(2.3)–(2.5), p. 536 — S_λh solves wf(w) − λf(w + 1) = h(w) − 𝒫_λh and has the tail form
ProvedPoissonDepTrials.MixSqrt.stein_solutionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1poisson-approximationstein-chen-methodstein-equation
Let and let be a bounded real function on the nonnegative integers, with Poisson expectation . Let
and . Then:
- solves the Stein equation (2.3): for every ,
- for every it has the tail representation (second line of (2.5))
At the equation involves only , which is why the value of at plays no role. Substituting into the basic identity (2.2) gives the error representation (2.6); the two forms of are what Lemmas 3.1–3.3 bound in the regimes and .
Formalization Note The hypothesis is implicit on the page, where divides by . Boundedness of is given as for some .
Preamble
import Mathlib import Definitions.Def_PoissonDepTrials_MixSqrt_Setting open MeasureTheory ProbabilityTheory
Formal statement
namespace PoissonDepTrials.MixSqrt
/-- Chen (1975), §2, p. 536, (2.3)–(2.5). For `λ > 0` and bounded `h`, the function `S_λh` of
(2.5) (first line) solves the Stein equation `wf(w) − λf(w + 1) = h(w) − 𝒫_λh` for every `w ≥ 0`,
and for `w ≥ 1` it also has the tail form
`S_λh(w) = (w − 1)! λ^{−w} Σ_{k=w}^∞ [h(k) − 𝒫_λh] λ^k / k!` (second line of (2.5)). -/
theorem stein_solution (lam : ℝ) (hlam : 0 < lam) (h : ℕ → ℝ) (M : ℝ)
(hM : ∀ k, |h k| ≤ M) :
(∀ w : ℕ, (w : ℝ) * stein lam h w - lam * stein lam h (w + 1) = h w - poissonExp lam h) ∧
(∀ w : ℕ, 1 ≤ w → stein lam h w = ((w - 1).factorial : ℝ) * (lam ^ w)⁻¹ *
∑' k : ℕ, (h (k + w) - poissonExp lam h) * lam ^ (k + w) / ((k + w).factorial : ℝ)) := by sorry
end PoissonDepTrials.MixSqrt
Source
Chen, Poisson approximation for dependent trials, Ann. Probab. 3 (1975), p. 536, §2, (2.3), (2.4), (2.5)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.