[Paper Review] Poisson statistics at the edge of Gaussian beta-ensembles at high temperature

Cambyse Pakzad|arXiv (Cornell University)|Apr 23, 2018

Random Matrices and Applications13 references3 citations

TL;DR

This paper establishes that, in the high-temperature limit where the inverse temperature $\beta \to 0$ as $n \to \infty$, the edge statistics of the Gaussian $\beta$-ensemble converge in distribution to a Poisson point process. Using asymptotic analysis of correlation functions and extreme value theory, the authors identify a specific decay regime of $\beta$ for which the largest eigenvalues behave like a Poisson process, extending classical extreme value results to general $\beta > 0$ and confirming Poisson statistics at the edge under high-temperature conditions.

ABSTRACT

We study the asymptotic edge statistics of the Gaussian $β$-ensemble, a collection of $n$ particles, as the inverse temperature $β$ tends to zero as $n$ tends to infinity. In a certain decay regime of $β$, the associated extreme point process is proved to converge in distribution to a Poisson point process as $n o +\infty$. We also extend a well known result on Poisson limit for Gaussian extremes by showing the existence of an edge regime that we did not find in the literature.

Motivation & Objective

• To investigate the asymptotic behavior of the largest eigenvalues in the Gaussian $\beta$-ensemble as the system size $n \to \infty$ and the inverse temperature $\beta \to 0$.
• To determine the regime of $\beta$ decay under which the extreme point process converges to a Poisson process at the edge of the spectrum.
• To extend classical extreme value results for Gaussian order statistics to the general $\beta$-ensemble framework, particularly in the high-temperature (low-correlation) limit.
• To establish a rigorous link between the local edge statistics of the $\beta$-ensemble and Poisson point processes under a specific scaling regime.

Proposed method

• Analyzes the correlation functions of the Gaussian $\beta$-ensemble using the tridiagonal matrix model introduced by Dumitriu and Edelman for general $\beta > 0$.
• Applies asymptotic analysis to the joint density and correlation functions under the condition $n\beta \ll 1$, corresponding to high temperature.
• Uses extreme value theory and Mill's ratio to approximate tail probabilities of i.i.d. standard normal variables, which model the eigenvalue behavior in the $\beta \to 0$ limit.
• Derives a normalization sequence $a_n = \delta_n \sqrt{2\log n}$, $b_n = \sqrt{2\log n} - \frac{1}{2}\frac{\log\log n + 2\log\delta_n + \log(4\pi)}{\sqrt{2\log n}}$ to center and scale the extreme eigenvalues.
• Employs bounds on partition function ratios and exponential moment estimates to control the correlation function decay and prove convergence to Poisson statistics.
• Establishes convergence in distribution of the rescaled point process $\sum_{i=1}^n \delta_{a_n(\lambda_i - b_n)}$ to a homogeneous Poisson process on $\mathbb{R}$ with intensity 1.

Experimental results

Research questions

• RQ1 Under what decay regime of $\beta$ as $n \to \infty$ does the edge point process of the Gaussian $\beta$-ensemble converge to a Poisson process?
• RQ2 How does the high-temperature limit ($\beta \to 0$) affect the local statistics of the largest eigenvalues in the $\beta$-ensemble?
• RQ3 Can the classical extreme value limit for i.i.d. Gaussians be extended to the correlated $\beta$-ensemble model in the $\beta \to 0$ regime?
• RQ4 What is the precise scaling of the edge eigenvalues that leads to Poisson statistics in the $\beta$-ensemble at high temperature?

Key findings

• The rescaled point process $\sum_{i=1}^n \delta_{a_n(\lambda_i - b_n)}$ converges in distribution to a homogeneous Poisson point process on $\mathbb{R}$ with intensity 1, under the condition $\log \delta_n \ll \log n$.
• The convergence holds when $\beta$ decays as $n\beta \ll 1$, which corresponds to the high-temperature regime where eigenvalue repulsion vanishes.
• The normalization sequences are $a_n = \delta_n \sqrt{2\log n}$ and $b_n = \sqrt{2\log n} - \frac{1}{2}\frac{\log\log n + 2\log\delta_n + \log(4\pi)}{\sqrt{2\log n}}$, ensuring proper centering and scaling of the extreme eigenvalues.
• The proof relies on controlling the correlation functions via bounds on partition function ratios and exponential moment estimates, showing that correlations decay sufficiently fast to yield Poisson statistics.
• The result extends the classical Poisson limit for i.i.d. Gaussian extremes to the correlated $\beta$-ensemble model, confirming universality of Poisson edge statistics in the high-temperature limit.
• The paper identifies a previously unreported edge regime in the literature where Poisson statistics emerge at the edge of the spectrum under high-temperature conditions.