Theorem 1 — six equivalent characterizations of (Strassen)
ProvedStochIneqPO.Comparison.theorem1_tfaeLet be a partially ordered Polish space: a Polish space with a closed partial order and its Borel -algebra. Let be probability measures on . The following six conditions are equivalent.
- .
- There is a probability measure on with support in , first marginal and second marginal .
- There are a real random variable (on some probability space) and measurable maps with for all , such that has law and has law .
- There are -valued random variables on a common probability space with almost surely and ().
- There is an upward stochastic kernel on with , the second marginal of .
- for every closed increasing set .
The equivalence (i) (iv) is the monotone coupling characterization of the stochastic order; (i) (ii) is a special case of Strassen's theorem (1965). Every comparison result of the paper reduces to this theorem. The platform item PalmQueueing.Ordering.strassen_st is the special case (i) (iv) for with the coordinatewise order.
Formalization Note "Partially ordered Polish space" is [PolishSpace E] [BorelSpace E] [PartialOrder E] [OrderClosedTopology E] (the order's graph is closed). Random variables live on some probability space Ω : Type. "Support in " is ; "" is pointwise on all of ; is (P₁ ⊗ₘ k).snd; the kernel in (v) is a Markov kernel. A closed set is Borel, so (vi) needs no separate measurability hypothesis.
import Mathlib import Definitions.Def_StochIneqPO_Comparison_StochLE import Definitions.Def_StochIneqPO_Comparison_IsUpward
namespace StochIneqPO.Comparison
open MeasureTheory ProbabilityTheory
theorem theorem1_tfae {E : Type*} [TopologicalSpace E] [PolishSpace E] [MeasurableSpace E]
[BorelSpace E] [PartialOrder E] [OrderClosedTopology E]
(P₁ P₂ : Measure E) [IsProbabilityMeasure P₁] [IsProbabilityMeasure P₂] :
List.TFAE
[ -- (i)
StochLE P₁ P₂,
-- (ii)
∃ lam : Measure (E × E), IsProbabilityMeasure lam ∧ lam {z : E × E | z.1 ≤ z.2}ᶜ = 0 ∧
lam.map Prod.fst = P₁ ∧ lam.map Prod.snd = P₂,
-- (iii)
∃ (Ω : Type) (_ : MeasurableSpace Ω) (μ : Measure Ω), IsProbabilityMeasure μ ∧
∃ (Z : Ω → ℝ) (f g : ℝ → E), Measurable Z ∧ Measurable f ∧ Measurable g ∧
(∀ t, f t ≤ g t) ∧ μ.map (fun ω => f (Z ω)) = P₁ ∧ μ.map (fun ω => g (Z ω)) = P₂,
-- (iv)
∃ (Ω : Type) (_ : MeasurableSpace Ω) (μ : Measure Ω), IsProbabilityMeasure μ ∧
∃ X₁ X₂ : Ω → E, Measurable X₁ ∧ Measurable X₂ ∧ (∀ᵐ ω ∂μ, X₁ ω ≤ X₂ ω) ∧
μ.map X₁ = P₁ ∧ μ.map X₂ = P₂,
-- (v)
∃ k : Kernel E E, IsMarkovKernel k ∧ IsUpward k ∧ P₂ = (P₁ ⊗ₘ k).snd,
-- (vi)
∀ B : Set E, IsClosed B → IsUpperSet B → P₁ B ≤ P₂ B ] := by sorry
end StochIneqPO.Comparison
Confirmed by the mission captain (proposal self-audit).
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.