Sharp Bounds on Lengths of Linear Recolouring Sequences

Abstract

A recolouring sequence, between $k$-colourings $\alpha$ and $\beta$ of a graph $G$, transforms $\alpha$ into $\beta$ by recolouring one vertex at a time, such that after each recolouring step we again have a proper $k$-colouring of $G$. The diameter of the $k$-recolouring graph, $\text{diam } \mathcal{C}_k(G)$, is the maximum over all pairs $\alpha$ and $\beta$ of the minimum length of a recolouring sequence from $\alpha$ to $\beta$. Much previous work has focused on determining the asymptotics of $\text{diam } \mathcal{C}_k(G)$: Is it $\Theta(|G|)$? Is it $\Theta(|G|^2)$? Or even larger? Here we focus on graphs for which $\text{diam } \mathcal{C}_k(G)=\Theta(|G|)$, and seek to determine more precisely the multiplicative constant implicit in the $\Theta()$. In particular, for each $k\ge 3$, for all positive integers $p$ and $q$ we exactly determine $\text{diam } \mathcal{C}_k(K_{p,q})$, up to a small additive constant. We also sharpen a recolouring lemma that has been used in multiple papers, proving an optimal version. This improves the multiplicative constant in various prior results. Finally, we investigate plausible relationships between similar reconfiguration graphs.