You are on page 1of 26

Regularity results for local solutions

to some anisotropic elliptic equations


G. di Blasio∗– F. Feo†– G. Zecca ‡
arXiv:2011.13412v2 [math.AP] 17 Jun 2021

Abstract
In this paper we study the higher integrability of local solutions for a class of
anisotropic equations with lower order terms whose growth coefficients lay in Marcinkiewicz
spaces. A condition for the boundedness of such solutions is also given.

Mathematics Subject Classifications: 35J60, 35B65


Key words: Anisotropic nonlinear equations, higher integrability, boundedness of solutions

1 Introduction
Our aim is to obtain regularity results for the following class of anisotropic elliptic equa-
tions
XN N
X
∂xi (Ai (x, ∇u) + Bi (x, u)) = ∂xi (|Fi |pi −2 Fi ) in Ω, (1.1)
i=1 i=1

where Ω is a domain of RN ,
N > 2, pi > 1 for every i = 1, ..., N with 1 < p̄ < N , denoting
by p the harmonic mean of p1 , · · · , pN , i.e.
N
1 1 X 1
= . (1.2)
p N pi
i=1

Throughout this paper, we make the following assumptions for any i = 1, ..., N

(H1) Ai : Ω × RN → R is a Carathéodory function that satisfies

|Ai (x, ξ)| 6 βi |ξi |pi −1 (1.3)


N
X N
X
α |ξi |pi 6 (Ai (x, ξ) ξi ) (1.4)
i=1 i=1

Dipartimento di Matematica e Fisica, Università degli Studi della Campania “L. Vanvitelli”, Viale
Lincoln, 5 - 81100 Caserta, Italy. E–mail: giuseppina.diblasio@unicampania.it

Dipartimento di Ingegneria, Università degli Studi di Napoli “Parthenope”, Centro Direzionale Isola
C4, 80143 Napoli, Italy. E–mail: filomena.feo@uniparthenope.it

Dipartimento di Matematica e Applicazioni “R. Caccioppoli”, Università degli Studi di Napoli Federico
II, Complesso Universitario di Monte S. Angelo, Via Cintia, 80126, Napoli, Italia. E–mail: g.zecca@unina.it

1
for a.e. x ∈ Ω and for any vector ξ in RN , where 0 < α 6 βi are constants;

(H2) Bi : Ω × R → R is a Carathéodory function such that




|Bi (x, s)| 6 bi (x)|s| pi (1.5)

for a.e. x ∈ Ω and for every s ∈ R, where bi is a non negative function in the Marcinkiewicz
N p′i
,∞ p
space L p̄ (Ω), where p′ = p−1 ;

(H3) Fi : Ω → R is such that Fi ∈ Lploc


i
(Ω).

Our model operator is


N
X  p̄

pi −2 p′
−1
L(u) = ∂xi |∂xi u| ∂xi u + bi (x)|u| i u ,
i=1

with bi as in (H2). The anisotropy of L is due to different power growths with respect to the
partial derivatives of the unknown u and it coincides with the so-called pseudo-Laplacian
operator when bi = 0 and pi = p for i = 1, ...., N .
Let us point out that term anisotropy is used in various scientific disciplines and could
have a different meaning when it is related to equations as well. The interest in anisotropic
problems has deeply increased in the last years and many results in different directions have
been obtained. We quote a list of references that is obviously not exhaustive and we refer
the reader to references therein to extend it: [1, 2, 3, 4, 7, 9, 10, 12, 13, 14, 15, 18, 19, 25, 24].
Several regularity results depending on the summability of datum are well-known in
literature for isotropic counterpart of (1.1)

div |∇u|p−2 ∇u + b(x)|u|p−2 u = div (|F|p−2 F) in Ω,

where p > 1, b and F are vector fields with suitable summability. In the isotropic case
for linear equations, assuming the coefficient of the lower order term in suitable Lebesgue
spaces, Stampacchia in [28] proves that if the datum F belongs to (Lr (Ω))N with 2 <

r < N , then the solution u belongs to Lr (Ω). Otherwise when r > N it follows that
the solution u is bounded. Similar results have been obtained also for isotropic nonlinear
operators taking the coefficient of the lower order term b in Lebesgue spaces in [5] and in
the Marcinkiewicz spaces in [6], [16], [21] and [22]. In this paper we prove such kind of
regularity results for anisotropic equation (1.1) dealing with local solution, whose definition
is recalled in what follows.

Definition 1.1 If Fi ∈ Lploc


i 1,~
(Ω) for i = 1, .., N , we say that u ∈ Wlocp
(Ω) is a local solution
to (1.1) provided
N Z
X N Z
X
(Ai (x, ∇u) + Bi (x, u)) ∂xi ϕ dx = (|Fi |pi −2 Fi ) ∂xi ϕ dx (1.6)
i=1 Ω i=1 Ω

∀ϕ ∈ C0∞ (Ω).

2
1,~
p
For the definition of the anisotropic Sobolev space Wloc (Ω) we refer to Section 4.
To give an idea of our result, let us consider, for simplicity, equation (1.1) without the
lower order terms, i.e. Bi ≡ 0. When the datum Fi ∈ Lploc i
(Ω) for i = 1, .., N with p̄ < N ,
p∗
a local solution u belongs to Lloc (Ω), where
N p̄
p̄∗ = , (1.7)
N − p̄
as suggested by the anisotropic imbedding (see Section 4). Otherwise if Fi ∈ Lrloci
(Ω) with
ri > pi for i = 1, .., N , we expect that the summability of u improves. In order to analyze
the higher summability of u we consider the following minimum
 
ri
µ = min , (1.8)
i pi
(pµ)∗
first introduced in [8], and we are able to prove that u ∈ Lloc (Ω) if p̄µ < N . Then the
regularity of u depends on µ.
These regularity results are stated in Theorem 2.1 taking into account the lower order
terms under the assumption that coefficients bi have a suitable distance to L∞ sufficiently
small for i = 1, · · · , N .
The principal difficulties are due to the anisotropy, to the managing of local solutions
and to the presence of the lower order terms. In our proof, one of the key tool is a new
anisotropic Sobolev inequality in Lorentz spaces that involves the product of different
powers of two functions (see Proposition 4.1). This inequality naturally appears in the
anisotropic framework, it is of independent interest and gives an estimate in terms of the
norm of the geometric mean of the partial derivatives instead of the geometric mean of
the norms of the partial derivatives as usual in literature (see (4.2)).
Moreover in Theorem 2.4 we give a sufficient condition in terms of µ for the bounded-
ness of the solutions to (1.1).
We also make comments on the regularity of weak solutions of Dirichlet problems in
a bounded open set Ω ⊂ RN with Lipschitz boundary. In this case when Fi ∈ Lpi (Ω) for
i = 1, .., N with p̄ < N a weak solution u belongs to Lp∞ (Ω), where p∞ = max{p∗ , pmax }
with pmax = max{p1 , · · · , pN } and p̄ defined as in (1.7). It is evident that the regularity of
u depends on how much the anisotropy is concentrated, so the situation is more diversified
than the isotropic case. Otherwise if Fi ∈ Lri (Ω) with ri > pi for i = 1, .., N and µp̄ < N ,
we get u ∈ Ls (Ω) with s = max{(pµ)∗ , µpmax } and the regularity again depends on how
much pi are spread out and on µ.
The paper is organized as follows. The main results are stated in Section 2. In Section
3 we recall same properties of Lorentz spaces and in Section 4 we introduce the anisotropic
Sobolev spaces and the related inequalities. The proofs of main results (Theorem 2.1 and
Theorem 2.4) are given in Section 5 and 6. We conclude the paper with a technical lemma
contained in the Appendix.

2 Main results
The first result of this paper states the regularity of local solutions to (1.1) in terms of the
summability of Fi , replacing the usual smallness assumption on the norm of the coefficients

3
of the lower order terms with a weaker one given in terms of the distance of a function
f ∈ Lp,∞ (Ω) to L∞ (Ω), denoted by distLp,∞ (Ω) (f, L∞ (Ω)) and defined by (3.11) in Section
3.

