Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 2, p. 7 — online gradient descent without feasibility: regret ≤ ‖x₁ − x‖²/(2Tη) + (η/(2T)) Σ‖f_t‖² against every x ∈ S

Proved
ApproxOLO.FullInfo.lemma_2

by mikedeng1 · Oct 22, 2026 · Mathlib 0df444a (Lean v4.33.1)

online-gradient-descentonline-learningp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1regret

Let S⊂Rd\mathcal S\subset\mathbb R^dS⊂Rd be convex and compact, η>0\eta>0η>0, T≥1T\ge1T≥1, and let f1,…,fT∈Rdf_1,\dots,f_T\in\mathbb R^df1​,…,fT​∈Rd be arbitrary. Consider plays x1,…,xTx_1,\dots,x_Tx1​,…,xT​ of online gradient descent without feasibility (Algorithm 1): x1∈Sx_1\in\mathcal Sx1​∈S, and for each ttt the next point xt+1x_{t+1}xt+1​ is an infeasible projection of yt+1y_{t+1}yt+1​ onto S\mathcal SS, i.e.

∀z∈S:∥z−xt+1∥2≤∥z−yt+1∥2,(3)\forall z\in\mathcal S:\quad\|z-x_{t+1}\|^2\le\|z-y_{t+1}\|^2,\tag{3}∀z∈S:∥z−xt+1​∥2≤∥z−yt+1​∥2,(3)

where yt+1=xt−ηfty_{t+1}=x_t-\eta f_tyt+1​=xt​−ηft​ for losses and yt+1=xt+ηfty_{t+1}=x_t+\eta f_tyt+1​=xt​+ηft​ for payoffs. Then for every x∈Sx\in\mathcal Sx∈S,

1T∑t=1Txt⋅ft−1T∑t=1Tx⋅ft≤12Tη∥x1−x∥2+η2T∑t=1T∥ft∥2(losses),\frac1T\sum_{t=1}^Tx_t\cdot f_t-\frac1T\sum_{t=1}^Tx\cdot f_t\le\frac1{2T\eta}\|x_1-x\|^2+\frac{\eta}{2T}\sum_{t=1}^T\|f_t\|^2\quad\text{(losses)},T1​t=1∑T​xt​⋅ft​−T1​t=1∑T​x⋅ft​≤2Tη1​∥x1​−x∥2+2Tη​t=1∑T​∥ft​∥2(losses), 1T∑t=1Tx⋅ft−1T∑t=1Txt⋅ft≤12Tη∥x1−x∥2+η2T∑t=1T∥ft∥2(payoffs).\frac1T\sum_{t=1}^Tx\cdot f_t-\frac1T\sum_{t=1}^Tx_t\cdot f_t\le\frac1{2T\eta}\|x_1-x\|^2+\frac{\eta}{2T}\sum_{t=1}^T\|f_t\|^2\quad\text{(payoffs)}.T1​t=1∑T​x⋅ft​−T1​t=1∑T​xt​⋅ft​≤2Tη1​∥x1​−x∥2+2Tη​t=1∑T​∥ft​∥2(payoffs).

The points x2,…,xTx_2,\dots,x_Tx2​,…,xT​ need not lie in S\mathcal SS; the classical regret bound of online gradient descent survives with this weaker projection, which is what makes approximate projections usable.

Formalization Note ∥⋅∥\|\cdot\|∥⋅∥ is Euclidean. The infeasible-projection condition is required for t=1,…,T−1t=1,\dots,T-1t=1,…,T−1, the steps that produce played points x2,…,xTx_2,\dots,x_Tx2​,…,xT​; Algorithm 1 also computes xT+1x_{T+1}xT+1​, which is never played, so the statement covers every TTT-round run of Algorithm 1. T≥1T\ge1T≥1 is the paper's "TTT rounds" with the factor 1/T1/T1/T.

Preamble
import Mathlib
import Definitions.Def_ApproxOLO_FullInfo_Model
Formal statement
namespace ApproxOLO.FullInfo

open Matrix

/-- **Lemma 2 (Online gradient descent without feasibility), Garber, arXiv:1709.03093v1, p. 7.**
Let `S ⊂ ℝ^d` be convex and compact, `η > 0`, `T ≥ 1`, and let `x_1 ∈ S`. If the plays
`x_1, …, x_T` of Algorithm 1 satisfy the infeasible-projection condition (3),
`‖z − x_{t+1}‖² ≤ ‖z − y_{t+1}‖²` for all `z ∈ S`, with `y_{t+1} = x_t − η f_t` (losses) resp.
`x_t + η f_t` (payoffs), then for every `x ∈ S`
`(1/T) Σ x_t · f_t − (1/T) Σ x · f_t ≤ (1/(2Tη)) ‖x_1 − x‖² + (η/(2T)) Σ ‖f_t‖²` (losses), and
`(1/T) Σ x · f_t − (1/T) Σ x_t · f_t ≤ (1/(2Tη)) ‖x_1 − x‖² + (η/(2T)) Σ ‖f_t‖²` (payoffs). -/
theorem lemma_2 {d : ℕ} (S : Set (Vec d)) (hS : Convex ℝ S) (hSc : IsCompact S)
    (eta : ℝ) (heta : 0 < eta) (T : ℕ) (hT : 1 ≤ T) (xs fs : ℕ → Vec d) (hx1 : xs 1 ∈ S) :
    ((∀ t, 1 ≤ t → t < T → ∀ z ∈ S,
        euclNorm (z - xs (t + 1)) ^ 2 ≤ euclNorm (z - (xs t - eta • fs t)) ^ 2) →
      ∀ x ∈ S,
        (1 / (T : ℝ)) * ∑ t ∈ Finset.Icc 1 T, xs t ⬝ᵥ fs t
            - (1 / (T : ℝ)) * ∑ t ∈ Finset.Icc 1 T, x ⬝ᵥ fs t
          ≤ 1 / (2 * T * eta) * euclNorm (xs 1 - x) ^ 2
            + eta / (2 * T) * ∑ t ∈ Finset.Icc 1 T, euclNorm (fs t) ^ 2) ∧
    ((∀ t, 1 ≤ t → t < T → ∀ z ∈ S,
        euclNorm (z - xs (t + 1)) ^ 2 ≤ euclNorm (z - (xs t + eta • fs t)) ^ 2) →
      ∀ x ∈ S,
        (1 / (T : ℝ)) * ∑ t ∈ Finset.Icc 1 T, x ⬝ᵥ fs t
            - (1 / (T : ℝ)) * ∑ t ∈ Finset.Icc 1 T, xs t ⬝ᵥ fs t
          ≤ 1 / (2 * T * eta) * euclNorm (xs 1 - x) ^ 2
            + eta / (2 * T) * ∑ t ∈ Finset.Icc 1 T, euclNorm (fs t) ^ 2) := by sorry

end ApproxOLO.FullInfo
Source
Garber, Efficient Online Linear Optimization with Approximation Algorithms, arXiv:1709.03093v1, p. 7, Lemma 2 (Algorithm 1, Eq. (3))
Human review
  • Endorsed by Shuze Chen · Oct 22, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 22, 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