Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2, p. 7 — liminf_n P(μ_x̂ ≤ G_n(x̂) + z_α s_n(x*_n)/√n) ≥ 1 − α for the SRP (needs σ²_x̂(x*_max) > 0 when α > 1/2)

Proved
SolutionQuality.SRP.theorem2_srp_valid

by mikedeng1 · Oct 5, 2026 · Mathlib 0df444a (Lean v4.33.1)

asymptotic-validityconfidence-intervaloptimality-gapp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-programming

Consider the stochastic program z∗=min⁡x∈XEf(x,ξ~)z^*=\min_{x\in X}Ef(x,\tilde\xi)z∗=minx∈X​Ef(x,ξ~​) under assumptions (A1)–(A3), a candidate solution x^∈X\hat x\in Xx^∈X with optimality gap μx^=Ef(x^,ξ~)−z∗\mu_{\hat x}=Ef(\hat x,\tilde\xi)-z^*μx^​=Ef(x^,ξ~​)−z∗, and observations ξ~1,ξ~2,…\tilde\xi^1,\tilde\xi^2,\dotsξ~​1,ξ~​2,… i.i.d. as ξ~\tilde\xiξ~​. The single replication procedure (SRP) solves the sampled problem (SPn_nn​) once, obtaining an optimal solution xn∗x_n^*xn∗​, and computes

  1. the gap estimate Gn(x^)=fˉn(x^)−min⁡x∈Xfˉn(x)G_n(\hat x)=\bar f_n(\hat x)-\min_{x\in X}\bar f_n(x)Gn​(x^)=fˉ​n​(x^)−minx∈X​fˉ​n​(x) of (2), with fˉn(x)=1n∑i=1nf(x,ξ~i)\bar f_n(x)=\frac1n\sum_{i=1}^n f(x,\tilde\xi^i)fˉ​n​(x)=n1​∑i=1n​f(x,ξ~​i);
  2. the sample variance sn2(xn∗)=1n−1∑i=1n[(f(x^,ξ~i)−f(xn∗,ξ~i))−(fˉn(x^)−fˉn(xn∗))]2s_n^2(x_n^*)=\frac1{n-1}\sum_{i=1}^n\big[(f(\hat x,\tilde\xi^i)-f(x_n^*,\tilde\xi^i))-(\bar f_n(\hat x)-\bar f_n(x_n^*))\big]^2sn2​(xn∗​)=n−11​∑i=1n​[(f(x^,ξ~​i)−f(xn∗​,ξ~​i))−(fˉ​n​(x^)−fˉ​n​(xn∗​))]2.

Given 0<α<10<\alpha<10<α<1 and zαz_\alphazα​ with P(N(0,1)≤zα)=1−αP(N(0,1)\le z_\alpha)=1-\alphaP(N(0,1)≤zα​)=1−α, the one-sided interval [0, Gn(x^)+zαsn(xn∗)/n][0,\,G_n(\hat x)+z_\alpha s_n(x_n^*)/\sqrt n][0,Gn​(x^)+zα​sn​(xn∗​)/n​] of (5) is asymptotically valid:

lim inf⁡n→∞P(μx^≤Gn(x^)+zαsn(xn∗)n)≥1−α.(6)\liminf_{n\to\infty}P\left(\mu_{\hat x}\le G_n(\hat x)+\frac{z_\alpha s_n(x_n^*)}{\sqrt n}\right)\ge1-\alpha. \qquad (6)n→∞liminf​P(μx^​≤Gn​(x^)+n​zα​sn​(xn∗​)​)≥1−α.(6)

The theorem shows that a single sample-average problem gives enough information for a valid statistical upper bound on the optimality gap of a candidate solution, even though Gn(x^)G_n(\hat x)Gn​(x^) need not be asymptotically normal.

Formalization Note. The decision dimension is ddd; the sample is the first nnn terms ξ 0, …, ξ (n-1) of one i.i.d. sequence. xn∗x_n^*xn∗​ is a measurable map that, almost surely, minimizes fˉn\bar f_nfˉ​n​ over XXX on the same sample. f(x,⋅)f(x,\cdot)f(x,⋅) is measurable for each xxx, and (A2) is read as the existence of an integrable majorant of sup⁡x∈Xf2(x,⋅)\sup_{x\in X}f^2(x,\cdot)supx∈X​f2(x,⋅). zαz_\alphazα​ is any real with cdf (gaussianReal 0 1) zα = 1 - α. Probabilities are in ℝ≥0∞, where liminf is genuine. At n≤1n\le1n≤1 the factors 1/n1/n1/n, 1/(n−1)1/(n-1)1/(n−1) and 1/n1/\sqrt n1/n​ take Lean's junk value 000, which a liminf ignores.

