On existence and asymptotic behavior of solutions of elliptic equations with nearly critical exponent
and singular coefficients
Shiyu Li
1, Gongming Wei
B1and Xueliang Duan
21College of Science, University of Shanghai for Science and Technology, Shanghai 200093, China
2School of Mathematical Sciences, Zhejiang University, Hangzhou 310027, China
Received 5 August 2020, appeared 8 September 2021 Communicated by Dimitri Mugnai
Abstract. In this paper we study the existence and asymptotic behavior of solutions of
−∆u=µ u
|x|2+|x|αup(α)−1−ε, u>0 inBR(0)
with Dirichlet boundary condition. Here,−2<α<0, p(α) = 2(N+α)N−2 , 0<ε<p(α)−1 and p(α)−1−εis a nearly critical exponent. We combine variational arguments with the moving plane method to prove the existence of a positive radial solution. Moreover, the asymptotic behaviour of the solutions, asε→0, is studied by using ODE techniques.
Keywords: asymptotic behavior, critical Sobolev exponent, Hardy exponents, singular coefficient.
2020 Mathematics Subject Classification: 35A15, 35A24, 35B40.
1 Introduction
In this paper, we consider the following elliptic problem:
−∆u=µ u
|x|2 +|x|αup(α)−1−ε, x∈ Ω,
u>0, x∈ Ω,
u=0, x∈ ∂Ω,
(1.1)
where Ω is a ball BR(0) in RN(N ≥ 3), −2 < α < 0, p(α) = 2(NN−+2α), 0 < ε < p(α)−1, 0≤µ<µ= N−222.
The equation in problem (1.1) is the Euler–Lagrange equation of the energy functional E:H01(Ω)→Rdefined by
E(u) = 1 2
Z
Ω|∇u|2−µ u2
|x|2
− 1
p(α)−ε Z
Ω|x|αup(α)−ε, ∀u∈ H10(Ω).
BCorresponding author. Email: gmweixy@163.com
It is known that critical points of functionalE(u)correspond to solutions of (1.1).
We denote
kuk, Z
Ω|∇u|2−µ u2
|x|2 12
, ∀u∈ H01(Ω).
Let us recall the Sobolev–Hardy inequality (see Lemma2.1in this paper), which using the fact 0≤µ<µ¯ implies thatkukis equivalent to the norm ofH01(Ω).
In the caseµ=0 andα=0, a prototype of problem (1.1) is
−∆u=u2∗−1−ε, x ∈Ω, u >0, x ∈Ω,
u =0, x ∈∂Ω.
(1.2)
Whenε=0, it is well known that the solution of problem (1.2) is bounded in the neighbor- hood of the origin. Gidas, Ni and Nirenberg [17] proved that all the solutions with reasonable behavior at infinity, namely
u=O(|x|2−N), (1.3)
are radially symmetric about some point. So, the form of the solutions may be assumed as u(x) = [N(N−2)λ2]N−42
(λ2+|x−x0|2)N2−1 for someλ>0 andx0 ∈RN.
Later in [7, Corollay 8.2] and [9, Theorem 2.1], the growth assumption (1.3) was removed, which implies that, for positiveC2 solutions of problem (1.2), we have the same result.
When ε > 0, Atkinson and Peletier [2] used ODE arguments to obtain exact asymptotic estimates of the radially symmetric solution of problem (1.2) asε→0. The following are their principal results
lim
ε→0εu2(0,ε) = 4
N−2{N(N−2)}N−22 Γ(N) Γ(N2)2
1 RN−2 and forx6=0
lim
ε→0ε−
1
2u(x,ε) = 1
2NN4−2(N−2)N4RR−22 Γ(N2) [Γ(N)]12
1
|x|N−2 − 1 RN−2
. In the caseµ=0 andα>0, problem (1.1) is known as the Hénon equation
−∆u=|x|αup−1, x∈Ω,
u>0, x∈Ω,
u=0, x∈∂Ω,
(1.4)
where p ∈ (2, 2∗). Equation (1.4) was proposed by Hénon when he studied rotating stellar structures and readers can refer to Ni [24], Smets [26] and Cao–Peng [11]. Among these works, for equations with critical, supercritical and slightly subcritical growth, the existence and multiplicity of non-radial solutions, the symmetry and asymptotic behavior of ground states were studied by variational method (forp→ N2N−2 orα→∞).
