pipette
ENEnglish

A noise-level-free parameter-choice rule for iterated Lavrentiev regularization of nonlinear potential monotone ill-posed problems

Ankit singh, Hemant Sharma

Preprint

En palabras de los autores

We consider nonlinear ill-posed operator equations in a Hilbert space, where is monotone with a Lipschitz continuous Fr\'echet derivative that is self-adjoint near the solution (a locally potential operator), and only noisy data with unknown noise norm is available. For the nonstationary iterated Lavrentiev regularization \[ x_{k+1}^\delta=x_k^\delta-\bigl(F'(x_k^\delta)+\alpha_kI\bigr)^{-1}\bigl(F(x_k^\delta)-y^\delta+\alpha_k(x_k^\delta-x_0)\bigr), \] all classical stopping rules require . We analyze a noise-level-free parameter-choice rule of retrospective type, selecting a minimizer of over a finite computed window. Under a standard noise condition (necessary in view of the Bakushinskii veto) the rule is well defined and . If moreover and , we prove with the computable satisfying , yielding the baseline rate along admissible growing-horizon families, and whenever ; no range-invariance or tangential cone conditions are imposed. Rates under H\"older source conditions with (including saturation) follow; the case , the removal of the self-adjointness assumption, and a full oracle inequality are identified as open problems.

Resultado principalLimitación que admiten los autores

Apareció: miércoles, 21 de octubre. arXiv. Preprint, todavía sin revisión por pares.