The continuous review policy under normally distributed lead-time demand: inventory level, service levels and cost rate
DefinitionInventoryControl_rqThe steady-state model of a continuous review policy facing continuous, normally distributed demand, in the notation of Axsäter, Inventory Control, Chapter 5.
An order for units is placed whenever the inventory position (stock on hand plus
outstanding orders minus backorders) reaches the reorder point , and it arrives after a
constant lead-time. The book's standing approximation for continuous demand (Sect. 5.3.1,
p. 75) is that in steady state the inventory position is uniformly distributed on the interval
; this is rqPosition R Q, the uniform probability measure on that interval. The
lead-time demand is normal with mean and standard deviation and is
independent of the inventory position; it is the same law newsboyDemand m s as in the newsboy
model, with and .
The fundamental relationship of Sect. 5.3.2, Eq. (5.35), is
the inventory level a lead-time later equals the inventory position now minus the demand in
between. Accordingly rqLevel R Q m s is the law of , the image of the product of the
two laws above under subtraction. From it,
rqCDF R Q m s xis the distribution function of Eq. (5.39);rqReadyRate R Q m sis the ready rate of Eq. (5.50), which for continuous demand coincides with the fill rate (Sect. 5.7.2);rqBatchBackorders R Q m sis , the expected backordered quantity covered by one batch, defined in Sect. 5.8 by the three cases for , for and for , where is the lead-time demand; that is ;rqCost h b1 R Q m sis the expected cost rate of Eq. (5.56), , with holding cost per unit and time unit and backorder cost per unit and time unit;normalLossH xis the second loss function of Eq. (5.64), built on the loss function of Eq. (5.40) published with the newsboy model.
Two lemmas record that rqPosition R Q and rqLevel R Q m s are probability measures whenever
.
Formalization Note rqPosition R Q is Lebesgue measure conditioned on , which
is the zero measure when ; rqLevel is the pushforward of the product measure under
. Every statement assumes and . The mean is an
arbitrary real: the book has , but none of the formulas depends on its sign.
Costs and are unconstrained in the definitions; positivity is a hypothesis of each
theorem. The integrals are Bochner integrals, which return for a non-integrable function;
integrability of the cost integrand is therefore a theorem of this mission rather than an
assumption.
import Definitions.Def_InventoryControl_newsboy
open MeasureTheory ProbabilityTheory
namespace InventoryControl
/-- The steady-state inventory position of a continuous review `(R, Q)` policy under continuous
demand: uniform on `[R, R + Q]`. Axsäter, *Inventory Control*, Sect. 5.3.1 p. 75. -/
noncomputable def rqPosition (R Q : ℝ) : Measure ℝ := volume[|Set.Icc R (R + Q)]
/-- The law of the inventory level `IL = IP - D(L)`, Eq. (5.35): the inventory position is
uniform on `[R, R + Q]` and the lead-time demand is normal with mean `m` and standard deviation
`s`, the two being independent. Axsäter, *Inventory Control*, Sects. 5.3.2 and 5.3.4. -/
noncomputable def rqLevel (R Q m s : ℝ) : Measure ℝ :=
Measure.map (fun p : ℝ × ℝ => p.1 - p.2) ((rqPosition R Q).prod (newsboyDemand m s))
/-- The distribution function `F(x) = P(IL ≤ x)` of the inventory level, Eq. (5.39). -/
noncomputable def rqCDF (R Q m s x : ℝ) : ℝ := (rqLevel R Q m s).real (Set.Iic x)
/-- The ready rate `S₃ = P(IL > 0)`, Eq. (5.50); equal to the fill rate `S₂` for continuous
demand, Sect. 5.7.2. -/
noncomputable def rqReadyRate (R Q m s : ℝ) : ℝ := (rqLevel R Q m s).real (Set.Ioi 0)
/-- `E(B)`: the expected backordered quantity covered by one batch, Sect. 5.8 p. 83. With
lead-time demand `u`, `B = 0` if `u ≤ R`, `B = u - R` if `R < u ≤ R + Q`, and `B = Q` if
`u > R + Q`. -/
noncomputable def rqBatchBackorders (R Q m s : ℝ) : ℝ :=
∫ u, min (max (u - R) 0) Q ∂(newsboyDemand m s)
/-- The second loss function `H(x) = ∫_x^∞ G(v) dv`, Eq. (5.64). -/
noncomputable def normalLossH (x : ℝ) : ℝ := ∫ v in Set.Ioi x, normalLoss v
/-- The expected holding-plus-backorder cost rate `h E(IL)⁺ + b₁ E(IL)⁻`, Eq. (5.56) and
(5.62). -/
noncomputable def rqCost (h b1 R Q m s : ℝ) : ℝ :=
∫ x, (h * max x 0 + b1 * max (-x) 0) ∂(rqLevel R Q m s)
lemma isProbabilityMeasure_rqPosition (R Q : ℝ) (hQ : 0 < Q) :
IsProbabilityMeasure (rqPosition R Q) := by
unfold rqPosition
refine cond_isProbabilityMeasure_of_finite ?_ ?_ <;>
simp [Real.volume_Icc, hQ]
lemma isProbabilityMeasure_rqLevel (R Q m s : ℝ) (hQ : 0 < Q) :
IsProbabilityMeasure (rqLevel R Q m s) := by
have := isProbabilityMeasure_rqPosition R Q hQ
unfold rqLevel
exact Measure.isProbabilityMeasure_map (by fun_prop)
end InventoryControl