A NONLINEAR LIOUVILLE THEOREM FOR FRACTIONAL EQUATIONS IN THE HEISENBERG GROUP ELEONORA CINTI AND JINGGANG TAN Abstract. We establish a Liouville-type theorem for a subcritical nonlinear problem, involving a fractional power of the subLaplacian in the Heisenberg group. To prove our result we will use the local realization of fractional CR covariant operators, which can be constructed as the Dirichlet-to-Neumann operator of a degenerate elliptic equation in the spirit of Caffarelli and Silvestre [4], as established in [9]. The main tools in our proof are the CR inversion and the moving plane method, applied to the solution of the lifted problem in the half-space Hn × R+ . 1. Introduction and Main Results In this paper we establish a Liouville-type result for the following fractional nonlinear problem in the Heisenberg group: P 1 u = up 2 in Hn . (1.1) Here P 1 denotes the CR square root of the Heisenberg Laplacian −∆H , 2 which is a CR covariant operator of order 1/2 in Hn . In [9] Frank, Gonzalez, Monticelli and one of the author, study CR covariant operators of fractional orders on orientable and strictly pseudoconvex CR manifolds. In this context, the Heisenberg group Hn plays the same role as Rn in conformal geometry. Given a Kähler-Einstein manifold X , CR covariant operators of fractional order γ are pseudodifferential operators whose principal symbol agrees with the pure fractional powers of the CR sub-Laplacian on the boundary M = ∂X . They can be defined using scattering theory, as done in [7, 15, 14, 13]. One of the main results in [9] establishes that it is possible to characterize fractional CR covariant operators on some CR manifold M = 2010 Mathematics Subject Classification. Primary: 35A01,35B50,35J70. Secondary: 35B53,35J50. Key words and phrases. Fractional sublaplacian, Heisenberg group, Louville theorem, moving plane method . 1 2 ELEONORA CINTI AND JINGGANG TAN ∂X , as the Dirichlet-to-Neumann map of a degenerate elliptic equation in the interior of X , in the spirit of Caffarelli-Silvestre result [4]. In [9], in order to construct fractional CR covariants operators in the specific case of the Heisenberg group, Hn is identified with the boundary of the Siegel domain in R2n+2 (see section 2 for the precise definition) and it is crucial to use its underlying complex hyperbolic structure. Another possible approach in the construction of fractional powers of the sub-Laplacian consists in using purely functional analytic tools as done in [8]. In [8] Ferrari and Franchi proved an extension result for fractional operators defined by using the spectral resolution of the sub-Laplacian in general Carnot groups. The operators considered in [8] are different in nature from the ones in [9], since they are not CR covariant. In this paper, the CR covariance of P 1 will be crucial, this is the 2 reason why the operator that we consider is the one constructed in [9]. Since it will be of utmost importance in the sequel, we recall here the extension result proven in [9]. Theorem 1.1 (see Theorem 1.1 in [9]). Let γ ∈ (0, 1), a = 1 − 2γ. For each u ∈ C ∞ (Hn ), there exists a unique solution U for the extension problem 2 2 a ∂U ∂ U 2∂ U + + λ + ∆H U = 0 in Hn × R+ , (1.2) ∂λ2 λ ∂λ ∂t2 U = u n on H × {λ = 0}. Moreover, ∂U , λ→0 ∂λ where cγ is a constant depending only on γ. Pγ u = −cγ lim λa Observe that, differently from the extension result established in 2 [8], here we have the additional term λ2 ∂∂tU2 which appears when one considers CR sub-Laplacians. Our Liouville-type theorem is the analogue, for the fractional operator P 1 , of a result by Birindelli and Prajapat [2], for the sublaplacian 2 ∆H . In [2], the authors establish a nonexistence result for a class of positive solution of the equation − ∆H u = up , Q+2 , Q−2 (1.3) for p subcritical (i.e. 0 < p < where Q = 2n + 2 denotes the homogeneous dimension of Hn ). The technique they used is based on the moving plane method (which goes back to Alexandrov and Serrin NONLOCAL LIOUVILLE THEOREM 3 [20]), adapted to the Heisenberg group setting. This method requires two basic tools: the maximum principle and invariance under reflection with respect to a hyperplane. Since the operator −∆H is not invariant under the usual reflection with respect to hyperplanes, Birindelli and Prajapat needed to introduce a new reflection, called H-reflection, under which −∆H is invariant. Since it will be important in the sequel, we recall here the definition of H-reflection. Definition 1.2. For any ξ = (x, y, t) ∈ Hn , we consider the plane Tµ := {ξ ∈ Hn : t = µ}. We define ξµ := (y, x, 2µ − t), to be the H-reflection of ξ with respect to the plane Tµ . Due to the use of this reflection, the proof of the non existence result in [2] requires the solution u of (1.3) to be cylindrical, that is u(x, y, t) = 1 u(r, t) must depend only on r and t where r = (x2 + y 2 ) 2 . We can now state our main result, which is the analogue for the operator P 1 of the Liouville result contained in [2]. 