Dispersive estimates in Rn (n ≥ 2) for the Schrödinger and the wave equations Simon Moulin Nantes, Laboratoire de Mathématiques Jean Leray 11 avril 2008 - Orléans Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 1 / 43 1 Introduction 2 New results Wave equation (n ≥ 3) Schrödinger equation (n ≥ 4) Dimension n = 2, high frequencies 3 Ideas of the proofs 4 Conclusion Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 2 / 43 1 Introduction 2 New results Wave equation (n ≥ 3) Schrödinger equation (n ≥ 4) Dimension n = 2, high frequencies 3 Ideas of the proofs 4 Conclusion Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 3 / 43 Goals Plan of study for Cauchy problem associated to evolution problems. To study the linear problem. To obtain a priori estimates for the linear problem. To study only some equations and to obtain a priori estimates for these equations. To search what properties could be generalized. To study non-linear problems by perturbative methods. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 4 / 43 A priori estimates Energy conservation. Dispersive estimates: Lpx (Rn ) → Lqx (Rn ). Strichartz estimates. Strichatz estimates are a priori estimates, for example for one-order PDE: L2x (Rn ) → Lpt (0, t; Lqx (Rn )). For two-order PDE 2, one have to have more regularity on initial conditions. Ingredients to obtain these estimates: dispersive estimates, T ∗ T argument, Sobolev embeddings (for example Hardy-Littlewood-Sobolev inequality). Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 5 / 43 Schrödinger equation Schrödinger equation i∂t u − ∆u = 0 with initial condition u(0, x) = u0 (x). Conservation of the L2 norm: ku(t)kL2 = kû(t)kL2 = kû0 kL2 = ku0 kL2 . Kernel: u(t, x) = e−it∆ u0 (x) = 1 (4πit)n/2 Z ei |x−y |2 4t u0 (y) dy. Rn The L1 − L∞ estimate: ku(t)kL∞ ≤ C|t|−n/2 ku0 kL1 , t 6= 0. Interpolation between L2 − L2 estimate and L1 − L∞ estimate. 0 Lp − Lp estimates with 2 ≤ p ≤ +∞ and α = 1 − 2/p: n −it∆ e p0 p ≤ C|t|−α 2 . L →L Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 6 / 43 Equations Wave equation: ∂t2 u − ∆u = 0, u(0, x) = u0 (x), ∂t u(0, x) = u1 (x). Schrödinger equation: i∂t u − ∆u = 0, u(0, x) = u0 (x). Dispersion relation: Schrödinger ω = kξk2 and wave ω = ±kξk. Group speed: Schrödinger vg = 2ξ and wave vg = ±ξ/kξk. Phase speed: Schrödinger ξ · (vφ − ξ) = 0 and wave ξ · (vφ ∓ ξ/kξk) = 0. We try to establish global in time decay estimates of the solution u(t, x) according to initial data u0 . Dispersion for Schrödinger ku(t)kL∞ ≤ C|t|−n/2 ku0 kL1 , t 6= 0. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 7 / 43 Equations Wave equation: ∂t2 u − ∆u = 0, u(0, x) = u0 (x), ∂t u(0, x) = u1 (x). Schrödinger equation: i∂t u − ∆u = 0, u(0, x) = u0 (x). Dispersion relation: Schrödinger ω = kξk2 and wave ω = ±kξk. Group speed: Schrödinger vg = 2ξ and wave vg = ±ξ/kξk. Phase speed: Schrödinger ξ · (vφ − ξ) = 0 and wave ξ · (vφ ∓ ξ/kξk) = 0. We try to establish global in time decay estimates of the solution u(t, x) according to initial data u0 . Dispersion for Schrödinger ku(t)kL∞ ≤ C|t|−n/2 ku0 kL1 , t 6= 0. