Stochastic nonlinear Schrödinger equations and modulation of solitary waves

Transcription

1 Stochastic nonlinear Schrödinger equations and modulation of solitary waves A. de Bouard CMAP, Ecole Polytechnique, France joint work with R. Fukuizumi (Sendai, Japan) Deterministic and stochastic front propagation Banff, mars 22-26, 2010

2 Optically confined Bose-Einstein condensates Stamper-Kurn et al., Phys. Rev. Lett, 1998 Advantages Obtain different geometrical configurations study magnetic properties of atoms (trapping not limited to specific magnetic states)

3 Drawbacks ex : fluctuations of the laser intensity introduce stochasticity in the dynamical behavior of the condensate, which has to be taken into account in real situations Dynamics of BEC under regular variations of trap parameters : widely studied Castin and Dum, Phys. Rev. Lett Kagan, et. al ; Phys. Rev. A, 1996 Ripoll, Perez-Garcia, Phys. Rev. A, 1999, etc... Mean field theory : fluctuations of laser field intensity regarded as modulations of the harmonic trap NLS equation (Gross-Pitaevskii) with harmonic potential and noise

4 Randomly varying optical trap potential Abdullaev, Baizakov, Konotop, in Nonlinearity and disorder, D radially symmetric Gross-Pitaevskii equation : i ψ t = 1 r r r ψ r + (1 + e(t))r 2 ψ + χ ψ 2 ψ iγψ in dimensionless variables, with χ = ±1 (related to the sign of the s-wave scattering length a of atoms) γ damping coefficient (thermal cloud) e(t) = E(t) E 0 E 0 with E(t) = laser field intensity, mean value E 0. Assume e(t) is δ-correlated with zero mean : e(t) = 0, e(t)e(t ) = σ 2 δ(t t )

5 Mathematical description of the equation (Ω, F, P) probability space W (t) real valued standard Brownian motion ; e(t) = εẇ (t) ψ macroscopic wave function (complex valued) Use of a Stratonovich product : idψ + ( ψ x 2 ψ)dt χ ψ 2 ψdt + iγψdt = ε x 2 ψ dw Conservation of the squared L 2 norm (total number of atoms) in the absence of damping (W is real valued) limit case of processes with non vanishing correlation length Equivalent equation in Itô form : idψ+( ψ x 2 ψ)dt χ ψ 2 ψdt+iγψdt = ε x 2 ψdw i ε2 2 x 4 ψdt

6 More generally : Consider the equation idψ+( ψ x 2 ψ)dt χ ψ 2α ψdt+iγψdt+i ε2 2 x 4 ψdt = ε x 2 ψdw α > 0, γ 0, x R d, d = 1 or 2. We consider solutions with paths having finite energy almost surely (Σ : energy space) H(ψ) = 1 ( ψ 2 + x 2 ψ 2 )dx + χ ψ 2α+2 dx 2 2α + 2 H : hamiltonian for the corresponding deterministic equation (without damping) : combination of the energy of the wave packet, and mean square width of the atomic cloud

7 Existence result (up to now...) Theorem : db, Fukuizumi, 2007 Assume α > 0, γ 0 and χ = ±1. Assume ψ 0 Σ if d = 1, or ψ 0 Σ 2 and 1/2 α 1 if d = 2. Then there exist a stopping time τ (ψ 0 ) and a unique solution ψ ε (t) of idψ+( ψ x 2 ψ)dt χ ψ 2α ψdt+iγψdt+i ε2 2 x 4 ψdt = ε x 2 ψdw with ψ ε (0) = ψ 0, such that ψ ε C([0, τ]; Σ) for any τ < τ (ψ 0 ), and ψ ε is adapted w.r. to the filtration generated by W. Moreover, we have almost surely, τ (ψ 0, ω) = + or lim sup ψ ε (t) Σ = +. t τ (ψ 0,ω)

8 where : ψ 0 Σ 2 if ψ 0 L 2, ψ 0 L 2 and x 2 ψ 0 L 2 Moreover : χ = +1 or χ = 1 and α < 2/d or χ = 1, α = 2/d and ψ 0 4/d < 1/C L 2 α the solution in Σ exists for all t i.e. τ (ψ 0 ) = + a.s. Pathwise conservation of L 2 norm Energy equality (Itô formula) for all τ < τ (ψ 0 ) a.s. H(ψ(τ)) = H(ψ 0 ) 2εIm From now on : τ 0 τ x. ψ ψdxdw (s)+2ε 2 xψ(s) 2 L ds 2 χ = 1 (attractive condensate), γ = 0 (no damping) 0

9 Standing wave of the deterministic equation Two parameter family of solutions (ε = 0) ψ µ,θ (t, x) = e i(µt+θ) φ µ (x), θ, µ R φ µ localized profile, positive, radially symmetric (ground state), critical point of S µ (u) = H(u) + µ 2 u 2 L 2 φ µ exists and is unique provided µ inf { u 2 L 2 + xu 2 L 2, u Σ, u L 2 = 1 } = d Moreover as µ d, φ µ (α + 1) d/4α π d/4 (µ + d) 1/2α Φ 0 where Φ 0 ground state of + x 2 (first Hermite function in 1-D)

