Improved Bounds for Perfect Sampling of $k$-Colorings in Graphs

TL;DR

This paper introduces a new randomized algorithm for perfectly sampling proper k-colorings of graphs with improved bounds, reducing the number of colors needed for efficient, exact sampling compared to previous methods.

Contribution

The paper presents a novel bounding chain approach that achieves polynomial expected running time for perfect sampling of k-colorings with fewer colors than prior algorithms.

Findings

Algorithm works for k > 3Δ, improving previous bounds.

Expected running time is polynomial in n and k.

Achieves perfect uniform sampling of proper colorings.

Abstract

We present a randomized algorithm that takes as input an undirected $n$-vertex graph $G$ with maximum degree $\Delta$ and an integer $k > 3\Delta$, and returns a random proper $k$-coloring of $G$. The distribution of the coloring is \emph{perfectly} uniform over the set of all proper $k$-colorings; the expected running time of the algorithm is $\mathrm{poly}(k,n)=\widetilde{O}(n\Delta^2\cdot \log(k))$. This improves upon a result of Huber~(STOC 1998) who obtained a polynomial time perfect sampling algorithm for $k>\Delta^2+2\Delta$. Prior to our work, no algorithm with expected running time $\mathrm{poly}(k,n)$ was known to guarantee perfectly sampling with sub-quadratic number of colors in general. Our algorithm (like several other perfect sampling algorithms including Huber's) is based on the Coupling from the Past method. Inspired by the \emph{bounding chain} approach, pioneered independently by Huber~(STOC 1998) and H\"aggstr\"om \& Nelander~(Scand.{} J.{} Statist., 1999), we employ a novel bounding chain to derive our result for the graph coloring problem.