On the role of geometry in scattering theory for nonlinear Schrödinger equations Rémi Carles (CNRS & Université Montpellier 2) Orléans, April 9, 2008 Free Schrödinger equation on Rn: 1 i∂tu + ∆u = 0 ; u|t=0 = u0 ∈ L2(Rn). 2 Solution known explicitly, and long time asymptotics available: i 2t ∆ e u0(x) ∼ t→±∞ |x|2 i 2t e 1 x . F (u0) n/2 t (2iπt) Defocusing nonlinear Schrödinger equation on Rn: 1 i∂tu + ∆u = |u|2σ u 2 ; u|t=0 = u0 ∈ H 1 ∩ F (H 1)(Rn). 2 , u ∈ C(R; H 1 ∩ F (H 1 )). If 0 < σ < n−2 Large time behavior: 1 < σ < 2 , there is scattering (at least in this weak sense): • If n n−2 (1) ∃u± ∈ L2(Rn), t u(t) − ei 2 ∆u± −→ 0. 2 t→±∞ L 1 : (1) is possible only in the trivial case u = u = 0. • If 0 < σ 6 n ± Long range effect. Tool: dispersive property of the free group, i t ∆ e 2 1 L →L∞ . 1 |t|n/2 . Goal of this talk. What happens if we add a potential in NLS? What happens in the case of non-Euclidean geometries? Introducing an external potential. 1 i∂tu + ∆u = |u|2σ u + V u, 2 with V = V (x) ∈ R. We mention four particular cases: • Linear potential: V (x) = E · x, for E ∈ Rn. • Harmonic potential: V (x) = +ω 2|x|2. • Repulsive harmonic potential: V (x) = −ω 2|x|2. 2 x2 , n > 2. • Mixture of the above two: e.g., V (x) = −ω12x2 + ω 1 2 2 Linear potential: V (x) = E · x. (With Yoshihisa Nakamura.) This case is trivial, in view of Avron–Herbst formula. Set v(t, x) = u t, x − t2 2 ! 3 t 2 i tE·x− 3 |E| E e . Then the equation for u is equivalent to: 1 i∂tv + ∆v = |v|2σ v. 2 The potential does not change the scattering theory. 2 Harmonic potential: V (x) = + ω2 |x|2. 2 ω 1 Local in time dispersive properties: set H = − 2 ∆ + 2 |x|2. !n/2 ω −itH . . e 1 L →L∞ | sin(ωt)| H has eigenvalues: no dispersion for large time, hence no scattering (at least in the usual sense). 2 Repulsive harmonic potential: V (x) = − ω2 |x|2. 2 1 ω Global in time dispersive properties: set H = − 2 ∆ − 2 |x|2. !n/2 ω −itH . . e 1 L →L∞ | sinh(ωt)| Exponential decay, instead of algebraic decay. Scattering in Σ = H 1 ∩ F (H1), with no long range effect: 2 1 ω |x|2u, i∂tu + ∆u = |u|2σ u − 2 2 2 : u(t) ∼ e−itH u . for all 0 < σ < n−2 ± t→±∞ Repulsive harmonic potential: sketch of the proof. −itH e 1 L →L∞ . ω | sinh(ωt)| !n/2 6 C(n, d) 1 |t|d/2 , ∀d > n. Same Strichartz estimates as on Rd, ∀d > n. On Rd, wave operators in H 1 for 2/d 6 σ < 2/(d − 2). Since d > n is arbitrary, 0 < σ < 2/(n − 2). Wave operators in Σ. −itH itH ∇ replaced by J(t) = e ∇ e · = −iω sinh(ωt)x + cosh(ωt)∇. Evolution law for kJ(t)uk2 (// pseudo-conformal law): L2 asymptotic completeness in Σ. NB: The (free) energy is not signed: J(t) is more geometric. ω12 2 ω22 2 A hybrid model: V (x) = − 2 x1 + 2 x2. 1∆ + V , Mehler: H = − 2 −itH e 1 L →L∞ . ω1 | sinh(ω1t)| !1/2 ω2 | sin(ω2t)| !1/2 1 |t| !(n−2)/2 = w(t)n/2. w ∈ L1w (R): HLS+usual proof for Strichartz estimates ⇒ Strichartz estimates like in Rn with V ≡ 0 wn/d ∈ L1w (R), ∀d > n: Strichartz estimates like in Rd, ∀d > n. Example. Global existence and scattering in L2 for small data for all 0 < σ 6 2/n: no long range. If ω1 is large enough (compared to 1, and to ω2), global existence and scattering in Σ. Otherwise?? Another open question. If V (x) = |x|α or hxiα, for 0 < α < 2. Local existence of solution to the nonlinear equation: OK. Scattering theory? With J.