Near Optimal Sized Weight Tolerant Subgraph for Single Source Shortest Path

TL;DR

This paper introduces a method to construct sparse subgraphs that preserve shortest path distances from a source under bounded weight increments, addressing a gap in fault-tolerant network design with novel bounds and constraints.

Contribution

It presents the first study of constructing sparse subgraphs that maintain distances under weight increments, with tight bounds and practical constraints on edge weights.

Findings

Constructed subgraph with at most e*(k-1)!*2^k*n edges

Proved a lower bound of c*2^k*n edges for such subgraphs

Showed the necessity of integer weight restrictions for preserving distances

Abstract

In this paper we address the problem of computing a sparse subgraph of a weighted directed graph such that the exact distances from a designated source vertex to all other vertices are preserved under bounded weight increment. Finding a small sized subgraph that preserves distances between any pair of vertices is a well studied problem. Since in the real world any network is prone to failures, it is natural to study the fault tolerant version of the above problem. Unfortunately, it turns out that there may not always exist such a sparse subgraph even under single edge failure [Demetrescu \emph{et al.} '08]. However in real applications it is not always the case that a link (edge) in a network becomes completely faulty. Instead, it can happen that some links become more congested which can easily be captured by increasing weight on the corresponding edges. Thus it makes sense to try to construct a sparse distance preserving subgraph under the above weight increment model. To the best of our knowledge this problem has not been studied so far. In this paper we show that given any weighted directed graph with $n$ vertices and a source vertex, one can construct a subgraph that contains at most $e \cdot (k-1)!2^kn$ many edges such that it preserves distances between the source and all other vertices as long as the total weight increment is bounded by $k$ and we are allowed to have only integer valued (can be negative) weight on each edge and also weight of an edge can only be increased by some positive integer. Next we show a lower bound of $c\cdot 2^kn$, for some constant $c \ge 5/4$, on the size of the subgraph. We also argue that restriction of integer valued weight and integer valued weight increment are actually essential by showing that if we remove any one of these two restrictions we may need to store $\Omega(n^2)$ edges to preserve distances.