Arc Consistency in SAT

I describe and study the 'support encoding' of binary constraint satisfaction problems (CSP's) into boolean satisfiability (SAT). This is based on work by Kasif in 1990, in which clauses encode support information, rather than the encoding of conflicts in the standard 'direct encoding'. This enables arc consistency in the original CSP to be established by propagation in the translated SAT instance, providing an optimal worst case time algorithm for arc consistency [14]. I prove that the DP algorithm in SAT applied to the suport encoding behaves exactly like the MAC algorithm in CSP's. Using the SAT solver Chaff, I show that the support encoding can be used effectively to solve hard instances of random binary CSP's, and more effectively than the direct encoding on hard instances. Finally, I show that the local search algorithm WalkSAT can perform many times better on the support encoding than on the direct encoding.