Morse theory for indefinite nonlinear elliptic problems
✩Kung-Ching Chang, Mei-Yue Jiang
∗LMAM, School of Mathematical Sciences, Peking University, Beijing, 100871, PR China Received 22 February 2007; accepted 22 August 2007
Available online 3 December 2007
Abstract
Using the heat flow as a deformation, a Morse theory for the solutions of the nonlinear elliptic equation:
−u−λu=a+(x)|u|q−1u−a−(x)|u|p−1u+h(x, u)
in a bounded domainΩ⊂RN with the Dirichlet boundary condition is established, wherea±0, supp(a−)∩supp(a+)= ∅, supp(a+)= ∅, 1< q <2∗−1 andp >1. Various existence and multiplicity results of solutions are presented.
©2007 Elsevier Masson SAS. All rights reserved.
1. Introduction
We study the nonlinear elliptic equations with indefinite nonlinearities. Arising from differential geometry and biology, the problem has been received much attention in recent years, see [1–5,8,9,16–18,21,22,24,26].
One of the modelling problems can be stated as follows: LetΩbe a bounded domain inRNwith smooth boundary, we study the existence and multiplicity of positive, negative and sign-changing solutions of the following elliptic boundary value problem:
−u=λu+a+(x)|u|q−1u−a−(x)|u|p−1u+h(x, u) inΩ,
u=0 on∂Ω, (1.1)
wherea±:Ω→Rare continuous functions andh:Ω×R→Ris aC1function, andλis a real parameter. We assume (A1) a±0,Ω+∩Ω−= ∅andΩ+= ∅, whereΩ±=supp(a±),
(A2) 1< q <2∗−1=NN+−22,p >1, (A3) there exists a constantC >0 such that
h(x, ξ )C 1+ |ξ|
, ξ∈R.
The case 1< pq <2∗−1 has been studied by many of the previous papers, while the case 1< q < p and q <2∗−1 by [5], but only for positive solutions.
✩ This work was supported by NSFC, RFDP of the Ministry of Education of China.
* Corresponding author.
E-mail address:mjiang@math.pku.edu.cn (M.-Y. Jiang).
0294-1449/$ – see front matter ©2007 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.anihpc.2007.08.004
The paper is a continuation of our previous paper [16]. The simple decomposition lemma in [16], on which the computation of the critical groups at infinity of the associated functionalI (see below) relies, plays an important role in dealing with this kind of indefinite nonlinearities.
In [14], the first author observed that there are several advantages if we use the heat flow for Eq. (1.1) to establish the Morse theory for isolated critical points, i.e., instead of the gradient flow:
∂tv(t )=v(t )−(−)−1
λv+a+(x)|v|q−1v−a−(x)|v|p−1v+h(x, v) for the associated functional of (1.1):
I (u)=
Ω
1 2
|∇u|2−λu2
−a+(x)
q+1|u|q+1+a−(x)
p+1|u|p+1−H (x, u)
dx,
whereH (x, u)=u
0 h(x, s) ds, inR×H01(Ω), the following heat semi-flow:
∂tv(t, x)=(+λ)v+a+(x)|v|q−1v−a−(x)|v|p−1v+h(x, v), (t, x)∈R+×Ω
inR+×C01(Ω)is considered. The heat semi-flow is theL2gradient flow ofI and can be used as deformation of the level sets ofI as the gradient flow.
The disadvantage of this method, to our knowledge, is that a more restrictive exponent on the nonlinear term
|u|q−1u is needed, i.e.,q < p∗1<2∗−1, where p∗1 is defined in Section 3. But one of the advantages of the heat semi-flow is the positive invariance of the cones± ˜P, whereP˜= {u∈C01(Ω)|u(x)0, x∈Ω}. This was observed in [1] and plays an important role in the existence of sign-changing solutions of (1.1).
