The probability of almost all eigenvalues being real for the elliptic real Ginibre ensemble

Gernot Akemann, Sung-Soo Byun, Yong-Woo Lee · 2025

Read The probability of almost all eigenvalues being real for the elliptic real Ginibre ensemble by Gernot Akemann, Sung-Soo Byun, Yong-Woo Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-18310.txt.

Introduction

Gernot Akemann, Sung-Soo Byun, and Yong-Woo Lee present a 2025 mathematics paper on random matrices from the elliptic Ginibre ensemble. These matrices model systems whose values, called eigenvalues, can be real or come in complex pairs. The authors ask how likely it is that almost all eigenvalues turn out real. They study the chance as the matrix grows, allowing a fixed small number of exceptions.

They derive full formulas for this probability in two regimes: one where the matrix stays strongly non-symmetric, and one where it nears symmetry as it grows. In the special case of a single allowed exception pair, they obtain the expansion to every order. The proofs use different tools in each regime, from energy methods to special polynomial techniques. The work sharpens knowledge of real eigenvalues in random matrix theory.

Catalog record

CREATORGernot Akemann, Sung-Soo Byun, Yong-Woo Lee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE94 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-18310

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The probability of almost all eigenvalues being real for the elliptic real Ginibre ensemble Thanks: Corresponding author: Yong-Woo Lee ( [email protected] ) Gernot Akemann, Sung-Soo Byun, Yong-Woo Lee Abstract . We investigate real eigenvalues of real elliptic Ginibre matrices of size n n , indexed by the parameter of asymmetry τ ∈ [ 0 , 1 ] \tau\in[0,1] . In both the strongly and weakly non-Hermitian regimes, where τ ∈ [ 0 , 1 ) \tau\in[0,1) is fixed or 1 − τ = O ⁡ ( 1 / n ) 1-\tau=O(1/n) , respectively, we derive the asymptotic expansion of the probability p n , n − 2 ​ l p_{n,n-2l} that all but a finite number 2 ​ l 2l of eigenvalues are real. In particular, we show that the expansion is of the form log ⁡ p n , n − 2 ​ l = { a 1 ​ n 2 + a 2 ​ n + a 3 ​ log ⁡ n + O ⁡ ( 1 ) at strong non-Hermiticity , b 1 ​ n + b 2 ​ log ⁡ n + b 3 + o ⁡ ( 1 ) at weak non-Hermiticity , \displaystyle\log p_{n,n-2l}=\begin{cases}a_{1}n^{2}+a_{2}n+a_{3}\log n+O(1)&\textup{at strong non-Hermiticity},\\ b_{1}n+b_{2}\log n+b_{3}+o(1)&\textup{at weak non-Hermiticity},\end{cases} and we determine all coefficients explicitly. Furthermore, in the special case where l = 1 l=1 , we derive the full-order expansions. For the proofs, we employ distinct methods for the strongly and weakly non-Hermitian regimes. In the former case, we utilise potential-theoretic techniques to analyse the free energy of elliptic Ginibre matrices conditioned to have n − 2 ​ l n-2l real eigenvalues, together with the strong Szegő limit theorems. In the latter case, we utilise the skew-orthogonal polynomial formalism and the asymptotic behaviour of the Hermite polynomials. 1. Introduction and Main results In the study of point processes, one of the most natural observables of interest is their counting or number statistics, which addresses fundamental questions about how many particles can be observed within a given domain. From a probabilistic perspective, the natural questions that arise can be explored step by step as follows: • What is the typical number of particles? This involves determining both the expectation and variance of the observable, which are central to the law of large numbers. • What are the fluctuations around the typical number of particles? This question provides a more refined statistical description and relates to the central limit theorem. • What is the probability of rare events? Here, one typically investigates the probability of observing an extremely small or large number of particles compared to the typical number. This falls within the scope of large deviation theory. These types of questions have been extensively studied in the literature and serve as central topics in the field. In this work, we address a question in this direction for a point process arising from non-Hermitian random matrix theory. Specifically, we focus on asymmetric random matrices with real entries, with the real Ginibre matrix (denoted by GinOE) serving as a prototypical example [ 11 ] . For asymmetric random matrices, an interesting domain of study in the number statistics is the real axis. In other words, we examine the