Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§7.2.2, p. 30 — every run of Algorithm 2 (Semi-APGM) is a run of the semi-implicit scheme (115)

Proved
NAGFlow.APGM.algorithm_2_eq_115

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

accelerated-gradientode-discretizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1proximal-gradient

Let VVV be a real Hilbert space, μ≥0\mu\ge0μ≥0, L>0L>0L>0, ∇h:V→V\nabla h:V\to V∇h:V→V a map and g:V→R∪{+∞}g:V\to\mathbb R\cup\{+\infty\}g:V→R∪{+∞} with domain DDD. Let (αk,γk,xk,yk,wk,vk)(\alpha_k,\gamma_k,x_k,y_k,w_k,v_k)(αk​,γk​,xk​,yk​,wk​,vk​) be generated by Algorithm 2 (Semi-APGM): γ0>0\gamma_0>0γ0​>0 and, for every kkk, αk>0\alpha_k>0αk​>0 with Lαk2=γk(1+αk)L\alpha_k^2=\gamma_k(1+\alpha_k)Lαk2​=γk​(1+αk​),

γk+1=γk+μαk1+αk,yk=xk+αkvk1+αk,wk=γkvk+μαkykγk+μαk,\gamma_{k+1}=\frac{\gamma_k+\mu\alpha_k}{1+\alpha_k},\quad y_k=\frac{x_k+\alpha_kv_k}{1+\alpha_k},\quad w_k=\frac{\gamma_kv_k+\mu\alpha_ky_k}{\gamma_k+\mu\alpha_k},γk+1​=1+αk​γk​+μαk​​,yk​=1+αk​xk​+αk​vk​​,wk​=γk​+μαk​γk​vk​+μαk​yk​​, xk+1=prox⁡g/L(yk−1L∇h(yk)),vk+1=wk+γkγk+1 xk+1−ykαk.x_{k+1}=\operatorname{prox}_{g/L}\big(y_k-\tfrac1L\nabla h(y_k)\big),\qquad v_{k+1}=w_k+\frac{\gamma_k}{\gamma_{k+1}}\,\frac{x_{k+1}-y_k}{\alpha_k}.xk+1​=proxg/L​(yk​−L1​∇h(yk​)),vk+1​=wk​+γk+1​γk​​αk​xk+1​−yk​​.

Then (xk,yk,vk,γk)(x_k,y_k,v_k,\gamma_k)(xk​,yk​,vk​,γk​) satisfies the semi-implicit scheme (115): for every kkk,

yk−xkαk=vk−yk,xk+1=Sf(yk),vk+1−vkαk=μγk(yk−vk+1)−1γkGf(yk),γk+1−γkαk=μ−γk+1,\frac{y_k-x_k}{\alpha_k}=v_k-y_k,\qquad x_{k+1}=S_f(y_k),\qquad \frac{v_{k+1}-v_k}{\alpha_k}=\frac{\mu}{\gamma_k}(y_k-v_{k+1})-\frac{1}{\gamma_k}\mathcal G_f(y_k),\qquad \frac{\gamma_{k+1}-\gamma_k}{\alpha_k}=\mu-\gamma_{k+1},αk​yk​−xk​​=vk​−yk​,xk+1​=Sf​(yk​),αk​vk+1​−vk​​=γk​μ​(yk​−vk+1​)−γk​1​Gf​(yk​),αk​γk+1​−γk​​=μ−γk+1​,

where Gf(yk)=L(yk−xk+1)\mathcal G_f(y_k)=L(y_k-x_{k+1})Gf​(yk​)=L(yk​−xk+1​).

This is the statement that Algorithm 2 "summarizes" (115): it lets every estimate proved for (115) be applied to the algorithm. The vvv-update matches because of the step-size relation Lαk2=γk(1+αk)L\alpha_k^2=\gamma_k(1+\alpha_k)Lαk2​=γk​(1+αk​).

Formalization Note. Only ∇h\nabla h∇h enters the scheme, so no class assumption on hhh or ggg is needed. The prox is the predicate of the definitions file (a minimiser over dom⁡g\operatorname{dom}gdomg).

Preamble
import Mathlib
import Definitions.Def_NAGFlow_APGM_Setting
Formal statement
namespace NAGFlow.APGM

/-- Algorithm 2 is the semi-implicit scheme (115) (Luo & Chen, arXiv:1909.03145v4, §7.2.2, p. 30).
Let `0 ≤ μ` and `0 < L`. Every run `(α, γ, x, y, w, v)` of Algorithm 2 (Semi-APGM, with `γ₀ > 0`,
`α_k > 0`, `Lα_k² = γ_k(1 + α_k)`, `η = 1/L`) satisfies (115): for every `k`,
`(y_k − x_k)/α_k = v_k − y_k`, `x_{k+1} = S_f(y_k)`,
`(v_{k+1} − v_k)/α_k = (μ/γ_k)(y_k − v_{k+1}) − (1/γ_k)G_f(y_k)` with `G_f(y_k) = L(y_k − x_{k+1})`, and
`(γ_{k+1} − γ_k)/α_k = μ − γ_{k+1}`. -/
theorem algorithm_2_eq_115 {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]
    (gradh : V → V) (g : V → ℝ) (D : Set V) (μ L : ℝ) (hμ : 0 ≤ μ) (hL : 0 < L)
    (α γ : ℕ → ℝ) (x y w v : ℕ → V) (hrun : IsAPGMRun gradh g D μ L α γ x y w v) :
    IsSemiImplicitRun gradh g D μ L α γ x y v := by sorry

end NAGFlow.APGM
Source
Luo & Chen, arXiv:1909.03145v4, §7.2.2, (115) and Algorithm 2, p. 30
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