Lemma 2, p. 15 — a log-concave pdf lies below its exponential envelope γ₀e^{γ₁(x−t)}
ProvedDataDrivenNV.WMS.lemma2_exponential_envelopedensitylog-concavep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a log-concave probability density on and let be a point with (so is in the support of ). Suppose , i.e. whenever . Then for every ,
Fixing a supergradient of bounds how fast the density can grow or decay; this envelope is what makes the truncated exponential (12) dominate every density of the class .
Formalization Note " in its support" is expressed by . The bound is claimed at every , including the points where .
Preamble
import Mathlib import Definitions.Def_LogConcaveOn import Definitions.Def_DataDrivenNV_WMS_Setting
Formal statement
namespace DataDrivenNV.WMS
open MeasureTheory
/-- **Lemma 2** (p. 15): if `f` is a log-concave pdf, `f(t) = γ₀ > 0` and `γ₁ ∈ ∂ log f(t)`, then
`f(x) ≤ γ₀ e^{γ₁ (x − t)}` for every `x`. -/
theorem lemma2_exponential_envelope (f : ℝ → ℝ) (hf : IsPdf f)
(hlc : ConvexOptimization.LogConcaveOn Set.univ f)
(t γ₀ γ₁ : ℝ) (hγ₀ : 0 < γ₀) (hft : f t = γ₀) (hsg : IsLogSupergradient f t γ₁) :
∀ x, f x ≤ γ₀ * Real.exp (γ₁ * (x - t)) := by sorry
end DataDrivenNV.WMS
Source
Levi, Perakis & Uichanco, The Data-Driven Newsvendor Problem: New Bounds and Insights, authors' accepted manuscript (MIT DSpace), p. 15, Lemma 2; proof EC.4, p. ec9
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.