Theorem 2.1 Assume that (1.3)-(1.5) are fulfilled. Let 1 < p̄ < N , p̄∗ > pmax and let
r1 , · · · , rN be such that
N
1<µ< , (2.1)

where µ is defined in (1.8). There exists a positive constant d = d(~r, N, α, p~) such that if
( )
max dist Np′
i ,∞
(bi , L∞ (Ω)) <d (2.2)
i
L p̄ (Ω)

1,~
p
and u ∈ Wloc (Ω) is a local solution to (1.1) with Fi ∈ Lrloc
i
(Ω) for i = 1, .., N , then
N p̄(µ−1)
+1 1,~
p
|u| pi (N−p̄µ) ∈ Wloc (Ω). In particular u ∈ Lsloc (Ω), where
N µp̄
s = (µp̄)∗ = . (2.3)
N − µp̄
 +
Note that if µ goes to Np , then s → ∞ as expected. In the isotropic case condition

p̄ > pmax does not turn up and assumption (2.1) reads as isotropic assumption r < N
considering the datum F = (F1 , · · · , FN ) ∈ (Lrloc (Ω))N (see [22] and [16]).
Some comments on the case pmax ≥ p∗ are contained in Remark 5.2, Remark 5.3 and
Remark 6.1, where weak solutions for Dirichlet problems are taking into account.
It is clear that in Theorem 2.1 regularity ofnu is o related n
to µpo 1 , · · · , µpN . Indeed if ~r
ri qi
and ~q are two different vectors such that mini pi = mini pi = µ, then the solution
(µp̄)∗
u ∈ Lloc (Ω) if either Fi ∈ Lrloc i
(Ω) or Fi ∈ Lqloc
i
(Ω), as in the case Fi ∈ Lµp i
loc (Ω) for
i = 1, .., N .
A standard approach to treat the presence of lower order terms is to require a smallness
on the norm of bi , which is avoid using assumption (2.2), firstly introduced by [21] in the
isotropic case. We stress that the value of d in (2.2) really depends on ~r (see Example 3.1
in [22]).
For example if Ω is the ball centered at the origin with radius R > 0, we can take
− pp̄′
bi (x) = γi |x| i + hi (x) with γi > 0 and hi ∈ L∞ (Ω). It is not difficult to see that bi ∈
N p′i
,∞
L p̄ (Ω) and verifies (2.2) for suitable γi . We emphasize that condition (2.2) is trivially
satisfied whenever bi belongs to a Lebesgue space or to any Lorentz space contained in
Np′i
L p̄ ,∞ (Ω), i = 1, ..., N . Then as a corollary of Theorem 2.1 we have immediately the
following result.

Corollary 2.2 Assume that (1.3)-(1.5) are fulfilled. Let 1 < p̄ < N , p̄∗ > pmax , 1 < q <
∞,
Np′i
,q
bi ∈ L p̄ (Ω) i = 1, .., N
and let r1 , · · · , rN be such that (2.1) holds. If Fi ∈ Lrloc
i
(Ω) for i = 1, .., N , then any local
1,~
p s
solution u ∈ Wloc (Ω) to (1.1) belongs to Lloc (Ω), where s is defined in (2.3).

4
Without lower order terms in [8] the authors have proved that the boundedness of a
weak solution of Dirichlet problems is guaranteed under the assumption
N
µ> , (2.4)

where µ is defined in (1.8). However if Bi 6≡ 0 for i = 1, · · · , N the boundedness is not


assured assuming that (2.2) is in force as showed in Example 4.8 of [21] (when pi = 2 for
i = 1, · · · , N ). The smallness of kbi k Np′i for i = 1, · · · , N neither is sufficient to get
p̄ ,∞
L
boundedness, as the following example shows.

Example 2.3 Let us consider the following Dirichlet problem



 div(∇u + bu) = divF in Ω

u=0 on ∂Ω,
γx
where N > 2, Ω = B(0, 1) is the unit ball of RN , b(x) = |x| 2 , F(x) = ((2 + γ)x1 , · · · , (2 + γ)xN )
−γ 2
and γ > 0. We stress that the solution u(x) = |x| − |x| is unbounded even if kbkLN,∞ =
1/N
γωN can be small as we want taking γ small enough.

In order to obtain an L∞ −regularity result we require extra summability on the coefficients


of lower order terms.
ri
pi −1
Theorem 2.4 Assume that (1.3)-(1.5) holds. Let 1 < p̄ < N , p̄∗
> pmax , bi ∈ Lloc (Ω)
and Fi ∈ Lrloc
i
(Ω) for i = 1, .., N with r1 , · · · , rN satisfying (2.4). Then any local solution
1,~
p
u ∈ Wloc (Ω) to equation (1.1) is locally bounded.

3 Some properties of Lorentz spaces


In this section we recall the definitions of Lorentz spaces and their properties (see [27] for
more details). There are various definitions of Lorentz spaces but all of them manage the
notion of rearrangement.
Here we assume that Ω ⊂ RN , N > 2 is an open set. Let v be a measurable function
defined in Ω. The distribution function µv : [0, +∞) → [0, +∞) of v is defined as

µv (λ) := {x ∈ Ω : |v(x)| > λ} for λ > 0.

The decreasing rearrangement of v is the map v ∗ : [0, +∞) → [0, +∞] given by

v ∗ (s) = sup {t > 0 : µv (t) > s} for s > 0. (3.1)

By v ∗∗ we denote the maximal function of v ∗ , i.e.


Z
∗∗ 1 s ∗
v (s) = v (σ) dσ for s > 0. (3.2)
s 0

5
For 1 6 p < +∞ and 0 < q 6 +∞ the Lorentz space Lp,q (Ω) consists in all measurable
functions v : Ω → R such that
 Z |Ω| 

  1/p ∗ q dt 1/q

 t v (t) if q < +∞,
0 t
kvkLp,q (Ω) =

  1/p ∗ 

 sup t v (t) if q = +∞
t∈(0,|Ω|)

is finite. Notice that in general k · kLp,q (Ω) is a quasinorm, but replacing v ∗ with v ∗∗ one
obtains an equivalent norm and Lp,q (Ω) becomes a Banach space. We observe that it holds
k |v|s kLp,q = kvksLsp,sq for 0 < s < +∞.

The Lorentz spaces are a refinement of Lebesgue spaces. Indeed for p = q, Lp,p(Ω) re-
duces to the standard Lebesgue space Lp (Ω), and Lp,∞(Ω) is also known as Marcinkiewicz
space Mp (Ω), or weak-Lp (Ω). If Ω is bounded, such inclusions follow
Lr (Ω) ⊂ Lp,q (Ω) ⊂ Lp,p (Ω) ≡ Lp (Ω) ⊂ Lp,r (Ω) ⊂ Lp,∞ (Ω) ⊂ Lq (Ω)

for 1 < q < p < r 6 +∞, with continuous injections. We observe that Lp,∞ (Ω) ⊃ Lp (Ω).
−N
For example, if Ω ⊂ RN contains the origin, the function v(x) = |x| p 6∈ Lp (Ω) but
v ∈ Lp,∞ (Ω) with ||v||pLp,∞ = ωN , where ωN stands for the Lebesgue measure of the unit
ball of RN .
Now we introduce a suitable characterization of Lorentz spaces that is useful to gen-
eralize Sobolev embedding theorems in anisotropic framework.
Let k > 1 fixed, for any measurable function v defined in Ω such that for every λ > 0

µv (λ) < +∞, (3.3)


we choose the levels avn ≥ 0, n ∈ Z, such that

|{x ∈ Ω : |v(x)| > avn }| ≤ k−n ≤ |{x ∈ Ω : |v(x)| ≥ avn }|


or equivalently
avn ∈ [v ∗ ((k−n )+ ); v ∗ ((k−n )− )] (3.4)
where v ∗ is the decreasing rearrangement of v defined in (3.1) and where plus and minus
denote the right and the left-hand limit respectively.
Now we consider a particular sequence of functions ωn (v) defined by

 0 for 0 ≤ |v| ≤ avn−1
ωn (v) = |v| − an−1 for avn−1 < |v| ≤ avn
v (3.5)
 v
an − avn−1 for |v| > avn ,
with the levels avn defined for n ∈ Z as in (3.4). We observe that
ωn (v) ≤ (avn − avn−1 )χ{|v|>avn−1 } and ωn (v) ≥ (avn − avn−1 )χ{|v|≥avn } , (3.6)

where χ is the characteristic function, so one finds for 0 < r < ∞, that
Z 1
r n−1
−n v v r
k r (an − an−1 ) ≤ |ωn (v)| dx ≤ k− r (avn − avn−1 ).
RN

6
P
Recalling that a sequence {an }n∈Z belongs to lq (Z) for q > 1 iff n∈Z |an |q < +∞, we are
in position to state the following announced characterization of Lorentz spaces.

Proposition 3.1 Let be 1 < p < ∞ and 1 ≤ q ≤ ∞. For any v extended by 0 outside Ω
satisfying (3.3), one has
n
v ∈ Lp,q (RN ) is equivalent to k− p (avn+1 − avn ) ∈ lq (Z) (3.7)

and in particular
!1/q !1/q
X − nq
X q − nq
kvkLp,q (Ω) = kvkLp,q (RN ) ≤ C1 [avn ]q k p ≤ C2 avn − avn−1 k p , (3.8)
n∈Z n∈Z

with C1 , C2 are positive constants independent on v and avn is defined in (3.4).

