Gaussian profiles , , and
DefinitionLostSalesRQ_gaussianapproximation-guaranteeinventorylost-salesoperations-research
The standard normal density and distribution function , together with two profiles used in the appendix comparisons:
- the Brownian profile of Appendix D,
- the Gaussian rank profile of Lemma G.1,
is the expected running maximum of a Brownian motion with drift over (Lemma D.1), and is a Gaussian counterpart of the rank bound (Lemma G.1).
Formalization Note is the integral of over against Lebesgue measure. The formula for is also applied for ; the paper uses only . is a real infimum, bounded below by for .
Definition code
import Mathlib.Probability.Distributions.Gaussian.Real
/-!
# Gaussian profiles used in the appendix comparisons
The standard normal density `φ` and distribution function `Φ`, the Brownian profile `w`
(Appendix D) and the Gaussian rank profile `G` (Appendix G).
-/
namespace LostSalesRQ
open MeasureTheory Set
/-- Standard normal density `φ(z) = e^{-z²/2} / √(2π)`. -/
noncomputable def stdNormalPdf (z : ℝ) : ℝ :=
Real.exp (-(z ^ 2) / 2) / Real.sqrt (2 * Real.pi)
/-- Standard normal distribution function `Φ(z) = ∫_{-∞}^{z} φ(x) dx`. -/
noncomputable def stdNormalCdf (z : ℝ) : ℝ :=
∫ x in Iic z, stdNormalPdf x
/-- Brownian profile `w(z) = φ(z) - z (1 - Φ(z)) + (Φ(z) - 1/2) / z` for `z ≠ 0`,
with `w(0) = √(2/π)` (Appendix D). -/
noncomputable def brownianProfile (z : ℝ) : ℝ :=
if z = 0 then Real.sqrt (2 / Real.pi)
else stdNormalPdf z - z * (1 - stdNormalCdf z) + (stdNormalCdf z - 1 / 2) / z
/-- Gaussian rank profile `G(s) = inf_{z ∈ ℝ} {φ(z) + z Φ(z) + s (1 - Φ(z))}` (Lemma G.1). -/
noncomputable def gaussianRankProfile (s : ℝ) : ℝ :=
⨅ z : ℝ, (stdNormalPdf z + z * stdNormalCdf z + s * (1 - stdNormalCdf z))
end LostSalesRQ
Source
M. Hu, "A 1.932 Guarantee for (r,Q) Policies in Continuous-Review Lost-Sales Inventory Systems", arXiv:2610.34576, 2026; Appendix D (p. A7); Lemma G.1 (p. A12).