Stein's Method and Random Character Ratios

Abstract

Stein's method is used to prove limit theorems for random character ratios. Tools are developed for four types of structures: finite groups, Gelfand pairs, twisted Gelfand pairs, and association schemes. As one example an error term is obtained for a central limit theorem of Kerov on the spectrum of the Cayley graph of the symmetric group generated by i-cycles, or equivalently for the character ratio of a Plancherel distributed representation on an i-cycle. Other main examples include an error term for a central limit theorem of Ivanov on character ratios of random projective representations of the symmetric group, and a new central limit theorem for the spectrum of certain graphs whose vertices are the set of perfect matchings on 2n symbols. The error terms in the resulting limit theorems are typically O(n^{-1/4}) or better. The results are obtained with remarkably little information: a character formula for a single representation close to the trivial representation and estimates on two step transition probabilities of a random walk. Although the limit theorems stated in this paper are all for the case of normal approximation, many of the tools developed are quite general. Indeed, both the construction of an exchangeable pair used for Stein's method and lemmas computing certain moments are useful for arbitrary distributional approximation.