[Paper Review] Mesoscopic eigenvalue statistics of Wigner matrices

Yukun He, Antti Knowles|arXiv (Cornell University)|Mar 4, 2016

Random Matrices and Applications16 references3 citations

TL;DR

This paper establishes the universality of Wigner-Dyson-Mehta (WDM) mesoscopic eigenvalue statistics for general Wigner matrices across all mesoscopic spectral scales $N^{-1} \ll \eta \ll 1$, proving convergence of linear eigenvalue statistics to a universal Gaussian process with $H^{1/2}$-Sobolev covariance structure. The proof combines self-consistent equations for resolvent moments with the local semicircle law and extends previous results beyond the $\eta \gg N^{-1/2}$ regime via refined cumulant expansions.

ABSTRACT

We prove that the linear statistics of the eigenvalues of a Wigner matrix converge to a universal Gaussian process on all mesoscopic spectral scales, i.e. scales larger than the typical eigenvalue spacing and smaller than the global extent of the spectrum.

Motivation & Objective

• To establish the universality of mesoscopic eigenvalue statistics for general Wigner matrices beyond the Gaussian or restricted spectral scale regimes.
• To resolve the gap in prior work that required $\eta \gg N^{-1/2}$ by extending the analysis to all mesoscopic scales $N^{-1} \ll \eta \ll 1$.
• To prove that linear statistics of eigenvalues converge to a universal Gaussian process with $H^{1/2}$-Sobolev covariance, independent of the specific distribution of Wigner matrix entries.
• To extend the validity of WDM mesoscopic statistics to general test functions with $1+o(1)$ derivatives and mild decay, under finite moments of order $4+o(1)$.

Proposed method

• Derives a new family of self-consistent equations for moments of linear statistics of the resolvent, adapted to general test functions via the Helffer-Sjöstrand representation.
• Applies the local semicircle law to control the resolvent and its moments down to the smallest mesoscopic scales.
• Uses a cumulant expansion (real and complex versions) to handle non-Gaussian entries, with precise error estimates on remainder terms.
• Implements a two-step analysis: first controlling resolvent traces, then extending to general linear statistics via functional calculus.
• Establishes convergence in distribution to a Gaussian process by showing tightness and identifying the limiting covariance structure.
• Adapts the method to the complex Hermitian case using complex cumulant expansion and adjusts the leading term by a factor of $1/2$ due to complex structure.

Experimental results

Research questions

• RQ1 Does the Wigner-Dyson-Mehta mesoscopic eigenvalue statistics universal for all Wigner matrices on all mesoscopic scales $N^{-1} \ll \eta \ll 1$?
• RQ2 Can the convergence of linear eigenvalue statistics to a universal Gaussian process be established for general test functions with $1+o(1)$ derivatives and mild decay?
• RQ3 What is the precise limiting covariance structure of the mesoscopic eigenvalue statistics, and is it scale-invariant as in the $H^{1/2}$-Sobolev norm?
• RQ4 How do the results differ in the complex Hermitian case, particularly in the leading-order term of the covariance?
• RQ5 Can the proof technique overcome the $\eta \gg N^{-1/2}$ restriction present in prior works using similar self-consistent equations?

Key findings

• The linear eigenvalue statistics of Wigner matrices converge in distribution to a universal Gaussian process on all mesoscopic scales $N^{-1} \ll \eta \ll 1$, regardless of the specific distribution of matrix entries.
• The limiting covariance is given by the $H^{1/2}$-Sobolev norm squared: $\mathbb{E}Z(f)^2 = \frac{1}{\pi}\int |\xi| |\hat{f}(\xi)|^2 d\xi$, confirming scale invariance.
• The result holds for all Wigner matrices with finite moments of order $4+o(1)$, extending beyond Gaussian or finite moment assumptions.
• The proof overcomes the $\eta \gg N^{-1/2}$ barrier by refining the analysis of self-consistent equations and using optimal bounds from the local semicircle law.
• For the complex Hermitian case, the leading term in the covariance is halved compared to the real case, consistent with the $1/2$ factor in the complex cumulant expansion.
• The convergence is established via tightness and moment matching, with remainder terms controlled using cumulant expansion estimates under $L^p$-integrability and decay conditions.