2 Theorem 1.3. Let 0 < p < Q+1 , where Q = 2n+2 is the homogeneous Q−1 n dimension of H . Then there exists no bounded cylindrical solution u ∈ C 2 (Hn ) of ( P 1 u = up in Hn 2 . (1.4) u>0 in Hn Using the local formulation (1.2) established in [9], the above theorem will follow as a corollary of the following Liouville-type result for a nonlinear Neumann problem in the half-space Hn × R+ . Q+1 Theorem 1.4. Let 0 < p < Q−1 and let U ∈ C 2 (Hn ×R+ )∩C(Hn × R+ ) be a nonnegative solution of 2 2 ∂ U + λ2 ∂ U + ∆H U = 0 in Hn × R+ ∂λ2 ∂t2 (1.5) ∂U p n − =U on H × {λ = 0}. ∂λ Suppose that U (x, y, t, λ) = U (r, t, λ) depends only on r, t, λ, where 1 r = (|x|2 + |y|2 ) 2 . Then U ≡ 0. In the Euclidean case, classical nonexistence results for subcritical nonlinear problems in the all space Rn are contained in two works by Gidas and Spruck [12] and by Chen and Li [5]. Analogue results for nonlinear Neumann problems in the half-space Rn+ where established in [17, 18], using the methods of moving planes and moving spheres. 4 ELEONORA CINTI AND JINGGANG TAN In the Heisenberg group setting there are several papers concerning nonexistence results for problem (1.3). In [10] Garofalo and Lanconelli proved some nonexistence results for positive solutions of (1.3) when p is subcritical, but their results require some integrability conditions on u and ∇u. In [16, 21] similar nonexistence results for positive solutions of (1.3) in the half-space are established for the critical exponent p = Q+2 . In [3], a Liouville-type result for solution of (1.3) is proved without Q−2 Q requiring any decay condition on u, but only for 0 < p < Q−2 . As explained before, in [2] Birindelli and Prajapat extends this last result but only in the class of cylindrical solution. A to any 0 < p < Q+2 Q−2 last more recent result in this context was proven by Xu in [22], who established that there are no positive solution of (1.3) for 0 < p < Q(Q+2) . This result uses a different technique, based on the vector filed (Q−1)2 method, and improves the results contained in [10] and [3], since it does not require any decay on the solution u and it improves the exponent p. Nevertheless it does not let to reach the optimal exponent Q+2 (observe Q−2 Q < Q(Q+2) < Q+2 ). that Q−2 (Q−1)2 Q−2 In this paper we aim to establish a first Liouville-type result for a CR fractional power of −∆H ; this is, to our knowledge, the first nonexistence result in this fractional setting. Let us comment now on the basic tools in the proof of our main result. Following [2], in order to get a nonexistence result, we combine the method of moving planes with the CR inversion of the solution u. The CR inversion was introduced by Jerison and Lee in [19], and it is the analogue of the Kelvin transform, in the Heisenberg group context. In Section 3 we will give the precise definition of CR inversion and we will show which problem is satisfied by the CR inversion of a solution of (1.5). As said before, the moving plane method is based on several version of the maximum principles. More precisely we will need to prove that our operator satisfies a weak maximum principle (see Proposition 4.2) and two versions of the Hopf’s Lemma (see Propositions 4.4 and 4.6). The paper is organized as follows: • in Section 2 we recall some basic facts on the Heisenberg group and we will introduce the fractional CR operator P 1 ; 2 • in Section 3 we will introduce the CR inversion of a function u and prove a lemma concerning the CR inversion of a solution of our problem (1.5); NONLOCAL LIOUVILLE THEOREM 5 • in Section 4 we establish Maximum principles and Hopf’s Lemma for our operator, which will be basic tools in the method of moving plane; • in Section 5 we will prove our main result (Theorem 1.4). 