Proof of Theorem 11, p. 941 ("the key step") — κ̄_t(v, ζ | d) is jointly convex in (v, ζ, d)
ProvedZipkinLostSales.Variability.kappa_jointly_convexFix a lead time , costs and a discount factor . Let , with the convention , and let (standing for the continuation cost ) be convex and nonnegative on . For , and let
which is program (4). Then is jointly convex in on the convex set
The paper calls this "the key step" of the proof of Theorem 11: since is convex in the demand , its expectation increases when demand becomes larger in the convex order.
Formalization Note The continuation is generic; the result is applied with , which is convex on (Theorem 4) and nonnegative. Nonnegativity of keeps the infimum in (4) a genuine infimum rather than Lean's junk value. The domain is : for the constraint set of (4) is empty. The infimum is over , which is the constraint set of (4).
import Mathlib import Definitions.Def_ZipkinLostSales_Variability_Model open MeasureTheory
namespace ZipkinLostSales.Variability
/-- Proof of Theorem 11 (Zipkin 2008, p. 941, "the key step"): if the continuation `F` (standing
for `f̄_{t+1}`) is convex and nonnegative on `V`, then `κ̄(v, ζ | d)` of program (4) is jointly
convex in `(v, ζ, d)` on `{v ∈ ZipkinLostSales.LNatural.V, ζ ≤ 0, d ≥ 0}`. -/
theorem kappa_jointly_convex {L : ℕ} (hL : 0 < L) (hh p γ : ℝ) (hhh : 0 ≤ hh) (hp : 0 ≤ p)
(hγ0 : 0 < γ) (hγ1 : γ ≤ 1) (F : (Fin L → ℝ) → ℝ) (hFconv : ConvexOn ℝ (ZipkinLostSales.LNatural.V L) F)
(hFnonneg : ∀ v ∈ ZipkinLostSales.LNatural.V L, 0 ≤ F v) :
ConvexOn ℝ {w : (Fin L → ℝ) × ℝ × ℝ | w.1 ∈ ZipkinLostSales.LNatural.V L ∧ w.2.1 ≤ 0 ∧ 0 ≤ w.2.2}
(fun w => ZipkinLostSales.LNatural.kappa hh p γ F w.2.2 w.1 w.2.1) := by sorry
end ZipkinLostSales.Variability
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.