6 Compactness results and applications to 0 0 2 some “zero mass” elliptic problems n a J 7 A. Azzollini & A. Pomponio ∗ † 1 ] P A . In ricordodiGiulioMinervini h t a m [ 1 1 Introduction and statement of the main results v 0 1 Inthis paperwestudy the elliptic problem, 4 1 0 ∆v = f′(v) in Ω, (1) 6 − 0 / inthesocalled“zeromasscase”thatis,roughlyspeaking,whenf′′(0) = 0. h A particular exampleis t a m : ∆v = vNN−+22 in RN, v − i X with N > 3. This problem has been studied very intensely (see [4,16,24]) r and weknow the explicit expression ofthe positive solutions a [N(N 2)λ2](N−2)/4 v(x) = − , with λ > 0, x RN. [λ2 + x x 2](N−2)/2 0 ∈ 0 | − | If f is not the critical power, we are led to require particular growth conditions onthenonlinearity f. Infact,whileinthe “positive masscase” (namely when f′′(0) < 0) the natural functional setting is H1(Ω) and we havesuitablecompactembeddingsjustassumingasubcriticalbehaviorof ∗DipartimentodiMatematica,Universita`degliStudidiBari,ViaE.Orabona4,I-70125 Bari,Italy,e-mail: [email protected] †DipartimentodiMatematica,PolitecnicodiBari,ViaAmendola126/B,I-70126Bari, Italy,e-mail: [email protected] 1 2 A. Azzollini and A. Pomponio f,inthe“zeromasscase”theproblemisstudiedin 1,2(Ω)thatisdefined D asthe completion ofC∞(Ω) with respectto the norm 0 1 2 u = u 2dx . k k |∇ | (cid:18)ZΩ (cid:19) Inordertorecoveranalogouscompactnessresults,weneedtoassumethat f issupercritical nearthe origin andsubcritical atinfinity. With these assumptions on f, the problem (1) has been dealt with by Berestycki & Lions [13–15], when Ω = RN, N > 3, and existence and multiplicity results have beenproved. Recently, Benci & Fortunato [8] have introduced a new functional set- ting, namely the Orlicz space Lp + Lq, which arises very simply from the growth conditions on f and seemsto be the natural framework forstudy- ing“zeromass” problems asshown also byPisani in [23]. Usingthisnewfunctional setting, Benci &Micheletti in[9]studied the problem (1), with Dirichlet boundary conditions, in the case of exterior domain, namely when RN Ω is contained into a ball B . Under suitable ε \ assumptions,iftheballradiusεissufficientlysmall,theyareabletoprove the existence of apositive solution. The functional setting introduced in [8] seems to be the natural one alsofor studying the nonlinearSchro¨dinger equationswith vanishing po- tentials, namely ∆v +V(x)v = f′(v), in RN, (2) − with lim V(x) = 0. x→∞ SomeexistenceresultsforsuchaproblemhavebeenfoundbyBenci,Grisanti &Micheletti [10,11]and byGhimenti & Micheletti [18]. Even if in a different context, we need also to mention the paper of Ambrosetti, Felli & Malchiodi [2], where problem (2) is studied when the nonlinearityf(v)isreplacedbyafunction f(x,v)ofthetype K(x)vp,with K vanishingat infinity. In this paper, we study problem (1) in two different situations. In Sec- tion 