain that
Z Z
Θ(u
b ε ) v dx → Θ(U)
b v dx as ε → 0.
Ω Ω

p0j
bj (uε )}ε is uniformly bounded in L (Ω) with respect to ε, we observe from (3.32) that (up to a
Since {A
subsequence) A bj (U) (weakly) in L p0j (Ω) as ε → 0 for each 1 ≤ j ≤ N. It follows that
bj (uε ) * A
N Z
X N Z
X
bj (uε ) ∂ j v dx →
A bj (U) ∂ j v dx as ε → 0.
A
j=1 Ω j=1 Ω

By letting ε → 0 in (3.31), we conclude that

XN Z Z Z
A j (U) ∂ j v dx +
b Φ(U) v dx +
b Θ(U)
b v dx = hBU, vi (3.33)
j=1 Ω Ω Ω

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 21
→
−
for every v ∈ W01, p (Ω) ∩ L∞ (Ω). Hence, U is a solution to (1.1) with f = 0.
It remains to prove (1.11) for v = u = U. Since U may not be in L∞ (Ω), we cannot directly use
→
−
v = U in (3.33). Nevertheless, for every k > 0, we have T k (U) ∈ W01, p (Ω) ∩ L∞ (Ω). Hence, by taking
v = T k (U) in (3.33), we have
Z Z
hAU, T k (U)i + Φ(U)
b T k (U) dx + Θ(U)
b T k (U) dx = hBU, T k (U)i. (3.34)
Ω Ω

Notice that kT k (U)kW 1,→−p (Ω) ≤ kUkW 1,→−p (Ω) for all k > 0. Moreover, ∂ j (T k (U)) → ∂ j U a.e. in Ω as k → ∞,
0 0
→
−
for every 1 ≤ j ≤ N, so that T k (U) * U (weakly) in W01, p (Ω) as k → ∞. Since AU and BU belong to
→
−0
W −1, p (Ω), it follows that

lim hAU, T k (U)i = hAU, Ui and lim hBU, T k (U)i = hBU, Ui.
k→∞ k→∞

Recalling that Φ(U)

b U ∈ L1 (Ω) and (1.5) holds, from the Dominated Convergence Theorem, we can
pass to the limit k → ∞ in (3.34) to conclude the proof. 

4. Proof of the second assertion in Theorem 1.3

Suppose for the moment only (1.2), (1.5), (1.8), and (1.9). Let B be in the class BC. Overall, to
prove Theorem 1.3 (ii), we follow similar arguments to those developed for proving Theorem 1.3 (i) in
Section 3. But there are several differences that appear when introducing a function f ∈ L1 (Ω) in the
equation in (1.1). We first approximate f by a “nice” function fε ∈ L∞ (Ω) with the properties that

| fε | ≤ | f | a.e. in Ω and fε → f a.e. in Ω as ε → 0. (4.1)

Then, by the Dominated Convergence Theorem, we find that

fε → f (strongly) in L1 (Ω) as ε → 0. (4.2)

For example, for every ε > 0, we could take fε (x) = f (x)/(1 + ε| f (x)|) for a.e. x ∈ Ω. This
approximation is done so that we can apply Theorem 1.3 (i) for the problem generated by (1.1) with fε
in place of f . Then such an approximate problem admits at least a solution Uε , namely,

AUε + Φ(Uε ) + Θ(Uε ) = BUε + fε in Ω,



 b b
(4.3)
U ∈ W 1,→

 −p
(Ω), Φ(U
b ) ∈ L1 (Ω).
ε 0 ε

→
− →
−0
To see this, we observe that Bε : W01, p (Ω) → W −1, p (Ω) belongs to the class BC, where
Z
→−
hBε u, vi = hBu, vi + fε v dx for every u, v ∈ W01, p (Ω). (4.4)
Ω

By Theorem 1.3 (i) applied for Bε instead of B, we obtain a solution Uε for (4.3). Thus,
N Z
X Z Z Z
bj (Uε ) ∂ j v dx +
A Φ(U
b ε ) v dx + Θ(U
b ε ) v dx = hBUε , vi + fε v dx (4.5)
j=1 Ω Ω Ω Ω

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 22
→
−
for every v ∈ W01, p (Ω) ∩ L∞ (Ω). However, unlike Theorem 1.3 (i), to obtain that Uε is uniformly
→
−
bounded in W01, p (Ω) with respect to ε, we need the following:
(i) B to satisfy the extra condition (P3 ), that is, B is chosen in the class BC+ ;
(ii) the additional hypothesis (1.10), which we recall below:
there exist positive constants τ and γ such that for a.e. x ∈ Ω and every ξ ∈ RN
N
X
|Φ(x, t, ξ)| ≥ γ |ξ j | p j for all |t| ≥ τ. (4.6)
j=1

Without any loss of generality, we can assume τ > 0 large such that τγ ≥ ν0 , where ν0 appears in
the coercivity condition of (1.8).
For the rest of this section, besides (1.2), (1.5), (1.8) and (1.9), we also assume (i) and (ii) above.
To avoid repetition, we understand that all the computations in Section 3 are done here replacing uε ,
U and Φε by Uε , U0 and Φ, respectively. We only stress the differences that appear compared with the
developments in Section 3.

