Approximation Algorithms for the Generalized Point-To-Point Problem

Abstract

We consider the Generalized Point-to-Point (GP2P) problem in which we have an edge-weighted graph G with (possibly negative) node charges ϕ(v) ∈ ℤ. The goal is to find a minimum-cost set of edges such that each component has nonnegative total charge. Viewing the positive charges as specifying supply and negative charges as demand quantities at various nodes, the problem is equivalent to build the cheapest network so that it is possible to satisfy all demands by routing supplies across the network.
This problem is a significant generalization of other network design problems such as the well-studied Steiner Forest problem. Even the special case of only having one single demand point (having charge -k and all the other nodes having charge +1) is capturing the k-Minimum Spanning Tree problem. Earlier work by Hajiaghayi et al. (2016) [Hajiaghayi et al., 2016] gave an O(log n) approximation in pseudo-polynomial time with further improved guarantees if the total supply is not much larger than the total demand, and also a 2-approximation if the total supply equals the total demand.
Our contributions are four-fold: (a) we show how known k-Minimum Spanning Tree approximations can be extended to GP2P approximations while losing only a ε-factor if the number of demand points in the instance is bounded by a constant, (b) we improve the running time to be Fixed-Parameter Tractable (FPT) in the number of demand points in constant-dimensional Euclidean metrics, (c) we give a 2-approximation in instances where edge costs are all 1 and ϕ(v) = ± 1 for each node v and show such instances are APX-hard, and (d) we show how the logarithmic approximations in earlier work can be modified to run in truly polynomial time.