Recall the case of definite nonlinearity, i.e.,a−=0,Ω+=Ω, if there is a positive invariant setDfor the gradient flow, then one can estimate the number of solutions inside and outside of D, separately. Few abstract critical point theorems can be applied in the study of multiple solutions, see [13]. In this paper, we shall extend these results to fit the indefinite nonlinearity by the above two ingredients. Our main purpose is to develop the above tools in dealing with the multiple solution problems for indefinite nonlinearities. The main results are stated in Section 5, among other things, the following theorem will be proved:
Theorem.Under the assumption(A1), (A2), (A3), (A4)and (A6), to Eq. (0.1)λ,γ there existsγ∗>0 such that
∀γ > γ∗,∃ − ∞< λ−(γ ) < λ1< λ+(γ )such that:
(1) For0< λ < λ−(γ ), there exist at least one positive, one negative and one sign-changing solutions.
(2) Forλ−(γ ) < λ < λ1, there exist at least three positive, three negative and one sign-changing solutions.
(3) Forλ1< λ < λ+(γ ), there exist at least two positive, two negative and one sign-changing solutions.
(4) Forλ2< λ < λ+(γ ), there exist at least two positive, two negative and three sign-changing solutions.
Results in [1,4,5,16,22] are extended.
2. A decomposition lemma and the critical groups at infinity
We extend the decomposition lemma in [16] to the problem (1.1). The main difference is that the exponentpmay be greater than 2∗−1. For Eq. (1.1), the associated functional is defined on the Banach spaceE=H01(Ω)∩Lpa−+1(Ω), where
Lpa−+1(Ω)= u∈D(Ω)
Ω
a−(x)u(x)p+1dx <∞
.
ThusE=H01(Ω)ifΩ−= ∅andp >2∗−1, andE=H01(Ω)ifp2∗−1 by the embedding theorem. The norm onEis defined byu =(∇u22+ u2
Lp+1a− )2. The space is the closed subspace
E=
(u1, u2)∈H01(Ω)×Lpa−+1(Ω)|u1=u2
ofH01(Ω)×Lpa−+1(Ω), and bothH01(Ω)andLpa−+1(Ω)are reflexive, according to Pettis Theorem, we have
Lemma 2.1.The Banach spaceEis reflexive.
The following decomposition lemma was proved in [16] ifp2∗−1. Let E1=H01(Ω0∪Ω−)∩Lpa−+1(Ω0∪Ω−)
and
E2=
u∈H01(Ω0∪Ω+)|u(x)=0∀x∈Ω0 .
Theorem 2.2.
E=E1⊕E2.
Proof. (1)∀u∈E, let
v(x)= u(x), x∈Ω−, u(x)−w0(x), x∈Ω0 and
w(x)= w0(x), x∈Ω0, u(x), x∈Ω+, wherew0∈H1(Ω0)is given by
w0=0, x∈Ω0,
w0(x)=0, x∈∂Ω0∩∂Ω−, w0(x)=u(x), x∈∂Ω0∩∂Ω+. Thenv∈E1andw∈E2, andu=v+w.
(2) The decomposition is unique, i.e., ifu=v+w=0 forv∈E1andw∈E2, thenv=w=0.
Indeed we havev(x)=u(x)=0∀x∈Ω−andw(x)=u(x)=0∀x∈Ω+, thenw0|∂Ω0=0, hencew0=0 by the maximum principle, andv=w=0.
(3) Define the mappingπ:u→(v, w)fromEtoE1⊕E2. It is linear and bounded. Moreover, it is also surjective.
Indeed, for(v, w)∈E1⊕E2, let u(x)=
v(x), x∈Ω−, v(x)+w(x), x∈Ω0, w(x), x∈Ω+,
thenu∈Eandπ(u)=(v, w). ThereforeEis isomorphic toE1⊕E2by Banach Theorem. 2 Again, the spacesE1andE2are decomposable. Indeed, let
E3= v3∈E1
Ω
∇v3· ∇φ dx=0, ∀φ∈H01(Ω0)∪H01(Ω−)
and
E4= v4∈E2
Ω
∇v4· ∇φ dx=0, ∀φ∈H01(Ω+)
.
