Title:Random Levy Matrices Revisited

Abstract: We compare two formulations of large but symmetric random matrices exhibiting a power law spectrum: a Free Random Levy (FRL) formulation with a rotationally symmetric measure, and a formulation due to Bouchaud and Cizeau (BC) based on a measure whereby each matrix entry is independently sampled from a symmetric stable Levy distribution. Both ensembles are spectrally stable with respect to matrix addition. After correcting the BC analysis at large N, we find a perfect agreement between the numerically sampled spectra and the analytical results for all symmetric distributions already for N=100. The numerical spectra show very weak dependence on the matrix size N, and are displayed for N=400. After pertinent rescaling, the BC and symmetric FRL spectra have the same tail behaviour. In bulk they deviate in the Cauchy limit, but rapidly agree in the Gauss limit. We illustrate the relation between the two types of stability and show that the addition of rotated BC matrices leads by a matrix central limit theorem to FRL spectra.