Proof. Equivalence (3.7) is contained in [30], Proposition 3. We prove (3.8). Since the
rearrangement is non increasing, we have

XZ k −(n−1) h iq ds X (n−1)q
− p
Z k−(n−1)
ds
kvkqLp,q = 1/p ∗
s v (t) ≤ k ∗ −n + q
[v ((k ) )] (3.9)
s s
n∈Z k k −n
−n
n∈Z
X n−1 X − n−1 q X −n
= [v ∗ ((k−n )+ )k− p ]q log k ≤ log k (k p avn )q = k p log k (k p avn )q .
n∈Z n∈Z n∈Z

Now following the idea of Tartar [30], we put bn = avn − avn−1 . We observe that avn =
P n v
m=−∞ bm since the measure of the level sets is finite. We have that
n
X m−n
−n
p av
−m
p bv )
k n = k p (k m
m=−∞

and by the convolution Young inequality, we deduce the following inequality


!1   !1
X −n q X m X −n q

|k p avn |q ≤ kp |k p (avn − avn−1 )|q


n∈Z m≤0 n∈Z
  !1 (3.10)
1 X m X n
q

= k− p  kp |k− p (avn+1 − avn )|q .


m≤0 n∈Z

1 P m
Combining (3.10) and (3.9), inequality (3.8) follows with C2 = (log k) q m≤0 k
p .

We remark that L∞ (Ω) is not dense in Lp,∞ (Ω), p ∈ ]1, +∞[. We define the distance
of a given function f ∈ Lp,∞ (Ω) to L∞ (Ω) as

distLp,∞ (Ω) (f, L∞ (Ω)) = inf kf − gkLp,∞ (Ω) . (3.11)


g∈L∞ (Ω)

7
Note that, since k kp,∞ is not a norm, distLp,∞ (Ω) is just equivalent to a metric. In [11] is
proved that
distLp,∞ (Ω) (f, L∞ (Ω)) = lim kf − TM f kLp,∞ (Ω) , (3.12)
M →+∞
where the truncation at level M > 0 is defined as
y
TM (y) = min{|y|, M }. (3.13)
|y|
At the end of this section we recall the following useful lemma.

Lemma 3.2 (see [26, page 43] ) P Let X be a rearrangement invariant space and let 0 ≤
θi ≤ 1 for i = 1, ..., M, such that M
i=1 θi = 1, then
M
Y M
Y
|fi |θi ≤ kfi kθXi ∀fi ∈ X.
i=1 X i=1

4 Anisotropic inequalities
Let p~ = (p1 , p2 , ..., pN ) with pi > 1 for i = 1, ..., N and Ω be a bounded open set. As usual
the anisotropic Sobolev space is the Banach space defined as

W 1,~p (Ω) = {u ∈ W 1,1 (Ω) : ∂xi u ∈ Lpi (Ω), i = 1, ..., N }


equipped with
N
X
kukW 1,~p (Ω) = kukL1 (Ω) + k∂xi ukLpi (Ω)
i=1
and
1,~
p
Wloc (Ω) = {u ∈ W 1,~p (Ω′ ) : Ω′ ⊂⊂ Ω open set}.
It is well-known that in the anisotropic setting a Poincarè type inequality holds true
(see [18]). Indeed for every u ∈ C0∞ (Ω) with Ω a bounded open set with Lipschitz boundary
we have
kukLpi (Ω) ≤ CP k∂xi ukLpi (Ω) , i = 1, ..., N (4.1)
for a constant CP depending on the diameter of Ω. Moreover, for u ∈ C0∞ (RN ) the
following anisotropic Sobolev inequality holds true (see[30])
N
Y 1
kukLp,q (RN ) ≤ SN k∂xi ukLNpi (RN ) , (4.2)
i=1

where SN is an universal constant and p = p̄∗ and q = p̄ whenever p̄ < N , where p̄ is


defined in (1.2). Using the inequality between
PN geometric and arithmetic mean we can
replace the right-hand-side of (4.2) with i=1 k∂xi ukLpi .
When p < N P and Ω is a bounded open set with Lipschitz boundary, the space

− N
k∂x uk p
W01, p (Ω) = C0∞ (Ω) i=1 i L i is continued embedding into Lq (Ω) for q ∈ [1, p∞ ], with
p∞ := max{p∗ , pmax } as a consequence of (4.2) and (4.1).
In the following proposition we generalize the anisotropic Sobolev inequality (4.2) to
the product of functions.

8
Proposition 4.1 Let αi , βi ≥ 0 (but not both identically zero) and p1 , · · · , pN ≥ 1 be such
that 1 ≤ p < N . Then for every nonnegative functions φ, u ∈ C0∞ (RN ) we have

N
!1/N
Y ∂
φα uβ ≤C (φαi uβi ) , (4.3)
Lp∗ ,p ∂xi
i=1
Lp

where p∗ is defined in (1.7) and α and β are defined by


N N
1 X 1 X
α= αi , β= βi
N N
i=1 i=1

and C is a positive constant independent on u and φ.


It is clear that for example for βi = 1 the sign assumption on u can be dropped.
We point out that Lemma 3.2 with X = Lp (Ω) and θi = pipN yields that

N
Y 1/N

φα uβ ≤C (φαi uβi ) ∀φ, u ∈ C0∞ (RN ) (4.4)
Lp∗ ,p ∂xi L pi
i=1

with a positive constant C independent on u and φ. We emphasize that (4.3) is sharper


than (4.4) as the following example shows.
 N

P θ
Example 4.2 Let us consider 1 < p̄ < N , Ω = x : |xi | < R , φ ≡ 1, αi ≡ 1
i

i=1

N
!−γ N
!−γ
X X
u(x) = |xi |θi − R θi
i=1 i=1
Q 1
N ∂ N
with R > 0, θi > 1 for i = 1, · · · , N . Lemma 7.1 in the Appendix yields i=1 ∂xi u ∈
Lp̄ (Ω) taking
N
X 1 N −p
γ< (4.5)
θj pN
j=1

and ∂xi u ∈ Lpi (Ω) taking
N
X 1 1
γ< − (4.6)
θ j pi θ i
j=1

for i = 1, · · · , N . We note that (4.6) implies (4.5) and under latest assumption we get
∗ ∗
u ∈ Lp̄ (Ω), but u 6∈ Lp̄ +ε (Ω) for every ε > 0 (using again Lemma 7.1).
In order to prove Proposition 4.1 as first step we need the following result.

Lemma 4.3 For every gi ∈ C0∞ (RN ), it follows that


N
Y N
Y 1/N
1/N ∂
gi ≤ gi . (4.7)
∂xi L1
i=1 L1∗ i=1

9
Proof. The proof is a generalization to product of functions of the Troisi’s one ( see
[31] Theorem 1.2). We have, for q > 0 that will be chosen later, the following inequality
Z N
Y Z N
Y
|gi |q/N dx ≤ sup |gi |q/N dx.
RN RN xi
i=1 i=1

Since
Z N N Z  N −1 ! N−1
1
Y Y
sup |gi |q/N dx ≤ sup |gi |q/N dx′ ,
RN i=1 xi i=1 RN−1 xi

where x = (xi , x′ ) (see Lemma 4.1 of [20]), we get


Z N
!N −1 N Z N Z
Y Y q(N−1) Y q(N − 1) q(N−1) ∂gi
q/N
|gi | dx ≤ sup |gi | N dx′ ≤ |gi | N −1 dx.
RN i=1 RN−1 xi N RN ∂xi
i=1 i=1

q(N −1)
Now choosing q such that N − 1 = 0, we obtain inequality (4.7).

Proof of Proposition 4.1. Applying inequality (4.7) to the function


φ u αi βi
gi = ωn (φαi uβi ) + an−1 ,

we obtain
!1/N
∂ h i
N
Y N
Y 1/N
αi βi φαi uβi φαi uβi
ωn (φ u ) + an−1 ≤ ωn (φαi uβi ) + an−1 , (4.8)
∂xi L1
i=1 i=1
L1∗

where ωn and an are defined as in (3.5) and (3.4), respectively. We stress that

N
!1/N
Y
φα u β φαi uβi
an = an . (4.9)
i=1

Moreover using the Hardy-Lyttlewood inequality, we have


∂ h φαi uβi
i ∂ h i
ωn (φαi uβi ) + an−1 = ωn (φαi uβi )
∂xi L1 ∂xi L1
Z Z k −(n−1) ∗
∂ ∂
= αi βi αi uβi
(φαi uβi ) dx ≤ (φαi uβi ) (s) ds.
φ u
an−1 φ
<|φαi uβi |≤an ∂xi 0 ∂xi

Let us consider the left-hand side of (4.8). By (3.6) and (4.9) we get
!1/N N 
N
Y Y 1/N
φαi uβi α β
ωn (φαi uβi ) + an−1 ≥ anφ i u i χ φ αi uβi
{φαi uβi >an }
i=1 i=1 L1∗
L1∗
N 
Y 1/N
αi uβi n α uβ n
≥ anφ k− 1∗ = aφn k− 1∗ .
i=1

