Minimum Cuts in Surface Graphs

TL;DR

This paper presents efficient algorithms for computing minimum cuts in surface-embedded graphs, extending known planar graph algorithms to graphs on surfaces with higher genus, and introduces methods for related homology-based subgraph problems.

Contribution

The authors develop algorithms for minimum cuts in surface graphs with improved time complexity and connect these to homology-based subgraph problems, including NP-hardness results.

Findings

Algorithms run in $g^{O(g)} n \, log \log n$ or $2^{O(g)} n \log n$ time.

Algorithms match best known times for planar graphs when genus is constant.

Finding a minimum-weight subgraph homologous to a cycle is NP-hard.

Abstract

We describe algorithms to efficiently compute minimum $(s,t)$-cuts and global minimum cuts of undirected surface-embedded graphs. Given an edge-weighted undirected graph $G$ with $n$ vertices embedded on an orientable surface of genus $g$, our algorithms can solve either problem in $g^{O(g)} n \log \log n$ or $2^{O(g)} n \log n$ time, whichever is better. When $g$ is a constant, our $g^{O(g)} n \log \log n$ time algorithms match the best running times known for computing minimum cuts in planar graphs. Our algorithms for minimum cuts rely on reductions to the problem of finding a minimum-weight subgraph in a given $\mathbb{Z}_2$-homology class, and we give efficient algorithms for this latter problem as well. If $G$ is embedded on a surface with $b$ boundary components, these algorithms run in $(g + b)^{O(g + b)} n \log \log n$ and $2^{O(g + b)} n \log n$ time. We also prove that finding a minimum-weight subgraph homologous to a single input cycle is NP-hard, showing it is likely impossible to improve upon the exponential dependencies on $g$ for this latter problem.