Documentation

SHSLib.Core.DavisMarkedJumpRelaxedRegime

The model-level statements with a random lower bound on the holding times #

The model-level statements for the observation filtration of the marked jump history ask for a deterministic strictly positive lower bound d on every holding time. That requirement enters only through the integrability of the elapsed occupation envelope, where it is used to bound the number of completed jumps by t / d; the relaxed regime already replaces it by a random bound d(ω) > 0 whose reciprocal is integrable.

This module runs the relaxed regime with the observation filtration: the deterministic-cap increment law comes from the bounded intensity, the envelope integrability from the random lower bound, and the two together give the martingale property of the occupation-compensated test-function process and the Dynkin identity. The uniform intensity bound is retained — it is what bounds the centered reset profile below an age horizon — but the holding times are now only required to stay away from zero pathwise.

theorem SHS.martingale_davisElapsedCompensatedGeneratorProcess_markedJump_of_relaxedRegime {Q : Type u_1} [MeasurableSpace Q] [MeasurableSingletonClass Q] {n m : ℕ} {S : SHS Q n m} {hflow : S.HasGlobalFlow} {hf : S.DavisFlowRegular hflow} {hr : S.DavisResetRegular} [S.IntensitiesMeasurable] {Ω : Type u_2} [mΩ : MeasurableSpace Ω] {P : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure P] {ξ : DavisJumpNoise m Ω} {hξ : ξ.MeasurableNoise} (hne : ξ.NonExplosive) {initial : DavisState Q n} {a g : DavisState Q n → ℝ} (hholding : DavisJumpNoise.ConditionalHoldingTimeLaw S hflow hf hr P ξ hξ initial) (hchan : DavisJumpNoise.ConditionalChannelLaw S hflow P ξ initial) (hcont : S.DavisHazardContinuousAlongFlow hflow) (hgcont : S.DavisResetOccupationContinuousAlongFlow hflow g) (hreg : S.DavisResetOccupationRegular hflow g) (hg : Measurable g) (Cg : ℝ) (hgC : ∀ (z : DavisState Q n), |g z| ≤ Cg) (hEcont : ∀ (z : DavisState Q n), Continuous fun (s : ℝ) => S.davisExpectedResetIncrement g (S.davisFlow hflow s.toNNReal z)) (Lam : ℝ) (hLam : ∀ (z : DavisState Q n), S.totalIntensity z ≤ Lam) (Lg : ℝ) (hLg0 : 0 ≤ Lg) (hLg : ∀ (z : DavisState Q n) (u v : NNReal), |S.davisExpectedResetIncrement g (S.davisFlow hflow u z) - S.davisExpectedResetIncrement g (S.davisFlow hflow v z)| ≤ Lg * |↑u - ↑v|) (d : Ω → NNReal) (hdpos : ∀ (ω : Ω), 0 < d ω) (hd : ∀ (i : ℕ) (ω : Ω), d ω ≤ ξ.holdingTime i ω) (hdint : ∀ (t : NNReal), MeasureTheory.Integrable (fun (ω : Ω) => ↑t / ↑(d ω)) P) (hga : S.DavisFlowIntegralIdentity hflow g a) :
MeasureTheory.Martingale (S.davisElapsedCompensatedGeneratorProcess hflow ξ hne initial a g) (DavisJumpNoise.markedJumpFiltration S hflow hf hr ξ hξ hne initial) P

The occupation-compensated test-function process is a martingale for the observation filtration of the marked jump history, under a bounded test function, a bounded intensity, a Lipschitz expected reset increment along the flow, and holding times bounded below by a random strictly positive variable whose reciprocal is integrable.

theorem SHS.integral_davisElapsedCompensatedGeneratorProcess_eq_zero_markedJump_of_relaxedRegime {Q : Type u_1} [MeasurableSpace Q] [MeasurableSingletonClass Q] {n m : ℕ} {S : SHS Q n m} {hflow : S.HasGlobalFlow} {hf : S.DavisFlowRegular hflow} {hr : S.DavisResetRegular} [S.IntensitiesMeasurable] {Ω : Type u_2} [mΩ : MeasurableSpace Ω] {P : MeasureTheory.Measure Ω} [MeasureTheory.IsFiniteMeasure P] {ξ : DavisJumpNoise m Ω} {hξ : ξ.MeasurableNoise} (hne : ξ.NonExplosive) {initial : DavisState Q n} {a g : DavisState Q n → ℝ} (hholding : DavisJumpNoise.ConditionalHoldingTimeLaw S hflow hf hr P ξ hξ initial) (hchan : DavisJumpNoise.ConditionalChannelLaw S hflow P ξ initial) (hcont : S.DavisHazardContinuousAlongFlow hflow) (hgcont : S.DavisResetOccupationContinuousAlongFlow hflow g) (hreg : S.DavisResetOccupationRegular hflow g) (hg : Measurable g) (Cg : ℝ) (hgC : ∀ (z : DavisState Q n), |g z| ≤ Cg) (hEcont : ∀ (z : DavisState Q n), Continuous fun (s : ℝ) => S.davisExpectedResetIncrement g (S.davisFlow hflow s.toNNReal z)) (Lam : ℝ) (hLam : ∀ (z : DavisState Q n), S.totalIntensity z ≤ Lam) (Lg : ℝ) (hLg0 : 0 ≤ Lg) (hLg : ∀ (z : DavisState Q n) (u v : NNReal), |S.davisExpectedResetIncrement g (S.davisFlow hflow u z) - S.davisExpectedResetIncrement g (S.davisFlow hflow v z)| ≤ Lg * |↑u - ↑v|) (d : Ω → NNReal) (hdpos : ∀ (ω : Ω), 0 < d ω) (hd : ∀ (i : ℕ) (ω : Ω), d ω ≤ ξ.holdingTime i ω) (hdint : ∀ (t : NNReal), MeasureTheory.Integrable (fun (ω : Ω) => ↑t / ↑(d ω)) P) (hga : S.DavisFlowIntegralIdentity hflow g a) (t : NNReal) :
∫ (ω : Ω), S.davisElapsedCompensatedGeneratorProcess hflow ξ hne initial a g t ω ∂P = 0

The Dynkin identity for the observation filtration of the marked jump history, under a bounded test function, a bounded intensity, a Lipschitz expected reset increment along the flow, and holding times bounded below by a random strictly positive variable whose reciprocal is integrable.