Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.1, p. 11 — under (A) and (B), ∃ β > 0 with limsup_N N⁻¹ log P(ℰ^c_N) ≤ −β

Proved
SAARate.SharpLD.theorem_3_1

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

large-deviationsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1sample-average-approximationsharp-minimumstochastic-programming

Let PPP be a probability measure on a measurable space (Ω,F)(\Omega,\mathcal F)(Ω,F), let h:Rm×Ω→Rh:\mathbb R^m\times\Omega\to\mathbb Rh:Rm×Ω→R, and let Θ⊆Rm\Theta\subseteq\mathbb R^mΘ⊆Rm. Write f(x)=EP h(x,ω)f(x)=\mathbb E_P\,h(x,\omega)f(x)=EP​h(x,ω) for the true objective (1.1) and f^N(x)=N−1∑j=1Nh(x,ωj)\hat f_N(x)=N^{-1}\sum_{j=1}^N h(x,\omega^j)f^​N​(x)=N−1∑j=1N​h(x,ωj) for the sample average objective (1.2) built from an i.i.d. sample ω1,ω2,…\omega^1,\omega^2,\dotsω1,ω2,… with law PPP. The assumptions of Theorem 2.1 are: (i) h(⋅,ω)h(\cdot,\omega)h(⋅,ω) is convex for every ω\omegaω; (ii) fff is well defined and finite valued; (iii) Θ\ThetaΘ is closed and convex; (iv) Assumption (A): fff has the unique minimizer xˉ\bar xxˉ on Θ\ThetaΘ and there is c>0c>0c>0 with f(x)≥f(xˉ)+c∥x−xˉ∥f(x)\ge f(\bar x)+c\|x-\bar x\|f(x)≥f(xˉ)+c∥x−xˉ∥ for all x∈Θx\in\Thetax∈Θ. Assumption (B) asks for a κ>0\kappa>0κ>0 with sup⁡d∈Sm−1∣hω′(xˉ,d)∣≤κ\sup_{d\in S^{m-1}}|h'_\omega(\bar x,d)|\le\kappasupd∈Sm−1​∣hω′​(xˉ,d)∣≤κ for PPP-almost every ω\omegaω, where hω′(xˉ,d)h'_\omega(\bar x,d)hω′​(xˉ,d) is the directional derivative of h(⋅,ω)h(\cdot,\omega)h(⋅,ω) at xˉ\bar xxˉ in direction ddd and Sm−1S^{m-1}Sm−1 is the unit sphere.

Let ANA_NAN​ be the set of optimal solutions of the SAA problem min⁡x∈Θf^N(x)\min_{x\in\Theta}\hat f_N(x)minx∈Θ​f^​N​(x) and let EN\mathcal E_NEN​ be the event that ANA_NAN​ is nonempty and AN={xˉ}A_N=\{\bar x\}AN​={xˉ} (3.7), i.e. that the SAA problem has the unique optimal solution x^N=xˉ\hat x_N=\bar xx^N​=xˉ.

Theorem 3.1. Suppose that the assumptions (i)–(iv) of Theorem 2.1 are satisfied and that Assumption (B) holds. Then there is a constant β>0\beta>0β>0 such that

lim sup⁡N→∞1Nlog⁡P(ENc)≤−β.\limsup_{N\to\infty}\frac1N\log P(\mathcal E_N^c)\le-\beta .N→∞limsup​N1​logP(ENc​)≤−β.

So at a sharp minimum the probability that the sample average approximation fails to return exactly the true solution decays exponentially in the sample size.

Formalization Note The sample ω1,ω2,…\omega^1,\omega^2,\dotsω1,ω2,… is one i.i.d. sequence of Ω\OmegaΩ-valued maps on a separate probability space (S,Q)(S,Q)(S,Q), each with law PPP; f^N\hat f_Nf^​N​ uses its first NNN terms. The rate is stated in the equivalent ε\varepsilonε-form: there is β>0\beta>0β>0 such that for every ε>0\varepsilon>0ε>0, eventually in NNN, Q(ENc)≤e−(β−ε)NQ(\mathcal E_N^c)\le e^{-(\beta-\varepsilon)N}Q(ENc​)≤e−(β−ε)N; this avoids the logarithm of a probability that may be 000. ENc\mathcal E_N^cENc​ is not shown to be measurable, so QQQ is its outer measure, which makes the upper bound stronger. "f finite valued" is the hypothesis that each h(x,⋅)h(x,\cdot)h(x,⋅) is PPP-integrable. The directional derivatives in (B) are true limits because h(⋅,ω)h(\cdot,\omega)h(⋅,ω) is convex.

Preamble
import Mathlib
import Definitions.Def_SAARate_SharpLD_Setting

open MeasureTheory ProbabilityTheory Filter Topology
Formal statement
namespace SAARate.SharpLD

theorem theorem_3_1 {m : ℕ} {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω)
    [IsProbabilityMeasure P] (h : SAARate.Sharp.E m → Ω → ℝ)
    (hconv : ∀ ω, ConvexOn ℝ Set.univ (fun x => h x ω))
    (hint : ∀ x, Integrable (h x) P)
    (Θ : Set (SAARate.Sharp.E m)) (hΘc : IsClosed Θ) (hΘv : Convex ℝ Θ)
    (xbar : SAARate.Sharp.E m) (hA : SAARate.Sharp.AssumptionA P h Θ xbar)
    (κ : ℝ) (hB : AssumptionB P h xbar κ)
    {S : Type*} [MeasurableSpace S] (Q : Measure S) [IsProbabilityMeasure Q]
    (ω : ℕ → S → Ω) (hω : IsIIDSample Q P ω) :
    ∃ β > 0, ∀ ε > 0, ∀ᶠ N : ℕ in atTop,
      Q {s | SAARate.Sharp.argminOn (SAARate.Sharp.saaObj h (fun j => ω j s) N) Θ ≠ {xbar}} ≤
        ENNReal.ofReal (Real.exp (-(β - ε) * N)) := by sorry

end SAARate.SharpLD
Source
Shapiro & Homem-de-Mello, On Rate of Convergence of Optimal Solutions of Monte Carlo Approximations of Stochastic Programs, preprint (SPEPS copy, edoc.hu-berlin.de), p. 11, Theorem 3.1, (3.7), (3.8); hypotheses of Theorem 2.1, p. 5; Assumption (A), p. 4; Assumption (B), p. 9
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 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