Computing Vertex-Weighted Multi-Level Steiner Trees

11/28/2018 ∙ by Faryad Darabi Sahneh, et al. ∙ 0

In the classical vertex-weighted Steiner tree problem (VST), one is given an undirected graph G = (V,E) with nonnegative vertex weights, and a set T ⊆ V of terminals. The objective is to compute a minimum-weight tree that spans T. The VST problem is NP-hard and it is NP-hard to approximate VST to within factor (1-ε) |T|, but nearly-best approximation algorithms exist including the 2 |T|-approximation algorithm of [Klein& Ravi, 1995]. Steiner tree problems and their variants have many applications, from combinatorial optimization and network routing to geometry and visualization. In some applications, the terminals may have different levels, priorities, or rate-of-service requirements. For problems of this type, we study a natural generalization of the VST problem to multiple levels, referred to as the vertex-weighted, multi-level Steiner tree (V-MLST) problem: given a vertex-weighted graph G = (V,E) and ℓ> 2 nested terminal sets T_ℓ⊂ T_ℓ-1⊂...⊂ T_1⊆ V, compute a nested set of trees G_ℓ⊆ G_ℓ-1⊆...⊆ G_1 where each tree G_i spans its corresponding terminal set T_i, with minimum total weight. We introduce a simple heuristic with approximation ratio O(ℓ |T_1|), which runs in a top-down manner using a single-level VST subroutine. We then show that the V-MLST problem can be approximated to within 2 |T_1| of the optimum with a greedy algorithm that connects "level-respecting trees" at each iteration with a minimum cost-to-connectivity ratio. This result is counterintuitive as it suggests that the seemingly harder multi-level version is not indeed harder than the single-level VST problem to approximate. The key tool in the analysis of our greedy approximation algorithm is a new "tailed spider decomposition."



There are no comments yet.


page 1

page 2

page 3

page 4

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.