Littlewood-Offord problems for Ising models

TL;DR

This paper investigates concentration probabilities in one-dimensional Ising models, establishing bounds on the likelihood of sums of spins falling within a specific interval, and applies these results to eigenvalue bounds of correlation matrices.

Contribution

It provides universal bounds on the concentration function for sums of spins in Ising models, extending Littlewood-Offord theory to this setting.

Findings

Upper bound on concentration probability: O(n^{-1/2})

Lower bound on the probability: inom{n}{[n/2]}2^{-n}

Application to eigenvalue bounds of correlation matrices

Abstract

We consider the one-dimensional Littlewood-Offord problem for general Ising models. More precisely, we consider the concentration function \[Q_n(x,v)=P\left(\sum_{i=1}^{n}\varepsilon_iv_i\in(x-1,x+1)\right),\] where $x\in\mathbb{R}$, $v_1,v_2,\ldots,v_n$ are real numbers such that $|v_1|\geq 1, |v_2|\geq 1,\ldots, |v_n|\geq 1$, and $(\varepsilon_i)_{i=1,2,\ldots,n}\in\{-1,1\}^{n}$ are random spins of some Ising model. Let $Q_n=\sup_{x,v}Q_n(x,v)$. Under natural assumptions, we show that there exists a universal constant $C$ such that for all $n\geq 1$, \[\binom{n}{[n/2]}2^{-n}\leq Q_n\leq Cn^{-\frac{1}{2}}.\] As an application of the method, under the same assumption, we give a lower bound on the smallest eigenvalue of the truncated correlation matrix of the Ising model.