2. Preliminary facts on the Heisenberg group In this section we recall some basic notions and properties concerning the Heisenberg group. We will denote the points in Hn using the notation ξ = (x, y, t) = (x1 , ..., xn , y1 , ..., yn , t) ∈ Rn × Rn × R. The Heisenberg group Hn is the space R2n+1 endowed with the group law ◦ defined in the following way: n X ˆ ξ ◦ ξ := (x̂ + x, ŷ + y, t̂ + t + 2 (xi ŷi − yi x̂i )). i=1 The natural dilation of the group is given by δ` (ξ) := (`x, `y, `2 t), and ˆ ◦ δ` (ξ). it satisfies δ` (ξˆ ◦ ξ) = δ` (ξ) n In H we will consider the gauge norm defined as 41 !2 n X |ξ|H := (x2i + yi2 ) + t2 , i=1 which is homogeneous of degree one with respect to δ` . Using this norm, one can define the distance between two points in the natural way: ˆ ξ) = |ξˆ−1 ◦ ξ|H , dH (ξ, where ξˆ−1 denotes the inverse of ξˆ with respect to the group action. We denote the ball associated to the gauge distance by BH (ξ0 , R) := {ξ ∈ Hn : dH (ξ, ξ0 ) < R}. Denoting by |A| the Lebesgue measure of the set A, we have that |BH (ξ0 , R)| = |BH (0, R)| = RQ |BH (0, 1)|. Here Q = 2n + 2 denotes the homogeneous dimension of Hn . For every i = 1, · · · , n, we denote by Xi , Yi , and T the following vector fields: ∂ ∂ ∂ ∂ ∂ Xi = + 2yi , Yi = − 2xi , T = . ∂xi ∂t ∂yi ∂t ∂t They form a basis of the Lie Algebra of left invariant vector fields. Moreover, an easy computions shows that [Xi , Yj ] = −4δij T . The 6 ELEONORA CINTI AND JINGGANG TAN Heisenberg gradient of a function f is given by ∇H f = (X1 f, ..., Xn f, Y1 f, ..., Yn f ). Finally, we define the sublaplacian as ∆H := n X (Xi2 + Yi2 ). i=1 It can be written also in the form ∆H = div(A∇T ), where A = aij is the (2n+1)×(2n+1) symmetric matrix given by aij = δij for i, j = 1, ..., 2n, aj(2n+1) = (2n + 1)j = 2yj for j = 1, ..., n, aj(2n+1) = (2n + 1)j = −2xj for j = n + 1, ..., 2n and a(2n+2)(2n+2) = 4(|x|2 + |y|2 ). It is easy to observe that A is positive semidefinite for any (x, y, t) ∈ Hn . This operator is degenerate elliptic, and it is hypoelliptic since it satisfies the Hormander condition. We pass now to describe CR covariant operators of fractional orders in Hn . For a rigourous geometric description of CR covariant operators on CR manifold, we remind to [9] and references therein. Here we just consider the case of the Heisenberg group, since it is the one of interest. Introducing complex coordinate ζ = x + iy ∈ Cn , we can identify the Heisenberg group Hn with the boundary of the Siegel domain Ωn+1 ⊂ Cn+1 , which is given by Ωn+1 := {(ζ1 , . . . , ζn+1 ) = (ζ, ζn+1 ) ∈ Cn × C | q(ζ, ζn+1 ) > 0} , with q(ζ, ζn+1 ) = Im ζn+1 − n X |ζj |2 , j=1 through the map (ζ, t) ∈ Hn → (ζ, t + i|ζ|2 ) ∈ ∂Ωn+1 . It is possible to see that X = Ωn+1 is a Kähler-Einstein manifold, endowed with a Kähler metric g+ , which can be identified with the complex hyperbolic space. The boundary manifold M = ∂Ωn+1 inherits a natural CR structure from the complex structure of the ambient manifold. Scattering theory tells us that for s ∈ C, Re(s) ≥ m2 , and except for a set of exceptional values, given f smooth on M, the eigenvalue equation −∆g+ u − s(m − s)u = 0, in X has a solution u with the expansion ( u = q (m−s) F + q s G for some F, G ∈ C ∞ (X ), F |M = f. NONLOCAL LIOUVILLE THEOREM 7 The scattering operator is defined as S(s) : C ∞ (M) → C ∞ (M) by S(s)f := G|M . For γ ∈ (0, m)\N, we set s = m+γ . The conformal fractional sub2 Laplacian on Hn is defined in the following way: Pγ f = cγ S(s)f, (2.1) for a constant cγ = 2γ Γ(γ) . Γ(−γ) For γ = 1 and γ = 2 we have: P1 = ∆H and P2 = ∆2H + T 2 . A crucial property of Pγ is its conformal covariance. As explained in the introduction, one of the main result in [9], is the characterization of these fractional operators via the extension problem (1.2). Since, throughout this paper, we will work on this lifted problem in the halfspace, let us introduce some notations in Hn × R+ . We will denote by dλ the distance between two points in Hn × R+ . More precisely, for any z = (x1 , ..., xn , y1 , ..., yn , t, λ) and any ẑ = (x̂1 , · · · , x̂n , ŷ1 , ..., ŷn , t̂, λ̂), let us define n h X i 41 2 dλ (z, ẑ) = (|xi − x̂i |2 + |yi − ŷi |2 ) + |λ − λ̂|2 + |t − t̂|2 . i=1 Observe that when λ = λ̂ = 0, that is z and ẑ lies on Hn , dλ (z, ẑ) = dH (z, ẑ). Moreover, given z0 ∈ Hn × {0}, we set B + (z0 , R) = {z ∈ Hn × R+ | dλ (z, z0 ) < R , λ > 0}. The operator of our interest will be 2 ∂2 2 ∂ + λ , ∂λ2 ∂t2 which, writing explicitly all the terms, becomes n 2 X ∂2 ∂ ∂2 ∂2 ∂2 L= + + 2 + 4yj − 4xj ∂λ2 j=1 ∂x2j ∂yj ∂xj ∂t ∂yj ∂t L = ∆H + n X ∂2 (x2j + yj2 )) 2 . + 4(λ + ∂t j=1 2 8 ELEONORA CINTI AND JINGGANG TAN Also in this case, we can write L = div(A∇T ), where now A is the (2n+2)×(2n+2) symmetric matrix given by aij = δij if i, j = 1, · · · , 2n, aj(2n+1) = a(2n+1)j = 2yj if j = 1, · · · , n, aj(2n+1) = a(2n+1)j = −2xj if j = n + 1, · · · , 2n, a(2n+1)(2n+1) = 4(|x|2 + |y|2 + λ2 ), a(2n+2)(2n+2) = 1, aj(2n+2) = a(2n+2)j = 0 if i = 1, · · · , n. In the sequel it will be useful to express L for cylindrical and radial functions. Let r = (|x|2 + |y|2 + λ2 )1/2 , ρ = (r4 + t2 )1/4 . Suppose that Ψ is a radial function, that is, Ψ depends only on ρ; then a direct computations gives: ∂Ψ ∂ρ ∂Ψ ∂Ψ = = ρ−3 r2 xj , ∂xj ∂ρ ∂xj ∂ρ ∂Ψ ∂Ψ ∂ρ ∂Ψ = = ρ−3 r2 λ . ∂λ ∂ρ ∂λ ∂ρ ∂Ψ ∂Ψ ∂ρ ∂Ψ = = ρ−3 r2 yj , ∂yj ∂ρ ∂yj ∂ρ Then we deduce that r4 x2j ∂ 2 Ψ ρ4 r2 + 2ρ2 x2j − 3r2 x2j ∂Ψ ∂ 2Ψ + , = 6 ∂x2j ρ ∂ρ2 ρ7 ∂ρ r4 yj2 ∂ 2 Ψ ρ4 r2 + 2ρ2 yj2 − 3r2 yj2 ∂Ψ ∂ 2Ψ + , = 6 ∂yj2 ρ ∂ρ2 ρ7 ∂ρ t2 ∂ 2 Ψ 2ρ4 − 3t2 ∂Ψ ∂ 2Ψ = + . ∂t2 4ρ6 ∂ρ2 4ρ7 ∂ρ Hence, by using that n 1 X 4 2 r2 4 2 4 2 2 2 ( (r x + r y ) + r λ + r t ) = j j ρ6 j=1 ρ2 and X 2 1 4 2 4 4 4 2 (2n + 1)ρ r + 2ρ − 3r + (2ρ − 3t ) ( (xj + yj2 ) + λ2 ) 7 ρ j=1 = Qr2 1 4 2 4 2 6 2 4 2 [(2n + 1)ρ r + 2ρ r − 3r + r (2ρ − 3t )] = , ρ7 ρ3 we conclude that r2 LΨ(ρ) = 2 ρ ∂ 2 Ψ(ρ) Q ∂Ψ(ρ) + ∂ρ2 ρ ∂ρ . (2.2) In a similar way, we deduce that for cylindrical symmetric functions φ = φ(r, t), we have Lφ = 2 ∂ 2 φ Q − 2 ∂φ 2∂ φ + + 4r . ∂r2 r ∂r ∂t2 NONLOCAL LIOUVILLE THEOREM 9 Using the radial form (2.2) for L, an easy computation yelds the following lemma. 1 1 Lemma 2.1. For ρ 6= 0, let ψ(ρ) = ρQ−1 = ρ2n+1 . Then we have that ( Lψ(ρ) = 0 ∈ Hn × R+ \ {0} (2.3) ∂ ψ(ρ) = 0 on Hn \ {0} × {λ = 0}. − ∂λ 3. CR inversion Following [2] and [19], we define the CR inversion in the half-space Hn × R+ . 1 For any (x, y, t, λ) ∈ Hn × R+ , let as before r = (|x|2 + |y|2 + λ2 ) 2 1 and ρ = (r4 + t2 ) 4 . We set x ei = x i t + yi r 2 y i t − xi r 2 e t e λ , y e = , t = − 4, λ = 2. i 4 4 ρ ρ ρ ρ The CR inversion of a function U defined on Hn × R+ , is given by 1 v(x, y, t, λ) = Q−1 U (x̃, ỹ, t̃, λ̃). ρ The following lemma shows which equation is satisfied by the CR inversion of a solution of problem (1.5). Lemma 3.1. Suppose that U ∈ C 2 (Hn × R+ \ {0}) ∩ C(Hn × R+ \ {0}) is a solution of (1.5). Then the CR inversion v of U satisfies Lv = 0 in Hn × R+ \ {0} (3.1) ∂v − = ρp(Q−1)−(Q+1) v on Hn \ {0} × {λ = 0}. ∂λ Proof. Since it is just a long computation, we will give the details of the proof for cylindrical solutions U (r, t), but the statement holds true for any solution. It is clear that r̃2 = |x̃|2 + |ỹ|2 + λ̃2 t2 (|x|2 + |y|2 + λ2 ) r4 (|x|2 + |y|2 + λ2 ) r2 = + = . ρ8 ρ8 ρ8 ρ4 The following relations are useful ∂r̃ t2 − r6 ∂r̃ −rt = , = 6 , 6 ∂r ρ ∂t ρ 3 2 ∂ t̃ 4tr ∂ t̃ 2t − t2 − r4 t2 − r 4 = 8 , = = . ∂r ρ ∂t ρ8 ρ8 10 ELEONORA CINTI AND JINGGANG TAN Now we have ∂v (1 − Q)r3 4tr3 ∂U 1 t2 − r4 ∂U = + ] U + [ ∂r ρQ+3 ρQ−1 ρ6 ∂r̃ ρ8 ∂ t̃ and ∂ 2v 2(1 − Q)r3 t2 − r4 ∂U 4r3 t ∂U ∂ (1 − Q)r3 U+ +( 8 ) ] = [( ) ∂r2 ∂r ρQ+3 ρQ+3 ρ6 ∂r̃ ρ ∂ t̃ 1 ∂ t2 − r4 ∂U ∂ 4r3 t ∂U ] + Q−1 [ ( + ( ) ) ρ ∂r ρ6 ∂r̃ ∂r ρ8 ∂ t̃ 1 h t2 − r4 2 ∂ 2 U 4r3 t 2 ∂U i t2 − r4 4r3 t ∂ 2 U + Q−1 ( + ( ) + ( )( ) ) ρ ρ6 ∂r̃2 ρ6 ρ8 ∂r̃∂ t̃ ρ8 ∂ t̃2 1 ∂ t2 − r4 ∂U ∂ 4r3 t ∂U + Q−1 [ ( ) + ( ) ]. ρ ∂r ρ6 ∂r̃ ∂r ρ6 ∂ t̃ The derivative with respect to t are given by (1 − Q)t 1 −rt ∂U t2 − r4 ∂U ∂v = U + [ + ] ∂t 2ρQ+3 ρQ−1 ρ6 ∂r̃ ρ8 ∂ t̃ and ∂ (1 − Q)t t2 − r4 ∂U 2(1 − Q)t −rt ∂U ∂ 2v = [( ) + ( ) ] U + + ∂t2 ∂r 2ρQ+3 2ρQ+3 ρ6 ∂r̃ ρ8 ∂ t̃ 1 ∂ −rt ∂U ∂ t2 − r4 ∂U + Q−1 [ ( 6 ) + ( ) ] ρ ∂r ρ ∂r̃ ∂r ρ8 ∂ t̃ 1 h −rt 2 ∂ 2 U −rt t2 − r4 ∂ 2 U t2 − r4 2 ∂U i + Q−1 ( 6 ) + ( 6 )( ) +( ) 2 ρ ρ ∂r̃2 ρ ρ8 ρ8 ∂r̃∂ t̃ ∂ t̃ 2 4 1 ∂ −rt ∂U ∂ t − r ∂U + Q−2 [ ( 6 ) + ( ) ]. ρ ∂t ρ ∂r̃ ∂t ρ6 ∂ t̃ Let us denote ∂ 2U ∂ 2U ∂ 2U ∂U ∂U + b2 LU = a1 2 + a2 + a3 2 + b 1 . ∂r̃ ∂r̃ ∂r̃∂ t̃ ∂ t̃ ∂ t̃ Then a1 = a2 = a3 = 1 1 ρ [(t2 − r4 )2 + 4r2 (rt)2 ] = Q+11 1 ρQ+13 1 ρ ρQ+3 , [2(t2 − r4 )(4r3 t) − 8r2 (rt)(t2 − r4 )2 ] = 0, [16r6 t3 + 4r2 (t2 − r4 )2 ] = Q+13 1 4r2 . ρQ+3 ρ4 NONLOCAL LIOUVILLE THEOREM 11 By using that ∂ t2 − r4 −4r3 (t2 + r4 ) −6(t2 − r4 )r3 ( ) = − , ∂r ρ6 ρ10 ρ10 ∂ −rt r(t2 + r4 ) 3rt2 ( )=− + 10 , ∂t ρ6 ρ10 ρ we can see that 2(1 − Q)r4 (t2 − r4 ) 1 ∂ t2 − r4 ( + ) rρQ+9 ρQ−1 ∂r ρ6 4r2 ∂ −rt Q − 2 (t2 − r4 )(t2 + r4 ) 4r2 (1 − Q)(−r2 t2 ) + + ( ) + r ρQ+9 rρQ+9 ρQ−1 ∂t ρ6 2(1 − Q)r4 (t2 − r4 ) −4r4 (t2 + r4 ) −6(t2 − r4 )r4 = + ( − ) rρQ+9 rρQ+9 rρQ+9 Q − 2 (t2 − r4 )(t2 + r4 ) 4(1 − Q)(−r4 t2 ) + + r ρQ+9 rρQ+9 4 2 4 4 2 4r (t + r ) 12r t − + Q+9 rρQ+9 rρ 2 1 (Q − 2) 1 (Q − 2)ρ = Q+3 . = Q+3 ρ r ρ r̃ b1 = We have that 12r2 t(t2 + r4 ) 32r6 t ∂ 4r3 t ( 8 )= − 12 , ∂r ρ ρ12 ρ ∂ t2 − r4 2t(t2 + r4 ) 4t(t2 − r4 ) ( ) = − . ∂t ρ8 ρ12 ρ12 Then we see that 8(1 − Q)r6 t 1 ∂ 4r3 t 4(Q − 2)r3 t(t2 + r4 ) + ( ) + ρQ+11 ρQ−1 ∂r ρ8 rρQ+11 4r2 ∂ t2 − r4 4(1 − Q)r2 t(t2 − r4 ) + + ( ) ρQ+11 ρQ−1 ∂t ρ8 8(1 − Q)r6 t 12r2 t(t2 + r4 ) 32r3 t 4(Q − 2)r2 t(t2 + r4 ) = + ( − ) + ρQ+11 ρQ+11 ρQ+11 ρQ+11 4(1 − Q)r2 t(t2 − r4 ) 8r2 t(t2 + r4 ) 16r2 t(t2 − r4 ) + + ( − ) = 0. ρQ+11 ρQ+11 ρQ+11 b2 = 12 ELEONORA CINTI AND JINGGANG TAN 1 e = 0. Finally, the This shows that Lv(x, y, t, λ) = ρ(Q+3) LU (e x, ye, e t, λ) Neumann data becomes: 1 ∂U ∂v e (x, y, t, λ) = − lim Q−1 ρ2 (e x, ye, e t, λ) − lim e λ→0 ∂λ e ρ λ→0 ∂λ 1 = (Q+1) U p ((e x, ye, e t, 0) = ρp(Q−1)−(Q+1) v(x, y, t, 0). ρ 4. Maximum principles and Hopf’s Lemma Basic tools in the method of moving planes are the maximum principle and Hopf’s Lemma. In this section we will prove that our operator L satisfies both of them. We start by stating the maximum principle for the classical solution of the problem related to L. Proposition 4.1. Let D be a bounded domain in Cn+1 , and let b ∈ C(D) be a nonnegative function. Assume that v ∈ C 2 (D) ∩ C 1 (D) is a solution of ( −Lv + bv ≥ 0 in D, (4.1) v≥0 on D. Then v ≥ 0 in D. Proof. Suppose that ξ0 ∈ D ⊂ Cn+1 is a local minimum point of v satisfying v(ξ0 ) < 0 then we have Dv(ξ0 ) = 0 and the Hessian matrix D2 v(ξ0 ) > 0. Writing L = div(A∇T ) and using the semipositivity of A, we reach a contradiction. We also give the maximum principle for weak solutions of the equations. For D ⊂ Cn+1 , we say that U ∈ H 1 (D) if it satisfies Z X n ∂U 2 ∂U 2 (|Xi U |2 + |Yi U |2 ) + + λ2 + |U |2 dz < ∞. (4.2) ∂λ ∂t D i=1 Proposition 4.2. Assume that D is a subset of Cn+1 with D0 := D ∩ {λ = 0} ⊂ ∂D. Let c ∈ L∞ (Ω) be a nonnegative function. Assume that v ∈ H 1 (D) ∩ C 2 (D) ∩ C 1 (D) is a solution of Lv = 0 in D, ∂v (4.3) − + c(ξ)v ≥ 0 on D0 = D ∩ {λ = 0}, ∂λ v≥0 on ∂D ∩ {λ 6= 0}. Then v ≥ 0 in D and v(ξ, 0) ≥ 0 in Ω. NONLOCAL LIOUVILLE THEOREM 13 Proof. Let v − denotes the negative part of the function v. We multiply the first equation in (4.3) by v − and integrate by parts to get − 2 − 2 ! Z Z n X ∂v − 2 − 2 2 ∂v − (|Xi v | + |Yi v | ) + c(ξ)|v − |2 dξ = 0, + λ ∂t dz− ∂λ D D0 i=1 which implies v − ≡ 0 in D0 = D ∩ {λ = 0}. We prove now two Hopf’s Lemmas. The first one is Hopf’s Lemma for the operator L in a subset D of Cn+1 . We first define the interior ball condition in this setting. Definition 4.3. Let D ⊂ Cn+1 . We say that D satisfies the interior dλ -ball condition at P ∈ ∂D if there exists a constant R > 0 and a point z0 ∈ D, such that the ball B(z0 , R) ⊂ D and P ∈ ∂B(z0 , R), where B(z0 , R) = {z ∈ Cn+1 | dλ (z, z0 ) < R}. Lemma 4.4. Let D ⊂ Cn+1 satisfy the interior dλ -ball condition at the point P0 ∈ ∂D and let U ∈ C 2 (D) ∩ C 1 (D), be a solution of − LU ≥ c(z)U in D (4.4) with c ∈ L∞ (D). Suppose that U (z) > U (P0 ) = 0 for every z ∈ D. Then U (P0 ) − U (P0 − sν) < 0. lim s→0 s where ν is the outer normal to ∂D in P0 . Proof. By assumption, there exist a point z0 = (x̂1 , ..., x̂n , ŷ1 , ..., ŷn , t̂, λ̂) and a radius R > 0 such that the ball B(z0 , R) ⊂ D and P0 ∈ ∂B(z0 , R). We consider the function 2 ψ = U e−K(x1 −x̂1 ) , for K > 0. An easy computations yields ∂U ∂ 2ψ ∂ 2U −K(x1 −x̂1 )2 2 2 = e [4K (x − x̂ ) U − 2KU − 4K(x − x̂ ) + ], 1 1 1 1 ∂x21 ∂x1 ∂x21 and 2 ∂ 2ψ −K(x1 −x̂1 )2 ∂ U = e , ∂x2j ∂x2j 2 ∂ 2ψ −K(x1 −x̂1 )2 ∂ U = e , ∂yk2 ∂yk2 where j = 2, · · · , n, k = 1, · · · , n. 2 ∂ 2ψ −K(x1 −x̂1 )2 ∂ U = e , ∂λ2 ∂λ2 14 ELEONORA CINTI AND JINGGANG TAN Moreover, ∂ 2ψ ∂U ∂ 2U 2 = e−K(x1 −x̂1 ) [−2K(x1 − x̂1 ) + ], ∂x1 ∂t ∂t ∂xj ∂t 2 ∂ 2ψ 2 ∂ U = e−K(x1 −x̂1 ) , ∂xj ∂t ∂xj ∂t Therefore, we have 2 Lψ + 4K(x1 − x̂1 )X1 ψ = e−K(x1 −x̂1 ) [−4K 2 (x1 − x̂1 )2 U + LU − 2KU ]. and hence for K sufficiently large, we deduce −Lψ − 4K(x1 − x̂1 )X1 ψ ≥ 0. 2 2 We introduce now the function φ = e−αR − e−αρ , where ρ = dλ (z, z0 ), and 0 < ρ < R. We can choose α sufficiently large such that − Lφ − 4K(x1 − x̂i )X1 φ 2 =[ r 2 (4α2 ρ2 − 2Qα) − 8Kα(x1 − x̂1 )X1 ρ]e−αρ ≥ 0. 