Theorem 2 — the discrete-time comparison theorem: coupled random sequences with for all a.s.
ProvedStochIneqPO.Comparison.comparison_theoremLet be partially ordered Polish spaces and with the coordinatewise order. Let be probability measures on , and for let be stochastic kernels from to . Assume and
Then there are random variables , with values in , , on a common probability space, such that , , the conditional distribution of given is , that of given is , and
This extends O'Brien's comparison theorem (and Kalmykov's) to random sequences in partially ordered Polish spaces: two processes whose initial laws and transition kernels are ordered can be realized on one probability space so that one path dominates the other for all times simultaneously.
Formalization Note The initial and conditional distributions determine the joint law of , which is the Ionescu-Tulcea measure; the conclusion states that the law of the whole sequence is Kernel.trajMeasure P₁ p and that of is trajMeasure Q₁ q (equivalent to the conditional-law formulation, which holds only up to null sets). Indices are 0-based: the paper's , are E (i - 1), X (i - 1), and the paper's kernel () is p (n - 2) : Kernel (Π i : Iic (n - 2), E i) (E (n - 1)). Hypothesis (4) is pointwise for all ordered pairs of histories; no stochastic monotonicity of the kernels is assumed. The conclusion (5) is one almost-sure event for all indices jointly. Every is a partially ordered Polish space by the paper's standing assumption (Sec. 1), although the theorem says "Polish spaces".
import Mathlib import Definitions.Def_StochIneqPO_Comparison_StochLE
namespace StochIneqPO.Comparison
open MeasureTheory ProbabilityTheory Finset
theorem comparison_theorem {E : ℕ → Type*} [∀ n, TopologicalSpace (E n)]
[∀ n, PolishSpace (E n)] [∀ n, MeasurableSpace (E n)] [∀ n, BorelSpace (E n)]
[∀ n, PartialOrder (E n)] [∀ n, OrderClosedTopology (E n)]
(P₁ Q₁ : Measure (E 0)) [IsProbabilityMeasure P₁] [IsProbabilityMeasure Q₁]
(p q : (n : ℕ) → Kernel (Π i : Iic n, E i) (E (n + 1)))
[∀ n, IsMarkovKernel (p n)] [∀ n, IsMarkovKernel (q n)]
(hPQ : StochLE P₁ Q₁)
(hpq : ∀ n, ∀ x y : Π i : Iic n, E i, x ≤ y → StochLE (p n x) (q n y)) :
∃ (Ω : Type) (_ : MeasurableSpace Ω) (μ : Measure Ω), IsProbabilityMeasure μ ∧
∃ (X Y : (n : ℕ) → Ω → E n), (∀ n, Measurable (X n)) ∧ (∀ n, Measurable (Y n)) ∧
μ.map (fun ω n => X n ω) = Kernel.trajMeasure P₁ p ∧
μ.map (fun ω n => Y n ω) = Kernel.trajMeasure Q₁ q ∧
∀ᵐ ω ∂μ, ∀ n, X n ω ≤ Y n ω := by sorry
end StochIneqPO.Comparison
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.