-F. Bony, D. Häfner and L. Michel: in the linear case, complete answer. The usual short range assumption |Vpert(x)| . hxi−γ with γ > 1 becomes α |Vpert(x)| . hxi−γ , with γ > 1 − . 2 Similar “interpolation” for nonlinear equations, when changing the geometry instead of introducing an external potential. Changing the geometry. Many results on compact manifolds: • Case of Tn: J. Bourgain (Strichartz estimates). • Compact manifold without boundary: N. Burq-P. Gérard-N. Tzvetkov. More recently, R. Anton, M. Blair-H. Smith-C. Sogge. No scattering theory available (more complicated dynamics): there are eigenvalues. Nonlinear Schrödinger on non-compact manifolds (except Rn. . . ). • Asymptotically flat metric: D. Tataru-G. Staffilani, N. Burq, L. RobbianoC. Zuily, D. Tataru, BGT, J.-M. Bouclet-N. Tzvetkov. . . • Asymptotically conic metric: A. Hassell-T. Tao-J. Wunsch. • Case of the hyperbolic space Hn: V. Banica, V. Banica-RC-G. Staffilani, A. Ionescu-G. Staffilani, J.-P. Anker-V. Pierfelice. • More general Damek–Ricci spaces: V. Pierfelice. • Asymptotically hyperbolic manifolds: J.-M. Bouclet. • Variations on hyperbolic: N. Burq-C. Guillarmou-A. Hassell. More reasonable to expect scattering. An intuitive parallel. Potential Manifold 0 Rn : non-perturbed case +|x|2 Sn : strongly trapping geometry −|x|2 Hn : strong dispersion Scattering on hyperbolic space. Hn = n Rn+1 3 Ω = (x0, . . . , xn) = (cosh r, sinh r ω), r > o n−1 0, ω ∈ S . Def. Radial functions: f (Ω) independent of ω ∈ Sn−1, f (Ω) = f (r). NB: r = distHn (0, Ω). We denote U (t) = eit∆Hn . We consider the nonlinear equation: i∂tu + ∆Hn u = |u|2σ u ; U (−t)u(t) t=t0 = ϕ. NB: The nonlinearity is defocusing. 2 Strichartz estimates global existence in H 1 (Hn) for 0 < σ < n−2 (V. Banica). Weighted global in time Strichartz estimates in Hn. n−1 sinh r 2 w(r) := ; U (t) = eit∆Hn . r 2d and Def. Let d > 2: (p, q) is d-admissible if 2 6 q 6 d−2 ! 2 1 1 = δ(q) := d − , p 2 q (p, q) 6= (2, ∞) . Proposition (V. Banica, V. Pierfelice). Let n > 3. 1. For any n-admissible pair (p, q), there exists Cq such that 1− 2 w q U (·)φ p L (R;Lq ) n for every radial function φ ∈ L2 rad (H ). 6 Cq kφkL2 2. Similar weighted estimates for inhomogeneous equations. Corollary (with V. Banica and G. Staffilani). Let d > n > 3. Strichartz estimates hold for d-admissible pairs and radial functions on Hn: 1. For any d-admissible pair (p, q), there exists Cq = Cq (n, d) s.t. kU (·)φkLp(R;Lq ) 6 Cq kφkL2 , n ). ∀φ ∈ L2 ( H rad 2. For any d-admissible pairs (p1, q1) and (p2, q2) and any interval I, there exists Cq1,q2 = Cq1,q2 (n, d) independent of I such that Z U (t − s)F (s)ds I∩{s6t} p L 1 (I;Lq1 ) 0 0 q 2 (Hn ) . for every F ∈ Lp2 I; Lrad 6 Cq1,q2 kF k p0 L 2 q0 , I;L 2 Remark. True also when n = 2 ([BCS]). Remark. Extended recently to the non-radial case: [AP], [IS]. How does one prove scattering in H 1(Rd)? • Strichartz estimates for d-admissible pairs. • Hölder inequality. 