Generalized Fritz John Necessary Conditions (§2), p. 41 — at a solution of (1.1) there are ū ≥ 0, v̄, (ū, v̄) ≠ 0, satisfying (2.9)–(2.10)
ProvedMangasarianFJ.GenFJ.generalized_fritz_johnConsider the program (1.1): minimize subject to , , and , , where have continuous first partial derivatives on . Let be a solution of (1.1), i.e. is feasible and is the minimum of on the feasible set . Then there exist and such that
This is the main result of the paper: Fritz John's necessary conditions extended to problems with equality constraints, in a form that does not become trivial when the equalities are present (splitting each equality into two inequalities makes the original conditions hold at every feasible point). It is the basis of the Mangasarian–Fromovitz constraint qualification of §3.
Formalization Note is EuclideanSpace ℝ (Fin n), is Mathlib's gradient, and "continuous first partial derivatives on " (the standing assumption of §1) is ContDiff ℝ 1, equivalent in finite dimension. Indices are Fin m, Fin k (0-based). is split into u0 (multiplier of ) and u; (2.11) is and for all ; (2.12) is " or some or some ". A solution of (1.1) is a global minimizer over (definition IsSolution). and are allowed.
import Mathlib import Definitions.Def_MangasarianFJ_GenFJ_Setting
namespace MangasarianFJ.GenFJ
theorem generalized_fritz_john {n m k : ℕ} (θ : EuclideanSpace ℝ (Fin n) → ℝ)
(g : Fin m → EuclideanSpace ℝ (Fin n) → ℝ) (h : Fin k → EuclideanSpace ℝ (Fin n) → ℝ)
(hθ : ContDiff ℝ 1 θ) (hg : ∀ i, ContDiff ℝ 1 (g i)) (hh : ∀ j, ContDiff ℝ 1 (h j))
(xbar : EuclideanSpace ℝ (Fin n)) (hsol : IsSolution θ g h xbar) :
∃ (u0 : ℝ) (u : Fin m → ℝ) (v : Fin k → ℝ),
u0 • gradient θ xbar + ∑ i, u i • gradient (g i) xbar
+ ∑ j, v j • gradient (h j) xbar = 0 ∧
∑ i, u i * g i xbar = 0 ∧
0 ≤ u0 ∧ (∀ i, 0 ≤ u i) ∧
(u0 ≠ 0 ∨ (∃ i, u i ≠ 0) ∨ ∃ j, v j ≠ 0) := by sorry
end MangasarianFJ.GenFJ
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.