Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Extremal random matrices with independent entries and matrix superconcentration inequalities

Tatiana Brailovskaya, Ramon van Handel

TL;DR

This work develops sharp nonasymptotic matrix concentration bounds for the spectral norm of matrices with independent entries across arbitrary variance patterns, achieving fluctuations at the Tracy–Widom scale and enabling precise control across sparse to dense regimes. The central tool is an extremum principle that reduces moment control of a nonhomogeneous matrix to that of a smaller homogeneous Gaussian model, with block-diagonal structures maximizing moments. A detailed Wick-based moment analysis for Hermitian and symmetric models, together with Ledoux-type tail translation, yields tight small- and large-deviation bounds matching known Tracy–Widom behavior in Wishart-type settings. The paper further provides sharp moment recursions and bounds for complex and real Wishart matrices that quantify the dependence on the aspect ratio $c=m/n$, including regime-specific bounds that reproduce the correct asymptotics. Together, these results offer a unified, sharp framework for matrix concentration beyond homogeneous settings and illuminate the role of variance patterns in spectral norm fluctuations.

Abstract

We prove nonasymptotic matrix concentration inequalities for the spectral norm of (sub)gaussian random matrices with centered independent entries that capture fluctuations at the Tracy-Widom scale. This considerably improves previous bounds in this setting due to Bandeira and Van Handel, and establishes the best possible tail behavior for random matrices with an arbitrary variance pattern. These bounds arise from an extremum problem for nonhomogeneous random matrices: among all variance patterns with a given sparsity parameter, the moments of the random matrix are maximized by block-diagonal matrices with i.i.d. entries in each block. As part of the proof, we obtain sharp bounds on large moments of Gaussian Wishart matrices.

Extremal random matrices with independent entries and matrix superconcentration inequalities

TL;DR

This work develops sharp nonasymptotic matrix concentration bounds for the spectral norm of matrices with independent entries across arbitrary variance patterns, achieving fluctuations at the Tracy–Widom scale and enabling precise control across sparse to dense regimes. The central tool is an extremum principle that reduces moment control of a nonhomogeneous matrix to that of a smaller homogeneous Gaussian model, with block-diagonal structures maximizing moments. A detailed Wick-based moment analysis for Hermitian and symmetric models, together with Ledoux-type tail translation, yields tight small- and large-deviation bounds matching known Tracy–Widom behavior in Wishart-type settings. The paper further provides sharp moment recursions and bounds for complex and real Wishart matrices that quantify the dependence on the aspect ratio , including regime-specific bounds that reproduce the correct asymptotics. Together, these results offer a unified, sharp framework for matrix concentration beyond homogeneous settings and illuminate the role of variance patterns in spectral norm fluctuations.

Abstract

We prove nonasymptotic matrix concentration inequalities for the spectral norm of (sub)gaussian random matrices with centered independent entries that capture fluctuations at the Tracy-Widom scale. This considerably improves previous bounds in this setting due to Bandeira and Van Handel, and establishes the best possible tail behavior for random matrices with an arbitrary variance pattern. These bounds arise from an extremum problem for nonhomogeneous random matrices: among all variance patterns with a given sparsity parameter, the moments of the random matrix are maximized by block-diagonal matrices with i.i.d. entries in each block. As part of the proof, we obtain sharp bounds on large moments of Gaussian Wishart matrices.
Paper Structure (31 sections, 54 theorems, 186 equations, 1 figure)

This paper contains 31 sections, 54 theorems, 186 equations, 1 figure.

Table of Contents

  1. Introduction
  2. Main results
  3. Small deviations
  4. Large deviations
  5. Extremum principle
  6. Self-adjoint models and outline
  7. Discussion and open questions
  8. Intrinsic freeness
  9. Superconcentration
  10. Universality
  11. Notation
  12. The Hermitian case
  13. Extremum principle
  14. Small deviations
  15. Large deviations
  16. ...and 16 more sections

Key Result

Theorem 1.1

Let $n\le m$, and let $X$ be the $n\times m$ random matrix whose entries $X_{ij}\sim N(0,b_{ij}^2)$ are independent. Then we have for all $t\ge 0$ and $0<\varepsilon\le\frac{1}{2}$, where the constant $C_\varepsilon$ depends only on $\varepsilon$.

Figures (1)

  • Figure 1.1: An extremal block-diagonal matrix.

Theorems & Definitions (107)

  • Theorem 1.1: BvH16
  • Example 1.2
  • Theorem 1.4: Small deviations
  • Remark 1.5
  • Theorem 1.6: Large deviations
  • Theorem 1.7: Extremum principle
  • Corollary 1.8
  • Corollary 1.9
  • Lemma 1.10
  • proof
  • ...and 97 more