Consecutive level spacings in the chiral Gaussian unitary ensemble: From the hard and soft edge to the bulk

TL;DR

This paper analytically and numerically investigates the spacings between consecutive eigenvalues at the spectral edges of the chiral Gaussian unitary ensemble, revealing they are similar to bulk spacings and that the Wigner surmise is a highly accurate approximation.

Contribution

It provides new analytical results for eigenvalue spacings at the spectral edges in the chiral GUE, connecting edge and bulk statistics, with rapid convergence and validation through simulations and lattice QCD data.

Findings

Edge spacings are nearly indistinguishable from bulk spacings.

Wigner surmise approximates these spacings with less than 1% deviation.

Results are validated by Monte Carlo simulations and lattice QCD data.

Abstract

The local spectral statistics of random matrices forms distinct universality classes, strongly depending on the position in the spectrum. Surprisingly, the spacing between consecutive eigenvalues at the spectral edges has received little attention, where the density diverges or vanishes, respectively. This different behaviour is called hard or soft edge. We show that the spacings at the edges are almost indistinguishable from the spacing in the bulk of the spectrum. We present analytical results for consecutive spacings between the $k$th and $(k+1)$st smallest eigenvalues in the chiral Gaussian unitary ensemble, both for finite- and large-$n$. The result depends on the number of the generic zero modes $\nu$ and the number of flavours $N_{\rm f}$, which are given in terms of characteristic polynomials, as motivated by Quantum Chromodynamics (QCD). We find that the convergence in $n$ is very rapid. The same can be said separately about the limit $k\to\infty$ (limit to the bulk) and $\nu\to\infty$ (limit to the soft edge). Interestingly, the Wigner surmise is a very good approximation for all these cases and, apart from $k=1$, shows a deviation below one percent. These findings are corroborated with Monte-Carlo simulations. We finally compare for $k=1$ with data from QCD on the lattice, being in this symmetry class.