Documentation

TauCeti.Probability.DeFinetti.BlockFactorization

Block-product factorisation and the de Finetti summit #

The block-product factorisation of the conditional expectation, and — built directly on it — the de Finetti summit for the reverse-martingale route.

Given that the selected coordinates of a contractable process are conditionally independent over the tail σ-algebra, the conditional expectation of a block-indicator product factors as a product of the directing measure on the coordinate sets:

μ[blockIndicatorProd X k C | tailProcess X] =ᵐ fun ω => ∏ i, (directingMeasure μ X ω).real (C i).

This chains Mathlib's iCondIndepFun_iff_condExp_inter_preimage_eq_mul (conditional independence ⟺ product of indicator conditional expectations) with Contractable.directingMeasure_ae_eq_condExp_coord (each coordinate's conditional law is the directing measure). The tail factorization condExp_blockIndicatorProd_tailProcess_ae_eq_prod (from TailFactorization) then discharges the finite-block rectangle identity for directingProbabilityMeasure μ X, exactly what mixedIIDWith_of_forall_rectangles consumes — so the whole chain assembles here.

Main results #

The ..._of_iCondIndepFun_tailProcess theorems expose the intermediate reduction (de Finetti given tail conditional independence of the coordinates). All of the rectangle-mixture staging lemmas and the standard-Borel-Ω existential are private (proof staging); the integrability and real/ℝ≥0∞ conversion facts this file runs on are DirectingMeasure/Integral.lean.

The reverse-martingale ("third") proof follows Kallenberg, Probabilistic Symmetries and Invariance Principles, Theorem 1.1 (pp. 26–28). Adapted from cameronfreer/exchangeability (DeFinetti/ViaMartingale/CommonEnding.lean and DeFinetti/TheoremViaMartingale.lean, pin e0532e59ceff23edab44dda9ab0655debbc9cc22); here the factorisation is obtained by reuse of Mathlib's iCondIndepFun characterisation rather than a hand-rolled π-system.

theorem TauCeti.Probability.condExp_blockIndicatorProd_ae_eq_prod_of_iCondIndepFun_tailProcess {Ω : Type u_1} {α : Type u_2} {mΩ : MeasurableSpace Ω} [MeasurableSpace α] [StandardBorelSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) {m : ℕ} {k : Fin m → ℕ} {C : Fin m → Set α} (hC : ∀ (i : Fin m), MeasurableSet (C i)) (hCI : ProbabilityTheory.iCondIndepFun (tailProcess X) ⋯ (fun (i : Fin m) => X (k i)) μ) :
μ[blockIndicatorProd X k C | tailProcess X] =ᵐ[μ] fun (ω : Ω) => ∏ i : Fin m, (directingMeasure μ X ω).real (C i)

Block-product factorisation of the conditional expectation (given tail conditional independence). If the selected coordinates fun i => X (k i) are conditionally independent given the tail, then the conditional expectation of the block-indicator product factors as ∏ i, (directingMeasure μ X ·).real (C i).

theorem TauCeti.Probability.blockLaw_eq_lintegral_prod_directingMeasure_of_iCondIndepFun_tailProcess {Ω : Type u_1} {α : Type u_2} {mΩ : MeasurableSpace Ω} [MeasurableSpace α] [StandardBorelSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) {m : ℕ} {k : Fin m → ℕ} {C : Fin m → Set α} (hC : ∀ (i : Fin m), MeasurableSet (C i)) (hCI : ProbabilityTheory.iCondIndepFun (tailProcess X) ⋯ (fun (i : Fin m) => X (k i)) μ) :
(blockLaw μ X k) (Set.univ.pi C) = ∫⁻ (ω : Ω), ∏ i : Fin m, (directingMeasure μ X ω) (C i) ∂μ

Block law of a rectangle as a directing-measure mixture (given tail conditional independence). The block law of the rectangle ∏ᵢ C i is the μ-average of the directing-measure product ∏ i, directingMeasure μ X ω (C i).