10
Putting all together, inequality (4.8) becomes
N Z !1/N
Y k −(n−1) ∗
φα u β − 1n∗ ∂
an k ≤C (φαi uβi ) (s) ds ,
0 ∂xi
i=1

where C denotes a positive constant that could change from line to line.
n
Multiplying for k p′ , elevating to the power p and summing, we obtain
Z k−(n−1) !p/N
X h α β ip − np XY N n
∂ ∗
φ u p ′ α β (4.10)
an k p ≤C

k i (φ i u i ) (s) ds .
0 ∂xi
n∈Z n∈Z i=1

On the other hand, by Proposition 3.1 with p = p∗ and q = p, it follows that


!1/p
X h α β ip − np
α β φ u
φ u p∗ ,p
≤C an k p ∗
.
L
n∈Z

We have, by the previous inequality and (4.10), that


 !p/N 1/p
XY N n
Z k−(n−1) ∗

φα uβ p∗ ,p ≤ C  (φαi uβi ) (s) ds  .

k pi
L 0 ∂x i
n∈Z i=1

We observe that by (3.2), since k > 1 it follows that


 !p/N 1/p
XY N n
Z k−(n−1) ∗

φα uβ p∗ ,p ≤C  (φαi uβi ) (s) ds 

k pi
L 0 ∂x i
n∈Z i=1

N 
XY ∗∗ p/N !1/p
1− pn ∂
= k i (φαi uβi ) (k −(n−1)
)
∂xi
n∈Z i=1
N  p/N Z !1/p
X kp Y ∗∗ k −(n−1)
∂ ds
= (φαi uβi ) (k −(n−1)
)
log k ∂xi k −n s
n∈Z i=1
 1/p N  p/N !1/p
kp XZ k −(n−1) Y
∂ ∗∗
ds
≤ (φαi uβi ) (s)
log k ∂xi s
n∈Z k
−n
i=1
N
!1/N
Y ∂
≤C (φαi uβi ) ,
∂xi
i=1
Lp

that is (4.3).

We stress that using inequality (4.7) we can prove the following well-known anisotropic
Sobolev inequalities in Lebesgue spaces. For every u ∈ C0∞ (RN ) there exists two positive
constant C1 , C2 independent of u such that
N
Y 1/N

kukLp̄∗ ≤ C1 u when p̄ < N
∂xi L pi
i=1

11
" N
#
Y 1/N

kukLq̄ ≤ C2 kukLp0 + u when p̄ = N
∂xi L pi
i=1

for every σi
PN q ∈ [p0 , ∞) and p0 ≥ 1. Indeed the starting point is to take gi = u in (4.7)
with i=1 σi = N t for suitable t > 0.

5 Proof of Theorem 2.1


The proof consists in two steps. First we prove our regularity result assuming that
kbi k Np′i are small enough for i = 1, · · · , N . Then, last assumption is removed thanks
,∞
L p
(2.2).
Step 1. Proof assuming that kbi k Np′
i ,∞
are small enough. Let us fix ϕ ∈ C01 (RN ), ϕ ≥
L p
0, supported in a ball contained in Ω. Let v = u − Tk (u), where Tk is defined as in (3.13).
for all s ∈ R. We use w = ϕq v with q > 0 (that we will choose later) as test function in
the weak formulation (1.6) and we denote by Ωk = {|u| > k}. Using assumptions (1.3),
(1.4) and (1.5), we get
N Z N Z N Z
X ∂ pi q X ∂ pi −1
∂ q−1
X p

(vϕq ) dx

α v ϕ dx ≤ βi v ϕ ϕ vdx + |bi ||u| pi
Ωk ∂xi Ωk ∂xi ∂xi Ω ∂x i
i=1 i=1 i=1 k

XN Z XN Z
∂ ∂
+ |Fi |pi −1 v ϕq dx + |Fi |pi −1 v ϕ qϕq−1 dx
∂x i ∂x i
i=1 Ωk i=1 Ωk
XN Z pi −1 XN Z p
∂ ∂ ∂
ϕq−1 (vϕq ) dx

≤ βi u ϕ vdx+ C |bi |k pi
Ωk ∂x i ∂x i ∂x i
i=1 i=1 Ωk
X N Z p XN Z p
∂ ∂
v ϕq dx + C ϕ qϕq−1 dx
′ ′
+C |bi ||v| pi |bi ||v| pi v
Ω k
∂xi Ω k
∂xi
i=1 i=1
XN Z XN Z
∂ ∂
+ |Fi |pi −1 v ϕq dx + |Fi |pi −1 v ϕ qϕq−1 dx
Ωk ∂x i Ωk ∂x i
i=1 i=1
:= I1 + I2 + I3 + I4 + I5 + I6 , (5.1)

where C is a positive constant independent of u. By Young inequality, we have


N Z
X pi N Z
X pi
∂ q ∂
I1 ≤ ε v ϕ + C(ε) ϕq−pi ϕ v pi , (5.2)
Ωk ∂xi Ωk ∂xi
i=1 i=1

where here and in what follows ε denotes a positive constant that will be chosen later.

12
Using Young inequality, Hölder inequality and (4.4), we get
N Z
X pi N Z
X
∂ ′
I3 ≤ε v ϕq dx + C(ε) |bi |pi |v|p ϕq dx
∂xi
i=1 Ωk i=1 Ωk
XN Z pi XN
∂ p′i q
≤ε v ϕq dx + C(ε) kbi k Np′
kvϕ p kpLp∗ ,p (5.3)
Ωk ∂xi i ,∞
i=1 i=1 L p
N Z 1 !p
∂  pq 
X pi N
X N
Y
∂ p′i N
≤ε v ϕq dx + C(ε) kbi k Np′
vϕ i
Ωk ∂xi i
p ,∞
∂xi L pi
i=1 i=1 L i=1

and
N Z 1 !p
∂  pq 
X pi N
X N
Y
∂ p′i N
I4 ≤ ε v pi ϕ ϕq−pidx+C(ε) kbi k Np′
vϕ i . (5.4)
∂xi i ∂xi L pi
i=1 Ωk i=1 L p ,∞ i=1

Finally Young and Hölder inequality yields


N Z pi N Z N Z
X
pi ∂ q−pi
X ∂ pi q X
I5 +I6 ≤ ε v ϕ ϕ dx+ε v ϕ dx+C(ε) |Fi |piϕq dx. (5.5)
Ωk ∂xi Ωk ∂xi Ωk
i=1 i=1 i=1

Now putting together inequalities (5.2), (5.3), (5.4) and (5.5), rearranging and choosing ε
small enough then inequality (5.1) becomes
"N Z !
N Z
X ∂ pi q X p
∂ XN
′ YN
∂  q  N1 p
p
(vϕq ) dx +

v ϕ dx ≤C |bi |k pi kbi k iNp′ vϕ pi
∂xi ∂x i i ,∞ ∂x i L pi
i=1 Ωk i=1 Ωk i=1 L p i=1
N Z N Z
#
X pi X
pi ∂ q−pi
+ v ϕ ϕ dx + |Fi |pi ϕq dx , (5.6)
Ωk ∂xi Ωk
i=1 i=1

for suitable constant C = C(~ ~ N ) that from now on could change from line to line.
p, α, β,
Now we note that
N Z N Z N Z
X ∂  pq  pi X ∂ pi q X ∂ pi
vϕ i dx ≤ 2pi −1 v ϕ dx + 2pi −1 v pi ϕ ϕq−pi dx
i=1 Ωk ∂xi i=1 Ωk
∂xi
i=1 Ωk
∂xi
 !
XN Z p
∂ XN
′ YN
∂  q  N1 p
q p
≤C p ′
|bi |k i (vϕ ) dx + kbi k Np′
i
vϕ pi
Ω k
∂xi p
i ,∞ ∂xi L pi
i=1 i=1 L i=1
N Z N Z
#
X pi X

+ v pi ϕ ϕq−pi dx + |Fi |pi ϕq dx ,
Ωk ∂xi Ωk
i=1 i=1

13
and so
Z    1 N Z
! p1
∂ q
pj
pj pj X ∂  pq  pi j

vϕ dx ≤ vϕ i dx (5.7)
Ωk ∂xj Ω ∂xi
i=1 k
"N Z N
1 X p
∂ X p′ p
(vϕq ) dx +

≤ C pj |bi |k pi kbi k iNp′ A N
∂xi i ,∞
i=1 Ωk i=1 L p
N Z N Z
# p1
X pi X j

+ v pi ϕ ϕq−pi dx + |Fi |pi ϕq dx ,
Ωk ∂xi Ωk
i=1 i=1

QN  q 

where A = i=1 ∂xi vϕ pi . Making the product on the left and right sides of (5.7),
L pi
we get
 ! Np ! Np
N Z
X p N
X
∂ p′i
A≤ C |bi |k p′
i (vϕq ) dx + kbi k Np′
A
Ωk ∂xi i ,∞
i=1 i=1 L p

! Np ! Np  (5.8)
N Z
X pi N Z
X

+ v pi ϕ ϕq−pi dx + |Fi |pi ϕq dx .
Ωk ∂xi Ωk
i=1 i=1

p′i
The assumption that the norms kbi k Np′
are small than a suitable constant depending
i ,∞
L p
~ N and (5.8) allow us to obtain
on p~, α, β,
 ! Np ! Np
N Z
X p N Z
X pi
∂ ∂
A ≤C  |bi |k p′
i (vϕq ) dx + v pi ϕ ϕq−pi dx
Ωk ∂xi Ωk ∂xi
i=1 i=1
!N  (5.9)
N Z
X p