4.1. A priori estimates

In Lemma 3.1 we gave a priori estimates for the solution uε to (3.3), corresponding to the
problem (1.1) with f = 0 and Φε instead of Φ. We next get a priori estimates for Uε solving (4.3), that
is, (1.1) with fε instead of f .
Lemma 4.1. Let Uε be a solution to (4.3).
(a) For a positive constant C, independent of ε, we have
Z
kUε kW 1,→−p (Ω) + |Φ(U
b ε )| dx ≤ C. (4.7)
0 Ω
→
−
(b) There exists U0 ∈ W01, p (Ω) such that, up to a subsequence of {Uε },
→
−
Uε * U0 (weakly) in W01, p (Ω), Uε → U0 a.e. in Ω as ε → 0. (4.8)

Proof. (a) The choice of fε gives that k fε kL1 (Ω) ≤ k f kL1 (Ω) . Let τ > 0 be as in (4.6). We have ∂ j T τ (Uε ) =
χ{|Uε |<τ} ∂ j Uε a.e. in Ω for every 1 ≤ j ≤ N. We now define
N Z
X Z
Kτ,ε := bj (Uε ) ∂ j Uε χ{|Uε |<τ} dx + τ
A |Φ(U
b ε )| χ{|Uε |≥τ} dx − hBUε , T τ (Uε )i.
j=1 Ω Ω

→
−
By taking v = T τ (Uε ) ∈ W01, p (Ω) ∩ L∞ (Ω) in (4.5) and using the sign-condition of Φ in (1.9), we obtain
that  
Kτ,ε ≤ τ k f kL1 (Ω) + CΘ meas (Ω) . (4.9)
By virtue of (4.6) and the coercivity condition in (1.8), we see that
N Z
X N Z
X
ν0 |∂ j Uε | χ{|Uε |<τ} dx + τγ
pj
|∂ j Uε | p j χ{|Uε |≥τ} dx − hBUε , T τ (Uε )i ≤ Kτ,ε .
j=1 Ω j=1 Ω

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 23

By our choice of τ, we have τγ > ν0 . Then, the above estimates lead to

XN Z  
ν0 |∂ j Uε | p j dx − hBUε , T τ (Uε )i ≤ τ k f kL1 (Ω) + CΘ meas (Ω) .
j=1 Ω

→
−
This fact, jointly with the property (P3 ), gives the boundedness of {Uε }ε>0 in W01, p (Ω). Since B is a
→
−
bounded operator from W01, p (Ω) into its dual, we have |hBUε , T τ (Uε )| ≤ C1 , where C1 is a positive
constant independent of ε. Using (4.9), we find that
Z
|Φ(U
b ε )| χ{|Uε |≥τ} dx ≤ C1 τ−1 + k f kL1 (Ω) + CΘ meas (Ω) = C2 . (4.10)
Ω

Now, using the growth condition of Φ in (1.9), we obtain a positive constant C3 such that
|Φ(Uε )| χ{|Uε |≤τ} dx ≤ C3 for every ε > 0. This completes the proof of (4.7).
R
b
Ω
(b) The assertion in (4.8) follows from (4.7) (see the proof of (b) in Lemma 3.1). 

4.2. Strong convergence of T k (Uε )

The game plan is closely related to that in Subsection 3.1. As mentioned before, when adapting
the calculations, we need to replace uε , U and B in Section 3 by Uε , U0 and Bε , respectively. The
counterpart of Lemma 3.2 holds so that we obtain the following.
Lemma 4.2. There exists a subsequence of {Uε }ε , relabeled {Uε }ε , such that
→
−
∇Uε → ∇U0 a.e. in Ω and T k (Uε ) → T k (U0 ) (strongly) in W01, p (Ω) as ε → 0
for every positive integer k.
Proof. The computations in Subsection 3.1 can be carried out with Φ instead of Φε since the upper
bounds used for |Φε | were derived from those satisfied by |Φ| and the sign-condition of Φ is the same
as for Φε (see (3.2)). A small change arises in the proof of (3.20) because of the introduction of fε
in (4.3). Using the definition of Bε in (4.4), the inequalities in (3.22) and (3.25) must R be read with
Bε instead of B. We note that hBε Uε , ϕλ (zε,k )i is the sum between hBUε , ϕλ (zε,k )i and Ω fε ϕλ (zε,k ) dx.
The latter Rterm, like the former, converges to 0 as ε → 0. The new claim regarding the convergence
to zero of Ω fε ϕλ (zε,k ) dx follows from the Dominated Convergence Theorem using (4.1), |ϕλ (zε,k )| ≤
2k exp (4λk2 ) and ϕλ (zε,k ) → 0 a.e. in Ω as ε → 0. The remainder of the proof of (3.20) carries over
easily to our setting. 

