The Doob decomposition theorem
ProvedFamousTheorems.doob_decomposition_6bThe Doob decomposition theorem. Let be an adapted sequence of integrable random variables, with values in a real Banach space, on a filtered probability space . Then , where is a martingale and is predictable, meaning that is -measurable, with . Explicitly,
The decomposition separates a process into a "fair game" part and a predictable drift. It is the discrete-time forerunner of the Doob–Meyer decomposition, and it is how martingale methods are applied to submartingales. A submartingale corresponds to an increasing . The decomposition is also unique almost surely.
Formalization note. Mathlib's MeasureTheory.martingale_martingalePart, together with the decomposition martingalePart_add_predictablePart, stronglyAdapted_predictablePart and predictablePart_zero. The measure is assumed -finite on each (SigmaFiniteFiltration), which holds for probability measures. Predictability is stated as adaptedness of .
import Mathlib
namespace FamousTheorems
open MeasureTheory
theorem doob_decomposition_6b {Ω E : Type*} {m0 : MeasurableSpace Ω} {μ : Measure Ω} [NormedAddCommGroup E] [NormedSpace ℝ E]
[CompleteSpace E] {ℱ : Filtration ℕ m0} [SigmaFiniteFiltration μ ℱ] {f : ℕ → Ω → E}
(hf : StronglyAdapted ℱ f) (hfint : ∀ n, Integrable (f n) μ) :
∃ M A : ℕ → Ω → E, Martingale M ℱ μ ∧ StronglyAdapted ℱ (fun n => A (n + 1)) ∧ A 0 = 0 ∧ f = M + A := by sorry
end FamousTheorems