Lemma 1:
ProvedHeldKarp.Ascent.lemma_1lagrangian-relaxationp2o-batch-p200bp2o-gran-per-chapterp2o-plan-paperp2o-v1subgradienttraveling-salesman
Let , let be symmetric real edge weights on , and let be the 1-tree bound. Let satisfy , and let be any minimum-weight 1-tree at the point , with degree-excess vector . Then
Geometrically, the hyperplane through with normal bounds a closed half-space containing every point at least as good as , in particular every maximizer of : is a supergradient of the concave function at . This justifies stepping from along .
Formalization Note The inner product is Euclidean, written as a coordinate sum. The minimum-weight 1-tree may be any of the tied minimizers.
Preamble
import Mathlib import Definitions.Def_HeldKarp_Ascent_IsOneTree import Definitions.Def_HeldKarp_Ascent_oneTreeBound
Formal statement
namespace HeldKarp.Ascent
/-- **Lemma 1** — Held & Karp, *The traveling-salesman problem and minimum spanning trees: Part II*,
Math. Programming 1 (1971), §2, p. 9 (PDF p. 4): "Let π̄ and π be such that w(π̄) ≥ w(π). Then
(π̄ − π) · v_{k(π)} ≥ w(π̄) − w(π) ≥ 0."
Here `G` plays the role of the 1-tree `k(π)`: any minimum-weight 1-tree at the point `π`.
Formalization Note: `·` is the Euclidean inner product, written as a sum over coordinates. The
choice of minimum-weight 1-tree is arbitrary among ties (every choice is covered). `3 ≤ n` is
implicit in the paper (1-trees exist only for `n ≥ 3`). -/
theorem lemma_1 {n : ℕ} [NeZero n] (hn : 3 ≤ n) (c : Sym2 (Fin n) → ℝ) (πbar π : Fin n → ℝ)
(G : SimpleGraph (Fin n)) (hG : IsMinOneTree c π G)
(hw : oneTreeBound c π ≤ oneTreeBound c πbar) :
oneTreeBound c πbar - oneTreeBound c π ≤ ∑ i, (πbar i - π i) * degExcess G i ∧
0 ≤ oneTreeBound c πbar - oneTreeBound c π := 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, §2, p. 9 (PDF p. 4), Lemma 1
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.