Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 5, p. 8 — under the proximal-PL inequality, proximal gradient with step 1/L1/L1/L gives F(xk)−F∗≤(1−μ/L)k[F(x0)−F∗]F(x_k) - F^* \le (1 - \mu/L)^k [F(x_0) - F^*]F(xk​)−F∗≤(1−μ/L)k[F(x0​)−F∗]

Proved
KarimiPL.Prox.theorem_5

by mikedeng1 · Oct 8, 2026 · Mathlib 0df444a (Lean v4.33.1)

composite-optimizationlinear-convergencep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1polyak-lojasiewiczproximal-gradient

Consider problem (11), minimizing F(x)=f(x)+g(x)F(x) = f(x) + g(x)F(x)=f(x)+g(x) over x∈Rdx \in \mathbb R^dx∈Rd, where:

  1. f:Rd→Rf : \mathbb R^d \to \mathbb Rf:Rd→R is differentiable and its gradient is LLL-Lipschitz continuous, ∥∇f(x)−∇f(y)∥≤L∥x−y∥\|\nabla f(x) - \nabla f(y)\| \le L\|x - y\|∥∇f(x)−∇f(y)∥≤L∥x−y∥, with L>0L > 0L>0;
  2. g:Rd→Rg : \mathbb R^d \to \mathbb Rg:Rd→R is convex;
  3. FFF has a non-empty solution set: there is x∗x^*x∗ with F(x∗)≤F(y)F(x^*) \le F(y)F(x∗)≤F(y) for all yyy; write F∗=F(x∗)F^* = F(x^*)F∗=F(x∗);
  4. FFF satisfies the proximal-PL inequality (12) with some μ>0\mu > 0μ>0: 12Dg(x,L)≥μ(F(x)−F∗)\tfrac12\mathcal D_g(x, L) \ge \mu(F(x) - F^*)21​Dg​(x,L)≥μ(F(x)−F∗) for all xxx, where
Dg(x,α)=−2αmin⁡y[⟨∇f(x),y−x⟩+α2∥y−x∥2+g(y)−g(x)].\mathcal D_g(x, \alpha) = -2\alpha \min_y \Bigl[\langle \nabla f(x), y - x\rangle + \frac{\alpha}{2}\|y - x\|^2 + g(y) - g(x)\Bigr].Dg​(x,α)=−2αymin​[⟨∇f(x),y−x⟩+2α​∥y−x∥2+g(y)−g(x)].

Let (xk)k≥0(x_k)_{k \ge 0}(xk​)k≥0​ be generated by the proximal-gradient method with step size 1/L1/L1/L from an arbitrary x0x_0x0​:

xk+1∈argmin⁡y[⟨∇f(xk),y−xk⟩+L2∥y−xk∥2+g(y)−g(xk)].(14)x_{k+1} \in \operatorname*{argmin}_y \Bigl[\langle \nabla f(x_k), y - x_k\rangle + \frac L2\|y - x_k\|^2 + g(y) - g(x_k)\Bigr]. \tag{14}xk+1​∈yargmin​[⟨∇f(xk​),y−xk​⟩+2L​∥y−xk​∥2+g(y)−g(xk​)].(14)

Then for every k≥0k \ge 0k≥0,

F(xk)−F∗≤(1−μL)k[F(x0)−F∗].F(x_k) - F^* \le \Bigl(1 - \frac{\mu}{L}\Bigr)^k \bigl[F(x_0) - F^*\bigr].F(xk​)−F∗≤(1−Lμ​)k[F(x0​)−F∗].

This is the paper's linear convergence result for the proximal-gradient method without strong convexity. When ggg is constant it reduces to Theorem 1 for gradient descent under the PL inequality.

Formalization Note ggg is real-valued, so the indicator functions mentioned on p. 8 as examples of ggg are outside the formal scope. The update (14) is encoded by requiring each xk+1x_{k+1}xk+1​ to be a minimizer; the minimizer is unique for convex ggg, but uniqueness is not used. No relation between μ\muμ and LLL is assumed, as on the page.

Preamble
import Mathlib
import Definitions.Def_KarimiPL_Prox_Setting

open InnerProductSpace
Formal statement
namespace KarimiPL.Prox

theorem theorem_5 {d : ℕ} (f g : EuclideanSpace ℝ (Fin d) → ℝ)
    (hdiff : Differentiable ℝ f) (L : ℝ) (hL : 0 < L)
    (hLip : ∀ x y, ‖gradient f x - gradient f y‖ ≤ L * ‖x - y‖)
    (hg : ConvexOn ℝ Set.univ g)
    (xstar : EuclideanSpace ℝ (Fin d)) (hxstar : ∀ y, f xstar + g xstar ≤ f y + g y)
    (μ : ℝ) (hμ : 0 < μ) (hPPL : ProxPL f g L μ (f xstar + g xstar))
    (x : ℕ → EuclideanSpace ℝ (Fin d)) (hrun : IsProxGradRun f g L x) :
    ∀ k : ℕ, (f (x k) + g (x k)) - (f xstar + g xstar)
      ≤ (1 - μ / L) ^ k * ((f (x 0) + g (x 0)) - (f xstar + g xstar)) := by sorry

end KarimiPL.Prox
Source
Karimi, Nutini, Schmidt, arXiv:1608.04636v4, Theorem 5, p. 8
Human review
  • Endorsed by Shuze Chen · Oct 9, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 9, 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