Using Interval Arithmetic To Model Finite Domain CSPs Where Domain Generation Is Expensive

It is generally assumed that the variables, domains and constraints of a Finite-Domain Constraint Satisfaction Problem are all pre-computed inputs to a black-box constraint satisfaction algorithm. The obvious advantages of such an assumption is the freedom in developing generic constraint solvers and the declarative use of constraint technology. However, it is useful to examine applications of CSPs to evaluate if they can beneet further from on-they constraint and domain speciication. We consider particular classes of numerically computable discrete CSPs for which the variables' domain generation is costly. The domain generation and its cost are usually application dependent. However , in several numeric discrete CSPs, some general measures of cost can be estimated. We propose to reduce the generation cost of the discrete nite domains by combining interval narrowing techniques with local consistency techniques in discrete domains. We present a new algorithm HetCSP for the above task and illustrate its use with an example and several experimental results .