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 7 / 43 High and low frequencies Free operator spectra σ(G0 ) = σess (G0 ) = [0, +∞[. High frequencies: λ 1, h 1. Low frequencies: λ 1, h 1. R0 (λ) = (G0 − λ2 )−1 = h2 (h2 G0 − z 2 )−1 = h2 R0,h (z) with z = hλ. Cut-off functions: ηa for low frequencies and χa for high frequencies. With ηa (G) + χa (G) = Pac . High frequencies: oscillations, low regularity. Low frequencies: slow variations, high regularity. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 8 / 43 Dispersion for the wave equation Dispersive estimates for the free wave equation L1 − L∞ low frequencies estimates (n ≥ 3, ∀t): √ it G0 −(n+1)/4+ε G0 ηa (G0 ) ≤ Cε hti−(n−1)/2 , e 1 ∞ L →L L1 − L∞ high frequencies estimates (n ≥ 2, ∀t 6= 0): √ n−1 it G0 −(n+1)/4−ε G0 χa (G0 ) ≤ Cε |t|− 2 , e 1 ∞ L →L 0 < ε 1. 0 < ε 1. 0 Lp − Lp estimates (n ≥ 3, 2 ≤ p < +∞ et α = 1 − 2/p): √ n−1 it G0 −α(n+1)/4 G0 Pac 0 ≤ C|t|−α 2 . e Lp →Lp Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 9 / 43 Problem: Which properties of dispersive estimates are preserved if we perturb equations with a potential ? Perturbed equations: i∂t u − ∆u + V (x)u = 0, ∂t2 u − ∆u + V (x)u = 0, x ∈ Rn , n ≥ 2. Obstacles to decay for the perturbed problem: Eigenvalues. Waves u(t, x) = eiλt u0 (x). No decay in time. Resonances. Results of Jensen-Kato (79) and Rauch (78): decay in hti−1/2 instead of hti−3/2 for Schrödinger equation with a potential (in 3D) if 0 resonance. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 10 / 43 Problem: Which properties of dispersive estimates are preserved if we perturb equations with a potential ? Perturbed equations: i∂t u − ∆u + V (x)u = 0, ∂t2 u − ∆u + V (x)u = 0, x ∈ Rn , n ≥ 2. Obstacles to decay for the perturbed problem: Eigenvalues. Waves u(t, x) = eiλt u0 (x). No decay in time. Resonances. Results of Jensen-Kato (79) and Rauch (78): decay in hti−1/2 instead of hti−3/2 for Schrödinger equation with a potential (in 3D) if 0 resonance. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 10 / 43 How to well-pose the problem ? Assumptions on V to define eitG et eit √ G . Assumptions on V to have: no eigenvalues > 0 and no resonances > 0, σ(G) = [0, ∞[∪{λj ≤ 0}. Typical assumption: V ∈ L∞ (Rn ), |V (x)| ≤ Chxi−δ , δ > 1. (V (δ)) Assumption 0 not eigenvalue nor resonance: 0 regular point. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 11 / 43 Known results Schrödinger equation (0 regular) Same estimations as the free problem under the following assumptions: Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 12 / 43 Known results Schrödinger equation (0 regular) Same estimations as the free problem under the following assumptions: Journée, Soffer, Sogge (91): dim ≥ 3, V (δ) avec δ > n + 4 + technical conditions, V̂ ∈ L1 . High frequencies: iteration of Duhamel formula and regularization estimates. δ > n et V̂ ∈ L1 . Low frequencies: expansions of the resolvant near 0 proved by Jensen-Kato. δ > n + 4. Yajima (95): same assumptions with δ > n + 2. Let W = s − lim e−itG eitG0 the usual wave operator. t→∞ eitG Pac = WeitG0 W ∗ and if kW kL∞ →L∞ is finite, then keitG Pac k1→∞ = kWeitG0 W ∗ k1→∞ ≤ C|t|−n/2 Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 12 / 43 Known results Schrödinger equation (0 regular) Same estimations as the free problem under the following assumptions: Journée, Soffer, Sogge (91): dim ≥ 3, V (δ) avec δ > n + 4 + technical conditions, V̂ ∈ L1 . Yajima (95): same assumptions with δ > n + 2. Schlag (04): dim 2, V (δ) with δ > 3. Schlag, Rodnianski (04): dim 3, small potential (small Rollnik norm and Kato norm) Vodev (05), Yajima (05): dim 3, V (δ) with δ > 5/2. Vodev: use of Φ(t; h) = eitG φ(h2 G) − eitG0 φ(h2 G0 ) and semi-classical type estimates with small h. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 12 / 43 Known results Schrödinger equation (0 regular) Same estimations as the free problem under the following assumptions: Journée, Soffer, Sogge (91): dim ≥ 3, V (δ) avec δ > n + 4 + technical conditions, V̂ ∈ L1 . Yajima (95): same assumptions with δ > n + 2. Schlag (04): dim 2, V (δ) with δ > 3. Schlag, Rodnianski (04): dim 3, small potential (small Rollnik norm and Kato norm) Vodev (05), Yajima (05): dim 3, V (δ) with δ > 5/2. Golberg (06): dim 3, V ∈ L3/2− (R3 ) ∩ L3/2+ (R3 ), 0 < 1. Expansion in Born series by iterating R(z) = R0 (z) − R0 (z)VR(z), plus limiting absorption principle. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 12 / 43 Wave equation (0 regular) Beals, Strauss (93): dim ≥ 3, V (δ) avec δ > n + 1 et V small enough or δ > 3n, V ≥ 0 plus conditions on the derivatives. Beals (94): dim ≥ 3, V ∈ S(Rn ) + V ≥ 0 or small potential. (Iteration of Duhamel formula plus estimates on the kernels.) Cardoso, Cuevas, Vodev (05): dim 2 and 3, V (δ) with δ > (n + 1)/2 + in dim 2 no resonances > 0, high frequencies estimates. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 13 / 43 Known results in dimension n ≥ 4 Yajima (Schrödinger): δ > n + 2 and V̂ ∈ L1 (+ technical conditions). Beals, Strauss (Ondes): decay (δ > n + 1) + small potential or decay + regularity (V ∈ S(Rn )) + V ≥ 0. Some methods: Resolvant properties (limiting absorption principle, Bessel function properties). Resolvant expansion around 0. Duhamel formula, iteration of Duhamel formula. Wave operator (Yajima). Goldberg, Visan (06) The usual dispersive estimate for the Schrödinger equation in n ≥ 4 is not true for a compact support potential V , V ∈ C α , α < (n − 3)/2. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 14 / 43 Known results in dimension n ≥ 4 Yajima (Schrödinger): δ > n + 2 and V̂ ∈ L1 (+ technical conditions). Beals, Strauss (Ondes): decay (δ > n + 1) + small potential or decay + regularity (V ∈ S(Rn )) + V ≥ 0. Some methods: Resolvant properties (limiting absorption principle, Bessel function properties). Resolvant expansion around 0. Duhamel formula, iteration of Duhamel formula. Wave operator (Yajima). Goldberg, Visan (06) The usual dispersive estimate for the Schrödinger equation in n ≥ 4 is not true for a compact support potential V , V ∈ C α , α < (n − 3)/2. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 14 / 43 Problems Problem 1 Which dispersive estimate can we obtain with decay assumption only ? Problem 2 Which regularity assumption do we have to make to obtain usual dispersive estimate ? Problem 3 Which is the minimal assumption that we have to make ? Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 15 / 43 High frequencies results Vodev (06): in dim n ≥ 4, at high frequencies, V (δ) with δ > (n + 1)/2 for the wave equation, δ > (n + 2)/2 for the Schrödinger equation, same decay as the free problem but with a loss of (n − 3)/2 derivatives. Dispersive estimates (for all a > 0): √ it G −(n−1)/2−ε G χa (G) 1 ∞ ≤ Cε |t|−(n−1)/2 , ∀t 6= 0, ∀0 < ε 1, e L →L itG −(n−3)/4 χa (G) e G L1 →L∞ Simon Moulin (Nantes, LMJL) ≤ C|t|−n/2 , Dispersive estimates ∀t 6= 0, 11 avril 2008 - Orléans 16 / 43 1 Introduction 2 New results Wave equation (n ≥ 3) Schrödinger equation (n ≥ 4) Dimension n = 2, high frequencies 3 Ideas of the proofs 4 Conclusion Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 17 / 43 Dispersive estimates (Wave n ≥ 3) Assumption linked with the resolvant kernel (n ≥ 3): Z sup |x − y |−n+2 + |x − y |−(n−1)/2 |V (x)| dx < +∞. y∈Rn (VOn) Rn Integrability assumption (n = 3): V ∈ L3/2−ε (R3 ), 0 < ε 1. (L3/2−) Low frequencies Assumptions (VOn) in dim n ≥ 4 and (VOn)+(L3/2-) in dim n = 3, 0 regular. There exists a0 > 0 as for all 0 < a ≤ a0 : √ n−1 it G −(n+1)/4+ε G ηa (G) 1 ∞ ≤ Cε hti− 2 , ∀t, ∀0 < ε 1. e L →L Dispersive estimate Assumptions V(δ) with δ > (n + 1)/2 and 0 regular. For all 0 < ε 1, t 6= 0: √ it G −(n+1)/4+ε G hGi−(n−3)/4−2ε Pac 1 ∞ ≤ Cε |t|−(n−1)/2 , e L →L Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 18 / 43 Dispersive estimates (Wave n ≥ 3) Assumption linked with the resolvant kernel (n ≥ 3): Z sup |x − y |−n+2 + |x − y |−(n−1)/2 |V (x)| dx < +∞. y∈Rn (VOn) Rn Integrability assumption (n = 3): V ∈ L3/2−ε (R3 ), 0 < ε 1. (L3/2−) Low frequencies Assumptions (VOn) in dim n ≥ 4 and (VOn)+(L3/2-) in dim n = 3, 0 regular. There exists a0 > 0 as for all 0 < a ≤ a0 : √ n−1 it G −(n+1)/4+ε G ηa (G) 1 ∞ ≤ Cε hti− 2 , ∀t, ∀0 < ε 1. e L →L Dispersive estimate Assumptions V(δ) with δ > (n + 1)/2 and 0 regular. For all 0 < ε 1, t 6= 0: √ it G −(n+1)/4+ε G hGi−(n−3)/4−2ε Pac 1 ∞ ≤ Cε |t|−(n−1)/2 , e L →L Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 18 / 43 Dispersive estimates (Schrödinger n ≥ 4) Assumption linked with the resolvant kernel with 0 < ε 1: Z sup |x − y |−n+2 + |x − y |−(n−2)/2+ε |V (x)| dx < +∞, y∈Rn (VSn) Rn Low frequencies Assumptions (VSn) and 0 regular. There exists a0 > 0 as for all 0 < a < a0 : itG 6 0. e ηa (G) 1 ∞ ≤ C|t|−n/2 , ∀t = L →L Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 19 / 43 Dispersive estimates (Schrödinger n ≥ 4) Dispersive estimate Assumptions V(δ) with δ > (n + 2)/2 and 0 regular: itG e hGi−(n−3)/4 Pac 1 ∞ ≤ C|t|−n/2 , L →L ∀t 6= 0. Dispersive estimate with more regularity 0 regular, V(δ) with δ > n − 1 and V̂ ∈ L1 (Rn ). itG e Pac 1 ∞ ≤ C|t|−n/2 , L →L Simon Moulin (Nantes, LMJL) Dispersive estimates ∀t 6= 0. 11 avril 2008 - Orléans 20 / 43 Dispersive estimates (Schrödinger n ≥ 4) Dispersive estimate Assumptions V(δ) with δ > (n + 2)/2 and 0 regular: itG e hGi−(n−3)/4 Pac 1 ∞ ≤ C|t|−n/2 , L →L ∀t 6= 0. Dispersive estimate with more regularity 0 regular, V(δ) with δ > n − 1 and V̂ ∈ L1 (Rn ). itG e Pac 1 ∞ ≤ C|t|−n/2 , L →L Simon Moulin (Nantes, LMJL) Dispersive estimates ∀t 6= 0. 