Theorem 3.1(c)–(d) — additive demand: and are -concave, and an policy is optimal
ProvedChenSimchiLevi.Additive.theorem_3_1Consider the finite-horizon joint pricing and inventory model of Chen and Simchi-Levi (2004) with fixed ordering cost , under Assumptions 1–5 (with , , ) and additive demand . Let be the profit-to-go (2) and the one-period objective (3). Then for every period :
(c) For every , attains its maximum over at some , and both
are -concave.
(d) There exist such that the following policy is optimal in period : if the initial inventory satisfies , order up to and set the expected demand to ; otherwise order nothing and set the expected demand to . Precisely, writing if and otherwise, for every :
- and maximizes over ;
- ;
- some expected demand maximizes ; the corresponding price is .
This is the paper's main result: with additive demand, the policy conjectured by Thomas (1974) is optimal, and the price depends on the inventory level after ordering.
Formalization Note. -concavity of is the platform's BertsekasKConvex k (−f) (Definition 2.1). The values and are defined as suprema (real sSup); the attainment statements in (c) and (d) show that these suprema are maxima. The model is in expected-demand space; the price is . The hypotheses , and are added to the paper's Assumptions 1–5 (see the model definition).
import Mathlib import Definitions.Def_BertsekasKConvex import Definitions.Def_ChenSimchiLevi_Additive_Model
namespace ChenSimchiLevi.Additive
/-- Theorem 3.1(c)–(d) of Chen–Simchi-Levi (2004), p. 890: under additive demand, for every period
`t = T, …, 1`,
(c) `g_t(·, d)` attains its maximum over `[d_t, d̄_t]` at every `y`, and `g_t(y, d_t(y))` and
`v_t(x)` are `k`-concave;
(d) there are `s_t ≤ S_t` such that the `(s, S)` order — order up to `S_t` if `x < s_t`, otherwise
do not order — attains the maximum in (2), `v_t(x)` equals `c_t x` plus that maximum, and the
expected demand is set to a maximizer of `g_t(y, ·)` at the post-order level `y`. -/
theorem theorem_3_1 (M : Model) (hA : M.Assumptions) (hadd : M.IsAdditive) :
∀ t ∈ Finset.Icc 1 M.T,
((∀ y : ℝ, ∃ d ∈ Set.Icc (M.dlo t) (M.dhi t),
IsMaxOn (M.g t y) (Set.Icc (M.dlo t) (M.dhi t)) d) ∧
BertsekasKConvex M.k (fun y => -M.Gstar t y) ∧
BertsekasKConvex M.k (fun x => -M.v t x)) ∧
(∃ s S : ℝ, s ≤ S ∧ ∀ x : ℝ,
x ≤ sSOrder s S x ∧
(∀ y : ℝ, x ≤ y → M.orderObj t x y ≤ M.orderObj t x (sSOrder s S x)) ∧
M.v t x = M.c t * x + M.orderObj t x (sSOrder s S x) ∧
∃ d ∈ Set.Icc (M.dlo t) (M.dhi t),
IsMaxOn (M.g t (sSOrder s S x)) (Set.Icc (M.dlo t) (M.dhi t)) d) := by sorry
end ChenSimchiLevi.Additive
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.