Efficient implementations of DPLL with the addition of clause learning are the fastest complete Boolean satisfiability solvers and can handle many significant real-world problems, such as verification, planning and design. Despite its importance, little is known of the ultimate strengths and limitations of the technique. This paper presents the first precise characterization of clause learning as a proof system (CL), and begins the task of understanding its power by relating it to the well-studied resolution proof system. In particular, we show that with a new learning scheme, CL can provide exponentially shorter proofs than many proper refinements of general resolution (RES) satisfying a natural property. These include regular and Davis-Putnam resolution, which are already known to be much stronger than ordinary DPLL. We also show that a slight variant of CL with unlimited restarts is as powerful as RES itself. Translating these analytical results to practice, however, presents a challenge because of the nondeterministic nature of clause learning algorithms. We propose a novel way of exploiting the underlying problem structure, in the form of a high level problem description such as a graph or PDDL specification, to guide clause learning algorithms toward faster solutions. We show that this leads to exponential speed-ups on grid and randomized pebbling problems, as well as substantial improvements on certain ordering formulas. 1.

We analyse the possibility that a system that simulates Resolution is automatizable. We call this notion "weak automatizability". We prove

We analyze size and space complexity of Res(k), a family of propositional proof systems introduced by Kraj'icek in [21] which extends Resolution by allowing disjunctions of conjunctions of up to k 1 literals. We show that the treelike Res(k) proof systems form a strict hierarchy with respect to proof size and also with respect to space. Moreover

Abstract. This paper is concerned with the complexity of proofs and of searching for proofs in resolution. We show that the recently proposed algorithm of Ben-Sasson & Wigderson for searching for proofs in resolution cannot give better than weakly exponential performance. This is a consequence of our main result: we show the optimality of the general relationship called size-width tradeoff in Ben-Sasson & Wigderson. Moreover we obtain the optimality of the size-width tradeoff for the widely used restrictions of resolution: regular, Davis-Putnam, negative, positive.

A fruitful connection between algorithm design and proof complexity is the formalization of the ¤¦¥¨§© § approach to satisfiability testing in terms of tree-like resolution proofs. We consider extensions of the ¤¦¥¨§© § approach that add some version of memoization, remembering formulas the algorithm has previously shown unsatisfiable. Various versions of such formula caching algorithms have been suggested for satisfiability and stochastic satisfiability ([10, 1]). We formalize this method, and characterize the strength of various versions in terms of proof systems. These proof systems seem to be both new and simple, and have a rich structure. We compare their strength to several studied proof systems: tree-like resolution, regular resolution, general resolution, and ���������� �. We give both simulations and separations. 1

Efficient implementations of DPLL with the addition of clause learning are the fastest complete satisfiability solvers and can handle many significant real-world problems, such as verification, planning, and design. Despite its importance, little is known of the ultimate strengths and limitations of the technique. This paper presents the first precise characterization of clause learning as a proof system, and begins the task of understanding its power. In particular, we show that clause learning using any nonredundant scheme and unlimited restarts is equivalent to general resolution. We also show that without restarts but with a new learning scheme, clause learning can provide exponentially smaller proofs than regular resolution, which itself is known to be much stronger than ordinary DPLL. 1

Abstract. Propositional proof complexity is the study of the sizes of propositional proofs, and more generally, the resources necessary to certify propositional tautologies. Questions about proof sizes have connections with computational complexity, theories of arithmetic, and satisfiability algorithms. This is article includes a broad survey of the field, and a technical exposition of some recently developed techniques for proving lower bounds on proof sizes. Contents

Abstract. In 1997 we presented ten challenges for research on satisfiability testing [1]. In this paper we review recent progress towards each of these challenges, including our own work on the power of clause learning and randomized restart policies. 1

The width of a resolution proof is the maximal number of literals in any clause of the proof. The space of a proof is the maximal number of clauses kept in memory simultaneously if the proof is only allowed to infer new clauses from clauses currently in memory. Both of these measures have previously been studied and related to the resolution refutation size of unsatisfiable CNF formulas. Also, the refutation space of a formula has been proven to be at least as large as the refutation width, but it has been open whether space can be separated from width or the two measures coincide asymptotically. We prove that there is a family of k-CNF formulas for which the refutation width in resolution is constant but the refutation space is non-constant, thus solving a problem mentioned in several previous papers.

DPLL based clause learning algorithms for satisfiability testing are known to work very well in practice. However, like most branch-and-bound techniques, their performance depends heavily on the variable order used in making branching decisions. We propose a novel way of exploiting the underlying problem structure to guide clause learning algorithms toward faster solutions. The key idea is to use a higher level problem description, such as a graph or a PDDL specification, to generate a good branching sequence as an aid to SAT solvers.