VectorSpaceOpt.restrictedConcaveConjugate
DefinitionVectorSpaceOpt_restrictedConcaveConjugateLet be a real normed space, , , and . The restricted concave conjugate is the real infimum
A theorem that treats this as finite also assumes that is nonempty and that the displayed set is bounded below, exactly expressing membership of in Luenberger's . Keeping the concave conjugate distinct from a sign-transformed whole-space convex conjugate preserves the book's notation and makes primal-attainment equalities readable. The definition supplies the second half of the dual objective . It is reusable for minimax statements with genuinely restricted domains.
import Mathlib
namespace VectorSpaceOpt
/-- The concave conjugate of a finite-valued function restricted to a set. -/
noncomputable def restrictedConcaveConjugate
{X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X]
(D : Set X) (g : X → ℝ) (ℓ : X →L[ℝ] ℝ) : ℝ :=
sInf ((fun x : X ↦ ℓ x - g x) '' D)
end VectorSpaceOptRead-back
What the Lean code literally says, in plain math · gpt-5
For every real normed vector space , set , real-valued function on , and continuous real-linear functional on , this defines the restricted concave conjugate to be the real infimum . This uses the total real-valued , not an extended-real infimum: when the displayed image is nonempty and bounded below it is the usual infimum, while if is empty or the image is unbounded below Mathlib's totalized real has value . No nonemptiness or boundedness hypothesis is built into the definition, and remains finite-valued.
Confirmed by the mission captain (proposal self-audit).