Distributed splitting of constraint satisfaction problems

Constraint propagation aims to reduce a constraint satisfaction problem into an equivalent but simpler one. However, constraint propagation must be interleaved with a splitting mechanism in order to compose a complete solver. In [13] a framework for constraint propagation based on a control-driven coordination model was presented. In this paper we extend this framework in order to integrate a distributed splitting mechanism. This technique has three main advantages: 1) in a single distributed and generic framework, propagation and splitting can be interleaved in order to realize complete distributed solvers, 2) by changing only one agent, we can perform different kinds of search, and 3) splitting of variables can be dynamically triggered before the fixed point of a propagation is reached.