Message passing in random satisfiability problems

TL;DR

This paper reviews message passing algorithms, especially the cavity method, for solving constraint satisfaction problems like satisfiability, highlighting their ability to analyze solution clustering and phase transitions.

Contribution

It introduces the cavity method as a generalization of belief propagation, providing new analytical tools and algorithmic frameworks for satisfiability problems.

Findings

Analytic phase diagrams for satisfiability problems.

Cavity method extends belief propagation to clustered solution spaces.

Framework for deriving new algorithms based on statistical physics insights.

Abstract

This talk surveys the recent development of message passing procedures for solving constraint satisfaction problems. The cavity method from statistical physics provides a generalization of the belief propagation strategy that is able to deal with the clustering of solutions in these problems. It allows to derive analytic results on their phase diagrams, and offers a new algorithmic framework.