Technical Report 019-2004

Title
Complexity and Approximability of k-splittable flow problems
Authors
Ronald Koch and Ines Spenke
Source
Download as [PDF] [ps.gz]
Classification
not available
Keywords
network flows, k-splittable flows, complexity, approximation algorithms
Abstract
We consider s,t-flow problems in connected graphs with the additional requirement that one can decompose the flow into at most k paths for a given number k. Such flows are called k-splittable flows. In the multi-commodity case they generalize unsplittable flows.
In this paper, we consider the problem MkSF, which means to find a maximum k-splittable s,t-flow. We show complexity and approximability results for this problem depending on k. Results work for directed and undirected graphs. Firstly, we show that for all constant integer k > 1 MkSF is strongly NP-hard and cannot be approximated with a guarantee better than 5/6. This is the first constant bound for the non-approximability of this problem. Secondly, we study MkSF with k depending on m, the number of edges, and n, the number of nodes of the graph. We prove that MkSF is NP-hard for 2 <= k <= m-n+1. We show that it is polynomially solvable for k >= m-n+2. Thus, there is no gap in the analysis. For the special case of simple graphs we show that MkSF is polynomially solvable for k = m - c for each integer constant c. That also holds for k=m-(log log p(m,n))^eps for every polynom p in n and m, eps in (0,1).
It is well known that every flow in a graph can be decomposed into at most m paths and cycles. We show as a general result that the number of paths in such a decomposition can always be limited to m - n + 2.