[Paper Review] Lecture Notes on "Random Matrices"

Roland Speicher|arXiv (Cornell University)|Sep 10, 2020

Random Matrices and Applications4 references4 citations

TL;DR

This lecture series provides a comprehensive introduction to random matrix theory, covering foundational topics such as Wigner's semicircle law, the Tracy–Widom distribution for the largest eigenvalue, and the circular law. It employs combinatorial, analytic, and probabilistic methods—including the Stieltjes transform, Wick's formula, and concentration inequalities—to establish the universality and asymptotic behavior of eigenvalue distributions in Gaussian and Wigner ensembles.

ABSTRACT

This in an introduction to random matrix theory, giving an impression of some of the most important aspects of this modern subject. In particular, it covers the basic combinatorial and analytic theory around Wigner's semicircle law, featuring also concentration phenomena, and the Tracy-Widom distribution of the largest eigenvalue. The circular law and a discussion of Voiculescu's multivariate extension of the semicircle law, as an appetizer for free probability theory, also make an appearance. The material here was presented in the winter term 2019/20 at Saarland University in 24 lectures of 90 minutes each. The lectures were recorded and can be found online at https://www.math.uni-sb.de/ag/speicher/web_video/index.html.

Motivation & Objective

• To provide a rigorous yet accessible introduction to random matrix theory for graduate students and researchers.
• To establish the semicircle law for Wigner and Gaussian random matrices using combinatorial and analytic techniques.
• To explore the limiting distribution of the largest eigenvalue via the Tracy–Widom distribution and its connection to non-intersecting Brownian motion.
• To introduce the circular law and asymptotic freeness in the context of independent GUE matrices, leading into free probability theory.
• To equip readers with tools for analyzing eigenvalue statistics through moment methods, Stieltjes transforms, and concentration inequalities.

Proposed method

• Uses the Wick formula and non-crossing pairings to compute moments of Gaussian random matrices and derive the semicircle law combinatorially.
• Applies the Stieltjes transform to analyze weak convergence of spectral measures and prove the semicircle law analytically.
• Employs Poincaré and logarithmic Sobolev inequalities to establish concentration of measure for linear spectral statistics.
• Utilizes the Karlin–McGregor and Gessel–Viennot identities to connect eigenvalue statistics to non-intersecting paths and determinantal point processes.
• Derives the Tracy–Widom distribution via heuristic and rigorous approaches, including the Harer–Zagier recursion and Painlevé II equation.
• Applies the RSK correspondence and path decomposition to link the longest increasing subsequence to eigenvalue fluctuations in GUE matrices.

Experimental results

Research questions

• RQ1 How do the eigenvalues of Wigner and Gaussian random matrices behave in the large-N limit, and what determines the semicircle law?
• RQ2 What is the limiting distribution of the largest eigenvalue in GUE and GOE ensembles, and how does it relate to the Tracy–Widom distribution?
• RQ3 How do concentration inequalities and logarithmic Sobolev inequalities control the fluctuations of spectral linear statistics?
• RQ4 What is the connection between non-intersecting Brownian motion, determinantal processes, and the eigenvalue statistics of random matrices?
• RQ5 How does the circular law emerge for non-Hermitian matrices such as the Ginibre ensemble, and what is its universality?

Key findings

• The eigenvalue distribution of Wigner and GUE matrices converges almost surely to the semicircle law as N → ∞, with the limiting density given by ρ(x) = (1/(2π))√(4−x²) on [−2,2].
• The largest eigenvalue of GUE matrices, properly scaled, converges almost surely to 2, and its fluctuations are governed by the Tracy–Widom distribution.
• The joint eigenvalue density of GUE can be expressed using Hermite polynomials and the Vandermonde determinant, leading to a determinantal point process structure.
• The distribution of the longest increasing subsequence in random permutations converges to the Tracy–Widom distribution, linking combinatorics and random matrix theory.
• For non-Hermitian i.i.d. matrices with mean zero and finite variance, the eigenvalues converge to the uniform distribution on the unit disk (circular law), as N → ∞.
• The eigenvalues of independent GUE matrices become asymptotically free in the sense of free probability, with joint moments governed by free independence.