Constraint satisfaction problems (CSPs) provide a flexible and powerful framework for modeling and solving many decision problems of practical importance. Consistency properties and the algorithms for enforcing them on a problem instance are at the heart of Constraint Processing and best distinguish this area from other areas concerned with the same combinatorial problems.
In this thesis, we study path consistency (PC) and investigate several algorithms for enforcing it on binary finite CSPs. We also study algorithms for enforcing consistency properties that are related to PC but are stronger or weaker than PC.
We identify and correct errors in the literature and settle an open question. We propose two improvements that we apply to the well-known algorithms PC-8 and PC-2001, yielding PC-8+ and PC-2001+. Further, we propose a new algorithm for enforcing partial path consistency, σ-∆-PPC, which generalizes features of the well-known algorithms DPC and PPC. We evaluate over fifteen different algorithms on both benchmark and randomly generated binary problems to empirically demonstrate the effectiveness of our approach.
Source: University of Nebraska
Authors: Christopher G. Reeson