Stationarity and complementary slackness extraction
ProvedVectorSpaceOpt.kkt_complementary_slacknessLet a cone-positive continuous functional z₀ accompany a feasible constraint value xConstraint. Suppose the separating inequality
holds for every direction h in the decision space. Then the part linear in h must vanish, and the remaining constant is forced to vanish by feasibility and positivity. Consequently, f' + z₀ ∘ G' = 0 and z₀(xConstraint) = 0. This theorem isolates the sign-sensitive algebra at the end of p. 250. It is a small but valuable reusable interface: geometric separation need only produce the universal affine inequality, after which stationarity and complementary slackness follow without any further topology or differentiability.
import Definitions.Def_VectorSpaceOpt_cone_optimization
namespace VectorSpaceOpt
/-- The stationarity and complementary-slackness extraction on p. 250. -/
theorem kkt_complementary_slackness
{X Z : Type*}
[NormedAddCommGroup X] [NormedSpace ℝ X]
[NormedAddCommGroup Z] [NormedSpace ℝ Z]
(P : ConvexCone ℝ Z) (xConstraint : Z)
(f' : X →L[ℝ] ℝ) (G' : X →L[ℝ] Z) (z₀ : Z →L[ℝ] ℝ)
(hz : dualPositive P z₀) (hfeas : coneLE P xConstraint 0)
(hsep : ∀ h : X, 0 ≤ f' h + z₀ (xConstraint + G' h)) :
f' + z₀.comp G' = 0 ∧ z₀ xConstraint = 0 := by
sorry
end VectorSpaceOptRead-back
What the Lean code literally says, in plain math · gpt-5
Let be real normed spaces, a convex cone in , , and continuous real-linear maps, and a continuous real-linear functional. Assume is nonnegative on , assume —that is, —and assume that for every , . Then is the zero functional on and . The separator inequality includes and both arbitrary and its negative; there are no differentiability, completeness, interior, or regularity hypotheses in this theorem.
Confirmed by the mission captain (proposal self-audit).