Long time Anderson localization for the nonlinear random Schrödinger equation (Q1022367): Difference between revisions

The paper studies the lattice nonlinear equation in one dimension \[ i\dot q_j = v_j q_j + \varepsilon_1 (q_{j-1}+q_{j+1}) + \varepsilon_2 |q_j|^2 q_j, \quad j\in \mathbb Z, \] where \(v_j\) is a collection of i.i.d. uniform random variables on \([0,1]\) and \(\epsilon_1\), \(\epsilon_2\) are positive parameters. It proves long-time Anderson localisation for arbitrary \(\ell^2\) initial data in the form inspired by the RAGE theorem. Namely, it shows that given \(A>1\), \(\delta >0\), initial datum \(q_j(0)\in \ell^2\) and \(j_0\) satisfying \[ \sum_{|j|>j_0}|q_j(0)|^2< \delta, \] there exist \(C\), \(\epsilon\) and \(N\) depending on \(A\) such that for all \(t\leq (\delta/C)\epsilon^{-A}\) one has, with a high probabililty, \[ \sum_{|j|>j_0+N}|q_j(t)|^2< 2\delta. \] The proof uses a Birkhoff normal type transform to create a barrier where there is essentially no propagation. One of the novelties of this transform is that it is in small neighbourhoods only, enabling to treat ``rough'' data without any moment condition.