Documentation

Atlas.HighDimensionalStatistics.code.Chapter4.ExpCauchySchwarz

theorem exp_cauchy_schwarz_bound {Ω : Type u_1} [MeasurableSpace Ω] (μ : MeasureTheory.Measure Ω) [MeasureTheory.IsProbabilityMeasure μ] (A B : Ω) (s : ) (hA_int : MeasureTheory.Integrable (fun (ω : Ω) => Real.exp (s * A ω)) μ) (hB_int : MeasureTheory.Integrable (fun (ω : Ω) => Real.exp (-s * B ω)) μ) :
(ω : Ω), Real.exp (s / 2 * (A ω - B ω)) μ (( (ω : Ω), Real.exp (s * A ω) μ) * (ω : Ω), Real.exp (-s * B ω) μ)

Cauchy-Schwarz inequality for exponential expectations.

For random variables A, B and any real s, we have 𝔼[exp(s/2 · (A - B))] ≤ √(𝔼[exp(s·A)] · 𝔼[exp(-s·B)])

This follows from Hölder's inequality with p = q = 2 applied to f = exp(s/2 · A) and g = exp(-s/2 · B), using the identities f² = exp(s · A) and g² = exp(-s · B).