Identifying Efficiently Solvable Cases of Max CSP

In this paper we study the complexity of the maximum constraint satisfaction problem (Max CSP) over an arbitrary finite domain. We describe a novel connection between this problem and the supermodular function maximization problem (which is dual to the submodular function minimization problem). Using this connection, we are able to identify large classes of efficiently solvable subproblems of Max CSP arising from certain restrictions on the constraint types. Until now, the only known polynomial-time solvable cases for this form of optimization problem were restricted to constraints over a 2-valued (Boolean) domain. Here we obtain the first examples of general families of efficiently solvable cases of Max CSP for arbitrary finite domains, by considering supermodular functions on finite lattices. Finally, we show that the equality constraint over a non-Boolean domain is non-supermodular, and, when combined with some simple unary constraints, gives rise to cases of Max CSP which are hard even to approximate.