Critical point result of Schechter type in a Banach space
Hannelore Lisei
Band Orsolya Vas
Babes,–Bolyai University, Faculty of Mathematics and Computer Science, Kog˘alniceanu Str. 1, Cluj-Napoca, RO – 400084 , Romania
Received 11 June 2015, appeared 22 March 2016 Communicated by Gennaro Infante
Abstract. Using Ekeland’s variational principle we give a critical point theorem of Schechter type for extrema on a sublevel set in a Banach space. This result can be applied to localize the solutions of PDEs which contain nonlinear homogeneous opera- tors.
Keywords: critical point, Ekeland’s variational principle, Palais–Smale type compact- ness condition,p-Laplacian.
2010 Mathematics Subject Classification: 47J30, 47J05, 58E05, 34B15.
1 Introduction
Finding the set of solutions of certain PDEs is closely related to the investigation of the critical points of a certain functional defined on an appropriate Hilbert or Banach space. Mountain pass theorems, saddle point theorems, linking theorems, mountain cliff theorems give suffi- cient conditions for the existence of a minimizer for a certain differentiable functional defined on the whole space or on a bounded region (for example, see [3,16–19,21]).
In [13–15] R. Precup studies critical point theorems of Schechter type forC1functionals on a closed ball and also on a closed conical shell in a Hilbert space by using Palais–Smale type compactness conditions and also Leray–Schauder conditions on the boundary. These results can be used successfully to localize the solutions of PDEs involving the Laplace operator.
In our paper we improve the above mentioned Schechter type results (on a ball) for sub- level sets in locally uniformly convex Banach spaces and then apply our result for localizing the solutions for p-Laplace type equations on bounded, and also on unbounded domains.
The paper is structured as follows: Section 2 contains certain preliminaries concerning duality mappings on Banach spaces and the assumptions for the critical point problem which we are investigating. Section 3 states the main result of our paper. Section 4 presents two examples of localizing the solutions for problems containing the p-Laplacian.
BCorresponding author. Email: hanne@math.ubbcluj.ro
2 Preliminaries
Let X be a real Banach space, X∗ its dual, h·,·idenotes the duality between X∗ and X. The norm onXand onX∗ is denoted by k · k.
A continuous function ϕ : R+ → R+ is called a normalization function if it is strictly increasing, ϕ(0) =0 andϕ(r)→∞forr→∞.
The duality mapping corresponding to the normalization function ϕis the set valued operator Jϕ: X→ P(X∗)defined by
Jϕx =nx∗ ∈ X∗ :hx∗,xi= ϕ(kxk)kxk,kx∗k= ϕ(kxk)o, x ∈X.
Assumption (A1): XandX∗ are locally uniformly convex reflexive Banach spaces.
Observe that X∗ is strictly convex, because a locally uniformly convex Banach space is also strictly convex, see [5, Theorem 3, p. 31]. Then it follows that card(Jϕx) = 1 by [6, Proposition 1, p. 342]. Hence, Jϕ :X→X∗
hJϕx,xi= ϕ(kxk)kxk and kJϕxk= ϕ(kxk). The following result holds.
Theorem 2.1. [6, Theorem 5, p. 345]Let X be a reflexive, locally uniformly convex Banach space and Jϕ: X→X∗. Then Jϕis bijective and its inverse J−ϕ1is bounded, continuous and monotone. Moreover, it holds Jϕ−1 = χ−1J∗ϕ−1, whereχ : X → X∗∗ is the canonical isomorphism between X and X∗∗ and J∗
ϕ−1 :X∗ →X∗∗is the duality mapping on X∗ corresponding to the normalization function ϕ−1. We consider ¯J : X∗ → Xdefined by ¯J = J−ϕ1. By Theorem2.1 it follows that ¯J is bounded, continuous and monotone. Forw∈ X∗denote v= Jϕ−1wand compute
hw, ¯Jwi=hJϕv,vi= ϕ(kvk)kvk=kJϕvkkvk=kwkkJ−ϕ1wk and
kJw¯ k=kJ−ϕ1wk=kχ−1J∗ϕ−1wk=kJϕ∗−1wkX∗∗ = ϕ−1(kwk). We conclude
hw, ¯Jwi= ϕ−1(kwk)kwk and kJw¯ k= ϕ−1(kwk) for eachw∈ X∗. (2.1) Fix 0<R.
Assumption (A2): Let H:X →Rbe of classC1 such that the level set NR={u∈ X: H(u) =R}is non-void and bounded ,
uinf∈NRhH0(u),ui>0, and the operatorH0 maps bounded sets into bounded sets.
We denote
XR= {u∈X: H(u)≤R}.
Assumption (A3): Let F:XR→Rbe of classC1 such thatFis bounded by below onXR.
