ion of a given partial assignment of values to variables. Compared with other symmetry breaking techniques, the big advantage of dynamic symmetry breaking is that it can accommodate dynamic variable and value orderings. Dynamic orderings have been shown to be vastly superior to static orderings in many different types of constraint satisfaction problems. However, robust heuristics for the selection of variables and values are hard to come by. For the task of variable selection, a bias towards variables with smaller domains often works comparably well, but there always remains a fair probability that we hit instances on which a solver gets trapped in extremely long runs. Particularly, heavy-tailed runtime distributions have been reported [9]. One way to circumvent this problematic situation is to randomize the solver and to restart the search when a run takes too long [10]. We show how symmetry nogoods can be used in restarted methods and introduce practical enhancements of SSB for its ...
Daniel S. Heller, Meinolf Sellmann