Correction of the page. The statement carries one hypothesis the page does not print: α≤1/2\alpha\le 1/2α≤1/2, or σx^2(x)>0\sigma^2_{\hat x}(x)>0σx^2​(x)>0 for some x∈X∗x\in X^*x∈X∗ (equivalently σx^2(xmax⁡∗)>0\sigma^2_{\hat x}(x^*_{\max})>0σx^2​(xmax∗​)>0). The page's proof of the case α>1/2\alpha>1/2α>1/2 replaces xmin⁡∗x^*_{\min}xmin∗​ by xmax⁡∗x^*_{\max}xmax∗​ and fails when σx^2≡0\sigma^2_{\hat x}\equiv0σx^2​≡0 on X∗X^*X∗, where (6) itself is false: for X=[−1,1]X=[-1,1]X=[−1,1], f(x,ξ)=x2−2xξf(x,\xi)=x^2-2x\xif(x,ξ)=x2−2xξ, ξ~∼N(0,1)\tilde\xi\sim N(0,1)ξ~​∼N(0,1), x^=0\hat x=0x^=0 and α=0.9\alpha=0.9α=0.9 one has Gn(x^)=ξˉn2G_n(\hat x)=\bar\xi_n^2Gn​(x^)=ξˉ​n2​, sn(xn∗)≈2∣ξˉn∣s_n(x_n^*)\approx2|\bar\xi_n|sn​(xn∗​)≈2∣ξˉ​n​∣, and the coverage tends to P(∣N(0,1)∣≥2∣z0.9∣)≈0.010<0.1P(|N(0,1)|\ge 2|z_{0.9}|)\approx0.010<0.1P(∣N(0,1)∣≥2∣z0.9​∣)≈0.010<0.1.

Preamble
import Mathlib
import Definitions.Def_SolutionQuality_SRP_Setting
Formal statement
namespace SolutionQuality.SRP

open MeasureTheory ProbabilityTheory Filter

/-- Theorem 2 (p. 7), display (6), with the one correction recorded in the moderation:
the hypothesis `hdeg` (`α ≤ 1/2`, or some optimal `x` has `σ²_x̂(x) > 0`, i.e.
`σ²_x̂(x*_max) > 0`). The page states (6) for every `0 < α < 1`, but its proof of the case
`α > 1/2` ("replace `x*_min` with `x*_max`") breaks down when `σ²_x̂ ≡ 0` on `X*`, and (6) is
then false: `X = [-1, 1]`, `f(x, ξ) = x² - 2xξ`, `ξ̃ ~ N(0, 1)`, `x̂ = 0`, `α = 0.9` gives
coverage `→ P(|N(0,1)| ≥ 2·z_{0.1}) ≈ 0.010 < 0.1`. -/
theorem theorem2_srp_valid {d : ℕ} {Ξ : Type*} [MeasurableSpace Ξ]
    (μ : Measure Ξ) [IsProbabilityMeasure μ]
    (f : E d → Ξ → ℝ) (hfm : ∀ x, Measurable (f x))
    (X : Set (E d)) (hA : Assumptions μ f X)
    {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (ξ : ℕ → Ω → Ξ) (hξ : IsIIDSample P μ ξ)
    (xhat : E d) (hxhat : xhat ∈ X)
    (xn : ℕ → Ω → E d) (hxn : IsSAAMinimizerSeq P f X ξ xn)
    (α : ℝ) (hα0 : 0 < α) (hα1 : α < 1)
    (zα : ℝ) (hzα : cdf (gaussianReal 0 1) zα = 1 - α)
    (hdeg : α ≤ 1 / 2 ∨ ∃ x ∈ optSet μ f X, 0 < gapVar μ f xhat x) :
    ENNReal.ofReal (1 - α) ≤
      liminf (fun n : ℕ => P {ω | optGap μ f X xhat ≤
        gapEstimate f X xhat (fun i => ξ i ω) n +
          zα * Real.sqrt (sampleVar f xhat (fun i => ξ i ω) n (xn n ω)) / Real.sqrt n}) atTop := by sorry

end SolutionQuality.SRP
Source
Bayraksan & Morton, Assessing Solution Quality in Stochastic Programs, preprint (January 26, 2005), p. 7, Theorem 2, (6); SRP and (5) on pp. 5–6
Human review
  • Endorsed by Shuze Chen · Oct 5, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 5, 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