The limiting spectral distribution of the product of the Wigner matrix and a nonnegative definite matrix (Q990879): Difference between revisions

The main problem considered in this paper is that of finding and characterizing the limiting spectral distribution of a certain class of large dimensional random matrices. By limiting spectral distribution, the authors are referring to the limiting distribution to which the empirical distribution of the eigenvalues of the matrices converges weakly as the dimension of the matrices goes to infinity. If \(B_n\) is an \(n \times n\) matrix having only real eigenvalues, denoted by \(\lambda_1(B_n), \ldots \lambda_n(B_n)\), then, the empirical spectral distribution of \(B_n\) is \[ F^{B_n}(x)=(1/n)\# \{1 \leq i \leq n : \lambda_i(B_n) \leq x \}, \] where \(\# \{ \cdot \}\) denotes the number of elements included in the set \(\{ \cdot \}\). The authors consider matrices of the form \(A_n=n^{-1/2}T_n^{1/2}W_nT_n^{1/2}\), where \(W_n\) is an \(n \times n\) Hermitian matrix whose entries are independent complex random variables satisfying a Lindeberg type condition, and \(T_n\) is an \(n \times n\) nonnegative definite matrix independent of \(W_n\). They assume that almost surely, as \(n \rightarrow \infty\), the empirical distribution of the eigenvalues of \(T_n\) converges weakly to a non-random probability distribution. The authors show, with the aid of the Stieltjes transforms, that almost surely, as \(n \rightarrow \infty\), the empirical distribution of the eigenvalues of \(A_n\) also converges weakly to a non-random probability distribution, a system of two equations determining the Stieltjes transform of the limiting distribution. Some analytic properties of this limiting spectral distribution are then derived by means of those equations. The authors also show that the limiting spectral distribution is continuously differentiable everywhere on the real line except only at the origin and they derive a necessary and sufficient condition for determining its support. Finally, the authors calculate the density function of the limiting spectral distribution for two cases of \(T_n\), when \(T_n\) is a sample covariance matrix and when \(T_n\) is the inverse of a sample covariance matrix.