+ |Fi |pi ϕq dx .
i=1 Ωk

Substituting (5.9) in (5.6), by easy calculations it follows


N Z
X pi N Z
X p X N Z pi
∂ q ∂ ∂
v ϕ dx ≤C bi k p′
i (vϕq ) dx + |v|pi ϕ ϕq−pi dx
Ωk ∂xi Ωk ∂xi Ωk ∂xi
i=1 i=1 i=1
!
XZ
N
+ |Fi |pi ϕq dx .
i=1 Ωk

At this point we multiply both sides of previous inequality by kγ for fixed γ > 0 that
will be chosen later and we integrate with respect to k over [0, K] for fixed K > 0. A
repeated use of Fubini’s theorem gives

14
N Z
" N Z  
X pi X p
∂ q γ+1 +γ+1 ∂ ∂
u ϕ |TK u| dx ≤ C bi |TK u| p′
i u ϕq + qϕq−1 u ϕ dx
Ω ∂xi Ω ∂xi ∂xi
i=1 i=1
N Z N Z
#
X pi X
pi ∂ q−pi γ+1
+ |u| ϕ ϕ |TK u| dx + |Fi |pi ϕq |TK u|γ+1 dx (5.10)
Ω ∂xi Ω
i=1 i=1

~ Now by Hölder inequality and Young inequality we get


where C = C(γ, α, N, p~, β).
N Z
X p
 
+γ+1 ∂ ∂
bi |TK u| p′
i u ϕq + qϕq−1 u ϕ dx ≤
Ω ∂xi ∂xi
i=1
N Z
! p1 N Z
! 1′
X pi i X p
∂ p′i i
u |TK u|γ+1 ϕq dx bi |TK u|γ+1+p ϕq dx
Ω ∂xi Ω
i=1 i=1
N Z
! p1 N Z
! 1′
X pi i X p
∂ p′ i
+ u ϕ |TK u|γ+1 ϕq−pi dx bi i |TK u|γ+1+p ϕq dx
∂xi
i=1 Ω i=1 Ω

XN Z pi N Z
X pi
∂ γ+1 q ∂
≤ǫ u |TK u| ϕ dx + ǫ u ϕ |TK u|γ+1 ϕq−pi dx
Ω ∂xi Ω ∂xi
i=1 i=1
γ+1 q
+1
+ 2C(ǫ)kbi k Np′
i ,∞
kTK u p ϕ kpLp∗ ,p ,
p

L p

where ǫ is a positive constant small enough. Substituting the previous inequality in (5.10)
and by (4.4), we obtain
N Z
"N Z
X pi X pi
∂ q γ+1 ∂
u ϕ |TK u| dx ≤ C u ϕ |TK u|γ+1 ϕq−pi dx (5.11)
Ω ∂xi Ω ∂xi
i=1 i=1

X N N
Y ∂   1 !p X N Z
#
γ+1 q N
+1 p q γ+1
+ kbi k Np′i |TK u| pi ϕ pi + |Fi | i ϕ |TK u| dx .
L p
,∞ ∂xi L pi Ω
i=1 i=1 i=1
 
QN ∂
γ+1
+1 q
Denoting B = i=1 ∂xi |TK u| pi
ϕ pi , it follows
L pi

N Z
X   pi
∂ γ+1
+1 q
|TK u| pi
ϕ pi
dx (5.12)
Ω ∂xi
i=1
N Z
X pi N Z
X pi
pi −1 ∂ q γ+1 pi −1 ∂
≤ 2 u ϕ |TK u| dx + 2 u ϕ |TK u|γ+1 ϕq−pi dx
Ω ∂xi Ω ∂xi
i=1 i=1
" N Z N N Z
#
X pi X X
∂ p
≤C u ϕ |TK u|γ+1 ϕq−pi dx + kbi k Np′ B N + |Fi |pi ϕq |TK u|γ+1 dx .
Ω ∂xi L p
i ,∞

i=1 i=1 i=1

Previous inequality give us an estimate of the j − th addendum of the sum at the left-hand
side of (5.12) as well. Then elevating to the power p1j , making the product on the left and

15
right sides of (5.12), we get
 ! Np ! Np
N Z
X pi N
X

B ≤C u ϕ |TK u|γ+1 ϕq−pi dx + kbi k Np′ B
Ω ∂xi L p
i ,∞
i=1 i=1
! Np  (5.13)
N Z
X
+ |Fi |pi ϕq |TK u|γ+1 dx .
i=1 Ω

p′i
Using again that the norms kbi k Np′
are small than a suitable constant depending on
i ,∞
L p

γ, α, N, p~, β~ from (5.13) we obtain

 ! Np ! Np 
N Z
X pi N Z
X

B ≤ C u ϕ |TK u|γ+1ϕq−pi dx + |Fi |pi ϕq |TK u|γ+1 dx  . (5.14)
Ω ∂xi Ω
i=1 i=1

Substituting (5.14) in (5.11) it follows that

N
X   pi
∂ γ+1
+1 q
|TK u| pi ϕ pi
∂xi L pi
i=1
"N Z # (5.15)
X pi N Z
X

≤C u ϕ |TK u|γ+1 ϕq−pi dx + |Fi |pi ϕq |TK u|γ+1 dx .
Ω ∂xi Ω
i=1 i=1

At this point, let us assume that u ∈ Lµp loc


max
(Ω) , where µ is defined in (1.8). Recalling
that Fi ∈ Lloc (Ω) for every i = 1, ..., N , then Fi ∈ Lploc
ri iµ
(Ω). Hence by Hölder inequality

with exponents µ and µ , from (5.15) we have
  Z 1
 
N 1 1
Y 1− µ
∂ γ+1
+1 q N µ
(γ+1) µ−1 p∗ q p
|TK u| pi
ϕ pi
≤C |TK u| ϕ p dx
∂xi L pi Ω
i=1
" N Z  1 N Z  1 # (5.16)
X pi µ X
∂ p̄µ pµ
× u ϕ ϕα1 dx + |Fi |pi µ ϕα2 dx
Ω ∂xi Ω
i=1 i=1
h ∗
 i h ∗
 i
with α1 = 1 − pp 1 − µ1 µq − pi µ, and α2 = 1 − pp 1 − µ1 qµ. Now, by (2.1) we
have α2 > 0 and we can now choose q large enough to have α1 > 0. Finally, we choose
now γ = NN−p̄µ

(µ − 1) − 1 so that
 
µ p
(γ + 1) = p∗ .
µ−1 γ +1+p

16
Thus, using Proposition 4.1, (5.16) becomes
N
Y   1
∂ γ+1
+1 q N
|TK u| pi
ϕ pi
∂xi L pi
i=1
 
p∗ N pi
N pi
!
γ+1
+1 q
1
1− µ p
X ∂ α1 p X α2
p
≤ C |TK u| p ϕ p u ϕ ϕ pi µ + |Fi |ϕ pi µ
Lp

∂xi L pi µ L pi µ
i=1 i=1
N   1− 1 p∗
 
N pi
N pi
!
Y ∂ γ+1
+1 q µ pN X ∂ α1 p X α2
p
≤C |TK u| pi
ϕ pi
u ϕ ϕ pi µ + |Fi |ϕ pi µ
.
∂xi L pi ∂xi L pi µ L pi µ
i=1 i=1 i=1

Rearranging the previous inequality it follows that

YN   1 !ρ XN pi
N
X pi
!
∂ γ+1
+1 q N ∂ α1 p α2
p
|TK u| pi
ϕ pi
≤C u ϕ ϕ pi µ + |Fi |ϕ pi µ
∂xi L pi ∂xi L pi µ L pi µ
i=1 i=1 i=1
(5.17)
h  i
p∗
where ρ := 1 − p 1 − µ1 that is positive by (2.1). Finally letting K → +∞ in (5.17),
we conclude
N   1 !ρ N pi
N pi
!
Y ∂ γ+1
+1 q N X ∂ α1 p X α2
p
|u| pi
ϕ pi
≤C u ϕ ϕ pi µ + |Fi |ϕ pi µ
.
∂xi L pi ∂xi L pi µ L pi µ
i=1 i=1 i=1
 
γ+1 N p̄µ
Since p̄ + 1 p̄∗ = N −p̄µ , using again Proposition 4.1 we obtain that
s(µ)
u ∈ Lloc (Ω) with s(µ) := (p̄µ)∗ . (5.18)
We observe that previous argument works directly when µpmax ≤ p∗ , since in this case
1,~
p
u ∈ Wloc (Ω) implies that u ∈ Lµp i
loc (Ω) for every i = 1, ...., N . If otherwise there exist

i ∈ {1, ..., N } such that µpi > p we use a bootstrap procedure.
Precisely, if there exist ij ∈ {1, ..., N } such that µpij > p∗ , j = 1, ..., m, with m ≤ N ,
we repeat the previous argument (from (5.15) to (5.18) ) with rij , j = 1, ..., m replaced by
p̄∗
s(1) = p∗ and µ replaced by the corresponding new minimum in (1.8), i.e. µ1 = pmax . In
s(µ )
this way we find u ∈ Lloc 1 (Ω) and

s(µ1 ) − s(1) > p̄∗ − pmax .


