Exponential anticoncentration of the permanent

Abstract

Let $A\in\mathbb{R}^{n\times n}$ be a random matrix with independent entries, and suppose that the entries are "uniformly anticoncentrated" in the sense that there is a constant $\varepsilon>0$ such that each entry $a_{ij}$ satisfies $\sup_{z}\Pr[a_{ij}=z]\le1-\varepsilon$ (for example, $A$ could be a uniformly random $n\times n$ matrix with $\pm1$ entries). Significantly improving previous bounds of Tao and Vu, we prove that the permanent of $A$ is exponentially anticoncentrated: there is $c_{\varepsilon}>0$ such that $\sup_{z}\Pr[\operatorname{per}(A)=z]\le\exp(-c_{\varepsilon}n)$. Our proof also works for the determinant, giving an alternative proof of a classical theorem of Kahn, Koml\'os and Szemer\'edi. As a consequence, we see that there are at least exponentially many different permanents of $n\times n$ matrices with $\pm1$ entries, resolving a problem of Ingram and Razborov.