number statistics of real eigenvalues. The presence of real eigenvalues with non-zero probability is particularly characteristic of the orthogonal symmetry class, distinguishing it from the unitary and symplectic symmetry classes (see e.g. [ 47 ] ). This question has been addressed in various models including the GinOE [ 21 , 32 ] and its variants including the elliptic GinOE [ 33 , 12 ] , truncated orthogonal ensemble [ 28 , 29 , 46 ] , products of GinOE matrices [ 1 , 54 , 24 , 27 ] , spherical Ginibre matrices [ 31 , 21 ] , asymmetric Wishart matrices [ 15 ] , as well as matrices with non-Gaussian entries [ 56 ] . We also refer the reader to [ 16 , 5 , 2 , 39 , 19 , 40 , 3 ] and the references therein for the counting statistics of various two-dimensional point processes. In addition, we mention that the eigenvalue statistics of the elliptic GinOE finds applications in various areas such as ecosystem stability [ 6 , 35 ] and complexity of random energy landscapes [ 45 , 34 ] . We first introduce our model of study, the elliptic GinOE. The GinOE is an n\times n matrix G with independent and identically distributed standard normal entries. Its elliptic extension is then defined as (1.1) X:=\frac{\sqrt{1+\tau}}{2}(G+G^{T})+\frac{\sqrt{1-\tau}}{2}(G-G^{T}), where \tau\in[0,1) is the asymmetry parameter. This parameter governs the extent to which the matrix deviates from being symmetric, allowing the elliptic GinOE to interpolate between the GinOE ( \tau=0 ) and the Gaussian Orthogonal Ensemble (GOE) in the limit as \tau\to 1 . We investigate the asymptotic properties of the real eigenvalues of the elliptic GinOE as n\to\infty . When examining the asymptotic behaviour of the elliptic GinOE, two distinct regimes of interest emerge: • Strong Non-Hermiticity : This is the case where \tau\in[0,1) is fixed as n\to\infty . • Weak Non-Hermiticity : This corresponds to the case where \tau\to 1 at a specific rate. In our study, we consider \tau=1-\alpha^{2}/n , where \alpha\in[0,\infty) is a fixed parameter. In this regime, one can observe critical transitions between asymmetric and symmetric matrices. This regime was introduced in the papers [ 36 , 37 , 38 ] . For the readers’ convenience, we provide a brief summary of the development of the study of real eigenvalues of the elliptic GinOE, both at strong and weak non-Hermiticity. The first natural question concerns the law of large numbers. To address this, let \mathcal{N}\equiv\mathcal{N}_{\tau} denote the (random) number of real eigenvalues of the elliptic GinOE. In order to establish the law of large numbers for the random variable \mathcal{N} (viewed as a sequence indexed by N ), one needs to determine the asymptotic behaviour of its mean and variance. It was established in [ 33 , 12 ] (cf. see [ 21 ] for an earlier study on the GinOE case) that (1.2) \mathbb{E}\mathcal{N}=\begin{cases}\mathfrak{b}(\tau)\sqrt{n}+O(1),&\textup{for }\tau\in[0,1)\text{ fixed},\\ \mathfrak{c}(\alpha)\,n+O(1),&\textup{for }\tau=1-\frac{\alpha^{2}}{n},\end{cases} where the expectation is taken with respect to the elliptic GinOE, and (1.3) \mathfrak{b}(\tau):=\sqrt{\frac{2}{\pi}\frac{1+\tau}{1-\tau}},\qquad\mathfrak{c}(\alpha):=e^{-\frac{\alpha^{2}}{2}}\Big(I_{0}\big(\tfrac{\alpha^{2}}{2}\big)+I_{1}\big(\tfrac{\alpha^{2}}{2}\big)\Big). Here, I_{\nu} is the modified Bessel function of the first kind given by (1.4) I_{\nu}(z):=\sum_{k=0}^{\infty}\frac{(z/2)^{2k+\nu}}{k!\,\Gamma(\nu+k+1)}, see e.g. [ 50 , Chapter 10] . We mention that the coefficients ( 1.2 ) can be realised as normalisation constants in the limiting macroscopic densities [ 22 , 12 , 13 ] . It is evident from ( 1.2 ) that a non-trivial proportion of eigenvalues is expected to be real in weak non-Hermiticity, whereas only a subdominant O(\sqrt{n}) number of eigenvalues are real in strong non-Hermiticity. This observation aligns with the intuition that the number of real eigenvalues increases as the ensemble becomes more symmetric. In addition to the mean asymptotic of \mathcal{N} , its variance has been studied in [ 33 , 12 ] as well, which reads (1.5) \lim_{n\to\infty}\frac{\textup{Var}\,\mathcal{N}}{\mathbb{E}\mathcal{N}}=\begin{cases}2-\sqrt{2},&\textup{for }\tau\in[0,1)\text{ fixed},\\ 2-2\dfrac{\mathfrak{c}(\sqrt{2}\alpha)}{\mathfrak{c}(\alpha)},&\textup{for }\tau=1-\frac{\alpha^{2}}{n}.\end{cases} Combining ( 1.2 ) and ( 1.5 ), one can immediately observe the law of large numbers for \mathcal{N} in probability. The next question concerns the fluctuations of the number of

Read the complete text