What makes normalized weighted satisfiability tractable

TL;DR

This paper characterizes the parameterized complexity of weighted satisfiability problems on normalized circuits of bounded depth, showing how genus constraints influence tractability and hardness.

Contribution

It provides a precise complexity classification of weighted satisfiability problems based on circuit genus and depth, revealing conditions for fixed-parameter tractability.

Findings

FPT for {\

W-hardness persists with high genus

Abstract

We consider the weighted antimonotone and the weighted monotone satisfiability problems on normalized circuits of depth at most $t \geq 2$, abbreviated {\sc wsat$^-[t]$} and {\sc wsat$^+[t]$}, respectively. These problems model the weighted satisfiability of antimonotone and monotone propositional formulas (including weighted anitmonoone/monotone {\sc cnf-sat}) in a natural way, and serve as the canonical problems in the definition of the parameterized complexity hierarchy. We characterize the parameterized complexity of {\sc wsat$^-[t]$} and {\sc wsat$^+[t]$} with respect to the genus of the circuit. For {\sc wsat$^-[t]$}, which is $W[t]$-complete for odd $t$ and $W[t-1]$-complete for even $t$, the characterization is precise: We show that {\sc wsat$^-[t]$} is fixed-parameter tractable (FPT) if the genus of the circuit is $n^{o(1)}$ ($n$ is the number of the variables in the circuit), and that it has the same $W$-hardness as the general {\sc wsat$^-[t]$} problem (i.e., with no restriction on the genus) if the genus is $n^{O(1)}$. For {\sc wsat$^+[2]$} (i.e., weighted monotone {\sc cnf-sat}), which is $W[2]$-complete, the characterization is also precise: We show that {\sc wsat$^+[2]$} is FPT if the genus is $n^{o(1)}$ and $W[2]$-complete if the genus is $n^{O(1)}$. For {\sc wsat$^+[t]$} where $t > 2$, which is $W[t]$-complete for even $t$ and $W[t-1]$-complete for odd $t$, we show that it is FPT if the genus is $O(\sqrt{\log{n}})$, and that it has the same $W$-hardness as the general {\sc wsat$^+[t]$} problem if the genus is $n^{O(1)}$.