4.3. Passing to the limit

We aim to pass to the limit as ε → 0 in (4.5) to obtain that U0 is a solution to (1.1). Since fε
→
−
satisfies (4.2) and Uε * U0 (weakly) in W01, p (Ω) as ε → 0, we readily have the convergence of the
→
−
right-hand side of (4.5) to hBU0 , vi + Ω f v dx for every v ∈ W01, p (Ω) ∩ L∞ (Ω). Moreover, because of
R

the convergence ∇Uε → ∇U0 a.e. in Ω, we can use the same argument as in Lemma 3.4 to deduce that,
as ε → 0,
Z Z XN Z N Z
X
Θ(uε ) v dx →
b Θ(U0 ) v dx,
b A j (Uε ) ∂ j v dx →
b bj (U0 ) ∂ j v dx
A
Ω Ω j=1 Ω j=1 Ω

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 24
→
−
for every v ∈ W01, p (Ω). What is here different compared with Subsection 3.2 is the proof of the
convergence
Φ(U
b ε ) → Φ(U
b 0 ) (strongly) in L1 (Ω) as ε → 0. (4.11)
To prove that U0 is a solution of (1.1), it remains to justify (4.11). Instead of Lemma 3.3, we establish
the following.
Lemma 4.3. We have Φ(U b 0 ) ∈ L1 (Ω) and (4.11) holds.
Proof. From Lemma 4.2, the pointwise convergence in (4.8) and the continuity of Φ(x, ·, ·) in the last
two variables, we infer that |Φ(U
b ε )| → |Φ(U b 0 )| a.e. in Ω as ε → 0. Then, (4.7) and Fatou’s Lemma
ensure that Φ(U0 ) ∈ L (Ω).
b 1

Proof of (4.11). We will use Vitali’s Theorem. To this end, we need to show that {Φ(U b ε )}ε is uniformly
integrable over Ω. We can only partially imitate the proof of the uniform integrability of {Φ bε (uε )}ε in
Lemma 3.3. Fix M > 1 arbitrary. For any measurable subset ω of Ω, using the growth condition of Φ
in (1.9), we find that
Z  N 
X 
b ε )| χ{|Uε |≤M} dx ≤ φ(M)  k∂ j T M (Uε )k p pj j + kckL1 (ω)  .
|Φ(U (4.12)
L (ω)
ω
 
j=1

Since ∂ j T M (Uε ) → ∂ j T M (U0 ) (strongly) in L (Ω) as ε → 0 for every 1 ≤ j ≤ N and c ∈ L1 (Ω), we see
pj

that the right-hand side of (4.12) is as small as desired uniformly in ε when the measure of ω is small.
We next bound from above ω |Φ(U b ε )| χ{|Uε |>M} dx. This is where the modification appears since we
R

don’t have anymore that {Φ(U b ε ) Uε }ε is uniformly bounded in L1 (Ω) with respect to ε. We adapt an
approach from [13]. In (4.5) we take
→
−
v = T 1 (G M−1 (Uε )) ∈ W01, p (Ω) ∩ L∞ (Ω).
Then, using (1.5), the coercivity condition in (1.8) and the sign-condition of Φ in (1.9), we obtain the
estimate
Z Z
|Φ(Uε )|χ{|Uε |>M} dx ≤ (| fε | + CΘ ) χ{|Uε |≥M−1} dx + |hBUε , T 1 (G M−1 (Uε ))i|.
b (4.13)
Ω Ω
Now, up to a subsequence of {Uε }, from (4.8), we have
→
−
T 1 (G M−1 (Uε )) * T 1 (G M−1 (U0 )) (weakly) in W01, p (Ω) as ε → 0.
Using this in (4.13), jointly with (4.1) and the property (P2 ) for B, we find that
Z Z
lim sup |Φ(U
b ε )|χ{|Uε |>M} dx ≤ (| f | + CΘ ) χ{|U0 |≥M−1} dx + |hBU0 , T 1 (G M−1 (U0 ))i|.
ε→0 Ω Ω

Recall that f ∈ L (Ω). Since ∂ j T 1 (G M−1 (U0 )) = χ{M−1<|U0 |<M} ∂ j U0 a.e. in Ω for every 1 ≤ j ≤ N, from
1

the above inequality, we infer that Z

|Φ(U
b ε )| χ{|Uε |>M} dx
ω
is small, uniformly in ε and ω, when M is sufficiently large. Thus, using the comments after (4.12), we
conclude the uniform integrability of {Φ(U
b ε )}ε over Ω. The proof of Lemma 4.3 is complete. 
By letting ε → 0 in (4.5), we conclude that U0 is a solution of (1.1). This ends the proof of
Theorem 1.3 (ii). 

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 25

5. Strong convergence of uε in Theorem 1.3 (i)

We show that in the setting of Theorem 1.3 (i), up to a subsequence of {uε }, not only the assertions
of Lemma 3.2 hold, but also the strong convergence in (1.13), that is
→
−
uε → U (strongly) in W01, p (Ω) as ε → 0. (5.1)

Lemma 5.1. Up to a subsequence of {uε }ε , relabeled {uε }ε , we have (5.1).

