Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(2.2), pp. 535–536 — the basic identity for E[Wf(W) − λf(W + 1)]

Proved
PoissonDepTrials.MixSqrt.basic_identity

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

bernoulli-trialsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1poisson-approximationstein-chen-method

Let X1,…,XnX_1,\dots,X_nX1​,…,Xn​ be Bernoulli random variables on a probability space (Ω,F,P)(\Omega,\mathcal F,P)(Ω,F,P), possibly dependent, with pi=P(Xi=1)p_i=P(X_i=1)pi​=P(Xi​=1), and put W=∑i=1nXiW=\sum_{i=1}^n X_iW=∑i=1n​Xi​ and λ=∑i=1npi\lambda=\sum_{i=1}^n p_iλ=∑i=1n​pi​. As on p. 535, XiX_iXi​ is taken to be identically zero when i≤0i\le 0i≤0 or i≥n+1i\ge n+1i≥n+1.

Fix a nonnegative integer mmm and let V(i)V^{(i)}V(i), YijY_{ij}Yij​, Yij′Y'_{ij}Yij′​ and Δ\DeltaΔ be as in the setting. Then for every real-valued function fff on the nonnegative integers,

E[Wf(W)−λf(W+1)]=∑∑0<∣i−j∣≤mE[XiXj Δf(Yi,j−1′+1)]+∑i=1nE[(Xi−pi)f(V(i)+1)]−∑∑∣i−j∣≤mpi E[Xj Δf(Yi,j−1+1)].\begin{aligned} E\bigl[Wf(W)-\lambda f(W+1)\bigr] &=\sum\sum_{0<|i-j|\le m}E\bigl[X_iX_j\,\Delta f(Y'_{i,j-1}+1)\bigr] +\sum_{i=1}^nE\bigl[(X_i-p_i)f(V^{(i)}+1)\bigr]\\ &\quad-\sum\sum_{|i-j|\le m}p_i\,E\bigl[X_j\,\Delta f(Y_{i,j-1}+1)\bigr]. \end{aligned}E[Wf(W)−λf(W+1)]​=∑0<∣i−j∣≤m∑​E[Xi​Xj​Δf(Yi,j−1′​+1)]+i=1∑n​E[(Xi​−pi​)f(V(i)+1)]−∑∣i−j∣≤m∑​pi​E[Xj​Δf(Yi,j−1​+1)].​

No assumption on the dependence between the trials is made.

This is the identity from which the whole method starts: choosing fff as the solution of the Stein equation turns its left side into Eh(W)−PλhEh(W)-\mathcal P_\lambda hEh(W)−Pλ​h, and choosing f(w)=wf(w)=wf(w)=w gives the variance identity (4.7).

Formalization Note The trials are a sequence X:N→Ω→NX:\mathbb N\to\Omega\to\mathbb NX:N→Ω→N that is measurable, almost surely {0,1}\{0,1\}{0,1}-valued, and identically 000 at index 000 and above nnn (the paper's padding convention, which changes no σ\sigmaσ-algebra). pip_ipi​ is defined as P(Xi=1)P(X_i=1)P(Xi​=1), not a free parameter. The double sums run over i,j∈{1,…,n}i,j\in\{1,\dots,n\}i,j∈{1,…,n}; terms with jjj outside this range vanish because Xj=0X_j=0Xj​=0 there. Index differences are computed in Z\mathbb ZZ.

Preamble
import Mathlib
import Definitions.Def_PoissonDepTrials_MixSqrt_Setting

open MeasureTheory ProbabilityTheory
Formal statement
namespace PoissonDepTrials.MixSqrt

/-- Chen (1975), §2, pp. 535–536, the basic identity (2.2). For an arbitrary sequence of
dependent Bernoulli trials and every real function `f` on the nonnegative integers,
`E[Wf(W) − λf(W + 1)] = ΣΣ_{0<|i−j|≤m} E[X_iX_jΔf(Y'_{i,j−1} + 1)]
  + Σ_i E[(X_i − p_i)f(V^{(i)} + 1)] − ΣΣ_{|i−j|≤m} p_i E[X_jΔf(Y_{i,j−1} + 1)]`. -/
theorem basic_identity {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (n : ℕ) (X : ℕ → Ω → ℕ) (hX : IsBernoulliTrials P n X) (m : ℕ) (f : ℕ → ℝ) :
    ∫ ω, ((W n X ω : ℝ) * f (W n X ω) - lam P n X * f (W n X ω + 1)) ∂P =
      (∑ i ∈ Finset.Icc 1 n, ∑ j ∈ Finset.Icc 1 n with 0 < |(i : ℤ) - (j : ℕ)| ∧ |(i : ℤ) - (j : ℕ)| ≤ m,
          ∫ ω, (X i ω : ℝ) * X j ω * delta f (Y' n m X i ((j : ℤ) - 1) ω + 1) ∂P)
      + (∑ i ∈ Finset.Icc 1 n, ∫ ω, ((X i ω : ℝ) - prob P X i) * f (V n m X i ω + 1) ∂P)
      - ∑ i ∈ Finset.Icc 1 n, ∑ j ∈ Finset.Icc 1 n with |(i : ℤ) - (j : ℕ)| ≤ m,
          prob P X i * ∫ ω, (X j ω : ℝ) * delta f (Y n m X i ((j : ℤ) - 1) ω + 1) ∂P := by sorry

end PoissonDepTrials.MixSqrt
Source
Chen, Poisson approximation for dependent trials, Ann. Probab. 3 (1975), pp. 535–536, §2, identity (2.2) (derived from (2.1))
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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