Project Details
Projekt Print View

Complexity measures for solving propositional formulas.

Subject Area Theoretical Computer Science
Term from 2019 to 2024
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 430150230
 
Final Report Year 2024

Final Report Abstract

The satisfiability problem for propositional logic, SAT, is the best-known NP-complete problem and has traditionally been considered very hard to solve. In the last decades, however, there has been an impressive advance in the development of programs that solve SAT instances. Today, these programs solve real-world applications with literally hundreds of thousands of variables. Proof complexity provides the tools necessary to analyze SAT solvers from a theoretical perspective. The goal of the project was to obtain a better understanding of this gap between theory and practice by focusing on how several complexity measures for the proof systems at the root of modern SAT solvers (mainly Resolution) translate into estimations for running times and search strategies for the SAT programs. We have considered several central families of contradictory formulas, mainly pebbling and graph isomorphism formulas, and analyzed different complexity parameters for these formulas for their refutation in some well-known proof systems: Resolution, Cutting Planes, and algebraic proof systems. We found that combinatorial measures on the underlying graphs described in the formulas directly translate into complexity parameters of the refutations in the different systems. This allows to obtain an estimation of the complexity of the refutations directly from the graphs, without going into the concrete refutations. We have also analyzed several relationships and trade-offs between the different complexity measures and gained a better understanding of the formulas for which SAT solvers perform well. On a more practical side, we have also investigated stochastic local search solvers (SLS) and analyzed their performance using space measures for Resolution. For this, we have implemented an SLS solver and several software tools that allowed experimental performance results.

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung