Lemma 5 — Gyöngy–Krylov maximal inequality for dominated processes
ProvedSabanisEuler.UniformRate.maximal_inequalityLet be a filtered probability space with right-continuous filtration, and let . Let and be nonnegative, -adapted processes with continuous paths on . Suppose that for every constant and every stopping time ,
Then for every stopping time and every ,
This is a Lenglart-type domination inequality: a bound on expectations at stopping times upgrades to a bound on the expected running maximum, at the price of a fractional power . In the proof of Theorem 3 it turns an Itô-formula estimate for at stopping times into a bound on the expected supremum over .
Formalization Note Nonnegativity is built into the value type ; expectations are Lebesgue integrals in , so both sides may be . Stopping times take values in and are required to satisfy everywhere. Nothing is required of after . The process is not assumed nondecreasing, and the right-hand side is , as printed. The right-continuity of the filtration is the paper's standing usual condition; completeness of the filtration is not imposed.
import Mathlib open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal Topology BigOperators
namespace SabanisEuler.UniformRate
/-- Sabanis (2016), arXiv:1308.1796v4, p. 19, Lemma 5 (the Gyöngy–Krylov / Lenglart-type
maximal inequality). Let `T ∈ [0, ∞)` and let `f = (f_t)_{t ∈ [0, T]}`,
`g = (g_t)_{t ∈ [0, T]}` be nonnegative `ℱ`-adapted processes with continuous paths on
`[0, T]` such that, for every constant `c > 0` and every stopping time `τ ≤ T`,
`𝔼[f_τ 𝟙_{g₀ ≤ c}] ≤ 𝔼[g_τ 𝟙_{g₀ ≤ c}]`. Then for every stopping time `τ ≤ T` and every
`γ ∈ (0, 1)`,
`𝔼[sup_{t ≤ τ} f_t^γ] ≤ (2 - γ)/(1 - γ) · 𝔼[sup_{t ≤ τ} g_t^γ]`.
Nonnegativity is built into the value type `ℝ≥0`; expectations are lower Lebesgue integrals
in `[0, ∞]`. Nothing is required of `f`, `g` after `T`. `g` is not assumed nondecreasing,
and the right-hand side is `sup_{t ≤ τ} g_t^γ`, as printed. -/
theorem maximal_inequality
{Ω : Type*} [mΩ : MeasurableSpace Ω]
(P : Measure Ω) [IsProbabilityMeasure P]
(ℱ : Filtration ℝ≥0 mΩ) [ℱ.IsRightContinuous]
(T : ℝ≥0)
(f g : ℝ≥0 → Ω → ℝ≥0)
(hf_cont : ∀ ω, ContinuousOn (fun t => f t ω) (Set.Icc 0 T))
(hg_cont : ∀ ω, ContinuousOn (fun t => g t ω) (Set.Icc 0 T))
(hf_adapt : ∀ t ≤ T, Measurable[ℱ t] (f t))
(hg_adapt : ∀ t ≤ T, Measurable[ℱ t] (g t))
(hdom : ∀ c : ℝ≥0, 0 < c → ∀ τ : Ω → WithTop ℝ≥0, IsStoppingTime ℱ τ →
(∀ ω, τ ω ≤ (T : WithTop ℝ≥0)) →
∫⁻ ω, {ω | g 0 ω ≤ c}.indicator (fun ω => ((stoppedValue f τ ω : ℝ≥0) : ℝ≥0∞)) ω ∂P
≤ ∫⁻ ω, {ω | g 0 ω ≤ c}.indicator (fun ω => ((stoppedValue g τ ω : ℝ≥0) : ℝ≥0∞)) ω ∂P) :
∀ τ : Ω → WithTop ℝ≥0, IsStoppingTime ℱ τ → (∀ ω, τ ω ≤ (T : WithTop ℝ≥0)) →
∀ γ : ℝ, 0 < γ → γ < 1 →
∫⁻ ω, (⨆ (t : ℝ≥0) (_ : (t : WithTop ℝ≥0) ≤ τ ω), (f t ω : ℝ≥0∞) ^ γ) ∂P
≤ ENNReal.ofReal ((2 - γ) / (1 - γ)) *
∫⁻ ω, (⨆ (t : ℝ≥0) (_ : (t : WithTop ℝ≥0) ≤ τ ω), (g t ω : ℝ≥0∞) ^ γ) ∂P := by sorry
end SabanisEuler.UniformRate
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.