[Paper Review] Transformations of polynomial ensembles

Arno B. J. Kuijlaars|arXiv (Cornell University)|Jan 22, 2015

Advanced Mathematical Theories and Applications4 citations

TL;DR

This paper establishes that several fundamental random matrix transformations—such as multiplication by a Ginibre matrix, rank-one modifications, and matrix extensions with complex Gaussians—preserve the polynomial ensemble structure of eigenvalue or singular value distributions. The key contribution is deriving explicit formulas for the transformed weight functions, enabling inductive proofs of classical random matrix results like the joint eigenvalue density of GUE matrices.

ABSTRACT

A polynomial ensemble is a probability density function for the position of $n$ real particles of the form $\frac{1}{Z_n} \, \prod_{j

Motivation & Objective

• To identify and characterize transformations in random matrix theory that preserve the polynomial ensemble structure of eigenvalue or singular value distributions.
• To provide a unified framework for understanding how operations like multiplication by Ginibre matrices or matrix extensions affect the joint density of eigenvalues.
• To derive explicit expressions for the new weight functions in the transformed ensembles, enabling analytical tractability.
• To apply these results to give an inductive proof of the joint eigenvalue density of the Gaussian Unitary Ensemble (GUE).

Proposed method

• Uses the determinant structure of polynomial ensembles, expressed as $ \mathcal{P}(x_1,\ldots,x_n) = \frac{1}{Z_n} \Delta_n(x) \det[f_k(x_j)]_{j,k=1}^n $, to analyze transformations.
• Applies the Andreief identity to average over the original ensemble and derive the new joint density in terms of integrals of the original functions.
• Employs Mellin convolution and integral transforms to characterize the new functions $ g_k(y) $ after transformations such as multiplication by a Ginibre matrix.
• Uses interlacing determinant identities to handle the eigenvalue distribution after adding a row and column with complex Gaussians.
• Derives explicit forms for the transformed weight functions, such as $ g_k(y) = e^{-y^2/2} \int_0^y e^{x^2/2} f_{k-1}(x) dx $, in the rank-one extension case.
• Relies on known results from random matrix theory and biorthogonal ensembles to validate the structure preservation under these operations.

Experimental results

Research questions

• RQ1 Which matrix transformations preserve the polynomial ensemble structure of eigenvalue or singular value distributions?
• RQ2 How do the weight functions of the polynomial ensemble transform under multiplication by a complex Ginibre matrix?
• RQ3 Can the eigenvalue distribution of a GUE matrix be derived inductively using transformations that preserve the polynomial ensemble structure?
• RQ4 What is the explicit form of the new weight functions after a rank-one modification or extension of a Hermitian matrix with complex Gaussians?
• RQ5 How can the interlacing property of eigenvalues after matrix extension be used to derive the new joint density?

Key findings

• Multiplication of a random matrix $ X $ with squared singular values in a polynomial ensemble by a complex Ginibre matrix results in a new polynomial ensemble with transformed weight functions given by the Mellin convolution $ g_k(y) = \int_0^\infty x^\nu e^{-x} f_k(y/x) \frac{dx}{x} $.
• The rank-one modification of a Hermitian matrix by adding a complex Gaussian row and column preserves the polynomial ensemble structure, with new weight functions $ g_1(y) = e^{-y^2/2} $ and $ g_{k+1}(y) = e^{-y^2/2} \int_0^y e^{x^2/2} f_k(x) dx $.
• The eigenvalue distribution of the extended matrix $ Y $ is a polynomial ensemble with the same Vandermonde determinant and a modified determinant of functions, explicitly derived via interlacing and averaging techniques.
• The method yields an inductive proof of the GUE eigenvalue density $ \frac{1}{Z_n} \Delta_n(x)^2 \prod_{j=1}^n e^{-x_j^2/2} $, starting from the polynomial ensemble with $ f_k(x) = x^{k-1} e^{-x^2/2} $.
• The transformation rules are general and can be applied iteratively, preserving the polynomial ensemble structure across multiple operations.
• The results are validated by recovering known results in random matrix theory, such as the joint eigenvalue density of GUE, through structural preservation rather than direct computation.