In the case 0≤µ<µ= N2−22 andα=0, problem (1.1) can be written as
−∆u=µ u
|x|2 +u2∗−1−ε, x∈Ω,
u>0, x∈Ω,
u=0, x∈∂Ω.
(1.5)
By using Moser iteration and a generalized comparison principle, Cao and Peng [10] proved u(x)∈ H01(Ω)satisfying
(u(x)|x|ν ≥C1, ∀x∈ Ω0 ⊂⊂Ω, u(x)|x|ν ≤C2, ∀x∈ Ω,
where C1 and C2 are two positive constants, ν = pµ−pµ−µ. When Ω = BR, u(x) is radially symmetric. Hence, they converted (1.5) to ODE and obtained the following:
lim
ε→0 lim
|x|→0εu2ε|x|2ν=4(2p
µ−µ)N−1NN−22(N−2)−N+22 Γ(N) Γ(N2)2
1 R2
√
µ−µ
and for x6=0 lim
ε→0ε−
1
2uε(x) = 1 2√
2(2p
µ−µ)N−23NN−42(N−2)−N4−6R
√
µ−µ Γ(N2) [Γ(N)]12
× 1
|x|
√
µ+√
µ−µ
− 1
|x|
√
µ−√
µ−µ|R|2
√
µ−µ
! .
Motivated by the previous works and remark 4.2 in [10], we first prove the existence and radial symmetry of positive solution of (1.1). Then we focus on the asymptotic behavior of the solutions of problem (1.1) asε→0.
To state our main results, for convenience, we set p = p(α)−1−ε, ν = pµ−pµ−µ, Ω= BR = {x ∈ RN : |x| < R}, R > 0. We denote byuε(x)the solution of (1.1) andΓ(x)is the Gamma function.
Theorem 1.1. Suppose that−2< α< 0, 0≤ µ<µ, 0¯ < ε< p(α)−1. Then problem(1.1) has a radially symmetric solution in H01(Ω).
For the proof of this Theorem1.1, we first obtain a solution by the Mountain Pass Lemma.
Then, by moving plane method for elliptic equations with variable coefficients in [14], we can prove that the posotive solution is radially symmetric. For problem (1.2), the solution satisfies Gidas–Ni–Nirenberg Theorem in [17] and hence all solutions of (1.2) are radial symmetric.
However, here we cannot use Gidas–Ni–Nirenberg theorem directly since problem (1.1) in- cludes the hardy termµ|xu|2 and singular coefficient |x|α. Luckily, through a transformation of the original solution uε(x), the new equation satisfied by the new solution v(x)satisfies the conditions of a Corollary in [14] and we obtain the result. To be more precise, set
v(x) =|x|−
√
µ+√
µ−µuε(x),
using Moser iteration and a generalized comparison principle introduced by Merle and Peletier [22], we prove that v ∈ L∞(Ω) and is bounded from below and above. Thus we obtain that the precise singularity of uε(x)at the origin is like |x|−
√
µ+√
µ−µ
. Then applying
Lemma2.5 in Section 2 to this new equation, we deduce thatv(x)is radially symmetric and satisfies the following ODE:
v00+ N−1−2ν
r v0+ 1
r(p(α)−2−ε)ν−αvp(α)−1−ε =0, for 0<r <R,
v(r)>0, for 0<r <R,
v(R) =0.
(1.6)
Because (1.6) is still singular at the origin, we can use the well-known shooting argument introduced by Atkinson and Peletier [2] to convert (1.6) to the following ODE:
y00(t) =−t−k(α,ε)yp(α)−1−ε, y(t)>0, forT< t<∞, y(T) =0,
(1.7)
wherek(α,ε) = 2mm−+1α −(p(αm)−−21−ε)ν,m=1+2pµ−µ= N−2ν−1,T = (mR−1)m−1,p(α)−1= 2k(α,ε)−3−m2νε−1.
