Decoupling of clusters in independent sets in a percolated hypercube

TL;DR

This paper analyzes the structure and count of independent sets in a percolated hypercube for all percolation probabilities above 0.465, revealing complex clustering behavior and providing new probabilistic tools.

Contribution

The authors develop a novel probabilistic framework to understand clustering in independent sets of a percolated hypercube for low percolation probabilities, extending previous results.

Findings

Sharp approximation for the number of independent sets in the percolated hypercube

Sampling algorithm for independent sets at low percolation probabilities

Identification that p=1/2 is not a phase transition point in this model

Abstract

Independent sets in graphs are sets of vertices containing no neighbors, and they represent a canonical spin system with hardcore constraints. Of particular interest is the setting of the boolean hypercube, where counting independent sets was the original motivator for Sapozhenko's famous graph container method. A modern perspective on such problems is to consider the effect of disorder, and the study of independent sets in random subgraphs of the hypercube obtained via bond percolation with parameter $p$ was initiated by Kronenberg and Spinka. They employed tools from statistical mechanics to obtain detailed information about the moments of the number of independent sets (now a random variable), and posed many interesting questions. Previous work by the authors addressed many of these questions in the regime $p \geq \frac{2}{3}$, where the behavior is relatively simple and can be modeled well by a related family of independent particles. As $p$ decreases, though, typical independent sets become larger and feature more intricate clustering behavior. In the present article we overcome many of the challenges presented by this phenomenon and analyze the model for all $p> 0.465$. We obtain a sharp in-probability approximation for the number of independent sets in the percolated hypercube in terms of explicit random variables, as well as provide a sampling algorithm. Note that this shows, curiously, that $p = \frac{1}{2}$ is not a natural barrier for this problem unlike in many other problems where it appears as a point of a phase transition. A key contribution of this work is the introduction of a new probabilistic framework to handle the clustering behavior for these low values of $p$. Although our analysis is restricted to $p > 0.465$, our arguments are expected to be helpful for studying this model at even lower values of $p$, and possibly for other related problems.