Approximating 1-in-3 SAT by linearly ordered hypergraph 3-colouring is NP-hard

Abstract: Given a satisfiable instance of 1-in-3 SAT, it is NP-hard to find a satisfying assignment for it, but it may be possible to efficiently find a solution subject to a weaker (not necessarily Boolean) predicate than `1-in-3'. There is a folklore conjecture predicting which choices of weaker predicates lead to tractability and for which the task remains \NP-hard. One specific predicate, corresponding to the problem of linearly ordered $3$-colouring of 3-uniform hypergraphs, has been mentioned in several papers as an obstacle to further progress in proving this conjecture. We prove that the problem for this predicate is NP-hard, as predicted by the conjecture. We use the Promise CSP framework, where the complexity analysis is performed via the algebraic approach, by studying the structure of polymorphisms, which are multidimensional invariants of the problem at hand. The analysis of polymorphisms is in general a highly non-trivial task, and topological combinatorics was recently discovered to provide a useful tool for this. There are two distinct ways in which it was used: one is based on variations of the Borsuk-Ulam theorem, and the other aims to classify polymorphisms up to certain reconfigurations (homotopy). Our proof, whilst combinatorial in nature, shows that our problem is the first example where the features behind the two uses of topology appear together. Thus, it is likely to be useful in guiding further development of the topological method aimed at classifying Promise CSPs. An easy consequence of our result is the hardness of another specific Promise CSP, which was recently proved by Filakovsk\'y et al. by employing a deep topological analysis of polymorphisms.