Cesàro Pair Factorization via MET #
This file proves the pair factorization lemma for conditional expectations using Mean Ergodic Theorem (MET) and exchangeability.
Main result:
condexp_pair_factorization_MET: For exchangeable measures, CE[f(ω₀)·g(ω₁)|ℐ] factors into CE[f(ω₀)|ℐ]·CE[g(ω₀)|ℐ].
Proof strategy (CORRECTED - avoids false k=0 lag constancy):
- Apply tower property directly on g₁ (via Cesàro from index 1): CE[f(ω₀)·g(ω₁)|ℐ] = CE[f(ω₀)·CE[g(ω₀)|ℐ]|ℐ] (uses h_tower_of_lagConst_from_one which only needs k ≥ 1 lag constancy)
- Apply pull-out property: CE[f(ω₀)·CE[g(ω₀)|ℐ]|ℐ] = CE[g(ω₀)|ℐ]·CE[f(ω₀)|ℐ] (CE[g(ω₀)|ℐ] is ℐ-measurable)
Key insight: This requires EXCHANGEABILITY (via hExch), not just stationarity.
Split from: CesaroConvergence.lean (lines 1561-2014)
Local notation #
Pair factorization via MET + Exchangeability (Kallenberg's approach).
For EXCHANGEABLE measures μ on path space, the conditional expectation of f(ω₀)·g(ω₁) given the shift-invariant σ-algebra factors into the product of the individual conditional expectations.
Proof strategy (CORRECTED - avoids false k=0 lag constancy):
- Apply tower property directly on g₁ (via Cesàro from index 1): CE[f(ω₀)·g(ω₁)|ℐ] = CE[f(ω₀)·CE[g(ω₀)|ℐ]|ℐ] (uses h_tower_of_lagConst_from_one which only needs k ≥ 1 lag constancy)
- Apply pull-out property: CE[f(ω₀)·CE[g(ω₀)|ℐ]|ℐ] = CE[g(ω₀)|ℐ]·CE[f(ω₀)|ℐ] (CE[g(ω₀)|ℐ] is ℐ-measurable)
Key insight: This requires EXCHANGEABILITY (via hExch), not just stationarity.
The original k=0 lag constancy approach was FALSE. See Infrastructure.lean for details.