Lemma 2(iii): comparisons at calibrated penalties above mean demand
ProvedLostSalesRQ.endpoint_calibratedapproximation-guaranteeinventorylost-salesoperations-research
Fix . For every integer write , where is the Erlang normalizer. Then
and the sequence () increases to infinity.
Consecutive calibrated penalties bracket intervals on which a single base-stock level satisfies the comparison, covering every penalty with .
Formalization Note The weight is erlangNormalizer μ j , which equals . Strict increase is stated on and divergence along all .
Preamble
import Definitions.Def_LostSalesRQ_lower_curve import Definitions.Def_LostSalesRQ_poisson_process
Formal statement
namespace LostSalesRQ
theorem endpoint_calibrated (μ : ℝ) (hμ : 0 < μ) :
(∀ j : ℕ, μ ≤ (j : ℝ) →
max (baseStockCost μ (calibratedPenalty μ (erlangNormalizer μ j)) j)
(baseStockCost μ (calibratedPenalty μ (erlangNormalizer μ j)) (j + 1)) <
Real.sqrt 2 * rankBound μ (erlangNormalizer μ j) ∧
Real.sqrt 2 * rankBound μ (erlangNormalizer μ j) ≤
Real.sqrt 2 * lowerCurve μ (calibratedPenalty μ (erlangNormalizer μ j))) ∧
StrictMonoOn (fun j : ℕ => calibratedPenalty μ (erlangNormalizer μ j))
{j : ℕ | μ ≤ (j : ℝ)} ∧
Filter.Tendsto (fun j : ℕ => calibratedPenalty μ (erlangNormalizer μ j))
Filter.atTop Filter.atTop := by sorry
end LostSalesRQSource
M. Hu, "A 1.932 Guarantee for (r,Q) Policies in Continuous-Review Lost-Sales Inventory Systems", arXiv:2610.34576, 2026; Lemma 2(iii), eq. (19) (p. 17); proof in Appendix F (pp. A10-A12).
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.