Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proof of Theorem A.1, p. 26 (corrected) — λ† = sup{λ ∈ Λ : R(λ) > α} lies in Λ and R(λ†) ≥ α

Proved
RiskControl.UCB.failure_point

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

calibrationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1risk-controlupper-confidence-bound

Let Λ⊆R‾=R∪{±∞}\Lambda \subseteq \overline{\mathbb R} = \mathbb R \cup \{\pm\infty\}Λ⊆R=R∪{±∞} be closed, let R:R‾→RR : \overline{\mathbb R} \to \mathbb RR:R→R be continuous on Λ\LambdaΛ, and let α∈R\alpha \in \mathbb Rα∈R. Suppose some λ∈Λ\lambda \in \Lambdaλ∈Λ has R(λ)>αR(\lambda) > \alphaR(λ)>α, and set

λ†=sup⁡{λ∈Λ:R(λ)>α}.\lambda^\dagger = \sup\{\lambda \in \Lambda : R(\lambda) > \alpha\}.λ†=sup{λ∈Λ:R(λ)>α}.

Then

λ†∈ΛandR(λ†)≥α.\lambda^\dagger \in \Lambda \quad\text{and}\quad R(\lambda^\dagger) \ge \alpha.λ†∈ΛandR(λ†)≥α.

The point λ†\lambda^\daggerλ† is the deterministic place where a failure of UCB calibration is detected: it depends only on RRR and α\alphaα, not on the data, so the pointwise confidence bound (3) can be applied at it.

Formalization Note This replaces a step of the printed proof of Theorem A.1, which introduces λ∗=inf⁡{λ∈Λ:R(λ)≤α}\lambda^* = \inf\{\lambda \in \Lambda : R(\lambda) \le \alpha\}λ∗=inf{λ∈Λ:R(λ)≤α} and asserts R(λ∗)=αR(\lambda^*) = \alphaR(λ∗)=α "by continuity". That assertion fails when Λ\LambdaΛ is not an interval: for Λ={0,1}\Lambda = \{0, 1\}Λ={0,1}, R(0)=1R(0) = 1R(0)=1, R(1)=0R(1) = 0R(1)=0, α=1/2\alpha = 1/2α=1/2 one gets λ∗=1\lambda^* = 1λ∗=1 and R(λ∗)=0R(\lambda^*) = 0R(λ∗)=0. The corrected point λ†\lambda^\daggerλ† coincides with λ∗\lambda^*λ∗ when Λ\LambdaΛ is an interval. No monotonicity of RRR is assumed. The supremum is taken in the extended reals, and continuity is with respect to the subspace topology of Λ\LambdaΛ.

Preamble
import Mathlib
Formal statement
namespace RiskControl.UCB

/-- Proof of Theorem A.1, p. 26, corrected: for a closed `Λ ⊆ EReal` and a function `R`
continuous on `Λ`, if some `λ ∈ Λ` has `R(λ) > α`, then `λ† = sup {λ ∈ Λ : R(λ) > α}` lies in
`Λ` and `R(λ†) ≥ α`. (The page uses `λ* = inf {λ ∈ Λ : R(λ) ≤ α}` with `R(λ*) = α`, which fails
when `Λ` is not an interval.) -/
theorem failure_point (Λ : Set EReal) (hΛ : IsClosed Λ) (R : EReal → ℝ)
    (hcont : ContinuousOn R Λ) (α : ℝ) (hU : {l ∈ Λ | α < R l}.Nonempty) :
    sSup {l ∈ Λ | α < R l} ∈ Λ ∧ α ≤ R (sSup {l ∈ Λ | α < R l}) := by sorry

end RiskControl.UCB
Source
Bates, Angelopoulos, Lei, Malik & Jordan, arXiv:2101.02703v3, Proof of Theorem A.1, p. 26 (the point λ* and 'R(λ*) = α (by continuity)'), corrected to λ† = sup{λ ∈ Λ : R(λ) > α}
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 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