Proposition 4 — the lemma on
ProvedResetInventory.proposition_4_sS_closed_intervaldynamic-programminginventory-controlp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1threshold-policy
Proposition 4. Let and , and let be continuous and convex on . Then there exists such that for all . Furthermore, let
If such exists, then
- ;
- is non-increasing on ;
- for all ;
- for all with .
This is the classical lemma (Bertsekas, Lemma 4.2.1) for a closed feasible interval. Below , ordering up to saves more than the fixed cost . From up to , no reachable level saves more than , so nothing is ordered.
Differences from the source:
- generalization: the paper's function G(·,t) = c z + H(z,t) + γ E J((z−w)⁺,t+1) → any function G continuous and convex on [0, c_max]; reason: the proposition's hypotheses are exactly continuity and convexity of G(·,t); the paper's G is an instance
- encoding: let s = inf{…}; if such s exists → if the set {z ∈ [0,S] | G S + K = G z} is nonempty, then for s its infimum (sInf); reason: the set is bounded below by 0, so its infimum exists exactly when it is nonempty; the guard avoids Lean's sInf ∅ = 0
Preamble
import Mathlib
Formal statement
namespace ResetInventory
/-- **Proposition 4** of Lee, Mintz, Aswani, Shen, Yang, arXiv:2209.03571v1, pp.18–19. If `G(·, t)`
is continuous and convex on `[0, c_max]`, there exists `S` such that `G(S, t) ≤ G(z, t)` for all
`z ∈ [0, c_max]`. Furthermore, let `s = inf{z ∈ [0, S] | G(S, t) + K = G(z, t)}`. If such `s`
exists, then
1. `G(S, t) + K = G(s, t)`;
2. `G(·, t)` is non-increasing on `[0, s]`;
3. `G(S, t) + K ≤ G(z, t)` for all `z ∈ [0, s]`;
4. `G(y, t) ≤ G(z, t) + K` for all `y, z` with `s ≤ y ≤ z ≤ c_max`.
**Formalization Note** `G` is `G(·, t)` for the fixed `t`, an arbitrary function continuous and
convex on `[0, c_max]`. "`s` exists" is the nonemptiness of the set (it is bounded below by `0`),
and `s` is its `sInf`; nonemptiness is required because Lean's `sInf ∅ = 0`. `c_max ≥ 0`
(Assumption 4) and `K ≥ 0` (Assumption 2) are the paper's standing assumptions. -/
theorem proposition_4_sS_closed_interval (cmax K : ℝ) (G : ℝ → ℝ)
(hcmax : 0 ≤ cmax) (hK : 0 ≤ K)
(hcont : ContinuousOn G (Set.Icc 0 cmax)) (hconv : ConvexOn ℝ (Set.Icc 0 cmax) G) :
∃ S ∈ Set.Icc 0 cmax, (∀ z ∈ Set.Icc 0 cmax, G S ≤ G z) ∧
({z : ℝ | z ∈ Set.Icc 0 S ∧ G S + K = G z}.Nonempty →
∀ s : ℝ, s = sInf {z : ℝ | z ∈ Set.Icc 0 S ∧ G S + K = G z} →
G S + K = G s ∧
AntitoneOn G (Set.Icc 0 s) ∧
(∀ z ∈ Set.Icc 0 s, G S + K ≤ G z) ∧
(∀ y z : ℝ, s ≤ y → y ≤ z → z ≤ cmax → G y ≤ G z + K)) := by sorry
end ResetInventory
Source
Lee, Mintz, Aswani, Shen, Yang, Optimal Policy for Inventory Management with Periodic and Controlled Resets, arXiv:2209.03571v1, Proposition 4, pp.18–19; proof, p.32
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.