Multi-Commodity Max-Flow Min-Cut Theorems and Applications

David Wajc
Abstract:Consider a single-commodity network flow instance, N = (G, s, t, c). The maximum amount of flow which can be sent from the source s to the sink t is clearly upper bounded by the capacity of the minimum cut separating s and t. The celebrated Maximum-Flow Min-Cut Theorem of Ford and Fulkerson [4] (also proven in the same year by Dantzig and Fulkerson [2] and Elias, Feinstein and Shannon [3]) asserts that these quantities coincide. This fundamental theorem has found myriad applications over the years, including simple proofs of classic combinatorial theorems such as Menger’s, Hall’s, König’s and Dilworth’s theorems, as well as countless algorithmic applications. One might therefore wonder which generalizations of maximum flow admit similar max-flow min-cut theorems and what applications they may yield. A natural generalization of the maximum flow problem to consider is multi-commodity flow problems. One such generalization is the concurrent multi-commodity flow, where we have source-sink pairs {(si, ti)}i=1 and demand Di for each pair (si, ti). This problems asks to send the maximum common fraction f of each demand Di along the set of paths Pi = {p : si ti}, while respecting the edge capacities, ce. (Singlecommodity flow is the special case of a single source-sink pair with demand one.) Written as an LP, the concurrent multi-commodity flow problem and its dual are the following.2
Mathematics,Computer Science
What problem does this paper attempt to address?