It is easy to verify that E1=
H01(Ω−)∩Lpa−+1(Ω−)
⊕H01(Ω0)⊕E3 and
E2=H01(Ω+)⊕E4.
These decompositions were used in [16] in the computation of the critical groups ofI at infinity. In order to compute these groups, we follow the method in Section 4 of [16], by introducing a family of functionalsIs,s∈ [0,1], as follows:
Is(u)=1 2
Ω
|∇v|2+2s∇v· ∇w+ |∇w|2−λ
v2+2sv·w+w2 dx
−
Ω
a+(x)
q+1|w|q+1−a−(x)
p+1|v|p+1+sH (x, v+w)
dx
for(v, w)∈E1×E2.
We note thatI1(v, w)=I (v+w)=I (u)and thatI0(v, w)=J−(v)+J+(w)is of separable variables, where J−(v)=
Ω
1 2
|∇v|2−λv2
+a−(x) p+1|v|p+1
dx
and
J+(w)=
Ω
1 2
|∇w|2−λw2
−a+(x) q+1|w|q+1
dx.
We shall compute the critical groups ofIat infinity via those ofI0. One can easily figure out the critical groups forJ± and so does forI0. Thus it remains to show that the critical groups forIs are invariant alongs∈ [0,1].
Definition 2.3.LetCbe a constant. A sequence{uk} = {(vk, wk)} ⊂E1⊕E2is said a weak Palais–Smale sequence forIs, if
Is(vk, wk)C and Is(vk, wk)
E∗=o ukE
=o
vkE1+ wkE2
.
Obviously, a Palais–Smale sequence is a weak Palais–Smale sequence.
Lemma 2.4.Assume(A1), (A2),
(A3) h(x, ξ )=o(|ξ|)as|ξ| → +∞uniformly inx, (A4) λ /∈σ (Ω0).
Then any weak Palais–Smale sequence forIsis bounded inE.
Proof. From the definition
Is(vk, wk)C and Is(vk, wk)
E∗=o
vkE1+ wkE2
, we have
Is(vk, wk)=1 2
Ω
|∇vk|2+2s∇vk· ∇wk+ |∇wk|2−λ
vk2+2svk·wk+wk2 dx
−
Ω
a+(x)
q+1|wk|q+1−a−(x)
p+1|vk|p+1+sH (x, vk+wk)
dxC and
Is(vk, wk), (vk, wk)
=
Ω
|∇vk|2+2s∇vk· ∇wk+ |∇wk|2−λ
vk2+2svk·wk+wk2 dx
−
Ω
a+(x)|wk|q+1−a−(x)|vk|p+1+sh(x, vk+wk)(vk+wk) dx
=o
vk2E1+ wk2E2 .
These imply 1
2− 1 q+1
Ω
a+(x)|wk|q+1dx− 1
2− 1
p+1
Ω
a−(x)|vk|p+1dxC+o
vk2E1+ wk2E2
. (2.1) Now letη∈C∞(Ω)satisfy
η(x)= 1, x∈Ω−, 0, x∈Ω+, again, we have
Is(vk, wk), (ηvk, ηwk)
=
Ω
∇vk· ∇(ηvk)+s∇(ηvk)· ∇wk+s∇(ηwk)· ∇vk+ ∇wk· ∇(ηwk) dx
+
Ω
a−(x)|vk|p+1dx+O uk22
=o ukE
ηukE=o uk2E
. (2.2)
The first integral on the right-hand side of (2.2) equals
Ω
1
2∇v2k· ∇η+η|∇vk|2+sη∇vk· ∇wk
dx
+
Ω
sη∇vk· ∇wk+s∇η· ∇(vkwk)+1
2∇w2k· ∇η+η|∇wk|2
dx
=
Ω
(1−s)
|∇vk|2+ |∇wk|2
+s∇(vk+wk)2
η dx−1 2
Ω
|vk|2+ |wk|2+2svkwk
η dx. (2.3)
Substituting (2.3) into (2.2) it follows
Ω
a−(x)|vk|p+1dxO uk22
+o uk2E
. (2.4)
Combining (2.4) with (2.1) we get
Ω
a+(x)|wk|q+1dxC+O uk22
+o uk2E
. (2.5)
The assumptionIs(vk, wk)Cand (2.5) imply 1
2
Ω
|∇vk|2+2s∇vk· ∇wk+ |∇wk|2 dx+
Ω
a−(x)|vk|p+1dxC+O uk22
+o uk2E
. (2.6)
However, there is a constantC1such that
Ω
|∇v|2+2s∇v· ∇w+ |∇w|2
dx(1−s)
Ω
|∇v|2+ |∇w|2 dx+s
Ω
|∇u|2dx C1
Ω
|∇v|2+ |∇w|2
dx. (2.7)
Inserting (2.7) into (2.6) we have
Ω
|∇vk|2+ |∇wk|2 dx+
Ω
a−(x)|vk|p+1dxC+O uk22
+o uk2E
. (2.8)
Therefore
uk2EC
1+ uk22
(2.9) for some constantC.
