Real Eigenvalues of Asymmetric Wishart Matrices: Expected Number, Global Density and Integrable Structure

Sung-Soo Byun, and, Kohei Noda · 2025

Read Real Eigenvalues of Asymmetric Wishart Matrices: Expected Number, Global Density and Integrable Structure by Sung-Soo Byun, and, Kohei Noda (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-2503-14942.txt.

Introduction

Sung-Soo Byun and Kohei Noda study the real eigenvalues of asymmetric Wishart matrices, random arrays used across statistics and physics. Their 2025 paper varies two knobs: how rectangular the matrix is, and how far it sits from symmetry, in both fixed and scaling regimes.

The authors find a decomposition for the correlation kernel of the real eigenvalues, which form a Pfaffian point process. In the symmetric limit their structure reduces to a known rank-one form from earlier work. Counting then shows the expected number of real eigenvalues grows like the square root of the size in the strongly asymmetric regime, and in direct proportion in the near-symmetric one. Limit densities also follow, matching a classic law at symmetry.

Catalog record

CREATORSung-Soo Byun, and, Kohei Noda
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE128 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-14942

Browse the full library · About the Tower of Babel

Full text

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

Real Eigenvalues of Asymmetric Wishart Matrices: Expected Number, Global Density and Integrable Structure Sung-Soo Byun, and, Kohei Noda Abstract We investigate the real eigenvalues of asymmetric Wishart matrices of size N N , indexed by the rectangular parameter ν ∈ ℕ \nu\in\mathbb{N} and the non-Hermiticity parameter τ ∈ [ 0 , 1 ] \tau\in[0,1] . The rectangular parameter ν \nu is either fixed or proportional to N N . The non-Hermiticity parameter τ \tau is either fixed or τ = 1 − O ⁡ ( 1 / N ) \tau=1-O(1/N) , corresponding to the strongly and weakly non-Hermitian regimes, respectively. We establish a decomposition structure for the finite- N N correlation kernel of the real eigenvalues, which form Pfaffian point processes. Taking the symmetric limit τ = 1 \tau=1 , where the model reduces to the Laguerre orthogonal ensemble, this decomposition structure reduces to the known rank-one perturbation structure established by Adler, Forrester, Nagao, and van Moerbeke, as well as by Widom. Using the decomposition structure, we show that the expected number of real eigenvalues is proportional to N \sqrt{N} in the strongly non-Hermitian regime and to N N in the weakly non-Hermitian regime, providing explicit leading coefficients in both cases. Furthermore, we derive the limiting real eigenvalue densities, which recovers the Marchenko-Pastur distribution in the symmetric limit. 1 Introduction and main results For given non-negative integers N and \nu , let \mathrm{P} and \mathrm{Q} be N\times(N+\nu) random matrices with independent real Gaussian entries, each having mean 0 and variance 1/(2N) . The matrices \mathrm{P} and \mathrm{Q} are independent and are referred to as rectangular Ginibre matrices [ 22 ] . By definition, the asymmetric Wishart matrix X is given by X:=X_{+}X_{-}^{\mathsf{T}},\hskip 20.00003ptX_{\pm}=\sqrt{1+\tau}\,\mathrm{P}\pm\sqrt{1-\tau}\,\mathrm{Q}, (1.1) where \tau\in[0,1] and A^{\mathsf{T}} denotes the transposition of a matrix A . Here, the rectangular parameter \nu\equiv\nu_{N} and the non-Hermiticity parameter \tau\equiv\tau_{N} possibly depend on N . The model X appears in the literature also under the name of sample cross-covariance matrix or chiral elliptic Ginibre ensemble , and finds applications in various areas such as quantum chromodynamics and ecosystem stability; see e.g. [ 19 , 61 , 14 , 62 ] and references therein. Note in particular that the parameter \tau\in[0,1] quantifies the asymmetry of the model. In the extremal, symmetric case when \tau=1 , the matrix model X becomes the classical Wishart matrix (the Laguerre orthogonal ensemble) [ 39 , Chapter 3] . Beyond the matrix with real elements, the complex and quaternionic counterparts of non-Hermitian Wishart matrices have also been extensively studied in the literature; see e.g. [ 65 , 27 , 9 , 4 , 5 , 6 , 15 , 11 , 29 , 56 ] and references therein. A remarkable property of the eigenvalues in the complex and quaternionic cases is their interpretation as one-component Coulomb gas ensembles [ 39 , 67 , 22 ] . In contrast, such a one-component