Proof. For every k ≥ 1, we define
h i
Lk := ν0−1 |hBU, Gk (U)i| + CΘ kGk (U)kL1 (Ω) . (5.2)

We next show that, up to a subsequence of {uε }, we have

N
X 1/p j
lim sup kGk (uε )kW 1,→−p (Ω) ≤ Lk . (5.3)
ε→0 0
j=1

Proof of (5.3). Let k ≥ 1 be a fixed integer. Since Gk (uε ) = uε − T k (uε ) and ∂ j T k (uε ) = ∂ j uε χ{|uε |<k} for
every 1 ≤ j ≤ N, from the coercivity assumption in (1.8), we see that
N Z
X
hAuε , Gk (uε )i = bj (uε ) ∂ j uε dx
A
j=1 {|uε |>k}

N Z
X N
X p
≥ ν0 |∂ j uε | dx = ν0
pj
k∂ jGk (uε )kL pj j (Ω) .
j=1 {|uε |>k} j=1

Using (3.2) and t Gk (t) ≥ 0 for every t ∈ R, we observe that Gk (t) Φ bε (t) ≥ 0 for all t ∈ R. Then, by
testing (3.4) with v = Gk (uε ) and using (1.5), we find that
Z
hAuε , Gk (uε )i ≤ hAuε , Gk (uε )i + Gk (uε ) Φ
bε (uε ) dx
Ω
Z
≤ |hBuε , Gk (uε )i| + CΘ |Gk (uε )| dx.
Ω

→
−
From (3.6), the boundedness of {uε } in W01, p (Ω) and Remark A.2, we can pass to a subsequence of {uε }
(relabeled {uε }) such that, as ε → 0, we have
→
−
T k (uε ) → T k (U) a.e. in Ω and T k (uε ) * T k (U) (weakly) in W01, p (Ω),
→
−
Gk (uε ) → Gk (U) a.e. in Ω and Gk (uε ) * Gk (U) (weakly) in W01, p (Ω),
Gk (uε ) → Gk (U) strongly in Lr (Ω) with 1 ≤ r < p∗ .

Hence, using the property (P2 ), we derive that

limhBuε , Gk (uε )i = hBU, Gk (U)i and lim kGk (uε )kL1 (Ω) = kGk (U)kL1 (Ω) .
ε→0 ε→0

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 26

Consequently, for every 1 ≤ j ≤ N, we have

 h i1/p j 1/p
lim sup k∂ j (Gk (uε ))kL p j (Ω) ≤ ν0−1 |hBU, Gk (U)i| + CΘ kGk (U)kL1 (Ω) = Lk j .
ε→0

This establishes the inequality in (5.3).

Recall that {uε }ε stands for a sequence {uε` }`≥1 with ε` & 0 as ` → ∞. By Lemma 3.1 and (5.3), as
well as from the proof of Lemma 3.2, we get that for any given integer k ≥ 1, there exists a subsequence
of {uε }ε that depends on k, say {u(k) (k)
ε` }`≥1 , for which (5.3) and (3.11) hold with uε` in place of {uε }. This
means that
N
X 1/p j
lim sup kGk (u(k)
ε` )kW 1, p (Ω)
→− ≤ Lk , ε` ) − T k (U)kW 1, p (Ω) = 0.
lim kT k (u(k) →− (5.4)
`→∞ 0 `→∞ 0
j=1

We proceed inductively with respect to k, at each step (k + 1) selecting the subsequence {u(k+1) ε` }`≥1 from
{uε` }`≥1 , the subsequence of {uε } with the properties in (5.4). Then, {uε` }`≥k is a subsequence of {u(ε`j) }`≥1
(k) (`)

for every 1 ≤ j ≤ k. Hence, by a standard diagonal argument, there exists a subsequence of {uε }ε , that
is, {u(`)
ε` }` , relabeled {uε }ε , such that (5.3) and (3.11) hold for every k ≥ 1, namely

N
X 1/p j
lim sup kGk (uε )kW 1,→−p (Ω) ≤ Lk , lim kT k (uε ) − T k (U)kW 1,→−p (Ω) = 0. (5.5)
ε→0 0 ε→0 0
j=1

→
−
Using the weak convergence of Gk (uε ) to Gk (U) in W01, p (Ω) as ε → 0, we see that
N
X 1/p j
kGk (U)kW 1,→−p (Ω) ≤ lim inf kGk (uε )kW 1,→−p (Ω) ≤ Lk . (5.6)
0 ε→0 0
j=1

We now complete the proof of (5.1). From the definition of Gk in (1.17), we find that

kuε − UkW 1,→−p (Ω) ≤ kGk (uε )kW 1,→−p (Ω) + kGk (U)kW 1,→−p (Ω) + kT k (uε ) − T k (U)kW 1,→−p (Ω) .
0 0 0 0

Then, in view of (5.5) and (5.6), for every k ≥ 1, we obtain that

N
X 1/p j
lim sup kuε − UkW 1,→−p (Ω) ≤ 2 Lk . (5.7)
ε→0 0
j=1

→
−
Remark that Lk (defined in (5.2)) converges to 0 as k → ∞ since Gk (U) * 0 (weakly) in W01, p (Ω) and
Gk (U) → 0 (strongly) in L1 (Ω) as k → ∞. Hence, by letting k → ∞ in (5.7), we obtain (5.1). 

Acknowledgments

The first author has been supported by the Sydney Mathematical Research Institute via the
International Visitor Program (August–September 2019) and by Programma di Scambi Internazionali
dell’Università degli Studi di Napoli Federico II. The research of the second author is supported by the
Australian Research Council under the Discovery Project Scheme (DP190102948 and DP220101816).
The authors would like to thank the referees for helping improve the presentation.

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 27

Conflict of interest

The authors declare no conflict of interest.

