Shortest Paths, Convexity, and Treewidth in Regular Hyperbolic Tilings
Sándor Kisfaludi-Bak, Tze-Yang Poon, Geert van Wordragen
TL;DR
This work studies shortest paths in regular hyperbolic tilings $G_{p,q}$ with $\frac{1}{p}+\frac{1}{q}<\frac{1}{2}$, introducing geodesic convex hulls and minimal isometric closures of terminal sets. It develops line-based, boundary-driven techniques to compute shortest paths and isometric closures in near-linear time, showing the convex hull has $\mathcal{O}(N)$ vertices and a treewidth bounded by $\max\{12, \mathcal{O}(\log\frac{n}{p+q})\}$. The paper then proves that optimal Steiner trees and Subset TSP tours can be realized within a minimal isometric closure and provides an $\mathcal{O}(N \log N)$-time algorithmic framework, with an additional polynomial dependence on $n/(p+q)$. These results culminate in near-linear algorithms for key NP-hard problems on hyperbolic tilings by leveraging small treewidth and outerplanarity properties, offering significant improvements over the planar case. Overall, the work advances efficient geometric and graph-theoretic methods on hyperbolic graphs and highlights the algorithmic potential of hyperbolic geometry in combinatorial optimization.
Abstract
Hyperbolic tilings are natural infinite planar graphs where each vertex has degree $q$ and each face has $p$ edges for some $\frac1p+\frac1q<\frac12$. We study the structure of shortest paths in such graphs. We show that given a set of $n$ terminals, we can compute a so-called isometric closure (closely related to the geodesic convex hull) of the terminals in near-linear time, using a classic geometric convex hull algorithm as a black box. We show that the size of the convex hull is $O(N)$ where $N$ is the total length of the paths to the terminals from a fixed origin. Furthermore, we prove that the geodesic convex hull of a set of $n$ terminals has treewidth only $\max(12,O(\log\frac{n}{p + q}))$, a bound independent of the distance of the points involved. As a consequence, we obtain algorithms for subset TSP and Steiner tree with running time $O(N \log N) + \mathrm{poly}(\frac{n}{p + q}) \cdot N$.
