Théorème 1 (= Lemme 2) — every weak cluster point of the regularized iterates solves the variational inequality (1)
ProvedMartinetReg.VI.theorem1Let be a real Hilbert space with dual , let be convex, closed and bounded, and let be monotone and hemicontinuous on . Let be generated by the regularization algorithm: and, for each , satisfies
Then every weak cluster point of is a solution of the variational inequality (1):
This is the main result of Martinet's note: the first convergence theorem for what is now called the proximal point algorithm. Since is bounded, weak cluster points exist, so the theorem guarantees that the iteration approaches the solution set.
Formalization Note. is real (InnerProductSpace ℝ H, CompleteSpace H), is StrongDual ℝ H and is φ y. Hemicontinuity is weak-* continuity of on for . The sequence is any sequence satisfying (2) at every step (IsRegSeq, which includes ). A weak cluster point is MapClusterPt in WeakSpace ℝ H. No hypothesis that (1) has a solution and no boundedness of is assumed. Lemme 2 of the paper (p. 156) is the same statement, since is the solution set of (1).
import Mathlib import Definitions.Def_MartinetReg_VI_Setting
namespace MartinetReg.VI
open Filter Topology
/-- Théorème 1, p. 155 (= Lemme 2, p. 156): every weak cluster point of the
regularized iterates `{xⁿ}` is a solution of the variational inequality (1). -/
theorem theorem1 {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
(T : H → StrongDual ℝ H) (C : Set H) (hS : Standing T C)
(x : ℕ → H) (hx : IsRegSeq T C x) :
∀ xt : H, IsWeakClusterPt xt x → xt ∈ solSet T C := by sorry
end MartinetReg.VI
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.