Stochastic and long-distance level spacing statistics in many-body localization

Abstract: From the random matrix theory all the energy levels should be strongly correlated due to the presence of all off-diagonal entries.In this work we introduce two new statistics to more accurately characterize these long-distance interactions in the disordered many-body systems with only short-range interaction. In the $(p, q)$ statistics, we directly measure the long distance energy level spacings, while in the second approach, we randomly eliminate some of the energy levels, and then measure the reserved $\eta\%$ energy levels using nearest-neighbor level spacings. We benchmark these results using the results in standard Gaussian ensembles. Some analytical distribution functions with extremely high accuracy are derived, which automatically satisfy the inverse relation and duality relation. These two measurements satisfy the same universal scaling law during the transition from the Gaussian ensembles to the Poisson ensemble, with critical disorder strength and corresponding exponent are independent of these measurements. These results shade new insight into the stability of many-body localized phase and their universal properties in the disordered many-body systems.