The inverse satisfiability problem over a set of Boolean relations Gamma (Inv-SAT(Gamma)) is the computational decision problem of, given a set of models R, deciding whether there exists a SAT(Gamma) instance with R as its set of models. This problem is co-NP-complete in general and a dichotomy theorem for finite Γ containing the constant Boolean relations was obtained by Kavvadias and Sideri. In this paper we remove the latter condition and prove that Inv-SAT(Gamma) is always either tractable or co-NP-complete for all finite sets of relations Gamma, thus solving a problem open since 1998. Very little of the techniques used by Kavvadias and Sideri are applicable and we have to turn to more recently developed algebraic approaches based on partial polymorphisms. We also consider the case when Γ is infinite, where the situation differs markedly from the case of SAT. More precisely, we show that there exists infinite Gamma such that Inv-SAT(Gamma) is tractable even though there exists finite Delta is subset of Gamma such that Inv-SAT(Delta) is co-NP-complete.

The parameterized satisfiability problem over a set of Boolean relations Gamma (SAT(Gamma)) is the problem of determining whether a conjunctive formula over Gamma has at least one model. Due to Schaefer's dichotomy theorem the computational complexity of SAT(Gamma), modulo polynomial-time reductions, has been completely determined: SAT(Gamma) is always either tractable or NP-complete. More recently, the problem of studying the relationship between the complexity of the NP-complete cases of SAT(Gamma) with restricted notions of reductions has attracted attention. For example, Impagliazzo et al. studied the complexity of k-SAT and proved that the worst-case time complexity increases infinitely often for larger values of k, unless 3-SAT is solvable in subexponential time. In a similar line of research Jonsson et al. studied the complexity of SAT(Gamma) with algebraic tools borrowed from clone theory and proved that there exists an NP-complete problem SAT(R^{neq,neq,neq,01}_{1/3}) such that there cannot exist any NP-complete SAT(Gamma) problem with strictly lower worst-case time complexity: the easiest NP-complete SAT(Gamma) problem. In this paper we are interested in classifying the NP-complete SAT(Gamma) problems whose worst-case time complexity is lower than 1-in-3-SAT but higher than the easiest problem SAT(R^{neq,neq,neq,01}_{1/3}). Recently it was conjectured that there only exists three satisfiability problems of this form. We prove that this conjecture does not hold and that there is an infinite number of such SAT(Gamma) problems. In the process we determine several algebraic properties of 1-in-3-SAT and related problems, which could be of independent interest for constructing exponential-time algorithms.