Central limit theorem for linear eigenvalue statistics of random matrices with independent entries (Q1035862)

The central limit theorem (CLT) is proved for two classes of random matrices, namely the Wigner matrices \(n^{-1/2}W\), where \(W\) are \(n\times n\) real symmetric random matrices with independent (modulo symmetry) entries and the sample covariance matrices \(n^{-1}X^TX\) where \(X\) are \(m\times n\) matrices with independent entries. The authors use a basic characteristic function and a generalized Fourier transform (in fact the \(-\pi/2\) rotated Laplace transform) instead of a collection of moments and of the Stieltjes transform employed in some previous papers. This enables them to prove the CLT for linear eigenvalue statistics with sufficiently regular (essentially \(C^5\)) test functions and assuming the existence of the fourth moments of entries satisfying a Lindeberg type condition. There are three levels of theorems. Firstly the CLT is proved for matrices with Gaussian entries (in the case of Gaussian orthogonal ensemble, where the entries of \(W\) are Gaussian and the probability law of \(W\) is orthogonal invariant and in the case of Wishart ensemble, where the entries of \(X\) are i.i.d. Gaussian). Then the CLT is obtained for matrices with zero excess of entries, by using the previous theorems and an ``interpolation trick''. Finally the same ideas provide the CLT in the general case of nonzero excess.