s(µ )
At this point, if µpmax ≤ s(µ1 ) we can use the information u ∈ Lloc 1 (Ω) to conclude
our proof as before. Otherwise we repeat the procedure again, that is: if there exist
ij ∈ {1, ..., N } such that µpij > s(µ1 ), we repeat the previous argument (again, from
(5.15) to (5.18) ) with rij replaced by s(µ1 ) and µ repleaded by µ2 := s(µ 1)
pmax so that we find
 ∗
s(µ )
u ∈ Lloc 2 (Ω) with s(µ2 ) = p̄ s(µ 1)
pmax . Using the convexity of the function f (p) = NN−p p

we find
N2 p̄ ∗ p̄∗
s(µ2 ) − s(µ1 ) ≥ (s(µ 1 ) − p̄ ) > (p̄∗ − pmax ) ,
(N − p̄)2 pmax pmax

17
h−1 s(µ )
and so on. Since, if necessary, at the h-th step one has µh = pmax and
 ∗ h−1

s(µh ) − s(µh−1 ) > (p̄∗ − pmax ) ,
pmax
it is now clear that, in a finite number of times, we can conclude the proof of Step 1.
Step 2. Dropping the smallness assumption
Now we remove the smallness assumptions on kbi k Np′i assuming that
,∞
( L p )

max dist Np′


i ,∞
(bi , L∞ (Ω))
i
L p̄ (Ω)
is sufficiently small. Setting
(
TM bi (x)
bi (x) if bi (x) 6= 0
θi (x) =
1 if bi (x) = 0
for M > 0, we rewrite equation (1.6) in the following form
XN Z N Z
X
(Ai (x, ∇u)+(1 − θi (x))Bi (x, u)) ∂xi ϕ dx = (|Fi |pi −2 Fi − θi (x)Bi (x, u)) ∂xi ϕ dx.
i=1 Ω i=1 Ω
(5.19)
Assumption (1.5) yields


|(1 − θi (x))Bi (x, u)| 6 |bi (x) − TM bi (x)||u| pi i = 1, ..., N
and p̄

|θi (x)Bi (x, u)| 6 M |u| pi . (5.20)
Using (3.12) we can find a sufficiently large constant M > 0 independent on i, such
that for every i = 1, ..., N the norm kbi − TM bi k Np′i is small in order to verify the
p̄ ,∞
L
smallness assumption of Step 1.
Hence, we can apply the result obtained in the previous step to (5.19) whenever
ri

Gi (x, u) := [|Fi |pi −2 Fi (x) − θi (x)Bi (x, u)] ∈ Lloci (Ω), for every i = 1, ...., N.
p −1

Using (5.20), it is enough to have



|u| pi ∈ Lrloc
i
(Ω), for every i = 1, ...., N. (5.21)
p∗
At this point we observe that if ri ≤ p pi for every i = 1, ...., N , then (5.21) holds

directly by Sobolev inequality. If otherwise there exist i ∈ {1, ..., N } such that ri > pp pi
we use a bootstrap procedure similar to the one used to conclude the proof of Step 1.

Precisely, if there exist ij ∈ {1, ..., N } such that rij > pp pij , j = 1, ..., m, with m ≤ N ,
p∗ ∗
we replace rij , j = 1, ..., m with p p ij and µ with µ1 := min{µ, pp } respectively. Now, if
s(µ)
µ1 = µ the proof is completed, since in this case we find u ∈ Lloc (Ω) with s(µ) as in (5.18).
∗ s(µ )
Otherwise, if µ1 = pp we find u ∈ Lloc 1 (Ω) with s(µ1 ) = p̄∗∗ and s(µ1 ) − s(1) > p̄∗ − pmax .
∗∗ s(µ )
Now, if rij ≤ pp pij for every j = 1, ..., m, then u ∈ Lloc 1 (Ω) gives (5.21). Otherwise we
repeat the procedure again. It is clear that, in a finite number of times, we can conclude
our proof. 

18
Remark 5.1 In the previous proof the definition of µ in (1.8) as a minimum obviously
appears. Indeed, using Hölder inequality in (5.14) with exponent ζ and ζ ′ , it is enough to
require pi ζ ≤ ri , i = 1, ..., N. Hence, the best choice of such ζ is µ.
Remark 5.2 We stress that assumption p∗ > pmax is essential for our technique. Indeed,
at the end of the Step 1 of previous proof, in order to start with the bootstrap argument
we need u ∈ Lrloc
i
(Ω) with ri ≤ p̄∗ for every i = 1, .., N that means pi < p̄∗ , i = 1, ..., N.
Remark 5.3 If Ω is bounded open set with Lipschitz boundary and we consider homoge-
neous Dirichlet problems we can argue as in Theorem 2.1 to obtain a regularity of solutions
without restriction pmax < p∗ . Precisely we have the following result. Assume that (1.3)-
(1.5) are fulfilled, let 1 < p̄ < N and let r1 , · · · , rN be such that (2.1) holds. There exists
a positive constant d = d(~r, N, α, p~) such that if
( )
max dist Np′
i ,∞
(bi , L∞ (Ω)) : <d
i
L p̄ (Ω)

and u ∈ W01,~p (Ω) is a weak solution to (1.1) with Fi ∈ Lri (Ω), then
u ∈ Ls (Ω) with s = max{(µp̄)∗ , µpmax }. (5.22)
Indeed when the max{(µp̄)∗ , µpmax } = (µp̄)∗ the proof of (5.22) runs as Theorem 2.1 taking
into account that we are managing not local solutions. Otherwise if max{(µp̄)∗ , µpmax } =
µpmax we can reason as in Step 1 of Theorem 2.1 but instead of (5.15) we obtain

N Z  1 Z 
 
1
∂  γ+1
 pmax X µ µ
1− µ
|TK u| pmax +1 ≤C |Fi |pi µ dx |TK u| (γ+1) µ−1
dx .
∂xj Lpmax Ω Ω
i=1
(5.23)
Now choosing γ > 0 such that
 
µ pmax
(γ + 1) = pmax , (5.24)
µ−1 γ + 1 + pmax
by Poincaré inequality (4.1), (5.23) becomes
pmax
∂   N
X ri
γ+1 µ
+1
|TK u| pmax ≤C kFi kLµri , (5.25)
∂xj Lpmax i=1
where the constant C depends
 now
 also by Ω. Letting K → +∞ in (5.25), observing that
γ+1
equality (5.24) implies pmax + 1 = µ, and using again Poincaré inequality (4.1), we
obtain (5.22) with s = µpmax .
Remark 5.4 Let us consider the homogeneous Dirichlet problem in a bounded open set
with Lipschitz boundary Ω under the assumptions (1.3)-(1.4), when (1.5) is replaced by

|Bi (x, s)| 6 bi (x)|s|pi −1 ∀i (5.26)


for a.e. x ∈ Ω and for every s ∈ R and bi ∈ L∞ (Ω) for all i. In this case we obtain the
same regularity result as in Remark 5.3. Indeed one can obtain the analogous inequality
of (5.15) using Poincaré inequality (4.1) instead of Sobolev inequalities. Starting from the
obtained estimate, one can conclude the proof as before under assumption (5.26).

19
6 Proof of Theorem 2.4
The first step of our proof is the boundedness of a local weak solution without the lower
order terms in order to adapt Stampacchia’s arguments [28] to the anisotropic case in
the same spirit of Lemma 5.4 in [23] when one deals with local solutions. Finally, our
assumptions on the summability of the coefficients bi allow us to apply Corollary 1.3
concluding also when Bi 6≡ 0.
Step 1. Proof assuming that Bi vanishes. Using the same test function w as in Theorem
2.1 (Step 1) and assumptions (1.3) and (1.4), we get
N Z N Z N Z
X ∂ pi q X ∂ pi −1 q−1 ∂ X ∂
α v ϕ dx ≤ βi v ϕ ϕ vdx+ |Fi |pi −1 v ϕq dx
Ωk ∂x i Ωk ∂x i ∂x i ∂x i
i=1 i=1 i=1 Ωk
XN Z

+ |Fi |pi −1 v ϕ qϕq−1 dx := L1 + L2 + L3 . (6.1)
Ωk ∂x i
i=1

Since L1 = I1 and L2 + L3 = I5 + I6 , where I1 , I5 , I6 are defined in Theorem 2.1 (Step


1), putting together inequalities (5.2), (5.5) and choosing ε small enough, inequality (6.1)
became
N Z
"N Z N Z
#
X ∂ pi q X ∂ pi X
v ϕ dx ≤C v pi ϕ ϕq−pi dx + |Fi |pi ϕq dx , (6.2)
i=1 Ωk ∂xi Ωk ∂xi
i=1 Ωk i=1