11 avril 2008 - Orléans 20 / 43 High frequencies dispersive estimate in dim 2 Assumption linked with the resolvant kernel: Z sup |x − y |−1/2 |V (x)|dx ≤ C < +∞. y ∈R2 (V 2) R2 Wave Assumption (V2): there exists a0 > 0 as for all a ≥ a0 : √ it G −3/4− G χa (G) 1 ∞ ≤ C |t|−1/2 , t 6= 0, ∀ 0 < ε 1, e L →L Schrödinger Assumption (V2): there exists a0 > 0 as for all a ≥ a0 : itG e χa (G) 1 ∞ ≤ C|t|−1 . L →L Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 21 / 43 1 Introduction 2 New results Wave equation (n ≥ 3) Schrödinger equation (n ≥ 4) Dimension n = 2, high frequencies 3 Ideas of the proofs 4 Conclusion Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 22 / 43 Bessel and Hankel function properties The free resolvant (h2 G0 − z 2 )−1 is defined for ±Im z > 0. Its kernel is of the form Rh± (|x − y|, z) where Rh± (σ, z) = ±h−2 iσ −2ν ± H (σz/h) = h−n R1± (σh−1 , z), 4(2π)ν ν with ν = (n − 2)/2, Hν± (λ) = λν Hν± (λ), Hν± are the outgoing and incoming Hankel functions of order ν. These functions behaves differently around 0 and at infinity. For large λ, Hν± (λ) = e±iλ bν± (λ), where bν± (λ) is a symbol of order (n − 3)/2. ± ± Near λ = 0, Hν± (λ) = aν,1 (λ) + λn−2 log λ aν,2 (λ), ± ± ± where aν,1 et aν,2 analytic functions, aν,2 ≡ 0 if n odd. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 23 / 43 Kernels associated to the free problem Thanks to the Stone formula Z √ 1 ∞ itλ eit G0 ψ(h2 G0 ) = e ψ(h2 λ2 )(R0+ (λ) − R0− (λ))λdλ, iπ 0 √ the kernel of the operator eit G0 ψ(h2 G0 ) is of the form Kh (|x − y |, t) with Z ∞ σ −2ν (w) Kh (σ, t) = eitλ Jν (σλ)ψ(h2 λ2 )λdλ = h−n K1 (σh−1 , th−1 ), (2π)ν+1 0 where Jν (z) = z ν Jν (z), Jν (z) = (Hν+ (z) + Hν− (z))/2 is the Besel function of order ν = (n − 2)/2. We have the following estimates for all σ > 0, t 6= 0 and h > 0: (w) |K1 (σ, t)| ≤ Chti−s hσis−(n−1)/2 , s ≥ 0, for n ≥ 2. |Kh (σ, t)| ≤ C|t|−s h−(n+1)/2 σ −(n−1)/2+s , for n ≥ 2, Moreover if 0 ≤ s ≤ (n − 1)/2, (w) (S) |Kh (σ, t)| ≤ Chs−(n−1)/2 |t|−s−1/2 σ −(n−1)/2+s , Simon Moulin (Nantes, LMJL) Dispersive estimates for n ≥ 3. 11 avril 2008 - Orléans 24 / 43 Perturbative methods Resolvant formulaes: R ± (λ) − R0± (λ) = R0± (λ)VR ± (λ) = R ± (λ)VR0± (λ). If inverse exists: R(z) = (1 + R0 (z)V )−1 R0 (z) = R0 (z)(1 + VR0 (z))−1 , R + (λ) − R − (λ) = (1 + R0+ (λ)V )−1 R0+ (λ) − R0− (λ) (1 + R0− (λ)V )−1 . Duhamel formula: eitG = eitG0 + i Z t ei(t−τ )G Veiτ G0 dτ, 0 e it √ √ G =e it G0 √ √ Z t p sin(t G0 ) √ sin((t − τ ) G0 ) iτ √G √ +i √ ( G − G0 ) − Ve dτ. G0 G0 0 Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 25 / 43 Assumption 0 regular Assumption 0 regular: the operator T = (1 − V ∆−1 )−1 : L1 → L1 is bounded. We have (n ≥ 3) ± VR (λ) − VR ± (0) 1 1 ≤ Cλβ , 0 < λ ≤ 1, β > 0. 0 0 L →L As 1 + VR0± (0) = 1 − V ∆−1 is invertible on L1 , 1 + VR0± (λ) is invertible on L1 for 0 < λ ≤ λ0 . Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 26 / 43 Low frequencies comparators (n ≥ 3) Ψ(t; h) = eitG ψ(h2 G) − T ∗ eitG0 ψ(h2 G0 )T (Schrödinger) √ √ Φ(t; h) = eit G ψ(h2 G) − T ∗ eit G0 ψ(h2 G0 )T (wave) Estimates on ψ(h2 G0 ) et ψ(h2 G). Estimates on perturbation terms: Z ∞ itG0 Ve ψ(h2 G0 ) 1 1 dt, L →L −∞ Z ∞ −∞ itG Ve ψ(h2 G) ∞ Z −∞ Z L1 →L dt, 1 ∞ −∞ √ it G0 ψ(h2 G0 ) Ve L1 →L1 it √G ψ(h2 G) Ve L1 →L1 dt, dt. Estimates on comparators. Estimates on propagators. