Documentation

TauCeti.Probability.Exchangeability.Arrays.DeFinetti

De Finetti's theorem for exchangeable arrays #

The consequences of de Finetti's theorem for the array symmetries: over a nonempty standard Borel state space, the rows and columns of a separately exchangeable array are conditionally i.i.d., as are the rows of a block of a jointly exchangeable one, and the directing measures produced this way inherit the symmetry of the array.

This is the array subtree's only direct import of the de Finetti summit. Arrays.Coding reaches DeFinetti.Theorem too, transitively through this module, which is as it should be: the coding representation is a de Finetti consequence. What changed is that the dependency now arrives through the one file whose subject it is.

Arrays.Basic carries the symmetry predicates and their elementary theory, Arrays.Block.Basic the combinatorics of blocks, and Arrays.MixingLaw the results that hold of any supplied mixing representative. Each of those is now independent of DeFinetti.Theorem, so a file needing only array symmetry — for instance Arrays.AldousHoover.Basic, which uses four declarations from Arrays.Basic — no longer depends on the representation theory at all.

The layering is therefore

array symmetry  →  consequences of a supplied mixture  →  de Finetti supplies one  →  coding

Main results #

De Finetti's theorem for the rows of a separately exchangeable array. Over a nonempty standard Borel state space α, the rows of a separately exchangeable array are conditionally i.i.d. as random elements of path space ℕ → α.

This is the first step of the standard route to the Aldous–Hoover representation. Path space is standard Borel because α is (StandardBorelSpace.pi_countable), so the hypotheses are exactly de Finetti's.

De Finetti's theorem for the columns of a separately exchangeable array.

De Finetti's theorem for the rows of a block of a jointly exchangeable array. Over a nonempty standard Borel state space, the rows of a block along injections with disjoint ranges are conditionally i.i.d. as random elements of path space.

De Finetti's theorem for the rows of a block of pairs. The conclusion simultaneously describes both orientations of the selected rectangular cross-block.

theorem TauCeti.Probability.SeparatelyExchangeable.exists_directing_arrayRow_mixingLaw_invariant {α : Type u_1} {Ω : Type u_2} [MeasurableSpace α] [MeasurableSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ × ℕ → Ω → α} (h : SeparatelyExchangeable μ X) (hX : ∀ (p : ℕ × ℕ), AEMeasurable (X p) μ) :
∃ (ν : Ω → MeasureTheory.ProbabilityMeasure (ℕ → α)), ConditionallyIIDWith μ (arrayRow X) ν ∧ ∀ (τ : Equiv.Perm ℕ), MeasureTheory.Measure.map (fun (ω : Ω) => (ν ω).map fun (x : ℕ → α) (k : ℕ) => x (τ k)) μ = MeasureTheory.Measure.map ν μ

De Finetti for the rows, with the inherited mixing-law symmetry. A separately exchangeable array has a directing measure for its row process whose law is invariant under every permutation of the path coordinates.

theorem TauCeti.Probability.SeparatelyExchangeable.exists_directing_arrayCol_mixingLaw_invariant {α : Type u_1} {Ω : Type u_2} [MeasurableSpace α] [MeasurableSpace Ω] [StandardBorelSpace α] [Nonempty α] {μ : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure μ] {X : ℕ × ℕ → Ω → α} (h : SeparatelyExchangeable μ X) (hX : ∀ (p : ℕ × ℕ), AEMeasurable (X p) μ) :
∃ (ν : Ω → MeasureTheory.ProbabilityMeasure (ℕ → α)), ConditionallyIIDWith μ (arrayCol X) ν ∧ ∀ (σ : Equiv.Perm ℕ), MeasureTheory.Measure.map (fun (ω : Ω) => (ν ω).map fun (x : ℕ → α) (k : ℕ) => x (σ k)) μ = MeasureTheory.Measure.map ν μ

De Finetti for the columns, with the inherited mixing-law symmetry.