We introduce some auxiliary mappings:
D: NR →X∗, D(u) =F0(u)− hF0(u),ui hH0(u),uiH
0(u), E: NR →X, E(u) = JD¯ (u)− hH0(u), ¯JDui
hH0(u),ui u.
Lemma 2.2. Assume that (A1)holds, and that F : XR → Rand H : X → Rare C1functions. For all u∈ NRthe following properties hold:
(1) hH0(u),E(u)i=0,
(2) hF0(u),E(u)i= ϕ−1(kD(u)k)kD(u)k. Proof. Letu∈ NR be arbitrary. We compute
hH0(u),E(u)i=
H0(u), ¯JD(u)−hH0(u), ¯JDui hH0(u),ui u
=0.
Observe that
hD(u),ui=
F0(u)− hF0(u),ui hH0(u),uiH
0(u),u
=0. (2.2)
By using the statement (1) of this lemma, by (2.2) and (2.1) we have hF0(u),E(u)i=
F0(u)− hF0(u),ui hH0(u),uiH
0(u),E(u)
=hD(u),E(u)i
=
D(u), ¯JD(u)i − hH0(u), ¯JDui
hH0(u),ui hD(u),u
= hD(u), ¯JD(u)i= ϕ−1(kD(u)k)kD(u)k.
3 Main result
Theorem 3.1. Assume that(A1),(A2)and(A3)are satisfied. Then, there exists a sequence(xn)n ⊂ XRsuch that F(xn)→infF(XR)and one of the following statements hold
(a) F0(xn)→0as n→∞;
(b) for each n∈Nwe have H(xn) =R,hH0(xn), ¯JF0(xn)i ≤0and F0(xn)− hF0(xn),xni
hH0(xn),xniH
0(xn)→0 as n→∞. If, in addition, there exists a∈R+such that
hH0(x), ¯JF0(x)i ≥ −a for each x∈ NR,
and F satisfies a Palais–Smale type compactness condition (i.e. any sequence satisfying(a)or(b)has a convergent subsequence) and the following boundary condition holds
F0(x) +µH0(x)6=0 for anyµ>0, x∈ NR, (3.1) then there exists x ∈XR such that
F(x) =infF(XR) and F0(x) =0.
Proof. By Ekeland’s variational principle, see [8, Theorem 1, p. 444], applied for XR (we use here thatHis continuous, henceXRis a closed set), the distanced(x,y) =kx−yk, the function F(which is continuous and bounded by below, see (A3)),ε= n1 and foru∈ XRsuch that
F(u)≤infF(XR) + 1 n, it follows that there exists a sequence(xn)nin XRsuch that
F(xn)≤F(u)≤infF(XR) + 1 n, and
F(xn)<F(y) + 1
nkxn−yk for eachy ∈XR\ {xn}. (3.2) This yieldsF(xn)→infF(XR).
Since(xn)nbelongs toXR, we distinguish two cases:
(1) there exists a subsequence of(xn)n, still denoted by(xn)n, such that H(xn)< Rfor each n∈N;
(2) there exists a subsequence of(xn)n, still denoted by(xn)n, such that H(xn) =Rfor each n∈N.
Case (1)Fix n∈ N. Lett >0 and z∈ X such thatkzk=1. Since His a continuous function and H(xn)< R, we have that there exits δ > 0 (small enough) such that H(xn−tz) < Rfor eacht ∈(0,δ). Hencexn−tz ∈XR\ {xn}for eacht∈ (0,δ)and by (3.2) it holds
F(xn)−F(xn−tz)< t n. By takingt &0 it follows
hF0(xn),zi ≤ 1 n.
Butz ∈ X with kzk = 1 was arbitrary chosen, hence kF0(xn)k ≤ n1, which yields F0(xn) → 0 as n → ∞. Hence we constructed a sequence (xn)n which satisfies the statement (a) of this theorem.
Case (2)Fix n∈ N. We have H(xn) = R. Letz ∈ Xsuch that kzk= 1. We use the definition of the Fréchet derivative of H: for eachε >0 there exists δε >0 such that for eacht ∈ (0,δε) we have
−εt< H(xn−tz)−H(xn) +hH0(xn),tzi<εt.
Hence,
R−εt−thH0(xn),zi< H(xn−tz)<R+εt−thH0(xn),zi for eacht ∈(0,δε). (3.3)
•IfhH0(xn),zi>0: by takingε=hH0(xn),ziin (3.3) we get H(xn−tz)< R for each t∈(0,δε).
Hencexn−tz∈ XR\ {xn}fort ∈(0,δε). By (3.2) it follows that fort ∈(0,δε) F(xn)−F(xn−tz)< t
n,
then byt&0 we get hF0(xn),zi ≤ 1
n for eachz∈ Xwithkzk=1 and hH0(xn),zi>0. (3.4)
• If hH0(xn),zi = 0: we approximate z by a sequence (zk)k such that kzkk = 1 and hH0(xn),zki > 0 for each k ∈ N, while kz−zkk → 0 as k → ∞. Since hH0(xn),·i is con- tinuous, we havehH0(xn),zki → hH0(xn),zi=0 ask→∞.
