Eq. (10.7)-(10.8): is convex and is optimal iff
ProvedInventoryControl.cs_stage1_optimalWith echelon holding costs , shortage cost and , the stage-1 cost of Eq. (10.6) is a convex function of on , it is differentiable with
and a level minimizes over if and only if
where , and .
Convexity comes from that of the loss function (Eq. 5.41), and the derivative from . The fractile is a newsboy critical fractile with overage cost and underage cost : a unit held at installation 1 rather than at installation 2 costs the value added , and a unit short costs plus the holding cost it would have incurred anyway. When the fractile is , no finite minimizer exists, and the book notes that installation 2 then never carries stock.
Formalization Note The fractile condition is stated as the characterization of the minimizers; existence and uniqueness of the solution for follow from the strict monotonicity of as in the newsboy mission.
import Definitions.Def_InventoryControl_clarkScarf
namespace InventoryControl
theorem cs_stage1_optimal (e1 e2 b1 mu sigma : ℝ) (he1 : 0 ≤ e1) (he2 : 0 ≤ e2) (hb : 0 < b1)
(hs : 0 < sigma) (L1 : ℕ) (S1 : ℝ) :
ConvexOn ℝ Set.univ (csStage1Cost e1 e2 b1 mu sigma L1)
∧ HasDerivAt (csStage1Cost e1 e2 b1 mu sigma L1)
(e1 + (e1 + e2 + b1)
* (ProbabilityTheory.cdf (ProbabilityTheory.gaussianReal 0 1)
((S1 - (L1 + 1) * mu) / (Real.sqrt (L1 + 1) * sigma)) - 1)) S1
∧ ((∀ y : ℝ, csStage1Cost e1 e2 b1 mu sigma L1 S1 ≤ csStage1Cost e1 e2 b1 mu sigma L1 y)
↔ ProbabilityTheory.cdf (ProbabilityTheory.gaussianReal 0 1)
((S1 - (L1 + 1) * mu) / (Real.sqrt (L1 + 1) * sigma))
= (e2 + b1) / (e1 + e2 + b1)) := by sorry
end InventoryControl
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.