Declarative modelling of constraint propagation strategies

Abstract

Constraint propagation is a generic method for combining monotone operators over lattices. In this paper, a family of constraint propagation strategies are designed as instances of a single algorithm shown to be correct, finite, and strategy-independent. The main idea is to separate complex reasoning processes (that exploit some knowledge of constraints and solvers in order to accelerate the convergence) from basic ffxed-point algorithms. Some sequential interval narrowing strategies are expressed in this framework and parallel implementations in different execution models are discussed.