Lemma 2 (first claim), p. 11 — OPT(L) ≥ C_L(r_L)
OpenLostSalesExpGap.ConstOrder.opt_ge_CLConsider the periodic-review lost-sales inventory model with i.i.d. nonnegative demands (finite positive mean, not a point mass), lead time , holding cost and lost-sales penalty , started empty. Let be the infimum of the long-run average cost over all admissible policies. For put
and let be the infimum of the set of minimisers of over . Then
This is the first assertion of Lemma 2 of Xin and Goldberg, the lower bound on the optimal cost that, combined with formula (3) for the constant-order cost, yields the decomposition (5) of the optimality gap. In the paper it follows from Theorem 2 (the stationary inventory and pipeline of an optimal policy) together with the convexity/Jensen argument of Observation 1.
Formalization Note. The model is the published CappedBaseStock model (zero-based time, empty initial state, costs in ); ranges over all measurable history-dependent randomised policies, a class containing the paper's Markov class, so the bound is at least as strong as printed. and are CL and rL of the published LostSalesExpGap_ConstOrder_ConstantOrder definitions, and the real number is embedded in via ENNReal.ofReal.
import Mathlib import Definitions.Def_LostSalesExpGap_ConstOrder_ConstantOrder open MeasureTheory ProbabilityTheory open scoped ENNReal NNReal BigOperators
namespace LostSalesExpGap.ConstOrder
/-- Lemma 2, first claim, p. 11: `OPT(L) ≥ C_L(r_L)`, where `C_L(r) = h E[I^r_L] + p E[D] − p r`
and `r_L` is the least minimiser of `C_L` over `[0, E[D]]`. `OPT(L)` is the infimum of the long-run
average cost over all admissible policies, in `[0, ∞]`. -/
theorem opt_ge_CL (P : CappedBaseStock.DemandLaw) (c : CappedBaseStock.Parameters)
(hP : LostSalesConstantOrder.LargeLeadTime.NonDeterministic P) :
ENNReal.ofReal (CL P c (rL P c c.L) c.L) ≤ CappedBaseStock.OPT P c := by sorry
end LostSalesExpGap.ConstOrder