Geometric Littlewood-Offord problems via lattice point counting

Abstract

Consider nonzero vectors $a_{1},\dots,a_{n}\in\mathbb{C}^{k}$, independent Rademacher random variables $\xi_{1},\dots,\xi_{n}$, and a set $S\subseteq\mathbb{C}^{k}$. What upper bounds can we prove on the probability that the random sum $\xi_{1}a_{1}+\dots+\xi_{n}a_{n}$ lies in $S$? We develop a general framework that allows us to reduce problems of this type to counting lattice points in $S$. We apply this framework with known results from diophantine geometry to prove various bounds when $S$ is a set of points in convex position, an algebraic variety, or a semialgebraic set. In particular, this resolves conjectures of Fox-Kwan-Spink and Kwan-Sauermann. We also obtain some corollaries for the polynomial Littlewood-Offord problem, for polynomials that have bounded Chow rank (i.e., can be written as a polynomial of a bounded number of linear forms). For example, one of our results confirms a conjecture of Nguyen and Vu in the special case of polynomials with bounded Chow rank: if a bounded-degree polynomial $F\in\mathbb{C}[x_{1},\dots,x_{n}]$ has bounded Chow rank and ''robustly depends on at least $b$ of its variables'', then $\mathbb{P}[F(\xi_{1},\dots,\xi_{n})=0]\le O(1/\sqrt{b})$. We also prove significantly stronger bounds when $F$ is ''robustly irreducible'', towards a conjecture of Costello.