Theorem 4.3, p. 237 — a semismooth, strongly BD-regular zero that is an accumulation point of the damped Newton method attracts the whole sequence superlinearly, and α_k eventually becomes 1
ProvedQiNonsmoothEq.Damped.theorem_4_3Let be locally Lipschitz and B-differentiable at every point, and let . Fix and , and let be any sequence generated by the damped Newton method (Algorithm 4.1) with : solves , with the first nonnegative integer passing the Armijo test (4.1), and . Assume that for all .
Suppose is an accumulation point of and a zero of , that is semismooth and strongly BD-regular at , and that is semismooth on a neighbourhood of . Then:
- the entire sequence converges, ;
- the convergence is superlinear,
- for all sufficiently large .
The theorem answers three questions Pang raised about his damped Newton method for nonsmooth equations: whether the iteration converges to one point, at what rate, and when the step size becomes 1.
Formalization Note "Theorem 4.2 (b) holds" (each generalized Newton equation has a solution) is implied by the run hypothesis, which supplies a solution at every iterate. Global B-differentiability is the setting of Theorem 4.2, on which Algorithm 4.1's analysis rests. Semismoothness on a neighbourhood of is an added hypothesis inherited from Corollary 3.4, whose proof uses (2.16) at points near ; Lemma 2.6 proves (2.16) only at semismooth points. An accumulation point is a cluster point of the sequence ( is frequently in every neighbourhood of ).
Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.
import Mathlib import Definitions.Def_QiNonsmoothEq_Damped_Setting
namespace QiNonsmoothEq.Damped
open Filter Topology NonsmoothNewton.Local NonsmoothNewton.Shared
theorem theorem_4_3 {n : ℕ} (F : EuclideanSpace ℝ (Fin n) → EuclideanSpace ℝ (Fin n))
(hF : LocallyLipschitz F) (hB : ∀ y, BDiffAt F y)
(β σ : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1) (hσ0 : 0 < σ) (hσ1 : σ < 1 / 2)
(x d : ℕ → EuclideanSpace ℝ (Fin n)) (α : ℕ → ℝ)
(hrun : IsDampedNewtonRun F 1 β σ x d α) (hne : ∀ k, F (x k) ≠ 0)
(xstar : EuclideanSpace ℝ (Fin n))
(hclu : MapClusterPt xstar atTop x) (hzero : F xstar = 0)
(hss : SemismoothAt F xstar) (hreg : StronglyBDRegularAt F xstar)
(hssN : ∀ᶠ y in 𝓝 xstar, SemismoothAt F y) :
Tendsto x atTop (𝓝 xstar) ∧
(fun k => x (k + 1) - xstar) =o[atTop] (fun k => x k - xstar) ∧
∀ᶠ k in atTop, α k = 1 := by sorry
end QiNonsmoothEq.Damped
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.