Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 29 — β(t)/3≤α(t)≤β(t)\beta(t)/\sqrt3\le\alpha(t)\le\beta(t)β(t)/3​≤α(t)≤β(t)

Proved
NonconvexSaddle.PSGD.lemma_29

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

nonconvex-optimizationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1saddle-points

Let a=ηγ∈(0,1]a=\eta\gamma\in(0,1]a=ηγ∈(0,1] and define

α(t)=[∑τ=0t−1(1+a)2(t−1−τ)]1/2,β(t)=(1+a)t2a.\alpha(t)=\Big[\sum_{\tau=0}^{t-1}(1+a)^{2(t-1-\tau)}\Big]^{1/2},\qquad \beta(t)=\frac{(1+a)^t}{\sqrt{2a}}.α(t)=[τ=0∑t−1​(1+a)2(t−1−τ)]1/2,β(t)=2a​(1+a)t​.

Then (1) α(t)≤β(t)\alpha(t)\le\beta(t)α(t)≤β(t) for every t∈Nt\in\mathbb Nt∈N, and (2) α(t)≥β(t)/3\alpha(t)\ge\beta(t)/\sqrt3α(t)≥β(t)/3​ for every t∈Nt\in\mathbb Nt∈N with t≥ln⁡(2)/at\ge\ln(2)/at≥ln(2)/a.

α(t)\alpha(t)α(t) is, up to the factor 2ηr/d2\eta r/\sqrt d2ηr/d​, the standard deviation of the perturbation term qp(t)q_p(t)qp​(t) along e1e_1e1​; the lemma says it grows like (1+ηγ)t(1+\eta\gamma)^t(1+ηγ)t.

Formalization Note The page allows ηγ∈[0,1]\eta\gamma\in[0,1]ηγ∈[0,1]; at ηγ=0\eta\gamma=0ηγ=0 the paper's β(t)\beta(t)β(t) is +∞+\infty+∞, while Lean's 1/0=01/\sqrt0=01/0​=0 would make part (1) false for t≥1t\ge1t≥1, so ηγ>0\eta\gamma>0ηγ>0 is required. Lemmas 30 and 31 apply the lemma with ηγ>0\eta\gamma>0ηγ>0.

Preamble
import Mathlib
import Definitions.Def_NonconvexSaddle_PSGD_Setting

open MeasureTheory ProbabilityTheory
open scoped RealInnerProductSpace
Formal statement
namespace NonconvexSaddle.PSGD

/-- Lemma 29 (arXiv:1902.04811v2, App. B.3, p. 26), with `a = ηγ ∈ (0, 1]`:
(1) `α(t) ≤ β(t)` for every `t ∈ ℕ`, and (2) `α(t) ≥ β(t)/√3` for `t ≥ ln 2/(ηγ)`. -/
theorem lemma_29 (a : ℝ) (ha : 0 < a) (ha1 : a ≤ 1) :
    (∀ t : ℕ, alphaL a t ≤ betaL a t) ∧
      ∀ t : ℕ, Real.log 2 / a ≤ (t : ℝ) → betaL a t / Real.sqrt 3 ≤ alphaL a t := by sorry

end NonconvexSaddle.PSGD
Source
Jin, Netrapalli, Ge, Kakade, Jordan, On Nonconvex Optimization for Machine Learning: Gradients, Stochasticity, and Saddle Points, arXiv:1902.04811v2, p. 26, Lemma 29
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