Constraint propagation as the core of local search

Constraint programming is a powerful paradigm for solving constraint satisfaction problems, using various techniques. Amongst them, local search is a prominent methodology, particularly for large instances. However, it lacks uniformity, as it includes many variations accompanied by complex data structures, that cannot be easily brought under the same "umbrella." In this work we embrace their wide diversity by adopting propagation algorithms. Our constraint based local search (CBLS) system provides declarative alternative tools to express search methods, by exploiting conflict-sets of constraints and variables. Their maintenance is straightforward as it does not employ queues, unlike the state of the art CBLS systems. Thus, the propagation complexity is kept linear in the number of changes required after each assignment. Experimental results illustrate the capabilities, not only of the already implemented methods, such as hill climbing, simulated annealing, etc., but also the robustness of the underlying propagation engine.