Using a Subsumption-Based Taxonomy to Construct Networks of Cooperating Decision Procedures

Costly general decision procedures can be sped up by incorporating eecient decision procedures for restricted classes of constraints. We show how to make use of these specialized decision procedures by constructing a network of cooperating decision procedures. Of course, in order to gain eeciency the expensive general decision procedures should not be called as far as possible. But they should always agree with the answers produced by the specialized decision procedures. Another problem is that each specialized decision procedure can handle only a certain class of constraints and hence the decision made by any one specialized decision procedure is based on a subset of all the constraints actually there. Nevertheless this local decision making can be used in short cutting the decision making process. This is accomplished by deening (i) a network topology, and (ii) a constraint distribution scheme, such that there is guaranteed to be consensus amongst the decision procedures. In this way when a specialized decision procedure decides local consistency (or inconsistency) then indeed we have global consistency (or inconsistency).