Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Gaussian Thompson sampling: large posterior tail at one exact rank

Proved
BanditAlgorithm.gaussian_exact_rank_large_posterior_tail_probability_bound

by Harry_Xu · Jul 31, 2026 · Mathlib 0df444a (Lean v4.33.1)

banditsconcentrationgaussianthompson-sampling

Consider Gaussian Thompson sampling with independent N(μi,1)\mathcal N(\mu_i,1)N(μi​,1) rewards and posterior variance 1/s1/s1/s after s≥1s\ge1s≥1 observations of arm iii. Fix an optimal arm i0i_0i0​, a suboptimal arm iii with gap Δi>0\Delta_i>0Δi​>0, a horizon n≥2n\ge2n≥2, and an exact positive rank sss. On histories where arm iii has been pulled more than sss times, the probability that its rank-sss posterior assigns more than 1/n1/n1/n mass above the midpoint μ∗−Δi/2\mu_* - \Delta_i/2μ∗​−Δi​/2 is at most

{exp⁡ ⁣[−s(Δi/2−2log⁡(n)/s)22],2log⁡n(Δi/2)2<s,1,otherwise.\begin{cases} \exp\!\left[-\dfrac{s\left(\Delta_i/2-\sqrt{2\log(n)/s}\right)^2}{2}\right], & \dfrac{2\log n}{(\Delta_i/2)^2}<s,\\ 1, & \text{otherwise.} \end{cases}⎩⎨⎧​exp​−2s(Δi​/2−2log(n)/s​)2​​,1,​(Δi​/2)22logn​<s,otherwise.​

The event reduction uses the Gaussian posterior tail bound and the stopped reward-stack concentration inequality for the first sss rewards of arm iii.

Preamble
import Definitions.Def_ThompsonSampling
import Definitions.Def_GaussianBandit

open MeasureTheory ProbabilityTheory
Formal statement
theorem BanditAlgorithm.gaussian_exact_rank_large_posterior_tail_probability_bound :
    ∀ (k : ℕ) [NeZero k] (μvec : Fin k → ℝ) (π : BanditAlgorithm.BanditPolicy k),
      ∀ (i₀ : Fin k),
        BanditAlgorithm.banditArmMean (BanditAlgorithm.gaussianBandit μvec) i₀ =
          BanditAlgorithm.banditOptimalMean (BanditAlgorithm.gaussianBandit μvec) →
      ∀ (i : Fin k), 0 < BanditAlgorithm.banditGap (BanditAlgorithm.gaussianBandit μvec) i →
      ∀ n : ℕ, 2 ≤ n →
      ∀ s : ℕ, 0 < s →
        (BanditAlgorithm.banditMeasure (BanditAlgorithm.gaussianBandit μvec) π n).real
          {h : BanditAlgorithm.BanditHistory k n |
            s < BanditAlgorithm.armPullCount i h ∧
              1 / (n : ℝ) <
                BanditAlgorithm.gaussianTSTailProb i s
                  (BanditAlgorithm.banditArmMean (BanditAlgorithm.gaussianBandit μvec) i₀ -
                    BanditAlgorithm.banditGap (BanditAlgorithm.gaussianBandit μvec) i / 2) h} ≤
          if 2 * Real.log n /
                (BanditAlgorithm.banditGap (BanditAlgorithm.gaussianBandit μvec) i / 2) ^ 2 < (s : ℝ)
          then
            Real.exp
              (-((s : ℝ) *
                (BanditAlgorithm.banditGap (BanditAlgorithm.gaussianBandit μvec) i / 2 -
                  Real.sqrt (2 * Real.log n / s))) ^ 2 /
                ((2 : ℝ) * (s : ℝ) * 1))
          else 1 := by sorry
Source
Lattimore and Szepesvári, Bandit Algorithms (CUP 2020), Exercise 36.6(b), printed p. 475 / PDF p. 484; reward-stack construction in §4.6, printed p. 65 / PDF p. 74, and Gaussian Thompson-sampling analysis in Theorem 36.2, printed pp. 463–465 / PDF pp. 472–474.

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