~ that will change from line to line.


for suitable positive constant C = C(α, N, p~, β)
Now we choose the cut-off function ϕ. Let σ > τ > 0, we fix two concentric balls
Bτ ⊂ Bσ ⊂⊂ Ω of radii τ and σ respectively. We consider ϕ such that 0 ≤ ϕ ≤ 1, ϕ ≡ 1
2
in Bτ , ϕ ≡ 0 in Ω \ Bσ and |∇ϕ| ≤ σ−τ . We put Ωk,σ = Ωk ∩ Bσ . By (6.2), taking
q > maxi pi and using Hölder inequality we get
N Z
X pi N
X Z N
X
∂ 1 1
1− µ
v dx ≤C (|u| − k)pi dx + C kFi kpLiµpi (Ωk,σ ) |Ωk,σ | .
Ωk,τ ∂xi (σ − τ )pi Ωk,σ
i=1 i=1 i=1

We stress that since Fi ∈ Lrloc


i
(Ω),
then Fi ∈ Lploc

(Ω)
for every i = 1, ..., N . Moreover
there exists b
k > 0 such that |Ωk,σ | < 1 for k ≥ b
k, and then
1 p
|Ωk,σ |1− µ ≤ |Ωk,σ |1− N

under the assumption (2.4). Then, for k > b k


N Z N
" Z #
X ∂ pi X
−pi pi −δN p
1− N +δ pi
v dx ≤ C (σ − τ ) (|u| − k) dx+σ |Ωk,σ | kFi kLµpi (Ωk,σ ) ,
Ω ∂xi Ω
i=1 k,τ i=1 k,σ

(6.3)
where δ is a positive constant that we choose later. Inequality (6.3) is the key ingredient
to conclude that we can choose k0 > 0 such that
Z
(u − 2k0 ) dx = 0, (6.4)
Ω2k0 ,σ−σ0

20
for a fixed 0 < σ0 < σ, i.e. u is locally bounded when the norms of bi are small enough.
The claim (6.4) will be proved in the next step following the idea of Lemma 5.4 of [23].
Step 2. Proof of (6.4).
Let us consider the sequence of concentric balls Bρh , with ρh = σ − σ0 + 2σh0 and the
sequence kh = 2k0 − 2kh0 , where k0 > 0 will be fixed later. We will denote for h = 0, 1, ...
Z  
Jhi = (|u| − kh )pi dx and ζh (x) = ζ 2h+1 (|x| − σ + σ0 ) ,
Ωkh ,ρh

where ζ(s) is a continuously differentiable nonincreasing function on R that is equal to 1


for s ≤ σ0 and equal to 0 for s ≥ 23 σ0 . We stress that ζn ≡ 1 inside the ball Bρh+1 and
ζn ≡ 0 outside the ball Bρh , where ρh = 12 (ρh+1 + ρh ). We have
Z
i
Jh+1 ≤ (|u| − kh+1 )pi ζhpi dx.
Ωkh+1 ,ρh

Moreover by Poincaré inequality obtained by Sobolev and Hölder inequalities we get


 (Z 
YN  pj ) pj1N pi
i
pi
1− +
pi ∂
Jh+1 ≤ C|Ωkh+1,ρh | p N  [(|u| − kh+1 )ζh ] 
Ω kh+1 ,ρh
∂x j
j=1
 " # 1 pi (6.5)
YN Z pj pj N
pi pi ∂
≤ C|Ωkh+1,ρh |1− p + N  u dx + ν j 2pj h Jhj  ,
Ωk ∂x j

j=1 h+1 h

~ and
where C = C(α, N, p~, β)
νi = max [ζ ′ (s)]pi .
s∈[σ0 , 23 σ0 ]

Now we write (6.3) with k = kh+1 and σ = ρh and τ = ρh :

N Z
X pi

v dx (6.6)
Ωkh+1 ,ρh ∂xi
i=1
"N Z N
#
X 1 p X
≤C (|u| − kh+1 ) dx + pi
ρ−δN
h |Ωkh+1 ,ρh |1− N +δ kFi kpLiµpi
(ρh − ρh )pi Ωkh+1 ,ρh
i=1 i=1
N
!
p X
≤C ρ−δN
h |Ωkh+1 ,ρh |1− N +δ + 2(h+3)pi Jhi ,
i=1

~ Let us find a bound for the measure of the set Ωk ,ρ .


where C = C(α, kFi kLµpi , N, p~, β). h+1 h
We have
Z
Jhi ≥ (|u| − kh )pi ≥ (kh+1 − kh )pi |Ωkh+1 ,ρh | = 2−(h+1)pi k0pi |Ωkh+1 ,ρh |. (6.7)
Ωkh+1 ,ρh

21
Putting together (6.7), (6.6), inequality (6.5) can be rewrite as
 " N Z # p 1N pi
pi  Y 
N X pm XN j
i
pi
1− + ∂
Jh+1 ≤ C|Ωkh+1,ρh | p N u dx + ν m 2pm h Jhm
 Ωk ∂xm 
,ρ j=1 m=1 h+1 h m=1
" N Z
# ppi
 1− pi + pi X ∂ pm N
X
≤ C 2(h+1)pi k0−pi Jhi
p N
u dx + ν m 2pm h Jhm
Ωkh+1 ,ρh ∂xm
m=1 m=1
" # ppi
 1− pi + pi N 
X  p
≤ C 2(h+1)pi k0−pi Jhi
p N
2(h+3)pm Jhm + ν m 2pm h Jhm + |Ωkh+1 ,ρh |1− N +δ
m=1
" # ppi
 1− pi + pi N 
X   1− p +δ
≤ C 2(h+1)pi k0−pi Jhi 2(h+3)pm Jhm + ν m 2pm h Jhm + 2(h+1)pi k0−pi Jhi
p N N

m=1
 ! ppi
p2 2 N
X 2 p2 p2
hpmax −pmin + max h pmax
p p
1+ Ni − pi h(1+δ) pmax
max − min (1+δ)
1+δ
pi
≤ C2 k0 p∗ 2 p (Jhi ) Jhm +2 p k0 N p
(Jhi ) p

m=1

N
! ppi 
pmax 2 pi pi pi X
+ 2h p (max ν m ) p (Jhi )1+ N − p Jhm ,
m
m=1

where pmin = mini {p1 , ..., pN }. Since p∗ > pmax the exponent 1 + pNi − pi
p > 0 and setting
P
Yh = N m
m=1 Jh , one can rewrite the previous inequality as follows
2
" p2
pmax pi 2 pi p 2 p2
max min (1+δ)
i −pmin + 1+δ h pmax −δ pi h(1+δ) pmax −
Jh+1 ≤C2hpmax k0 p∗
(Yh ) p 2 p (Yh ) N +2 p k0 N p

2 pi pi p

h pmax m −δ pi
+2 p max((max ν ) ) (Yh ) p N (6.8)
i m
p2
max ) −pmin + pmax pi h pi pi i
h(pmax + p∗ 1+δ −δ
≤C2 p k0 (Yh ) p (Y0 ) N p +1 ,

where the last inequality follows taking δ < Np and observing that Jhm are decreasing.
Putting δ′ = δpp i and summarizing left and right side of (6.8), we get

p2 p2
max h i
h(pmax + max ) −pmin + ′ pi ′
Yh+1 ≤ N C2 p k0 p∗
(Yh )1+δ max (Y0 ) N −δ + 1 .
i
h pi
i 2 p2
(pmax + max
Denoting C1 = N C maxi (Y0 ) N −δ + 1 , ω = pmin − pmax )

p ∗ and b = 2 p the previ-
ous inequality became
Yh+1 ≤ C1 bh k0−ω (Yh )1+δ .

1/ω
Choosing k0 such that k0 = max{b
′ ′ ′
k, 1, C1 b1/[ωδ (1+δ )] aδ /ω }, we obtain
′ ω/δ′ −1/δ′ −1/(δ′ )2
Y1 ≤ C1 k0−ω Y01+δ ≤ k0 C1 b .

Now we are in position to apply Lemma 4.7 of [23] in order to obtain (6.4).

22
Step 3. Dropping the assumption that Bi ≡ 0.
ri
p −1
In the general case where bi ∈ Lloc
i
(Ω), with r1 , · · · , rN satisfying (2.4), we rewrite
equation (1.6) as
N Z
X N Z
X
(Ai (x, ∇u)) ∂xi ϕ dx = (|Fi |pi −2 Fi − Bi (x, u)) ∂xi ϕ dx. (6.9)
i=1 Ω i=1 Ω

Then, we can apply the result obtained in the previous steps to (6.9). In fact, by (2.4)
q
and by Theorem 2.1, we have
n othat u ∈ Lloc (Ω) for every q < +∞. Hence we can find
si N
pi < si < ri such that mini pi > p̄ and
si

Gi (x, u) := [|Fi |pi −2 Fi (x) − Bi (x, u)] ∈ Lloci (Ω), for every i = 1, ...., N.
p −1

Remark 6.1 If Ω is a bounded open set with Lipschitz boundary and we consider homo-
geneous Dirichlet problems we can argue as in Theorem 2.4 to obtain the boundedness of
solutions. Precisely, instead of (6.2), we obtain
N Z
X pi N Z
X

v dx ≤ C |Fi |pi dx,
Ωk ∂xi Ωk
i=1 i=1

~ N ) > 0. At this point, we can proceed as in Theorem 2 in [29] and


where C = C(α, p~, β,
one can conclude the proof using the Stampacchia’s Lemma (see [28]) instead of Step 2 of
Theorem 2.4.

