On the asymptotic distribution of the largest eigenvalue of unbalanced complex Jacobi ensembles

Abstract

Let $\mathbf{A}\sim \mathbb{C}{\mathcal{W}}_p(I,m)$ and $\mathbf{B}\sim \mathbb{C}{\mathcal{W}}_p(I,n)$ be two independent complex Wishart random matrices. In this work, we investigate the asymptotic distribution of the largest eigenvalue ${\theta }_{\max }$ of the random matrix $\mathbf{J}=(\mathbf{A}+\mathbf{B}{)}^{-1}\mathbf{B}$. The high-dimensional regime is potentially unbalanced in the sense that, as $p\to \infty$, $m$ may grow faster than $p$, whereas $n$ grows proportionally to $p$. The asymptotic behavior of ${\theta }_{\max }$ is characterized by means of a Riemann-Hilbert (RH) problem for Jacobi polynomials with modified Jacobi weights. We establish the type 2 Tracy--Widom asymptotic distribution of ${\theta }_{\max }$. This provides an alternative, RH-based proof of the corresponding result of Johnstone [Ann. Statist. 36 (2008), 2638-2716] in the balanced regime, and establishes a conjecture of Jiang [Probab. Theory Relat. Fields 144 (2009), 221-246] in the unbalanced regime. In addition, we obtain an explicit correction of order $p^{-2/3}$ to the Tracy--Widom approximation, with a remainder of order $\mathcal{O}(p^{-1})$, and show exponential decay, as a function of $p$, of the upper tail of the distribution of ${\theta }_{\max }$.

arXiv comment64 pages, 8 figures