Asymptotic expansions for the Gaussian Unitary Ensemble

Abstract

Let g:{\mathbb R} --> {\mathbb C} be a C^{\infty}-function with all derivatives bounded and let tr_n denote the normalized trace on the n x n matrices. In the paper [EM] Ercolani and McLaughlin established asymptotic expansions of the mean value E{tr_n(g(X_n))} for a rather general class of random matrices X_n,including the Gaussian Unitary Ensemble (GUE). Using an analytical approach, we provide in the present paper an alternative proof of this asymptotic expansion in the GUE case. Specifically we derive for a GUE random matrix X_n that E{tr_n(g(X_n))}= \frac{1}{2\pi}\int_{-2}^2 g(x)\sqrt{4-x^2} dx +\sum_{j=1}^k\frac{\alpha_j(g)}{n^{2j}}+ O(n^{-2k-2}), where k is an arbitrary positive integer. Considered as mappings of g, we determine the coefficients \alpha_j(g), j\in{\mathbb N}, as distributions (in the sense of L. Schwarts). We derive a similar asymptotic expansion for the covariance Cov{Tr_n[f(X_n)],Tr_n[g(X_n)]}, where f is a function of the same kind as g, and Tr_n=n tr_n. Special focus is drawn to the case where g(x)=1/(z-x) and f(x)=1/(w-x) for non-real complex numbers z and w. In this case the mean and covariance considered above correspond to, respectively, the one- and two-dimensional Cauchy (or Stieltjes) transform of the GUE(n,1/n).