Lemma 3.3 — is lower semicontinuous on
ProvedCalamaiMore.ActiveSet.projGrad_norm_lsclower-semicontinuityp2o-batch-p200ap2o-gran-per-chapterp2o-plan-paperp2o-v1projected-gradient
Let be a nonempty closed convex subset of a finite-dimensional real inner product space , and let be continuously differentiable on . Then the map
is lower semicontinuous on : for every sequence converging to , .
The projected gradient itself need not be continuous, since the tangent cone jumps when the point reaches a new face of ; lower semicontinuity of its norm is what survives, and it is what shows that the limit of a sequence with is stationary.
Formalization Note Lower semicontinuity on is Mathlib's LowerSemicontinuousOn, i.e. relative to .
Preamble
import Mathlib import Definitions.Def_CalamaiMore_ActiveSet_projGrad
Formal statement
namespace CalamaiMore.ActiveSet
/-- Calamai–Moré, Lemma 3.3 (p. 104): for `Ω` nonempty closed convex and `f` continuously
differentiable on `Ω`, the map `x ↦ ‖∇_Ω f(x)‖` is lower semicontinuous on `Ω`. -/
theorem projGrad_norm_lsc {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E]
(Ω : Set E) (hΩne : Ω.Nonempty) (hΩc : IsClosed Ω) (hΩcv : Convex ℝ Ω)
(f : E → ℝ) (hfd : ∀ x ∈ Ω, DifferentiableAt ℝ f x) (hfc : ContinuousOn (gradient f) Ω) :
LowerSemicontinuousOn (fun y => ‖projGrad f Ω y‖) Ω := by sorry
end CalamaiMore.ActiveSet
Source
Calamai & Moré, Projected gradient methods for linearly constrained problems, Math. Programming 39 (1987), p. 104, Lemma 3.3
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.