Network Satisfaction Problems Solved by k-Consistency
Manuel Bodirsky, Simon Knäuer
TL;DR
This work addresses whether the network satisfaction problem for finite relation algebras can be solved by the $k$-consistency procedure. It proves undecidability in general, while giving a practical, polynomial-time criterion for a large, important subclass: finite symmetric relation algebras with a normal representation and a flexible atom, via a Siggers-type condition and a binarisation technique that leverages finite binary conservative CSP results. The authors connect NSP to CSP through normal representations, deploy universal-algebraic tools (polymorphisms, atom structures, conservative clones), and derive a concrete tractability boundary: NSP is solvable by $(4,6)$-consistency in the tractable class and NP-complete otherwise. These results yield both a robust negative result about the meta-decidability of NSP and a sharp positive dichotomy for a broad, practically relevant subclass, with implications for qualitative spatial-temporal reasoning and related CSP frameworks.
Abstract
We show that the problem of deciding for a given finite relation algebra A whether the network satisfaction problem for A can be solved by the k-consistency procedure, for some natural number k, is undecidable. For the important class of finite relation algebras A with a normal representation, however, the decidability of this problem remains open. We show that if A is symmetric and has a flexible atom, then the question whether NSP(A) can be solved by k-consistency, for some natural number k, is decidable (even in polynomial time in the number of atoms of A). This result follows from a more general sufficient condition for the correctness of the k-consistency procedure for finite symmetric relation algebras. In our proof we make use of a result of Alexandr Kazda about finite binary conservative structures.
