Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices

DOI10.1214/009117905000000233MaRDI QIDQ2571691zbMATH OpenOpenAlexWikidataFDO

Authors Jinho Baik, Gerard Ben Arous, S. Péché

Publication date 14 November 2005

Published in The Annals of Probability (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/math/0403022

Abstract: We compute the limiting distributions of the largest eigenvalue of a complex Gaussian sample covariance matrix when both the number of samples and the number of variables in each sample become large. When all but finitely many, say

r

, eigenvalues of the covariance matrix are the same, the dependence of the limiting distribution of the largest eigenvalue of the sample covariance matrix on those distinguished

r

eigenvalues of the covariance matrix is completely characterized in terms of an infinite sequence of new distribution functions that generalize the Tracy-Widom distributions of the random matrix theory. Especially a phase transition phenomena is observed. Our results also apply to a last passage percolation model and a queuing model.

Review text

The largest eigenvalue distribution is investigated for \(N\)-dimensional random Gram matrices (``covariance matrices) \(S = M^{-1}(X - \bar X)(X -\bar X)^T\), where \(X\) are rectangular \(N \times M\) sample matrices whose columns are observation vectors distributed as \(N(0,\Sigma)\), and \(\bar X\) are sample mean vectors. It is assumed that \(N, M \rightarrow\infty\) and \(M/N \rightarrow\gamma^2 > 1\). In this asymptotics, when \(\Sigma\) are unity matrices, \textit{V. A. Marchenko} and \textit{L. A. Pastur} [Math. USSR, Sb. 1, 457--483 (1967; Zbl 0162.22501)] derived well-known limiting distribution density of eigenvalues of \(S\); its upper bound was found to be \((1 + 1/\gamma)^2\). For this case also, the largest eigenvalue \(\lambda_1\) of the limiting distribution was found by \textit{P. J. Forrester} [Nucl. Phys. B 402, No.~3, 709--728 (1993; Zbl 1043.82538)] \[ \mathbb P(aM^{2/3}(\lambda_1- (1 + 1/\gamma)^2< x) \rightarrow F(x),\tag{1} \] where \(a\) depends only on \(\gamma\). The authors find the limiting distribution of the largest eigenvalue \(\lambda_1\) for complex matrices \(S\) when there is a finite number of eigenvalues of \(\Sigma\) different from 1. Let \(r \leq N\) and the eigenvalues \(l_1 \geq l_2 \geq\dots l_N\) of \(\Sigma\) be such that \(l_{r+1} = l_{r+2} = \dots = l_N.\) Theorem 1 states that (i) If \(k\leq r\), \(l_1=l_2=\dots=l_k=1+1/\gamma\) and all \(\lambda_j\), \(j=k+1,k+2\dots,r\), are located on a compact within \((0,1 + 1/\gamma)\), then the largest eigenvalue \(\lambda_1\) of \(S\) has the same limiting distribution (1); (ii) If for some \(k\leq r\), the eigenvalues \(l_1=l_2=\dots=l_k\) are located on a compact within \((1 + 1/\gamma,\infty)\) and all eigenvalues \(l_j\), \(j=k+1,k+2\dots,r\), are on a compact within \((0,l_1)\), then \[ \mathbb P(aM^{1/2}(\lambda_1-c)\leq x)\to G_k(x), \] where \(a > 0\) and \(c < l_1\) depend on \(\gamma\) and \(l_1\), and \(G_k(x)\) is the distribution function for the largest eigenvalue of \(k\times k\) random matrix when \(\Sigma\) is unity. Theorem 2 establishes the limiting distribution (1) in an intermediate case. Two applications are considered: finding the last passage time in a percolation problem and a queueing model of a sequential service of \(M\) customers in the line of \(N\) tellers under a fixed order of serving.

Recommendations

Cites work

Cited in

(only showing first 100 items - show all)

This page was built for publication: Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2571691)