A Local-to-Global Theorem for Congested Shortest Paths

TL;DR

This paper extends a local-to-global shortest path theorem from DAGs to undirected graphs, explores its limitations in directed graphs, and applies it to solve congestion problems in shortest path collections.

Contribution

It generalizes a local-to-global shortest path theorem to undirected graphs and provides a roundtrip analogue for directed graphs, with implications for congestion problems.

Findings

The theorem holds for undirected graphs with a constant difference in node pairs.

Counterexample shows the theorem does not hold for directed graphs.

Application to polynomial-time solutions for congestion problems when k-c is constant.

Abstract

Amiri and Wargalla (2020) proved the following local-to-global theorem in directed acyclic graphs (DAGs): if $G$ is a weighted DAG such that for each subset $S$ of 3 nodes there is a shortest path containing every node in $S$, then there exists a pair $(s,t)$ of nodes such that there is a shortest $st$-path containing every node in $G$. We extend this theorem to general graphs. For undirected graphs, we prove that the same theorem holds (up to a difference in the constant 3). For directed graphs, we provide a counterexample to the theorem (for any constant), and prove a roundtrip analogue of the theorem which shows there exists a pair $(s,t)$ of nodes such that every node in $G$ is contained in the union of a shortest $st$-path and a shortest $ts$-path. The original theorem for DAGs has an application to the $k$-Shortest Paths with Congestion $c$ (($k,c$)-SPC) problem. In this problem, we are given a weighted graph $G$, together with $k$ node pairs $(s_1,t_1),\dots,(s_k,t_k)$, and a positive integer $c\leq k$. We are tasked with finding paths $P_1,\dots, P_k$ such that each $P_i$ is a shortest path from $s_i$ to $t_i$, and every node in the graph is on at most $c$ paths $P_i$, or reporting that no such collection of paths exists. When $c=k$ the problem is easily solved by finding shortest paths for each pair $(s_i,t_i)$ independently. When $c=1$, the $(k,c)$-SPC problem recovers the $k$-Disjoint Shortest Paths ($k$-DSP) problem, where the collection of shortest paths must be node-disjoint. For fixed $k$, $k$-DSP can be solved in polynomial time on DAGs and undirected graphs. Previous work shows that the local-to-global theorem for DAGs implies that $(k,c)$-SPC on DAGs whenever $k-c$ is constant. In the same way, our work implies that $(k,c)$-SPC can be solved in polynomial time on undirected graphs whenever $k-c$ is constant.