Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(32) — the one-step bound with the bias δ = g − ∇_θφ_γ separated, for α ≤ 1/L_φ

Proved
CertDRO.SGD.eq_32

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

biased-gradientdescent-lemmap2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let F:Rd→RF:\mathbb R^d\to\mathbb RF:Rd→R be differentiable with LφL_\varphiLφ​-Lipschitz gradient, Lφ≥0L_\varphi\ge0Lφ​≥0, and let 0≤α0\le\alpha0≤α with Lφα≤1L_\varphi\alpha\le1Lφ​α≤1. For all θ,g,u∈Rd\theta,g,u\in\mathbb R^dθ,g,u∈Rd, writing δ=g−u\delta=g-uδ=g−u,

F(θ−αg)≤F(θ)−α2∥∇F(θ)∥22+α(1−Lφα)⟨∇F(θ),∇F(θ)−u⟩+α(1+Lφα)2∥δ∥22+Lφα2∥u−∇F(θ)∥22.F(\theta-\alpha g)\le F(\theta)-\frac\alpha2\|\nabla F(\theta)\|_2^2+\alpha(1-L_\varphi\alpha)\langle\nabla F(\theta),\nabla F(\theta)-u\rangle+\frac{\alpha(1+L_\varphi\alpha)}2\|\delta\|_2^2+L_\varphi\alpha^2\|u-\nabla F(\theta)\|_2^2 .F(θ−αg)≤F(θ)−2α​∥∇F(θ)∥22​+α(1−Lφ​α)⟨∇F(θ),∇F(θ)−u⟩+2α(1+Lφ​α)​∥δ∥22​+Lφ​α2∥u−∇F(θ)∥22​.

In the proof of Theorem 2, u=∇θφγ(θt;zt)u=\nabla_\theta\varphi_\gamma(\theta^t;z^t)u=∇θ​φγ​(θt;zt) is the exact stochastic gradient and δt=gt−u\delta^t=g^t-uδt=gt−u is the bias caused by solving the inner maximization only approximately; the inequality separates a mean-zero term, the bias and the variance.

Formalization Note The paper assumes αt≤1/Lφ\alpha_t\le1/L_\varphiαt​≤1/Lφ​ "in the rest of the proof" (p. 40); this is the hypothesis Lφα≤1L_\varphi\alpha\le1Lφ​α≤1 together with α≥0\alpha\ge0α≥0 and Lφ≥0L_\varphi\ge0Lφ​≥0, which the step ±⟨a,b⟩≤12∥a∥2+12∥b∥2\pm\langle a,b\rangle\le\frac12\|a\|^2+\frac12\|b\|^2±⟨a,b⟩≤21​∥a∥2+21​∥b∥2 needs.

Preamble
import Definitions.Def_CertDRO_SGD_model
Formal statement
namespace CertDRO.SGD

/-- (32), §B.3, p. 40: for a differentiable `F` on `ℝ^d` with `L_φ`-Lipschitz gradient (`L_φ ≥ 0`) and a
stepsize `0 ≤ α ≤ 1/L_φ`, the step `θ⁺ = θ − α g` satisfies, for every vector `u` (in the proof,
`u = ∇_θ φ_γ(θ^t; z^t)`) and with `δ = g − u`,
`F(θ⁺) ≤ F(θ) − (α/2)‖∇F(θ)‖² + α(1 − L_φ α)⟨∇F(θ), ∇F(θ) − u⟩ + (α(1 + L_φ α)/2)‖δ‖²
  + L_φ α² ‖u − ∇F(θ)‖²`. -/
theorem eq_32 {d : ℕ} (F : Param d → ℝ) (Lφ : ℝ) (hF : Differentiable ℝ F)
    (hL : ∀ θ θ', ‖gradient F θ - gradient F θ'‖ ≤ Lφ * ‖θ - θ'‖) (hLφ : 0 ≤ Lφ)
    (θ g u : Param d) (α : ℝ) (hα : 0 ≤ α) (hαL : Lφ * α ≤ 1) :
    F (θ - α • g) ≤ F θ - α / 2 * ‖gradient F θ‖ ^ 2
      + α * (1 - Lφ * α) * inner ℝ (gradient F θ) (gradient F θ - u)
      + α * (1 + Lφ * α) / 2 * ‖g - u‖ ^ 2
      + Lφ * α ^ 2 * ‖u - gradient F θ‖ ^ 2 := by sorry

end CertDRO.SGD
Source
Sinha, Namkoong, Volpi, Duchi, Certifying Some Distributional Robustness with Principled Adversarial Training, arXiv:1710.10571v5, p. 40, §B.3, (32)
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