[论文解读] A Local-To-Global Theorem for Congested Shortest Paths

Shyan Akmal, Nicole Wein|arXiv (Cornell University)|Nov 13, 2022

Advanced Graph Theory Research被引用 1

一句话总结

本文将阿米里与沃加拉关于有向无环图中最短路径的局部到全局定理推广至一般图。对于无向图，若每组四个节点均位于同一条最短路径上，则所有节点均位于同一条最短路径上；对于有向图，原始定理不成立，但存在一种往返类比定理——每个节点均位于从 s 到 t 的最短路径或从 t 到 s 的最短路径上。该结果使得在 k−c 为常数时，可在无向图上设计多项式时间算法求解 (k, c)-SPC 问题。

ABSTRACT

Amiri and Wargalla proved the following local-to-global theorem about shortest paths in directed acyclic graphs (DAGs): if G is a weighted DAG with the property 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). However, we prove a roundtrip analogue of the theorem which guarantees 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 local-to-global 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),… ,(s_k,t_k), and a positive integer c ≤ k, and tasked with finding a collection of paths P_1,… , 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, there are no congestion constraints, and the problem can be solved easily by running a shortest path algorithm for each pair (s_i,t_i) independently. At the other extreme, when c = 1, the (k,c)-SPC problem is equivalent to the k-Disjoint Shortest Paths (k-DSP) problem, where the collection of shortest paths must be node-disjoint. For fixed k, k-DSP is polynomial-time solvable on DAGs and undirected graphs. Amiri and Wargalla interpolated between these two extreme values of c, to obtain an algorithm for (k,c)-SPC on DAGs that runs in polynomial time when k-c is constant. In the same way, we prove that (k,c)-SPC can be solved in polynomial time on undirected graphs whenever k-c is constant. For directed graphs, because of our counterexample to the original theorem statement, our roundtrip local-to-global result does not imply such an algorithm (k,c)-SPC. Even without an algorithmic application, our proof for directed graphs may be of broader interest because it characterizes intriguing structural properties of shortest paths in directed graphs.

研究动机与目标

• 确定阿米里与沃加拉关于有向无环图中最短路径的局部到全局定理是否可推广至一般无向图与有向图。
• 解决结构问题：若小规模节点集在最短路径中局部包含，是否意味着其在单一最短路径中全局包含？
• 建立局部到全局性质在无向图上 k-最短路径带拥塞 c ((k, c)-SPC) 问题中的算法影响。
• 刻画有向图中最短路径结构的局限性，尤其在拥塞与路径覆盖的背景下。

提出的方法

• 通过证明：若每组四个节点均位于同一条最短路径上，则所有节点位于同一条最短路径上，从而建立无向图的局部到全局定理。
• 构造反例以证明有向图中原始 DAG 定理不成立，采用边权不对称的双向环结构。
• 为有向图引入往返类比定理：对某对节点 (s, t)，每个节点均位于从 s 到 t 的最短路径或从 t 到 s 的最短路径上。
• 运用归纳路径交换论证与环序引理，证明在局部条件下路径存在的性质。
• 应用关键节点引理与路径分解技术，验证复杂节点排序下的路径覆盖性。
• 通过最短路径排序的结构分析与路径交集研究，推导出假设路径配置中的矛盾。

实验结果

研究问题

• RQ1 有向无环图中最短路径的局部到全局性质是否可推广至无向图？若可，所需的局部条件规模是多少？
• RQ2 原始 DAG 局部到全局定理是否可推广至一般有向图？是否存在根本性障碍？
• RQ3 有向图中是否存在有意义的局部到全局定理类比，可保持全局路径覆盖性质？
• RQ4 局部到全局性质对无向图上 (k, c)-SPC 问题的算法影响是什么？
• RQ5 在局部路径包含条件下，有向图中最短路径的结构特性是否可被完全刻画？

主要发现

• 对于无向图，若每组四个节点均位于同一条最短路径上，则图中所有节点均位于同一条最短路径上。
• 无向图中常数 4 无法减小至 3，反例为 4-圈结构。
• 对于有向图，原始局部到全局定理不成立：存在使用边权不对称双向环的反例。
• 有向图中存在往返局部到全局定理：对某对节点 (s, t)，每个节点均位于从 s 到 t 的最短路径或从 t 到 s 的最短路径上。
• 当 k−c 为常数时，无向图上的 (k, c)-SPC 问题可多项式时间求解，扩展了此前在 DAG 上的结果。
• 有向图中的结构分析揭示了最短路径排序的深层约束，尤其通过路径交换与环序论证。