Corollary 1(c) — is a time-homogeneous Markov process
OpenHarrisonReimanRBM.Orthant.corollary_1_markovLet , let be a covariance matrix (symmetric and nonnegative definite), , and let be a nonnegative matrix with zeros on the diagonal and spectral radius strictly less than unity; . There is a family of Markov transition kernels on , depending only on , with the following property.
Let be any probability space carrying a -dimensional Brownian motion with covariance matrix , drift vector and almost surely, independent of the increments of . Let , and let , on the set where , with for all on the exceptional set. Then is a Markov process with respect to with stationary transition probabilities : for all and Borel ,
This is part (c) of Corollary 1 of Harrison and Reiman; parts (a) (adaptedness) and (b) (pathwise existence and uniqueness of ) are proved separately in the reduction of HarrisonReimanRBM.Orthant.corollary_1.
Formalization Note The conventions are exactly those of HarrisonReimanRBM.Orthant.corollary_1: the kernel family is quantified before the probability space, the Brownian motion and the initial law, so it depends on alone and not on (stationarity). The filtration is the uncompleted natural filtration of . The conditional probability is the conditional expectation of the indicator of . Brownian motion with a random initial state and the processes are the definitions of HarrisonReimanRBM.Orthant.Process.
import Mathlib import Definitions.Def_Reiman84_QueueLength_Paths import Definitions.Def_HarrisonReimanRBM_Orthant_Basic import Definitions.Def_HarrisonReimanRBM_Orthant_Process
namespace HarrisonReimanRBM.Orthant
open MeasureTheory ProbabilityTheory Reiman84.QueueLength
/-- Corollary 1(c), p. 305. Fix `K ≥ 1`, `Q` as in §1, a covariance matrix `A` (symmetric,
nonnegative definite) and a drift vector `b`. There is a family `κ = (κ_t)` of Markov kernels on
`ℝ^K`, depending only on `(Q, A, b)`, such that for every probability space `(Ω, 𝓕, P)`, every
Brownian motion `X` with covariance `A`, drift `b`, `X(0) ∈ S` a.s. and `X(0)` independent of
the increments, and `Y = ψ(X)`, `Z = φ(X)` (set to `0` off `{X ∈ C_S}`), the process `Z` is a
Markov process with respect to `𝓕ₜ = 𝓕(X(s); 0 ≤ s ≤ t)` with stationary transition
probabilities `κ`: `P[Z(s + t) ∈ B | 𝓕ₛ] = κ_t(Z(s), B)` almost surely, for all `s, t ≥ 0` and
Borel `B`. -/
theorem corollary_1_markov {K : ℕ} (hK : 0 < K) (Q : Matrix (Fin K) (Fin K) ℝ)
(hQ : IsReflectionMatrix Q) (A : Matrix (Fin K) (Fin K) ℝ) (hA : A.PosSemidef)
(b : Fin K → ℝ) :
∃ κ : ℝ → Kernel (Fin K → ℝ) (Fin K → ℝ), (∀ t, IsMarkovKernel (κ t)) ∧
∀ (Ω : Type) [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(X Y Z : Ω → ℝ → Fin K → ℝ),
IsBMFrom b A P X → IsRBMConstruction Q X Y Z →
∀ s t : ℝ, 0 ≤ s → 0 ≤ t → ∀ B : Set (Fin K → ℝ), MeasurableSet B →
P[Set.indicator {ω | Z ω (s + t) ∈ B} (fun _ => (1 : ℝ)) | naturalFiltration X s]
=ᵐ[P] fun ω => (κ t (Z ω s) B).toReal := by sorry
end HarrisonReimanRBM.Orthant