Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 1 — AG on smooth Ψ: min‖∇Ψ(x^md_k)‖² ≤ (Ψ(x_0) − Ψ*)/Σλ_kC_k; for convex Ψ, Ψ(x^ag_N) − Ψ(x*) ≤ Γ_N‖x_0 − x*‖²/(2λ_1)

Proved
NonconvexAG.Smooth.theorem_1

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

accelerated-gradientnonconvex-optimizationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let Ψ:Rn→R\Psi:\mathbb R^n\to\mathbb RΨ:Rn→R be differentiable, bounded from below, with ∥∇Ψ(y)−∇Ψ(x)∥≤LΨ∥y−x∥\|\nabla\Psi(y)-\nabla\Psi(x)\|\le L_\Psi\|y-x\|∥∇Ψ(y)−∇Ψ(x)∥≤LΨ​∥y−x∥ for all x,yx,yx,y, where LΨ>0L_\Psi>0LΨ​>0, and let Ψ∗=inf⁡xΨ(x)\Psi^*=\inf_x\Psi(x)Ψ∗=infx​Ψ(x). Let {xk,xkmd,xkag}\{x_k,x^{md}_k,x^{ag}_k\}{xk​,xkmd​,xkag​} be computed by Algorithm 1 from x0x_0x0​ with step sizes α1=1\alpha_1=1α1​=1, αk∈(0,1)\alpha_k\in(0,1)αk​∈(0,1) (k≥2k\ge2k≥2), βk>0\beta_k>0βk​>0, λk>0\lambda_k>0λk​>0, and let Γk\Gamma_kΓk​ be defined in (2.6). Fix N≥1N\ge1N≥1.

  1. If Ck>0C_k>0Ck​>0 for k=1,…,Nk=1,\dots,Nk=1,…,N, where
Ck=1−LΨλk−LΨ(λk−βk)22αkΓkλk(∑τ=kNΓτ)(2.7),C_k=1-L_\Psi\lambda_k-\frac{L_\Psi(\lambda_k-\beta_k)^2}{2\alpha_k\Gamma_k\lambda_k}\Big(\sum_{\tau=k}^N\Gamma_\tau\Big)\qquad(2.7),Ck​=1−LΨ​λk​−2αk​Γk​λk​LΨ​(λk​−βk​)2​(τ=k∑N​Γτ​)(2.7),

then

min⁡k=1,…,N∥∇Ψ(xkmd)∥2≤Ψ(x0)−Ψ∗∑k=1NλkCk.(2.8)\min_{k=1,\dots,N}\|\nabla\Psi(x^{md}_k)\|^2\le\frac{\Psi(x_0)-\Psi^*}{\sum_{k=1}^N\lambda_kC_k}.\qquad(2.8)k=1,…,Nmin​∥∇Ψ(xkmd​)∥2≤∑k=1N​λk​Ck​Ψ(x0​)−Ψ∗​.(2.8)
  1. Suppose moreover that Ψ\PsiΨ is convex and that x∗x^*x∗ is a minimizer of Ψ\PsiΨ on Rn\mathbb R^nRn. If for k=1,…,Nk=1,\dots,Nk=1,…,N
αkλk≤βk<1LΨ(2.9),α1λ1Γ1≥α2λ2Γ2≥⋯≥αNλNΓN(2.10),\alpha_k\lambda_k\le\beta_k<\frac1{L_\Psi}\quad(2.9),\qquad \frac{\alpha_1}{\lambda_1\Gamma_1}\ge\frac{\alpha_2}{\lambda_2\Gamma_2}\ge\dots\ge\frac{\alpha_N}{\lambda_N\Gamma_N}\quad(2.10),αk​λk​≤βk​<LΨ​1​(2.9),λ1​Γ1​α1​​≥λ2​Γ2​α2​​≥⋯≥λN​ΓN​αN​​(2.10),

then

min⁡k=1,…,N∥∇Ψ(xkmd)∥2≤∥x0−x∗∥2λ1∑k=1NΓk−1βk(1−LΨβk),(2.11)\min_{k=1,\dots,N}\|\nabla\Psi(x^{md}_k)\|^2\le\frac{\|x_0-x^*\|^2}{\lambda_1\sum_{k=1}^N\Gamma_k^{-1}\beta_k(1-L_\Psi\beta_k)},\qquad(2.11)k=1,…,Nmin​∥∇Ψ(xkmd​)∥2≤λ1​∑k=1N​Γk−1​βk​(1−LΨ​βk​)∥x0​−x∗∥2​,(2.11) Ψ(xNag)−Ψ(x∗)≤ΓN∥x0−x∗∥22λ1.(2.12)\Psi(x^{ag}_N)-\Psi(x^*)\le\frac{\Gamma_N\|x_0-x^*\|^2}{2\lambda_1}.\qquad(2.12)Ψ(xNag​)−Ψ(x∗)≤2λ1​ΓN​∥x0​−x∗∥2​.(2.12)