Till now, study on behaviors and precise properties of the original solution uε(x)can be reduced to deal with (1.7). Based on this, we have
Theorem 1.2. Let uε(x)∈ H01(Ω)be a solution of problem(1.1). Then lim
ε→0 lim
|x|→0εu2ε|x|2ν =2(α+2)(2p
µ−µ)2Nα++α2−2(N+α)Nα+−22(N−2)−2α+α+N2+2 Γ(2(N+2)
α+2 ) Γ(Nα++2α)2
1 R2
√
µ−µ
. Theorem 1.3. Let uε(x)∈ H01(Ω)be a solution of problem(1.1). Then, for every x 6=0,
lim
ε→0ε−
1
2uε(x) = 1
2(α+2)−12(2p
µ−µ)2N2α−+α4−6(N+α)2αN−+24(N−2)2α2α−N++46R
√
µ−µ Γ(N+α
α+2) hΓ
2(N+α) α+2
i12
× 1
|x|
√
µ+√
µ−µ
− 1
|x|
√
µ−√
µ−µ|R|2
√
µ−µ
! . Notations:
• C,Ci,i=0, 1, 2, . . . denote positive constants, which may vary from line to line;
• k · k and k · kLq denote the usual norms of the spaces H01(Ω) and Lq(Ω), respectively, Ω∈RN;
• Some of the notations that will appear in the following paragraphs:
m=1+2p
µ−µ= N−2ν−1, T =m−1 R
m−1
, k=k(α,ε) = 2m+α
m−1 − (p(α)−2−ε)ν
m−1 , k1(α,ε) = (k−1)k−12, k2(α,ε) = k−1
k−2, Tα,ε = γ
p(α)−2−ε k−2
k1(α,ε) = γ
2−(mm−−11)(−k2ν−2)ε
k1(α,ε) , τ(α,ε) = t
Tα,ε k−2
, ϕ(α,ε) = m−1+2ν
(m−1)(k−2)ε,
Cα,β,ε = β
(1+βk−2)k−12, dα,β,ε
= (1−Cα,β,ε)(1+2ν/(m−1)) Cα,β,ε2+(1+2ν/(m−1))ε .
2 Preliminary results and existence of solution
In this section,we shall provide some preliminaries which will be used in the sequel and prove the existence of solution to problem (1.1).
Lemma 2.1(see [16, Lemma 3.1 and 3.2]). Suppose−2< α<0,2≤ q≤ p(α), and0≤ µ< µ.
Then
(i) (Hardy inequality)
Z
Ω
u2
|x|2 ≤ 1 µ
Z
Ω|∇u|2, ∀u∈ H01(Ω); (ii) (Sobolev–Hardy inequality)
there exists a constant C>0such that Z
Ω|x|α|u|q 1q
≤Ckuk, ∀u∈ H01(Ω); (iii) the map u 7→ |x|αqu from H01(Ω)into Lq(Ω)is compact for q< p(α).
Lemma 2.2(see [5, Theorem 2.2]). Let J be a C1function on a Banach space X. Suppose there exists a neighborhood U of0in X and a constantρsuch that J(u)≥ρfor every u in the boundary of U,
J(0)< ρ and J(v)<ρ for some v∈/U.
Set
c= inf
g∈Γmax
ω∈g J(ω)≥ρ, whereΓ={g ∈C([0, 1],X):g(0) =0,g(1) =v,J(g(1))< ρ}.
Conclusion: there is a sequence{un}in X such that J(un)→c and J0(un)→0in X∗.
Lemma 2.3(The Caffarelli–Kohn–Nirenberg inequalities, see [8] and [12]). For all u∈ C0∞(RN), Z
RN|x|−bq|uq| pq
≤Ca,b Z
RN|x|−ap|Du|pdx, where (i) for n > p,
−∞< a< n−pp, 0≤ b−a≤1, and q= np
n−p+p(b−a)
and (ii) for n≤ p,
−∞< a< n−pp, p−pn ≤b−a ≤1, and q= np
n−p+p(b−a).
