Lemma 3.4, pp. 121–122 — broken convergence: if f*ⱼ < f*ⱼ₊₁ then the first j vertices are eventually fixed
ProvedNelderMeadLD.Dim2.lemma_3_4Let () be bounded below, let the coefficients satisfy (2.1), and let be a run of Algorithm NM on from a nondegenerate initial simplex in which no shrink step occurs. Suppose that for some with the limits and satisfy
Then there is such that for all the change index satisfies , i.e.
Property (3.2) is called broken convergence for vertex ; the lemma says it freezes the best vertices.
Formalization Note The paper's index is a Lean j : Fin n with j.val , so is j.castSucc and is j.succ; "the first vertices" are the indices i ≤ j.castSucc. The limits , are taken as hypotheses (Tendsto … (𝓝 a), Tendsto … (𝓝 b), a < b); their existence is Lemma 3.3 (3). The change index is stated through its meaning (2.8): the first vertices do not change.
import Mathlib import Definitions.Def_NelderMeadLD_Dim2_Algorithm open Filter Topology
namespace NelderMeadLD.Dim2
theorem lemma_3_4 {n : ℕ} [NeZero n] (f : E n → ℝ) (ρ χ γ σ : ℝ) (hpar : NelderMeadLD.Conv1D.ParamsOK ρ χ γ σ)
(Δ : ℕ → Fin (n + 1) → E n) (hrun : IsNMRun f ρ χ γ σ Δ) (hnd : Nondegenerate (Δ 0))
(hbdd : BddBelow (Set.range f)) (hnoshrink : ∀ k, ¬ IsShrinkAt f ρ χ γ Δ k)
(j : Fin n) (a b : ℝ)
(ha : Tendsto (fun k => f (Δ k j.castSucc)) atTop (𝓝 a))
(hb : Tendsto (fun k => f (Δ k j.succ)) atTop (𝓝 b)) (hab : a < b) :
∃ K : ℕ, ∀ k ≥ K, ∀ i : Fin (n + 1), i ≤ j.castSucc → Δ (k + 1) i = Δ k i := by sorry
end NelderMeadLD.Dim2
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.