4, we look for complexvalued solutions ofthe following problem ∆v = f′(v) in R3, (3) − assumingthat f C1(C,R)satisfies the following assumptions: ∈ (f1) f(0) = 0; Compactness resultsand “zeromass”ellipticproblems 3 (f2) M > 0 such that f(M) > 0; ∃ (f3) ξ C : f′(ξ) 6 cmin( ξ p−1, ξ q−1); ∀ ∈ | | | | | | (f4) f(eiαρ) = f(ρ),for all ξ = eiαρ C; ∈ where 1 < p < 6 < q andc > 0. Observe that an example of function satisfying the previous hypothe- ses can be obtained as follows. Let us consider the function f˜ : R+ R → definedas atp +b if t > 1 ˜ f(t) := tq if t 6 1, (cid:26) with a,b R chosen in order to have f˜ C1 and let us define f : C R ∈˜ ∈ → asf(ξ) = f( ξ ). | | Introducing the cylindrical coordinates (r,z,θ), for all n Z, we look ∈ for solutions ofthe type vn(x,y,z) = un(r,z)einθ with un R. (4) ∈ Weobtain the following existence result for problem (3): Theorem 1.1. Let f satisfy the hypotheses (f1-f4). Then there exists a sequence (vn) of complex-valued solutions of problem (3), such that, for every n Z, n ∈ vn(x,y,z) = un(r,z)einθ, with un R. ∈ Actually,anexistenceresultinthesamespiritofoursispresentin[22]. However, in [22] the problem is studied using different tools and the de- tails are omitted. Moreover in [12] an interesting physical interpretation has been given to the complex valued solutions of the equation (3) in the positivemasscase. Infacttherehasbeenshownthestrictrelationbetween such solutions and the standing waves of the Schro¨dinger equation with nonvanishing angularmomentum. In Section 5,we study ∆v = f′(v) in R2 I, − × (5) v = 0 in R2 ∂I, (cid:26) × whereI isaboundedintervalofRandf C1(R,R)satisfiesthefollowing ∈ assumptions: (f1’) f(0) = 0; (f2’) ξ R :f(ξ) > c min( ξ p, ξ q); 1 ∀ ∈ | | | | 4 A. Azzollini and A. Pomponio (f3’) ξ R : f′(ξ) 6 c min( ξ p−1, ξ q−1); 2 ∀ ∈ | | | | | | (f4’) there exists α > 2 such that ξ R : αf(ξ) 6 f′(ξ)ξ; ∀ ∈ with 2 < p < 6 < q and c ,c > 0. 1 2 Wewill prove the following multiplicity result: Theorem 1.2. Let f satisfy the hypotheses (f1’-f4’). Then there exist infinitely many solutions withcylindricalsymmetryof problem(5). In order to approach to our problems, we use a functional framework related to the Orlicz space Lp + Lq. The main difficulty in dealing with suchspacesconsistsinthelackofsuitablecompactnessresults. Inviewof this,thekeypointsofthispaperaretwocompactnesstheoremspresented in Section 3. They are obtained adapting a well known lemma of Esteban &Lions [17]to our situation. The paperisorganized asfollows: Section 2 isdevoted to a briefrecall on the space Lp +Lq; in Section 3, we present our compactness results; in Sections 4 and 5 we solve problems (3) and (5); finally, in the Appendix we prove a compact embedding theorem using similar arguments as in Section 3. 