Thouless and relaxation time scales in many-body quantum systems

TL;DR

This paper analytically and numerically investigates the time scales of relaxation in many-body quantum systems, revealing exponential growth with system size and differences between chaotic and localized regimes.

Contribution

It introduces a spectral correlation-based method to analytically determine Thouless and relaxation times, linking them to universal dynamical features in many-body quantum systems.

Findings

Thouless and relaxation times grow exponentially with system size.

In chaotic systems, Thouless time is much less than relaxation time.

In localized systems, Thouless and relaxation times converge.

Abstract

A major open question in studies of nonequilibrium quantum dynamics is the identification of the time scales involved in the relaxation process of isolated quantum systems that have many interacting particles. We demonstrate that long time scales can be analytically found by analyzing dynamical manifestations of spectral correlations. Using this approach, we show that the Thouless time, $t_{\text{Th}}$, and the relaxation time, $t_{\text{R}}$, increase exponentially with system size. We define $t_{\text{Th}}$ as the time at which the spread of the initial state in the many-body Hilbert space is complete and verify that it agrees with the inverse of the Thouless energy. $t_{\text{Th}}$ marks the point beyond which the dynamics acquire universal features, while relaxation happens later when the evolution reaches a stationary state. In chaotic systems, $t_{\text{Th}}\ll t_{\text{R}}$, while for systems approaching a many-body localized phase, $t_{\text{Th}}\rightarrow t_{\text{R}}$. Our analytical results for $t_{\text{Th}}$ and $t_{\text{R}}$ are obtained for the survival probability, which is a global quantity. We show numerically that the same time scales appear also in the evolution of the spin autocorrelation function, which is an experimental local observable. Our studies are carried out for realistic many-body quantum models. The results are compared with those for random matrices.