VectorSpaceOpt.lagrangeDualValue
DefinitionVectorSpaceOpt_lagrangeDualValueLet and fix an objective , a constraint map , and a continuous linear multiplier . The Lagrange dual value is the extended-real infimum
Lean takes an indexed infimum in EReal: it is when is empty and may be when the Lagrangian is unbounded below. The main theorem separately assumes is nonempty and identifies an attained multiplier whose value is the finite primal optimum. This definition preserves the boundary cases needed for a genuine dual function and is reusable independently of strong duality.
import Definitions.Def_VectorSpaceOpt_lagrangian
namespace VectorSpaceOpt
/-- The possibly infinite Lagrange dual value, as an infimum over `Ω`. -/
noncomputable def lagrangeDualValue
{X Z : Type*} [NormedAddCommGroup Z] [NormedSpace ℝ Z]
(Ω : Set X) (f : X → ℝ) (G : X → Z) (zstar : Z →L[ℝ] ℝ) : EReal :=
⨅ x : Ω, ((lagrangian f G zstar x : ℝ) : EReal)
end VectorSpaceOptRead-back
What the Lean code literally says, in plain math · gpt-5
For an arbitrary type , every real normed vector space , set , function , map , and continuous real-linear functional on , this defines the Lagrange dual value as the infimum in the complete ordered type of the embedded finite real numbers over all subtype elements . Thus in extended reals. If is empty, the indexed infimum is ; if is nonempty but the displayed real values are unbounded below, it is . No algebraic or topological structure on is required.
Confirmed by the mission captain (proposal self-audit).