Strong and Hiding Distributed Certification of $k$-Coloring

A locally checkable proof (LCP) is a non-deterministic distributed algorithm designed to verify global properties of a graph $G$. It involves two key components: a prover and a distributed verifier. The prover is an all-powerful computational entity capable of performing any Turing-computable operation instantaneously. Its role is to convince the distributed verifier -- composed of the graph's nodes -- that $G$ satisfies a particular property $\Pi$. We study the problem of certifying whether a graph is $k$-colorable with an LCP that is able to hide the $k$-coloring from the verifier. More precisely, we say an LCP for $k$-coloring is hiding if, in a yes-instance, it is possible to assign certificates to nodes without revealing an explicit $k$-coloring. Motivated by the search for promise-free separations of extensions of the LOCAL model in the context of locally checkable labeling (LCL) problems, we also require the LCPs to satisfy what we refer to as the strong soundness property. This is a strengthening of soundness that requires that, in a no-instance (i.e., a non-$k$-colorable graph) and for every certificate assignment, the subset of accepting nodes must induce a $k$-colorable subgraph. We focus on the case of $2$-coloring. We show that strong and hiding LCPs for $2$-coloring exist in specific graph classes and requiring only $O(\log n)$-sized certificates. Furthermore, when the input is promised to be a cycle or contains a node of degree $1$, we show the existence of strong and hiding LCPs even in an anonymous network and with constant-size certificates. Despite these upper bounds, we prove that there are no strong and hiding LCPs for $2$-coloring in general, regardless of certificate size. The proof relies on a Ramsey-type result as well as an intricate argument about the realizability of subgraphs of the neighborhood graph consisting of the accepting views of an LCP.