§3, (11) — Ekeland's theorem with the distance estimate
ProvedAubinLipschitz.ekeland_with_distance_boundekelandp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1variational-principle
Let be a complete metric space, lower semicontinuous and bounded below, and . Then there is such that
The first inequality is what bounds the distance from to the starting point in the proof of Theorem 2 of §3.
Differences from the source:
- generalization: the continuous function V(x,y) = ‖y₂ − y‖ on graph(F) with the metric ‖x − x'‖ + ‖y − y'‖ → any lower semicontinuous bounded-below f on any complete metric space; reason: this is the cited theorem, and the instance Z = graph(F), z₁ = (x₁,y₁) gives (11).
Preamble
import Mathlib
Formal statement
namespace AubinLipschitz
/-- Aubin, Lipschitz Behavior of Solutions to Convex Minimization Problems, IIASA WP-81-76 (1981), §3, display (11), p.25 (Ekeland's theorem, in the form applied there): for a complete
metric space `Z`, a lower semicontinuous `f : Z → ℝ` bounded below, `ε > 0` and `z₁ ∈ Z`, there
is `zbar` with `f(zbar) + ε d(zbar, z₁) ≤ f(z₁)` and `f(zbar) ≤ f(z) + ε d(zbar, z)` for all `z`. -/
theorem ekeland_with_distance_bound {Z : Type*} [MetricSpace Z] [CompleteSpace Z] (f : Z → ℝ)
(hf : LowerSemicontinuous f) (m : ℝ) (hm : ∀ z, m ≤ f z) (ε : ℝ) (hε : 0 < ε) (z₁ : Z) :
∃ zbar : Z, f zbar + ε * dist zbar z₁ ≤ f z₁ ∧ ∀ z : Z, f zbar ≤ f z + ε * dist zbar z := by sorry
end AubinLipschitz
Source
Aubin, Lipschitz Behavior of Solutions to Convex Minimization Problems, IIASA WP-81-76 (1981), §3, display (11), p.25
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.