Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3 — G_n and W_n are K-convex and coercive, an (r, S) rule solves the n-stage problem, and W_n, G_n increase in n

Proved
SongZipkinFluct.FixedCost.theorem3

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

inventoryk-convexityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1ss-policy

Consider the fixed-cost model: the standing hypotheses, Assumption 1 (αcˉ<p\alpha\bar c < pαcˉ<p), Kˉ>0\bar K > 0Kˉ>0 (so K=KˉF~L(α)>0K = \bar K\tilde F_L(\alpha) > 0K=KˉF~L​(α)>0), and the terminal cost W0=W∞W_0 = W_\inftyW0​=W∞​ of the linear model. For any fixed iii and any n≥1n \ge 1n≥1:

  1. Gn(i,y)G_n(i, y)Gn​(i,y) is KKK-convex as a function of yyy, and lim⁡∣y∣→∞Gn(i,y)=+∞\lim_{|y|\to\infty} G_n(i, y) = +\inftylim∣y∣→∞​Gn​(i,y)=+∞;
  2. the parameters Sn∗(i)S^*_n(i)Sn∗​(i) = the smallest global minimizer of Gn(i,y)G_n(i, y)Gn​(i,y) and rn∗(i)r^*_n(i)rn∗​(i) = the largest y<Sn∗(i)y < S^*_n(i)y<Sn∗​(i) with Gn(i,y)>K+Gn(i,Sn∗(i))G_n(i, y) > K + G_n(i, S^*_n(i))Gn​(i,y)>K+Gn​(i,Sn∗​(i)) exist, and the (r,S)(r, S)(r,S) rule with these parameters attains the minimum in (10): for every xxx, with y=Sn∗(i)y = S^*_n(i)y=Sn∗​(i) if x≤rn∗(i)x \le r^*_n(i)x≤rn∗​(i) and y=xy = xy=x otherwise,
Wn(i,x)=Kδ(y−x)+Gn(i,y);W_n(i, x) = K\delta(y - x) + G_n(i, y);Wn​(i,x)=Kδ(y−x)+Gn​(i,y);
  1. Wn(i,x)W_n(i, x)Wn​(i,x) is KKK-convex in xxx and lim⁡x→+∞Wn(i,x)=+∞\lim_{x\to+\infty} W_n(i, x) = +\inftylimx→+∞​Wn​(i,x)=+∞;
  2. Wn(i,x)≥Wn−1(i,x)≥0W_n(i, x) \ge W_{n-1}(i, x) \ge 0Wn​(i,x)≥Wn−1​(i,x)≥0;
  3. Gn+1(i,y)≥Gn(i,y)≥0G_{n+1}(i, y) \ge G_n(i, y) \ge 0Gn+1​(i,y)≥Gn​(i,y)≥0.

This is the finite-horizon structure theorem: an (r,S)(r, S)(r,S) policy with world-dependent parameters is optimal for every nnn-stage problem, and the value iterates increase to the infinite-horizon limits.

Formalization Note WnW_nWn​ is by definition the optimal nnn-stage cost (the recursion (10)), so "an optimal policy for the nnn-stage problem is an (r,S)(r, S)(r,S) policy" is stated as: the (r,S)(r, S)(r,S) decision attains the minimum defining Wn(i,x)W_n(i, x)Wn​(i,x) at every xxx. The existence of Sn∗(i)S^*_n(i)Sn∗​(i) and rn∗(i)r^*_n(i)rn∗​(i) is part of the conclusion.

Preamble
import Mathlib
import Definitions.Def_SongZipkinFluct_FixedCost_Bounds

open Filter Topology
Formal statement
namespace SongZipkinFluct.FixedCost

open Model

/-- Theorem 3 (p. 358) of Song and Zipkin (1993). Fixed-cost model of §3.2: `K = K̄F̃_L(α) > 0`
and terminal cost `W₀ = W_∞` of the linear model. For any fixed `i` and any `n ≥ 1`:

* (a) `G_n(i, ·)` is `K`-convex (Definition 2) and `G_n(i, y) → +∞` as `|y| → ∞`;
* (b) the `(r, S)` parameters exist (`S*_n(i)` the smallest global minimizer of `G_n(i, ·)`,
  `r*_n(i)` the largest `y < S*_n(i)` with `G_n(i, y) > K + G_n(i, S*_n(i))`), and the
  `(r, S)` rule with these parameters attains the minimum in (10) at every `x`;
* (c) `W_n(i, ·)` is `K`-convex and `W_n(i, x) → +∞` as `x → +∞`;
* (d) `W_n(i, x) ≥ W_{n−1}(i, x) ≥ 0`;
* (e) `G_{n+1}(i, y) ≥ G_n(i, y) ≥ 0`.

Formalization Note: `W_n` is the optimal `n`-stage cost by its definition through (10)
(p. 356), so "an optimal policy for the `n`-stage problem is an `(r, S)` policy" is encoded as:
the decision `y = S*_n(i)` if `x ≤ r*_n(i)`, `y = x` otherwise, attains the infimum defining
`W_n(i, x)`. Assumption 1 (`αc̄ < p`) is the paper's standing assumption from p. 356 on. -/
theorem theorem3 {I : Type*} [Countable I] [Nonempty I] [DecidableEq I] (M : Model I)
    (hM : M.Standing) (hA1 : M.α * M.cbar < M.p) (hK : 0 < M.Kbar) (i : I) (n : ℕ)
    (hn : 1 ≤ n) :
    (LConvex M.K (Gfix M n i) ∧ Tendsto (Gfix M n i) atTop atTop ∧
        Tendsto (Gfix M n i) atBot atTop) ∧
      ((∃ r S : ℤ, IsRSParams M.K (Gfix M n i) r S) ∧
        ∀ r S : ℤ, IsRSParams M.K (Gfix M n i) r S → ∀ x : ℤ,
          Wfix M n i x = M.K * SongZipkinFluct.Linear.delta ((if x ≤ r then S else x) - x)
            + Gfix M n i (if x ≤ r then S else x)) ∧
      (LConvex M.K (Wfix M n i) ∧ Tendsto (Wfix M n i) atTop atTop) ∧
      (∀ x : ℤ, Wfix M (n - 1) i x ≤ Wfix M n i x ∧ 0 ≤ Wfix M (n - 1) i x) ∧
      (∀ y : ℤ, Gfix M n i y ≤ Gfix M (n + 1) i y ∧ 0 ≤ Gfix M n i y) := by sorry

end SongZipkinFluct.FixedCost
Source
Song and Zipkin, Inventory Control in a Fluctuating Demand Environment, Oper. Res. 41(2):351–370 (1993), DOI 10.1287/opre.41.2.351, p. 358, Theorem 3
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me