References

1. A. Alberico, I. Chlebicka, A. Cianchi, A. Zatorska-Goldstein, Fully anisotropic elliptic problems

with minimally integrable data, Calc. Var., 58 (2019), 186. https://doi.org/10.1007/s00526-019-
1627-8
2. A. Alberico, G. di Blasio, F. Feo, Comparison results for nonlinear anisotropic parabolic
problems, Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur., 28 (2017), 305–322.
https://doi.org/10.4171/RLM/764
3. A. Alvino, V. Ferone, A. Mercaldo, Sharp a priori estimates for a class of nonlinear
elliptic equations with lower order terms, Annali di Matematica, 194 (2015), 1169–1201.
https://doi.org/10.1007/s10231-014-0416-4
4. A. Alvino, A. Mercaldo, Nonlinear elliptic equations with lower order terms and symmetrization
methods, Boll. Unione Mat. Ital., 1 (2008), 645–661.
5. S. N. Antontsev, M. Chipot, Anisotropic equations: uniqueness and existence results, Differential
Integral Equations, 21 (2008), 401–419. https://doi.org/10.57262/die/1356038624
6. S. N. Antontsev, J. I. Dı́az, S. Shmarev, Energy methods for free boundary problems, Boston, MA:
Birkhäuser Boston, Inc., 2002. https://doi.org/10.1007/978-1-4612-0091-8
7. S. Antontsev, S. Shmarev, Chapter 1 Elliptic equations with anisotropic nonlinearity and
nonstandard growth conditions, In: Handbook of differential equations: stationary partial
differential equations, North-Holland, 2006. https://doi.org/10.1016/S1874-5733(06)80005-7
8. P. Baroni, A. Di Castro, G. Palatucci, Intrinsic geometry and De Giorgi classes
for certain anisotropic problems, Discrete Contin. Dyn. Syst. S, 10 (2017), 647–659.
https://doi.org/10.3934/dcdss.2017032
9. M. Bendahmane, K. H. Karlsen, Nonlinear anisotropic elliptic and parabolic equations in RN with
advection and lower order terms and locally integrable data, Potential Anal., 22 (2005), 207–227.
https://doi.org/10.1007/s11118-004-6117-7
10. A. Bensoussan, L. Boccardo, Nonlinear systems of elliptic equations with natural
growth conditions and sign conditions, Appl. Math. Optim., 46 (2002), 143–166.
https://doi.org/10.1007/s00245-002-0753-3
11. A. Bensoussan, L. Boccardo, F. Murat, On a nonlinear partial differential equation having natural
growth terms and unbounded solution, Ann. Inst. H. Poincaré Anal. Non Linéaire, 5 (1988), 347–
364. https://doi.org/10.1016/S0294-1449(16)30342-0
12. M. F. Betta, A. Mercaldo, F. Murat, M. M. Porzio, Existence of renormalized solutions to nonlinear
elliptic equations with a lower-order term and right-hand side a measure, J. Math. Pure. Appl., 82
(2003), 90–124. https://doi.org/10.1016/S0021-7824(03)00006-0

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 28

13. L. Boccardo, T. Gallouët, F. Murat, A unified presentation of two existence results for problems
with natural growth, In: Progress in partial differential equations: the Metz surveys, 2 (1992),
Harlow: Longman Sci. Tech., 1993, 127–137.
14. L. Boccardo, P. Marcellini, C. Sbordone, L∞ -regularity for variational problems with sharp
nonstandard growth conditions, Bollettino U. M. I., 4 (1990), 219–225.
15. L. Boccardo, F. Murat, J.-P. Puel, Existence of bounded solutions for nonlinear elliptic unilateral
problems, Annali di Matematica, 152 (1988), 183–196. https://doi.org/10.1007/BF01766148
16. L. Boccardo, T. Gallouët, Strongly nonlinear elliptic equations having natural growth terms and L1
data, Nonlinear Anal., 19 (1992), 573–579. https://doi.org/10.1016/0362-546X(92)90022-7
17. M.-M. Boureanu, A new class of nonhomogeneous differential operators and
applications to anisotropic systems, Complex Var. Elliptic Equ., 61 (2016), 712–730.
https://doi.org/10.1080/17476933.2015.1114614
18. M.-M. Boureanu, A. Vélez-Santiago, Fine regularity for elliptic and parabolic anisotropic
Robin problems with variable exponents, J. Differ. Equations, 266 (2019), 8164–8232.
https://doi.org/10.1016/j.jde.2018.12.026
19. B. Brandolini, F. C. Cı̂rstea, Singular anisotropic elliptic equations with gradient-dependent lower
order terms, arXiv:2001.02887.
20. H. Brézis, Équations et inéquations non linéaires dans les espaces vectoriels en dualité, Ann. Inst.
Fourier, 18 (1968), 115–175.
21. F. E. Browder, Existence theorems for nonlinear partial differential equations, In: Global analysis,
Providence, R.I.: Amer. Math. Soc., 1970, 1–60.
22. A. Cianchi, Symmetrization in anisotropic elliptic problems, Commun. Part. Diff. Eq., 32 (2007),
693–717. https://doi.org/10.1080/03605300600634973
23. F. C. Cı̂rstea, J. Vétois, Fundamental solutions for anisotropic elliptic equations:
existence and a priori estimates, Commun. Part. Diff. Eq., 40 (2015), 727–765.
https://doi.org/10.1080/03605302.2014.969374
24. G. di Blasio, F. Feo, G. Zecca, Regularity results for local solutions to some anisotropic elliptic
equations, Isr. J. Math., in press.
25. A. Di Castro, E. Montefusco, Nonlinear eigenvalues for anisotropic quasilinear degenerate elliptic
equations, Nonlinear Anal., 70 (2009), 4093–4105. https://doi.org/10.1016/j.na.2008.06.001
26. X. Fan, Anisotropic variable exponent Sobolev spaces and → −p (·)−Laplacian equations, Complex
Var. Elliptic Equ., 56 (2011), 623–642. https://doi.org/10.1080/17476931003728412
27. X. Fan, Local boundedness of quasi-minimizers of integral functions with variable exponent
anisotropic growth and applications, Nonlinear Differ. Equ. Appl., 17 (2010), 619–637.
https://doi.org/10.1007/s00030-010-0072-3
28. F. Feo, J. L. Vazquez, B. Volzone, Anisotropic p-Laplacian evolution of fast diffusion type, Adv.
Nonlinear Stud., 21 (2021), 523–555. https://doi.org/10.1515/ans-2021-2136
29. V. Ferone, B. Messano, Comparison and existence results for classes of nonlinear elliptic
equations with general growth in the gradient, Adv. Nonlinear Stud., 7 (2007), 31–46.
https://doi.org/10.1515/ans-2007-0102

