§2.2, p. 4 — 1 − P(Ŝ^ε_N ⊂ 𝒮^ε) ≤ Σ_{x∈𝒮} P{|ĝ_N(x) − g(x)| ≥ α(ε)/2}
ProvedKleywegtSAA.ExpRate.union_boundp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1sample-average-approximationunion-bound
Let be a nonempty finite set, a probability space carrying the sample , , and suppose is nonempty. Then for every sample size ,
The bound reduces the failure probability of the SAA method to finitely many one-dimensional deviation probabilities, each controlled by the large-deviations bound (2.4).
Formalization Note The statement holds for any sample sequence, so independence, identical distribution, measurability and integrability are omitted (a stronger statement). Probabilities are values of the measure in , with truncated subtraction ; for a probability measure this is the usual complement probability.
Preamble
import Mathlib import Definitions.Def_KleywegtSAA_ExpRate_Setting open MeasureTheory ProbabilityTheory Filter Topology
Formal statement
namespace KleywegtSAA.ExpRate
/-- §2.2, p. 4: `1 − P(Ŝ^ε_N ⊂ S^ε) ≤ ∑_{x ∈ S} P{|ĝ_N(x) − g(x)| ≥ α(ε)/2}` for every sample
size `N`. -/
theorem union_bound {X : Type*} (S : Finset X) (hS : S.Nonempty)
{𝒲 : Type*} (G : X → 𝒲 → ℝ)
{Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P] (W : ℕ → Ω → 𝒲)
(ε : ℝ) (hε : 0 ≤ ε) (hne : (nonOptSet S hS (trueObj G P W) ε).Nonempty) :
∀ N : ℕ,
1 - P {ω | epsSet S hS (sampleObj G W N ω) ε ⊆ epsSet S hS (trueObj G P W) ε} ≤
∑ x ∈ S, P {ω | alpha S hS (trueObj G P W) ε hne / 2 ≤
|sampleObj G W N ω x - trueObj G P W x|} := by sorry
end KleywegtSAA.ExpRate
Source
Kleywegt & Shapiro, The sample average approximation method for stochastic discrete optimization, preprint (two-author version, sha256 56657748…), p. 4, §2.2, display after "it follows that"
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.