Theorem 1, p. 11 — total variation and zero-count bounds
ProvedChenStein.OneVar.theorem_1chen-stein-methoddependent-indicatorsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1poisson-approximation
Let count a family of dependent Bernoulli events and let be Poisson with their common mean . Using the paper's total variation norm, which is twice the supremum over events,
This controls both the complete count distribution and the probability that no event occurs, with the paper's exact coefficients.
Formalization Note The index set is countable; the indicators are measurable and take values in ; all are positive with HasSum equal to ; and every neighbourhood contains its own index. The four clauses use extended nonnegative error sums. Only the strict final comparison is conditional on , since strict is false.
Preamble
import Mathlib import Definitions.Def_TotalVariationDist import Definitions.Def_ChenStein_OneVar_Setting
Formal statement
namespace ChenStein.OneVar
open MeasureTheory ProbabilityTheory
open scoped ENNReal NNReal
/-- Arratia--Goldstein--Gordon (1989), p. 11, Theorem 1. -/
theorem theorem_1 {Ω I : Type*} [MeasurableSpace Ω] [Countable I]
(P : Measure Ω) [IsProbabilityMeasure P]
(X : I → Ω → ℕ) (hXm : ∀ α, Measurable (X α))
(hX01 : ∀ α ω, X α ω ≤ 1)
(hp : ∀ α, 0 < p P X α)
(lam : ℝ) (hlam : HasSum (p P X) lam) (hlam0 : 0 < lam)
(B : I → Set I) (hB : ∀ α, α ∈ B α) :
let q := (1 - Real.exp (-lam)) / lam
let a := b1 P X B + b2 P X B
let c := b3 P X B
let c' := b3' P X B
let firstBound := 2 * (a * ENNReal.ofReal q +
c' * ENNReal.ofReal (min 1 (1.4 * lam ^ (-(1/2 : ℝ)))))
ENNReal.ofReal (2 * MarkovChainCLT.tvDist (P.map (W X))
(poissonMeasure lam.toNNReal)) ≤ firstBound ∧
firstBound ≤ 2 * (a + c) ∧
ENNReal.ofReal |P.real {ω | W X ω = 0} - Real.exp (-lam)| ≤
(a + c') * ENNReal.ofReal q ∧
(a + c < ⊤ →
(a + c') * ENNReal.ofReal q <
ENNReal.ofReal (min 1 lam⁻¹) * (a + c)) := by sorry
end ChenStein.OneVar
Source
Arratia, Goldstein and Gordon, Two moments suffice for Poisson approximations: the Chen-Stein method, Ann. Probab. 17 (1989), p. 11, Theorem 1; https://doi.org/10.1214/aop/1176991491
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.