Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 6, p. 20 — kl(p,q)≥(p−q)2/(2max⁡[p,q]x(1−x))≥(p−q)2/(2q)\mathrm{kl}(p,q)\ge(p-q)^2/(2\max_{[p,q]}x(1-x))\ge(p-q)^2/(2q)kl(p,q)≥(p−q)2/(2max[p,q]​x(1−x))≥(p−q)2/(2q)

Proved
ExploreFirst.Collective.lemma_6

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

kullback-leiblerp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1pinsker-inequalityregret-lower-bound

This is a local refinement of Pinsker's inequality for Bernoulli distributions.

Let kl(p,q)\mathrm{kl}(p,q)kl(p,q) be the Kullback–Leibler divergence between Bernoulli distributions of parameters ppp and qqq, as in (5). For 0≤p<q≤10\le p<q\le10≤p<q≤1,

kl(p,q) ≥ 12max⁡x∈[p,q]x(1−x) (p−q)2 ≥ 12q (p−q)2.\mathrm{kl}(p,q)\ \ge\ \frac{1}{2\max_{x\in[p,q]}x(1-x)}\,(p-q)^2\ \ge\ \frac{1}{2q}\,(p-q)^2.kl(p,q) ≥ 2maxx∈[p,q]​x(1−x)1​(p−q)2 ≥ 2q1​(p−q)2.

The classical form kl(p,q)≥2(p−q)2\mathrm{kl}(p,q)\ge2(p-q)^2kl(p,q)≥2(p−q)2 follows from x(1−x)≤1/4x(1-x)\le1/4x(1−x)≤1/4. The proof of Theorem 4 uses the outer inequality kl(p,q)≥(p−q)2/(2q)\mathrm{kl}(p,q)\ge(p-q)^2/(2q)kl(p,q)≥(p−q)2/(2q), which is sharper when qqq is small.

Formalization Note kl\mathrm{kl}kl takes values in [0,+∞][0,+\infty][0,+∞] (it is +∞+\infty+∞ at q=1q=1q=1, p<1p<1p<1), so the first inequality is stated in [0,+∞][0,+\infty][0,+∞] for the nonnegative real (p−q)2/(2max⁡)(p-q)^2/(2\max)(p−q)2/(2max), and the second inequality between the two real lower bounds is stated in R\mathbb RR. The maximum of x(1−x)x(1-x)x(1−x) over the compact interval [p,q][p,q][p,q] is written as a supremum of its image; it is attained and positive.

Preamble
import Mathlib
import Definitions.Def_ExploreFirst_Collective_Setting
Formal statement
namespace ExploreFirst.Collective

open MeasureTheory ProbabilityTheory BanditAlgorithm
open scoped ENNReal

/-- Lemma 6, p. 20 (local refinement of Pinsker's inequality): for `0 ≤ p < q ≤ 1`,
`kl(p, q) ≥ (p - q)² / (2 max_{x ∈ [p, q]} x(1 - x)) ≥ (p - q)² / (2q)`. -/
theorem lemma_6 (p q : ℝ) (hp : 0 ≤ p) (hpq : p < q) (hq : q ≤ 1) :
    ENNReal.ofReal ((p - q) ^ 2 / (2 * sSup ((fun x => x * (1 - x)) '' Set.Icc p q)))
        ≤ ExploreFirst.FundIneq.klBer p q ∧
      (p - q) ^ 2 / (2 * q) ≤ (p - q) ^ 2 / (2 * sSup ((fun x => x * (1 - x)) '' Set.Icc p q)) := by sorry

end ExploreFirst.Collective
Source
Garivier, Ménard, Stoltz, Explore First, Exploit Next, arXiv:1602.07182v3, p. 20, Appendix A, Lemma 6
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