Documentation

SHSLib.Category.Levy

Lévy-system presentations: the common shape of the hybrid generators #

Davis's PDP generator (26.15), Hespanha's SHS generator (Eq. 9, in the form of Bect's Example 18), and the GSHS generator (28) of Bujorianu–Lygeros all have the shape of Bect's Lévy-system generator (Eqs. (4)–(6) of Bect 2010):

Lφ(x) = (drift part)φ(x) + λ(x) · (∫ φ(ℓ_x(y)) J(x, dy) − φ(x))

where λ is the total jump intensity, J(x, ·) the post-jump law (on a space of jump targets), and ℓ_x lands a jump target in the state space (the identity for Davis, the inclusion X ↪ X̄ for a GSHS, and "keep the clock" on the time-augmented state of an SHS).

A LevyData packages these ingredients; LevyData.toGenPres is the corresponding generator presentation. The framework files prove that each framework's native presentation is isomorphic (or equal) to its Lévy form.

The Lévy form is also what connects to Bect's measure-valued theory: LevyData.weakDynkin_of_satisfiesDynkin turns Dynkin's formula for the presentation (a GenPres property, hence transportable along morphisms) into the hypothesis WeakDynkin of Bect's Theorem 8, via GSHS.FPKData.weakDynkin_of_dynkin12.

structure SHSCat.LevyData :
Type (u + 1)

Lévy-system data: test functions with their drift (differential) generator, a total jump intensity, a post-jump law on a space of jump targets, and the landing map from jump targets to states.

  • State : Type u

    The state space.

  • Jump : Type u

    The space on which the post-jump law lives.

  • instJump : MeasurableSpace self.Jump

    The σ-algebra on jump targets.

  • Test : Type u

    The test functions.

  • ev : self.Test → self.State → ℝ

    Evaluation of test functions.

  • drift : self.Test → self.State → ℝ

    The drift (differential) part of the generator.

  • rate : self.State → ℝ

    The total jump intensity λ.

  • jump : self.State → MeasureTheory.Measure self.Jump

    The post-jump law J(x, ·).

  • land : self.State → self.Jump → self.State

    Landing of a jump target from the pre-jump state x.

Instances For
    def SHSCat.LevyData.jumpTerm (L : LevyData) (φ : L.Test) (x : L.State) :

    The jump part λ(x) (∫ φ(ℓ_x y) J(x,dy) − φ(x)) of the generator.

    Equations
    Instances For
      def SHSCat.LevyData.gen (L : LevyData) (φ : L.Test) (x : L.State) :

      The full Lévy-system generator drift + λ (J − I).

      Equations
      Instances For

        The generator presentation of Lévy-system data.

        Equations
        Instances For
          theorem SHSCat.LevyData.jumpTerm_eq_integral_sub (L : LevyData) (φ : L.Test) (x : L.State) [MeasureTheory.IsProbabilityMeasure (L.jump x)] (hint : MeasureTheory.Integrable (fun (y : L.Jump) => L.ev φ (L.land x y)) (L.jump x)) :
          L.jumpTerm φ x = L.rate x * ∫ (y : L.Jump), L.ev φ (L.land x y) - L.ev φ x ∂L.jump x

          For a probability post-jump law and integrable test functions, the jump part can be written in Bujorianu's form λ(x) ∫ (φ(ℓ_x y) − φ(x)) J(x, dy).

          From Dynkin's formula to Bect's generalized Dynkin formula (13) #

          theorem SHSCat.LevyData.weakDynkin_of_satisfiesDynkin (L : LevyData) [MeasurableSpace L.State] (D : GSHS.FPKData L.State L.Test) {Ω : Type u_1} [MeasurableSpace Ω] {P : MeasureTheory.Measure Ω} [MeasureTheory.IsProbabilityMeasure P] {X : ℝ → Ω → L.State} (hX : Measurable (Function.uncurry X)) (hdyn : L.toGenPres.SatisfiesDynkin P X) (hev : D.ev = L.ev) (hdrift : D.gen = L.drift) (hK : ∀ (φ : L.Test) (x : L.State), ∫ (y : L.State), L.ev φ y ∂D.K x = ∫ (y : L.Jump), L.ev φ (L.land x y) ∂L.jump x) (hlaw : ∀ (s : ℝ), 0 ≤ s → MeasureTheory.Measure.map (X s) P = D.law s) {genBound : L.Test → ℝ} (hgen_le : ∀ (φ : L.Test) (x : L.State), |L.drift φ x| ≤ genBound φ) (hdrift_meas : ∀ (φ : L.Test), Measurable (L.drift φ)) (hrate_meas : Measurable L.rate) (hrate_nonneg : ∀ (x : L.State), 0 ≤ L.rate x) {rateBound : ℝ} (hrate_le : ∀ (x : L.State), L.rate x ≤ rateBound) (hr : ∀ (s : ℝ), D.r s = (D.law s).withDensity fun (x : L.State) => ENNReal.ofReal (L.rate x)) :

          Bridge to Bect's measure-valued framework.

          Let X be a jointly measurable process whose laws are D.law s, for Bect data D built on the same test functions as L (hev), with the drift part of L as differential generator (hdrift), with a Markov kernel D.K having the same test-function averages as the post-jump law of L (hK), and with mean jump intensity r_s = λ μ_s (hr). If X satisfies Dynkin's formula for the Lévy presentation L.toGenPres, then D satisfies the generalized Dynkin formula (13), the standing hypothesis of Theorem 8 of Bect (2010).