Balanced flows for transshipment problems

TL;DR

This paper introduces a polynomial-time method for finding balanced flows in transshipment problems, minimizing excess ratios lexicographically, extending previous results to directed graphs and simplifying the algorithm.

Contribution

The paper extends the balanced flow solution to directed graphs and simplifies the existing algorithm using Gale and Hoffman's criterion.

Findings

Constructed a polynomial-time algorithm for balanced flows in directed graphs.

Extended symmetric graph results to directed graphs.

Simplified the proof and algorithm for balanced flows.

Abstract

A transshipment problem (G, d, \lambda) is modeled by a directed graph G = (V, E) with weighted vertices d = (d_v | v \in V) and directed edges \lambda = (\lambda_e | e \in E) interpreted as follows: G is a communication or transportation network, e.g., a pipeline; each edge e \in E is a one-way communication line, road or pipe of capacity \lambda_e, while every vertex v \in V is a node of production d_v > 0, consumption d_v < 0, or transition d_v = 0. A non-negative flow x = (x_e \mid e \in E) is called weakly feasible if for each v \in V the algebraic sum of flows, over all directed edges incident to v, equals d_v; or shorter, if A_G x = d, where A_G is the vertex-edge incidence matrix of G. A weakly feasible flow x is called feasible if x_e \leq \lambda_e for all e \in E. We consider weakly feasible but not necessarily feasible flows, that is, inequalities x_e > \lambda_e are allowed. However, such an excess is viewed as unwanted (dangerous) and so we minimize the excess ratio vector r = (r_e = x_e / \lambda_e | e \in E) lexicographically. More precisely, first, we look for the weakly feasible flows minimizing the maximum of re over all e in E; among all such flows we look for those that minimize the second largest coordinate of r, etc. Clearly, |E| such steps define a unique balanced flow, which provides the lexmin solution for problem (G, d, \lambda). We construct it in polynomial time, provided vectors d and \lambda are integer. For symmetric digraphs the problem was solved by Gurvich and Gvishiani in 1984. Here we extend this result to directed graphs. Furthermore, we simplify the algorithm and proofs applying the classic criterion of existence of a feasible flow for (G, d, \lambda) obtained by Gale and Hoffman in late 1950-s.