Early Conflict Detection Based BCP for SAT Solving

This paper describes a new BCP algorithm that improves the performance of Chaff class solvers by reducing the total number of clauses the BCP procedure must evaluate. This is done by: detecting conflicts earlier; evaluating clauses better; and by increasing the controllability of the conflicts which the BCP procedure finds. Solvers like Limmat [10] include a simple Early Conflict Detection BCP (ECDB), however we introduce a new aggressive ECDB procedure and the MIRA solver that efficiently incorporates it while easily facilitating comparisons between ECDB modes. With the full ECDB procedure enabled, MIRA was able to reduce the number of evaluated clauses by 59% on average compared to the disabled ECDB version. This new procedure and other speedup techniques discussed here allow MIRA to solve problems 3.7 times faster on average than zChaff. Lastly, this paper shows how significant speedup can be attained relatively easily by incorporating a certain level of ECDB into other solvers.