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 29

30. V. Ferone, F. Murat, Nonlinear elliptic equations with natural growth in the gradient
and source terms in Lorentz spaces, J. Differ. Equations, 256 (2014), 577–608.
https://doi.org/10.1016/j.jde.2013.09.013
31. I. Fragalà, F. Gazzola, B. Kawohl, Existence and nonexistence results for anisotropic
quasilinear elliptic equations, Ann. Inst. H. Poincaré Anal. Non Linéaire, 21 (2004), 715–734.
https://doi.org/10.1016/j.anihpc.2003.12.001
32. I. Fragalà, F. Gazzola, G. Lieberman, Regularity and nonexistence results for anisotropic
quasilinear elliptic equations in convex domains, Discrete Contin. Dyn. Syst., 2005 (2005), 280–
286. https://doi.org/10.3934/proc.2005.2005.280
33. H. Gao, F. Leonetti, W. Ren, Regularity for anisotropic elliptic equations with degenerate
coercivity, Nonlinear Anal., 187 (2019), 493–505. https://doi.org/10.1016/j.na.2019.06.017
34. D. Gilbarg, N. S. Trudinger, Elliptic partial differential equations of second order, Berlin: Springer,
2001. https://doi.org/10.1007/978-3-642-61798-0
35. N. Grenon, F. Murat, A. Porretta, A priori estimates and existence for elliptic equations
with gradient dependent terms, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5), 13 (2014), 137–205.
ttps://doi.org/10.2422/2036-2145.201106 012
36. H. Le Dret, Nonlinear elliptic partial differential equations. An introduction, Cham: Springer,
2018. https://doi.org/10.1007/978-3-319-78390-1
37. J. Leray, J.-L. Lions, Quelques résultats de Višik sur les problèmes elliptiques nonlinéaires par les
méthodes de Minty-Browder, Bull. Soc. Math. France, 93 (1965), 97–107.
38. P. Marcellini, Regularity of minimizers of integrals of the calculus of variations with
nonstandard growth conditions, Arch. Rational Mech. Anal., 105 (1989), 267–284.
https://doi.org/10.1007/BF00251503
39. R. E. Showalter, Monotone operators in Banach space and nonlinear partial differential equations,
Providence, RI: American Mathematical Society, 1997. http://doi.org/10.1090/surv/049
40. M. Troisi, Teoremi di inclusione per spazi di Sobolev non isotropi, Ricerche Mat., 18 (1969), 3–24.
41. N. S. Trudinger, Recent developments in elliptic partial differential equations of Monge–Ampère
type, In: International Congress of Mathematicians. Vol. III, Zürich: Eur. Math. Soc., 2006, 291–
301. https://doi.org/10.4171/022-3/15
42. N. S. Trudinger, From optimal transportation to conformal geometry, In: Geometric analysis,
Cham: Birkhäuser, 2020, 511–520. https://doi.org/10.1007/978-3-030-34953-0 20
43. J. Vétois, Strong maximum principles for anisotropic elliptic and parabolic equations, Adv.
Nonlinear Stud., 12 (2012), 101–114. https://doi.org/10.1515/ans-2012-0106
44. E. Zeidler, Nonlinear functional analysis and its applications, New York: Springer, 1990.
https://doi.org/10.1007/978-1-4612-0981-2

A. Appendix

In this section, we prove some convergence results that have been used in Sections 2 and 3,
respectively. We assume (1.2) and (1.8).