realisation is no longer valid for matrices with real elements. The key distinction lies in the fact that, in the real case, there is a non-trivial probability of having purely real eigenvalues; see Figure 1 . In this paper, we investigate the real eigenvalues of asymmetric Wishart matrices. The main contributions of this work are summarised as follows: • Expected number of real eigenvalues (Theorem 1.1 ); • Macroscopic density of real eigenvalues (Theorem 1.2 ); • Decomposition structure of the correlation kernel (Theorem 2.1 ). Analogous results in the context of the elliptic Ginibre matrices were established in [ 36 , 46 , 24 , 37 ] . (a) \nu=0 (b) \nu=N Figure 1 : The plots show the eigenvalues of asymmetric Wishart matrices with N=400 and \tau=1/2 . In the zoomed-in images, the presence of purely real eigenvalues is clearly visible. Furthermore, in the case of \nu=0 , there is an accumulation of eigenvalues near the origin, reflecting the divergent density. In the study of asymmetric Wishart matrices, various interesting regimes emerge, depending on the scaling properties of the parameters. (a1) Strongly non-Hermitian regime . This is the case where \tau stays away from the extremal, symmetric case \tau=1 . In particular, we usually consider the case where \tau\in[0,1) is fixed. (a2) Weakly non-Hermitian regime . This is the case when \tau\uparrow 1 with a proper speed. This regime, first pioneered in [ 50 , 51 , 52 ] for the elliptic Ginibre matrices, allows one to observe a non-trivial transition to Hermitian matrices. In particular, we mainly consider the regime 1-\tau=O(N^{-1}) . (b1) Singular density regime . This is the case where \nu=o(N) . In this case, for general \tau\in[0,1) , the origin is contained in the limiting spectrum with divergent density; cf. Figure 1 (A). (b2) Regular density regime . This is the case where \nu=\varrho N with a positive \varrho>0 . Whether the origin is contained in the limiting spectrum is determined by the sign of \tau\sqrt{1+\varrho}-1 ; see ( 1.18 ) and ( 1.20 ). If the sign is positive, the origin is included in the spectrum; however, the density at the origin remains bounded. See Figure 1 (B). Following the notations in the previous literature, we use the parametrisation \tau=1-\frac{\alpha^{2}}{2N},\hskip 20.00003pt\alpha>0 (1.2) for the weakly non-Hermitian regime. (See also a recent work [ 10 ] where various mesoscopic regimes were explored.) We also note that when discussing macroscopic properties, it is often unnecessary to distinguish between the two cases of rectangular parameters. By combining these cases, one can assume that \lim_{N\to\infty}\frac{\nu}{N}=\varrho\in[0,\infty). (1.3) In the following, we focus on the case where N is an even integer, as this choice simplifies the overall presentation and analysis. While the case of odd N is also tractable, it requires additional computations. For instance, the odd N case was addressed in [ 70 , 43 ] for the real Ginibre matrix. Let \mathcal{N}\equiv\mathcal{N}_{\tau,\nu} denote the number of real eigenvalues. We denote by \mathbf{R}_{N,k} the k -point correlation function of real eigenvalues; see [ 22 , Chapter 7] . In particular, the 1 -point function \mathbf{R}_{N}\equiv\mathbf{R}_{N,1} can be characterised by \mathbb{E}\Big[\sum_{j=1}^{\mathcal{N}}f(x_{j})\Big]=\int_{\mathbb{R}}f(x)\,\mathbf{R}_{N}(x)\,dx, (1.4) where f is a test function. For general k , the correlation function \mathbf{R}_{N,k} can be written explicitly in terms of a certain Pfaffian; see ( 2.7 ) below. When exploring the real eigenvalues of asymmetric random matrices, a natural object of interest is their number/counting statistics . A fundamental aspect of this is determining the typical number of real eigenvalues. This has been studied in various models, beginning with the seminal work on the real Ginibre matrix [ 36 , 45 ] and extending to its variants, including the elliptic Ginibre matrix in both the strongly [ 46 ] and weakly [ 24 ] non-Hermitian regimes, as well as truncated matrices and their products [ 48 , 60 , 47 ] , spherical ensembles [ 36 , 44 ] , and products of Ginibre matrices [ 7 , 69 , 40 ] . Furthermore, [ 72 ] established that matrices with general i.i.d. elements asymptotically exhibit the same number of real eigenvalues, thereby demonstrating universality. (We also refer the reader to [ 30 , 16 , 8 , 53 , 33 , 55 ] and the references therein for the counting statistics of various

Read the complete text