Metric Travelling Salesman Problem

Metric Travelling Salesman Problem
Name	Metric Travelling Salesman Problem
Field	Theoretical computer science, Combinatorial optimization
Introduced	20th century
Related	Knapsack problem, Set cover problem, Hamiltonian cycle problem, Minimum spanning tree

Metric Travelling Salesman Problem

The Metric Travelling Salesman Problem is a variant of the Travelling Salesman Problem defined on a finite set of vertices with a distance function satisfying the triangle inequality; it asks for a shortest Hamiltonian cycle. The problem appears across work connected to Richard Karp, John Hopcroft, Michael Garey, David Johnson, Vaughan Jones and influenced algorithmic research at institutions like Bell Labs, MIT, Princeton University, Stanford University. Its study intersects literature involving Kruskal's algorithm, Prim's algorithm, Christofides algorithm, Held–Karp lower bound and results of researchers at IBM Research, Microsoft Research, AT&T Laboratories, Brown University, Harvard University.

Definition and Problem Statement

Formally, given a complete weighted undirected graph on vertex set V with nonnegative edge weights w satisfying w(u,v) + w(v,x) >= w(u,x) for all u,v,x, find a Hamiltonian cycle minimizing total weight. Early formalizations trace to work by William Rowan Hamilton in the 19th century and computational framing in texts by Stephen Cook, Leonid Levin, Richard Bellman, Jack Edmonds. Canonical textbook presentations appear in monographs by Christos Papadimitriou, Kenneth Steiglitz, Juris Hartmanis, Mihalis Yannakakis, and lecture notes from École Polytechnique Fédérale de Lausanne and University of Cambridge. Practical exact formulations often use integer linear programming introduced by Dantzig, Fulkerson and Johnson and dynamic programming approaches associated with Bellman and Held.

Properties and Metric Conditions

The metric condition enforces the triangle inequality, often arising from embedding vertices into metric spaces like Euclidean space studied by Carl Friedrich Gauss, Bernhard Riemann, David Hilbert, and applications in networks analyzed by Paul Baran and Donald Davies. Special metrics include Euclidean metrics linked to René Descartes and Gottfried Leibniz coordinate systems, Manhattan metrics tied to urban planning in New York City and Chicago, and ultrametrics connected to studies by John Tukey and Andrey Kolmogorov. Structural properties relate to spanning trees from Kruskal and Prim, matchings from work by Edmonds, and polyhedral studies by William Cook, M. Grötschel, László Lovász, Martin Grötschel and Mikko Koivisto. Metric closure, triangle inequality tightening, and 1-tree relaxations link to contributions from Jack Edmonds and Alexander Schrijver.

Algorithms and Approximation Results

Approximation algorithms include the 3/2-approximation by Christofides and improvements and analyses by researchers at University of Waterloo, Carnegie Mellon University, University of California, Berkeley, and ETH Zurich. Polynomial-time heuristics such as nearest neighbor, insertion heuristics, and 2-opt/3-opt local searches connect to programming efforts at Bell Labs, AT&T Laboratories Research, Google, and Amazon. Linear programming relaxations like the subtour elimination LP and Held–Karp bound were advanced by Dantzig, Wolfe, Held, and Karp. Recent algorithmic progress invokes metric embeddings studied by Jon Kleinberg, Sanjeev Arora, Piotr Indyk, Assaf Naor, and approximation schemes for special cases by Sanjeev Arora for Euclidean instances and related PTAS work at ETH Zurich and University of Pennsylvania.

Complexity and Hardness

The metric variant remains NP-hard via reductions related to Richard Karp's 21 NP-complete problems and completeness frameworks developed by Stephen Cook and Leonid Levin. Hardness of approximation results leverage PCP theorems by László Babai, Ilia Feige, Umesh Vazirani, Mikhail Rabinovich and techniques from Subhash Khot's unique games conjecture circle. Results linking inapproximability thresholds employ reductions involving Set Cover hardness studied by Vazirani and hardness frameworks by Arora, Safra, and Dinur. Complexity-theoretic classifications draw on structural results from Michael Sipser, Leonid Levin, Richard Lipton, Noam Nisan, and circuit complexity insights at Carnegie Mellon University.

Special Cases and Variants

Notable restricted cases include Euclidean TSP in fixed dimension with PTAS by Sanjeev Arora and Joseph Mitchell; graphic TSP tied to Euler tours and work by Christos Papadimitriou and András Sebő; asymmetric metrics related to Jonas Kärkkäinen and approximations by Robert Weismantel; and prize-collecting orienteering variants explored at Columbia University, University of Illinois Urbana–Champaign, and Cornell University. Variants like bottleneck TSP, k-TSP, and clustered TSP link to combinatorial studies by Karp, Papadimitriou, Shmuel Safra, Ravi Kumar, and heuristics developed at MIT Lincoln Laboratory and Sandia National Laboratories.

Practical Applications and Empirical Performance

Applications span logistics and routing in operations at United Parcel Service, FedEx, DHL, and fleet management at Maersk and Carnival Corporation; bioinformatics sequencing efforts at Broad Institute and Sanger Institute; and circuit board drilling and microchip layout studied at Intel Corporation, AMD, NVIDIA, and TSMC. Empirical performance comparisons appear in benchmark libraries maintained by DIMACS, TSPLIB and evaluated in competitions involving teams from Princeton University, ETH Zurich, National University of Singapore, Tokyo Institute of Technology, and University of British Columbia. Industrial solvers from Gurobi, CPLEX, GLPK, and academic implementations by Concorde TSP Solver and projects at University of Waterloo demonstrate state-of-the-art practical behavior for metric instances.

Category:Combinatorial optimization