Cusp universality for correlated random matrices (Q6933512)

This article presents a result of universality at cusp singularities, i.e.,\ at points where the limiting density exhibit a cubic root behavior. Together with the behavior at the bulk and the behavior at the edge, such behavior is one of the three classical spectral regimes for which universality was conjectured.\N\NCusp singularities in the spectral density were exhibited by \textit{E. Brézin} and \textit{S. Hikami} for matrices \(A + W\) with \(A\) a well-tuned deterministic matrix and \(W\) a Gaussian random matrix [``Level spacing of random matrices in an external source'', Phys. Rev. E 58, No. 6, 7176 (1998)]. The present article of Erdős et al. shows that many random matrix models display the same behavior at cusp singularities, under assumptions that allow for correlations between the entries. In particular, this improves over earlier work that concerned only Wigner-type matrices [\textit{L. Erdős} et al., Commun. Math. Phys. 378, No. 2, 1203--1278 (2020; Zbl 1475.60013); \textit{G. Cipolloni} et al., Pure Appl. Anal. 1, No. 4, 615--707 (2019; Zbl 1476.60012)].\N\NThe result rests on a precise local law for the matrix \(H + W\), of independent interest, and on the zigzag strategy already used to prove universality in several models. See [\textit{L. Erdős} and \textit{H.-T. Yau}, A dynamical approach to random matrix theory. Providence, RI: American Mathematical Society (AMS); New York, NY: Courant Institute of Mathematical Sciences (2017; Zbl 1379.15003)] for a review of the main ideas. This method consists in applying two steps repeatedly: the \emph{zig} step which allows to improve a local law at the cost of adding a small Gaussian part to the random matrix, and a \emph{zag} part which consists in removing this Gaussian part. The two steps are based on two different matrix Ornstein-Uhlenbeck-type diffusions that can be shown to coincide at a particular time, for a good choice of initial conditions.