The two-level serial system of the Clark-Scarf model: period costs, their reallocation, and the total cost of Sect. 10.1.1
DefinitionInventoryControl_clarkScarfThe two-echelon serial system of Axsäter, Inventory Control, Sect. 10.1.1, in the
infinite-horizon periodic-review form of Federgruen and Zipkin. Installation 1 faces the
customer demand and replenishes from installation 2, which replenishes from an outside
supplier with infinite supply. Period demands are independent and normally distributed with
mean and standard deviation , so the demand over periods, , is normal
with mean and standard deviation : this is csDemand mu sigma n, the
same Gaussian law as the newsboy model's newsboyDemand. Lead-times are integral
numbers of periods; unmet demand is backordered; there are echelon holding costs
per unit and period, so that the installation holding costs are and
, and a shortage cost per unit and period; there are no ordering costs.
In an arbitrary period , is the echelon inventory position of installation 2 after ordering, and is the realized echelon inventory position of installation 1 after its order in period , which cannot exceed the echelon stock then available (Eq. 10.1). Writing and ,
csPeriodCost e1 e2 b1 mu sigma L1 L2 y2 y1is the sum of the period costs (10.2) and (10.3), ;csStage2Cost e2 mu L2 y2is the reallocated cost of Eq. (10.4);csStage1Cost e1 e2 b1 mu sigma L1 y1is the reallocated cost of Eq. (10.5), which is also the function of Eq. (10.6) when is free;csTotalCost e1 e2 b1 mu sigma L1 L2 S1 y2is the total expected cost of Eq. (10.9) when installation 1 uses the echelon order-up-to level :
the integral being taken against the law of .
Formalization Note The costs are parametrized by the echelon costs , and , are written out, so the relation between the two cost systems is built in rather than assumed. All expectations are Bochner integrals against the Gaussian laws; integrability holds because the integrands grow at most linearly, and the arbitrary allocation rules of the main theorem carry an explicit integrability hypothesis. is allowed, in which case is the point mass at . The order-up-to level is a parameter of ; the theorems fix it by the fractile condition (10.8).
import Definitions.Def_InventoryControl_newsboy
open MeasureTheory ProbabilityTheory
namespace InventoryControl
/-- `D(n)`: the demand over `n` periods, normal with mean `n μ` and standard deviation `√n σ`.
Axsäter, *Inventory Control*, Sect. 10.1.1 p. 194. -/
noncomputable def csDemand (mu sigma : ℝ) (n : ℕ) : Measure ℝ :=
newsboyDemand (n * mu) (Real.sqrt n * sigma)
/-- Eq. (10.4): the reallocated cost at installation 2, `h₂ (y₂ − μ'₂)` with `h₂ = e₂`. -/
noncomputable def csStage2Cost (e2 mu : ℝ) (L2 : ℕ) (y2 : ℝ) : ℝ := e2 * (y2 - L2 * mu)
/-- Eq. (10.5) and (10.6): the reallocated cost at installation 1,
`e₁ y₁ − h₁ μ''₁ + (h₁ + b₁) E((y₁ − D(L₁+1))⁻)` with `h₁ = e₁ + e₂`. -/
noncomputable def csStage1Cost (e1 e2 b1 mu sigma : ℝ) (L1 : ℕ) (y1 : ℝ) : ℝ :=
e1 * y1 - (e1 + e2) * ((L1 + 1) * mu)
+ (e1 + e2 + b1) * ∫ x, max (x - y1) 0 ∂(csDemand mu sigma (L1 + 1))
/-- Eq. (10.2) + (10.3): the period costs before reallocation,
`h₂ E(y₂ − D(L₂) − y₁) + h₁ E((y₁ − D(L₁+1))⁺) + b₁ E((y₁ − D(L₁+1))⁻)`. -/
noncomputable def csPeriodCost (e1 e2 b1 mu sigma : ℝ) (L1 L2 : ℕ) (y2 y1 : ℝ) : ℝ :=
e2 * ((y2 - L2 * mu) - y1)
+ (e1 + e2) * ∫ x, max (y1 - x) 0 ∂(csDemand mu sigma (L1 + 1))
+ b1 * ∫ x, max (x - y1) 0 ∂(csDemand mu sigma (L1 + 1))
/-- Eq. (10.9): the total expected cost `Ĉ₂(y₂)` when installation 1 orders up to `S₁`. -/
noncomputable def csTotalCost (e1 e2 b1 mu sigma : ℝ) (L1 L2 : ℕ) (S1 y2 : ℝ) : ℝ :=
csStage2Cost e2 mu L2 y2 + csStage1Cost e1 e2 b1 mu sigma L1 S1
+ ∫ u in Set.Ioi (y2 - S1),
(csStage1Cost e1 e2 b1 mu sigma L1 (y2 - u) - csStage1Cost e1 e2 b1 mu sigma L1 S1)
∂(csDemand mu sigma L2)
end InventoryControl