Theorem 3.1, p. 330 — ε²-optimal feasible points satisfying the Lagrange multiplier rule up to ε
ProvedEkelandVP.Constraints.theorem_3_1Let be a real Banach space with dual , let be Fréchet-differentiable, and let be functions, , defining the problem
with feasible set (3.2). Assume the regularity assumption (3.4): at every the derivatives of the saturated constraints are linearly independent. Assume and that is bounded below on , (3.5). Then for every there exist a point with
and real numbers with
The problem (3.1) need have no solution in an infinite-dimensional space; the theorem asserts that nearly optimal points nevertheless satisfy the Karush–Kuhn–Tucker conditions up to an error in the dual norm.
Formalization Note. The infimum over is not formed in Lean: (3.5) is BddBelow (F '' 𝒞) and (3.6) is for every . Non-emptiness of is an added hypothesis: if the page's (3.5) holds () but no exists. Constraints are indexed by Fin m (constraint of the page is Lean index ), is fderiv ℝ F v and is the operator norm on V →L[ℝ] ℝ.
import Mathlib import Definitions.Def_EkelandVP_Constraints_feasibleSet import Definitions.Def_EkelandVP_Constraints_IsRegular
namespace EkelandVP.Constraints
/-- Ekeland (1974), Theorem 3.1, p. 330: on a Banach space `V`, let `F` be Fréchet-differentiable and
`G_1, …, G_m` be `C¹` with the regularity assumption (3.4), and let `F` be bounded below on the
(nonempty) feasible set `𝒞` (3.5). Then for every `ε > 0` there is `v_ε ∈ 𝒞` with
`F(v_ε) ≤ inf_𝒞 F + ε²` (3.6), and reals `λ_1, …, λ_m` with `λ_i ≥ 0` for the inequality constraints,
`λ_i = 0` when `G_i(v_ε) ≠ 0`, and `‖F'(v_ε) - Σ λ_i G_i'(v_ε)‖* ≤ ε` (3.7). -/
theorem theorem_3_1 {V : Type*} [NormedAddCommGroup V] [NormedSpace ℝ V] [CompleteSpace V]
{m : ℕ} (p : ℕ) (hp : p ≤ m) (F : V → ℝ) (G : Fin m → V → ℝ)
(hF : Differentiable ℝ F) (hG : ∀ i, ContDiff ℝ 1 (G i))
(hreg : IsRegular p G)
(hne : (feasibleSet p G).Nonempty) (hbdd : BddBelow (F '' feasibleSet p G))
(ε : ℝ) (hε : 0 < ε) :
∃ v ∈ feasibleSet p G,
(∀ w ∈ feasibleSet p G, F v ≤ F w + ε ^ 2) ∧
∃ lam : Fin m → ℝ,
(∀ i : Fin m, p ≤ i.val → 0 ≤ lam i) ∧
(∀ i : Fin m, G i v ≠ 0 → lam i = 0) ∧
‖fderiv ℝ F v - ∑ i, lam i • fderiv ℝ (G i) v‖ ≤ ε := by sorry
end EkelandVP.Constraints
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.