On diffusion in high-dimensional Hamiltonian systems and PDE (Q1580484)

The author investigates the nonlinear Schrödinger equation \(iu_t+\Delta u+\partial_{\bar u}P(|u|^2)=0\) for Nekhoroshev-type stability on the torus \(T^1\), where \(P(x)=\nu x^3+\) (higher-order terms) is a polynomial. He shows that for a typical initial data \(u(0)=\phi\) and large enough exponent \(s\) the solution of the Cauchy problem satisfies \(\|u(t)\|_{H^s}<C\varepsilon\) for \(-\varepsilon^{-B}<t<\varepsilon^{-B}\). If the exponent \(s\to\infty\), then \(B=B(s)\to 0\). The equation written in (infinite-dimensional) Hamiltonian form is considered in the perturbation theory framework and is simplified by the Birkhoff normal form of \textit{S. Kuksin} and \textit{J. Pöschel} [Ann. Math., II. Ser. 143, No. 1, 149-179 (1996; Zbl 0847.35130)]. Then the author confronts a small divisor problem. A similar long-time stability result was considered by \textit{D. Bambusi} [Commun. Math. Phys. 189, No. 1, 205-226 (1997; Zbl 0895.35095)], where a stronger Fourier localization of the data was assumed. There the topology was defined to be \(H^1\), while \(H^s\)-estimates of the present paper allow to control the perturbative terms and to weaken the localization of the Fourier modes. However tending \(s\to\infty\) spoils the diophantine properties of the modulated frequencies. This restricts the methods and permits to make conclusions only for low-dimensional tori \(T^d\) (actually in the paper \(d=1\)). The author however expects that several other PDEs, where the global Cauchy problem is well-posed, can be attacked via Hamiltonian perturbations of non-resonant linear systems on bounded domains or on \(T^d\). But the unperturbed system should not be ``too large''. At the end of the paper the author exhibits a diffusive Hamiltonian even in a finite-dimensional (though big) space, where a restriction on the \(\ell^2\)-bound comparing to the perturbation size \(\varepsilon\) is violated.