Unit-rate Poisson demand paths, and the calibrated penalty
DefinitionLostSalesRQ_poisson_processapproximation-guaranteeinventorylost-salesoperations-research
A demand path is the sequence of interarrival times of a rate-one Poisson process; under the canonical law the are independent exponential random variables with rate one. The -st arrival occurs at and counts arrivals in . A path is regular when all and .
For and the expected largest excess of the line over demand on is
and for the calibrated penalty (eq. (11)) is
Proposition 2 identifies with , the penalty at which both lower-bound terms coincide.
Formalization Note The expectation is a Bochner integral of a bounded function of the path. Arrival counts use Set.ncard, which is finite on regular paths (a full-measure set).
Definition code
import Mathlib.Probability.ProductMeasure
import Mathlib.Probability.Distributions.Exponential
/-!
# The unit-rate Poisson demand process
A demand path is the sequence of interarrival times of a rate-one Poisson process; under
`interarrivalLaw` they are independent exponential random variables with rate one.
-/
namespace LostSalesRQ
open MeasureTheory ProbabilityTheory Filter Set
/-- A demand path: the sequence `ξ 0, ξ 1, …` of unit-rate interarrival times. -/
abbrev DemandPath := ℕ → ℝ
/-- The law of the interarrival times: independent exponential distributions with rate one. -/
noncomputable def interarrivalLaw : Measure DemandPath :=
Measure.infinitePi (fun _ : ℕ => expMeasure 1)
/-- Epoch `ξ 0 + ⋯ + ξ k` of the `(k+1)`-st arrival of the unit-rate process. -/
def arrivalEpoch (ξ : DemandPath) (k : ℕ) : ℝ :=
∑ i ∈ Finset.range (k + 1), ξ i
/-- Number `N(s)` of unit-rate arrivals in `[0, s]`. -/
noncomputable def unitCount (ξ : DemandPath) (s : ℝ) : ℕ :=
{k : ℕ | arrivalEpoch ξ k ≤ s}.ncard
/-- A demand path is regular when every interarrival time is positive and the arrival epochs
diverge. -/
def RegularPath (ξ : DemandPath) : Prop :=
(∀ k, 0 < ξ k) ∧ Tendsto (arrivalEpoch ξ) atTop atTop
/-- Largest excess `sup_{0 ≤ t ≤ T} {f t - N(t)}` of the line `f t` over the arrival count. -/
noncomputable def maxExcess (T f : ℝ) (ξ : DemandPath) : ℝ :=
sSup ((fun t : ℝ => f * t - (unitCount ξ t : ℝ)) '' Icc 0 T)
/-- `W_T(ℓ) = E sup_{0 ≤ t ≤ T} {(1 - ℓ) t - N(t)}` (Appendix D). -/
noncomputable def expectedMaxExcess (T ℓ : ℝ) : ℝ :=
∫ ξ, maxExcess T (1 - ℓ) ξ ∂interarrivalLaw
/-- Calibrated penalty `α_μ(η) = η [1 + E sup_{0 ≤ t ≤ μ} {(1 - 1/η) t - N(t)}]` (eq. (11)). -/
noncomputable def calibratedPenalty (μ η : ℝ) : ℝ :=
η * (1 + expectedMaxExcess μ (1 / η))
end LostSalesRQ
Source
M. Hu, "A 1.932 Guarantee for (r,Q) Policies in Continuous-Review Lost-Sales Inventory Systems", arXiv:2610.34576, 2026; Section 4.3.2, eq. (11) (p. 13); Appendix D (p. A7).