Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 11.10: HiddenExploration is BIC for any target function as long as ε≤13E[G⋅1{G>0}]\varepsilon \le \frac13\mathbb{E}[G \cdot \mathbf 1\{G > 0\}]ε≤31​E[G⋅1{G>0}]

Proved
IntroBandits.hidden_exploration_bic

by naimengye · Sep 24, 2026 · Mathlib 0df444a (Lean v4.33.1)

bayesian-incentive-compatibilityhidden-explorationincentivized-exploration

Lemma 11.10. Algorithm 11.1 is BIC, for any target function atrga_{\mathrm{trg}}atrg​, as long as ε≤13 E[G⋅1{G>0}]\varepsilon \le \frac13\,\mathbb{E}[G \cdot \mathbf 1\{G > 0\}]ε≤31​E[G⋅1{G>0}], where G=E[μ2−μ1∣sig]G = \mathbb{E}[\mu_2 - \mu_1 \mid \mathrm{sig}]G=E[μ2​−μ1​∣sig] is the posterior gap.

Formally: Ω\OmegaΩ a finite signal space and QQQ a probability law of (μ,sig)(\mu, \mathrm{sig})(μ,sig) whose μ\muμ-marginal is supported on the finite F⊆[0,1]2F \subseteq [0,1]^2F⊆[0,1]2 with μ10≥μ20\mu^0_1 \ge \mu^0_2μ10​≥μ20​ (the standing assumption of the chapter's Preliminaries); trg\mathrm{trg}trg any randomized target, a distribution over the two arms for each signal; 0<ε≤13∑Smax⁡(0,E[(μ2−μ1)1{sig=S}])0 < \varepsilon \le \frac13 \sum_S \max\big(0, \mathbb{E}[(\mu_2 - \mu_1)\mathbf 1\{\mathrm{sig} = S\}]\big)0<ε≤31​∑S​max(0,E[(μ2​−μ1​)1{sig=S}]). Then the recommendation rule of HiddenExploration, Pr⁡[rec=a∣sig=S]=ε trg(S)(a)+(1−ε)1{a=min⁡arg⁡max⁡a′E[μa′∣sig=S]}\Pr[\mathrm{rec} = a \mid \mathrm{sig} = S] = \varepsilon\,\mathrm{trg}(S)(a) + (1-\varepsilon)\mathbf 1\{a = \min\arg\max_{a'} \mathbb{E}[\mu_{a'} \mid \mathrm{sig} = S]\}Pr[rec=a∣sig=S]=εtrg(S)(a)+(1−ε)1{a=minargmaxa′​E[μa′​∣sig=S]}, satisfies the single-round BIC property (11.5): for both orderings of the two arms, Pr⁡[rec=a]>0\Pr[\mathrm{rec} = a] > 0Pr[rec=a]>0 implies E[(μa−μa′)1{rec=a}]≥0\mathbb{E}[(\mu_a - \mu_{a'})\mathbf 1\{\mathrm{rec} = a\}] \ge 0E[(μa​−μa′​)1{rec=a}]≥0.

The bound is the printed one (non-strict); the book's proof ends with the strict ε<13F(G>0)\varepsilon < \frac13 F(G > 0)ε<31​F(G>0) because it proves the strict inequality (11.6), and at equality the same computation gives ≥0\ge 0≥0, which is what (11.5) asks.

Preamble
import Definitions.Def_IntroBandits_Agents

open MeasureTheory ProbabilityTheory BanditAlgorithm
Formal statement
namespace IntroBandits

theorem hidden_exploration_bic {Ω : Type} [Fintype Ω] [DecidableEq Ω] [MeasurableSpace Ω]
    [MeasurableSingletonClass Ω] (Q : Measure ((Fin 2 → ℝ) × Ω)) [IsProbabilityMeasure Q]
    (F : Finset (Fin 2 → ℝ)) (hF : Q ((↑F)ᶜ ×ˢ Set.univ) = 0)
    (hunit : ∀ μ ∈ F, ∀ a, μ a ∈ Set.Icc (0 : ℝ) 1)
    (hprior : priorMean Q.fst F 1 ≤ priorMean Q.fst F 0)
    (trg : Ω → Fin 2 → ℝ) (htrg : ∀ S, trg S ∈ stdSimplex ℝ (Fin 2))
    {ε : ℝ} (hε : 0 < ε) (hε3 : ε ≤ sigGapPlus Q F / 3) :
    IsSingleRoundBIC Q F (hiddenExploration ε trg Q F) := by sorry

end IntroBandits
Source
Slivkins, Introduction to Multi-Armed Bandits, arXiv:1904.07272 (FnT ML 12, 2019), §11.3 p. 149, Lemma 11.10, with the proof on pp. 149-150
Human review
  • Endorsed by Shuze Chen · Sep 25, 2026

    Confirmed by the moderator at approval.

  • Endorsed by naimengye · Sep 25, 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