Proof Compression and NP Versus PSPACE II

We upgrade [3] to a complete proof of the conjecture NP = PSPACE that is known as one of the fundamental open problems in the mathematical theory of computational complexity; this proof is based on [2]. Since minimal propositional logic is known to be PSPACE complete, while PSPACE to include NP, it suffices to show that every valid purely implicational formula ρ has a proof whose weight (= total number of symbols) and time complexity of the provability involved are both polynomial in the weight of ρ. As in [3], we use proof theoretic approach. Recall that in [3] we considered any valid ρ in question that had (by the definition of validity) a "short" tree-like proof π in the Hudelmaier-style cutfree sequent calculus for minimal logic. The "shortness" means that the height of π and the total weight of different formulas occurring in it are both polynomial in the weight of ρ. However, the size (= total number of nodes), and hence also the weight, of π could be exponential in that of ρ. To overcome this trouble we embedded π into Prawitz's proof system of natural deductions containing single formulas, instead of sequents. As in π, the height and the total weight of different formulas of the resulting tree-like natural deduction ∂1 were polynomial, although the size of ∂1 still could be exponential, in the weight of ρ. In our next, crucial move, ∂1 was deterministically compressed into a "small", although multipremise, dag-like deduction ∂ whose horizontal levels contained only mutually different formulas, which made the whole weight polynomial in that of ρ. However, ∂ required a more complicated verification of the underlying provability of ρ. In this paper we present a nondeterministic compression of ∂ into a desired standard dag-like deduction ∂0 that deterministically proves ρ in time and space polynomial in the weight of ρ. Together with [3] this completes the proof of NP = PSPACE. Natural deductions are essential for our proof. Tree-to-dag horizontal compression of π merging equal sequents, instead of formulas, is (possible but) not sufficient, since the total number of different sequents in π might be exponential in the weight of ρ − even assuming that all formulas occurring in sequents are subformulas of ρ. On the other hand, we need Hudelmaier's cutfree sequent calculus in order to control both the height and total weight of different formulas of the initial tree-like proof π, since standard Prawitz's normalization although providing natural deductions with the subformula property does not preserve polynomial heights. It is not clear yet if we can omit references to π even in the proof of the weaker result NP = coNP.

Cuts for circular proofs

One of the authors introduced in [1] a calculus ofcircular proofs for studying the computability arising from thefollowing categorical operations: finite products and coproducts,initial algebras, final coalgebras. The calculus of[1] is cut-free; yet, even if sound and complete forprovability, it lacks an important property for the semantics ofproofs, namely fullness w.r.t. the class of natural categorical modelscalled μ-bicomplete category in [2].We fix, with this work, this problem by adding the cut rule to thecalculus. To this goal, we need to modifying the syntacticalconstraints on the cycles of proofs so to ensure soundness of thecalculus and at same time local termination of cut-elimination. Theenhanced proof system fully represents arrows of the intended model, afree μ-bicomplete category. We also describe a cut-eliminationprocedure as a model of computation arising from the above mentionedcategorical operations. The procedure constructs a cut-freeproof-tree with infinite branches out of a finite circular proof withcuts.[1] Luigi Santocanale. A calculus of circular proofs and its categorical semantics. In Mogens Nielsen and Uffe Engberg, editors, FoSSaCS, volume 2303 of Lecture Notes in Computer Science, pages 357–371. Springer, 2002.[2] Luigi Santocanale. μ-bicomplete categories and parity games. Theoretical Informatics and Applications, 36:195–227, September 2002.

Linear Logic as a Tool for Qualitative and Quantitative Analysis of Workow Processes

This article presents a method for qualitative and quantitative analysis of WorkFlow nets based on the proof trees of Linear Logic. The qualitative analysis is concerned with the proof of Soundness correctness criterion defined for WorkFlow nets. To prove the Soundness property, a proof tree of Linear Logic is built for each different scenario of the WorkFlow net. The quantitative analysis is concerned with the resource planning for each task of the workflow process and is based on the computation of symbolic date intervals for task execution. In particular, such symbolic date intervals are computed using the proof trees used to prove Soundness property.

New Inference Rules for Max-SAT

Exact Max-SAT solvers, compared with SAT solvers, apply little inference at each node of the proof tree. Commonly used SAT inference rules like unit propagation produce a simplified formula that preserves satisfiability but, unfortunately, solving the Max-SAT problem for the simplified formula is not equivalent to solving it for the original formula. In this paper, we define a number of original inference rules that, besides being applied efficiently, transform Max-SAT instances into equivalent Max-SAT instances which are easier to solve. The soundness of the rules, that can be seen as refinements of unit resolution adapted to Max-SAT, are proved in a novel and simple way via an integer programming transformation. With the aim of finding out how powerful the inference rules are in practice, we have developed a new Max-SAT solver, called MaxSatz, which incorporates those rules, and performed an experimental investigation. The results provide empirical evidence that MaxSatz is very competitive, at least, on random Max-2SAT, random Max-3SAT, Max-Cut, and Graph 3-coloring instances, as well as on the benchmarks from the Max-SAT Evaluation 2006.