§4, p. 18 — the Tikhonov curve satisfies lim_{ε→0} x_ε = x*
ProvedTikhonovHDD.Ergodic.tikhonov_curve_tendstoconvex-optimizationhilbert-spacep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1tikhonov-regularization
Let be a real Hilbert space and let satisfy the General assumption (convex, twice Fréchet differentiable, gradient Lipschitz on bounded sets, ). Let be the element of minimum norm of the nonempty closed convex set , and for let be the unique minimizer of . Then the Tikhonov approximation curve converges strongly to :
The paper calls this fact well known and uses it in the proof of Theorem 4.2, where gives and hence .
Formalization Note The curve is a map on that is required to be a minimizer of for every ; its values for are irrelevant, since the limit is taken from the right (). Convergence is in norm.
Preamble
import Mathlib import Definitions.Def_TikhonovHDD_Ergodic_TikhonovCurve import Definitions.Def_TikhonovHDD_Ergodic_Setting open Set Filter Topology MeasureTheory
Formal statement
namespace TikhonovHDD.Ergodic
/-- §4, p. 18 ("well known"): the Tikhonov approximation curve `ε ↦ x_ε`,
`x_ε = argmin (g + (ε/2)‖·‖²)`, converges in norm to the minimum-norm element `x⋆` of
`argmin g` as `ε → 0⁺`. -/
theorem tikhonov_curve_tendsto {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H]
[CompleteSpace H]
(g : H → ℝ) (hg : GeneralAssumptionG g) (xstar : H) (hxstar : TikhonovHDD.Strong.IsMinNormMinimizer g xstar)
(xε : ℝ → H) (hxε : ∀ e > 0, IsTikhonovPoint g e (xε e)) :
Tendsto xε (𝓝[>] 0) (𝓝 xstar) := by sorry
end TikhonovHDD.Ergodic
Source
Boţ, Csetnek, László, Tikhonov regularization of a second order dynamical system with Hessian driven damping, arXiv:1911.12845v2, p. 18, §4 ("It is well known that the Tikhonov approximation curve ϵ −→ xϵ satisfies limϵ−→0 xϵ = x∗"); used on p. 20
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.