integral bilinear pairing
ProvedMathFin.integral_bilinear_pairingcontext-reviewlevystochcalcsource-import
The source theorem MathFin.integral_bilinear_pairing is extracted from MathFin.Foundations.PoissonCompensatedBilinear. The original Lean proposition is retained verbatim, and its proof is tracked separately for remote certification.
Preamble
import Definitions.Def_MathFin_Foundations_PoissonCompensatedIsometryAdapted_F001
/-
Copyright (c) 2026 Raphael Coelho. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Raphael Coelho
-/
/-!
# The compensated Poisson random measure is an orthogonal martingale measure
The diagonal second moment `𝔼[Ñ(B)²] = ν̂(B)` (`compensated_integral_sq`) is the `C = D` case of the
full **covariance** of the compensated Poisson random measure:
`𝔼[Ñ(C)·Ñ(D)] = ν̂(C ∩ D)` for measurable finite-intensity `C, D`.
`Ñ` is thus an *orthogonal martingale measure* with (deterministic) intensity `ν̂`: increments on
disjoint regions are uncorrelated, and the covariance of general increments is the intensity of their
overlap. The proof decomposes `C = (C∩D) ⊔ (C\D)` and `D = (C∩D) ⊔ (D\C)` into three disjoint pieces
whose counts are independent (the independent-scattering field `indep_of_disjoint_region`); the three
disjoint cross-pairs vanish (`integral_mul_compensated_eq_zero`) and the diagonal `𝔼[Ñ(C∩D)²]` is
`ν̂(C∩D)`.
This covariance is the bilinear core the compensated-integral isometry is built on (the jump analogue
of `WienerIntegralL2.covariance_increment_aux`), and the first rung of the Itô–Lévy integral
*operator* (the CLM over the characterised predictable `L²`).
## Provenance
Built on our own `PoissonRandomMeasure` / `PoissonCompensatedIsometryAdapted` (PRM field shape
consulted from `cgarryZA/LevyStochCalc`, Apache-2.0, cited — a rendering of Applebaum 2009 Def.
2.3.1; the proofs are ours). Reference: Applebaum, *Lévy Processes and Stochastic Calculus*, CUP
2009, Thm 4.2.3.
-/
@[expose] public section
namespace MathFin
open MeasureTheory ProbabilityTheory
open scoped NNReal ENNReal
variable {Ω : Type*} [MeasurableSpace Ω] {E : Type*} [MeasurableSpace E]
{P : Measure Ω} [IsProbabilityMeasure P] {ν : Measure E} [SigmaFinite ν]
/-! ### Weighted future pairings — the bilinear rung
The covariance above is the `φ ≡ 1` case of a **weighted** future pairing
`𝔼[φ · Ñ(C) · Ñ(D)] = 𝔼[φ] · ν̂(C ∩ D)` for `φ` adapted at a time `s` before both `C, D`
(`C, D ⊆ (s, ∞) × E`). It is the jump analogue of `ItoIsometryAdapted.rect_increment_pairing`
(Brownian rectangles — cited, not reused): the past-adapted weight factors out of every future
increment, and the increment covariance supplies `ν̂(C ∩ D)`. From it, splitting a box at the later
of two start times gives the **overlapping-box bilinear pairing** the Itô–Lévy isometry sums over.
All of it rests on the *single-increment* field `indep_of_disjoint_region` — no past-vs-future
`σ`-algebra independence is assumed. Integrability of the triple products is bought by the same
independence (a past-adapted factor is `⟂` any future increment it multiplies). -/
Formal statement
theorem _root_.MathFin.integral_bilinear_pairing (N : PoissonRandomMeasure P ν)
{sa ta sb tb : ℝ} (hab : sa ≤ sb)
{Aa Ab : Set E} (hAa : MeasurableSet Aa) (hAb : MeasurableSet Ab)
(hAafin : ν Aa ≠ ⊤) (hAbfin : ν Ab ≠ ⊤)
{φa φb : Ω → ℝ} {Ca Cb : ℝ}
(hφa : N.AdaptedAt sa φa) (hφb : N.AdaptedAt sb φb)
(hφaB : ∀ ω, |φa ω| ≤ Ca) (hφbB : ∀ ω, |φb ω| ≤ Cb) :
∫ ω, (φa ω * N.compensated (Set.Ioc sa ta ×ˢ Aa) ω)
* (φb ω * N.compensated (Set.Ioc sb tb ×ˢ Ab) ω) ∂P
= (∫ ω, φa ω * φb ω ∂P)
* (referenceIntensity ν ((Set.Ioc sa ta ×ˢ Aa) ∩ (Set.Ioc sb tb ×ˢ Ab))).toReal := by sorry
Source