Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma B.3 — exponential inequality for ratios (De la Peña, Klass and Lai 2004)

Proved
LinParamBandits.UEGeneral.exp_ineq_ratios

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

concentrationmartingalep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1self-normalized

Let AAA and BBB be random variables on a probability space such that B≥0B \ge 0B≥0 almost surely and

E[eγA−γ2B2/2]≤1for all γ∈R.\mathbb E\big[e^{\gamma A - \gamma^2B^2/2}\big] \le 1 \qquad \text{for all } \gamma \in \mathbb R.E[eγA−γ2B2/2]≤1for all γ∈R.

Then for all ζ≥2\zeta \ge \sqrt 2ζ≥2​ and y>0y > 0y>0,

Pr⁡{∣A∣≥ζ(B2+y)(1+12log⁡(1+B2y))}≤e−ζ2/2.\Pr\left\{|A| \ge \zeta\sqrt{(B^2 + y)\Big(1 + \frac12\log\Big(1 + \frac{B^2}{y}\Big)\Big)}\right\} \le e^{-\zeta^2/2}.Pr{∣A∣≥ζ(B2+y)(1+21​log(1+yB2​))​}≤e−ζ2/2.

This self-normalized tail bound is quoted by the paper from Corollary 2.2 of De la Peña, Klass and Lai (2004); it is the probabilistic input of the martingale inequality, Lemma B.4.

Formalization Note The expectation in the hypothesis is a lower Lebesgue integral of the nonnegative integrand, so the hypothesis cannot hold by a junk value of a non-integrable Bochner integral. AAA and BBB are measurable (they are random variables).

Preamble
import Mathlib
Formal statement
namespace LinParamBandits.UEGeneral

open MeasureTheory ProbabilityTheory

/-- Lemma B.3 (p. 30; De la Peña, Klass and Lai 2004, Corollary 2.2): exponential inequality
for ratios. -/
theorem exp_ineq_ratios {Ω : Type*} [MeasurableSpace Ω] {P : Measure Ω} [IsProbabilityMeasure P]
    {A B : Ω → ℝ} (hA : Measurable A) (hB : Measurable B) (hB0 : ∀ᵐ ω ∂P, 0 ≤ B ω)
    (hexp : ∀ γ : ℝ,
      ∫⁻ ω, ENNReal.ofReal (Real.exp (γ * A ω - γ ^ 2 * B ω ^ 2 / 2)) ∂P ≤ 1)
    (ζ : ℝ) (hζ : Real.sqrt 2 ≤ ζ) (y : ℝ) (hy : 0 < y) :
    P {ω | ζ * Real.sqrt ((B ω ^ 2 + y) * (1 + 1 / 2 * Real.log (1 + B ω ^ 2 / y))) ≤ |A ω|} ≤
      ENNReal.ofReal (Real.exp (-ζ ^ 2 / 2)) := by sorry

end LinParamBandits.UEGeneral
Source
Rusmevichientong, Tsitsiklis, Linearly Parameterized Bandits, arXiv:0812.3465v2, Lemma B.3, p. 30 (citing De la Peña, Klass, Lai 2004, Ann. Probab. 32, Corollary 2.2, p. 1908)
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 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