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 27 / 43 Use of low frequencies (h ≥ 1) A important point in the proofs is the following estimate (n ≥ 2): Z ∞ √ it G0 h ψ(h2 G0 )f 1 dt ≤ γn Cn (V )h−(n−3)/2 kf kL1 , Ve L −∞ h > 0, where ψ ∈ C0∞ ((0, +∞)), γn > 0 is a constant independant of V , h and f , and Z |V (x)|dx Cn (V ) := sup < +∞. (n−1)/2 n y∈R Rn |x − y | Our approch is based on the fact that if Cn (V )h−(n−3)/2 1, then under raisonnable assumptions on the potential, we have Z ∞ it √G e n (V )h−(n−3)/2 kf k 1 . ψ(h2 G)f 1 dt ≤ C h Ve L −∞ Simon Moulin (Nantes, LMJL) L Dispersive estimates 11 avril 2008 - Orléans 28 / 43 More details in the proof: wave n ≥ 3 Proposition 3 (dim n ≥ 3) kΦ(t; h)kL1 →L∞ ≤ Ch−(n+1)/2−β |t|−(n−1)/2 , Proof of the low frequency estimate (dim n ≥ 3): Z ∞ We write σ −(n+1)/4 ηa (σ) = ψ(σθ)θ(n+1)/4−1 dθ, h ≥ h0 . σ > 0. a−1 √ √ −(n+1)/4 Then eit G G−(n+1)/4 ηa (G) − T ∗ eit G0 G0 ηa (G0 )T 1 ∞ L →L Z ∞ √ ≤ Φ(t, θ) 1 ∞ θ(n+1)/4−1 dθ L →L a−1 ≤ C|t|−(n−1)/2 Z ∞ θ−1−β/2 dθ ≤ C|t|−(n−1)/2 . a−1 Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 29 / 43 More details in the proof: wave n ≥ 3 Proposition 3 (dim n ≥ 3) kΦ(t; h)kL1 →L∞ ≤ Ch−(n+1)/2−β |t|−(n−1)/2 , Proof of the low frequency estimate (dim n ≥ 3): Z ∞ We write σ −(n+1)/4 ηa (σ) = ψ(σθ)θ(n+1)/4−1 dθ, h ≥ h0 . σ > 0. a−1 √ √ −(n+1)/4 Then eit G G−(n+1)/4 ηa (G) − T ∗ eit G0 G0 ηa (G0 )T 1 ∞ L →L Z ∞ √ ≤ Φ(t, θ) 1 ∞ θ(n+1)/4−1 dθ L →L a−1 ≤ C|t|−(n−1)/2 Z ∞ θ−1−β/2 dθ ≤ C|t|−(n−1)/2 . a−1 Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 29 / 43 As the we have the resolvant formula ± ± 2 R± h (z)(1 + h V R0,h (z)) = R0,h (z), our proof follows the steps to study properties of Hankel functions Hν± (z), to deduce properties on R± 0,h (z), to deduce properties on h2 V R± 0,h (z), to deduce properties on R± h (z), to deduce properties on the different projectors. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 30 / 43 We can show (n ≥ 3, h ≥ 1): ± R (σ, z) ≤ Ch−2 σ −n+2 + σ −(n−1)/2 . h Then: kh2 V R0,h (z)kL1 →L1 ≤ C sup y∈Rn Z |V (x)| |x − y |−n+2 + |x − y|−(n−1)/2 dx. Rn Notice that, in dimension n = 2, we can show (0 < h < 1): Z kh2 V R0,h (z)kL1 →L1 ≤ Ch1/2 sup |V (x)||x − y |−1/2 dx. y∈Rn Simon Moulin (Nantes, LMJL) Rn Dispersive estimates 11 avril 2008 - Orléans 31 / 43 In dimension n ≥ 4, we obtain: 2 h V R0,h (z) − h2 V R0,h (0) 1 1 ≤ Ch−1/2 . L →L In dimension n = 3, we obtain: 2 h V R0,h (z) − h2 V R0,h (0) 1 1 ≤ Ch−1 sup L →L y∈R3 Z |V (x)|h(x − y)/hi−1 . R3 We have to make an other assumption V ∈ L3/2− (R3 ), 0 < 1, to obtain 2 h V R0,h (z) − h2 V R0,h (0) 1 1 ≤ Ch−γ , n = 3. L →L We also have ± R0,h (z) L1 →L1 ≤ C|Im z|−q , h > 0, Im z 6= 0, and the same estimate for R± h with restrictions on h. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 32 / 43 We then prove in dimension n ≥ 3: Proposition 1 ψ(h2 G0 ) 1 1 ≤ C, h > 0, L →L 2 ψ(h G) 1 1 ≤ C, h ≥ h0 , L →L 2 2 ψ(h G) − ψ(h G0 )T 1 1 ≤ Ch−β , h ≥ h0 . L →L Proposition 2 (dim ≥ 4) √ it G0 ψ(h2 G0 )f ≤ Ch−1−(n−1)/2 |t|−(n−1)/2 kf kL1 , h > 0, e L1 Z ∞ √ it G0 ψ(h2 G0 )f dt ≤ Ch−1−(n−3)/2 kf kL1 , h > 0, Ve 1 L −∞ Z ∞ −∞ it √G ψ(h2 G)f 1 dt ≤ Ch−1−β kf kL1 , Ve Simon Moulin (Nantes, LMJL) L Dispersive estimates h ≥ h0 . 11 avril 2008 - Orléans 33 / 43 We then prove in dimension n ≥ 3: Proposition 1 ψ(h2 G0 ) 1 1 ≤ C, h > 0, L →L 2 ψ(h G) 1 1 ≤ C, h ≥ h0 , L →L 2 2 ψ(h G) − ψ(h G0 )T 1 1 ≤ Ch−β , h ≥ h0 . L →L Proposition 2 (dim ≥ 4) √ it G0 ψ(h2 G0 )f ≤ Ch−1−(n−1)/2 |t|−(n−1)/2 kf kL1 , h > 0, e L1 Z ∞ √ it G0 ψ(h2 G0 )f dt ≤ Ch−1−(n−3)/2 kf kL1 , h > 0, Ve 1 L −∞ Z ∞ −∞ it √G ψ(h2 G)f 1 dt ≤ Ch−1−β kf kL1 , Ve Simon Moulin (Nantes, LMJL) L Dispersive estimates h ≥ h0 . 