Lemma 1, p. 163 — minimizing a convex function over one block of variables leaves a convex function
ProvedTopkisRation.Levels.lemma_1Let and be real vector spaces, a convex set and a convex function. Suppose that for every the values , , are bounded below. Then the function
is convex on the projection .
In the paper this well-known fact gives convexity of and from the recursion (1): the infimum over the decision of a jointly convex function of the state and the decision is convex in the state.
Formalization Note. The paper calls the function and the infimum ; they are renamed and here to avoid a clash with the model's , . The boundedness hypothesis is added: the paper's infimum may be , which is not a real number, and a real sInf of a set unbounded below is the junk value , for which the conclusion can fail. With the hypothesis every infimum is the real number the paper means.
import Mathlib
namespace TopkisRation.Levels
theorem lemma_1 {E F : Type*} [AddCommGroup E] [Module ℝ E] [AddCommGroup F] [Module ℝ F]
(A : Set (E × F)) (hA : Convex ℝ A) (φ : E × F → ℝ) (hφ : ConvexOn ℝ A φ)
(hbdd : ∀ u, BddBelow (φ '' {q | q ∈ A ∧ q.1 = u})) :
ConvexOn ℝ (Prod.fst '' A) (fun u => sInf (φ '' {q | q ∈ A ∧ q.1 = u})) := by sorry
end TopkisRation.Levels
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.