Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(2.10), p. 7 — under A1, E[⟨∇f(x_k), δ_k⟩ | ξ_[k−1]] = 0

Proved
GhadimiLan.RSG.eq_2_10

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

martingalesnonconvex-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1sgdstochastic-approximation

Let f∈CL1,1(Rn)f\in\mathcal C^{1,1}_L(\mathbb R^n)f∈CL1,1​(Rn), let GGG be a Borel oracle, and let x1,x2,…x_1,x_2,\dotsx1​,x2​,… be an RSG run from x1x_1x1​ satisfying Assumption A1 with respect to a filtration (Fk)(\mathcal F_k)(Fk​) (noise adapted; conditional unbiasedness E[G(xk,ξk)∣Fk−1]=∇f(xk)\mathbb E[G(x_k,\xi_k)\mid\mathcal F_{k-1}]=\nabla f(x_k)E[G(xk​,ξk​)∣Fk−1​]=∇f(xk​); E∥G(xk,ξk)−∇f(xk)∥2≤σ2\mathbb E\|G(x_k,\xi_k)-\nabla f(x_k)\|^2\le\sigma^2E∥G(xk​,ξk​)−∇f(xk​)∥2≤σ2). Write δk=G(xk,ξk)−∇f(xk)\delta_k=G(x_k,\xi_k)-\nabla f(x_k)δk​=G(xk​,ξk​)−∇f(xk​). Then for every k≥1k\ge1k≥1 the random variable ⟨∇f(xk),δk⟩\langle\nabla f(x_k),\delta_k\rangle⟨∇f(xk​),δk​⟩ is integrable and

E[⟨∇f(xk),δk⟩ ∣ Fk−1]=0almost surely.\mathbb E\big[\langle\nabla f(x_k),\delta_k\rangle\,\big|\,\mathcal F_{k-1}\big]=0\quad\text{almost surely}.E[⟨∇f(xk​),δk​⟩​Fk−1​]=0almost surely.

The cross term in the summed descent inequality (2.9) therefore has mean zero, which is what lets the expectation of (2.9) be bounded without assuming the noise is independent.

Formalization Note In the paper the conditioning is on the history ξ[k−1]\xi_{[k-1]}ξ[k−1]​; any filtration to which the noise is adapted is allowed here, the history filtration being one of them. No moment assumption on ∇f(xk)\nabla f(x_k)∇f(xk​) is made: its integrability is part of the claim.

Preamble
import Mathlib
import Definitions.Def_ConvexOptAlg_SmoothGD_Defs
import Definitions.Def_GhadimiLan_RSG_Model
open MeasureTheory ProbabilityTheory
open scoped InnerProductSpace
Formal statement
namespace GhadimiLan.RSG

/-- Eq. (2.10) (Ghadimi & Lan, arXiv:1309.5549v1, proof of Theorem 2.1, p. 7): under Assumption
A1 along an RSG run, for every `k ≥ 1` the scalar `⟨∇f(x_k), δ_k⟩` (with
`δ_k = G(x_k, ξ_k) − ∇f(x_k)`) is integrable and `E[⟨∇f(x_k), δ_k⟩ | ℱ_{k-1}] = 0` almost surely.
No integrability of `∇f(x_k)` is assumed: it is part of the claim. -/
theorem eq_2_10 {n : ℕ} (f : E n → ℝ) (g : E n → E n) (L : ℝ)
    (hf : ConvexOptAlg.SmoothGD.IsBetaSmooth f g L)
    {Ξ : Type*} [MeasurableSpace Ξ] (G : E n → Ξ → E n) (hG : Measurable (Function.uncurry G))
    {Ω : Type*} [MeasurableSpace Ω] (μ : Measure Ω) [IsProbabilityMeasure μ]
    (ℱ : Filtration ℕ ‹MeasurableSpace Ω›) (ξ : ℕ → Ω → Ξ) (σ : ℝ)
    (γ : ℕ → ℝ) (x1 : E n) (x : ℕ → Ω → E n) (hx : IsRSGRun G γ x1 ξ x)
    (hA1 : AssumptionA1 μ ℱ g G ξ x σ) (k : ℕ) (hk : 1 ≤ k) :
    Integrable (fun ω => ⟪g (x k ω), rsgNoise G g ξ x k ω⟫_ℝ) μ ∧
    μ[fun ω => ⟪g (x k ω), rsgNoise G g ξ x k ω⟫_ℝ | ℱ (k - 1)] =ᵐ[μ] 0 := by sorry

end GhadimiLan.RSG
Source
Ghadimi & Lan, arXiv:1309.5549v1, proof of Theorem 2.1, Eq. (2.10), p. 7
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 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