Let k ∈ N be fixed. By considering (3.3) for zk instead of z and ε = hH0(xn),zki for t∈(0,δε)we get
H(xn−tzk)< R.
Then, xn−tzk ∈ XR\ {xn}fort ∈(0,δε). By (3.2) we obtain fortsufficiently small F(xn)−F(xn−tzk)< t
n, which yields
hF0(xn),zki ≤ 1 n. Butkz−zkk →0 as k→∞, hence
hF0(xn),zi ≤ 1 n. This inequality and (3.4) imply
hF0(xn),zi ≤ 1
n for eachz∈ Xwithkzk=1 and hH0(xn),zi ≥0. (3.5) Further we have two possible cases.
Case (2a) There exists a subsequence of (xn)n, which we still denote by (xn)n, such that hH0(xn), ¯JF0(xn)i>0: by takingz = k 1
JF0(xn)kJF¯ 0(xn)in (3.5) we get hF0(xn), ¯JF0(xn)i ≤ 1
nkJF0(xn)k. By the property (2.1) of ¯J it follows that
hF0(xn), ¯JF0(xn)i= ϕ−1(kF0(xn)k)kF0(xn)k and kJF¯ 0(xn)k= ϕ−1(kF0(xn)k) which yields
kF0(xn)k ≤ 1 n,
hence F0(xn)→ 0 asn → ∞ and we obtained a sequence(xn)n which satisfies the statement (a) of this theorem.
Case (2b) There exists a subsequence of (xn)n, which we still denote by (xn)n, such that hH0(xn), ¯JF0(xn)i ≤0: by takingz= kE(1x
n)kE(xn)in (3.5) (ifkE(xn)k=0, then by Lemma2.2(2) we getkD(xn)k=0) and by Lemma2.2(2) we get
ϕ−1(kD(xn)k)kD(xn)k=hF0(xn),E(xn)i ≤ 1
nkE(xn)k. Hence,
ϕ−1(kD(xn)k)kD(xn)k ≤ 1
nkE(xn)k. (3.6)
Denote the kernel ofH0(xn)byKn={x ∈X:hH0(xn),xi=0}and the projection mapping Pn: X→ Kn by Pnv=v− hH0(xn),vi
hH0(xn),xnixn.
Since v ∈ X 7→ hH0(xn),vi is linear and continuous, it follows that Pn is also linear and continuous.
Since (xn)n ⊂ NR and the level set NR is bounded, it follows that (xn)n is a bounded sequence. By the assumption onH0 it follows that(H0(xn))nis also bounded and there exists
0< βR := inf
u∈NR
hH0(u),ui ≤ hH0(xn),xni for eachn∈N. We write
kPnvk ≤ kvk+ kH0(xn)kkvk
infn∈NhH0(xn),xnikxnk ≤
1+ kH0(xn)kkxnk βR
kvk for eachv ∈X.
Hence there existsαR>0 (independent ofn) such that
kPnvk ≤αRkvk for eachv∈ X.
We takev = JD¯ (xn)to get PnJD¯ (xn) =E(xn)and
kE(xn)k ≤αRkJD¯ (xn)k=αRϕ−1(kD(xn)k) for each n∈N.
Then by (3.6) we have
ϕ−1(kD(xn)k)kD(xn)k ≤ αR
n ϕ−1(kD(xn)k) for each n∈N.
This yields D(xn) → 0 as n → ∞. Hence we constructed a sequence (xn)n which satisfies statement (b) of this theorem.
If, in addition, Fsatisfies the (PS) type compactness condition.
Case (a) F0(xn) → 0 as n → ∞ and there exist x ∈ XR and a subsequence (xnk)k such that kxnk−xk →0 ask→∞. Since Fis aC1function, we get F0(x) =0 and by the construction of (xn)n we haveF(xn)→infF(XR), henceF(x) =infF(XR).
