AI & Computingarticle2026-08-14

Spectral Properties and Localization Barriers of Co-occurrence Gram Matrices in Set Families

Open access0 citations

Abstract

In this paper, we establish the spectral framework governing $G = V^T V$ for arbitrary $0\text{-}1$ matrices. Combining Rayleigh-Ritz variational bounds with Gershgorin disk localization, we prove a general spectral inequality relating element frequencies to local average set sizes:\[f_{\max} \ge \frac{\E[|S|^2]}{n \cdot \max_{i \in U} \bar{s}_i}.\]We construct an explicit counterexample to illustrate the \emph{Gershgorin Localization Barrier}, demonstrating why the maximum product $f_i \bar{s}_i$ is not necessarily attained at the element $x^*$ of maximum frequency $f_{\max}$. We emphasize that these spectral bounds hold for general set systems independent of union-closure, and we outline the precise structural alignment conditions required to leverage this framework toward Frankl's conjecture.

// Source

View paper (DOI)Open access versionOpenAlexZenodo (CERN European Organization for Nuclear Research)Published 2026-08-14

Authors: Terence Cheng