Global Search Algorithms and the Constraint Logic Programming Framework

Abstract

Global search is an improvement of the “generate & test“ paradigm, which is particularly efficient for solving combinatorial problems. This paradigm was studied in the functional programming framework in [Smith88], and in the logic programming framework in [Schaeck91].

The constraint logic programming (CLP) framework is also a very interesting tool for designing programs solving combinatorial problems. A method for designing CLP programs, based on a general program schema, is studied in [Deville-Van Hentenryck90].

In our research, we first introduce a particular version of the schema for global search algorithms. The interest of this version is to explicitely separate the generation of candidate solutions and their feasibility test. We also show that this particular schema is indeed totally equivalent to the one presented in our earlier work (see [Schaeck91]).

We then show that each concept used in global search algorithms finds its counterpart in the schema used for constructing CLP programs or in the mechanisms of constraint logic programming languages. This convinces us that a CLP program can be considered as an instance of a global search algorithm, even if it does not have such an appearance.

We then propose a new schema for the design of CLP programs, which uses the full power of the global search paradigm. The main interest of this new schema is that the decomposition into subproblems is now justified by a semantical argument, and no more by a purely syntactical one.

This work is sponsored by a research grant of the Belgian National Fund for Scientific Research.