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 30

We first recall an anisotropic Sobolev inequality for the case p < N, see [40].
Lemma A.1. Let N ≥ 2 be an integer. If (1.2) holds, then there exists a constant S = S(N,→
−p ) > 0
such that
N
Y
kukL p∗ (RN ) ≤ S k∂ j uk1/N
L p j (RN )
for all u ∈ Cc∞ (RN ).
j=1

Remark A.2. Let Ω be a bounded, open subset of RN (N ≥ 2). If (1.2) holds, then using a density
argument and the arithmetic-geometric mean inequality, we find that
N
Y S →
−
kukL p∗ (Ω) ≤ S k∂ j uk1/N
L p j (Ω)
≤ kuk 1,→−p for all u ∈ W01, p (Ω). (A.1)
j=1
N W0 (Ω)
→
−
Moreover, by Hölder’s inequality, the embedding W01, p (Ω) ,→ L s (Ω) is continuous for every s ∈ [1, p∗ ]
and compact for every s ∈ [1, p∗ ).
Remark A.3. Note that if Ω ⊂ RN is an open bounded domain with Lipschitz boundary and (1.2) holds,
then the “true” critical exponent is p∞ , the maximum between p∗ and pN . Indeed, Fragalà, Gazzola
→
−
and Kawohl [31] showed that the embedding W01, p (Ω) ,→ Lr (Ω) is continuous for every r ∈ [1, p∞ ]
and compact if r ∈ [1, p∞ ).

A.1. Notation
→
−
For v, w and {uε }ε in W01, p (Ω) and for a.e. x ∈ Ω, we define
N h
X i
Duε (v, w)(x) = A j (x, uε (x), ∇v(x)) − A j (x, uε (x), ∇w(x)) ∂ j (v − w)(x),
j=1
N
(A.2)
X
Huε (v, w)(x) = A j (x, uε (x), ∇v(x)) ∂ j w(x).
j=1

Hence, Duε (v, w) = Huε (v, v) − Huε (v, w) − Huε (w, v) + Huε (w, w). The monotonicity assumption in (1.8)
gives that Duε (v, w) ≥ 0 a.e. in Ω, whereas the coercivity condition in (1.8) yields that Huε (v, v) ≥
ν0 Nj=1 |∂ j v| p j , where ν0 > 0. We thus find that
P

N
X
Duε (v, w) ≥ ν0 |∂ j v| p j − |Huε (v, w)| − |Huε (w, v)|. (A.3)
j=1

Here, we establish Lemma A.4, which is invoked in the proof of Lemma 2.7. Further, we prove
Lemma A.5, which is useful in the proof of Theorem 1.3 (i) in Section 3. To prove Lemmata A.4 and
A.5, we adapt an argument from [15, Lemma 5], the proof of which goes back to Browder [21].
As previously often recalled, by Remark A.2, whenever
→
−
uε * u (weakly) in W01, p (Ω) as ε → 0, (A.4)
we can pass to a subsequence (always relabeled {uε }) such that
uε → u strongly in Lr (Ω) if r ∈ [1, p∗ ) and uε → u a.e. in Ω. (A.5)

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 31

A.2. Some convergence results

→
−
Lemma A.4. Let u, {uε }ε be in W01, p (Ω) such that (A.4) holds. If Duε (uε , u) → 0 a.e. in Ω as ε → 0,
then, up to a subsequence, ∇uε → ∇u a.e. in Ω as ε → 0.
Proof. Let Z be a subset of Ω with meas (Z) = 0 such that for every x ∈ Ω \ Z, we have |u(x)| < ∞,
|∇u(x)| < ∞, |η j (x)| < ∞ for all 1 ≤ j ≤ N, as well as

uε (x) → u(x), Duε (uε , u)(x) → 0 as ε → 0, (A.6)

where η j are the functions appearing in the growth condition in (1.8). For every x ∈ Ω \ Z, we claim
that
{|∇uε (x)|}ε is uniformly bounded with respect to ε. (A.7)
Proof of (A.7). We fix x ∈ Ω \ Z. In view of (A.3), we have
N
X
Duε (uε , u)(x) ≥ ν0 |∂ j uε (x)| p j − |Huε (uε , u)(x)| − |Huε (u, uε )(x)|. (A.8)
j=1

By Young’s inequality, for every δ > 0, there exists Cδ > 0 such that
N 
X 0 
|Huε (uε , u)(x)| ≤ δ |A j (x, uε , ∇uε )| p j + Cδ |∂ j u(x)| p j ,
j=1
N 
(A.9)
X 0
|Huε (u, uε )(x)| ≤ δ |∂ j uε (x)| p j + Cδ |A j (x, uε , ∇u)| p j .
j=1

We use the growth condition in (1.8) to bound from above the right-hand side of each inequality
in (A.9). Then, from (A.8), there exist positive constants C and Cbδ , both independent of ε (with Cbδ
depending on δ), such that
N
X
Duε (uε , u)(x) ≥ (ν0 − C δ) |∂ j uε (x)| p j − Cbδ guε (u)(x), (A.10)
j=1

p0
where guε (u)(x) = Nj=1 η j j (x) + |uε (x)| p + Nj=1 |∂ j u(x)| p j . Using (A.6) and choosing δ ∈ (0, ν0 /C), from
P ∗ P
(A.10) we conclude (A.7).
Proof of Lemma A.4 concluded. Let x ∈ Ω \ Z be arbitrary. Define ξε = ∇uε (x) and ξ = ∇u(x). To
show that ξε → ξ as ε → 0, it is enough to prove that any accumulation point of ξε , say ξ∗ , coincides
with ξ. From (A.7), we have |ξ∗ | < ∞. By (A.6) and the continuity of A j (x, ·, ·) with respect to the last
two variables, we find that
N h
X i
Duε (uε , u)(x) → A j (x, u(x), ξ∗ ) − A j (x, u(x), ξ) (ξ∗j − ξ j ) as ε → 0.
j=1

This, jointly with (A.6) and the monotonicity condition in (1.8), gives that ξ∗ = ξ. This ends the proof
since x ∈ Ω \ Z is arbitrary and meas (Z) = 0. 

