Stochastic Systematic Search Algorithms for Satisfiability

Abstract

This paper proposes the utilization of randomized backtracking within complete backtrack search algorithms for Prepositional Satisfiability (SAT). Given that state-of-the-art SAT algorithms often randomize variable selection heuristics, and given that randomization is not naturally applicable to the identification of necessary assignments, our approach provides a fully stochastic, but complete, search algorithm for SAT. In addition, different organizations of randomized backtracking are described and compared. Moreover, we relate randomized backtracking with a more general form of backtracking, referred to as unrestricted backtracking. Finally, experimental results provide empirical evidence that randomized backtracking can be a competitive technique in solving hard real-world instances of SAT.