Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 2.2(ii) (2.13), p. 234 — ε(z^N) ≤ (Θ(z₀) + Σε_t)/Σγ_t for the relaxed CPM

Proved
ProxMethodVI.Rate.prop2_2_ii_ergodic_bound

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

convergence-ratemirror-proxp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1prox-methodvariational-inequality

Let ZZZ be a nonempty convex compact subset of a finite-dimensional normed space EEE, let FFF be monotone on ZZZ, and let ω\omegaω be continuously differentiable on ZZZ and α\alphaα-strongly convex in the sense of (2.2). Consider a run of the relaxed conceptual prox-method: z0∈Zz_0\in Zz0​∈Z, and for every t≥1t\ge1t≥1, γt>0\gamma_t>0γt​>0, wt∈Zw_t\in Zwt​∈Z, zt=Pzt−1(γtF(wt))z_t=P_{z_{t-1}}(\gamma_tF(w_t))zt​=Pzt−1​​(γt​F(wt​)) and

⟨γtF(wt),wt−zt⟩+ω(zt−1)+⟨ω′(zt−1),zt−zt−1⟩−ω(zt)≤ϵt.(2.11)\langle\gamma_tF(w_t),w_t-z_t\rangle+\omega(z_{t-1})+\langle\omega'(z_{t-1}),z_t-z_{t-1}\rangle-\omega(z_t)\le\epsilon_t.\tag{2.11}⟨γt​F(wt​),wt​−zt​⟩+ω(zt−1​)+⟨ω′(zt−1​),zt​−zt−1​⟩−ω(zt​)≤ϵt​.(2.11)

Let N≥1N\ge1N≥1 and zN=(∑t=1Nγt)−1∑t=1Nγtwtz^N=\big(\sum_{t=1}^N\gamma_t\big)^{-1}\sum_{t=1}^N\gamma_tw_tzN=(∑t=1N​γt​)−1∑t=1N​γt​wt​. Then

ϵ(zN)≡max⁡u∈Z⟨F(u),zN−u⟩≤Θ(z0)+∑t=1Nϵt∑t=1Nγt,Θ(z0)=max⁡z∈Z[ω(z)−ω(z0)−⟨ω′(z0),z−z0⟩].\epsilon(z^N)\equiv\max_{u\in Z}\langle F(u),z^N-u\rangle\le\frac{\Theta(z_0)+\sum_{t=1}^N\epsilon_t}{\sum_{t=1}^N\gamma_t},\qquad \Theta(z_0)=\max_{z\in Z}\big[\omega(z)-\omega(z_0)-\langle\omega'(z_0),z-z_0\rangle\big].ϵ(zN)≡u∈Zmax​⟨F(u),zN−u⟩≤∑t=1N​γt​Θ(z0​)+∑t=1N​ϵt​​,Θ(z0​)=z∈Zmax​[ω(z)−ω(z0​)−⟨ω′(z0​),z−z0​⟩].

This is the convergence guarantee of the conceptual prox-method. Its quantity ϵ(zN)\epsilon(z^N)ϵ(zN) is the standard accuracy measure of an approximate weak solution of the variational inequality, and the basic implementation of §3 is analysed by verifying (2.11) with ϵt=0\epsilon_t=0ϵt​=0.

Formalization Note. The bound on ϵ(zN)\epsilon(z^N)ϵ(zN) is stated as ⟨F(u),zN−u⟩≤bound\langle F(u),z^N-u\rangle\le\text{bound}⟨F(u),zN−u⟩≤bound for every u∈Zu\in Zu∈Z, which is equivalent to the maximum being at most the bound. Only monotonicity of FFF is assumed; the Lipschitz half of (2.1) is not used by the proof. The hypothesis "the relaxed algorithm does not terminate in the course of NNN steps" is replaced by runs that never terminate, since (2.8) is not modelled. N≥1N\ge1N≥1 makes ∑γt>0\sum\gamma_t>0∑γt​>0.

Preamble
import Mathlib
import Definitions.Def_ProxMethodVI_Rate_Setting
Formal statement
namespace ProxMethodVI.Rate

theorem prop2_2_ii_ergodic_bound {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
    (Z : Set E) (hZc : Convex ℝ Z) (hZk : IsCompact Z) (hZne : Z.Nonempty)
    (ω : E → ℝ) (ω' : E → E →L[ℝ] ℝ) (α : ℝ) (hω : IsStrongDGF Z ω ω' α)
    (F : E → E →L[ℝ] ℝ) (hmono : IsMonotoneOnWRT Z F)
    (γ ε : ℕ → ℝ) (z w : ℕ → E) (hrun : IsRelaxedRun Z F ω ω' γ ε z w)
    (N : ℕ) (hN : 1 ≤ N) (u : E) (hu : u ∈ Z) :
    F u (ergodicAvg γ w N - u)
      ≤ (theta Z ω ω' (z 0) + ∑ t ∈ Finset.Icc 1 N, ε t) / ∑ t ∈ Finset.Icc 1 N, γ t := by sorry

end ProxMethodVI.Rate
Source
Nemirovski, Prox-Method with Rate of Convergence O(1/t), SIAM J. Optim. 15(1) (2004), p. 234, Proposition 2.2(ii), (2.11), (2.13)
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 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