2 Some properties of the Lp + Lq spaces Inthissection,wepresentsomebasicfactsontheOrliczspaceLp+Lq. For more details, see [8,19,23]. Let Ω R3. For 1 < p < 6 < q, denote by (Lp(Ω), ) and by Lp ⊂ k · k (Lq(Ω), ) the usual Lebesgue spaceswith their norms, andset Lq k·k Lp +Lq(Ω) := v : Ω R (v ,v ) Lp(Ω) Lq(Ω) s.t. v = v +v . 1 2 1 2 { → |∃ ∈ × } Thespace Lp +Lq(Ω) isa Banachspace with the norm v (Ω) : inf v + v (v ,v ) Lp(Ω) Lq(Ω),v +v = v Lp+Lq 1 Lp 2 Lq 1 2 1 2 k k {k k k k | ∈ × } anditsdualistheBanachspace Lp′(Ω)∩Lq′(Ω),k·kLp′∩Lq′ ,wherep′ = p−p1, q′ = q and q−1 (cid:0) (cid:1) kϕkLp′∩Lq′ : kϕkLp′ +kϕkLq′. In the sequel,for allv Lp +Lq(Ω), weset ∈ Ω> := x Ω v(x) > 1 , ∈ | | | Ω6 := x Ω v(x) 6 1 . (cid:8) (cid:9) ∈ | | | The following theorem su(cid:8)mmarizes some p(cid:9)roperties about Lp + Lq spaces Compactness resultsand “zeromass”ellipticproblems 5 Theorem 2.1. 1. Let v Lp +Lq(Ω). Then ∈ 1 max v 1, v k kLq(Ω6) − 1+meas(Ω>)1/rk kLp(Ω>) (cid:18) (cid:19) 6 v 6 max v , v (6) Lp+Lq Lq(Ω6) Lp(Ω>) k k k k k k wherer = pq/(q p). (cid:0) (cid:1) − 2. ThespaceLp +Lq iscontinuously embeddedinLp . loc 3. For everyr [p,q] :Lr(Ω) ֒ Lp +Lq(Ω) continuously. ∈ → 4. Theembedding 1,2(Ω) ֒ Lp +Lq(Ω) (7) D → iscontinuous. Proof 1. See Lemma1 in[8]. 2. See Proposition 6 of[23]. 3. See Corollary 9 in [23]. 4. Itfollows from the point 3and the Sobolev continuous embedding 1,2(Ω) ֒ L6(Ω). D → (cid:3) The following theorem hasbeenproved in [23]: Theorem2.2. Letf beaC1(C,R)function(resp. C1(R,R))satisfyingassump- tion(f3)(resp. (f3’)). Thenthefunctional v Lp +Lq(Ω) f(v)dx ∈ 7−→ ZΩ isof classC1. Moreover theNemytskioperator f′ : v Lp +Lq(Ω) f′(v) (Lp +Lq(Ω))′ ∈ 7−→ ∈ isbounded. UsingTheorem2.2wegetaveryusefulinequalityfortheLp+Lq-norm. 6 A. Azzollini and A. Pomponio Theorem2.3. For allR > 0,thereexistsapositiveconstantc = c(R)suchthat, forall v Lp +Lq(Ω) with v 6 R, Lp+Lq ∈ k k max v pdx, v qdx 6 c(R) v . (8) Lp+Lq | | | | k k (cid:18)ZΩ> ZΩ6 (cid:19) Proof Let us introduce g C1(R,R) such that g(0) = 0 and with the ∈ following growth conditions: (g1) ξ R :g(ξ) > c min( ξ p, ξ q); 1 ∀ ∈ | | | | (g2) ξ R : g′(ξ) 6 c min( ξ p−1, ξ q−1). 2 ∀ ∈ | | | | | | Integrating in (g1)we get g(v)dx > c v pdx+ v qdx . 1 | | | | ZΩ (cid:18)ZΩ> ZΩ6 (cid:19) ByLagrange theorem, there exists t [0,1]such that ∈ g′(tv)vdx = g(v)dx > c v pdx+ v qdx . 1 | | | | ZΩ ZΩ (cid:18)ZΩ> ZΩ6 (cid:19) Then, by the boundness of g′ (see Theorem 2.2), there exists M > 0 such that M v > g′(tv)v dx > c v pdx+ v qdx Lp+Lq 1 k k | | | | | | ZΩ (cid:18)ZΩ> ZΩ6 (cid:19) andhence the conclusion. (cid:3) Remark 2.4. Combining the inequality (6) with the estimate (8) we deduce that thefollowing statementsare equivalent: a) v v inLp +Lq(Ω), n → b) v v 0 and v v 0, k n − kLp(Ω>n) → k n − kLq(Ω6n) → whereΩ> = x Ω v (x) v(x) > 1 and Ω6 isanalogously defined. n { ∈ | | n − | } n Compactness resultsand “zeromass”ellipticproblems 7 3 Compactness results