Theorem 2.4.1 — coupling implies convergence in variation
ProvedPalmQueueing.Loynes.coupling_convergenceTheorem 2.4.1. If couples with a -compatible sequence , then the sequence converges in variation to as tends to :
where is the law of the whole shifted trajectory on the infinite product space.
The convergence is in variation on the law of the entire trajectory, not of a single coordinate and not merely in distribution — which is what makes it the right notion of "reaches the stationary regime". The book derives it from the coupling inequality
obtained by splitting each event on ; finiteness of the coupling time then gives (2.4.1), with no further hypothesis. In particular, if is -stationary, (2.4.3).
The statement quantifies the measurable set inside the index, which is exactly the supremum defining the distance in variation being small. is any measurable space: p.99 takes Polish and introduces the metric on only to identify that space's Borel field with the product -field, which is used directly here.
Formalization Note. The coupling time is an -valued random variable (measurable), and for all holds -almost surely.
import Mathlib import Definitions.Def_PalmQueueing_Loynes_Coupling
/-!
# Theorem 2.4.1: coupling implies convergence in variation (§2.4.1, p.99)
-/
namespace PalmQueueing.Loynes
open MeasureTheory Filter Topology
variable {Ω : Type*} [MeasurableSpace Ω]
/-- **Theorem 2.4.1** (§2.4.1, p.99). If `{X_n}` couples with a `θ`-compatible sequence
`{Z ∘ θⁿ}`, then the sequence `{X_{n+k}}_{n ≥ 0}` converges in variation to `{Z ∘ θⁿ}` as `k`
tends to `∞`:
`(2.4.1) lim_{k→∞} | P̃_{X,k} − P̃_{Z,k} | = 0`,
where `P̃_{X,k}` is the law on `(E^∞, E^∞)` of the whole shifted trajectory
`X̃_k = (X_k, X_{k+1}, …)`.
The convergence is **in variation on the law of the entire trajectory**, not of one coordinate and
not in distribution: `|P̃_{X,k} − P̃_{Z,k}| = sup_{C} |P̃_{X,k}(C) − P̃_{Z,k}(C)|` over all
measurable `C ⊆ E^∞`. That is why it is stated with the set quantified inside the index `K`, which
is exactly the supremum being small. The book derives it from the **coupling inequality**
`(2.4.2) |P̃_{X,k} − P̃_{Z,k}| ≤ P(N > k)`, which is finite-`N` plus nothing else.
`E` is any measurable space here. p.99 takes `(E, ℰ)` Polish with its Borel field and introduces
the metric `d_∞` on `E^∞` only to identify that space's Borel field with the product σ-field; the
statement uses the product σ-field directly, so the metric is not needed. -/
theorem coupling_convergence {E : Type*} [MeasurableSpace E]
(P : Measure Ω) [IsProbabilityMeasure P]
(θ : Ω → Ω) (X Z : ℕ → Ω → E)
(hXmeas : ∀ n, Measurable (X n)) (hZmeas : ∀ n, Measurable (Z n))
(hZcomp : IsShiftCompatible θ Z)
(hcouple : Couple P X Z) :
∀ ε : ℝ, 0 < ε → ∃ K : ℕ, ∀ k : ℕ, K ≤ k →
∀ C : Set (ℕ → E), MeasurableSet C →
|(P {ω | (fun n : ℕ => X (n + k) ω) ∈ C}).toReal
- (P {ω | (fun n : ℕ => Z (n + k) ω) ∈ C}).toReal| ≤ ε := by sorry
end PalmQueueing.Loynes
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.