Case (b)We haveD(xn)→ 0 asn → ∞, H(xn) = R andhH0(xn), ¯JF0(xn)i ≤ 0 for all n ∈ N and there exist x ∈ XR (XR is a closed set, since H is continuous) and a subsequence (xnk)k such that kxnk −xk → 0 as k → ∞. Hence F(x) = limk→∞F(xnk) = infF(XR), F0(x) = limk→∞F0(xnk)andx∈ NR, i.e. H(x) = R. ButD(xnk)→0 ask →∞, which implies
F0(x)− hF0(x),xi hH0(x),xiH
0(x) =0. (3.7)
Applying the operator ¯J we get
JF¯ 0(x)− hF0(x),xi hH0(x),xiJH¯
0(x) =0, which yields
hH0(x), ¯JF0(x)i − hF0(x),xi
hH0(x),xihH0(x), ¯JH0(x)i=0,
hH0(x), ¯JF0(x)i − hF0(x),xi hH0(x),xiϕ
−1(kH0(x)k)kH0(x)k=0. (3.8) Since we are investigating Case (b), it follows that(hH0(xn), ¯JF0(xn)i)n is a bounded sequence inR, hence there existb∈R,b≤0, and a subsequence, denoted again by(xnk)k, such that
hH0(xnk), ¯JF0(xnk)i →b.
Butkxnk−xk →0 ask →∞, hence
hH0(x), ¯JF0(x)i=b≤0.
Using (3.8) it follows
hF0(x),xi hH0(x),xiϕ
−1(kH0(x)k)kH0(x)k= b≤0.
Since hH0(x),xi > 0 (assumption on H0 and the fact that x ∈ NR as the limit of (xnk)k), we obtain
hF0(x),xi ≤0.
•IfhF0(x),xi=0, then (3.7) impliesF0(x) =0.
•IfhF0(x),xi<0, then (3.7) implies
F0(x) +µH0(x) =0 whereµ=−hF0(x),xi
hH0(x),xi >0 andx∈ NR, which contradicts the assumption (3.1) from the statement of this theorem.
4 Applications
4.1 Example 1
Consider the Sobolev space W01,p(Ω), where Ω is a bounded domain in Rk with Lipschitz continuous boundary and 1 < p<∞, equipped with the norm
kuk1,p = Z
Ω|∇u(x)|pdx 1p
, where
∇u= ∂u
∂x1, . . . , ∂u
∂xk
, |∇u|=
∑
k i=1∂u
∂xi 2!12
.
The Banach space (W01,p(Ω),k · k1,p)is uniformly convex, see [1, Theorem 3.6]. Moreover, it is also uniformly smooth (which is proved by using Clarkson’s inequalities [1, 2.38 Theorem, p. 44] and [4, Definition 2.4., p. 13]). The dual space(W01,p(Ω))∗ will be denoted byW−1,p0(Ω), where 1p+ p10 =1, and by [4, Theorem 2.10] it follows that it is uniformly convex.
The Rellich–Kondrachov Theorem states that the embeddingW01,p(Ω),→Lq(Ω)is compact for q∈ (1,p∗)(where p∗ = kkp−p if p < k and p∗ = ∞, if p ≥ k) and there existsCq > 0 such that
kukLq(Ω)≤Cqkuk1,p for eachu∈W01,p(Ω). (4.1)
In the context of our paper we choose (X,k · k) = (W01,p(Ω),k · k1,p), ¯J = J−ϕ1, where ϕ(t) =tp−1fort ∈R+and let H :W01,p(Ω)→Rbe given byH(u) = 1pkuk1,pp .
We consider the p-Laplacian operator−∆p :W01,p(Ω)→W−1,p0(Ω)defined by h−∆p(u),vi=
Z
Ω|∇u(x)|p−2∇u(x)∇v(x)dxfor all u,v∈W01,p(Ω).
It is known that the functional H is continuously Fréchet differentiable on W01,p(Ω) and H0 = −∆p. The operator −∆p is in fact the duality mapping Jϕ : W01,p(Ω) → W−1,p0(Ω) corresponding to the normalization function ϕ(t) = tp−1 fort ∈ R+, i.e. H0 = Jϕ, for details consult [6, Theorem 7 and Theorem 9]. In our example we have
NR =
v∈W01,p(Ω): 1
pkvk1,pp = R
and
XR =
v∈W01,p(Ω): 1
pkvk1,pp ≤ R
.
Assume that f : Ω×R → R is a Carathéodory function such that f(x, 0) 6= 0 for a.e.
x∈ Ωand
|f(x,s)| ≤a(x)|s|q−1+b(x) forx ∈Ω, s∈R,
where a ∈ L∞(Ω),b ∈ Lq−q1(Ω)are positive functions and q ∈ (1,p∗). Define the Nemytskii operatorNf :W01,p(Ω)→W−1,p0(Ω)by
Nf (u) (x) = f(x,u(x)).
We have Nf(W01,p(Ω)) ,→ Nf(Lq(Ω)) ⊂ Lq−q1(Ω) = (Lq(Ω))∗ ,→ W−1,p0(Ω) and Nf is a continuous function which maps bounded sets into bounded sets (see [9]).