In this section we present the main tools of this paper, namely a compact- ness theorem for sequences with “a particular symmetry” and a compact embeddingofasuitable subspace of 1,2 into Lp+Lq. The proofs ofthese D results are both modelled on that of Theorem 1 of [17], which states that a suitable subspace ofH1 iscompactly embeddedinto Lp, forp subcritical. Firstofall,foreveryintervalI ofR,possiblyunbounded,weintroduce the following subspace of 1,2(R2 I): D × 1,2(R2 I) = u 1,2(R2 I) u( , ,z)isradial, for a.e. z I . Dcyl × ∈ D × | · · ∈ (cid:8) (cid:9) Moreover weassume the following Definition3.1. Ifu : R2 I Risameasurablefunction,wecallz-symmetrical × → rearrangementofuin(x,y)theSchwarzsymmetricalrearrangementofthefunc- tion u(x,y, ) : z I u(x,y,z) R. · ∈ 7→ ∈ Moreover we call z-symmetrical rearrangement of u the function v defined as follows u˜ : (x,y,z) R2 I u˜ (z) x,y ∈ × 7→ whereu˜ is thez-symmetricalrearrangementof uin(x,y). x,y In our first compactness result, weconsider I = R. Theorem 3.2. Let (u ) be a bounded sequence in 1,2(R3) such that u is the j j Dcyl j z-symmetrical rearrangement of itself. Then (u ) possesses a converging subse- j j quence inLp +Lq(R3), for all1 < p < 6 < q. Proof With an abuse of notations, in the sequel for every v 1,2(R3), ∈ Dcyl we denote byv alsothe function defined in R+ R as × v( x2 +y2,z) = v(x,y,z). p Beingtheproofquitelongandinvolved,wedivideitintoseveralsteps,for reader’s convenience. Since (u ) is bounded in the 1,2(R3) norm, there j j D exists u 1,2(R3) such that ∈ Dcyl u ⇀ u weaklyin 1,2 R3 andin Lp +Lq(R3), 1 < p 6 6 6 q, (9) j Dcyl u u a.e. in R3, (10) j (cid:0) (cid:1) → u u inLp(K), for all K R3, 1 6 p < 6. (11) j → ⊂⊂ 8 A. Azzollini and A. Pomponio ByLions[20], C j > 1, r > 0,z = 0 : u (x,y,z) 6 , (12) ∀ ∀ 6 | j | r41 z 41 | | where r = x2 +y2. By (12), for R > 0 large enough, j > 1 and for all (r, z ) (R,+ ) (R,+ ),we have | | ∈ p∞ × ∞ u (x,y,z) < 1, j | | u(x,y,z) < 1, (13) | | (u u)(x,y,z) < 1. j | − | Let D := (r,z) R+ R r > R, z > R , 1 { ∈ × | | | } D := (r,z) R+ R 0 6 r 6 R, z 6 R , 2 { ∈ × | | | } D := (r,z) R+ R 0 6 r 6 R, z > R , 3 { ∈ × | | | } D := (r,z) R+ R r > R, z 6 R . 4 { ∈ × | | | } 4 Obviously D = R+ R. Moreover denote by χ the characteristic i × Di i=1 [ function ofD and observe that, since i 4 u u = (u u)χ k j − kLp+Lq j − Di Lp+Lq (cid:13)Xi=1 (cid:13) (cid:13)4 (cid:13) 4 (cid:13) (cid:13) 6 (u u)χ = u u , k j − DikLp+Lq k j − kLp+Lq(Di) i=1 i=1 X X then weget the conclusion ifweprove that, forall i = 1,...,4, u u in Lp +Lq(D ). j i → CLAIM 1: uj u inLp +Lq(D1). → Suppose for a moment that q > 8. By (13), for every (x,y,z) D , we 1 ∈ have (u u)(x,y,z) < 1,then the inequality(6)implies j | − | u u 6 u u . (14) k j − kLp+Lq(D1) k j − kLq(D1) On the other hand,since C u u a.e. and (u u)(r,z) q 6 L1(D ), j j q q 1 → | − | r 4 z 