Cumulants in rectangular finite free probability and beta-deformed singular values

TL;DR

This paper introduces rectangular cumulants in finite free probability, establishing their properties, convergence to q-rectangular free cumulants, and applications to polynomial root distributions.

Contribution

It defines rectangular cumulants, proves moment-cumulant formulas, and shows their convergence to q-rectangular free cumulants, extending free probability theory.

Findings

Rectangular cumulants linearize the rectangular convolution.

They converge to q-rectangular free cumulants as d→∞.

Application to polynomial root distributions and asymptotic free convolution with Gaussian.

Abstract

Motivated by the $(q,\gamma)$-cumulants, introduced by Xu [arXiv:2303.13812] to study $\beta$-deformed singular values of random matrices, we define the $(n,d)$-rectangular cumulants for polynomials of degree $d$ and prove several moment-cumulant formulas by elementary algebraic manipulations; the proof naturally leads to quantum analogues of the formulas. We further show that the $(n,d)$-rectangular cumulants linearize the $(n,d)$-rectangular convolution from Finite Free Probability and that they converge to the $q$-rectangular free cumulants from Free Probability in the regime where $d\to\infty$, $1+n/d\to q\in[1,\infty)$. As an application, we employ our formulas to study limits of symmetric empirical root distributions of sequences of polynomials with nonnegative roots. One of our results is akin to a theorem of Kabluchko [arXiv:2203.05533] and shows that applying the operator $\exp(-\frac{s^2}{n}x^{-n}D_xx^{n+1}D_x)$, where $s>0$, asymptotically amounts to taking the rectangular free convolution with the rectangular Gaussian distribution of variance $qs^2/(q-1)$.