Consider the following Dirichlet problem involving the p-Laplacian:
−∆pu= f(x,u) a.e. x ∈Ω and u
∂Ω =0. (4.2)
We callu∈W01,p(Ω)aweak solutionof (4.2) if for eachv∈W01,p(Ω)it holds Z
Ω|∇u(x)|p−2∇u(x)∇v(x)dx=
Z
Ω f(x,u(x))v(x)dx. (4.3) We introduceF:W01,p(Ω)→Rdefined by
F(u) = 1
pkuk1,pp −
Z
Ωh(x,u)dx, whereh:Ω×R→Rish(x,t) =
Z t
0 f(x,s)ds. We have (see [9, Theorem 7]) F0(u) =H0(u)−Nf (u).
The critical points ofFare the solutions of (4.3).
(A4) Assumptions for R: denote by C an upper bound for Cq and suppose that one of the following three assumptions is satisfied.
(1) If p>q: letR>0 be a solution of the inequality inR Rp
−1
p >Cqpq
−p
p kakL∞(Ω)Rq
−1
p +Cp1
−p p kbk
L
q q−1(Ω). (2) If p=q: assume 1>CpkakL∞(Ω) and let Rbe such that
R>
Cp1
−p p kbk
L
p p−1(Ω)
1−CpkakL∞(Ω)
p p−1
.
(3) If q> p: assume that 1> Cqpq−ppkakL∞(Ω)+Cp1−ppkbk
L
q
q−1(Ω) and letR > 0 to be a solution of the inequality inR
Rp
−1
p −Cqpq
−p
p kakL∞(Ω)Rq
−1
p >Cp1
−p p kbk
L
q q−1(Ω). Proposition 4.1. The following relation holds
F0(u) +µH0(u)6=0 for anyµ>0, u∈ NR, where R satisfies one of the three conditions mentioned in(A4).
Proof. We reason by contradiction: assume that there exist u ∈ NR and µ > 0 such that F0(u) +µH0(u) =0, which implies
(1+µ)hJϕ(u),ui=hNf(u),ui. (4.4) By our assumptions
hNf(u),ui=
Z
Ω f(x,u(x))u(x)dx
≤
Z
Ωa(x)|u(x)|q+b(x)|u(x)|dx≤ kakL∞(Ω)kukq
Lq(Ω)+kbk
L
q q−1(Ω)
kukLq(Ω). Using (4.1) and (4.4) we get
kuk1,pp ≤(1+µ)kuk1,pp ≤ kakL∞(Ω)kukq
Lq(Ω)+kbk
L
q q−1(Ω)
kukLq(Ω)
≤CqkakL∞(Ω)kukq1,p+Ckbk
L
q q−1(Ω)
kuk1,p.
Butu ∈NR implieskuk1,p= (pR)1p, and we obtain pR≤CqkakL∞(Ω)(pR)qp +Ckbk
L
q q−1(Ω)
(pR)1p, which yields
Rp
−1
p ≤Cqpq
−p
p kakL∞(Ω)Rq
−1
p +Cp1
−p p kbk
L
q
q−1(Ω). (4.5) The assumptions in (A4) imply that (4.5) cannot be satisfied.
Proposition 4.2. Suppose that R satisfies one of the three conditions mentioned in (A4). Then F satisfies the following Palais–Smale type compactness condition: if(un)n is a sequence from XR such that one of the following statements hold
(a) F0(un)→0as n→∞;
(b) for each n∈Nwe have H(un) =R,hH0(un), ¯JF0(un)i ≤0and F0(un)− hF0(un),uni
hH0(un),uniH
0(un)→0 as n→∞,
then(un)nadmits a convergent subsequence.
Proof. Since the sequence(un)nis bounded inW01,p(Ω)(it belongs toXR) and since the embed- dingW01,p(Ω),→ Lq(Ω)is compact forq∈(1,p∗), there existu∈W01,p(Ω)and a subsequence of (un)n, which we denote again by (un)n, which converges weakly in W01,p(Ω) to u and strongly inLq(Ω)tou. Then by Hölder’s inequality we have
hNf(un),un−ui ≤ kf(·,un)k
L
q
q−1(Ω)kun−ukLq(Ω)→0. (4.6) Case (a): F0(un)→0 asn→∞. Then,
hH0(un),un−ui=hF0(un),un−ui+hNf(un),un−ui →0.
The(S+)property ofH0 = Jϕ (see [6, Theorem 10]) implies(un)nconverges strongly tou.
Case (b): For eachn∈Nwe haveH(un) =R,hH0(un), ¯JF0(un)i ≤0 and F0(un)− hF0(un),uni
hH0(un),uniH
0(un)→0 asn→∞.
We denote
µ= lim
n→∞
hF0(un),uni hH0(un),uni ∈R.
Therefore,
F0(un)−µH0(un)→0 asn→∞. (4.7) Ifµ=0, the above convergence implies
F0(un)→0 asn→∞.
