Small eigenvalues of large Hankel matrices

Abstract

In this paper we investigate the smallest eigenvalue, denoted as $\la_N,$ of a $(N+1)\times (N+1)$ Hankel or moments matrix, associated with the weight, $w(x)=\exp(-x^{\bt}),x>0,\bt>0$, in the large $N$ limit. Using a previous result, the asymptotics for the polynomials, $P_n(z),z\notin[0,\infty)$, orthonormal with respect to $w,$ which are required in the determination of $\la_N$ are found. Adopting an argument of Szeg\"{o} the asymptotic behaviour of $\la_N$, for $\bt>1/2$ where the related moment problem is determinate, is derived. This generalises the result given by Szeg\"{o} for $\bt=1$. It is shown that for $\bt>1/2$ the smallest eigenvalue of the infinite Hankel matrix is zero, while for $0<\bt<1/2$ it is greater then a positive constant. This shows a phase transition in the corresponding Hermitian random matrix model as the parameter $\bt$ varies with $\bt=1/2$ identified as the critical point. The smallest eigenvalue at this point is conjectured.