Exponential singularity and rank bounds for random symmetric matrices

Source abstract

Let $A$ be an $n\times n$ real symmetric random matrix whose entries above the diagonal are independent, each with largest atom at most $β\in (0,1)$. The diagonal entries have an arbitrary joint distribution and are independent of the entries above the diagonal. We prove that $$\mathbb{P}[\mathrm{rank}(A)\le n-k]\le 2e^{-c_βnk},\qquad 1\le k\le n,$$ where $c_β>0$ depends only on $β$. This extends Han's corank bound to the full range $1\le k\le n$. No moment assumptions are needed, and the entries need not be centered or identically distributed. In particular, the case $k=1$ gives an exponential singularity bound under the uniform atom assumption. Our argument also gives a new proof of the exponential singularity bound for symmetric Rademacher matrices obtained by Campos, Jenssen, Michelen, and Sahasrabudhe. For every fixed $p\in (0,1)$, it yields the exponential singularity bound conjectured by Costello and Vu for the adjacency matrix of the Erdős--Rényi random graph $G(n,p)$, where the case $p=1/2$ was previously proved by Han.