As in Case (a) it follows that there existu∈W01,p(Ω) and a subsequence of(un)n, which we denote again by(un)n, which converges strongly inW01,p(Ω)tou. Since F0 is continuous, we haveF0(u) =0, which implieshH0(u),ui= hNf(u),ui. Butu∈ NR(since(un)nbelongs to the closed setNR), which yields
pR=hNf(u),ui ≤CqkakL∞(Ω)(pR)qp +Ckbk
L
q
q−1(Ω)(pR)1p.
This contradicts the assumption onRfrom (A4). Hence, the caseµ=0 is not possible.
Forµ6=0 we have by (4.7)
hF0(un)−µH0(un),un−ui →0 asn→∞.
(1−µ)hH0(un),un−ui+hNf(un),un−ui →0 asn→∞.
Ifµ6=1 we get by (4.6)
(1−µ)hH0(un),un−ui →0 asn→∞.
The (S+) property of H0 = Jϕ implies (un)n converges strongly to u. The convergence (4.7) and the strong convergenceun →uimplies F0(u)−µH0(u) =0.
hF0(u), ¯JF0(u)i=µhH0(u), ¯JF0(u)i.
But hH0(u), ¯JF0(u)i ≤ 0 and hF0(u), ¯JF0(u)i ≥ 0, therefore µ < 0 and the relation F0(u)− µH0(u) =0 contradicts the statement of Proposition4.1. Hence, the caseµ6=1 is not possible.
Forµ=1: by (4.7) it follows that
hF0(un), ¯JF0(un)i − hH0(un), ¯JF0(un)i →0 asn→∞.
Using (2.1) we have
hF0(un), ¯JF0(un)i= ϕ−1(kF0(un)k)kF0(un)k ≥0 and by the assumptions of Case (b) we have
−hH0(un), ¯JF0(un)i ≥0
Then F0(un) → 0 asn → ∞. As in Case (a) it follows that there exist u∈W01,p(Ω) and a subsequence of (un)n, which we denote again by(un)n, which converges strongly inW01,p(Ω) to u. This yields u ∈ NR, because (un)n belongs to the closed set NR. Since F0 and H0 are continuous, we have by the convergence (4.7) thatF0(u)−H0(u) =0,
hF0(u), ¯JF0(u)i=hH0(u), ¯JF0(u)i.
ButhH0(u), ¯JF0(u)i ≤0 (by the assumption of Case (b) and by the strong convergenceun→u) and hF0(u), ¯JF0(u)i ≥ 0, hence F0(u) = 0, which implies H0(u) = 0 and then u = 0. But 0 /∈ NR, contradictsu∈ NR. Hence the caseµ=1 is not possible.
We apply Theorem3.1 in order to localize the solution of (4.3).
Theorem 4.3. Suppose that R satisfies one of the three conditions mentioned in(A4). Then, equation (4.3)admits a weak solution u∈XR, which minimizes F on XR.
In what follows we discuss situations when the best Sobolev constantCqadmits an upper estimate which can be computed:
Denote the first eigenvalue of thep-Laplace operator by λp(Ω) = min
v∈W01,p(Ω)\{0}
R
Ω|∇v(x)|pdx R
Ω|v(x)|pdx . Then,
kukLpp(Ω)≤ 1
λp(Ω)kuk1,pp for allu∈W01,p(Ω). Hence the best embedding constant of W01,p(Ω),→ Lp(Ω) is Cp = 1
λp(Ω)
1/p
, while for q < p the best embedding constant of W01,p(Ω),→ Lq(Ω) verifies (via Hölder’s inequality)
Cq≤ |Ω|ppq−q 1
λp(Ω)
1/p
(here |Ω| denotes the Lebesgue measure, i.e. the k-dimensional vol- ume, of the setΩ). In order to obtain upper bounds forCq(q≤ p) we need lower bounds for λp(Ω).
By using the Faber–Krahn inequality [2, Theorem 1] it holds λp(Ω)≥ λp(Ω∗),
whereΩ∗ is thek-dimensional ball centered at the origin having the same volume asΩ. So it has the radiusr = √1
π |Ω|Γ(2k +1)1/k.
By [10] we have for the ballΩ∗ = Br⊂Rk of radiusrthe inequality λp(Br)≥
k rp
p
.
Then the best Sobolev constant has the following upper estimate, which can be computed:
Cp≤ p k√
π
|Ω|Γ k
2+1 1k
, and for 1<q< p
Cq≤ p k√ π
|Ω|k(ppq−q)+1Γ k
2+1 1k
.
Fork =1 andΩ= (0,T)⊂Rthe value of the first eigenvalue is known (see [7])
λp(Ω) = (p−1) 2π T psin(πp)
!p
, hence
Cp = T psin(πp) 2π(p−1)1p .
