Fact D.1 — Jensen on arbitrary intervals, convex and concave
ProvedEliasGlobalClock.LowerBound.jensen_intervalinformation-theoryleader-electionp2o-batch-stoc27ap2o-gran-per-chapterp2o-plan-paperp2o-v1probability
Let be an integrable real random variable on a probability space, taking values in a real interval almost surely, and suppose its mean belongs to . Let have integrable composition . Both Jensen assertions hold:
The interval may be open, closed, half-open, unbounded or a singleton. No continuity or boundedness hypothesis is imposed. This fact supplies the expectation and finite-law convexity bounds used by the source's failure estimate.
Differences from the source:
- encoding: a real interval I and a function φ:I→ℝ → a set I with ConvexOn or ConcaveOn and a total real extension of φ; reason: in one dimension the convex-domain clause is exactly the interval condition; the extension outside I has no effect on the almost-surely I-valued variable
Preamble
import Mathlib open MeasureTheory
Formal statement
namespace EliasGlobalClock.LowerBound
/-- Fact D.1, printed p.36 / PDF p.37. No continuity or closedness assumption.
ConvexOn/ConcaveOn include convexity of I, hence precisely the interval condition
in one real dimension. The extension of φ outside I is immaterial almost surely. -/
theorem jensen_interval {Ω : Type*} [MeasurableSpace Ω]
(μ : Measure Ω) [IsProbabilityMeasure μ] (I : Set ℝ) (X : Ω → ℝ) (φ : ℝ → ℝ)
(hX : Integrable X μ) (hXI : ∀ᵐ ω ∂μ, X ω ∈ I)
(hm : (∫ ω, X ω ∂μ) ∈ I) (hφX : Integrable (fun ω => φ (X ω)) μ) :
(ConvexOn ℝ I φ → φ (∫ ω, X ω ∂μ) ≤ ∫ ω, φ (X ω) ∂μ) ∧
(ConcaveOn ℝ I φ → (∫ ω, φ (X ω) ∂μ) ≤ φ (∫ ω, X ω ∂μ)) := by sorry
end EliasGlobalClock.LowerBound
Source
Kuowen Chen, Shengquan Du, Tsvi Kopelowitz, Seth Pettie, Ben Plosk, STOC 2027 submission #1311 (October 2026), printed p.36, PDF p.37, Fact D.1; https://github.com/sigact-stoc/stoc2027/releases/download/papers-2/1311.pdf
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.