Claim 1 — under any policy in Z(s), almost surely the state lies in X(s) from some period on, so all states outside X(s) are transient
ProvedLostSalesOldNew.StateReduction.claim_1Consider the lost-sales system with lead time and i.i.d. demands with common law on , where demand is not almost surely zero: . Let satisfy , let be a stationary policy, and let be any initial state. Let be the state in period under , started at . Then, almost surely,
The paper states: "For any policy in , all states outside are transient." We state: almost surely the state lies in from some period on, so every state outside (indeed the whole complement of ) is visited only finitely often. This is the reading the paper's proof establishes — the process started outside eventually reaches , and once it enters it never leaves.
Consequently, to solve the dynamic program one may restrict attention to the compact state space and the policies .
Formalization Note. The demand sequence is the coordinate process under the product measure (Measure.infinitePi), which encodes the paper's independent, stationary, nonnegative demands. The hypothesis is added: the paper assumes only nonnegative demands, but with a state with is never left and the claim fails. Policies are deterministic stationary functions of the state; the initial state is any , inside or outside .
import Mathlib import Definitions.Def_LostSalesOldNew_StateReduction_Model
namespace LostSalesOldNew.StateReduction
open MeasureTheory
theorem claim_1 {L : ℕ} (hL : 0 < L) (s : Fin (L + 1) → ℝ) (hs : IsLevelVector s)
(z : (Fin L → ℝ) → ℝ) (hz : z ∈ Zs s)
(D : Measure ℝ) (hD : IsDemand D) (hDpos : 0 < D (Set.Ioi 0))
(x : Fin L → ℝ) (hx : ∀ i, 0 ≤ x i) :
haveI := hD.isProb
∀ᵐ ds ∂(Measure.infinitePi (fun _ : ℕ => D)),
∃ T : ℕ, ∀ t, T ≤ t → traj z x ds t ∈ Xs s := by sorry
end LostSalesOldNew.StateReduction
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.