In view of (2.9), it remains to prove thatvk2+ wk2is bounded. This is proved by contradiction. Suppose not, we have a weak Palais–Smale sequence{(vk, wk)}satisfyingvk2+ wk2→ +∞ask→ ∞. Define
˜
uk= uk vk2+ wk2
, v˜k= vk vk2+ wk2
, w˜k= wk vk2+ wk2
,
(2.8) implies that{˜vk},{ ˜wk}, and{ ˜uk}are bounded. By Lemma 2.1,Eis reflexive, so we may assume, after a subse- quence that
˜
vk v0, w˜k w0, u˜k u0 weakly inH01(Ω) and
˜
vk→v0, w˜k→w0, u˜k→u0 strongly inL2(Ω)
withu0=v0+w0. Thenu0=0 asv02+ w02=1. Settingzk=∂vIs(vk, wk)andφ=(v vk
k2+wk2)2 we have zk, φ =
Ω
|∇ ˜vk|2−λv˜2k dx+s
Ω
(∇ ˜vk· ∇ ˜wk−sλv˜k· ˜wk) dx +
vk2+ wk2
p−1
Ω
a−(x)|˜vk|p+1dx+o(1).
Therefore
vk2+ wk2
p−1
Ω
a−(x)|˜vk|p+1dxC
and
Ω
a−(x)|˜vk|p+1dx→0
byp >1 andvk2+ wk2→ +∞.It follows fromv˜k v0that
Ω
a−(x)|v0|p+1dx lim
k→∞
Ω
a−(x)|˜vk|p+1=0 and then supp(v0)⊂Ω0.
Similarly, we have
Ω
a+(x)|w0|q+1dx=0
and supp(w0)⊂Ω0by computing∂wIs(vk, wk), wk
(vk2+wk2)2. Hencew0=0 provided byw0=0 inΩ0and then supp(u0)⊂Ω0.
Let us chooseφ∈H01(Ω0)as an element inE, then we have 1
vk2+ wk2
Is(vk, wk), φ
=
Ω
(∇ ˜vk· ∇φ−λv˜kφ) dx+s
Ω
(∇ ˜wk· ∇φ−λw˜kφ) dx
=
Ω
(∇ ˜uk· ∇φ−λu˜kφ) dx+s
Ω
(∇ ˜wk· ∇φ−λw˜kφ) dx=o(1).
Thus
Ω
(∇u0· ∇φ−λu0φ) dx=0 ∀φ∈H01(Ω0)
byw˜k0. According to the assumptionλ /∈σ (Ω0),u0=0 inΩ0andu0=0 inE. This is a contradiction. 2 Lemma 2.5.Under the assumption of Lemma2.4, every weak Palais–Smale sequence ofIs contains a convergent subsequence inE.
Proof. (1) Let{uk} = {(vk, wk)}be a weak Palais–Smale sequence ofIsfor somes∈ [0,1]. According to Lemma 2.4, {uk}is bounded inE.
(2) Applying Lemma 2.1, there is au∗∈Esuch thatuk u∗inE. After a subsequence we have uk(x)→u∗(x) a.e. inΩ
and
uk→u∗ inL2(Ω)and inLq+1(Ω).
