Abstract
Though there exist some rules of thumb for design of good models for solving constraint satisfaction problems, the modeling process still belongs more to art than to science. Moreover, as new global constraints and search techniques are being developed, the modeling process is becoming even more complicated and a lot of effort and experience is required from the user.
Hence (semi-) automated tools for improving efficiency of constraint models are highly desirable. The paper presents a low-information technique for discovering implied Boolean constraints in the form of equivalences, exclusions, and dependencies for any constraint model with (some) Boolean variables.
The technique is not only completely independent of the constraint model (therefore a low-information technique), but it is also easy to implement because it is based on ideas of singleton consistency. Despite its simplicity, the proposed technique proved itself to be surprisingly efficient in our experiments.