Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2.13 — projected OGD regret with nonincreasing steps

Proved
ModernOnlineLearning.OGD.theorem_2_13

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

online-gradient-descentonline-learningp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1regret

Let VVV be a nonempty closed convex set of finite diameter D=sup⁡x,y∈V∥x−y∥D=\sup_{x,y\in V}\|x-y\|D=supx,y∈V​∥x−y∥. For T≥1T\ge1T≥1, let ℓ1,…,ℓT\ell_1,\ldots,\ell_Tℓ1​,…,ℓT​ be convex and differentiable on open sets containing VVV, and let (xt,gt)(x_t,g_t)(xt​,gt​) follow Algorithm 2.1 with positive nonincreasing step sizes ηt\eta_tηt​. Then every u∈Vu\in Vu∈V satisfies

Regret⁡T(u)≤D22ηT+∑t=1Tηt2∥gt∥2−∥xT+1−u∥22ηT.\operatorname{Regret}_T(u)\le\frac{D^2}{2\eta_T}+\sum_{t=1}^T\frac{\eta_t}{2}\|g_t\|^2-\frac{\|x_{T+1}-u\|^2}{2\eta_T}.RegretT​(u)≤2ηT​D2​+t=1∑T​2ηt​​∥gt​∥2−2ηT​∥xT+1​−u∥2​.

This gives a time-varying-step regret guarantee while retaining the terminal correction term. The constant-step clause of the same theorem is a companion item.

Formalization Note DDD is Metric.diam V; boundedness ensures this is the actual finite diameter. The horizon is positive because the formula divides by ηT\eta_TηT​. The ambient space is a complete real inner-product space.


Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.

Preamble
import Mathlib
import Definitions.Def_ModernOnlineLearning_OGD_Defs
Formal statement
namespace ModernOnlineLearning.OGD

/-- Orabona, Theorem 2.13, p. 13: the variable-step regret bound, including the
terminal negative squared-distance term. -/
theorem theorem_2_13 {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
    [CompleteSpace E] (V : Set E) (hV : V.Nonempty) (hc : IsClosed V)
    (hv : Convex ℝ V) (hb : Bornology.IsBounded V)
    (loss : ℕ → E → ℝ) (η : ℕ → ℝ) (x g : ℕ → E) (T : ℕ)
    (hT : 1 ≤ T) (u : E) (hu : u ∈ V)
    (hloss : ∀ t ∈ Finset.Icc 1 T,
      ∃ U : Set E, IsOpen U ∧ V ⊆ U ∧ ConvexOn ℝ U (loss t) ∧ DifferentiableOn ℝ (loss t) U)
    (hη : ∀ t ∈ Finset.Icc 1 T, 0 < η t)
    (hmono : ∀ t ∈ Finset.Icc 1 (T - 1), η (t + 1) ≤ η t)
    (hrun : IsOGDRun V loss η x g T) :
    regret loss x u T ≤
      Metric.diam V ^ 2 / (2 * η T) +
        ∑ t ∈ Finset.Icc 1 T, (η t / 2) * ‖g t‖ ^ 2 -
          ‖x (T + 1) - u‖ ^ 2 / (2 * η T) := by sorry

end ModernOnlineLearning.OGD
Source
Orabona, arXiv:1912.13213v10, Theorem 2.13, p. 13, first displayed bound
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