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 32
→
−
Lemma A.5. Let k ≥ 1 be a fixed integer. Let u, {uε }ε be in W01, p (Ω) such that (A.4) holds. Suppose
that, up to a subsequence of {uε } (depending on k and relabeled {uε })

Duε (T k (uε ), T k (u)) → 0 in L1 (Ω) as ε → 0. (A.11)

Then, up to a subsequence of {uε }, as ε → 0, we have

∇T k (uε ) → ∇T k (u) a.e. in Ω, (A.12)

→−
T k (uε ) → T k (u) (strongly) in W01, p (Ω). (A.13)

Proof. By (A.4) and (A.11), up to a subsequence of {uε }, we have (A.5), as well as Duε (T k (uε ), T k (u)) →
0 a.e. in Ω as ε → 0. Let Z be a subset of Ω as in the proof of Lemma A.4, where Duε (T k (uε ), T k (u)))
replaces Duε (uε , u). We follow the same argument as in Lemma A.4 with the obvious modifications
suggested by the above replacement. Then, for every x ∈ Ω \ Z, we obtain
N
X
Duε (T k (uε ), T k (u))(x) ≥ν0 |∂ j T k (uε )(x)| p j
j=1 (A.14)
− |Huε (T k (uε ), T k (u))(x)| − |Huε (T k (u), T k (uε ))(x)|.
This leads to {|∇T k (uε )(x)|}ε being uniformly bounded with respect to ε and we also obtain (A.12).
We conclude the proof of Lemma A.5 by showing (A.13). From (A.12), we see that {|∂ j T k (uε ) −
∂ j T k (u)| p j }ε is a sequence of nonnegative integrable functions, converging to 0 a.e. on Ω. Thus, by
Vitali’s Theorem, we obtain that ∂ j T k (uε ) → ∂ j T k (u) in L p j (Ω) as ε → 0 for every 1 ≤ j ≤ N by
proving that  N 

X 
is uniformly integrable over Ω.
 
pj

 |∂ j T k (u ε )| 
 (A.15)
 
j=1 ε
The claim of (A.15) follows from (A.11) and (A.14) whenever {Huε (T k (uε ), T k (u))}ε and
{Huε (T k (u), T k (uε ))}ε converge in L1 (Ω) as ε → 0. We next establish that
N
X
Huε (T k (uε ), T k (u)) → A j (x, u, ∇T k (u)) ∂ j T k (u) in L1 (Ω) as ε → 0,
j=1
N
(A.16)
X
Huε (T k (u), T k (uε )) → A j (x, u, ∇T k (u)) ∂ j T k (u) in L1 (Ω) as ε → 0.
j=1

0
Proof of (A.16). Let 1 ≤ j ≤ N be arbitrary. We see that {A j (x, uε , ∇T k (uε ))}ε is bounded in L p j (Ω)
→
− ∗
from the growth condition in (1.8) and the boundedness of {uε }ε in W01, p (Ω) and, hence, in L p (Ω).
Moreover, A j (x, uε , ∇T k (uε )) → A j (x, u, ∇T k (u)) a.e. in Ω as ε → 0 using (A.12), the convergence
uε → u a.e. in Ω (from (A.5)) and the continuity of A j (x, ·, ·) in the last two variables. Thus, up to a
0
subsequence of {uε }, we infer that A j (x, uε , ∇T k (uε )) * A j (x, u, ∇T k (u)) (weakly) in L p j (Ω) as ε → 0.
This proves the first convergence in (A.16). We now prove the second one.
Using (A.12) and the continuity properties of A j , as ε → 0,

A j (x, uε , ∇T k (u)) ∂ j T k (uε ) → A j (x, u, ∇T k (u)) ∂ j T k (u) a.e. in Ω (A.17)

Mathematics in Engineering Volume 5, Issue 4, 1–33.

 33
0
for each 1 ≤ j ≤ N. Observe that {χ{|uε |<k} |A j (x, uε , ∇T k (u))| p j }ε is uniformly integrable over Ω (from the
growth condition of A j in (1.8)) and ∂ j T k (uε ) = χ{|uε |<k} ∂ j uε . Thus, since {∂ j uε }ε is bounded in L p j (Ω),
it follows from Hölder’s inequality that {A j (x, uε , ∇T k (u)) ∂ j T k (uε )}ε is uniformly integrable over Ω for
each 1 ≤ j ≤ N. From (A.17) and Vitali’s Theorem, we reach the claim of (A.16). 
From Lemma A.5 and a standard diagonal argument, we obtain the following.
Corollary A.6. Let (A.4) and (A.11) hold. Then, there exists a subsequence of {uε }ε , relabeled {uε }ε ,
→−
such that ∇uε → ∇u a.e. in Ω and T k (uε ) → T k (u) (strongly) in W01, p (Ω) as ε → 0 for every integer
k ≥ 1.

c 2023 The Author(s), licensee AIMS Press. This

is an open access article distributed under the
terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0)

Mathematics in Engineering Volume 5, Issue 4, 1–33.