Core Concepts
The complexity of the Quantified Constraint Satisfaction Problem (QCSP) over any finite constraint language is either in ΠP2 or PSpace-complete.
Abstract
The paper presents a complexity dichotomy for the Quantified Constraint Satisfaction Problem (QCSP) over finite constraint languages. The key findings are:
If the QCSP over a constraint language Γ is not PSpace-hard, then for any No-instance of QCSP(Γ), the Universal Player (UP) has a winning strategy on a polynomial-size set of tuples. This implies that QCSP(Γ) is in the complexity class ΠP2.
There exists a constraint language Γ on a 6-element domain such that QCSP(Γ) is ΠP2-complete, showing that the gap between PSpace and ΠP2 cannot be enlarged.
The paper provides a characterization of all PSpace-hard constraint languages in terms of "mighty tuples" - certain relations that can be q-defined over the constraint language and lead to PSpace-hardness.
The proof techniques involve reducing the QCSP to an exponential-size CSP instance and analyzing the properties required for the UP to have a polynomial-size winning strategy.