7 Appendix
In this appendix we prove a technical lemma (see also [17]).

Lemma
P 7.1 Let R > 0, α ≥ 0, β ≥ 0 and θi > 0 for all i = 1, · · · , N and Ω =
{ Ni=1 |x i |θi < R}. If
XN
1 β
+ > α, (7.1)
θ θj
i=1 i
P −α P −α
N θi β ∈ L1 (Ω). Otherwise N θi
then i=1 |x i | |x j | i=1 |x i | 6∈ L1 (Ω).
θ
Proof. Putting θ = max θi and xi = |yi | θi signyi we have
Z N
!−α Z N
!−α N
X X θ
βY θ θ
−1
θi β θ θj
|x i | |x j | dx = |y i | |y j | |yi | θi dy
PN
{ i=1 |xi |θi <R} { N θ
i=1 i
θ
P
i=1 i=1 |yi | <R} i=1
Z
−θα+ θθ β+ N θ
P
i=1 θi −N
≤ C1 P |y| j dy
N
{ i=1 |yi |θ <R}
Z PN
−θα+ θθ β+ θ
i=1 θi −N
≤ C1 |y| j dy
{|y|θ <C2 R}

23
for suitable positive constants C1 , C2 . The last integral is finite if (7.1) holds. Otherwise
θ
putting θ = min θi and xi = |yi | θi signyi we have
Z N
!−α Z N
!−α N
X X θ
β Y θ θ
−1
|xi |θi |xj |β dx = |yi |θ |yj | θj |yi | θi dy
{ N
P θi
i=1 |xi | <R} { N
P θ
i=1 |yi | <R}
θi
i=1 i=1 i=1
Z θ
−θα+ N θ β
P
−N θj
≥C3 |y| i=1 θi
|yj | dy
{|y|θ <R/N }

where C3 is a positive constant. Now we use spherical coordinates, in which the coordinates
consist of a radial coordinate ρ and N − 1 angular coordinates ψ1 , ψ2 , · · · , ψN −1 , where
the angles ψ1 , ψ2 , · · · , ψN −2 range over [0, π] and ψN −1 ranges over [0, 2π). For example
y1 coordinate becomes y1 = ρ cos ψ1 and then
Z
−θα+ N θ θ
P
i=1 θi −N β
|y| |y1 | θ1 dy
{|y|θ <R/N }
Z 2π Z R PN θ
N
Y −2
N θ
−θα+ i=1 θi −N β
= ρ |ρ cos ψ1 | θ1 ρN −1 sinN −1−i ψi dρ dψ1 · · · dψN −1
0 0 i=1
Z R PN θ θ
N −θα+ i=1 θi −N + θ1 β+N −1
= C(N ) ρ dρ.
0

The last integral is finite if (7.1) does not hold and for j = 2, · · · , N the proof runs
similarly.

Acknowledgements
The authors are members of Gruppo Nazionale per l’Analisi Matematica, la Probabilità e
le loro Applicazioni (GNAMPA) of the Istituto Nazionale di Alta Matematica (INdAM).
Research partially supported by project Vain-Hopes within the program VALERE: VAn-
viteLli pEr la RicErca.

References
[1] Alberico A., di Blasio G., Feo F., Estimates for fully anisotropic elliptic equations
with a zero order term, Nonlinear Analysis 181 (2019), 249-264.

[2] Alberico A., di Blasio G., Feo F., An eigenvalue problem for the anisotropic Φ-
Laplacian, J. Differential Equations 269 (2020), 4853-4883.

[3] Antontsew S., Chipot M., Anisotropic Equations: Uniqueness and existence results,
Diff. Int. Eq. 21 (2008), 401-419.

[4] Bendahmane M., Karlsen K. H, Nonlinear anisotropic elliptic and parabolic equations
in RN with advection and lower order terms and locally integrable data, Potential
Anal. 22 (2005), 207–227.

24
[5] Boccardo L., Finite energy solutions of nonlinear Dirichlet problems with discontin-
uous coefficients, Boll. Un. Mat. It. 5 (2012), 357-368.

[6] Boccardo L., Dirichlet problems with singular convection terms and applications, J.
Differential Equations, 258 (2015), 2290–2314.

[7] Boccardo L., Gallouet T., Marcellini P., Anisotropic Equations in L1 , Diff. Int. Eq. 9
(1996), 209-212.

[8] Boccardo L., Marcellini P., Sbordone C., L∞ − regularity for variational problems
with sharp non-standard growth conditions, Boll. Un. Mat. Ital. A, 4 (1990), 219-
225.

[9] Bousquet P., Brasco L., Lipschitz regularity for orthotropic functionals with nonstan-
dard growth conditions, Rev. Mat. Iberoam. 36 (2020), 1989–2032.

[10] Brandolini B., Cı̂rstea F. C., Anisotropic elliptic equations with gradient-dependent
lower order terms and L1 data, (2020) arXiv:2001.02754.

[11] Carozza M., Sbordone C., The distance to L∞ in some function spaces and applica-
tions, Diff. Int. Equ. 10 (1997), 599-607.

[12] Cianchi A., Local boundedness of minimizers of anisotropic functionals, Ann. Inst.
Henri Poincaré, Anal. Non Linéaire 17 (2000), 147-168.

[13] Cianchi A., Symmetrization in anisotropic elliptic problems, Comm. Partial Differen-
tial Equations 32 (2007), 693-717.

[14] Cupini G., Marcellini P., Mascolo E., Regularity under sharp anisotopic general
growth conditions, Discrete Contin. Dyn. Syst. Ser. B 11 (2009), 67-86.

[15] Di Castro A., Existence and regularity results for anisotropic elliptic problems,
Adv.Nonlinear Stud. 9 (2009), 367-393.

[16] Farroni F., Greco L., Moscariello G., Zecca G., Noncoercive quasilinear elliptic oper-
ators with singular lower order terms, Calc. Var. and Partial Differential Equations
(2021), DOI: 10.1007/s00526-021-1965-z.

[17] Feo F., Vàzquez J. L., Volzone B., Anisotropic Fast Diffusion Equations,
arXiv:2007.00122.

[18] Fragalà I., Gazzola F., Kawohl B., Existence and nonexistence results for anisotropic
quasilinear elliptic equations, Ann. I. H. Poincaré 21 (2004), 715-734.

[19] Fragalà I., Gazzola F., Lieberman G., Regularity and nonexistence results for
anisotropic quasilinear elliptic equations in convex domains, Disc. Cont. Dyn. Syst.
(2005), 280-286.

[20] Gagliardo E., Proprietà di alcune classi di funzioni in più variabili, Matematika 5
(1961), 87-116.

25
[21] Giannetti F., Greco L., Moscariello G., Linear elliptic equations with lower order
terms, Diff. Int. Equ. 26 (2013), 623–638.

[22] Greco L., Moscariello G., Zecca G., Regularity for solutions to nonlinear elliptic
equations, Diff. Int. Equ. 26 (2013), 1105-1113.

[23] Ladyzhenskaya O. A., Linear and Quasilinear Elliptic Equations, Academic Press
(1968).

[24] Leonetti F., Innamorati A., Global integrability for weak solutions to some anisotropic
elliptic equations, Nonlinear Analysis 113 (2015), 430-434.

[25] Moscariello G., Regularity results for quasi minima of functionals with nopolynomial
growth, J. Math. Anal. Appl. 168 (1992), 500–512

[26] Krein S. G., Petunin Yu. I., Semenov E. M., Interpolation of linear operators, Transl.
Math. Monogr. Amer. Math, Soc. 54 (1982).

[27] Pick L., Kufner A., John O., Fucik S., Function spaces, Walter de Gruyter, Berlin -
New York (2013).

[28] Stampacchia G., Le problème de Dirichlet pour les équations elliptiques du second
ordre à coefficients discontinus, Annales de l’Institut Fourier 15 (1965), 189-257.

[29] Stroffolini B., Global boundedness of solutions of anisotropic variational problems,


Boll. Un. Mat. Ital. A 5 (1991), 345-352.

[30] Tartar L., Imbedding theorems of Sobolev spaces into Lorentz spaces, BUMI Serie 8
(1998), 479-500.

[31] Troisi M., Teoremi di inclusione per spazi di Sobolev non isotropi, Ricerche Mat. 18
(1969), 3-24.

26

You might also like