LEMMA 1, p. 226 — R(t) = Y(t) − λtV(t) is a continuous-time martingale, in the form (5)
ProvedWolffPASTA.Main.lemma_1Throughout, is a probability space; is a process with values in a measurable space ; is a Poisson process at rate ; is a set with for every ; ; the sample paths of are left continuous with right-hand limits w.p. 1; is jointly measurable on ; and lack of anticipation holds with respect to the joint history: for each , is independent of , the σ-field generated by .
Let , (display (4)). Then is a continuous-time martingale with respect to its own history: every , , is integrable, and for , , almost surely
compares the number of arrivals who find in with times the time spent in ; Lemma 1 says the comparison has no drift, and Lemma 2 turns it into a strong law.
Formalization Note The conditioning σ-field is the one generated by , , exactly as in (5); the stronger martingale property with respect to is a consequence of (6) and is not stated here. Lack of anticipation is assumed against rather than against alone as in the printed LAA (p. 225). The page's proof of (6) uses this joint independence ("from LAA and the independent-increments property … for any set in "), and the printed LAA together with independent increments does not give it: with on , on , afterwards, where is the parity of masked by a fair bit computed from the arrival times in , is independent of the future increments after time , so the printed LAA holds, yet on an event of positive probability. The joint condition implies the printed one.
import Mathlib import Definitions.Def_WolffPASTA_Main_Setting open MeasureTheory ProbabilityTheory Filter Topology
namespace WolffPASTA.Main
/-- LEMMA 1, p. 226, in the form (5): `R` is a continuous-time martingale, i.e. each `R(t)` is
integrable and `E{R(t + h) - R(t) | R(s), 0 ≤ s ≤ t} = 0` for `t ≥ 0`, `h > 0`.
Lack of anticipation is taken against the joint history `ℱₜ` (`LAAJoint`), the independence the
page's proof of (6) uses; the printed LAA (future increments independent of the past of `U` only)
does not imply (6) or LEMMA 1 (see the Formalization Note). -/
theorem lemma_1
{Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
{S : Type*} [MeasurableSpace S] (N : ℝ → Ω → S)
(A : ℝ → Ω → ℕ) (lam : ℝ) (hlam : 0 < lam) (hA : IsPoissonProcess P A lam)
(B : Set S) (hB : ∀ t, 0 ≤ t → MeasurableSet {ω | N t ω ∈ B})
(hpath : PathRegular P (U N B))
(hjoint : JointlyMeasurable (U N B))
(hLAA : LAAJoint P A (U N B)) :
(∀ t, 0 ≤ t → Integrable (R (U N B) A lam t) P) ∧
∀ t h, 0 ≤ t → 0 < h →
P[fun ω => R (U N B) A lam (t + h) ω - R (U N B) A lam t ω | Rhist (R (U N B) A lam) t]
=ᵐ[P] 0 := by sorry
end WolffPASTA.Main
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.