New Diameter-Reducing Shortcuts and Directed Hopsets: Breaking the $\sqrt{n}$ Barrier

TL;DR

This paper introduces new algorithms for adding a small number of edges to directed graphs to significantly reduce their diameter, breaking previous barriers and improving efficiency in graph shortcutting.

Contribution

It presents the first algorithms to reduce directed graph diameters below the  barrier with near-linear shortcut edges, advancing the understanding of diameter-sparsity tradeoffs.

Findings

Reduced diameter to (n^{1/3}) with linear shortcut set.

Achieved diameter (n^{1/2}) with sublinear shortcut edges.

Provided bounds on shortcut set size for various diameters.

Abstract

For an $n$-vertex digraph $G=(V,E)$, a \emph{shortcut set} is a (small) subset of edges $H$ taken from the transitive closure of $G$ that, when added to $G$ guarantees that the diameter of $G \cup H$ is small. Shortcut sets, introduced by Thorup in 1993, have a wide range of applications in algorithm design, especially in the context of parallel, distributed and dynamic computation on directed graphs. A folklore result in this context shows that every $n$-vertex digraph admits a shortcut set of linear size (i.e., of $O(n)$ edges) that reduces the diameter to $\widetilde{O}(\sqrt{n})$. Despite extensive research over the years, the question of whether one can reduce the diameter to $o(\sqrt{n})$ with $\widetilde{O}(n)$ shortcut edges has been left open. We provide the first improved diameter-sparsity tradeoff for this problem, breaking the $\sqrt{n}$ diameter barrier. Specifically, we show an $O(n^{\omega})$-time randomized algorithm for computing a linear shortcut set that reduces the diameter of the digraph to $\widetilde{O}(n^{1/3})$. This narrows the gap w.r.t the current diameter lower bound of $\Omega(n^{1/6})$ by [Huang and Pettie, SWAT'18]. Moreover, we show that a diameter of $\widetilde{O}(n^{1/2})$ can in fact be achieved with a \emph{sublinear} number of $O(n^{3/4})$ shortcut edges. Formally, letting $S(n,D)$ be the bound on the size of the shortcut set required in order to reduce the diameter of any $n$-vertex digraph to at most $D$, our algorithms yield: \[ S(n,D)=\begin{cases} \widetilde{O}(n^2/D^3), & \text{for~} D\leq n^{1/3},\\ \widetilde{O}((n/D)^{3/2}), & \text{for~} D> n^{1/3}~. \end{cases} \] We also extend our algorithms to provide improved $(\beta,\epsilon)$ hopsets for $n$-vertex weighted directed graphs.