On the variance of linear statistics of Hermitian random matrices

Abstract

Linear statistics, a random variable build out of the sum of the evaluation of functions at the eigenvalues of a N times N random matrix,sum[j=1 to N]f(xj) or tr f(M), is an ubiquitous statistical characteristics in random matrix theory. Hermitian random matrix ensembles, under the eigenvalue-eigenvector decomposition give rise to the joint probability density functions of N random variables. We show that if f(.) is a polynomial of degree K, then the variance of trf(M), is of the form,sum[n=1 to K] n(d[n])square, and d[n] is related to the expansion coefficients c[n] of the polynomial f(x) =sum[n=0 to K] c[n] b Pn(x), where Pn(x) are polynomials of degree n, orthogonal with respect to the weights 1/[(b-x)(x-a)]^(1/2), [(b -x)(x -a)]^(1/2), [(b-x)(x-a)]^(1/2)/x; (0 < a < x < b), [(b-x)(x-a)]^(1/2)/[x(1-x)] ; (0 < a < x < b < 1), respectively.