The empirical distribution of the eigenvalues of a Gram matrix with a given variance profile (Q858482)

An \(N \times n\) random matrix \(Y_n=(Y_{ij}^n)\) is considered. Its entries are given by \(Y_{ij}^n= \frac{\sigma(i/N, j/n)}{\sqrt{n}}X_{ij}^n\), where \(\sigma: [0,1] \times [0,1] \to [0, \infty]\) is a function whose square is continuous, called a variance profile and the random variables \(X_{ij}^n\) are real, centered, independent and identically distributed (i.i.d.) with finite \(4 + \epsilon \) moment. Also a real deterministic \(N \times n\) matrix \(\Lambda = (\Lambda_{ij}^n)\) whose off-diagonal entries are zero and a matrix \(\Sigma_n = Y_n + \Lambda_n\) are considered. Then the convergence of the empirical distribution of the eigenvalues of the Gram random matrix \(\Sigma_n \Sigma_n^T\) when \(n \to + \infty\) and \(N \to + \infty\) and \(N/n \to c\), \(0 < c < \infty\) is studied. It is proved that if \(\lim_{n\to\infty}N/n x\to c > 0\) and if there exists a probability measure \(H\) on \([0,1] \times {\mathbb R}\) with compact support \(\mathcal H\) such that \[ \frac{1}{\min(N,n)} \sum_{i=1}^{\min(N,n)} \delta_{(i/\min(N,n),(\Lambda_{ii}^n)^2)} x@>>n\to\infty>{\mathcal D} H(\text{d}x,\text{d}\lambda), \] where \({\mathcal D}\) stands for convergence in distribution, then the empirical distribution of the eigenvalues of the random matrix \(\Sigma_n \Sigma_n^T\) (resp. \(\Sigma_n^T \Sigma_n\)) converges in distribution to a deterministic probability distributions which are characterized in terms of their Stieltjes transform.