Absence of true localization in many-body localized phases

TL;DR

This paper demonstrates that many-body localized phases exhibit persistent slow growth in number entropy, indicating that true localization does not occur even at very strong disorder, contrasting with Anderson localization.

Contribution

The study provides numerical evidence that number entropy in MBL phases continues to grow, showing that these systems are never truly localized at finite disorder levels.

Findings

Number entropy grows logarithmically with time in MBL phases.

Particle number distribution develops a growing tail over time.

MBL systems are not fully localized even at strong disorder.

Abstract

We have recently shown that the logarithmic growth of the entanglement entropy following a quantum quench in a many-body localized (MBL) phase is accompanied by a slow growth of the number entropy, $S_N\sim\ln\ln t$. Here we provide an in-depth numerical study of $S_N(t)$ for the disordered Heisenberg chain and show that this behavior is not transient and persists even for very strong disorder. Calculating the truncated R\'enyi number entropy $S_N^{(\alpha)}(t)=(1-\alpha)^{-1}\ln\sum_n p^\alpha(n)$ for $\alpha\ll 1$ and $p(n)>p_c$ -- which is sensitive to large number fluctuations occurring with low probability -- we demonstrate that the particle number distribution $p(n)$ in one half of the system has a continuously growing tail. This indicates a slow but steady increase of the number of particles crossing between the partitions in the interacting case, and is in sharp contrast to Anderson localization, for which we show that $S_N^{(\alpha\to 0)}(t)$ saturates for any cutoff $p_c>0$. We show, furthermore, that the growth of $S_N$ is $\mathit not$ the consequence of rare states or rare regions but rather represents typical behavior. These findings provide strong evidence that the interacting system is never fully localized even for very strong but finite disorder.