2 ρ Let A := B(z0 , R) \ B(z0 , R1 ) for 0 < R1 < R. For ε small enough ψ(z) + εφ(z) ≥ 0 in ∂A := ∂B(z0 , R) ∪ ∂B(z0 , R1 ) Then, by Lemma 4.2 we obtain that ψ(z) + εφ ≥ 0 in A. Therefore, using that ψ(P0 ) = φ(P0 ) = 0, we deduce that ψ(P0 ) − ψ(P0 − sν) + ε(φ(P0 ) − φ(P0 − sν)) ≤ 0. Using that φ is strictly increasing in ρ, we deduce ψ(P0 ) − ψ(P0 − sν) < 0, s→0 s which, in turns, implies that lim U (P0 − sν) − U (P0 ) < 0. s→0 s lim For any Ω ⊂ R2n+1 , we denote by C the infinite cylinder C = Ω × (0, +∞). Before proving our second Hopf’s Lemma, let us recall the notion of interior ball condition in the Heisenberg group. NONLOCAL LIOUVILLE THEOREM 15 Definition 4.5. Let Ω ⊂ R2n+1 . We say that Ω satisfies the interior Heisenberg ball condition at ξ ∈ ∂Ω if there exists a constant R > 0 and a point ξ0 ∈ Ω, such that the Heisenberg ball BH (ξ0 , R) ⊂ Ω and ξ ∈ ∂BH (ξ0 , R). Lemma 4.6. Let Ω ⊂ Hn satisfy the interior Heisenberg ball condition at the point P ∈ ∂Ω and let w ∈ C 2 (C) ∩ C 1 (C), be a nonnegative solution of ( −Lw ≥ c(z)w in C (4.5) −∂λ w ≥ d(z)w on Ω, with c ∈ L∞ (C) and d ∈ L∞ (Ω). Suppose that w((P, 0)) = 0 Then ∂ν w(P, 0) < 0 where ν is the outer normal to ∂Ω in P . (4.6) Proof. We follow the proof of Lemma 2.4 in [6]. Without loss of generality, we can assume that Ω intersects the positive ξ1 half-axes and that P = (b, 0, ..., 0) for some b > 0. We have to show that ∂ξ1 w(P, 0) < 0. (4.7) By the maximum principle and the Lemma 4.4 we have that w > 0 on C ∪ Ω. (4.8) We start by proving the lemma in the case c(z) = d(z) ≡ 0. Since Ω satisfies the interior Heisenberg ball condition at P , there exist z0 ∈ Ω × {0} and R > 0, such that the ball B + (z0 , R) is contained in the cylinder C and (∂C ∩ {λ > 0}) ∩ ∂B + (z0 , R) = {(P, 0)}. We may assume that z0 = (a, 0, ..., 0) and we consider the set + + A = B (z0 , b − a) \ B (z0 , (b − a)/2) ∩ {λ > 0}. We observe that {(P, 0)} = ∂A ∩ ∂C. 2 2 For z ∈ A we consider the function η(z) = e−αρ − e−α(b−a) , where ρ = dλ (z, z0 ). Writing L in radial coordinate as in (2.2), we have that r2 2 Lη(ρ) = 2 4α2 ρ2 − 2Qα e−αρ , ρ and therefore, for α sufficiently large, we have that − Lη ≤ 0. (4.9) By (4.8) we deduce that w > 0 on ∂B + (z0 , (b − a)/2) ∩ {λ ≥ 0}. Hence, we may choose ε > 0 such that w − εη ≥ 0 on ∂B + (z0 , (b − a)/2) ∩ {λ ≥ 0}. 16 ELEONORA CINTI AND JINGGANG TAN Claim: w − εη ≥ 0 in A. Indeed, using (4.9), we deduce that −L(w − εη) ≥ 0 in A. Hence, by the maximum principle, we have that the minimum of w − εη is attained on ∂A. Now, on one side we have that w − εη ≥ 0 on ∂A ∩ ∂B + (z0 , b − a) ∪ ∂B + (z0 , (b − a)/2) . On the other side, since ∂λ η = 0 on {λ = 0}, we deduce that −∂λ (w − εη) ≥ 0 on ∂A∩{λ = 0}. Thus, using Lemma 4.4, we conclude that the + minimum of w−εη cannot be achieved on B (z0 , b − a) \ B + (z0 , (b − a)/2) ∩ {λ = 0}. This conclude the proof of the claim. Finally, since (w−εη)((P, 0)) = 0, we deduce that ∂ξ1 (w−εη)((P, 0)) ≤ 0, which in turns implies that ∂ξ1 w((P, 0)) < 0 using that ∂ξ1 η((P, 0)) < 0. This concludes the proof of the theorem in the case c(z) = d(z) ≡ 0. In the general case, we introduce the function v = e−βλ w. An easy computation yields, for β large enough: −Lv = −e−βλ Lw + β 2 w ≥ (c(z) + β 2 )v ≥ 0 in C, and −∂λ v ≥ (β + d(z))v ≥ 0. The result finally follows after applying the first part of the proof to the function v. 5. Proof of Theorems 1.3 and 1.4 In this last section we give the proof of our Liouville-type result. We recall that U = U (|(x, y)|, t, λ) is a solution of in Hn × R+ , LU = 0 (5.1) −∂λ U = U p on Hn , U > 0. First of all, we consider the CR inversion of U . For z = (x, y, t, λ) ∈ R , we consider the function 1 z ), w(z) = Q−1 U (e ρ hP i1 n 1 2 2 2 2 2 2 2 2 4 where ze = ρ4 (xt + yr , yt − xr , −t, λρ ) and ρ = ( i=1 (xi + yi ) + λ ) + t . We have seen in Lemma 3.1 that w satisfies ( Lw = 0 in Hn × R+ \ {0}, (5.2) −∂λ w = ρp(Q−1)−(Q+1) wp on Hn . 2n+2 NONLOCAL LIOUVILLE THEOREM 17 Observe that the function w could be singular at the origin and it satisfies limρ→∞ ρQ−1 w(z) = U (0). We start now applying the moving plane method. We will move an hyperplane orthogonal to the t-direction and use the H-reflection. More precisely, for any µ ≤ 0, let Tµ = {z ∈ R2n+2 | t = µ}, and Σµ = {z ∈ R2n+2 : t < µ}. For z ∈ Σµ , we define zµ = (y, x, 2µ − t, λ). To avoid the singular point, we consider e µ = Σµ \ {eµ }, Σ where eµ = (0, 0, 2µ, 0) is the reflection of the origin. Let wµ (z) = wµ (|(x, y)|, t, λ) := w(|(x, y)|, 2µ−t, λ) = w(y, x, 2µ−t, λ) = w(zµ ), and Wµ (z) := wµ (z) − w(z) := w(zµ ) − w(z) z ∈ Σµ . By using