Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 3.1 — with ∣βk∣≤βkFR|\beta_k| \le \beta_k^{FR}∣βk​∣≤βkFR​ and (2.17), σ2<1/2\sigma_2 < 1/2σ2​<1/2, the directions are descent directions satisfying (3.2)

Proved
NonlinCG.FRBound.lemma_3_1

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

conjugate-gradientfletcher-reevesnonlinear-optimizationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let fff satisfy Assumptions 2.1 at x1x_1x1​. Consider a method of the form (1.2)–(1.3), d1=−g1d_1 = -g_1d1​=−g1​, dk=−gk+βkdk−1d_k = -g_k + \beta_k d_{k-1}dk​=−gk​+βk​dk−1​ (k≥2k \ge 2k≥2), xk+1=xk+αkdkx_{k+1} = x_k + \alpha_k d_kxk+1​=xk​+αk​dk​ with αk>0\alpha_k > 0αk​>0, in which every βk\beta_kβk​ satisfies

∣βk∣≤βkFR=∥gk∥2∥gk−1∥2(k≥2)(3.1)|\beta_k| \le \beta_k^{FR} = \frac{\|g_k\|^2}{\|g_{k-1}\|^2}\qquad (k \ge 2) \tag{3.1}∣βk​∣≤βkFR​=∥gk−1​∥2∥gk​∥2​(k≥2)(3.1)

and every steplength satisfies condition (2.17), ∣⟨g(xk+αkdk),dk⟩∣≤−σ2⟨gk,dk⟩|\langle g(x_k + \alpha_k d_k), d_k\rangle| \le -\sigma_2\langle g_k, d_k\rangle∣⟨g(xk​+αk​dk​),dk​⟩∣≤−σ2​⟨gk​,dk​⟩, with 0<σ2<120 < \sigma_2 < \tfrac120<σ2​<21​. Suppose no gradient gkg_kgk​ vanishes. Then every dkd_kdk​ is a descent direction, ⟨gk,dk⟩<0\langle g_k, d_k\rangle < 0⟨gk​,dk​⟩<0, and for every k≥1k \ge 1k≥1

−∑j=0k−1σ2j  ≤  ⟨gk,dk⟩∥gk∥2  ≤  −2+∑j=0k−1σ2j.(3.2)-\sum_{j=0}^{k-1}\sigma_2^j \;\le\; \frac{\langle g_k, d_k\rangle}{\|g_k\|^2} \;\le\; -2 + \sum_{j=0}^{k-1}\sigma_2^j . \tag{3.2}−j=0∑k−1​σ2j​≤∥gk​∥2⟨gk​,dk​⟩​≤−2+j=0∑k−1​σ2j​.(3.2)

At k=1k = 1k=1 all three terms equal −1-1−1. Since ∑jσ2j<1/(1−σ2)<2\sum_j \sigma_2^j < 1/(1-\sigma_2) < 2∑j​σ2j​<1/(1−σ2​)<2, the upper bound gives the sufficient descent condition (2.18), and the two bounds control the growth of ∥dk∥\|d_k\|∥dk​∥ in the proof of Theorem 3.2.

Formalization Note The paper does not state gk≠0g_k \ne 0gk​=0; it divides by ∥gk∥2\|g_k\|^2∥gk​∥2 in (3.2) and calls dkd_kdk​ a descent direction, which is impossible when gk=0g_k = 0gk​=0, so we assume gk≠0g_k \ne 0gk​=0 for all k≥1k \ge 1k≥1. Only condition (2.17) is assumed, as in the paper ("the Wolfe condition (2.17)"), with the range 0<σ20 < \sigma_20<σ2​ from (2.17). The paper's standing assumption (1.1) "f is smooth" is taken as global C1C^1C1. Indices start at 111; ∑j=0k−1\sum_{j=0}^{k-1}∑j=0k−1​ is a sum over j<kj < kj<k. The finite-dimensional real inner product space and positive steplengths are as in the definition file.

Preamble
import Mathlib
import Definitions.Def_NonlinCG_FRBound_Setting
Formal statement
namespace NonlinCG.FRBound

/-- Lemma 3.1 (Gilbert–Nocedal, INRIA RR-1268, p. 7): for a method (1.2)–(1.3) with
`|β_k| ≤ β_k^{FR}` (3.1) for `k ≥ 2` and steplengths satisfying (2.17) with `0 < σ₂ < 1/2`, every
`d_k` is a descent direction and
`−Σ_{j=0}^{k−1} σ₂^j ≤ ⟨g_k, d_k⟩/‖g_k‖² ≤ −2 + Σ_{j=0}^{k−1} σ₂^j` for `k ≥ 1` (3.2).
The hypothesis `g_k ≠ 0` is implicit in the paper, which divides by `‖g_k‖²`. -/
theorem lemma_3_1 {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
    [FiniteDimensional ℝ E] (f : E → ℝ) (hf : ContDiff ℝ 1 f) (β α : ℕ → ℝ) (x d : ℕ → E)
    (σ₂ : ℝ) (hA : Assumptions21 f (x 1)) (hrun : IsCGRun f β α x d)
    (hβ : ∀ k ≥ 2, |β k| ≤ betaFR f x k)
    (hσ : 0 < σ₂ ∧ σ₂ < 1 / 2)
    (hls : ∀ k ≥ 1,
      |inner ℝ (gradient f (x k + α k • d k)) (d k)| ≤ -σ₂ * inner ℝ (g f x k) (d k))
    (hg : ∀ k ≥ 1, g f x k ≠ 0) :
    ∀ k ≥ 1, inner ℝ (g f x k) (d k) < 0 ∧
      -(∑ j ∈ Finset.range k, σ₂ ^ j) ≤ inner ℝ (g f x k) (d k) / ‖g f x k‖ ^ 2 ∧
      inner ℝ (g f x k) (d k) / ‖g f x k‖ ^ 2 ≤ -2 + ∑ j ∈ Finset.range k, σ₂ ^ j := by sorry

end NonlinCG.FRBound
Source
Gilbert & Nocedal, Global convergence properties of conjugate gradient methods for optimization, INRIA Rapport de Recherche 1268 (June 1990), HAL inria-00075291v1, p. 7, Lemma 3.1, (3.2); (3.1) on p. 6
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