Comparing SAT preprocessing techniques

Preprocessing algorithms can dramatically improve the performance of even the most efficient SAT solvers on many problem instances [8]. Such algorithms simplify the formula by various means, including the deduction of necessary assignments, the addition of implied clauses, the deletion of redundant clauses, and the identification of equivalent literals. A number of preprocessors have been made publicly available, including Compact [2], Compactor [6], Simplify [9], and CompressLite [4]. While each preprocessor is very successful on some instances, they each combine a different selection of simplification techniques, making it difficult to understand which techniques are responsible for their success on a particular problem. We are in the process of implementing each of the main simplification techniques in JQuest, a Java framework designed to support comparisons of SAT techniques. JQuest has already been used to evaluate the performance of various data structures for SAT solvers [7]. In addition to comparing existing techniques, we are creating efficient implementations of those techniques and investigating novel techniques. Here we briefly discuss two existing techniques: length-restricted resolution and binary equivalence finding.