BBP transition     random-matrices

Leonid Petrov


Simulation Info

BBP transition     random-matrices

Leonid Petrov

Eigenvalue histogram of a rank-one perturbed GOE matrix demonstrating the BBP phase transition. As perturbation theta increases, the largest eigenvalue separates from the Wigner semicircle bulk. Point process plots show top 10, lowest 10, and near-zero eigenvalues. Adjust matrix size N and theta via synced slider pairs.

This simulation uses WebAssembly and the Eigen library to compute eigenvalues of a (modified) Gaussian Orthogonal Ensemble (GOE) matrix. We introduce a rank-1 perturbation governed by a parameter $\theta$: $$A\mapsto A + \theta \cdot e_1e_1^T,$$ where $A$ is the original GOE matrix, and $e_1$ is the first basis vector. There is the BBP phase transition phenomenon: for large enough $|\theta|$, the top eigenvalue “spikes” out of the traditional GOE spectrum.

100         0
Upper 10×10 Corner of Matrix:
Lowest 5 Eigenvalues:
5 Eigenvalues around zero:
Top 5 Eigenvalues:
Lowest 10 Eigenvalues:
20 Eigenvalues Around Zero:
Top 10 Eigenvalues:
100         0
Histogram of eigenvalues:

code

(note: parameters in the code might differ from the ones in simulation results below)

Dear colleagues:

Feel free to use code (unless otherwise specified next to the corresponding link), data, and visualizations to illustrate your research in talks and papers, with attribution (CC BY-SA 4.0 (opens in new tab)). Some images are available in very high resolution upon request. I can also produce other simulations upon request - email me at lenia.petrov@gmail.com
This material is based upon work supported by the National Science Foundation under Grant DMS-2153869