Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4 — Assumption 2(a) holds if X̄ ≠ ∅ and C1, C2 or C3 holds, or if f is strongly convex with Lipschitz gradient (C4)

Proved
TsengCGD.ErrorBound.theorem4

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

coordinate-descenterror-boundnonsmooth-optimizationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Assume the standing assumptions of problem (1): c>0c > 0c>0, P:Rn→(−∞,∞]P:\mathbb R^n\to(-\infty,\infty]P:Rn→(−∞,∞] proper, convex and lower semicontinuous, fff continuously differentiable on an open set containing dom⁡P\operatorname{dom}PdomP, and Fc=f+cPF_c = f + cPFc​=f+cP. Let Xˉ\bar XXˉ be the set of stationary points of FcF_cFc​ and dI(x)d_I(x)dI​(x) the residual (13).

Suppose that either

  1. Xˉ≠∅\bar X \neq \emptysetXˉ=∅ and one of the conditions C1, C2, C3 of Lemma 7 holds (PPP polyhedral, and fff quadratic, or f(x)=g(Ex)+q⊤xf(x) = g(Ex) + q^\top xf(x)=g(Ex)+q⊤x, or f(x)=max⁡y∈Y{(Ex)⊤y−g(y)}+q⊤xf(x) = \max_{y\in Y}\{(Ex)^\top y - g(y)\} + q^\top xf(x)=maxy∈Y​{(Ex)⊤y−g(y)}+q⊤x with ggg strongly convex with Lipschitz gradient and YYY polyhedral); or
  2. C4: fff is strongly convex and satisfies (22), ∥∇f(y)−∇f(z)∥≤L∥y−z∥\|\nabla f(y) - \nabla f(z)\| \le L\|y - z\|∥∇f(y)−∇f(z)∥≤L∥y−z∥ on dom⁡P\operatorname{dom}PdomP, for some L≥0L \ge 0L≥0.

Then Assumption 2(a) holds: Xˉ≠∅\bar X \neq \emptysetXˉ=∅ and for every ζ\zetaζ there are τ>0\tau > 0τ>0 and ϵ>0\epsilon > 0ϵ>0 such that

dist⁡(x,Xˉ)≤τ ∥dI(x)∥whenever x∈dom⁡P, Fc(x)≤ζ, ∥dI(x)∥≤ϵ.\operatorname{dist}(x,\bar X) \le \tau\,\|d_I(x)\| \quad\text{whenever } x \in \operatorname{dom}P,\ F_c(x) \le \zeta,\ \|d_I(x)\| \le \epsilon.dist(x,Xˉ)≤τ∥dI​(x)∥whenever x∈domP, Fc​(x)≤ζ, ∥dI​(x)∥≤ϵ.

This is the main result of §6 of the paper: it identifies problem classes for which the linear convergence theorem of the coordinate gradient descent method applies, among them ℓ1\ell_1ℓ1​-regularized least squares and other problems with polyhedral PPP and quadratic or composite fff.

Formalization Note Under C4 the conclusion includes Xˉ≠∅\bar X \neq \emptysetXˉ=∅, which is not assumed (the paper's proof notes that Xˉ\bar XXˉ is then a singleton). In C4 strong convexity and (22) are required on dom⁡P\operatorname{dom}PdomP only, which is weaker than the paper's "strongly convex" on Rn\mathbb R^nRn, so the statement is at least as strong as printed. Assumption 2(a) quantifies over every real ζ\zetaζ; the paper's "ζ≥min⁡xFc(x)\zeta \ge \min_x F_c(x)ζ≥minx​Fc​(x)" agrees with this since the condition is vacuous below inf⁡Fc\inf F_cinfFc​. dist⁡\operatorname{dist}dist is Metric.infDist.

Preamble
import Mathlib
import Definitions.Def_TsengCGD_Linear_Basic
import Definitions.Def_TsengCGD_ErrorBound_Conditions
Formal statement
namespace TsengCGD.ErrorBound

theorem theorem4 {n : ℕ} (f : TsengCGD.Global.Vec n → ℝ) (D : Set (TsengCGD.Global.Vec n)) (P : TsengCGD.Global.Vec n → ℝ) (c : ℝ)
    (hs : TsengCGD.Global.Standing f D P c)
    (hC : ((TsengCGD.Linear.statSet f D P c).Nonempty ∧ (C1 f D P ∨ C2 f D P ∨ C3 f D P)) ∨ C4 f D) :
    TsengCGD.Linear.Assumption2a f D P c := by sorry

end TsengCGD.ErrorBound
Source
Tseng and Yun, A coordinate gradient descent method for nonsmooth separable minimization, Math. Program. Ser. B 117 (2009), p. 412, Theorem 4 (Assumption 2(a), p. 404)
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