(3) By the definition of weak Palais–Smale sequence, we have Is(uk)−Is(u∗), uk−u∗
=o ukE
uk−u∗E+o(1)=o(1) (2.10)
provided by the boundedness of{uk}inE. On the other hand, there holds Is(uk)−Is(u∗), uk−u∗
= uk−u∗2s +
Ω
a−
|uk|p−1uk− |u∗|p−1u∗
(uk−u∗) dx+o(1), (2.11) where
u2s =
Ω
|∇v|2+2s∇v· ∇w+ |∇w|2 dx
is an equivalent norm ofuH1
0 =(
Ω|∇u|2dx)1/2by (2.7). From an elementary inequality, there is a constantC such that
C
Ω
a−|uk−u∗|p+1dx
Ω
a−
|uk|p−1uk− |u∗|p−1u∗
(uk−u∗) dx. (2.12)
Combining (2.10)–(2.12) we getuk→u∗inE. 2 As a consequence of Lemmas 2.4 and 2.5 we have
Corollary 2.6.Under the assumptions of Lemma2.5, the functionalI satisfies the Palais–Smale condition onE.
The proofs of Lemmas 2.4 and 2.5 also yield (see Proposition 3.3 in [16]).
Theorem 2.7.Under the assumptions of (A1), (A2), (A3)and (A4), there are constants Aand δ >0 such that
∀s∈ [0,1], Is(u)
E∗δuE ifIs(u)A.
Using the deformationIs we have the following theorem on the critical groups at infinity, its proof is referred to Theorem 4.1 in [16].
Theorem 2.8.Under the assumptions of (A1), (A2), (A3)and(A4), all critical groups ofI at infinity are trivial, i.e., C∗(I,∞):=H∗
E, Ia
= {0}, ∗ =0,1,2, . . .
foraA, whereAis the constant in Theorem2.7andIa= {u∈E|I (u)a}.
Also one defines I±(u)=
Ω
1 2
|∇u|2−λu2±
−a+(x)
q+1|u±|q+1+a−(x)
p+1|u±|p+1−H±(x, u)
dx,
whereu±=max{±u,0},H±(x, ξ )is the primitive ofh±(x, ξ )and h±(x, ξ )= h(x, ξ ), ±ξ 0,
0, otherwise.
We have the following facts forI±: (1)I±(u)=I (u)if±u0;
(2) LetK andK±be the critical sets ofI andI±, respectively, thenK±=K∩(±P ), whereP = {u∈H01(Ω)| u(x)0 a.e.x∈Ω}.
As we proved in [16], the following theorem holds.
Theorem 2.9. Under the assumptions of (A1), (A2), (A3)and (A4), the critical groups ofI± at infinity are well defined and trivial, i.e.,
C∗(I±,∞)= {0}, ∗ =0,1,2, . . . .
Remark 2.10.In the following sections we shall consider the functional on the spaceC01(Ω)rather than the spaceE.
LetI˜=I|C1
0(Ω)andI˜±=I±|C1
0(Ω). It is well known [25] that H∗
C01(Ω),I˜a
=H∗ E, Ia
= {0}, ∗ =0,1,2, . . . , and
H∗
C01(Ω),I˜±a
=H∗ E, I±a
= {0}, ∗ =0,1,2, . . . .
This means that the critical groups at infinity for bothIandI˜are trivial.
Remark 2.11.The results in this section hold if the term|u|q−1uis replaced byg(u)satisfying:
(g1) |g(u)|C(1+ |u|q) u∈R,
(g2) there are constantsθ >2 andR >0 such that g(u)uθ G(u) >0, |u|R
whereG(u)=u
0 g(s) ds.
