Convergence rate to the Tracy-Widom laws for the largest eigenvalue of Wigner matrices (Q2149067)

In the theory of random matrices, an already classical result is that the largest eigenvalue of a Wigner ensemble converges, after an appropriate rescaling, in distribution to the well-known probability distribution called the Tracy-Widom law. In fact, this limit is true when the entries of the Wigner matrix come from a quite general class of random variables, just specifying information on the moments and having independence of the entries up to symmetry; this is called a generalized Wigner matrix, thus the entries are not necessarily required to be Gaussian. This is the model consider here. In this paper, the authors study the problem of determining the speed of convergence of the largest eigenvalue, and they are able to improve convergence rates that were known before. Indeed, they find the exponent of convergence to be \(-1/3\), which improves the \(-2/3\) known before. To achieve this new rate, the authors carry out a careful analysis of a resolvent function (also called the Green function) coming from a matrix-valued Ornstein-Uhlenbeck process. Among the important tools that the authors need are the cumulants, the so-called Weingarten calculus (which has been proposed before as a natural tool in random matrix theory), and the determinantal structure of the eigenvalues. The analysis is quite elaborated to be able to obtain their result.