English

Extreme eigenvalues of sparse, heavy tailed random matrices

Probability 2015-06-23 v1

Abstract

We study the statistics of the largest eigenvalues of p×pp \times p sample covariance matrices Σp,n=Mp,nMp,n\Sigma_{p,n} = M_{p,n}M_{p,n}^{*} when the entries of the p×np \times n matrix Mp,nM_{p,n} are sparse and have a distribution with tail tαt^{-\alpha}, α>0\alpha>0. On average the number of nonzero entries of Mp,nM_{p,n} is of order nμ+1n^{\mu+1}, 0μ10 \leq \mu \leq 1. We prove that in the large nn limit, the largest eigenvalues are Poissonian if α<2(1+μ1)\alpha<2(1+\mu^{{-1}}) and converge to a constant in the case α>2(1+μ1)\alpha>2(1+\mu^{{-1}}). We also extend the results of Benaych-Georges and Peche [7] in the Hermitian case, removing restrictions on the number of nonzero entries of the matrix.

Keywords

Cite

@article{arxiv.1506.06175,
  title  = {Extreme eigenvalues of sparse, heavy tailed random matrices},
  author = {Antonio Auffinger and Si Tang},
  journal= {arXiv preprint arXiv:1506.06175},
  year   = {2015}
}

Comments

23 pages

R2 v1 2026-06-22T09:57:04.627Z