2d . • Sobolev embedding: H 1(Rd) ⊂ Lp(Rd) for 2 6 p 6 d−2 2 . Yields the existence of wave operators in H 1(Rd), for 2d 6 σ < d−2 1 (Hn ), n > 2! Claim: We have everything we need in Hrad 2 . Algebraic conditions: d > n > 2 and 2d 6 σ < d−2 2 is admissible: existence of wave operators in Any 0 < σ < n−2 1 (Hn ), n > 2. Hrad No smallness condition, and no long range effect! n ), for 0 < σ 6 2 : Remark. Similar approach for small data in L2 ( H rad n • existence of wave operators, • global existence, • asymptotic completeness, n in a small ball around 0 in L2 rad (H ). Asymptotic completeness in H 1(Hn). Morawetz inequality: 2 < σ < 2 (general), 0 < σ < 2 (radial). • [BCS] for n = 3: 3 • With V. Banica and T. Duyckaerts: n > 4, radial. • [Ionescu-Staffilani]: n > 2, no radial assumption. 2 . Asymptotic completeness for 0 < σ < n−2 From Euclidean to hyperbolic. (With V. Banica and T. Duyckaerts.) For n, k > 1, let φ(r) = k X 1 r2j+1. j=0 (2j + 1)! Denote by Mkn the rotationally symmetric manifold, with metric ds2 = dr2 + φ(r)2dω 2. Remark. For k = 0: Rn. For k = ∞: Hn. Laplace–Beltrami : φ0(r) 2 ∆M = ∂r + (n − 1) ∂r + φ(r) 1 ∆Sn−1 . 2 φ(r) NLS on Mkn. i∂tu + ∆M u = |u|2σ u, e x ∈ Mkn, u(t, x) = u(t, r). Introduce N = (2k + 1)(n − 1) + 1: N > n for k > 0. Existence of wave operators (V. Banica-T. Duyckaerts) and asymptotic completeness for 2 2 <σ< . N n−2 Remark. For k > 1, 2/N < 1/n: the frontier short range/long range has already moved. Remark. For σ 6 1/N , “formal” proof of the presence of long range effects: study of the free dynamics. Free dynamics. i∂tu + ∆M u = 0, x ∈ Mkn, u(0, x) = u0(r). n ) to L2 (Rn ), such that There exists L unitary from L2 (M rad k rad ku(t) − v(t)kL2(M ) −→ 0, with v(t, r) = 2 eir /(4t) tn/2 t→+∞ ! n−1 r φ(r) 2 r . (Lu0) t Idea of the proof. New unknown φ(r) v(t, r) = u(t, r) r ! n−1 2 : i∂tv + ∆Rn v = F v, n − 1 φ00(r) (n − 1)(n − 3) φ0(r) F (r) = + 2 φ(r) 4 φ(r) !2 1 1 − 2 =O . 2 r 1+r it∆ n Linear scattering for v: v(t) − e R v+ 2 n −→ 0. L (R ) t→+∞ Standard result: 2 /(4t) i|x| x e it∆Rn F ϕ . ϕ − Λ(t) 2 n −→ 0, with Λ(t, x) = e ( ) n/2 L (R ) t→+∞ 2t t Existence of long range effects. Suppose n > 2 and 0 < σ 6 1/N , where N = (2k + 1)(n − 1) + 1. Let n 2 n u ∈ C([T, ∞[; L2 rad (Mk )) sol. NLS, u+ ∈ Lrad (Mk ) with it∆ Mu u(t) − e + L2 −→ 0. t→+∞ Let ψ ∈ C0∞(M ) be radial, and t2 > t1 > T . By assumption, D ψ, e−it2∆M u(t2) − e−it1∆M u(t1) = −i E Z t D 2 t1 E it∆ 2σ e M ψ, |u| u (t) dt goes to zero as t1, t2 → +∞. Free dynamics: D E 1 Z ∞ eit∆M ψ, |u|2σ u (t) ≈ nσ+n t 0 for ϕ = Lψ|Lu+|2σ Lu+. r φ(r) !(n−1)(σ+1) r φ(r)n−1dr t ϕ Change of variable r 7→ tr: 1 Z ∞ tnσ+n−1 0 tr φ(tr) !(n−1)(σ+1) ϕ (r) φ(tr)n−1dr. For r > 1 and large t, the function at stake behaves like 1 tnσ+n−1 tr (tr)2k+1 !(n−1)(σ+1) r−(N −n)σ+n−1 (n−1)(2k+1) ϕ(r) (tr) = ϕ(r). N σ t This function of t is not integrable, unless ϕ ≡ 0. This means that Lu+ = 0 = u+ (Ker L = {0}). The assumption and the conservation of mass then imply u ≡ 0.