Multicommodity flow problem

Multicommodity flow problem
Name	Multicommodity flow problem
Field	Operations research, Computer science, Combinatorial optimization
Introduced	20th century
Notable	L. R. Ford Jr., D. R. Fulkerson, Jon Kleinberg, Éva Tardos

Multicommodity flow problem The multicommodity flow problem is a network optimization model studied in Operations research and Computer science that extends single-commodity flow formulations to handle multiple, simultaneous flow demands between pairs of nodes; it unifies concepts from Graph theory, Linear programming, Combinatorial optimization, and Network design. The problem has driven research at institutions such as Bell Labs, Massachusetts Institute of Technology, Princeton University, and Stanford University and connects to results by researchers including L. R. Ford Jr., D. R. Fulkerson, Jon Kleinberg, and Éva Tardos. It underpins applications in infrastructure projects linked to organizations like AT&T, Cisco Systems, and Deutsche Telekom.

Definition and problem formulation

A multicommodity flow instance is defined on a directed or undirected graph G = (V, E) with nonnegative capacities c(e) for edges e ∈ E, and a set of k commodities specified by distinct source–sink pairs (s_i, t_i) and demand values d_i; formalizations appear in classic texts from Princeton University and Cornell University and papers by L. R. Ford Jr. and D. R. Fulkerson. The canonical linear programming formulation maximizes aggregate throughput or minimizes congestion subject to edge capacity constraints and flow conservation at vertices; this LP influences work at Bell Labs and methods from George Dantzig's legacy in Linear programming. Variants introduce cost functions on edges, often using convex costs as in research from Stanford University and University of California, Berkeley.

Variants and special cases

Important variants include integral multicommodity flow studied in contexts like the Max-Flow Min-Cut theorem generalizations, unsplittable flow where each commodity must use a single path studied in collaborations involving Microsoft Research, and splittable flow allowing fractional routing as in early results from Princeton University. Other special cases comprise uniform-capacity instances analyzed in works associated with Courant Institute and planar graphs linked to results from University of Oxford; edge-disjoint paths, node-disjoint paths, and keyed formulations for survivable network design relate to research at IBM Research and AT&T Bell Laboratories.

Computational complexity and hardness

The decision and optimization versions of the multicommodity flow problem exhibit varied complexity: while the single-commodity max-flow problem is solvable in polynomial time via algorithms influenced by Jack Edmonds and Richard Karp's paradigms, the integral and unsplittable multicommodity flow problems are NP-hard, with hardness proofs connected to reductions from 3-SAT, Partition problem, and Graph coloring as demonstrated in papers from University of Illinois at Urbana–Champaign and Columbia University. Inapproximability results rely on PCP-theorem techniques developed by researchers at Princeton University and Rutgers University and hardness of approximation links to conjectures posed at workshops at DIMACS.

Algorithms and solution methods

Algorithmic approaches include linear and convex programming relaxations pioneered by George Dantzig and refined through interior-point methods from University of California, Berkeley; combinatorial algorithms derive from augmenting path techniques by D. R. Fulkerson and scaling methods associated with Andrew V. Goldberg. Approximation algorithms yielding polylogarithmic congestion guarantees were developed in influential papers by Jon Kleinberg and Éva Tardos, and randomized rounding methods trace to work at Microsoft Research and Massachusetts Institute of Technology. Combinatorial preflow-push variants, spectral sparsification from Stanford University and Massachusetts Institute of Technology, and multilevel heuristics used in industry at Cisco Systems provide practical solvers; integer programming formulations solved via branch-and-cut originated from techniques at IBM Research and Oak Ridge National Laboratory.

Applications

Multicommodity flow models underpin traffic engineering in metropolitan projects involving Federal Highway Administration, packet routing in backbone networks deployed by AT&T and Verizon Communications, supply chain logistics coordinated by Walmart and Maersk, and energy distribution explored by researchers at Lawrence Berkeley National Laboratory. In telecommunications, the model informs capacity planning at Cisco Systems and spectrum allocation studied at Bell Labs Research; in transportation, it supports urban planning efforts by agencies like New York City Department of Transportation and engineering firms collaborating with Arup Group.

Theoretical results and bounds

Key theoretical results include integrality gap bounds for multicommodity flows relative to cut relaxations proven in work involving Jon Kleinberg and Éva Tardos, flow–cut gap theorems with connections to metric embeddings developed at Princeton University and University of Chicago, and approximation ratio lower bounds derived using techniques from Boaz Barak and colleagues associated with Harvard University and MIT. Spectral and geometric methods from Stanford University and ETH Zurich yield upper bounds on concurrent flow values, while probabilistic constructions tracing to Paul Erdős provide families of graphs demonstrating worst-case behaviors.

Category:Network flow problems