Random Matrices¶
Job: spectral statistics of large structured random matrices — Wigner ensembles, universal bulk/edge laws, and the bridge to free probability and mathematical physics.
Core¶
A random matrix ensemble is a probability law on matrices (often Hermitian or symmetric) with prescribed entry distributions and symmetries. As size \(N\to\infty\), empirical spectral distributions converge; local eigenvalue statistics exhibit universality (GOE/GUE/GSE classes) matching integrable kernels.
| Layer | Content |
|---|---|
| Wigner | i.i.d. (up to symmetry) entries; semicircle law |
| Classical compact | GOE/GUE/GSE; Haar unitaries (CUE) |
| Bulk universality | sine kernel after unfolding |
| Edge | Tracy–Widom laws for largest eigenvalue |
| Free probability | asymptotic freeness of independent Wigner / unitaries |
Reading map¶
| Page | Focus |
|---|---|
| Foundations | Wigner, GOE/GUE, semicircle, Tracy–Widom edge, free probability bridge |
| Universality & Log-Gases | DBM; three-step (Erdős–Schlein–Yau); Tao–Vu replacement; rigidity; log-gas |
| Determinantal & β-Ensembles | DPP; GUE kernel / RH method; Airy / Tracy–Widom / Painlevé II; loop equations; β-stochastic Airy |
| Spiked Models & Statistics | BBP transition; deformed semicircle / free convolution; PCA / signal detection; zeta-GUE |
| Non-Hermitian & Circular | circular law (Girko / Tao–Vu); Ginibre / DPP; single ring / Brown measure; elliptic law |
Dependency order¶
- Measure-theoretic probability, weak convergence — probability.
- Linear algebra of Hermitian eigenproblems; optional FA — functional analysis.
- Free probability as the NC limit calculus — free probability.
- Physics motivation — mathematical physics.
Interfaces¶
| Area | Contact |
|---|---|
| Probability | LLN/CLT for spectra; concentration |
| Free probability | asymptotic freeness; Cauchy transform methods |
| Mathematical physics | quantum chaos; disordered systems; QCD Dirac spectra |
| Representation theory | classical compact groups; Schur–Weyl |
| Functional analysis | operators on Hilbert space; spectral measures |
Notation (paper-standard)¶
- ESD \(\mu_H=\frac1N\sum_{i=1}^N\delta_{\lambda_i}\); semicircle \(\sigma_{\mathrm{sc}}\)
- GOE/GUE/GSE (\(\beta=1,2,4\)); Tracy–Widom \(F_\beta\)
- Gap probability; correlation kernels \(K_N(x,y)\)
See it¶
Wigner semicircle. Real symmetric \(N\times N\) with i.i.d. upper-triangular entries mean \(0\), variance \(1/N\) (diagonal variance \(2/N\) variants): ESD \(\Rightarrow\) semicircle density \(\frac{1}{2\pi}\sqrt{4-x^2}\) on \([-2,2]\) almost surely. Local bulk gaps are order \(1/N\); largest eigenvalue sits at \(2+O(N^{-2/3})\) with Tracy–Widom fluctuations.
Check¶
Why normalize off-diagonal variances by \(1/N\) for a nontrivial bulk spectrum on \(O(1)\) scale?
Typical row \(\ell^2\) norm squared is \(\sum_j \operatorname{Var}(H_{ij})\sim 1\); eigenvalues of dense matrices with entry variance \(1\) would be order \(\sqrt{N}\). Scaling \(1/N\) places the spectrum on a fixed interval (semicircle radius \(2\)).
What is the symmetry distinction between GOE and GUE?
GOE: real symmetric (time-reversal invariant, \(\beta=1\)); GUE: complex Hermitian with independent real/imaginary Gaussian parts (\(\beta=2\)). Eigenvalue joint laws differ by the Jacobian factor \(\prod_{i<j}|\lambda_i-\lambda_j|^\beta\).
Pitfalls¶
- Global vs local: semicircle is macroscopic ESD; universality of spacings is microscopic (need unfolding).
- Heavy tails / sparse matrices: different limits (Lévy, tree-like) — not semicircle.
- Non-Hermitian: circular law (Girko) is a separate theory (Ginibre, etc.).
Refs¶
- Anderson–Guionnet–Zeitouni, An Introduction to Random Matrices
- Tao, Topics in Random Matrix Theory
- Mehta, Random Matrices
- Deift, Orthogonal Polynomials and Random Matrices
Next¶
Foundations · Universality & Log-Gases · Determinantal & β-Ensembles · Spiked Models & Statistics · Non-Hermitian & Circular · Free probability · Probability · Mathematical physics · Functional analysis · Representation theory