Proposition 2, p. 15 — every log-concave demand has weighted mean spread Δ(q*)f(q*) ≥ min(b,h)/(b+h)
ProvedDataDrivenNV.WMS.proposition2_wms_lower_boundLet be the unit underage cost and the unit overage cost of a newsvendor. Let the demand have a log-concave probability density with cdf , let be its quantile, and let
be its absolute mean spread. Then the weighted mean spread satisfies
The weighted mean spread governs the probability that the (biased) sample-average-approximation order quantity is -optimal (Theorem 3 of the paper); this proposition makes that guarantee uniform over all log-concave demand distributions, independently of their parameters.
Formalization Note Log-concavity is the published LogConcaveOn Set.univ f in the power form, which admits densities vanishing outside an interval (uniform, exponential). No integrability of is assumed: a log-concave density has exponential tails, so both conditional means exist. The value is meaningful pointwise: since , lies in the interior of the interval , where a log-concave is continuous and determined by its law. The case and the boundary cases of the paper's proof are covered by the statement, which has no restriction.
import Mathlib import Definitions.Def_LogConcaveOn import Definitions.Def_DataDrivenNV_WMS_Setting
namespace DataDrivenNV.WMS
open MeasureTheory
/-- **Proposition 2** (p. 15): if `D` has a log-concave pdf `f` with `b/(b+h)` quantile `q*`
and AMS `Δ(q*)`, then the weighted mean spread satisfies `Δ(q*) f(q*) ≥ min(b,h)/(b+h)`. -/
theorem proposition2_wms_lower_bound (b h : ℝ) (hb : 0 < b) (hh : 0 < h)
(f : ℝ → ℝ) (hf : IsPdf f) (hlc : ConvexOptimization.LogConcaveOn Set.univ f) :
min b h / (b + h) ≤
ams f (quantileOf f (b / (b + h))) * f (quantileOf f (b / (b + h))) := by sorry
end DataDrivenNV.WMS
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.