Lemma 2.4(see [25, page 4]). Suppose V is a reflexive Banach space with normk · k, and let M⊂V be a weakly closed subset of V. Suppose E : M → R∪ {+∞}is coercive and (sequentially) weakly lower semi-continuous on M with respect to V, that is, suppose the following conditions are fulfilled:
(1) (coercive) E(u)→∞askuk →∞, u∈ M.
(2) (W.S.L.S.C) For any u ∈ M, any sequence {um} in M such that um + u weakly in V there holds:
E(u)≤lim inf
m→∞ E(um).
Then E is bounded from below on M and attains its infimum in M.
Lemma 2.5(see [14, Corollary 1.6]). Let u be a bounded C2(BR\{0})∩C1(BR\{0})solution of
∂i(|x|b∂iu) +K|x|auq=0, x ∈BR\{0}, u>0, x ∈BR\{0},
u=0, x ∈∂BR\{0},
where K is a positive constant.Then u is radially symmetric in BRprovided q≥1, b(12b+N−2)≤0 and12b≥ aq.
Proof. Whenb<0, we have|x|bis singular at origin.
It’s clear that S(x)
|x|b − S(xλ)
|xλ|b = 1 2b
1
2b+N−2
(|x|b−2− |xλ|b−2)≥0 and
K |x|b
|xλ|b 12
|xλ|auq−K|x|a
|x|b
|xλ|b 12
u
q
= K|x|12b|xλ|a−12b
1− |xλ|
|x|
12bq−a
uq≥0, whereS(x) = 12(∆|x|b− 1
2|x|b|∇|x|b|2). From [14], we have
hλ(x) =
|xλ|b
|x|b 12
u(xλ)−u(x), wherexλ = (2λ−x1,x2, . . . ,xN).
By Lemma 4.2 in [14], we can obtainuhas a positive lower bound near the origin. Hence, we can get the estimate ofhλ(x)near the origin. Furthermore, ifxλ =0, we havehλ(x) =∞.
Now, we consider the case of u ∈ C2(B1\{0})∩C1(B1\{0}) in Proposition 1.3 of [14].
Analogically, we can also obtain u(x1,x2, . . . ,xN) ≤ u(−x1,x2, . . . ,xN) for x1 ∈ (−1, 0) and x1 ∈(0, 1). Hence, uis symmetric inx1. By Lemma 1.1 in [14], the above analysis and scaling transformation,u(x)is radially symmetric inBR.
Next, we shall prove the existence of solution to the problem (1.1). To start with, we prove the existence of nonnegative solution to the following Dirichlet problem:
(−∆u=µ|xu|2 +|x|α|u|p(α)−2−εu, x ∈Ω,
u=0, x ∈∂Ω, (2.1)
whereΩ is a ball inRN(N ≥ 3) centered at the origin,−2 < α< 0, p(α) = 2(NN−+2α), 0< ε <
p(α)−1,0≤ µ<µ= N−222.
The energy functional corresponding to problem (2.1) is J(u) = 1
2kuk2− 1 p(α)−ε
Z
Ω|x|α|u|p(α)−ε, u∈ H01(Ω). Lemma 2.6. The function J satisfies(PS)ccondition for every c∈R.
Proof. Takec ∈ R and assume that {un} is a Palais–Smale sequence at level c, namely such that
J(un)→c and J0(un)→0 (in H−1(Ω)). This implies that there is a constant M>0 such that
|J(un)| ≤ M. (2.2)
FromJ0(un)→0, we obtain
o(1)kunk=hJ0(un),uni=kunk2−
Z
Ω|x|α|un|p(α)−ε. (2.3) Calculating (2.2) − p(1
α)−ε(2.3), we have M+o(1)kunk ≥ 1
2kunk2− 1 p(α)−ε
Z
Ω|x|α|un|p(α)−ε
− 1
p(α)−εkunk2+ 1 p(α)−ε
Z
Ω|x|α|un|p(α)−ε
= 1
2− 1
p(α)−ε
kunk2,
which implies the boundedness of {un}. By usual arguments, we can assume that up to a subsequence, there existsu∈ H01(Ω)such that
• un*uin H10(Ω);
• |x|p(αα)−εun → |x|p(αα)−εuin Lp(α)−ε(Ω);
• ·un→ufor almost everyx ∈Ω.
