Directional stationarity at a local extremum
ProvedVectorSpaceOpt.gateaux_stationary_of_local_extremumcalculusgateaux-derivativelocal-extremumoptimization
Let be a real normed space, , and let . Suppose is a local minimum or a local maximum of . If the scalar function is differentiable at with derivative , then
This is the direction-by-direction form of the necessary condition for an extremum; it does not strengthen Gateaux differentiability to Fréchet differentiability. The result is the first milestone because it identifies the stationarity condition from which a first-variation identity must arise, while remaining reusable for functionals unrelated to the particular integral action in the mission goal.
Preamble
import Mathlib
Formal statement
namespace VectorSpaceOpt
/-- Luenberger, Chapter 7, §7.4, Theorem 1. -/
theorem gateaux_stationary_of_local_extremum
{X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X]
(f : X → ℝ) (x₀ h : X) (d : ℝ)
(hext : IsLocalMin f x₀ ∨ IsLocalMax f x₀)
(hdir : HasDerivAt (fun s : ℝ ↦ f (x₀ + s • h)) d 0) :
d = 0 := by sorry
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 7, §7.4, Theorem 1, printed p. 178; physical PDF p. 198.