On the Limit of the Tridiagonal Model for β-Dyson Brownian Motion

Alan Edelman, Sungwoo Jeong, Ron Nissim

Abstract

In previous work, a description of the result of applying the Householder tridiagonalization algorithm to a GβE random matrix is provided by Edelman and Dumitriu. The resulting tridiagonal ensemble makes sense for all β>0, and has spectrum given by the β-ensemble for all β>0. Moreover, the tridiagonal model has useful stochastic operator limits which was introduced and analyzed in subsequent studies. In this work, we analogously study the result of applying the Householder tridiagonalization algorithm to a GβE process which has eigenvalues governed by β-Dyson Brownian motion. We propose an explicit limit of the upper left $k \times k$ minor of the $n \times n$ tridiagonal process as $n \to \infty$ and $k$ remains fixed. We prove the result for β=1, and also provide numerical evidence for β=1,2,4. This leads us to conjecture the form of a dynamical β-stochastic Airy operator with smallest $k$ eigenvalues evolving according to the $n \to \infty$ limit of the largest, centered and re-scaled, $k$ eigenvalues of β-Dyson Brownian motion.

Introduction

This research explores the intersection of numerical linear algebra and stochastic calculus by extending the tridiagonal model for β-ensembles into the time-domain. The study analyzes how these models behave when the underlying matrices evolve as a stochastic process. Specifically, the authors characterize the large-scale limit of a tridiagonalized Gaussian random matrix process, proving that as the matrix size $n \to \infty$, the entries of the upper-left corner converge to a system of independent, classical Ornstein–Uhlenbeck (OU) processes.

Problem Statement

In 2002, Edelman and Dumitriu demonstrated that the eigenvalues of classical Gaussian ensembles (GOE, GUE, GSE) could be represented by a tridiagonal matrix with independent random variable entries. This model is valid for any β > 0. However, eigenvalues in dynamic systems often evolve over time, a phenomenon described by Dyson Brownian Motion (DBM). The central question addressed here is: if a matrix-valued process is tridiagonalized at every time step, what is the behavior of the resulting tridiagonal process as the matrix size becomes infinite? The goal is to find a "dynamical" version of the tridiagonal model that could lead to a time-evolving Stochastic Airy Operator.

Methods

The researchers apply the Householder tridiagonalization algorithm to a $G\beta E$ process. The evolution of matrix entries is tracked using Stochastic Differential Equations (SDEs). A key finding is the connection between the limit of tridiagonal entries and Chebyshev polynomials of the second kind, which are orthogonal polynomials associated with the Wigner semicircle law. High-dimensional concentration of measure is used to show that inner products of matrix-valued processes concentrate around their expectations as $n \to \infty$.

Schematic of the Householder tridiagonalization applied to a matrix-valued process, showing the convergence of the upper-left minor as $n \to \infty$.

Key Insights

  • Emergent Independence: Despite being coupled at any finite $n$, the entries of the tridiagonalized matrix become jointly independent in the $n \to \infty$ limit.
  • Scaling Laws: Diagonal entries $a_j$ converge directly to OU processes, while off-diagonal entries $b_j$ require centering and scaling (fluctuations of $b_j^2$ around $n$) to reach a non-trivial limit.
  • The Chebyshev Link: Tridiagonalization effectively performs a Gram-Schmidt-like procedure on matrix powers. Given the semicircle law spectral measure, entries converge to the recurrence coefficients for Chebyshev polynomials.
  • Dynamical Operator Challenges: While a "dynamical β-stochastic Airy operator" was sought, perturbative calculations suggest the true dynamical operator is more complex than this model's covariance initially indicated.

Results

The primary theoretical result (Theorem 3.1) establishes that for β = 1, the vector of the first $k$ diagonal and scaled off-diagonal entries converges weakly to a vector of independent OU processes. Specifically, the $j$-th diagonal entry $a_j(t)$ converges to $\sqrt{2}OU(2j-1)$, and the scaled off-diagonal $b_j(t)$ converges such that $\frac{1}{\sqrt{n}}(b_j(t)^2 - n)$ behaves like $\sqrt{2}OU(2j)$. These theoretical predictions were validated through large-scale numerical simulations of matrices up to $2000 \times 2000$ for β = 1, 2, 4.

Conclusion and Future Work

While the model accurately describes the "bulk" of the matrix, it does not perfectly capture the "edge" dynamics (the Airy process) governing the largest eigenvalues. Identifying the correct dynamical stochastic Airy operator remains an open challenge. Additionally, while numerical evidence is robust for various β values, the formal proof currently focuses on the β=1 case, leaving the rigorous extension to all β for future study.

Summary updated

Abstract

In previous work, a description of the result of applying the Householder tridiagonalization algorithm to a G$\beta$E random matrix is provided by Edelman and Dumitriu. The resulting tridiagonal ensemble makes sense for all $\beta>0$, and has spectrum given by the $\beta$-ensemble for all $\beta>0$. Moreover, the tridiagonal model has useful stochastic operator limits which was introduced and analyzed in subsequent studies. In this work, we analogously study the result of applying the Householder tridiagonalization algorithm to a G$\beta$E process which has eigenvalues governed by $\beta$-Dyson Brownian motion. We propose an explicit limit of the upper left $k \times k$ minor of the $n \times n$ tridiagonal process as $n \to \infty$ and $k$ remains fixed. We prove the result for $\beta=1$, and also provide numerical evidence for $\beta=1,2,4$. This leads us to conjecture the form of a dynamical $\beta$-stochastic Airy operator with smallest $k$ eigenvalues evolving according to the $n \to \infty$ limit of the largest, centered and re-scaled, $k$ eigenvalues of $\beta$-Dyson Brownian motion.
BibTeX
@misc{edelman2024limit,
  author = {Alan Edelman and Sungwoo Jeong and Ron Nissim},
  title = {{On the Limit of the Tridiagonal Model for β-Dyson Brownian Motion}},
  year = {2024},
  eprint = {2411.01633},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2411.01633},
}

Loading related content...

Figure