arXiv Analytics

Sign in

arXiv:2401.00492 [math.PR]AbstractReferencesReviewsResources

Edge statistics for random band matrices

Dang-Zheng Liu, Guangyi Zou

Published 2023-12-31Version 1

Consider Hermitian and symmetric random band matrices $H=(\sigma_{xy}A_{xy})$ on the $d$-dimensional lattice $\left(\mathbb{Z}/{L\mathbb{Z}}\right)^d$, where $A_{xy}=\overline{A_{yx}}$ are independent uniformly distributed random variables on $S^1$ or $\{+1, -1\}$, and the variance profile $\sigma^{2}_{xy}$ is characterized by the bandwidth $W$ and $\alpha$-stable density with $\alpha\in (0,2]$. We investigate local eigenvalue statistics at the spectral edge as $W\to \infty$ and observe the critical dimension $d_c=3\alpha$ and the critical bandwidth $W_c=L^{(1-\frac{d}{3\alpha})_{+}}$, possibly with a $\log L$ correction when $d=\alpha$ or $2\alpha$. In the Hermitian case, we establish that (i) when $d<2\alpha$, GUE edge, interpolating, and Poisson statistics emerge in the supercritical ($W\gg W_c$), critical ($W\sim W_c$), and subcritical ($W\ll W_c$) regimes, respectively; (ii) when $d\ge 2\alpha$, as long as $W\ge L^{\frac{1}{3}+\epsilon}$ for a small constant $\epsilon>0$, GUE edge universality holds. In the symmetric case, we also establish similar but subtle phenomena. In both $d=1$ and $\alpha=2$, the subcritical and supercritical results have been proven by Sodin for the band model with a cutoff variance profile \cite{sodin2010spectral}. Our proof builds upon Sodin's program and new techniques of taming the singularity of Feynman diagrams and graph integrals through a connection to the $\phi^3$ model.

Related articles: Most relevant | Search more
arXiv:1604.08188 [math.PR] (Published 2016-04-27)
Local eigenvalue statistics for random matrices with general short range correlations
arXiv:1308.1057 [math.PR] (Published 2013-08-05, updated 2014-03-18)
Universality of local eigenvalue statistics in random matrices with external source
arXiv:1909.06350 [math.PR] (Published 2019-09-13)
Bulk Universality for non-Hermitian Random Matrices