Theorem 2.5.4 — the converse: coupling produces renovating events
DisprovedPalmQueueing.Recurrence.renovating_from_couplingTheorem 2.5.4, which the book introduces as "the following converse to Corollary 2.5.1 holds".
Let be a -valued stochastic recurrent sequence with constant initial condition . If couples in the strong backwards sense to , where is a finite stationary solution of , then admits a -compatible sequence of renovating events of positive probability.
So Borovkov's condition is not merely sufficient. For -valued recurrences with a constant initial condition, strong backwards coupling to a stationary solution and the existence of renovating events are the same thing — which is what makes the renovating-events method a characterisation rather than a technique.
One correction to the printed page. The book prints the conclusion for ; its proof (p.119) builds the renovating events for , namely on , and the printed version is false: with , , if and otherwise, and i.i.d. exponential, couples with while admits no renovating event of positive probability. The statement is the one the proof proves, for .
The renovating events produced are -compatible, , with ; that is exactly the hypothesis Corollary 2.5.1 consumes, which is what closes the circle.
Formalization Note. is ergodic (§2.5.1) and holds -a.s. -valuedness is carried as and mapping the non-negative orthant into itself.
import Mathlib import Definitions.Def_PalmQueueing_Recurrence_Renovating
/-!
# Theorem 2.5.4: the converse — coupling produces renovating events (§2.5.4, p.118)
-/
namespace PalmQueueing.Recurrence
open MeasureTheory Filter Topology
variable {Ω : Type*} [MeasurableSpace Ω]
/-- **Theorem 2.5.4** (p.118), which the book introduces as "the following converse to Corollary
2.5.1 holds".
Let `{W_n^{[C]}}` be a `ℝ₊^K`-valued stochastic recurrent sequence with constant initial condition
`C`. If `{W_n^{[C]}}` couples in the strong backwards sense to `{Z ∘ θⁿ}`, where `Z` is a finite
stationary solution of `Z ∘ θ = h(Z, ξ)`, then `{W_n^{[C]}}` (printed: `{W_n^{[0]}}`, see below) admits a `θ`-compatible sequence of
renovating events of positive probability.
So Borovkov's condition is not merely sufficient: for `ℝ₊^K`-valued recurrences with a constant
initial condition, strong backwards coupling to a stationary solution and the existence of
renovating events are the same thing. That is what makes the renovating-events method a
characterisation rather than a technique.
**One correction to the page.** The conclusion prints `{W_n^{[0]}}`, but the proof on p.119
builds the renovating events for `{W_n^{[C]}}` (`W^{[C]}_{n+m} = g(ξ_n, …, ξ_{n+m-1})` on
`A_n = θ^{-n}A`), and with `W^{[0]}` the statement is false: for `K = 1`, `C = 1`,
`h(w, ξ) = 1` if `w = 1` and `w + ξ` otherwise, with `{ξ_n}` i.i.d. exponential, `W^{[1]} ≡ 1`
couples with `Z ≡ 1` while `W^{[0]}_n = ξ_0 + … + ξ_{n-1}` admits no renovating event of positive
probability. The item states the conclusion for `W^{[C]}`, as proved.
`herg` is the standing assumption of §2.5.1 (p.104), `(P⁰, θ)` ergodic; the stationary relation
`Z ∘ θ = h(Z, ξ)` is `P⁰`-a.s. -/
theorem renovating_from_coupling {K : ℕ} {F : Type*}
(P0 : Measure Ω) [IsProbabilityMeasure P0] (θ : Ω ≃ᵐ Ω) (herg : Ergodic θ P0)
(h : (Fin K → ℝ) → F → Fin K → ℝ)
(hnonneg : ∀ (x : Fin K → ℝ) (z : F), 0 ≤ x → 0 ≤ h x z)
(xi : ℕ → Ω → F) (xi0 : Ω → F)
(hxi : ∀ (n : ℕ) (ω : Ω), xi n ω = xi0 ((θ : Ω → Ω)^[n] ω))
(C : Fin K → ℝ) (hC : 0 ≤ C)
(WC : ℕ → Ω → Fin K → ℝ)
(hWC : IsRecurrentSequence h xi (fun _ => C) WC)
(Z : Ω → Fin K → ℝ)
(hZ : ∀ᵐ ω ∂P0, Z ((θ : Ω → Ω) ω) = h (Z ω) (xi0 ω))
(hcouple : StrongBackwardsCoupling P0 θ WC Z) :
∃ (m : ℕ) (Phi : (Fin m → F) → Fin K → ℝ) (A : ℕ → Set Ω),
(∀ n, MeasurableSet (A n)) ∧
(∀ n : ℕ, A n = ((θ : Ω → Ω)^[n]) ⁻¹' A 0) ∧
0 < P0 (A 0) ∧
IsRenovating xi WC m Phi A := by sorry
end PalmQueueing.Recurrence
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.