11 avril 2008 - Orléans 33 / 43 The first two estimates √ follows from (VOn) and estimates on the kernel Kh (|x − y|, t) of eit 0 ≤ s ≤ (n − 1)/2 G0 ψ(h2 G0 ), that is for all σ > 0, h > 0 and |Kh (σ, t)| ≤ Ch−(n+1)/2 |t|−s σ s−(n−1)/2 , Z ∞ |t|s |Kh (σ, t)| dt ≤ Ch−(n−1)/2 σ s−(n−1)/2 . −∞ For the third estimate, we can write Veit √ ψ(h2 G) = (iπh)−1 G X ± Z ± 0 t TVPh± (t − τ )Uh± (τ )dτ, ∞ where Ph± (t) = Z Uh± (t) = Z eitλ ϕ eh (λ) R0± (λ) − R0± (0) dλ, 0 0 ∞ eitλ ϕh (λ)T 1 + (VR0± (λ) − VR0± (0))T Simon Moulin (Nantes, LMJL) Dispersive estimates −1 dλ. 11 avril 2008 - Orléans 34 / 43 The kernel of the operator Ph± (t) is of the form A± h (|x − y |, t) where Z ∞ −1 A± (2π)−ν σ −n+2 eitλ ϕ eh (λ) Hν± (σλ)) − Hν± (0) dλ h (σ, t) = ±i4 0 which satisfied the following estimate: for all σ > 0 et h ≥ 1: Z ∞ ± A (σ, t) dt ≤ Ch−1/2 σ −n+5/2 + σ −(n−1)/2 . h −∞ We also have for some h0 > 0 and for all h ≥ h0 : Z Z ∞ ± U (t)f (x) dtdx ≤ Ckf kL1 . h Rn −∞ This follows from Uh± (t) = T ϕ bh (t) − Simon Moulin (Nantes, LMJL) Z 0 t TVPh± (t − τ )Uh± (τ )dτ . Dispersive estimates 11 avril 2008 - Orléans 35 / 43 Proposition 3 kΦ(t; h)kL1 →L∞ ≤ Ch−(n+1)/2−β |t|−(n−1)/2 , h ≥ h0 . We make use of the Duhamel formula √ √ √ √ √ p Z t sin((t − τ ) G0 ) sin(t G0 ) √ it G it G0 √ √ e =e +i G − G0 − Veiτ G dτ, G0 G0 0 which allows us to obtain the following decomposition: Φ(t; h) = Φ1 (t; h) + Φ2 (t, h). Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 36 / 43 Proposition 3 kΦ(t; h)kL1 →L∞ ≤ Ch−(n+1)/2−β |t|−(n−1)/2 , h ≥ h0 . We make use of the Duhamel formula √ √ √ √ √ p Z t sin((t − τ ) G0 ) sin(t G0 ) √ it G it G0 √ √ e =e +i G − G0 − Veiτ G dτ, G0 G0 0 which allows us to obtain the following decomposition: Φ(t; h) = Φ1 (t; h) + Φ2 (t, h). Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 36 / 43 √ Φ1 (t; h) = ψ1 (h2 G) − T ∗ ψ1 (h2 G0 ) eit G ψ(h2 G) √ +T ∗ ψ1 (h2 G0 )eit G0 ψ(h2 G) − ψ(h2 G0 )T p −iT ∗ ψ1 (h2 G0 ) sin t G0 ψ(h2 G) − ψ(h2 G0 )T p e 2 G) − ψ(h e 2 G0 )T . +iT ∗ ψe1 (h2 G0 ) sin t G0 ψ(h The already proved estimates allows us to obtain: kΦ1 (t; h)f kL∞ ≤ Ch−(n+1)/2−β |t|−(n−1)/2 kf kL1 + Ch−β kΦ(t; h)f kL∞ . Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 37 / 43 √ Φ1 (t; h) = ψ1 (h2 G) − T ∗ ψ1 (h2 G0 ) eit G ψ(h2 G) √ +T ∗ ψ1 (h2 G0 )eit G0 ψ(h2 G) − ψ(h2 G0 )T p −iT ∗ ψ1 (h2 G0 ) sin t G0 ψ(h2 G) − ψ(h2 G0 )T p e 2 G) − ψ(h e 2 G0 )T . +iT ∗ ψe1 (h2 G0 ) sin t G0 ψ(h The already proved estimates allows us to obtain: kΦ1 (t; h)f kL∞ ≤ Ch−(n+1)/2−β |t|−(n−1)/2 kf kL1 + Ch−β kΦ(t; h)f kL∞ . Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 37 / 43 On the other hand: Z Φ2 (t; h) = −h t √ p T ∗ ψe1 (h2 G0 ) sin (t − τ ) G0 Veiτ G ψ(h2 G)dτ. 0 We can have thanks to the already proved estimates: t (n−1)/2 kΦ2 (t; h)f kL∞ ≤ Ch−(n+1)/2−β kf kL1 √ +Ch−β sup τ (n−1)/2 eiτ G ψ(h2 G)f t/2≤τ ≤t L∞ . By summing up the two contributions: t (n−1)/2 kΦ(t; h)f kL∞ ≤ Ch−(n+1)/2−β kf kL1 + Ch−β t (n−1)/2 kΦ(t; h)f kL∞ +Ch−β sup τ (n−1)/2 kΦ(τ ; h)f kL∞ . t/2≤τ ≤t Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 38 / 43 On the other hand: Z Φ2 (t; h) = −h t √ p T ∗ ψe1 (h2 G0 ) sin (t − τ ) G0 Veiτ G ψ(h2 G)dτ. 0 We can have thanks to the already proved estimates: t (n−1)/2 kΦ2 (t; h)f kL∞ ≤ Ch−(n+1)/2−β kf kL1 √ +Ch−β sup τ (n−1)/2 eiτ G ψ(h2 G)f t/2≤τ ≤t L∞ . By summing up the two contributions: t (n−1)/2 kΦ(t; h)f kL∞ ≤ Ch−(n+1)/2−β kf kL1 + Ch−β t (n−1)/2 kΦ(t; h)f kL∞ +Ch−β sup τ (n−1)/2 kΦ(τ ; h)f kL∞ . t/2≤τ ≤t Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 38 / 43 On the other hand: Z Φ2 (t; h) = −h t √ p T ∗ ψe1 (h2 G0 ) sin (t − τ ) G0 Veiτ G ψ(h2 G)dτ. 0 We can have thanks to the already proved estimates: t (n−1)/2 kΦ2 (t; h)f kL∞ ≤ Ch−(n+1)/2−β kf kL1 √ +Ch−β sup τ (n−1)/2 eiτ G ψ(h2 G)f t/2≤τ ≤t L∞ . By summing up the two contributions: t (n−1)/2 kΦ(t; h)f kL∞ ≤ Ch−(n+1)/2−β kf kL1 + Ch−β t (n−1)/2 kΦ(t; h)f kL∞ +Ch−β sup τ (n−1)/2 kΦ(τ ; h)f kL∞ . t/2≤τ ≤t Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 38 / 43 Some ideas in the proof for high frequencies Estimates for the free operator L1 → L∞ and L2w → L∞ : ←− kernel explicit formula. Estimates for the perturbation terms with G0 and with G: ←− resolvant estimates and R ± = R0± (1 + VR0± )−1 . Z ∞ 2 −s itG hxi e ψ(h2 G)f dt ≤ Chkf k2L2 , ∀f ∈ L2 , ∀s > 1/2, 0 < h ≤ 1. L2 −∞ Proof bases on resovant estimates as −s ± hxi R (λ)hxi−s 2 2 ≤ Cλ−1 , L →L λ ≥ λ0 , ∀s > 1/2, λ0 > 0. Estimates for the comparators L2 → L2 , L1 → L∞ and L2w → L∞ : ←− Duhamel formula. Projectors estimates: kφ(h2 G) − φ(h2 G0 )kL1 →L1 ≤ Ch2 , Simon Moulin (Nantes, LMJL) Dispersive estimates ∀0 < h ≤ 1. 11 avril 2008 - Orléans 39 / 43 1 Introduction 2 New results Wave equation (n ≥ 3) Schrödinger equation (n ≥ 4) Dimension n = 2, high frequencies 3 Ideas of the proofs 4 Conclusion Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 40 / 43 L1 − L∞ wave estimates (n ≥ 3) Free operator: low frequencies: |t|−(n−1)/2 + (n + 1)/4 − ε loss. high frequencies: |t|−(n−1)/2 + (n + 1)/2 + ε loss. Perturbed operator: V(δ), δ > (n + 1)/2 and 0 regular Same estimates as the free operator, + (n − 3)/2-derivative loss at high frequencies. L1 − L∞ Schrödinger estimates (n ≥ 4) Free operator: |t|−n/2 decay. Perturbed operator: V(δ), δ > (n + 2)/2 and 0 regular |t|−n/2 decay, + (n − 3)/2-derivative loss at high frequencies, without loss if we add V̂ ∈ L1 and δ > n − 1. Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 41 / 43 Perspectives Can we adapt the proof of the dispersive estimate for Schrödinger with more regularity (V̂ ∈ L1 and (V(δ)) with δ > n − 1) to the dimension 2 ? Which are the minimal regularity assumptions to make to obtain the usual dispersive estimates for Schrödinger equation (without loss of derivatives at high frequencies) ? We hope an assumption like V ∈ C α , α ≥ (n − 3)/2, with still a decay assumption (V(δ)). Which are the minimal regularity assumptions for the wave equation ? ... Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 42 / 43 Some ideas (Cardoso, Cuevas, Vodev) T (h) := (Id + ψ1 (h2 G0 ) − ψ1 (h2 G))−1 = Id + O(h2 ), F0 (t, h) = ψ1 (h2 G0 )eitG0 ψ(h2 G), W0 (t, h) = eitG0 ψ1 (h2 G0 ). We can derived from Duhamel formula Z t eitG ψ(h2 G) = T (h)F0 (t, h) + iT (h)W0 (t − τ, h)Veiτ G ψ(h2 G)dτ. 0 Iterating m times leads to the identity eitG ψ(h2 G) = m X F̃j (t, h) + Rm+1 (t, h), j=0 where Z t W̃0 (t − τ, t)V F̃j−1 (τ, h)dτ, F̃j (t, h) = 0 itG 2 R0 (t, h) = e ψ(h G), Z Rm+1 (t, h) = t W̃0 (t − τ, h)V Rm (τ, h)dτ. 0 Simon Moulin (Nantes, LMJL) Dispersive estimates 11 avril 2008 - Orléans 43 / 43