Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 14 (29) — w.p. ≥ 1 − δ, E g(T) ≤ (1/N) Σ g(uⁱ) + ḡ√(log(1/δ)/2N) + ℜ_N(𝒢) for all g ∈ 𝒢 (0 ≤ g ≤ ḡ)

Proved
PrescAnalytics.ERM.theorem_14_eq29

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

generalization-boundp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1rademacher-complexityuniform-deviation

Let PPP be a probability measure on a measurable space U\mathcal UU, and let G\mathcal GG be a pointwise separable class of measurable functions g:U→Rg:\mathcal U\to\mathbb Rg:U→R with

0≤g(u)≤gˉfor all g∈G, u∈U.0\le g(u)\le\bar g\qquad\text{for all }g\in\mathcal G,\ u\in\mathcal U.0≤g(u)≤gˉ​for all g∈G, u∈U.

Let N≥1N\ge1N≥1, δ>0\delta>0δ>0, and let SN=(u1,…,uN)S_N=(u^1,\dots,u^N)SN​=(u1,…,uN) be an i.i.d. sample from PPP, i.e. drawn from the product measure PNP^NPN. Then, with probability at least 1−δ1-\delta1−δ,

E[g(T)]≤1N∑i=1Ng(ui)+gˉlog⁡(1/δ)2N+RN(G)for all g∈G,\mathbb E[g(T)]\le\frac1N\sum_{i=1}^N g(u^i)+\bar g\sqrt{\frac{\log(1/\delta)}{2N}}+\mathfrak R_N(\mathcal G)\qquad\text{for all }g\in\mathcal G,E[g(T)]≤N1​i=1∑N​g(ui)+gˉ​2Nlog(1/δ)​​+RN​(G)for all g∈G,

where T∼PT\sim PT∼P and RN(G)\mathfrak R_N(\mathcal G)RN​(G) is the marginal Rademacher complexity (factor 2/N2/N2/N, no absolute value).

This is the first half of the paper's restatement of the main uniform deviation bound of Bartlett and Mendelson (2003); applied to the cost class of a family of decision rules it gives inequality (26) of Theorem 13.

Formalization Note The paper assumes only ∣g∣≤gˉ|g|\le\bar g∣g∣≤gˉ​; the printed statement is false without g≥0g\ge0g≥0 (with G={u↦u}\mathcal G=\{u\mapsto u\}G={u↦u}, uuu uniform on {−1,+1}\{-1,+1\}{−1,+1}, N=1N=1N=1, δ=1/5\delta=1/5δ=1/5, the bound fails with probability 1/21/21/2), so the Lean adds 0≤g0\le g0≤g and keeps every printed constant. "With probability 1−δ1-\delta1−δ" is stated as: the set of samples on which the inequality fails for some g∈Gg\in\mathcal Gg∈G has outer PNP^NPN-measure at most δ\deltaδ. The inequality is written as max⁡(0,Eg−E^g−gˉ⋅)≤RN(G)\max(0,\mathbb E g-\hat{\mathbb E}g-\bar g\sqrt{\cdot})\le\mathfrak R_N(\mathcal G)max(0,Eg−E^g−gˉ​⋅​)≤RN​(G) in [0,∞][0,\infty][0,∞], which is equivalent since RN(G)≥0\mathfrak R_N(\mathcal G)\ge0RN​(G)≥0. Pointwise separability is the added measurability convention. For δ≥1\delta\ge1δ≥1, Lean's log⁡(1/δ)/(2N)\sqrt{\log(1/\delta)/(2N)}log(1/δ)/(2N)​ is 000 and the claim is trivial, as on the page.

Preamble
import Mathlib
import Definitions.Def_PrescAnalytics_ERM_Basic
Formal statement
namespace PrescAnalytics.ERM

open MeasureTheory

/-- Theorem 14, (29) (p. 40), with the corrected boundedness `0 ≤ g ≤ ḡ`. Let `G` be a pointwise
separable class of measurable functions `U → ℝ` with values in `[0, ḡ]`, `P` a probability measure on
`U`, `N ≥ 1`, `δ > 0`. Outside a set of `P^N`-measure at most `δ`, every sample
`S_N = (u^1, …, u^N)` satisfies
`E[g(T)] ≤ (1/N) Σ_i g(u^i) + ḡ √(log(1/δ)/(2N)) + ℜ_N(G)` for all `g ∈ G` simultaneously
(stated as: the bad set, an outer measure, is `≤ δ`; the inequality is written
`ofReal(E g − Ê g − ḡ√…) ≤ ℜ_N(G)`, equivalent since `ℜ_N(G) ≥ 0`). -/
theorem theorem_14_eq29 {U : Type*} [MeasurableSpace U] (P : Measure U) [IsProbabilityMeasure P]
    (G : Set (U → ℝ)) (gbar : ℝ) (hG_meas : ∀ g ∈ G, Measurable g)
    (hG_sep : IsPointwiseSeparable G) (hG_bdd : ∀ g ∈ G, ∀ u, 0 ≤ g u ∧ g u ≤ gbar)
    (N : ℕ) (hN : 1 ≤ N) (δ : ℝ) (hδ : 0 < δ) :
    (Measure.pi fun _ : Fin N => P)
      {s | ∃ g ∈ G, margRademacherR P N G <
        ENNReal.ofReal ((∫ u, g u ∂P) - empMean s g
          - gbar * Real.sqrt (Real.log (1 / δ) / (2 * N)))} ≤ ENNReal.ofReal δ := by sorry

end PrescAnalytics.ERM
Source
Bertsimas, Kallus, From Predictive to Prescriptive Analytics, arXiv:1402.5481v4, p. 40, Theorem 14, Eq. (29)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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