Held–Karp relaxation
Generated by GPT-5-miniExpansion Funnel Raw 62 → Dedup 0 → NER 0 → Enqueued 0
| Held–Karp relaxation | |
|---|---|
| Name | Held–Karp relaxation |
| Inventor | Richard M. Karp; Michael Held |
| Year | 1970s |
| Field | Combinatorial optimization |
| Related | Traveling Salesman Problem; linear programming; cutting-plane method |
Held–Karp relaxation is a linear programming relaxation used to bound the optimum of the Traveling Salesman Problem by embedding subtour elimination constraints and degree constraints into a polyhedral framework. It provides a provable lower bound that is central to theoretical analyses of exact and approximation algorithms, and serves as a bridge between combinatorial optimization and continuous methods developed in operations research and theoretical computer science.
Definition and formulation
The Held–Karp relaxation is defined for a complete weighted instance of the Travelling Salesman Problem by introducing variables x_{ij} for edges (i,j) and enforcing degree constraints and subtour inequalities to obtain a linear program over the edge-incidence polytope. In standard form the program minimizes sum_{i
History and origin
The relaxation emerged from collaborative research by Michael Held and Richard M. Karp in the early 1970s within the milieu of Bell Labs and the broader community that included contemporaries from IBM Research, University of California, Berkeley, and Princeton University. Motivated by landmark work on the Travelling Salesman Problem by Dantzig, Fulkerson, and Johnson in the 1950s and 1960s, Held and Karp formalized a Lagrangian relaxation and cutting-plane perspective that synthesized ideas from Jack Edmonds and Richard Karp's complexity theory. Subsequent refinement and dissemination occurred through conferences such as the ACM Symposium on Theory of Computing and the Mathematical Programming Society meetings, and influenced later algorithmic frameworks developed at institutions like MIT and Stanford University.
Mathematical properties
The relaxation yields a polytope whose extreme points correspond to fractional 1-tree structures, generalizing the concept of spanning trees introduced by Kirchhoff and later formalized by scholars at Harvard University and Yale University. The Held–Karp bound is always at least as strong as simple degree-based bounds and often coincides with the optimal tour value for special metrics including those satisfying properties akin to the triangle inequality instances studied by Vijay Vazirani and David Johnson. Integrality gap analyses link the relaxation to hardness results proved by Richard Karp and approximation thresholds studied in the context of NP-completeness championed by Stephen Cook. Polyhedral combinatorics results by researchers affiliated with INRIA and ETH Zurich established facet-defining conditions, while work by investigators at Bell Labs and AT&T characterized the dual variables as potentials on vertices and cuts, connecting to classical theorems in combinatorics by Paul Erdős and structural insights from William Tutte.
Computational methods and algorithms
Practical computation leverages branch-and-bound and branch-and-cut frameworks that integrate the Held–Karp relaxation as a bounding subroutine; these frameworks evolved in software originating at IBM and academic codebases from Cornell University and Columbia University. Lagrangian relaxation solvers use subgradient optimization techniques pioneered by researchers at Bell Labs and later refined at Sandia National Laboratories and Los Alamos National Laboratory. Cutting-plane algorithms add violated subtour constraints discovered via minimum cut computations reducible to max-flow routines attributed to Edmonds–Karp and implementations by engineers at Google and Microsoft Research. Heuristic embedding into primal heuristics and local search—lineages traceable to Holger H. Hoos and Thomas Stützle—combine with exact methods to produce state-of-the-art solvers used in competitions like the DIMACS Implementation Challenge.
Applications and impact
Beyond its central role in exact TSP solvers used in logistics operations at firms such as DHL, UPS, and Amazon (company), the Held–Karp relaxation influenced routing and network design research at Bell Labs, Siemens, and General Electric. In theoretical computer science it underpins approximation algorithms and hardness separations cited in works from ETH Zurich and Carnegie Mellon University, and it appears in computational biology problems addressed at Broad Institute and Salk Institute where Hamiltonian-path-like models arise. The relaxation also inspired advances in polyhedral theory taught at universities including Oxford University and Cambridge University and influenced curricula at the Courant Institute.
Variants and extensions
Variants include strengthened relaxations that incorporate comb inequalities and blossom facets developed following paradigms pioneered by Jack Edmonds and extended in research groups at INRIA and University of Waterloo. Metric-specific adaptations, such as those tailored for Euclidean instances studied at ETH Zurich, exploit geometric structure akin to results by Kurt Mehlhorn and Eugene Lawler. Lagrangian dual heuristics and semidefinite programming relaxations from teams at Princeton University and University of California, Berkeley offer alternative bounds, and hybrid methods combining the Held–Karp relaxation with constraint programming frameworks have been prototyped at Google Research and Microsoft Research.