The density of states and local eigenvalue statistics for random band matrices of fixed width

TL;DR

This paper proves that local eigenvalue statistics for one-dimensional fixed bandwidth random band matrices follow a Poisson process, identifying its intensity and extending localization bounds and density of states analysis.

Contribution

It extends localization bounds and density of states regularity results to fixed bandwidth random band matrices, establishing Poisson eigenvalue statistics and identifying the process intensity.

Findings

Eigenvalue statistics follow a Poisson point process

Density of states converges uniformly and smoothly

Results extend to higher dimensions under localization bounds

Abstract

We prove that the local eigenvalue statistics for $d=1$ random band matrices with fixed bandwidth and, for example, Gaussian entries, is given by a Poisson point process and we identify the intensity of the process. The proof relies on an extension of the localization bounds of Schenker \cite{schenker} and the Wegner and Minami estimates. These two estimates are proved using averaging over the diagonal disorder. The new component is a proof of the uniform convergence and the smoothness of the density of states function. The limit function, known to be the semicircle law with a band-width dependent error \cite{bmp,dps,dl,mpk}, is identified as the intensity of the limiting Poisson point process. The proof of these results for the density of states relies on a new result that simplifies and extends some of the ideas used by Dolai, Krishna, and Mallick \cite{dkm}. These authors proved regularity properties of the density of states for random Schr\"odinger operators (lattice and continuum) in the localization regime. The proof presented here applies to the random Schr\"odinger operators on a class of infinite graphs treated by in \cite{dkm} and extends the results of \cite{dkm} to probability measures with unbounded support. The method also applies to fixed bandwidth RBM for $d=2,3$ provided certain localization bounds are known.