We now show that the convergence ofunto uis strong.
First of all, from the above convergence properties, we obtain
un|x|p(αα)−ε −u|x|p(αα)−ε
Lp(α)−ε(Ω)→0, n→∞.
As J0(un) → 0 and un * u, we also have hJ0(un),un− ui → 0 and obviously hJ0(u),un−ui →0. Then, asn→∞, on the one hand,
hJ0(un)−J0(u),un−ui ≤ |hJ0(un),un−ui|+|hJ0(u),un−ui|=o(1). On the other hand,
hJ0(un)−J0(u),un−ui
=
Z
Ω|∇un− ∇u|2−
Z
Ωµ|un−u|2
|x|2 −
Z
Ω|x|α(|un|p(α)−2−εun− |u|p(α)−2−εu)(un−u)
=kun−uk2−
Z
Ω|x|α(|un|p(α)−2−εun− |u|p(α)−2−εu)(un−u). We claimR
Ω|x|α(|un|p(α)−2−εun− |u|p(α)−2−εu)(un−u)→0.
Indeed, by Hölder’s inequality, Z
Ω|x|α|un|p(α)−2−εun(un−u)
≤
Z
Ω|x|α|un|p(α)−1−ε|un−u|
=
Z
Ω|x|α·p
(α)−1−ε
p(α)−ε |un|p(α)−1−ε|x|α·p(α1)−ε|un−u|
≤
Z
Ω
|x|α·p
(α)−1−ε
p(α)−ε |un|p(α)−1−ε p(p(α)−ε
α)−1−ε
p(α)−1−ε p(α)−ε Z
Ω
|x|α·p(α1)−ε|un−u|p(α)−ε p(α1)−ε
= Z
Ω|x|α|un|p(α)−ε
p(pα()−1−ε
α)−ε
|x|p(αα)−ε|un−u|
Lp(α)−ε(Ω)
≤Ckunkp(α)−1−εun|x|p(αα)−ε −u|x|p(αα)−ε
Lp(α)−ε(Ω)
=o(1).
(2.4)
By (2.4), similar calculation also gives Z
Ω|x|α|u|p(α)−2−εu(un−u) =o(1). (2.5) From the above analysis, we obtain
o(1) =hJ0(un)−J0(u),un−ui=kun−uk2+o(1),
which impliesun→uinH01(Ω)and proves thatJsatisfies(PS)ccondition for everyc∈R.
Lemma 2.7. The function J admits a(PS)csequence in the cone of nonnegative function at the level c= inf
g∈Γmax
t∈[0,1]J(g(t)), whereΓ={g∈C([0, 1],H01(Ω)): g(0) =0,J(g(1))<0}.
Proof. We next prove thatJsatisfies all the hypotheses of the mountain pass lemma. Obviously, J(0) =0.
From the Sobolev–Hardy inequality, we obtain J(u) = 1
2kuk2− 1 p(α)−ε
Z
Ω|x|α|u|p(α)−1−εu
≥ 1
2kuk2−C1kukp(α)−ε.
For anyα, we can chooseεsmall enough such thatp(α)−ε>2. From the above analysis,there exist ρ,e > 0 such that J(u) ≥ ρ,∀u ∈ {u ∈ H10(Ω) : kuk = e}. Furthermore, for any u∈ H01(Ω),
J(tu) = t
2
2kuk2− t
p(α)−ε
p(α)−ε Z
Ω|x|α|u|p(α)−1−εu.
We obtain J(tu) → −∞ as t → ∞. Hence, we can choose t0 > 0 such that J(t0u) < 0.
Therefore, by Lemma 2.2, we infer that J admits a (PS)c sequence at level c, such sequence may be chosen in the set of nonnegative functions becauseJ(|u|)≤ J(u)for allu∈ H01(Ω).
