Eq. (2): — every 1-tree bound is below the optimum tour weight
ProvedHeldKarp.Ascent.eq_2lagrangian-relaxationlower-boundp2o-batch-p200bp2o-gran-per-chapterp2o-plan-paperp2o-v1traveling-salesman
Let , let be arbitrary symmetric real edge weights on , and let be the 1-tree bound. Then for every real -vector and every tour of ,
Equivalently, with the weight of a minimum tour,
This is the family of lower bounds on which the whole method rests: the best such bound is , and in particular is bounded above, so its maximum exists.
Formalization Note is not named; the statement quantifies over every tour instead, which is equivalent. No metric assumption is made on the weights. The hypothesis is implicit in the paper (for there are no tours and no 1-trees).
Preamble
import Mathlib import Definitions.Def_HeldKarp_Ascent_IsOneTree import Definitions.Def_HeldKarp_Ascent_oneTreeBound
Formal statement
namespace HeldKarp.Ascent
/-- **Eq. (2)** (with Eq. (1)) — Held & Karp, *The traveling-salesman problem and minimum spanning
trees: Part II*, Math. Programming 1 (1971), §1, p. 8 (PDF p. 3): "C* ≥ w(π)", where C* is the
weight of a minimum tour with respect to the weights `c_ij` and
`w(π) = min_k [c_k + π · v_k]` is the minimum over 1-trees.
For `n ≥ 3`, arbitrary real edge weights `c`, every real `n`-vector `π` and every tour `H`,
`w(π) ≤ weight c H`. Since this holds for every tour, it is `C* ≥ w(π)`.
Formalization Note: `C*` is not named; quantifying over every tour is equivalent (for `n ≥ 3`
tours exist, finitely many). `3 ≤ n` is implicit in the paper: for `n ≤ 2` there is neither a
tour nor a 1-tree. Weights are arbitrary reals; no metric assumption. -/
theorem eq_2 {n : ℕ} [NeZero n] (hn : 3 ≤ n) (c : Sym2 (Fin n) → ℝ) (π : Fin n → ℝ)
(H : SimpleGraph (Fin n)) (hH : IsTour H) :
oneTreeBound c π ≤ weight c H := by sorry
end HeldKarp.Ascent
Source
Held & Karp, The traveling-salesman problem and minimum spanning trees: Part II, Math. Programming 1 (1971), DOI 10.1007/BF01584070, §1, p. 8 (PDF p. 3), Eq. (1) and Eq. (2)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.