Computer Science and Engineering, Department of


Date of this Version



This article was published in AI Communications, The European Journal on Artificial Intelligence. 17.4 (2004)


Dechter et al. [5] proposed solving the Temporal Constraint Satisfaction Problem (TCSP) by modeling it as a meta-CSP, which is a finite CSP with a unique global constraint. The size of this global constraint is exponential in the number of time points in the original TCSP, and generalized-arc consistency is equivalent to finding the minimal network of the TCSP, which is NP-hard. We introduce _AC, an efficient consistency algorithm for filtering the meta-CSP. This algorithm significantly reduces the domains of the variables of the meta-CSP without guaranteeing arc-consistency. We use _AC as a preprocessing step to solving the meta-CSP. We show experimentally that it dramatically reduces the size of a meta-CSP and significantly enhances the performance of search for finding the minimal network of the corresponding TCSP