Documentation

TauCeti.Probability.Exchangeability.MixedIID.Mixture

The path law of a mixed i.i.d. process #

A mixed i.i.d. process has, as its path law, a mixture of infinite product measures: for a mixing representative ν,

pathLaw μ X = ∫ P^{⊗ℕ} d(μ.map ν)(P)

written in the Measure.bind idiom.

Main results #

The witness-level representation and uniqueness results need only [IsFiniteMeasure μ], a.e.-measurable coordinates, and the witness. MixedIID.existsUnique_mixingLaw assumes [IsProbabilityMeasure μ] only so that the unique mixing law can be bundled as a ProbabilityMeasure. No standard-Borel hypothesis appears: that is the cost of supplying a canonical witness, not of using one, so this file carries no de Finetti dependency.

Implementation #

MixedIIDWith constrains only the finite blocks, so the work is the passage from finite blocks to the whole path. The two sides are compared through their finite-dimensional prefix marginals via measure_eq_of_prefixProj_map_eq: on the left map_prefixProj_pathLaw and the block identity, on the right map_bind to push the marginal inside the mixture, map_prefixProj_infinitePi_const to recognise each prefix marginal of an infinite power as the corresponding finite power, and bind_map to re-index the mixture over μ rather than over μ.map ν.

This advances TauCetiRoadmap/Exchangeability/README.md, Layer 6, the directing-measure API bullet asking for the mixture-of-product-measures form. That bullet also asks for π to be the unique law of ν, which mixedIID_mixingLaw_unique now supplies: the mixture representation turns two witnesses into the same Measure.bind, and injectivity of π ↦ π.bind (P ↦ P^{⊗ℕ}) (Measure.ext_of_bind_infinitePi_eq) identifies the mixing laws. The roadmap name deFinetti_mixture, which derives the unique representation from exchangeability rather than assuming a witness, lives in TauCeti.Probability.DeFinetti.Representation.

The mixture representation of a path law. If ν is a mixing representative for X, the path law of X is the μ.map ν-mixture of the infinite product measures P^{⊗ℕ}.

theorem TauCeti.Probability.mixedIID_mixingLaw_eq_of_pathLaw_eq {Ω : Type u_1} {α : Type u_2} [MeasurableSpace Ω] [MeasurableSpace α] {Ω' : Type u_3} [MeasurableSpace Ω'] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {μ' : MeasureTheory.Measure Ω'} [MeasureTheory.IsFiniteMeasure μ'] {X : ℕ → Ω → α} {Y : ℕ → Ω' → α} {ν : Ω → MeasureTheory.ProbabilityMeasure α} {ν' : Ω' → MeasureTheory.ProbabilityMeasure α} (h : MixedIIDWith μ X ν) (h' : MixedIIDWith μ' Y ν') (hpath : pathLaw μ X = pathLaw μ' Y) :

The mixing law is a function of the path law. Two mixed i.i.d. processes with the same path law — on possibly different sample spaces — have mixing representatives with the same law.

This is the sharp form of mixing-law uniqueness: the mixture representation writes the path law as π.bind (P ↦ P^{⊗ℕ}) for π = μ.map ν, and that assignment is injective (Measure.ext_of_bind_infinitePi_eq), so the path law already determines π. Comparing two witnesses for one and the same process (mixedIID_mixingLaw_unique) is the special case hpath = rfl.

Uniqueness of the mixing law. Two mixing representatives for the same process induce the same law on ProbabilityMeasure α.

Only the law μ.map ν is unique, not the witness. For a nondegenerate mixing law an independent copy of a mixing representative is another one, so no witness-level a.e.-equality theorem can conclude ν =ᵐ[μ] ν' from MixedIIDWith alone; a.e. uniqueness of the witness belongs to the conditional predicate (conditionallyIID_ae_unique).

Finiteness of μ is load-bearing rather than decorative: for an infinite base measure, distinct mixing measures can produce identical ∞-valued finite-dimensional mixtures, so mixing-law uniqueness fails at the hypothesis-light generality the definitions otherwise enjoy.

Normalization, however, is not needed. The roadmap states this target with [IsProbabilityMeasure μ], but the justification it gives only separates finite from infinite base measures, and the proof goes through for any finite μ.

Existence and uniqueness of the mixing law. A mixed i.i.d. process under a probability law has a unique probability measure π on ProbabilityMeasure α such that its path law is the π-mixture of the infinite product measures P^{⊗ℕ}.

This identifies the mixing law intrinsically from the path law, rather than merely comparing the pushforwards of two supplied mixing representatives as mixedIID_mixingLaw_unique does.