By Lemma 2.6, 2.7 and mountain pass lemma, we get a nonnegative solution u ∈ H01(Ω) for (1.1), this solution is positive by the maximum principle.
3 Estimate of the singularity
First, we fix p = p(α)−1−ε > 0 in problem (1.1) and study the singularity and radial symmetry of the solution uε(x) ∈ H01(Ω). By standard elliptic regularity theory, uε(x) ∈ C2(Ω\{0})TC1(Ω\{0}). Hence the singular point ofuε(x)should be the origin.
Suppose thatuε(x)∈ H01(Ω)satisfies problem (1.1).
Letv(x) =|x|νuε(x), then
−∆u= (−ν2−2ν+Nν)|x|−ν−2v(x) +2ν|x|−ν−2x∇v(x)− |x|−ν∆v(x), µ u
|x|2 +|x|αup(α)−1−ε =µ|x|−ν−2v(x) +|x|α−(p(α)−1−ε)νv(x)p(α)−1−ε. From equation in (1.1),
(−ν2−2ν+Nν)|x|−ν−2v(x) +2ν|x|−ν−2x∇v(x)− |x|−ν∆v(x)
=µ|x|−ν−2v(x) +|x|α−(p(α)−1−ε)νv(x)p(α)−1−ε. Multiply both sides of the above equation by|x|−ν, then we get
[−ν2+ (N−2)ν]|x|−2ν−2v(x)−div(|x|−2ν∇v(x)) =µ|x|−2ν−2v(x) +|x|α−(p(α)−ε)νv(x)p(α)−1−ε. For ν=pµ−pµ−µ, we obtain
−div(|x|−2ν∇v) =|x|−(p(α)−ε)ν+αvp(α)−1−ε, x∈Ω,
v >0, x∈Ω,
v =0, x∈∂Ω.
(3.1)
By the regularity theory of elliptic equations,vε(x)∈C2(Ω\{0})TC1(Ω\{0}). Moreover, we have
Lemma 3.1.
(i) v(x)∈ H01(Ω,|x|−2ν). (ii) v(x)is bounded in Ω.
Proof. (i) For anyu(x)∈ H01(Ω)satisfying problem (1.1), by Hardy inequality, we have Z
Ω|x|−2ν|∇v|2=
Z
Ω|x|−2ν||x|ν∇u+ν|x|ν−2ux|2
≤2 Z
Ω|∇u|2+ν2 Z
Ω
u2
|x|2
≤ C.
Hence, we claimv(x) =|x|νu(x)∈ H10(Ω,|x|−2ν).
(ii) From Caffarelli–Kohn–Nirenberg inequality mentioned in Lemma2.3, we have
Z
Ω|x|m1|∇u|2 12
≥Cm1,n1 Z
Ω|x|n1|u|p(m1,n1) p(m1
1,n1)
, ∀u∈ H01(Ω,|x|m1), (3.2)
where
m1= −2ν, n1= −(p(α)−ε)ν+α, p(m1,n1) = p(α) + pεν µ−µ. Note that Z
Ω|x|m1∇v· ∇ϕ=
Z
Ω|x|n1vpϕ, ∀ϕ∈ H01(Ω,|x|m1).
For s,l > 1, define vl(x) = min{v(x),l}. Taking ϕ = v·v2l(s−1) ∈ H01(Ω,|x|m1)in the above equation, we have
Z
Ω|x|m1|∇v|2v2l(s−1)+2(s−1)
Z
Ω|x|m1|∇vl|2v2l(s−1) =
Z
Ω|x|n1vp+1v2l(s−1). Hence,
Z
Ω|x|n1(v·vsl−1)p(m1,n1) p(m2
1,n1)
≤Cm−12,n1 Z
Ω|x|m1|∇(v·vsl−1)|2
≤2Cm−12,n1
(s−1)2
Z
Ω|x|m1|∇vl|2v2l(s−1)+
Z
Ω|x|m1|∇v|2v2l(s−1)
≤2Cm−12,n1s Z
Ω|x|n1vp+2s−1.