Part 1 shows that the accelerated gradient method, with a suitable step-size policy, drives the gradient to zero on a nonconvex smooth problem at the rate of gradient descent; part 2 shows that the same method keeps the optimal O(1/N2)O(1/N^2)O(1/N2) rate of Nesterov's method when Ψ\PsiΨ happens to be convex (for instance with αk=2/(k+1)\alpha_k=2/(k+1)αk​=2/(k+1), ΓN=2/(N(N+1))\Gamma_N=2/(N(N+1))ΓN​=2/(N(N+1))).

Formalization Note The gradient is an explicit map g with IsBetaSmooth Ψ g LΨ (differentiability, gradient =g=g=g, and (1.2)). LΨ>0L_\Psi>0LΨ​>0 as in (1.2). "Bounded from below" is the standing assumption of §2.1 (BddBelow (Set.range Ψ)), and Ψ∗\Psi^*Ψ∗ is the infimum ⨅ x, Ψ x. The minimum over k=1,…,Nk=1,\dots,Nk=1,…,N is Finset.inf' over Finset.Icc 1 N. CkC_kCk​ depends on NNN. Conditions (2.7), (2.9) and (2.10) are imposed for k=1,…,Nk=1,\dots,Nk=1,…,N only, the indices the proof uses. The two parts are the two conjuncts of the conclusion.

Preamble
import Mathlib
import Definitions.Def_ConvexOptAlg_SmoothGD_Defs
import Definitions.Def_NonconvexAG_Smooth_AGRun
Formal statement
namespace NonconvexAG.Smooth

/-- Theorem 1 (p. 5): a) and b). -/
theorem theorem_1 {n : ℕ} (Ψ : E n → ℝ) (g : E n → E n) (LΨ : ℝ)
    (hΨ : ConvexOptAlg.SmoothGD.IsBetaSmooth Ψ g LΨ) (hL : 0 < LΨ)
    (α β lam : ℕ → ℝ) (hstep : AGStepsizes α β lam) (x0 : E n)
    (hbdd : BddBelow (Set.range Ψ)) (N : ℕ) (hN : 1 ≤ N) :
    -- a) under (2.7) for `k = 1, …, N`: (2.8)
    ((∀ k ∈ Finset.Icc 1 N, 0 < C LΨ α β lam N k) →
      (Finset.Icc 1 N).inf' (Finset.nonempty_Icc.mpr hN) (fun k => ‖g (xmdSeq g α β lam x0 k)‖ ^ 2) ≤
        (Ψ x0 - ⨅ x, Ψ x) / ∑ k ∈ Finset.Icc 1 N, lam k * C LΨ α β lam N k) ∧
    -- b) `Ψ` convex with a minimizer `x*`, under (2.9) and (2.10) for `k = 1, …, N`: (2.11), (2.12)
    (∀ xstar : E n, ConvexOn ℝ Set.univ Ψ → (∀ x, Ψ xstar ≤ Ψ x) →
      (∀ k ∈ Finset.Icc 1 N, α k * lam k ≤ β k ∧ β k < 1 / LΨ) →
      (∀ k ∈ Finset.Icc 1 N, 2 ≤ k →
        α k / (lam k * Gamma α k) ≤ α (k - 1) / (lam (k - 1) * Gamma α (k - 1))) →
      (Finset.Icc 1 N).inf' (Finset.nonempty_Icc.mpr hN) (fun k => ‖g (xmdSeq g α β lam x0 k)‖ ^ 2) ≤
          ‖x0 - xstar‖ ^ 2 /
            (lam 1 * ∑ k ∈ Finset.Icc 1 N, (Gamma α k)⁻¹ * β k * (1 - LΨ * β k)) ∧
        Ψ (xagSeq g α β lam x0 N) - Ψ xstar ≤ Gamma α N * ‖x0 - xstar‖ ^ 2 / (2 * lam 1)) := by sorry

end NonconvexAG.Smooth
Source
Ghadimi & Lan, Accelerated Gradient Methods for Nonconvex Nonlinear and Stochastic Programming, arXiv:1310.3787v1, p. 5, Theorem 1 (with §1 (1.1)–(1.2), p. 1, and the standing assumption of §2.1, p. 4)
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