Lemma 3.3, p. 10 — E[max_{i≤n}|X_i|] ≤ 2√(2 log(2n))·σ for centered Gaussians with variances ≤ σ²
ProvedFBMStrongSim.RecordBreak.lemma_3_3expected-supremumgaussian-maximumgaussian-processp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let , , be (not necessarily independent) centered Gaussian random variables whose variances are uniformly bounded by , . Then
with the natural logarithm.
Combined with Lemma 3.2 this gives the bound of (3.7), which removes the expectation from the Borell–TIS bound.
Formalization Note Only each is assumed Gaussian (no joint law), as in the lemma. The constant is the printed one.
Preamble
import Mathlib open MeasureTheory ProbabilityTheory open scoped ENNReal
Formal statement
namespace FBMStrongSim.RecordBreak
/-- Lemma 3.3, p. 10: if `X_1, …, X_n` (`n ≥ 1`, not necessarily independent) are centered
Gaussian random variables with variances at most `σ²`, then
`E[max_{1 ≤ i ≤ n} |X_i|] ≤ 2 √(2 log(2n)) · σ` (natural logarithm). -/
theorem lemma_3_3 {Ω : Type*} [MeasurableSpace Ω] {P : Measure Ω} [IsProbabilityMeasure P]
{n : ℕ} (hn : 0 < n) (X : Fin n → Ω → ℝ) (hX : ∀ i, HasGaussianLaw (X i) P)
(hcent : ∀ i, P[X i] = 0) {σ : ℝ} (hσ : 0 ≤ σ) (hvar : ∀ i, Var[X i; P] ≤ σ ^ 2) :
P[fun ω => Finset.univ.sup' (Finset.univ_nonempty_iff.mpr ⟨⟨0, hn⟩⟩) fun i => |X i ω|]
≤ 2 * Real.sqrt (2 * Real.log (2 * n)) * σ := by sorry
end FBMStrongSim.RecordBreak
Source
Chen, Dong & Ni, ε-Strong simulation of fractional Brownian motion and related stochastic differential equations, arXiv:1902.07824v1, p. 10, Lemma 3.3 (proof: Appendix A.2, p. 39)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.