Fast algorithms for Vizing's theorem on bounded degree graphs

TL;DR

This paper presents a linear-time algorithm for proper edge-coloring of graphs with bounded degree, improving previous algorithms and extending to distributed models with efficient deterministic and randomized solutions.

Contribution

It introduces the first linear-time algorithm for $( ext{max degree}+1)$-edge-coloring when degree is constant, and develops new distributed algorithms using entropy compression.

Findings

Linear-time $( ext{max degree}+1)$-edge-coloring algorithm for constant degree graphs.

Deterministic $ ilde{O}( ext{log}^5 n)$ and randomized $O( ext{log}^2 n)$ distributed algorithms.

Applicable for degrees up to $ ext{log}^{o(1)} n$ with polynomial dependence on $ ext{max degree}$.

Abstract

Vizing's theorem states that every graph $G$ of maximum degree $\Delta$ can be properly edge-colored using $\Delta + 1$ colors. The fastest currently known $(\Delta+1)$-edge-coloring algorithm for general graphs is due to Sinnamon and runs in time $O(m\sqrt{n})$, where $n :=|V(G)|$ and $m :=|E(G)|$. We investigate the case when $\Delta$ is constant, i.e., $\Delta = O(1)$. In this regime, the runtime of Sinnamon's algorithm is $O(n^{3/2})$, which can be improved to $O(n \log n)$, as shown by Gabow, Nishizeki, Kariv, Leven, and Terada. Here we give an algorithm whose running time is only $O(n)$, which is obviously best possible. Prior to this work, no linear-time $(\Delta+1)$-edge-coloring algorithm was known for any $\Delta \geq 4$. Using some of the same ideas, we also develop new algorithms for $(\Delta+1)$-edge-coloring in the $\mathsf{LOCAL}$ model of distributed computation. Namely, when $\Delta$ is constant, we design a deterministic $\mathsf{LOCAL}$ algorithm with running time $\tilde{O}(\log^5 n)$ and a randomized $\mathsf{LOCAL}$ algorithm with running time $O(\log ^2 n)$. Although our focus is on the constant $\Delta$ regime, our results remain interesting for $\Delta$ up to $\log^{o(1)} n$, since the dependence of their running time on $\Delta$ is polynomial. The key new ingredient in our algorithms is a novel application of the entropy compression method.