There are several powerful solvers for satisfiability (SAT), such as WSAT , Davis-Putnam, and RELSAT . However, in practice, the SAT encodings often have so many clauses that we exceed physical memory resources on attempting to solve them. This excessive size often arises because conversion to SAT, from a more natural encoding using quantifications over domains, requires expanding quantifiers.
This suggests that we should “lift” successful SAT solvers. That is, adapt the solvers to use quantified clauses instead of ground clauses. However, it was generally believed that such lifted solvers would be impractical: Partially, because of the overhead of handling the predicates and quantifiers, and partially because lifting would not allow essential indexing and caching schemes.
Here we show that, to the contrary, it is not only practical to handle quantified clauses directly, but that lifting can give exponential savings. We do this by identifying certain tasks that are central to the implementation of a SAT solver. These tasks involve the extraction of information from the set of clauses (such as finding the set of unsatisfied clauses in the case of WSAT ) and consume most of the running time. We demonstrate that these tasks are NP-hard with respect to their relevant size measure. Hence, they are themselves search problems, and so we call them “subsearch problems”.
Ground SAT solvers effectively solve these subsearch problems by naive enumeration of the search space. In contrast, a lifted solver can solve them using intelligent search methods. Consequently, lifting a solver will generally allow an exponential reduction in the cost of subsearch and so increase the speed of the search engine itself.
Experimental results are given for a lifted version of WSAT . We only use very simple backtracking for the subsearch, but we still find that cost savings in the subsearch can more than offset the overheads from lifting. The reduction in size of the formulas also allows us to solve problems that are too large for ground WSAT .
In summary, a lifted SAT solver not only uses far less memory than a ground solver, but it can also run faster.
Cited By
- Jain D, von Gleissenthall K and Beetz M Bayesian logic networks and the search for samples with backward simulation and abstract constraint learning Proceedings of the 34th Annual German conference on Advances in artificial intelligence, (144-156)
- Baumgartner P and Tinelli C (2008). The model evolution calculus as a first-order DPLL method, Artificial Intelligence, 172:4-5, (591-632), Online publication date: 1-Mar-2008.
- East D and Truszczyński M (2006). Predicate-calculus-based logics for modeling and solving search problems, ACM Transactions on Computational Logic (TOCL), 7:1, (38-83), Online publication date: 1-Jan-2006.
- Dixon H, Ginsberg M and Parkes A (2004). Generalizing Boolean satisfiability I, Journal of Artificial Intelligence Research, 21:1, (193-243), Online publication date: 1-Jan-2004.
- Baumgartner P First-order logic Davis-Putnam-Logemann-Loveland procedure Exploring artificial intelligence in the new millennium, (289-329)
- De Giacomo G and Levesque H Two approaches to efficient open-world reasoning Logic-based artificial intelligence, (59-78)
Recommendations
Orchestrating satisfiability engines
CP'11: Proceedings of the 17th international conference on Principles and practice of constraint programmingConstraint satisfaction problems arise in many diverse areas including software and hardware verification, type inference, static program analysis, test-case generation, scheduling, planning and graph problems. These areas share a common trait, they ...
A satisfiability procedure for quantified boolean formulae
The renesse issue on satisfiabilityWe present a satisfiability tester QSAT for quantified Boolean formulae and a restriction QSATCNF of QSAT to unquantified conjunctive normal form formulae. QSAT makes use of procedures which replace subformulae of a formula by equivalent formulae. By a ...
Satisfiability of mixed Horn formulas
In this paper the class of mixed Horn formulas is introduced that contain a Horn part and a 2-CNF (conjunctive normal form) (also called quadratic) part. We show that SAT remains NP-complete for such instances and also that any CNF formula can be ...