theorem TauCeti.Probability.mixedIIDWith_of_iCondIndepFun_tailProcess {Ω : Type u_1} {α : Type u_2} {mΩ : MeasurableSpace Ω} [MeasurableSpace α] [StandardBorelSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) (hCI : ∀ (m : ℕ) (k : Fin m → ℕ), Function.Injective k → ProbabilityTheory.iCondIndepFun (tailProcess X) ⋯ (fun (i : Fin m) => X (k i)) μ) :

de Finetti reduces to conditional independence over the tail (directing-measure form). If every finite injective selection of coordinates of a contractable process is conditionally independent given the tail σ-algebra, then the process is mixed i.i.d. with directing measure the tail conditional law directingProbabilityMeasure μ X.

theorem TauCeti.Probability.mixedIID_of_iCondIndepFun_tailProcess {Ω : Type u_1} {α : Type u_2} {mΩ : MeasurableSpace Ω} [MeasurableSpace α] [StandardBorelSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) (hCI : ∀ (m : ℕ) (k : Fin m → ℕ), Function.Injective k → ProbabilityTheory.iCondIndepFun (tailProcess X) ⋯ (fun (i : Fin m) => X (k i)) μ) :

de Finetti reduces to conditional independence over the tail. The existential form of mixedIIDWith_of_iCondIndepFun_tailProcess: under the same tail conditional-independence hypothesis, a contractable process is MixedIID.

theorem TauCeti.Probability.condExp_blockIndicatorProd_prefix_ae_eq_prod_directingMeasure {Ω : Type u_1} {α : Type u_2} {mΩ : MeasurableSpace Ω} [MeasurableSpace α] [StandardBorelSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) {r : ℕ} {C : Fin r → Set α} (hC : ∀ (i : Fin r), MeasurableSet (C i)) :
μ[blockIndicatorProd X (fun (i : Fin r) => ↑i) C | tailProcess X] =ᵐ[μ] fun (ω : Ω) => ∏ i : Fin r, (directingMeasure μ X ω).real (C i)

Prefix block factorization at the directing measure. For a contractable process, the conditional expectation of the length-r prefix indicator product given the tail σ-algebra tailProcess X is a.e. the product of directing-measure evaluations ∏ i, (directingMeasure μ X ω).real (C i) on the coordinate sets.

This is the base case for rectangle identities over a contractable process: a finite block of coordinates is replaced by the directing measure under the tail σ-algebra, and identities for arbitrary selections reduce to it.

theorem TauCeti.Probability.mixedIIDWith_of_contractable {Ω : Type u_1} {α : Type u_2} {mΩ : MeasurableSpace Ω} [MeasurableSpace α] [StandardBorelSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) :

de Finetti's theorem, directed form. A contractable process on a standard Borel space is mixed i.i.d. with witness the tail conditional law directingProbabilityMeasure μ X.

theorem TauCeti.Probability.mixedIID_of_contractable {Ω : Type u_3} {α : Type u_4} [MeasurableSpace Ω] [MeasurableSpace α] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Contractable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) :

de Finetti's theorem, mixture form: contractable ⇒ mixed i.i.d. A contractable process valued in a standard Borel space, on an arbitrary measurable sample space, is mixed i.i.d. This is the roadmap's mixedIID_of_contractable; the sharp summit conditionallyIID_of_contractable, which concludes the joint-law disintegration rather than only the mixture identity, is in TauCeti.Probability.DeFinetti.Theorem.

theorem TauCeti.Probability.mixedIID_of_exchangeable {Ω : Type u_3} {α : Type u_4} [MeasurableSpace Ω] [MeasurableSpace α] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ → Ω → α} (hX : Exchangeable μ X) (hX_meas : ∀ (n : ℕ), Measurable (X n)) :

de Finetti's theorem, mixture form: exchangeable ⇒ mixed i.i.d. An exchangeable process valued in a standard Borel space, on an arbitrary measurable sample space, is mixed i.i.d.