English

Edge universality of separable covariance matrices

Probability 2019-11-11 v2

Abstract

In this paper, we prove the edge universality of largest eigenvalues for separable covariance matrices of the form Q:=A1/2XBXA1/2\mathcal Q :=A^{1/2}XBX^*A^{1/2}. Here X=(xij)X=(x_{ij}) is an n×Nn\times N random matrix with xij=N1/2qijx_{ij}=N^{-1/2}q_{ij}, where qijq_{ij} are i.i.d.i.i.d. random variables with zero mean and unit variance, and AA and BB are respectively n×nn \times n and N×NN\times N deterministic non-negative definite symmetric (or Hermitian) matrices. We consider the high-dimensional case, i.e. n/Nd(0,){n}/{N}\to d \in (0, \infty) as NN\to \infty. Assuming Eqij3=0\mathbb E q_{ij}^3=0 and some mild conditions on AA and BB, we prove that the limiting distribution of the largest eigenvalue of Q\mathcal Q coincide with that of the corresponding Gaussian ensemble (i.e. the Q\mathcal Q with XX being an i.i.d.i.i.d. Gaussian matrix) as long as we have limss4P(qijs)=0\lim_{s \rightarrow \infty}s^4 \mathbb{P}(\vert q_{ij} \vert \geq s)=0, which is a sharp moment condition for edge universality. If we take B=IB=I, then Q\mathcal Q becomes the normal sample covariance matrix and the edge universality holds true without the vanishing third moment condition. So far, this is the strongest edge universality result for sample covariance matrices with correlated data (i.e. non-diagonal AA) and heavy tails, which improves the previous results in \cite{BPZ1,LS} (assuming high moments and diagonal AA), \cite{Anisotropic} (assuming high moments) and \cite{DY} (assuming diagonal AA).

Keywords

Cite

@article{arxiv.1809.04572,
  title  = {Edge universality of separable covariance matrices},
  author = {Fan Yang},
  journal= {arXiv preprint arXiv:1809.04572},
  year   = {2019}
}

Comments

58 pages, 1 figure

R2 v1 2026-06-23T04:04:16.588Z