Attribute-Specific Interchangeability in Constraint Satisfaction Problems

Constraint satisfaction problems consist of finding a value assignment to a set of variables such that all constraints on these variables are satisfied. In various application areas (e.g., engineering domains), domain values are actually used to denote complex objects, whose structure then must be derived by additional constraints. Greater effectiveness and a shorter problem description can be achieved by a component-oriented constraint formalism that allows explicit specification of the relationship between a component and its fixed attributes. To this goal, we introduce attribute functions into constraint networks. We define a variant of interchangeability suitable for these extended problems, based on computing domain partitions depending on attribute function values of the constraint variables, and give an algorithm for computing these partitions. By a slight modification of the key constraint filtering and search procedures, the domain partitions can be used to increase efficiency of the majority of existing algorithms for such problems.