Learning from conflicts in propositional satisfiability

Abstract

Learning is a general concept, playing an important role in many Artificial intelligence domains. In this paper, we address the learning paradigm used to explain failures or conflicts encountered during search. This explanation, derived by conflict analysis, and generally expressed as a new constraint, is usually used to dynamically avoid future occurrences of similar situations. Before focusing on clause learning in Boolean satisfiability (SAT), we first overview some important works on this powerful reasoning tool in other domains such as constraint satisfaction and truth maintenance systems. Then, we present a comprehensive survey of the most important works having led to what is called today—conflict driven clause learning (CDCL)—which is one of the key components of modern SAT solvers. In theory, current SAT solvers with clause learning are as powerful as general resolution proof systems. In practice, real-world SAT instances with millions of variables and clauses are now in the scope of this solving paradigm.