Theorem 6.4 (Shang and Song): and , for nonnegative demands
ProvedSupplyChainTheory.shang_song_bounds_v2Theorem 6.4 (Shang and Song 2003). In the serial system of §6.2 with echelon holding costs , stockout cost and nonnegative lead-time demands of finite mean, let be built by the sequential minimization of Theorem 6.3 and let and be the truncated cost (6.31) with every local holding cost replaced by (lower) or by (upper): the single-stage newsvendor cost with demand and stockout cost , plus the holding cost of the pipeline stock at the replaced rate. Then for any and :
(a) ;
(b) , where, as the book says on p. 200, minimizes and minimizes : the cheaper holding rate of the lower bound has the larger fractile, so it is the upper base-stock bound.
The bounds come from the truncated system of (6.31): with all local holding costs equal, all inventory is held at stage 1 and the -stage system collapses to a single stage with lead time . Their average (6.32) is the Shang–Song heuristic, accurate to a fraction of a percent on the authors' test instances.
Formalization Note. The retired version let the lead-time demand laws take negative values; the inventory models of the book assume nonnegative demands (the newsvendor and echelon-cost derivations and the pipeline-stock interpretation presuppose ), and with a negative demand the pipeline-stock term of turns negative while the echelon accounting of does not, so (a) fails (accepted disproof with ). The new statement adds the standing convention for every and is otherwise unchanged. The bounding functions keep the pipeline holding cost that (6.31) charges on stock in transit; the book drops it when it writes the minimizers, since it does not affect them, but part (a) compares values and fails without it (in Example 6.1, while the single-stage cost alone is ). Part (b) asserts the existence of a minimizer of and a minimizer of bracketing ; when the fractiles of are unique these are the book's values. In Example 6.2 at : . Pairing the minimizers the other way would assert .
import Definitions.Def_SupplyChainTheory_multiechelon
namespace SupplyChainTheory
/-- Snyder & Shen, *Fundamentals of Supply Chain Theory*, 2nd ed., p. 201, Theorem 6.4
(Shang & Song 2003): in the serial system of §6.2 with echelon holding costs `h_j > 0`, stockout
cost `p > 0` and **nonnegative** lead-time demands `D_j` of finite mean, with `S*` the
sequentially optimal vector of Theorem 6.3 and `gˡ_j`, `gᵘ_j` the newsvendor-type bounding
functions of p. 200: (a) `gˡ_j(y) ≤ g_j(y | S*) ≤ gᵘ_j(y)` for every `y`, and (b) a minimizer
`Sˡ_j` of `gᵘ_j` and a minimizer `Sᵘ_j` of `gˡ_j` bracket `S*_j`.
Corrected version: the demands are nonnegative random variables (`D_j([0, ∞)^c) = 0`), the
standing convention of the inventory models of the book; the retired statement allowed demand
laws with negative values, under which the pipeline-stock term of `gᵘ_j` turns negative and (a)
fails. -/
theorem shang_song_bounds_v2 (N : ℕ) (h : ℕ → ℝ) (p : ℝ) (D : ℕ → MeasureTheory.Measure ℝ)
[∀ j, MeasureTheory.IsProbabilityMeasure (D j)]
(hD : ∀ j, MeasureTheory.Integrable (fun x => x) (D j))
(hD0 : ∀ j, D j (Set.Iio 0) = 0)
(hh : ∀ j, 0 < h j) (hp : 0 < p) (Sstar : ℕ → ℝ) (hS : CSSequential N h p D Sstar)
(j : ℕ) (hj1 : 1 ≤ j) (hjN : j ≤ N) :
(∀ y, ssLower N h p D j y ≤ csG N h p D Sstar j y
∧ csG N h p D Sstar j y ≤ ssUpper N h p D j y)
∧ ∃ Sl Su : ℝ, IsMinOn (ssUpper N h p D j) Set.univ Sl
∧ IsMinOn (ssLower N h p D j) Set.univ Su ∧ Sl ≤ Sstar j ∧ Sstar j ≤ Su := by sorry
end SupplyChainTheory