Probability Conditions for Convergence of the Warning Propagation Algorithm

Abstract

Message propagation algorithms are very effective in finding satisfying assignments for random SAT instances, and hard region become narrower. However, message propagation algorithms do not always converge for graphs with cycles. Unfortunately, rigorous theory proof of this phenomenon is still lacking. Warning Propagation algorithm is the most basic message propagation algorithm, we analyses convergence of the warning propagation algorithm, and gives the conditions for convergence.