Fact 3.1 — minimizing a convex function over [x, ∞)
ProvedLeviSampling.BaseStock.min_above_convexLet be convex with a global minimizer , and let . Then
This is what makes base-stock policies optimal in the dynamic program (1)–(2): ordering up to is optimal from any inventory below , and not ordering is optimal above it.
Formalization Note is the infimum of over (minAbove).
import Mathlib import Definitions.Def_LeviSampling_BaseStock_Model
namespace LeviSampling.BaseStock
/-- **Fact 3.1** (Levi, Roundy & Shmoys 2007, p. 828). Let `f : ℝ → ℝ` be convex with a
minimizer `r`. Then `w(x) = min_{y ≥ x} f(y)` is convex, `w(x) = f(r)` for `x ≤ r`, and
`w(x) = f(x)` for `x > r`. -/
theorem min_above_convex (f : ℝ → ℝ) (hf : ConvexOn ℝ Set.univ f) (r : ℝ)
(hr : ∀ y, f r ≤ f y) :
ConvexOn ℝ Set.univ (minAbove f) ∧
(∀ x, x ≤ r → minAbove f x = f r) ∧
(∀ x, r < x → minAbove f x = f x) := by sorry
end LeviSampling.BaseStock
Read-back
What the Lean code literally says, in plain math · claude-opus-5-5
Let be convex on all of , and let be a global minimizer of : for all . The minimizer is not required to be the smallest one. Define
the real infimum over , with the convention that it is if unbounded below. The theorem asserts all three of the following:
- is convex on .
- for every .
- for every .
There are no further hypotheses. Since bounds every value from below, the infimum convention is never triggered. If has several minimizers, may be any of them, and the claims are asserted for that choice.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.