Hierarchical arc consistency: exploiting structured domains in constraint satisfaction problems

Abstract

Constraint satisfaction problems can be solved by network consistency algorithms that eliminate local inconsistencies before constructing global solutions. We describe a new algorithm that is useful when the variable domains can be structured hierarchically into recursive subsets with common properties and common relationships to subsets of the domain values for related variables. The algorithm, HAC, uses a technique known as hierarchical arc consistency. Its performance is analyzed theoretically and the conditions under which it is an improvement are outlined. The use of HAC in a program for understanding sketch maps, Mapsee3, is briefly discussed and experimental results consistent with the theory are reported.