3. The heat flow
LetT >0 andΩT =(0, T )×Ωandφ∈C01(Ω), in this section we study theL∞a priori estimate for the solution v∈C1(ΩT)of the nonlinear heat equation:
∂tv(t, x)=(+λ)v+a+(x)|v|q−1v−a−(x)|v|p−1v+h(x, v), (t, x)∈ΩT, v(t, x)=0, (t, x)∈ [0, T )×∂Ω,
v(0, x)=φ(x), x∈Ω, (3.1)
and explain how the heat flow can be used in Morse theory for the associated functional. It is well known that the solutionvmay blow up at finite time. However, if one adds a finite energy condition, the blow up phenomena can be ruled out, see Ackermann, Bartsch, Kaplicky and Quittner [1], Cazenave and Lions [11], Chang [14], Giga [19] and Quittner [27]. For technical reasons we assume
(A2) q < p∗1:=
⎧⎪
⎪⎪
⎨
⎪⎪
⎪⎩
+∞, n=1,
7, n=2,
18
5, n=3,
9n2−4n+16√ n(n−1)
(3n−4)2 , n >3
andp >1. We shall prove
Theorem 3.1. Assume(A1), (A2)and (A3). If v is a solution of (3.1)which blows up at a finite time T, then I (v(t,·))→ −∞ast→T −0.
Theorem 3.2.Assume(A1), (A2)and
(A3) h(x, ξ )=o(|ξ|)as|ξ| → ∞uniformly inx, (A4) λ /∈σ (Ω0),
whereΩ0=Ω\(Ω+∪Ω−)andσ (Ω0)is the spectrum of−onΩ0with Dirichlet boundary condition. Ifvis a global solution of (3.1)satisfyingI (v(t,·))−C0for some constantC0, then theω-limit setω(φ)= ∅and contains critical points ofI.
Combining Theorems 3.1 and 3.2 we have
Theorem 3.3.Assume(A1), (A2), (A3)and (A4), the Morse theory for isolated critical points of the functionalI holds in the Banach spaceC01(Ω). The Morse theory is related to the order preserving parabolic semi-flow.
In the remaining of this section we give details of the proofs.
Lemma 3.4.Ifv∈C1(ΩT)is a solution of(3.1)and ifI (v(t,·))−C0∀t∈ [0, T]for some constantC0, then there is a constantCT(φ, C0)depending only onφandC0such that
v(t,·)
2CT(φ, C0).
Proof. It follows from the Hölder inequality that v(t,·)−φ
2= t 0
∂sv(s,·) ds 2
T1/2 T
0
∂sv(s,·)2
2ds 1/2
T
I (φ)−I
v(T ,·)1/2
provided by T 0
∂sv(s,·)2
2ds= − T 0
d dtI
v(t,·)
dt=I (φ)−I v(T ,·)
.
Now the conclusion follows easily. 2
Lemma 3.5.Under the assumptions of(A1), (A2), (A3)and(A4), ifv∈C1(ΩT)is a global solution of (3.1)and if I (v(t,·))−C0∀t∈ [0,+∞)for some constantC0, then there is a constantC1such that
v(t,·)
2C1, t∈ [0,∞).
Proof. We prove it by contradiction. Suppose not, there exists a sequence{tk}such thatv(tk,·)2k. According to Lemma 3.4,tk→ +∞and we may assumetk+1−tk>1. By Lemma 2.4, a contradiction follows if we can construct a weak Palais–Smale sequence ofIclose to{v(tk,·)}.
(1) Claim:∃η >0 such that v(t,·)
2k
2, ∀t∈ [tk−η, tk]. Indeed,∀s < tk
v(s,·)2
2−v(tk,·)2
2=2
tk
s
Ω
v∂tv(t, x) dx dt
2
tk s
v(t,·)2
2dt
tk
s
∂tv(t,·)2
2dt 1/2
I (φ)+C0
+
tk
s
v(t,·)2
2dt.
From the Gronwall inequality v(tk,·)2
2v(s,·)2
2+C1
e(tk−s), stk, whereC1=I (φ)+C0. Thus we can find anη >0 satisfying
v(t,·)
2k
2, t∈ [tk−η, tk]andk1.
(2) Claim:∃sk∈ [tk−η, tk]such that
Ω
∂tv(sk, x)2dxI (φ)+C0
η .
This is due to the fact:
+∞
0
Ω
∂tv(t, x)2dx dtI (φ)+C0.
(3) Now letuk=v(sk,·), then I (uk)=I
v(sk,·) I (φ) and∀ψ∈E,
I(uk), ψ=
Ω
∂tv(sk, x)ψ dx
Ω
∂tv(sk, x)2dx 1/2
ψ2
η−1/2
I (φ)+C0
1/2
ψE. That is,
I(uk)
E∗η−1/2
I (φ)+C01/2
=o uk2
=o ukE
.
Therefore,{uk}is a weak Palais–Smale sequence and is bounded by Lemma 2.4. This is a contradiction. 2 Having Lemmas 3.4 and 3.5, now we can prove the main estimate.
Lemma 3.6.LetJ= [0, T]. Assume(A1), (A2)and(A3), ifv∈C1(ΩT)is a solution of (3.1)satisfying v(t,·)
2CJ, t∈ [0, T],
whereT is either finite or infinite, thenv(t,·)∞is bounded onJ. Proof. We estimatev(t,·)∞in various subdomains ofΩ separately.
(1) For >0, letΩ+, be the-neighborhood ofΩ+ inΩ. SinceΩ+∩Ω−= ∅, there exists >0 such that Ω+,∩Ω−= ∅, then∂Ω+,\∂Ω⊂Ω0and(∂Ω+,\∂Ω)∩Ω+= ∅.We fixfrom now on and∀(t0, x0)∈ΩT and
∀R >0, denoteQR(t0, x0)=(t0−R2, t0+R02)×BR(x0). Nowvsatisfies
∂tv=v+λv+h(x, v), (t, x)∈ [0, T] ×Ω0.
In case QR(t0, x0)⊂Ω0T, according to Moser’s iteration on the local boundedness of the weak solutionv onΩ0, see [23], there existR0andC >0 such that∀R∈(0, R0], there holds:
sup
QR
2
(t0,x0)
v(t, x) C
Rn+2
QR(t0,x0)
v(t, x)2dx dt 1/2
+C.
In casex0∈∂Ω∩Ω0andt0>0 we also have sup
QR
2
(t0,x0)∩ΩT
v(t, x) C
Rn+2
QR(t0,x0)∩ΩT
v(t, x)2dx dt 1/2
+C.
IfT is finite, one fixesR >0, then[R42, T] ×(∂Ω+,\∂Ω)is covered by finitely many, sayM, cylinders in the family{QR
2
(t0, x0)|(t0, x0)∈ [R42, T] ×(∂Ω+,\∂Ω)}. Thus sup
[R42,T]×(∂Ω+,\∂Ω)
v(t, x)M C
Rn+2
ΩT
v(t, x)2dx dt 1/2
+C
. (3.2)
IfT = +∞, applying the same arguments to the domains[k−1, k+1] ×Ω0,k=2,3, . . ., we get sup
[k−1,k+1]×(∂Ω+,\∂Ω)
v(t, x)M C
Rn+2
QR(t0,x0)∩ΩT
v(t, x)2dx dt 1/2
+C
.
The numberMand all constants are independent ofk, so we have sup
[R42,+∞)×(∂Ω+,\∂Ω)
v(t, x)M C
Rn+2
QR(t0,x0)∩ΩT
v(t, x)2dx dt 1/2
+C
. (3.3)
According to the assumption:
v(t,·)
2CJ, t∈ [0, T],
the right-hand sides of (3.2) and (3.3) are bounded byM[C+( C
Rn+2)1/2CJ], and thenv is bounded on [R42, T] × (∂Ω+,\∂Ω).
(2) By a standard argument of the variation of constant formula, we have sup
[0,R22]×Ω
v(t, x)C1sup
Ω
φ(x). (3.4)
Combining (3.4) with the estimates in the last step we obtain
v(t, x)C2, (t, x)∈J×(∂Ω+,\∂Ω), (3.5) whereC2is a constant depending onCJ.