4 ∈ | | | | Compactness resultsand “zeromass”ellipticproblems 9 byLebesgue theorem u uin Lq(D ). j 1 → If6 < q 6 8,thentake r q 6,4(q 6) andsetα = 6 and β = 6 . ∈ − − 6−q+r q−r Observe that 1 + 1 = 1 so,(cid:16)byHolder, (cid:17) α β u u qdxdydz u u r u u q−rdxdydz j j j | − | | − | | − | ZD1 ZD1 1 1 6 u u αr α u u (βq−r) β j j | − | | − | (cid:16)ZD1 (cid:17) (cid:16)ZD1 (cid:17) 6−q+r 6 u u 6−6qr+r 6 u u q−r. (15) | j − | k j − kL6 (cid:16)ZD1 (cid:17) Since (u ) is bounded in 1,2(R3),it isbounded in L6(R3). j j D Moreover, since q 6 < r < 4(q 6),certainly 6r > 8,and thenthe last − − 6−q+r integral in inequality(15)goesto zero. Hencethe Claim1 isproved. CLAIM 2: uj u in Lp +Lq(D2). → Itisenoughtoobserve that,sinceD hasfinitemeasure,Lp+Lq(D ) = 2 2 Lp(D ) (see [23, Remark5])and then we getthe conclusion by(11). 2 CLAIM 3: uj u in Lp +Lq(D3). → First suppose p < 4 and consider g C1(R,R), g(0) = 0, such that the ∈ following growth and strong convexity conditions hold (G) k > 0s.t. t R : g′(t) 6 k min( t p−1, t q−1), 1 1 ∃ ∀ ∈ | | | | | | (SC) k > 0s.t. s,t R :g(s) g(t) g′(t)(s t) 2 ∃ ∀ ∈ − − − > k min( s t p, s t q). 2 | − | | − | Since g(0) = 0, from (G)and (SC)wededuce that k ,k > 0s.t. s R : k min( s p, s q) 6 g(s) 6 k min( s p, s q). (16) 3 4 3 4 ∃ ∀ ∈ | | | | | | | | The condition (SC) has been introduced in [5], where an explicit example offunction satisfying (SC)is alsogiven. For almost every (x,y) R2, we set ux,y : R R defined as ux,y(z) := ∈ → u(x,y,z). Wegive an analogous definition forux,y, forall j > 1. j For almost every(x,y) R2 with (x2 +y2)1/2 6 R, we set ∈ w (x,y) := g(ux,y(z))dz. j j Z(−R,R)c 10 A. Azzollini and A. Pomponio Weshowthat w (x,y) g(ux,y(z))dz for a.e. (x,y) B . (17) j R → ∈ Z(−R,R)c Consider w (x,y) g(ux,y(z))dz 6 g(ux,y) g(ux,y) dz j − | j − | (cid:12) Z(−R,R)c (cid:12) Z(−R,R)c (cid:12) (cid:12) (cid:12) (cid:12) = g′(θx,y) ux,y ux,y dz (18) (cid:12) (cid:12) | j || j − | Z(−R,R)c where, for almost every (x,y) B , θx,y is a suitable convex combination ∈ R j of ux,y and ux,y. Since (u ) is bounded in Lp + Lq, g′(θx,y) is bounded in j j j j ′ Lp + Lq (see Theorem 2.2) so, by (18), to prove (17) we are reduced to s(cid:16)howtha(cid:17)t ux,y ux,y in Lp +Lq ( R,R)c for a.e. (x,y) B . j → − ∈ R For, define (cid:0) (cid:1) Ωx,y = z R z > R, ux,y(z) ux,y(z) > 1 j ∈ | | | | j − | sothat, by(6), (cid:8) (cid:9) ux,y ux,y 6 k j − kLp+Lq((−R,R)c) max kuxj,y −ux,ykLp(Ωxj,y),kuxj,y −ux,ykLq((−R,R)c\Ωxj,y) . (19) (cid:16) (cid:17) ByLebesgue theorem and by(12), kuxj,y −ux,ykLq((−R,R)c\Ωxj,y) → 0 for a.e. (x,y) ∈ BR. (20) Moreover there exists R′ = R′(x,y) Rsuch thatfor all z > R′ ∈ | | 2C ux,y(z) ux,y(z) 6 6 1. | j − | r1/4 z 1/4 | | LetR˜ = max(R,R′). Wehave ux,y ux,y p 6 ux,y ux,y pdz k j − kLp(Ωxj,y) Z(−R˜,−R)∪(R,R˜)| j − | andthen kuxj,y −ux,ykLp(Ωxj,y) → 0 for a.e. (x,y) ∈ BR, (21)