Théorème 3 — weak cluster points of the proximal iterates (13) solve (12), and the iterates form a minimizing sequence
ProvedMartinetReg.ConvexMin.theorem3Let be a real Hilbert space, a nonempty closed convex set, and convex and lower semicontinuous such that every sublevel set , , is bounded. Let be a sequence with and, for every ,
Then:
- every weak cluster point of is a solution of (12), i.e. and for all ;
- is a minimizing sequence: the infimum is a real number and
This is the main result of §IV: the proximal regularization of a convex minimization problem converges in value, and its weak cluster points are minimizers. Martinet remarks that it is more precise than the corresponding result for variational inequalities (§II), where only the weak cluster points are controlled.
Formalization Note "Minimisante" is stated as: there is a real that is the greatest lower bound of and ; this avoids the default value Lean's sInf takes on sets unbounded below, and asserts in particular that is bounded below on . Weak cluster points are cluster points in the weak topology (MapClusterPt in WeakSpace ℝ H), not norm cluster points. No existence of a minimizer, and no lower bound on , is assumed: both are consequences of the hypotheses. (implicit in the paper) is part of the standing hypotheses; is part of the sequence predicate, as in "".
import Mathlib import Definitions.Def_MartinetReg_ConvexMin_Setting
namespace MartinetReg.ConvexMin
open Filter Topology
/-- Théorème 3, §4.3, p. 158: every weak cluster point of a sequence {xⁿ} generated by (13) is a
solution of (12), and {xⁿ} is a minimizing sequence: f(xⁿ) converges to the infimum of f over C. -/
theorem theorem3 {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
[CompleteSpace H] (f : H → ℝ) (C : Set H) (hS : Standing f C)
(x : ℕ → H) (hx : IsProxSeq f C x) :
(∀ xt : H, MartinetReg.VI.IsWeakClusterPt xt x → xt ∈ solSet f C) ∧
∃ m : ℝ, IsGLB (f '' C) m ∧ Tendsto (fun n => f (x n)) atTop (𝓝 m) := by sorry
end MartinetReg.ConvexMin
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.