Doob's martingale convergence theorem (almost-sure form)
ProvedFamousTheorems.martingale_convergence_theoremDoob's martingale convergence theorem. Let be a real-valued submartingale on a finite measure space with . Then converges to a finite limit for almost every .
This is one of the fundamental theorems of probability theory. It applies in particular to nonnegative supermartingales and to -bounded martingales. Consequences include Lévy's upward and downward theorems, the strong law of large numbers by martingale methods, and many almost-sure convergence results in analysis.
Formalization note. Mathlib's MeasureTheory.Submartingale.exists_ae_tendsto_of_bdd. The bound is eLpNorm (f n) 1 μ ≤ R for a fixed R : NNReal. The conclusion says that for almost every ω there is a real c with . Mathlib also packages the limit as a measurable process, Filtration.limitProcess.
import Mathlib
namespace FamousTheorems
open MeasureTheory
theorem martingale_convergence_theorem {Ω : Type*} {m0 : MeasurableSpace Ω} {μ : Measure Ω} {ℱ : Filtration ℕ m0} {f : ℕ → Ω → ℝ}
{R : NNReal} [IsFiniteMeasure μ] (hf : Submartingale f ℱ μ) (hbdd : ∀ n, eLpNorm (f n) 1 μ ≤ R) :
∀ᵐ ω ∂μ, ∃ c : ℝ, Filter.Tendsto (fun n => f n ω) Filter.atTop (nhds c) := by sorry
end FamousTheorems