Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(75), p. 17 — one step of the implicit scheme gives ℒ_{k+1} − ℒ_k ≤ −α_kℒ_{k+1}

Proved
NAGFlow.Implicit.eq_75

by mikedeng1 · Oct 9, 2026 · Mathlib 0df444a (Lean v4.33.1)

convergence-rateconvex-optimizationimplicit-schemelyapunov-functionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Let VVV be a real Hilbert space and f∈Sμ1f\in\mathcal S^1_\muf∈Sμ1​ with μ≥0\mu\ge0μ≥0, i.e. fff is continuously differentiable and f(x)−f(y)−⟨∇f(y),x−y⟩≥μ2∥x−y∥2f(x)-f(y)-\langle\nabla f(y),x-y\rangle\ge\frac\mu2\|x-y\|^2f(x)−f(y)−⟨∇f(y),x−y⟩≥2μ​∥x−y∥2 for all x,yx,yx,y. Let x∗x^*x∗ be a global minimiser of fff, and let Lk=f(xk)−f(x∗)+γk2∥vk−x∗∥2\mathcal L_k=f(x_k)-f(x^*)+\frac{\gamma_k}2\|v_k-x^*\|^2Lk​=f(xk​)−f(x∗)+2γk​​∥vk​−x∗∥2 be the Lyapunov function (74). Fix kkk, suppose γk>0\gamma_k>0γk​>0, and suppose step kkk of the implicit scheme holds: αk>0\alpha_k>0αk​>0,

xk+1−xkαk=vk+1−xk+1,vk+1−vkαk=μγk(xk+1−vk+1)−1γk∇f(xk+1),γk+1−γkαk=μ−γk+1.\frac{x_{k+1}-x_k}{\alpha_k}=v_{k+1}-x_{k+1},\quad \frac{v_{k+1}-v_k}{\alpha_k}=\frac{\mu}{\gamma_k}(x_{k+1}-v_{k+1})-\frac1{\gamma_k}\nabla f(x_{k+1}),\quad \frac{\gamma_{k+1}-\gamma_k}{\alpha_k}=\mu-\gamma_{k+1}.αk​xk+1​−xk​​=vk+1​−xk+1​,αk​vk+1​−vk​​=γk​μ​(xk+1​−vk+1​)−γk​1​∇f(xk+1​),αk​γk+1​−γk​​=μ−γk+1​.

Then

Lk+1−Lk ≤ −αk Lk+1.\mathcal L_{k+1}-\mathcal L_k\ \le\ -\alpha_k\,\mathcal L_{k+1}.Lk+1​−Lk​ ≤ −αk​Lk+1​.

This one-step inequality is equivalent to the contraction Lk+1≤Lk/(1+αk)\mathcal L_{k+1}\le\mathcal L_k/(1+\alpha_k)Lk+1​≤Lk​/(1+αk​) of Theorem 4.1, which the page reduces to it ("It suffices to prove").

Formalization Note. The step is arbitrary: no relation between αk\alpha_kαk​ and γk\gamma_kγk​ is assumed, and the existence of (xk+1,vk+1)(x_{k+1},v_{k+1})(xk+1​,vk+1​) is not claimed. The hypothesis γk>0\gamma_k>0γk​>0 is the one-step form of the run's γ0>0\gamma_0>0γ0​>0 (it propagates through (73) because μ≥0\mu\ge0μ≥0, αk>0\alpha_k>0αk​>0; see Remark 4.1). Inner product and duality pairing are both the real inner product (Riesz); the gradient is an explicit map with HasGradientAt at every point.

Preamble
import Mathlib
import Definitions.Def_NAGFlow_Implicit_Setting
Formal statement
namespace NAGFlow.Implicit

/-- (75), proof of Theorem 4.1, p. 17: let `f ∈ S¹_μ` with gradient map `gradf` and `μ ≥ 0`, and
let `x*` be a global minimiser of `f`. If `γ_k > 0` and step `k` of (72)–(73) holds with
`α_k > 0`, then `ℒ_{k+1} - ℒ_k ≤ -α_k ℒ_{k+1}`. -/
theorem eq_75 {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V] [CompleteSpace V]
    (f : V → ℝ) (gradf : V → V) (μ : ℝ) (hf : NAGFlow.PredCorr.IsS1 f gradf μ)
    (xstar : V) (hxstar : ∀ y, f xstar ≤ f y)
    (α γ : ℕ → ℝ) (x v : ℕ → V) (k : ℕ) (hγ : 0 < γ k)
    (hstep : IsImplicitStep gradf μ α γ x v k) :
    lyap f xstar x v γ (k + 1) - lyap f xstar x v γ k ≤ -α k * lyap f xstar x v γ (k + 1) := by sorry

end NAGFlow.Implicit
Source
Luo & Chen, arXiv:1909.03145v4, (75), proof of Theorem 4.1, p. 17; (2) p. 2; (72)–(74) p. 16
Human review
  • Endorsed by Shuze Chen · Oct 10, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 10, 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