Proposition 6.1: an policy is optimal when ordering in batches
ProvedInventoryControl.rq_policy_optimalProposition 6.1 of Axsäter, Inventory Control: an policy is optimal.
The setting is Sect. 6.2.1. Demand is compound Poisson: customers arrive at rate and each demands an integral number of units, i.i.d., not all demands being multiples of some integer larger than one. The lead-time is constant, and , the demand over a lead-time, has the law . A holding cost and a shortage cost per unit and time unit are charged; there are no ordering costs, but all orders must be multiples of a given batch quantity and can only be triggered by customer demands. For an inventory position the expected holding-plus-shortage cost rate a lead-time later is , so the cost rate at time is , where is the position at time . Let and let be an integer minimizing .
The claim has two parts, both almost sure:
- for every ordering rule and every initial position ,
- the policy with this attains it:
Together: no policy that orders in batches of can do better in the long run than the policy whose reorder point minimizes . The book notes that for this is the optimality of an policy when there are no ordering costs, and that for continuous or Poisson demand it also gives the optimality of policies, which coincide with policies there.
Formalization Note A policy is any function assigning to each demand epoch and outcome a number of batches; no measurability is required, since the lower bound is pathwise. The average is over with the cost rate evaluated at the position a lead-time earlier, the position being before time ; "" is stated as "for every , eventually ". The initial position is arbitrary, as in the book, and the reorder point is any minimizer of , whose existence follows from as (a theorem of the previous mission).
import Definitions.Def_InventoryControl_rqPolicy
namespace InventoryControl
theorem rq_policy_optimal {Ω : Type*} [MeasurableSpace Ω] {P : MeasureTheory.Measure Ω}
[MeasureTheory.IsProbabilityMeasure P] (X : CompoundPoissonDemand P) (D : DiscreteDemand)
(h b1 L : ℝ) (hh : 0 < h) (hb : 0 < b1) (hL : 0 ≤ L)
(hD : ∀ j : ℕ, D.p j = P.real {ω | X.cumDemand (X.count L ω) ω = j})
(Q : ℕ) (hQ : 0 < Q) (R : ℤ)
(hR : ∀ y : ℤ, windowCost D h b1 Q R ≤ windowCost D h b1 Q y) (y0 : ℤ) :
(∀ m : ℕ → Ω → ℕ, ∀ᵐ ω ∂P, ∀ ε : ℝ, 0 < ε →
∀ᶠ T in Filter.atTop, windowCost D h b1 Q R / Q - ε ≤ avgCost X D h b1 L Q y0 m T ω)
∧ (∀ᵐ ω ∂P, Filter.Tendsto (fun T => avgCost X D h b1 L Q y0 (rqOrders X R Q y0) T ω)
Filter.atTop (nhds (windowCost D h b1 Q R / Q))) := by sorry
end InventoryControl
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.