Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2 — no order for a product stocked at or above its base-stock level

Disproved
BassokSubstitution.no_order_above_base_stock

by mikedeng1 · Sep 27, 2026 · Mathlib 0df444a (Lean v4.33.1)

base-stockinventoryoptimal-policyp2o-batch-p100bp2o-gran-per-chapterp2o-plan-paperp2o-v1substitution

Consider the single-period NNN-product model with downward substitution, satisfying Assumptions 1–3, with substitution cost b≥0b \ge 0b≥0 and unit costs si<ci<pi+πis_i < c_i < p_i + \pi_isi​<ci​<pi​+πi​, and with independent nonnegative demands that have densities and finite means and whose laws charge every nonempty open interval of [0,∞)[0,\infty)[0,∞).

Let y∗≥0y^* \ge 0y∗≥0 be a base-stock vector: a maximizer of P(0,⋅)P(0, \cdot)P(0,⋅) over the nonnegative orthant. Let x≥0x \ge 0x≥0 be any starting inventory and yˉ\bar yyˉ​ any optimal inventory level after ordering, i.e. a maximizer of P(x,⋅)P(x,\cdot)P(x,⋅) over {y:y≥x}\{y : y \ge x\}{y:y≥x}. Then for every product iii,

xi≥yi∗  ⟹  yˉi=xi.x_i \ge y^*_i \implies \bar y_i = x_i.xi​≥yi∗​⟹yˉ​i​=xi​.

No order is placed for a product whose starting inventory is at least its base-stock level, whatever the starting inventory of the other products. With Theorem 1 (order up to y∗y^*y∗ when x≤y∗x \le y^*x≤y∗) this describes the optimal ordering policy of the substitution model.

Formalization Note. y∗y^*y∗ is quantified over all maximizers of P(0,⋅)P(0,\cdot)P(0,⋅) on {y≥0}\{y \ge 0\}{y≥0}, which are exactly the vectors y∗y^*y∗ of Theorem 1; the conclusion is claimed for every optimal yˉ\bar yyˉ​. The paper uses without stating: independence of the demand classes, densities and finite means, si<ci<pi+πis_i < c_i < p_i + \pi_isi​<ci​<pi​+πi​, and a unique optimal level yˉ(x)\bar y(x)yˉ​(x); the full-support hypothesis stands in for the latter (without it, a demand density vanishing around the critical fractile gives a flat-topped profit and the universal statement fails already for N=1N = 1N=1).

Preamble
import Mathlib
import Definitions.Def_BassokSubstitution_Profit

open MeasureTheory
Formal statement
namespace BassokSubstitution

/-- Theorem 2 (goal): let `y*` be a base-stock vector, i.e. a maximizer of `P(0, ·)` over the
nonnegative orthant. For every starting inventory `x ≥ 0`, every optimal inventory level
`ȳ` after ordering (a maximizer of `P(x, ·)` over `{y | x ≤ y}`), and every product `i`
with `x_i ≥ y*_i`, no order is placed for product `i`: `ȳ_i = x_i`. -/
theorem no_order_above_base_stock {N : ℕ} (M : Model N)
    (hA1 : M.Assumption1) (hA2 : M.Assumption2) (hA3 : M.Assumption3) (hb : 0 ≤ M.b)
    (hcost : ∀ i, M.s i < M.c i ∧ M.c i < M.p i + M.penalty i)
    (ν : Fin N → Measure ℝ) [∀ i, IsProbabilityMeasure (ν i)] (hν : DemandLaw ν)
    (hsupp : FullSupport ν)
    (ystar : Fin N → ℝ) (hystar_nonneg : 0 ≤ ystar)
    (hystar : IsMaxOn (M.profit (Measure.pi ν) 0) {y | 0 ≤ y} ystar)
    (x : Fin N → ℝ) (hx : 0 ≤ x)
    (ybar : Fin N → ℝ) (hybar_ge : x ≤ ybar)
    (hybar : IsMaxOn (M.profit (Measure.pi ν) x) {y | x ≤ y} ybar)
    (i : Fin N) (hi : ystar i ≤ x i) :
    ybar i = x i := by sorry

end BassokSubstitution
Source
Bassok, Anupindi, Akella, Single-Period Multiproduct Inventory Models with Substitution, Operations Research 47(4):632–642 (1999), p. 636, Theorem 2
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Fix a natural number NNN and index the products by i∈{0,1,…,N−1}i \in \{0,1,\dots,N-1\}i∈{0,1,…,N−1}. All vectors below lie in RN\mathbb{R}^NRN, and every order between vectors is componentwise: u≤vu \le vu≤v means ui≤viu_i \le v_iui​≤vi​ for every iii, and 0≤u0 \le u0≤u means every entry of uuu is nonnegative.

The statement uses the following objects, which it takes from an imported definitions module whose code is not shown here. This read-back can only say how the statement uses them; it cannot say what they mean.