the invariance of the operator under the CR transform as in Lemma 3.1 and the fact that |zµ | ≤ |z|, we have that ( LWµ = 0 in R2n+2 , + n ∂λ Wµ ≥ c(z, µ)Wµ on H , Ψp−1 µ where c(z, µ) = ρ(Q+1)−p(Q−1) and Ψµ (z) is between w(z) and wµ (z). By the definition of wµ and w, we have that c(z, µ) ≈ C/ρ2 at infinity. The following lemma will let us to start moving the hyperplane Tµ . Proposition 5.1. Let Bρ+ = B + (0, ρ) be the half ball in R2n+2 . Assume + that w ∈ C 2 (R2n+2 )\{0} satisfies (5.2). Then (i) For µ < 0 )∩C 1 (R2n+2 + + with |µ| large enough, we have Wµ ≥ 0. (ii) For µ < 0 with |µ| large enough, if inf Σµ Wµ < 0, then the infimum is attained at some point z0 ∈ Σµ \ {eµ } such that |z0 | is bounded. Proof. Define for z ∈ B1+ \ Bρ+0 and 0 < ρ0 < 1, ε ελ ε ρQ−1 φ0 (z) = − 0 Q−1 + . 4 |z| 2 Let ψ = w − φ0 . An easy computation implies that ( Lψ ≤ 0 in B1+ \ Bρ+0 , ∂λ ψ = wp − 2ε on ∂(B1+ \ Bρ+0 ) ∩ ∂Rn+1 + . It is clear that ψ = w − (ε/4 − ε + ελ/2) > 0 on ∂Bρ+0 ∩ ∂Bρ0 and that ψ = w − [ε/4 − ρQ−1 ε 0 |z|Q−1 + ελ/2] > w − ε > 0 on ∂B1+ ∩ ∂B1 . We claim that ψ ≥ 0 in B1+ \ Bρ+0 . 18 ELEONORA CINTI AND JINGGANG TAN Suppose by contradiction that there exists P0 = (x0 , y0 , t0 , 0), ρ0 < |P0 | < 1 such that ψ(P0 ) = minB + \Bρ+ ψ < 0. Then, 1 0 −∂λ ψ |P0 ≤ 0. This inequality, together with the Neumann boundary condition for ψ, imply that ε p1 ε ≥ , w(P0 ) ≥ 2 4 which in turns implies that ε ψ(P0 ) = w(P0 ) − φ0 (P0 ) > w(P0 ) − > 0. 4 This give a contradiction with w(P0 ) < 0. Thus, we have proven that ψ ≥ 0 in B1+ \ Bρ+0 . By letting ρ0 → 0, we deduce that w(z) ≥ 4ε for + every z ∈ B 1 \ {0}. We observe that lim |z|Q−1 w(z) = lim U (z̃) = U (0). |z|→∞ |z|→∞ 1 For ε > 0, letting bε = max{U (z̃) | |z̃| = |z| ≥ 1ε }, we see that w(z) ≤ bε for all |z| ≥ 1ε . |z|Q−1 e µ = Σµ \ {eµ }, if |µ| is sufficiently large. Hence Wµ = wµ − w > 0 in Σ This completes part (i). To prove (ii), since by definition of w, w(z) → 0 as |z| → ∞, it follows that for µ < 0 large in absolute value, we have w ≤ 8ε in B1+ (eµ ). For such µ we therefore have that Wµ > 0 in B1+ (eµ ). Thus, combining that Wµ vanishes on the plane Tµ , if inf Σe µ Wµ < 0, then Wµ attains its + infimum in the compact set B R1 ∩ (Σµ \ B1+ (eµ )). 2 Proof of Theorem 1.4. By Proposition 5.1, we know that for µ negative and large in absolute value we have that Wµ ≥ 0 in Σµ . Let us define µ0 = sup{µ < 0 | Wµ ≥ 0}. We only need to prove that µ0 = 0. Suppose that µ0 6= 0 by contradiction. By continuity, Wµ0 ≥ 0 in Σµ0 . By the maximum principle, we deduce that Wµ0 ≡ 0 in Σµ0 or Wµ0 > 0 on Σµ0 ∪ (∂Σµ0 ∩ {λ = 0} ∩ {t < µ0 }) \ {zµ0 }. (5.3) Using the Neumann condition satisfied by Wµ0 and the assumption µ0 > 0, we see that Wµ0 ≡ 0 is impossible. Therefore (5.3) holds. By the definition of µ0 there exists µk → µ0 , µk < µ0 such that inf Σµk Wµk < 0. We observe that for some positive d: 2n+2 min Wµ0 (z) : z ∈ ∂B + (zµ0 , |µ0 |/2) ∩ R+ = d. NONLOCAL LIOUVILLE THEOREM 19 From this fact, using a similar argument to the one of point i) in Proposition 5.1, we deduce that Wµ0 ≥ d in B + (zµ0 , |µ0 |/2) \ {zµ0 }. Therefore, we have that lim inf Wµk (z) : z ∈ B + (zµk , |µ0 |/2) \ {zµk } ≥ d. k→∞ Using this bound and the fact that Wµk (z) → 0 as |z| → ∞, we deduce that for k large enough, the negative infimum of Wµk is attained at some point zk ∈ Σµk \ B + (zµk , |µ0 |/2). By Proposition 5.1 we know that the sequence {zk } is bounded and therefore, after passing to a subsequence, we may assume that zk → z0 . By (5.3) we have that Wµ0 (z0 ) = 0 and z0 ∈ ∂Σµ0 ∩ {t = µ0 }. If zk ∈ Σµk for an infinite number of k, then ∇Wµk (zk ) = 0, and therefore, by continuity ∇Wµ0 (z0 ) = 0. (5.4) (z0 ) < 0, If z0 ∈ ∂Σµ0 ∩ R2n+2 , then by Lemma 4.4, we have that ∂w + ∂t which give a contradiction. Analogously, using Lemma 4.6, we get a contradiction if we assume that z0 ∈ ∂Σµ0 ∩ {λ = 0} ∩ {t = µ0 }. In the case in which zk ∈ ∂Σµk ∩ {λ = 0} ∩ {t < µk }, we still have that the derivatives of Wµk at zk in all directions except the λ direction vanish. Passing to the limit and arguing as above, we get a contradiction. Hence we have established that µ0 = 0. This implies that v is even in t, but since the origin 0 on the t-axes is arbitrary, we can perform the CR transform with respect to any point and then we conclude that w is constant in the direction t. This shows that U is actually a solution of the following problem ( 2n+1 ∆U = 0 in R+ , (5.5) p 2n −∂λ U = U on R . Q+1 Since Q−1 = 2n+3 < 2n+1 , we conclude the proof by using the standard 2n+1 2n−1 Liouville type theorem for problem (5.5) (see [6, 17]). 