Computing and Sampling Restricted Vertex Degree Subgraphs and Hamiltonian Cycles

Abstract

Let $G=(V,E)$ be a bipartite graph embedded in a plane (or $n$-holed torus). Two subgraphs of $G$ differ by a {\it $Z$-transformation} if their symmetric difference consists of the boundary edges of a single face---and if each subgraph contains an alternating set of the edges of that face. For a given $ϕ: V \mapsto \mathbb Z^+$, $S_ϕ$ is the set of subgraphs of $G$ in which each $v\in V$ has degree $ϕ(v)$. Two elements of $S_ϕ$ are said to be adjacent if they differ by a $Z$-transformation. We determine the connected components of $S_ϕ$ and assign a {\it height function} to each of its elements. If $ϕ$ is identically two, and $G$ is a grid graph, $S_ϕ$ contains the partitions of the vertices of $G$ into cycles. We prove that we can always apply a series of $Z$-transformations to decrease the total number of cycles provided there is enough ``slack'' in the corresponding height function. This allows us to determine in polynomial time the minimal number of cycles into which $G$ can be partitioned provided $G$ has a limited number of non-square faces. In particular, we determine the Hamiltonicity of polyomino graphs in $O(|V|^2)$ steps. The algorithm extends to $n$-holed-torus-embedded graphs that have grid-like properties. We also provide Markov chains for sampling and approximately counting the Hamiltonian cycles of $G$.