Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4.2, p. 130 — in dimension 1 with ρ = 1, diam(Δ_{k+M}) ≤ ½ diam(Δ_k) for all k ≥ K, M depending only on χ and γ

Proved
NelderMeadLD.Rate1D.theorem_4_2

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

direct-searchlinear-convergencenelder-meadp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1strict-convexity

Fix an expansion coefficient χ>1\chi > 1χ>1 and a contraction coefficient 0<γ<10 < \gamma < 10<γ<1. There is an integer M≥0M \ge 0M≥0, depending only on χ\chiχ and γ\gammaγ, with the following property.

Let 0<σ<10 < \sigma < 10<σ<1, let f:R→Rf:\mathbb R \to \mathbb Rf:R→R be strictly convex with bounded level sets, and apply the one-dimensional Nelder–Mead method with reflection coefficient ρ=1\rho = 1ρ=1 and coefficients χ,γ,σ\chi, \gamma, \sigmaχ,γ,σ to fff, starting from a nondegenerate ordered initial interval Δ0\Delta_0Δ0​. Then the iteration index KKK of Lemma 4.2 exists — that is, there is a first iteration KKK with f1(K)≤fe(K)f_1^{(K)} \le f_e^{(K)}f1(K)​≤fe(K)​ — and

diam⁡(Δk+M)≤12diam⁡(Δk)for all k≥K.\operatorname{diam}(\Delta_{k+M}) \le \tfrac12 \operatorname{diam}(\Delta_k)\qquad\text{for all } k \ge K.diam(Δk+M​)≤21​diam(Δk​)for all k≥K.

Here Δk={x1(k),x2(k)}\Delta_k = \{x_1^{(k)}, x_2^{(k)}\}Δk​={x1(k)​,x2(k)​} is the interval at iteration kkk, diam⁡(Δk)=∣x1(k)−x2(k)∣\operatorname{diam}(\Delta_k) = |x_1^{(k)} - x_2^{(k)}|diam(Δk​)=∣x1(k)​−x2(k)​∣, and fe(K)f_e^{(K)}fe(K)​ is the value of fff at the expansion point x1(K)+χ(x1(K)−x2(K))x_1^{(K)} + \chi(x_1^{(K)} - x_2^{(K)})x1(K)​+χ(x1(K)​−x2(K)​). The theorem says that, once the minimizer is bracketed, the Nelder–Mead interval converges M-step linearly, with a step count MMM that does not depend on fff or on the initial interval.

Formalization Note. "MMM depending only on χ\chiχ and γ\gammaγ" is encoded by the quantifier order: MMM is chosen after χ,γ\chi, \gammaχ,γ and before σ\sigmaσ, fff and Δ0\Delta_0Δ0​. The page presupposes the index KKK of Lemma 4.2; here its existence is part of the conclusion, so the statement cannot hold vacuously for a run that is never bracketed. This makes the item slightly stronger than the printed sentence (the existence is Lemma 4.2, proved in the paper). The shrink coefficient σ\sigmaσ is a parameter of the algorithm and satisfies (2.1); it does not otherwise enter.

Preamble
import Mathlib
import Definitions.Def_NelderMeadLD_Conv1D_Algorithm
import Definitions.Def_NelderMeadLD_Rate1D_Algorithm
Formal statement
namespace NelderMeadLD.Rate1D

open NelderMeadLD.Conv1D

theorem theorem_4_2 :
    ∀ χ γ : ℝ, 1 < χ → 0 < γ → γ < 1 →
      ∃ M : ℕ,
        ∀ σ : ℝ, 0 < σ → σ < 1 →
        ∀ f : ℝ → ℝ, StrictConvexOn ℝ Set.univ f →
        (∀ μ : ℝ, Bornology.IsBounded {x | f x ≤ μ}) →
        ∀ p0 : ℝ × ℝ, IsStart f p0 →
        ∃ K : ℕ, FirstBracket f χ γ σ p0 K ∧
          ∀ k : ℕ, K ≤ k →
            diam (run f 1 χ γ σ p0 (k + M)) ≤ 1 / 2 * diam (run f 1 χ γ σ p0 k) := by sorry

end NelderMeadLD.Rate1D
Source
Lagarias, Reeds, Wright & Wright, Convergence properties of the Nelder–Mead simplex method in low dimensions, SIAM J. Optim. 9 (1998), p. 130, Theorem 4.2
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me