For the casek=1 andΩ= (0,T)the sharp Poincaré inequality is known (see [20], p. 357):
for each p>1,q>1 andu∈W01,p(0,T)it holds
kukLq(Ω) ≤Cqkuk1,p, where the embedding constant is given by
Cq= T
1 q+1
p0
2B(1q,p10) p
01q
qp10 p0+q1p−1q
,
p0 = p−p1 andBis the Beta function.
4.2 Example 2
For 1< p<∞we define the following subspace of radially symmetric functions ofW1,p(Rk) Wr1,p(Rk) =u∈W1,p(Rk):u(x) =u(x0)∀ x,x0 ∈ Rk,|x|=|x0| ,
endowed with the norm induced fromW1,p(Rk) kukp =
Z
Rk
|∇u(z)|p+|u(z)|pdz.
The spaceW1,p(Rk) is a separable, reflexive and uniformly convex Banach space [1, 3.6 The- orem, p. 61]; moreover, it is also uniformly smooth (which is proved by using Clarkson’s inequalities [1, 2.38 Theorem, p. 44] and [4, Definition 2.4., p. 13]).
In the context of Section 2 and Section 3 we consider X = Wr1,p(Rk) endowed with the above normk · k, Xis a closed subspace ofW1,p(Rk). Hence it is also uniformly smooth and by [4, Theorem 2.10] it follows that its dualX∗ is uniformly convex.
LetJϕ :X→ X∗ be the duality mapping corresponding to the weight functionϕ(t) =tp−1, t ∈ R+ where p ∈ (1,+∞) (see [3, Proposition 2.2.4]). It is well known that the duality mapping Jϕ satisfies the following conditions:
kJϕuk= ϕ(kuk) and hJϕu,ui=kJϕukkuk for all u∈X.
Moreover, the functional H : X →R defined by H(u) = 1pkukp is convex and Fréchet differ- entiable with H0 = Jϕ. We take ¯J = J−ϕ1.
It is known [11, Théorème II. 1] that the embedding Wr1,p(Rn) ,→ Lq(Rn)is compact for q∈ (p,p∗)(wherek ≥2, p∗ = kkp−p if p <kand p∗ =∞, if p ≥k) there existsCq>0 (the best embedding constant) such that
kukLq(Rk) ≤Cqkuk for eachu∈Wr1,p(Rk). (4.8) Let f :R×Rk →Rbe a Carathéodory function such that f(x, 0)6=0 for a.e.x∈Rwhich satisfies
|f(x,s)| ≤a(x)|s|q−1+b(x) for(x,s)∈Rk×R,
where a ∈ L∞(Rk),b ∈ Lq−q1(Rk)are positive functions and q ∈ (p,p∗)and f(x,·) = f(x0,·) for all x,x0 ∈ Rn,|x|= |x0|(f is radially symmetric in the first variable).
Consider the following problem involving the p-Laplacian:
−∆pu+|u|p−2u= f(x,u) a.e.x ∈Rk. (4.9) We call u∈W1,p(Rk)aweak solutionof (4.9) if for eachv ∈W1,p(Rk)it holds
Z
Ω|∇u(x)|p−2∇u(x)∇v(x) +|u(x)|p−2u(x)v(x)dx=
Z
Ω f(x,u(x))v(x)dx. (4.10) Define the Nemytskii operatorNf :X→ X∗ byNf (u) (x) = f(x,u(x))andF: X→Rby
F(u) = 1
pkuk1,pp −
Z
Rnh(x,u)dx, where h:Ω×R→Rish(x,t) =Rt
0 f(x,s)ds. We have F0(u) = H0(u)−Nf(u).
Let G= O(Rk)be the set of all rotations onRk. Observe that the elements of GleaveRk invariant, i.e.g(Rk) =Rk for allg∈G. Ginduces an isometric linear action overW1,p(Rk)by
(gu)(z) =u(g−1z), g ∈G, u∈W1,p(Rk), a.e. z∈Rk.
A functionφdefined onW1,p(Rk)is said to beG-invariant if
φ(gu) =φ(u) for allg ∈G, u∈W1,p(Rk). In factWr1,p(Rk)is the fixed point set ofW1,p(Rk)underGand the norm
kuk= Z
Rk(|∇u(z)|p+|u(z)|p)dz 1p
isG-invariant onW1,p(Rk).
Observe that by the assumption on f and the above remark, the functionalFisG-invariant and then by the principle of symmetric criticality [12] every critical point ofFis also a solution of (4.10).
Consider
XR=
v∈Wr1,p(Rk): 1
pkvkp ≤R
. Reasoning as in Section4.1 one has the following result.
Theorem 4.4. Suppose that R satisfies one of the three conditions mentioned in(A4). Then, F admits a critical point u∈XR, which minimizes F on XR. Moreover, this critical point is also a weak solution of (4.10).
