§2.1, p. 6 — Π_{1,α} is non-increasing and convex, Π_{1,α}(q) + q = m₁ for q ≤ 0, and Π_{1,α}(q) → 0 as q → ∞
ProvedHeavyTailNV.ZeroOrder.pi_propertiesconvexitydistributionally-robust-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1stop-loss
Let and , let be the set of laws on with mean and -th moment , and let for . Then:
- is non-increasing on ;
- is convex on ;
- for every , ;
These are the general properties of a worst-case stop-loss function noted in §2.1 of the paper for any ambiguity set with a specified mean, here for . Convexity is what lets the proof of Proposition 3.2 pass from the interval to all of .
Formalization Note The paper states these properties for a general ambiguity set with finite mean; this item states them for only, where .
Preamble
import Mathlib import Definitions.Def_HeavyTailNV_ZeroOrder_Model
Formal statement
namespace HeavyTailNV.ZeroOrder
open Filter Topology
/-- The properties of `Π` (arXiv:1806.05379v2, §2.1, p. 6), for the ambiguity set `F_{1,α}`:
`Π_{1,α}` is non-increasing and convex, `Π_{1,α}(q) + q = m₁` for every `q ≤ 0`, and
`Π_{1,α}(q) → 0` as `q → ∞`. -/
theorem pi_properties (α m1 ma : ℝ) (hα : 1 < α) (hm1 : 0 < m1) (hma : m1 ^ α < ma) :
Antitone (HeavyTailNV.Tail.worstCase m1 ma α) ∧
ConvexOn ℝ Set.univ (HeavyTailNV.Tail.worstCase m1 ma α) ∧
(∀ q : ℝ, q ≤ 0 → HeavyTailNV.Tail.worstCase m1 ma α q + q = m1) ∧
Tendsto (HeavyTailNV.Tail.worstCase m1 ma α) atTop (𝓝 0) := by sorry
end HeavyTailNV.ZeroOrder
Source
Das, Dhara and Natarajan, On the heavy-tail behavior of the distributionally robust newsvendor, arXiv:1806.05379v2, §2.1, p. 6, the properties of Π following (2.1)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.