  • MMM is a "model" for NNN products. It carries at least:
    • a quantity bbb;
    • per-product quantities sis_isi​, cic_ici​, pip_ipi​ and a per-product "penalty" πi\pi_iπi​;
    • three properties named Assumption 1, Assumption 2 and Assumption 3;
    • a function profit\mathrm{profit}profit. Given a probability measure on RN\mathbb{R}^NRN and a vector xxx, it returns a function y↦Pμ(x,y)y \mapsto P_\mu(x, y)y↦Pμ​(x,y). That function is maximised below.
  • "DemandLaw" and "FullSupport" are two properties of a family of measures, also defined in that module and not shown.

Hypotheses. The statement assumes all of the following:

  1. MMM satisfies Assumption 1, Assumption 2 and Assumption 3.
  2. 0≤b0 \le b0≤b. This is componentwise if bbb is a vector; its type is not visible here.
  3. For every product iii:
si<ciandci<pi+πi.s_i < c_i \quad\text{and}\quad c_i < p_i + \pi_i .si​<ci​andci​<pi​+πi​.
  1. ν0,…,νN−1\nu_0,\dots,\nu_{N-1}ν0​,…,νN−1​ are probability measures on R\mathbb{R}R, one per product. The family ν\nuν satisfies DemandLaw and FullSupport. The profit function is always evaluated at the product measure μ=ν0⊗⋯⊗νN−1\mu = \nu_0 \otimes \cdots \otimes \nu_{N-1}μ=ν0​⊗⋯⊗νN−1​ on RN\mathbb{R}^NRN, so the coordinates are independent with marginals νi\nu_iνi​.
  2. y∗∈RNy^* \in \mathbb{R}^Ny∗∈RN satisfies y∗≥0y^* \ge 0y∗≥0, and y∗y^*y∗ maximises y↦Pμ(0,y)y \mapsto P_\mu(0, y)y↦Pμ​(0,y) over the nonnegative orthant:
Pμ(0,y)≤Pμ(0,y∗)for all y≥0.P_\mu(0,y) \le P_\mu(0,y^*) \quad \text{for all } y \ge 0 .Pμ​(0,y)≤Pμ​(0,y∗)for all y≥0.
  • y∗y^*y∗ is only assumed to be some maximiser. It is not assumed to be unique.
  1. x∈RNx \in \mathbb{R}^Nx∈RN satisfies x≥0x \ge 0x≥0.
  2. yˉ∈RN\bar y \in \mathbb{R}^Nyˉ​∈RN satisfies x≤yˉx \le \bar yx≤yˉ​, and yˉ\bar yyˉ​ maximises y↦Pμ(x,y)y \mapsto P_\mu(x, y)y↦Pμ​(x,y) over {y:x≤y}\{y : x \le y\}{y:x≤y}:
Pμ(x,y)≤Pμ(x,yˉ)for all y≥x.P_\mu(x,y) \le P_\mu(x,\bar y) \quad \text{for all } y \ge x .Pμ​(x,y)≤Pμ​(x,yˉ​)for all y≥x.
  • yˉ\bar yyˉ​ is also any maximiser, not necessarily unique.
  1. iii is a product index with yi∗≤xiy^*_i \le x_iyi∗​≤xi​.

Conclusion. Under all of these hypotheses,

yˉi=xi.\bar y_i = x_i .yˉ​i​=xi​.

Nothing is claimed about any other coordinate of yˉ\bar yyˉ​. The statement does not assert that a maximiser y∗y^*y∗ or yˉ\bar yyˉ​ exists: it only speaks about maximisers that are given.

Degenerate cases.

  • N=0N = 0N=0: there is no product index iii, so the statement is vacuously true. The same holds for the per-product cost condition (hypothesis 3) and the vectors in R0\mathbb{R}^0R0.
  • No maximiser exists: if, for the given data, Pμ(0,⋅)P_\mu(0,\cdot)Pμ​(0,⋅) has no maximiser on the orthant, or Pμ(x,⋅)P_\mu(x,\cdot)Pμ​(x,⋅) has none on {y≥x}\{y \ge x\}{y≥x}, the statement is vacuous for those data.
  • Unsatisfiable hypotheses: if Assumptions 1–3, DemandLaw or FullSupport cannot all hold together with the cost inequalities, the statement is vacuous. Whether they can hold depends on definitions not shown here.
  • Flat profit function: if Pμ(x,⋅)P_\mu(x,\cdot)Pμ​(x,⋅) is constant on {y≥x}\{y \ge x\}{y≥x}, every yˉ≥x\bar y \ge xyˉ​≥x counts as a maximiser. The conclusion would then require yˉi=xi\bar y_i = x_iyˉ​i​=xi​ for all such yˉ\bar yyˉ​. This could happen if the profit is an integral that takes a default value such as 000 when the integrand is not integrable, or if it has some other junk value. Whether the unseen assumptions rule such cases out cannot be determined from the code shown.
  • Default values in general: because the profit function's definition is not visible, this read-back cannot say whether it involves division by zero, subtraction of natural numbers, integrals of non-integrable functions, suprema or extended-real arithmetic, or what default values those would produce.
Human review
  • Endorsed by Shuze Chen · Sep 27, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Sep 27, 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