2 Proof of Theorem 1.3. We observe that if u is a cilindrical function, that is u = u(|(x, y)|, t), then its extension U satisfying (1.2) is also cilindrical in the all halfspace Hn × R+ , in the sense that U = U (|(x, y)|, t, λ). This follows by the definition of the extension operator, which is constructed by taking the Fourier transform in Hn (see for details Section 5.3, and in particular formula (5.9) in [9]). Using this fact, the conclusion follows as a corallary of Theorem 1.4. 2 20 ELEONORA CINTI AND JINGGANG TAN Acknowledgements: Both authors were supported by Spain Government grant MTM2011-27739-C04-01; J.T. was supported by Chile Government grant Fondecyt 1120105, USM 121402; CMM in Universidad de Chile. References [1] Berestycki H.; Nirenberg L.; On the method of moving planes and the sliding method, Bol. Soc. Brasil. Mat. 22 (1991), 1-37. [2] Birindelli, I.; Prajapat, J.; Nonlinear Liouville theorems in the Heisenberg group via the moving plane method, Comm. Part. Diff. Equa. 24 (1999), no. 9-10, 1875-1890. [3] Birindelli, I.; Capuzzo Dolcetta, I.; Cutri, A Liouville theorems for semilinear equations on the Heisenberg group, Ann. Inst. H. Poin. Anal. Non Lin. 14 (1997), no. 3, 295-308. [4] Caffarelli, L.; Silvestre, L., An extension problem related to the fractional Laplacian, Comm. in Part. Diff. Equa. 2 (2007), 1245-1260. [5] Chen W.; Li C., Classification of solutions of some nonlinear elliptic equations, Duke Math. J. 63 (1991), 615–622. [6] Chipot M.; Chlebı́k M.; Fila M.; Shafrir I.; Existence of positive solutions of a semilinear elliptic equation in Rn+ with a nonlinear boundary condition, J. Math. Anal. Appl. 223 (1998), 429–471. [7] Epstein, C.; Melrose, R.; Mendoza, G.; Resolvent of the Laplacian on pseudoconvex domains, Acta Math. bf 167 (1991), 1-106. [8] Ferrari, F; Franchi, B; Harnack inequality for fractional laplacians in Carnot groups, preprint. [9] Frank, R.; Gonzalez M.d.M; Monticelli, D.; Tan, J.; Conformal fractional Laplacians on the Heisenberg group, preprint. [10] Garofalo, N; Lanconelli, E: Existence and nonexistence results for semilinear equations on the Heisenberg group, Indiana Univ. Math. J. 41 (1992), 71-98. [11] Gidas B.; Ni W.-M.; Nirenberg L.; Symmetry and related properties via the maximum principle, Comm. Math. Phys. 68 (1979), 209-243. [12] Gidas B., Spruck J. ; Global and local behavior of positive solutions of nonlinear elliptic equations, Comm. Pure App. Math. 35 (1982), 528–598. [13] Gover A., Graham C.R.; CR invariant powers of the sub-Laplacian, J. Reine Angew. Math. 583 (2005),1–2,. [14] Guillarmou, C. Sa Barreto, A. Scattering and inverse scattering on ACH manifolds, J. Reine Angew. Math. 622 (2008) 1–55. [15] Hislop, P.; Perry, P.; Tang, S., CR-invariants and the scattering operator for complex manifolds with boundary, Anal. PDE 1 (2008) 197–227. [16] Lanconelli E., Uguzzoni F., Asymptotic behavior and non-existence theorems for semilinear Dirichlet problems involving critical exponent on unbounded domains of the Heisenberg group, Boll. Unione Mat. Ital. Sez. B Artic. Ric. Mat. (8) 1 (1998), no. 1, 139-168. [17] Li, Y.Y.; Zhang, L.; Liouville-type theorems and Harnack-type inequalities for semilinear elliptic equations, J. Anal. Math. 90 (2003), 27-87. [18] Li, Y.Y.; Zhu, M.; Uniqueness theorems through the method of moving spheres, Duke Math. J. 80 (1995), 383-417. NONLOCAL LIOUVILLE THEOREM 21 [19] Jerison, D.S.; Lee, J.M.; The Yamabe problem on CR manifolds, J. Diff. Geo. 25, (1987), 167–197. [20] Serrin, J.; A symmetry problem in potential theory, Arch. Ration. Mech. Anal. 43 (1971), 304-318. [21] Uguzzoni, F., A non-existence theorem for a semilinear Dirichlet problem involving critical exponent on halfspaces of the Heisenberg group, NoDEA Nonlinear Diff. Equa. Appl. 6 (1999), 191-206. [22] Xu, L; Semi-linear Liouville theorems in the Heisenberg group via vector field methods, J. Differential Equations 247 (2009), 2799-2820. E.C., Dipartimento di Matematica, Università degli Studi di Bologna, Piazza di Porta San Donato 5, 40126 Bologna (Italy) E-mail address: [email protected] J.T., Departamento de Matemática, Universidad Técnica Federico Santa Marı́a, Avda. España 1680, Valparaı́so (Chile) E-mail address: [email protected]
© Copyright 2024 Paperzz