Theorem 12, p. 1878 — an (α, θ) partition coagulated by (β, θ/α) has the joint law of an (αβ, θ) partition fragmented by (α, −αβ)
ProvedLambdaCoalescent.CoagFrag.theorem_12Let , and . For a pair of random partitions of the following conditions are equivalent:
- is an partition and is a -coagulation of ;
- is an partition and is an -fragmentation of .
Here an partition is an exchangeable random partition with EPF of (15); is a -coagulation of if, given , is the -coagulation of for an independent with law (Definition 5); and is an -fragmentation of if, given , is obtained by restricting independent partitions with law to the successive blocks of (Definition 11).
Since each condition fixes the joint law of , the theorem says that two measures on coincide: if are the laws with EPFs , , , , then for every measurable ,
The theorem describes one joint law of a pair of exchangeable partitions, refining , from both ends; with it links the family to Ewens' family, and through Kingman's correspondence it gives the Poisson–Dirichlet statement of Corollary 13.
Formalization Note "(i) (ii) for every pair " is the equality of the two joint laws: each condition determines the joint law (conditional laws given are meaningful only almost surely, so the conditions are encoded through joint laws), and each law is realized on the canonical space, so the equivalence for all pairs and the equality of laws imply each other. The equality is stated through (outer) measures of preimages of measurable sets. The four laws exist by Lemma 9, so the hypotheses are not vacuous; all four parameter pairs satisfy (14) under the hypotheses (, ). is a real division with . The fragmenting partitions are an i.i.d. sequence, one sample of Measure.infinitePi (fun _ => κ), indexed from in the order of least elements of the blocks of . EPFs are in the cancelled form of the Setting module, which matters at .
import Mathlib import Definitions.Def_LambdaCoalescent_CoagFrag_Setting
namespace LambdaCoalescent.CoagFrag
open MeasureTheory
theorem theorem_12 (α β θ : ℝ) (hα0 : 0 < α) (hα1 : α < 1) (hβ0 : 0 ≤ β) (hβ1 : β < 1)
(hθ : -(α * β) < θ)
(μ ν ρ κ : Measure LambdaCoalescent.Rates.PInf) [IsProbabilityMeasure μ] [IsProbabilityMeasure ν]
[IsProbabilityMeasure ρ] [IsProbabilityMeasure κ]
(hμ : IsEPFLaw (pdEPF α θ) μ) (hν : IsEPFLaw (pdEPF β (θ / α)) ν)
(hρ : IsEPFLaw (pdEPF (α * β) θ) ρ) (hκ : IsEPFLaw (pdEPF α (-(α * β))) κ) :
∀ S : Set (LambdaCoalescent.Rates.PInf × LambdaCoalescent.Rates.PInf), MeasurableSet S →
(μ.prod ν) {x | (x.1, coag x.1 x.2) ∈ S} =
(ρ.prod (Measure.infinitePi (fun _ : ℕ => κ))) {y | (frag y.1 y.2, y.1) ∈ S} := by sorry
end LambdaCoalescent.CoagFrag
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.