Table of Contents
Fetching ...

Weakly universal dynamical correlations between eigenvalues of large random matrices

Kirone Mallick, Gabriel Téllez, Frédéric van Wijland

TL;DR

This work analyzes how universal features of eigenvalue correlations in large random matrices persist under Dyson Brownian motion. By formulating a fluctuating hydrodynamics for the resolvent and applying Macroscopic Fluctuation Theory, it derives a central dynamical relation for the resolvent correlations: $\Gamma(z,z';t) = -\frac{1}{\beta N^2}\frac{Q(\zeta)}{Q(z)}\frac{a^2- \zeta z'+\sqrt{\zeta^2-a^2}\sqrt{z'^2-a^2}}{\sqrt{z^2-a^2}\sqrt{z'^2-a^2}(\zeta-z')^2}$ with $\frac{d\zeta}{dt}=Q(\zeta)\sqrt{\zeta^2-a^2}$ and $\zeta(0)=z$, showing that dynamical correlations retain a large degree of universality through a characteristic trajectory. The results are worked out in detail for harmonic and quartic potentials, revealing explicit time-dependent forms and long-time decays, and are extended to higher-degree potentials where multiple relaxation channels emerge. The framework paves the way for dynamical large-deviation analyses and extensions to other ensembles, connecting stochastic eigenvalue dynamics to fluctuating hydrodynamics and dynamic free-probability constructs via the resolvent.

Abstract

It was shown roughly thirty years ago that the density correlations of eigenvalues of large random matrices display a universal form, independent of most of the details of the distribution of the random matrix itself. We show that when the matrix elements evolve according to a Dyson Brownian motion, dynamical correlations retain a large degree of the universality found at equal times when expressed in terms of the characteristics of some partial differential equation in the complex plane.

Weakly universal dynamical correlations between eigenvalues of large random matrices

TL;DR

This work analyzes how universal features of eigenvalue correlations in large random matrices persist under Dyson Brownian motion. By formulating a fluctuating hydrodynamics for the resolvent and applying Macroscopic Fluctuation Theory, it derives a central dynamical relation for the resolvent correlations: with and , showing that dynamical correlations retain a large degree of universality through a characteristic trajectory. The results are worked out in detail for harmonic and quartic potentials, revealing explicit time-dependent forms and long-time decays, and are extended to higher-degree potentials where multiple relaxation channels emerge. The framework paves the way for dynamical large-deviation analyses and extensions to other ensembles, connecting stochastic eigenvalue dynamics to fluctuating hydrodynamics and dynamic free-probability constructs via the resolvent.

Abstract

It was shown roughly thirty years ago that the density correlations of eigenvalues of large random matrices display a universal form, independent of most of the details of the distribution of the random matrix itself. We show that when the matrix elements evolve according to a Dyson Brownian motion, dynamical correlations retain a large degree of the universality found at equal times when expressed in terms of the characteristics of some partial differential equation in the complex plane.

Paper Structure

This paper contains 16 sections, 89 equations, 2 figures.

Figures (2)

  • Figure 1: Trajectories of the characteristics starting from right below the branch cut $[-a,a]$ for a quartic potential $V(\lambda)=\frac{\lambda^4}{4}$ ($Q(\lambda)=\lambda^2+a^2/2$ and $a^2=2\sqrt{2/3}$). All flow to the $z_1=-ia/\sqrt{2}$ root of $Q$ lying in the lower half plane (the initial values are $-0.7$ (green), $0.5$ (blue) and $1$ (orange)).
  • Figure 2: Trajectories of the characteristics starting from right below the branch cut $[-a,a]$ for a potential $V(\lambda)=\frac{\lambda^8}{8}$ ($Q(\lambda)=\lambda^6+\frac{a^2}{2}\lambda^4+\frac{3}{8}a^4 \lambda^2+\frac{5}{16}a^6$ and $a^8=64/35$). The roots are $z_1\simeq-0.9 i$, $z_2=0.7-0.6 i$ (and $z_3=-z_2^*$) and $a\simeq 1.1$. The characteristics are initialized with positive reals parts (due to the symmetries of the potential) at the following (just below the real axis): $z=0.1$ (blue), $z=0.5$ (orange), $z=0.7$ (green), $z=0.9$ (red) and $z=1.05$ (violet). The basins of attraction of each $z_1$ and $z_2$ are non trivial.