10 Di Menza, Hadj Selem, 2009

11

12

13 Dynamics of the deterministic equation The family {e iθ φ µ, θ R} is stable for µ close to d (Fukuizumi, Ohta, D.I.E. 2003) Consider the functional S µ as a Lyapunov functional ; then S µ(φ µ ) = L µ satisfies L µ v, v ν v 2 Σ for any v = (v 1, v 2 ) t, v i Σ, with (v, φ µ ) = (v, iφ µ ) = 0. Let ψ be a solution of the equation (with ε = γ = 0) and write ψ(t, x) = e iθ(t) [φ µ (x) + v(t, x)] with θ(t) such that (v 2, φ µ ) = (v, iφ µ ) = 0 ; then S µ (e iθ(t) ψ(t, x)) S µ (φ µ ) = S µ (ψ(0, x)) S µ (φ µ ) = S µ(φ µ )v (S µ(φ µ )v, v) + o( v 2 Σ ) In general, (v, φ µ ) = (v 1, φ µ ) 0, but conservation of L 2 norm (v, φ µ ) = O( v 2 Σ )

14 Dynamics of the stochastic equation db, Fukuizumi, 2009 Let ψ ε (0, x) = φ µ0 (x) (µ 0 close to d, α 1/2) ; in order to use the stability of the standing wave of the deterministic equation, write the solution ψ ε of the stochastic equation as ψ ε (t, x) = e iθε (t) (φ µ ε (t)(x) + εη ε (t, x)) with θ ε (t) and µ ε (t) random modulation parameters, chosen such that for all t, (η ε (t), φ µ0 ) = (η ε (t), iφ µ0 ) = 0. This decomposition holds as long as εη ε Σ δ and µ ε (t) µ 0 δ for δ > 0 sufficiently small. Let : τ ε δ = inf{t > 0, εηε H 1 δ or c ε (t) c 0 δ} then : C(α, µ 0 ) > 0, such that T > 0, ε 0 > 0 such that ε ε 0, P(τ ε δ < T ) exp( C ε 2 T )

15 Change of the orthogonality conditions Spectral projection on the generalized null-space of JL µ0 : defined for w = w 1 + iw 2 by P µ0 w = ( µ φ µ0, φ µ0 ) 1 [(w 1, φ µ0 ) µ φ µ0 + i(w 2, µ φ µ0 )φ µ0 ] preceding orthogonality conditions do not imply P µ0 η ε = 0 Change in the orthogonality conditions : by setting θ ε (t) = θ ε (t) εh(t), with h(t) a well chosen Itô process (driven by W ) one gets ψ ε (t, x) = e i θ ε (t) (φ µ ε (t)(x) + ε η ε (t, x)) for t τ ε α and with : P µ0 η ε = O(ε)

16 The modulation parameters At first order in ε, the equations for the modulation parameters are given by dµ ε (t) = o(ε), d θ ε (t) = µ 0 dt ε ( x 2 φ µ0, µ φ µ0 ) dw + o(ε). (φ µ0, µ φ µ0 ) This shows in particular that at first order the noise does not act on the frequency of the standing wave, but only on its phase. Note that coupling with η ε only at next order (due to the change in the orthogonality conditions)

17 Central limit Theorem Theorem : db, Fukuizumi, 2009 Assume d = 1 and α 1, or d = 2 and α = 1. Then, for any T > 0, the process ( η ε (t)) t [0,T τ ε α ] converges in probability, as ε goes to zero, to a process η satisfying ( ) 0 d η = JL µ0 ηdt (I P µ0 ) x 2 dw, φ µ0 with η(0) = 0, where P µ0 is the spectral projection onto the generalized null space of JL µ0. The convergence holds in C([0, τδ ε T ], L2 ). Moreover, η satisfies for any T > 0 the estimate ( ) E for some constant C > 0. sup η(t) 2 Σ t T CT (1)

18 Asymptotics on the frequency Assume d = 1 and α > 1 ; as the frequency decreases to 1, the operator JL µ0 converges to ( 2 J x + x 2 ) x 2 + x which has simple, purely imaginary eigenvalues ξ ± k = ±2i(k + 1), and a corresponding complete system of eigenfunctions Let η ± k be the corresponding component of the process η ; then as µ 0 goes to 1 π E( η 2 ± (t) 2 1 ) = (α + 1) 2α (µ0 + 1) 1/α t + O((µ 0 + 1) κ+1/α t), 4 E( η ± k (t) 2 ) = O((µ 0 + 1) κ+1/α t) for k 2. with κ = min{1 1/α, 1/2α} > 0.

19 Conclusion We have considered small multiplicative, time white noise perturbations of a NLS equation with confining potential and standing wave (ground state) as initial data The time scale on which the solution stays in a neighborhood of the randomly modulated standing wave is ε 2. We obtained at order one a simple behaviour for the modulation parameters A central limit theorem holds, i.e. the order one part of the remaining term converges as ε goes to 0 to a centered Gaussian process As the frequency tends to its minimal value, this latter process concentrates in the third eigenmode Open problem : parameters of dark solitons (positive scattering length) : much less invariances in the equation