VectorSpaceOpt.lagrangian
DefinitionVectorSpaceOpt_lagrangiandefinitionlagrangianmultiplieroptimization
Let be an objective, a vector constraint map, and a continuous linear multiplier. Their Lagrangian is
The plus sign is paired with the feasibility convention and dual-positive multipliers; consequently a feasible point satisfies . The definition does not mention the domain or the cone , because these enter through minimization and multiplier feasibility rather than through evaluation of . This small interface is reused by saddle-point, complementarity, sensitivity, and dual-value items. Its linear multiplier dependence makes saddle inequalities stateable without auxiliary coercions.
Definition code
import Mathlib
namespace VectorSpaceOpt
/-- The Lagrangian associated with an objective, a constraint map, and a multiplier. -/
def lagrangian
{X Z : Type*} [NormedAddCommGroup Z] [NormedSpace ℝ Z]
(f : X → ℝ) (G : X → Z) (zstar : Z →L[ℝ] ℝ) (x : X) : ℝ :=
f x + zstar (G x)
end VectorSpaceOptSource
David G. Luenberger, Optimization by Vector Space Methods, Wiley, 1969, public scan https://sites.science.oregonstate.edu/~show/old/142_Luenberger.pdf, Chapter 8, §8.3, Lagrangian appearing in Theorem 1 and its conclusion, printed pp. 217–218; physical PDF pp. 237–238.