On the hop-constrained Steiner tree problems

TL;DR

This paper compares two mathematical models for solving hop-constrained Steiner tree problems, demonstrating that the partial-ordering model offers superior theoretical and practical performance over the assignment model.

Contribution

It provides theoretical polyhedral insights and computational evidence showing the advantages of the partial-ordering model for hop-constrained Steiner tree problems.

Findings

Partial-ordering model has better LP relaxation than assignment model.

Partial-ordering model solves more instances in computational tests.

Theoretical polyhedral advantages of the partial-ordering model are established.

Abstract

The hop-constrained Steiner tree problem (HSTP) is a generalization of the classical Steiner tree problem. It asks for a minimum cost subtree that spans some specified nodes of a given graph, such that the number of edges between each node of the tree and its root respects a given hop limit. This NP-hard problem has many variants, often modeled as integer linear programs. Two of the models are so-called assignment and partial-ordering based models, which yield (up to our knowledge) the best two state-of-the-art formulations for the variant Steiner tree problem with revenues, budgets, and hop constraints (STPRBH). The solution of the HSTP and its variants such as the STPRBH and the hop-constrained minimum spanning tree problem (HMSTP) is a hop-constrained tree, a rooted tree whose depth is bounded by a given hop limit. This paper provides some theoretical results that show the polyhedral advantages of the partial-ordering model over the assignment model for this class of problems. Computational results in this paper and the literature for the HSTP, STPRBH, and HMSTP show that the partial-ordering model outperforms the assignment model in practice, too; it has better linear programming relaxation and solves more instances.