On the fractionality of the path packing problem

Given an undirected graph G=(N,E), a subset T of its nodes and an undirected graph (T,S), G and (T,S) together are often called a network. A collection of paths in G whose end-pairs lie in S is called an integer multiflow. When these paths are allowed to have fractional weight, under the constraint that the total weight of the paths traversing a single edge does not exceed 1, we have a fractional multiflow in G. The problems of finding the maximum weight of paths with end-pairs in S over all fractional multiflows in G is called the fractional path packing problem. In 1989, A. Karzanov had defined the fractionality of the fractional path packing problem for a class of networks {G,(T,S)} as the smallest natural D such that for any network from the class, the fractional path packing problem has a solution which becomes integer-valued when multiplied by D (see A. Karzanov in Linear Algebra Appl. 114–115:293–328, 1989). He proved that the fractional path packing problem has infinite fractionality outside a very specific class of networks, and conjectured that within this class, the fractionality does not exceed 4. A. Karzanov also proved that the fractionality of the path packing problem is at most 8 by studying the fractionality of the dual problem. Special cases of Karzanov’s conjecture were proved in or are implied by the works of L.R. Ford and D.R. Fulkerson, Y. Dinitz, T.C. Hu, B.V. Cherkassky, L. Lovãsz and H. Hirai. We prove Karzanov’s conjecture by showing that the fractionality of the path packing problem is at most 4. Our proof is stand-alone and does not rely on Karzanov’s results.