(3) Let us consider Eq. (3.1) on the subdomainΩ\Ω+,:
∂tv1(t, x)=(+λ)v1−a−(x)|v1|p−1v1+h(x, v1), ∀(t, x)∈J×(Ω\Ω+,), v1(t, x)=v(t, x), ∀(t, x)∈J×(∂Ω+,\∂Ω),
v1(t, x)=0, ∀(t, x)∈J×
∂Ω∩(Ω\Ω+,) , v1(0, x)=φ(x), ∀x∈Ω\Ω+,.
By the uniqueness,
v(t, x)=v1(t, x), ∀(t, x)∈J×(Ω\Ω+,).
We apply the weak maximum principle due to De Giorgi’s iteration [23], it follows
vL∞(J×(Ω\Ω+,))= v1L∞(J×(Ω\Ω+,))C3 (3.6)
for a constant depending onC2,φ,λ,handCJ.
(4) Finally, we consider Eq. (3.1) on the subdomainΩ+,:
∂tv2(t, x)=(+λ)v2+a+(x)|v2|q−1v2+h(x, v2), ∀(t, x)∈J×Ω+,, v2(t, x)=v(t, x), ∀(t, x)∈J×(∂Ω+,\∂Ω),
v2(t, x)=0, ∀(t, x)∈J×(∂Ω∩Ω+,), v2(0, x)=φ(x), ∀x∈Ω+,.
Sinceq < p1∗is assumed, after the iteration estimate due to Quittner, see [1], and (3.6), we have
vL∞(J×Ω+,)= v2L∞(J×Ω+,)C4. (3.7)
In summary we have proved the boundedness of supt∈Jv(t,·)∞. 2
Proof of Theorem 3.1. This follows from Lemmas 3.4 and 3.6. In fact, ifI (v(t,·))−C0for some constantC0>0, thenv(t,·)2CT. By Lemma 3.6,v(t,·)∞is bounded on[0, T]. This contradicts with the assumption thatT is the blow up time. 2
Proof of Theorem 3.2. Letv be the global solution of (3.1). After Lemma 3.6 we havev(t,·)∞C2 ∀t. Then by a standard argument, see [10], Theorem 9.4.2, ω(φ)= ∅and∀u∈ω(φ), it is a critical point ofI withI (u)= limt→+∞I (v(t,·))−C0. According to the regularity theory, the topology can be taken onC01(Ω). 2
Proof of Theorem 3.3. Assume now that the functionalI has only isolated critical points. If the global orbitO(φ)= {v(t,·)|t∈R+}exists and satisfiesI (v(t,·))−C0, again by Lemma 3.6 and Theorem 9.4.2 in [10], the limit set ω(φ)must be a singleton. Moreover, the limit exists inC01(Ω)topology. 2
The heat flow v(t, φ)with initial value φ is used to replace the pseudo-gradient flow. In order to establish the Morse theory for the isolated critical points, it is sufficient to prove the following deformation lemma via the heat flow: LetKbe the critical set ofI˜,a < d, ifK∩ ˜I−1(a, d] = ∅and ifKa=K∩ ˜I−1(a)is isolated, thenI˜ais a strong deformation retract ofI˜b, whereI˜=I|C1
0(Ω).
Indeed,∀φ∈ ˜I−1(a, d], letTφ>0 be the maximal existence time ofv(t, φ), and letO(φ)= {v(t, φ)|t∈ [0, Tφ)} be its orbit. Lettφ be the arriving time of the orbitO(φ) at the levelI˜−1(a). According to Theorems 3.1 and 3.2, tφ>0 is either finite,tφ< TφifI (ω(φ)) < a, or˜ tφ= +∞andω(φ)∈ ˜I−1(a). In both cases, we can rescale the time variable as follows: Let
τ =τ (t )=I (φ)˜ − ˜I (v(t, φ))
I (φ)˜ −a , ∀φ∈ ˜I1(a, d]. Then we have
τ (0)=0,
τ (+∞)=1 ifω(φ)∈ ˜I−1(a), τ (+∞) >1 ifI (ω(φ)) < a.˜