Tractable Fragments of Presburger Arithmetic

Abstract

In this paper we introduce a problem called Quantified Integer Programming, which generalizes the Quantified Satisfiability problem (QSAT). In a Quantified Integer Program (QIP) the program variables can assume arbitrary integral values, as opposed to the boolean values that are assumed by the variables of an instance of QSAT. QIPs naturally represent two-person integer matrix games. The Quantified Integer Programming problem is PSPACE-hard in general, since the QSAT problem is PSPACE-complete. Quantified Integer Programming can be thought of as a restriction of Presburger Arithmetic, in that we allow only conjunctions of linear inequalities. We focus on analyzing various special cases of the general problem, with a view to discovering subclasses that are tractable. Subclasses of the general QIP problem are obtained by restricting either the constraint matrix or quantifier specification or both. We show that if the constraint matrix is totally unimodular, the problem of deciding a QIP can be solved in polynomial time. We also establish the computational complexities of Oblivious strategy games and Clairvoyant strategy games.