Going Beyond Surfaces in Diameter Approximation

TL;DR

This paper develops new deterministic algorithms for approximating the diameter of various graph classes beyond planar graphs, using VC set systems and local treewidth, achieving efficient runtimes and extending previous methods.

Contribution

It introduces diameter approximation algorithms for apex-minor-free graphs using VC set systems, surpassing embedding-based techniques and extending to broader graph classes.

Findings

Achieved $(1+\varepsilon)$-approximation algorithms with polylogarithmic factors for apex-minor-free graphs.

Designed efficient approximate distance oracles for apex-minor-free graphs with improved preprocessing and query times.

Extended diameter approximation techniques beyond bounded genus graphs using local treewidth and VC set systems.

Abstract

Calculating the diameter of an undirected graph requires quadratic running time under the Strong Exponential Time Hypothesis and this barrier works even against any approximation better than 3/2. For planar graphs with positive edge weights, there are known $(1+\varepsilon)$-approximation algorithms with running time $poly(1/\epsilon, \log n) \cdot n$. However, these algorithms rely on shortest path separators and this technique falls short to yield efficient algorithms beyond graphs of bounded genus. In this work we depart from embedding-based arguments and obtain diameter approximations relying on VC set systems and the local treewidth property. We present two orthogonal extensions of the planar case by giving $(1+\varepsilon)$-approximation algorithms with the following running times: 1. $O_h((1/\varepsilon)^{O(h)} \cdot n \log^2 n)$-time algorithm for graphs excluding an apex graph of size h as a minor, 2. $O_d((1/\varepsilon)^{O(d)} \cdot n \log^2 n)$-time algorithm for the class of d-apex graphs. As a stepping stone, we obtain efficient (1+\varepsilon)-approximate distance oracles for graphs excluding an apex graph of size h as a minor. Our oracle has preprocessing time $O_h((1/\varepsilon)^8\cdot n \log n \log W)$ and query time $O((1/\varepsilon)^2 * \log n \log W)$, where $W$ is the metric stretch. Such oracles have been so far only known for bounded genus graphs. All our algorithms are deterministic.