(3.3)
From (3.3) and Levi’s theorem, we see thatv∈Lp+2s−1(Ω,|x|n1)impliesv∈Lsp(m1,n1)(Ω,|x|n1). For j=0, 1, 2, . . . , by induction we define
(p−1+2s0= p(m1,n1),
p−1+2sj+1= p(m1,n1)sj, (3.4)
M0= (C·Cm−21,n1)p
(m1,n1)
2 ,
Mj+1= (2Cm−12,n1sjMj)p
(m1,n1)
2 ,
(3.5) whereCis a fixed number such thatR
Ω|x|m1|∇v|2 ≤C.
From (3.4), we see that
sj = (2−1p(m1,n1))j+1(p(m1,n1)−p−1) +p−1 p(m1,n1)−2 . From (3.5), similar to the computation in [21], we can see that
∃d>0 anddis independent of j, such that Mj ≤edsj−1.
Since 2< p+1< p(m1,n1), it follows thatsj >1 for all j≥0, sj →+∞as j→+∞.
By (3.3), (3.4) and (3.5), Z
Ω|x|n1vp+2s1−1≤(2Cm−12,n1s0)p
(m1,n1) 2
Z
Ω|x|n1vp+2s0−1 p
(m1,n1) 2
≤(2Cm−12,n1s0)p(m1,2n1)
Cp(m1,2n1)C−m1p,n(m11,n1)
p(m1,2n1)
≤(2Cm−12,n1s0M0)p(m1,2n1)
≤ M1.
Similarly,
Z
Ω|x|n1vp+2sj−1 ≤ Mj.
Hence, byp+2sj+1−1= p(m1,n1)sj, denotingC(Ω,n1) =maxx∈Ω|x|−n1, we obtain
|v|
Lp(m1,n1)sj(Ω)≤ Z
Ω|v|p(m1,n1)sj|x|n1· |x|−n1 p(m1
1,n1)sj
≤ C(Ω,n1)
1 p(m1,n1)sj
|v|
1 p(m1,n1)sj
Lp(m1,n1)sj(Ω,|x|n1)
≤ C(Ω,n1)
1 p(m1,n1)sj
M
1 p(m1,n1)sj j+1
≤ C(Ω,n1)
1 p(m1,n1)sj
e
d p(m1,n1).
Taking limit on each side of the above inequality and usingsj →+∞, asj→+∞, we have
|v|L∞(Ω)≤ep(m1,dn1), which implies the conclusion.
From Lemma 3.1, we can see that v(x) = |x|νu(x)is bounded form above in Ω. For the lower bound ofv(x) =|x|νu(x), we have
Lemma 3.2. Suppose that u(x) ∈ H01(Ω) satisfies problem (1.1) and 0 ≤ µ < µ, then for any Bρ⊂⊂Ωthere exists a C(ρ)>0, such that
u(x)≥C(ρ)|x|−ν, ∀x∈ Bρ ⊂⊂Ω. Proof. Let f(x) =min{|x|αup(α)−1−ε(x),l}withl>0, then f ∈ L∞(Ω).
Letu1 ≥0 andu1∈ H01(Ω)be the solution of the following linear problem
−∆u1 =µ u1
|x|2 + f, x ∈Ω,
u1=0, x ∈∂Ω.
(3.6)
SetU=u−u1, thenU∈ H01(Ω)andUsatisfies the following problem
−∆U= µ U
|x|2 +g, x∈Ω,
U=0, x∈∂Ω,
(3.7)
where g≥0 and 0≤ µ<µ= (N−22)2.
From Lemma 2.4, there exist solutions for problem (3.6) and (3.7). From the Hardy in- equality and the comparison principle proved in [15], we obtain that u is a super-solution of problem (3.6) and 0 ≤ u1 ≤ u. Actually we can prove this as follows. Multiplying U−:=max{0,−U(x)}on both side of equation in (3.7) and integrating by parts, we have
−
Z
Ω|∇U−|2 =−
Z
Ωµ(U−)2
|x|2 +
Z
ΩgU−. It follows thatU− =0 inΩand hence U≥0.