The best constant order is a minimizer of and
OpenLostSalesConstantOrder.LargeLeadTime.zStar_specConsider the lost-sales model with holding cost , lost-demand penalty , and i.i.d. nonnegative demands with finite third moment and a non-degenerate law. For let and
(this is zObjective, the paper's objective shifted by the constant ). Let be the infimum of the set of minimizers of over (zStar). Then is itself a minimizer of , and .
This is the implicit well-posedness claim behind the paper's definition (p. 5) and its use of the stationary policy in Theorem 1. Why it holds: . is lower semicontinuous (the supremum is a supremum of continuous functions of ; apply Fatou). for , because the walk then has nonnegative drift and is non-degenerate, so its supremum is a.s. infinite. Hence the minimizers form a nonempty closed subset of , and that subset contains its infimum.
import Definitions.Def_LostSalesConstantOrder_LargeLeadTime_ConstantOrder open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal BigOperators
namespace LostSalesConstantOrder.LargeLeadTime
/-- §3.1, p. 5: the best constant order quantity `z = argmin_{v ≥ 0} (h E[I^v_∞] − c v)` is
well defined: the infimum `zStar` of the minimizers of `zObjective` is itself a minimizer, and it
lies strictly below the mean demand `E[D]` (so `π_z` is a stable constant-order policy). -/
theorem zStar_spec (P : CappedBaseStock.DemandLaw) (prm : CappedBaseStock.Parameters)
(h3 : FiniteThirdMoment P) (hnd : NonDeterministic P) :
(zStar P prm : ℝ) < CappedBaseStock.mean P ∧
∀ w : ℝ≥0, zObjective P prm (zStar P prm) ≤ zObjective P prm w := by sorry
end LostSalesConstantOrder.LargeLeadTime