Title:All or nothing: toward a promise problem dichotomy for constraint problems

Abstract:A finite constraint language $\mathscr{R}$ is a finite set of relations over some finite domain $A$. We show that intractability of the constraint satisfaction problem $\mathsf{CSP}(\mathscr{R})$ for $\mathscr{R}$ can, in all known cases, be replaced by an infinite hierarchy of intractable promise problems of increasingly disparate promise conditions: where instances are guaranteed to be either have no solutions, or are $k$-robustly satisfiable (for any fixed $k$), meaning that every "reasonable" partial instantiation on $k$ variables extends to a solution. For example, subject to the assumption $\texttt{P}\neq \texttt{NP}$, then for any $k$, we show that there is no polynomial time algorithm that can distinguish non-$3$-colorable graphs, from those for which any reasonable $3$-coloring of any $k$ of the vertices can extend to a full $3$-coloring.
The enormous flexibility granted by $k$-robust satisfiability enables the complexity of a large number of other computational problems for $\mathscr{R}$ to be tied precisely to that of $\mathsf{CSP}(\mathscr{R})$. A key example is the problem of deciding if an instance has \emph{implied constraints}: constraints not given explicitly, but which are preserved under any solution. Our results show that such instances cannot be efficiently distinguished from those with no solutions at all. Our result holds wherever the algebraic dichotomy conjecture has been shown to hold. In particular they are true for list homomorphism problems, for graphs, and for constraint languages on small domains.