We discuss situations when the Sobolev constant Cq admits an upper estimate which can be computed: by [20] we have that for 1< p<kand p∗= kkp−p it holds for allu ∈W1,p(Rk)
kukLp∗
(Rk) ≤CR Z
Rk|∇u(x)|pdx 1p
, where
CR= √1 πk1p
p−1 k−p
1−1p Γ(1+2k)Γ(k) Γ(kp)Γ(1+k− kp)
!1k . Obviously this implies
kukLp∗
(Rk)≤CRkukfor eachu∈W1,p(Rk).
For any q ∈ (p,p∗) there exists θ ∈ (0, 1) such that q = θp+ (1−θ)p∗, then by Hölder’s inequality
kukq
Lq(Rk)≤ kukθp
Lp(Rk)kuk(1−θ)p∗
Lp∗(Rk) ≤Ckq
1 p−1q R kukq,
for eachu ∈ W1,p(Rk). Then the Sobolev constant has the following upper estimate, which can be computed:
Cq≤Ck
1 p−1q
R forq∈
p, kp k−p
and 1< p<k.
Acknowledgement
The authors would like to express their special thanks to Prof. Cs. Varga from Babes,-Bolyai University in Cluj-Napoca for his constructive discussions.
References
[1] R. A. Adams, J. J. F. Fornier, Sobolev spaces, Elsevier/Academic Press, Second Edition, 2003.MR2424078
[2] T. Bhattacharya, A Proof of the Faber–Krahn inequality for the first eigenvalue of the p-Laplacian,Ann. Mat. Pura Appl. (4)177(1999), 225–240.MR1747632;url
[3] J. Chabrowski,Variational methods for potential operator equations, Walter de Gruyter, 1997.
MR1467724;url
[4] Ch. Chidume,Geometric properties of Banach spaces and nonlinear iterations, Springer Verlag London, 2009.MR2504478
[5] J. Diestel, Geometry of Banach spaces – Selected topics, Springer Verlag, Berlin, 1975.
MR0461094
[6] G. Dinca, P. Jebelean, J. Mawhin, Variational and topological methods for Dirichlet problems with p-Laplacian,Port. Math. (N.S.)58(2001), 339–378.MR1856715
[7] P. Drábek, R. Manásevich, On the closed solution to some nonhomogeneous eigenvalue problems with p-Laplacian,Differential Integral Equations12(1999), 773–788.MR1728030 [8] I. Ekeland, Nonconvex minimization problems, Bull. Amer. Math. Soc. (N.S.) 1(1979),
No. 3, 443–474.MR526967;url
[9] H. Goldberger, W. Kampowskoy, F. Troltzsch, On Nemytskij operators inLp-spaces of abstract functions,Math. Nachr.155(1992) 127–140.MR1231260;url
[10] B. Kawohl, V. Fridman, Isoperimetric estimates for the first eigenvalue of the p-Laplace operator and the Cheeger constant, Comment. Math. Univ. Carolin. 44(2003), No. 4, 659–
667.MR2062882
[11] P.-L. Lions, Symétrie et compacité dans les espaces Sobolev (in French) [Symmetry and compactness in Sobolev spaces],J. Funct. Anal.49(1982), 315–334.MR683027;url
[12] R. S. Palais, The principle of symmetric criticality, Comm. Math. Phys. 69(1979), 19–30.
MR547524
[13] R. Precup, A compression type mountain pass theorem in conical shells, J. Math. Anal.
Appl.338(2008), 1116–1130.MR2386486;url
[14] R. Precup, Critical point localization theorems via Ekeland’s variational principle, Dy- nam. Systems Appl.22(2013), 355–370.MR3100209
[15] R. Precup, On a bounded critical point theorem of Schechter, Stud. Univ. Babes,-Bolyai Math.58(2013), 87–95. MR3068817
[16] P. H. Rabinowitz, Minimax methods in critical point theory with applications to partial dif- ferential equations, CBMS Regional Conference Series in Mathematics, Vol. 65, American Mathematical Society, Providence Rhode Island, 1986.MR845785;url
[17] M. Schechter, The mountain pass alternative, Adv. in Appl. Math. 12(1991), 91–105.
MR1091826;url
[18] M. Schechter, A bounded mountain pass lemma without the (PS) condition and appli- cations,Trans. Amer. Math. Soc.331(1992), 681–703.MR1064270;url
[19] M. Schechter, Linking methods in critical point theory, Birkhäuser Boston, 1999.
MR1729208;url
[20] G. Talenti, Best constant in Sobolev inequality,Ann. Mat. Pura Appl.110(1976), 353–372.
MR0463908
[21] M. M. Vainberg,Variational methods for the study of nonlinear operators, Holden-Day, Inc., San Francisco, Calif.–London–Amsterdam, 1964.MR0176364