[Paper Review] Bulk Universality for Wigner Matrices

László Erdős, Sandrine Péché+3 more|ArXiv.org|May 26, 2009

Random Matrices and Applications15 references22 citations

TL;DR

This paper establishes bulk universality for large Wigner matrices with general distributions, proving that local eigenvalue statistics converge to the Dyson sine kernel under minimal moment and decay conditions. Using an approximate time reversal of Dyson Brownian motion and refined local semicircle law estimates, the authors extend universality beyond Gaussian ensembles to non-Gaussian, heavy-tailed distributions with exponential decay, resolving a long-standing open problem in random matrix theory.

ABSTRACT

We consider $N imes N$ Hermitian Wigner random matrices $H$ where the probability density for each matrix element is given by the density $ν(x)= e^{- U(x)}$. We prove that the eigenvalue statistics in the bulk is given by Dyson sine kernel provided that $U \in C^6(\RR)$ with at most polynomially growing derivatives and $ν(x) \le C e^{- C |x|}$ for $x$ large. The proof is based upon an approximate time reversal of the Dyson Brownian motion combined with the convergence of the eigenvalue density to the Wigner semicircle law on short scales.

Motivation & Objective

• To establish bulk universality of eigenvalue statistics for general Wigner matrices beyond Gaussian ensembles.
• To remove the need for Gaussian or sub-Gaussian tail assumptions in universality theorems.
• To prove that local eigenvalue correlations converge to the Dyson sine kernel under minimal regularity and decay conditions on the matrix entry distribution.
• To extend the validity of the local semicircle law to distributions with exponential decay, not just Gaussian tails.

Proposed method

• Uses an approximate time reversal of Dyson Brownian motion to compare eigenvalue statistics of a Wigner matrix with those of a matrix evolved under the Ornstein-Uhlenbeck process.
• Applies the local semicircle law on short scales, proven via a modified version of the approach in [11], adapted to non-Gaussian distributions.
• Implements a cutoff and rescaling technique to approximate heavy-tailed distributions by compactly supported ones with controlled variance.
• Establishes a total variation bound between the original and truncated measures, showing that the error is smaller than any negative power of $N$.
• Uses the fact that the eigenvalue density converges to the Wigner semicircle law on scales $\eta \geq N^{-1+\varepsilon}$, even under exponential decay assumptions.
• Relies on a refined large deviation estimate for the Stieltjes transform, with bounds deteriorating only by a logarithmic factor in the exponent.

Experimental results

Research questions

• RQ1 Does bulk universality hold for Wigner matrices with non-Gaussian, heavy-tailed entries under minimal regularity and decay conditions?
• RQ2 Can the Dyson sine kernel universality be established without assuming Gaussian or sub-Gaussian tails for matrix entries?
• RQ3 To what extent can the local semicircle law be extended to distributions with exponential decay instead of Gaussian tails?
• RQ4 Can the time scale for convergence to local equilibrium be reduced to $N^{-1+\lambda}$ for $\lambda > 0$ while preserving universality?
• RQ5 Is the total variation distance between the original and truncated measures negligible for the purpose of eigenvalue statistics?

Key findings

• Bulk universality holds for $N \times N$ Hermitian Wigner matrices with i.i.d. entries whose distribution $\nu$ satisfies $U \in C^6(\mathbb{R})$ with polynomially growing derivatives and $\nu(x) \leq C e^{-C|x|}$ for large $x$.
• The local eigenvalue statistics in the bulk converge to the Dyson sine kernel, even when the matrix entries have only exponential decay, not Gaussian tails.
• The local semicircle law is established on scales $\eta \geq N^{-1+\varepsilon}$ for any $\varepsilon > 0$, under the exponential decay assumption.
• The total variation distance between the original measure and its truncated version is smaller than any negative power of $N$, allowing passage from truncated to original measures.
• The key estimate for the Stieltjes transform deviation is bounded by $\exp\big(-c\delta\sqrt{N\eta}/\ell\big)$, with $\ell = (\log N)^2$, showing robustness under truncation.
• The result confirms that universality is determined by local relaxation to equilibrium, not global invariance, and extends the range of applicability of Dyson Brownian motion techniques.