Doob's maximal inequality for nonnegative submartingales (cited on p. 359)
ProvedAzumaWeightedSums.IteratedLog.doob_Lp_maximalmartingalemaximal-inequalityp2o-batch-pfp1bp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
Let be a submartingale on a probability space with respect to a filtration , and suppose for all and . Then for every and every real ,
Azuma applies this inequality, citing Doob's Stochastic Processes (1953), p. 317, to the nonnegative submartingale of weighted sums of martingale differences, to compare the moments of the maximal sum with those of . It is the classical form of Doob's maximal inequality.
Formalization Note Both expectations are lower Lebesgue integrals with values in , so the statement is meaningful (and trivially true) when . Powers are real powers of nonnegative reals.
Preamble
import Mathlib open MeasureTheory
Formal statement
namespace AzumaWeightedSums.IteratedLog
/-- Doob's `L^α` maximal inequality for nonnegative submartingales, as cited by Azuma 1967
(proof of Lemma 2, p. 359, from Doob, *Stochastic Processes*, p. 317): for a nonnegative
submartingale `(f_m)`, every `n` and every real `α > 1`,
`E{(max_{m ≤ n} f_m)^α} ≤ (α/(α−1))^α E{f_n^α}`. -/
theorem doob_Lp_maximal {Ω : Type*} {mΩ : MeasurableSpace Ω} {μ : Measure Ω}
[IsProbabilityMeasure μ] {ℱ : Filtration ℕ mΩ} {f : ℕ → Ω → ℝ}
(hf : Submartingale f ℱ μ) (hnonneg : ∀ m ω, 0 ≤ f m ω) (n : ℕ) {α : ℝ} (hα : 1 < α) :
∫⁻ ω, ENNReal.ofReal
(((Finset.range (n + 1)).sup' Finset.nonempty_range_add_one (fun m => f m ω)) ^ α) ∂μ
≤ ENNReal.ofReal ((α / (α - 1)) ^ α) * ∫⁻ ω, ENNReal.ofReal (f n ω ^ α) ∂μ := by sorry
end AzumaWeightedSums.IteratedLog
Source
Azuma, Weighted sums of certain dependent random variables, Tôhoku Math. J. 19 (1967), https://doi.org/10.2748/tmj/1178243286, p. 359, proof of Lemma 2, citing J. L. Doob, Stochastic Processes, Wiley 1953, p. 317
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.