Closures and dichotomies for quantified constraints

T. Feder,Phokion G. Kolaitis

Abstract:Quantified constraint satisfaction is the generalization of con- straint satisfaction that allows for both universal and existential quantifiers over constrained variables, instead of just existential quantifiers. We study quantified constraint satisfaction problems CSP(Q,S), where Q denotes a pattern of quantifier alternation ending in exists or the set of all possible alternations of quantifiers, and S is a set of relations constraining the com- binations of values that the variables may take. These problems belong to the corresponding level of the polynomial hierarchy or in PSPACE, depend- ing on whether Q is a fixed pattern of quantifier alternation or the set of all possible alternations of quantifiers. We also introduce and study the quan- tified constraint satisfaction problems CSP 0 (Q,S) in which the universally quantified variables are restricted to range over given subsets of the domain. We first show that CSP(Q,S) and CSP 0 (Q,S) are polynomial-time equiv- alent to the problem of evaluating certain syntactically restricted monadic second-order formulas on finite structures. After this, we establish three broad sufficient conditions for polynomial-time solvability of CSP 0 (Q,S) that are based on closure functions; these results generalize and extend earlier results by other researchers about polynomial-time solvability of CSP(Q,S). Our study culminates with a dichotomy theorem for the com- plexity of list CSP 0 (Q,S), that is, CSP 0 (Q,S) where the relations of S include every subset of the domain of S. Specifically, list CSP 0 (Q,S) is ei- ther solvable in polynomial-time or complete for the corresponding level of the polynomial hierarchy, if Q is a fixed pattern of quantifier alternation (or PSPACE-complete if Q is the set of all possible alternations of quantifiers). The proofs are based on a more general unique sink property formulation.