Regularization of linear approximate schemes by the gradient descent

DOI10.1137/S0036142999364522MaRDI QIDQ2706411zbMATH OpenOpenAlexFDO

Authors V. G. Karmanov, A. F. Izmailov, A. Tret'yakov

Publication date 19 March 2001

Published in SIAM Journal on Numerical Analysis (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1137/s0036142999364522

zbMATH Keywords

approximate schemeconvergencegradient descentgradient methodimplicit iterationLandweber methodlinear operator equationregularizationRichardson method

Mathematics Subject Classification ID

Equations and inequalities involving linear operators, with vector unknowns (47A50) Linear operators and ill-posed problems, regularization (47A52) Numerical solutions to equations with linear operators (65J10) Numerical solutions of ill-posed problems in abstract spaces; regularization (65J20)

Review text

An abstract approximate scheme for a linear operator equation \(Ax=b\) in a Banach space is considered. The scheme is not supposed to be stable. The approximate equation \(A_nx_n=b_n\) is solved by the gradient method with a stopping rule depending on the approximation error of the scheme. In the authors' previous paper [Zh. Vychisl. Mat. Mat. Fiz. 39, No. 9, 1453-1463 (1999; Zbl 0977.65047)] they proved the convergence of approximate solutions for the case of a continuous operator \(A\) and under some strong approximation conditions on \(\{A_n\}\). NEWLINENEWLINENEWLINEIn this paper similar results concerning regularization properties of the considered method are obtained under weaker approximation assumptions. Moreover, it is not assumed that \(A\) is continuous nor its domain is all the space. The results are obtained under the additional assumption that the solution is sourcewise representable, i.e. \(x\in\text{ im }A\). The problem of existence of sourcewise representable solution is also discussed.

Recommendations

• scientific article; zbMATH DE number 665289
• Gradient method for linear approximate schemes
• Regularization of some modifications of the conditional-gradient method
• On regularization procedures with linear accuracy estimates of approximations
• On the regularization effect of stochastic gradient descent applied to least-squares
• scientific article; zbMATH DE number 4143347
• Regularizing algorithms based on the conjugate-gradient method
• scientific article; zbMATH DE number 2210664
• Quadratic regularization methods with finite-difference gradient approximations
• scientific article; zbMATH DE number 3963111

Cited in

(9)

• Stabilizing properties of the gradient method for unstable approximate schemes.
• Regression-based variance reduction approach for strong approximation schemes
• On regularization procedures with linear accuracy estimates of approximations
• On the regularizing behavior of the SDA and SDC gradient methods in the solution of linear ill-posed problems
• scientific article; zbMATH DE number 3845978 (Why is no real title available?)
• Linear regression models, least-squares problems, normal equations, and stopping criteria for the conjugate gradient method
• Stochastic gradient descent for linear inverse problems in Hilbert spaces
• On the stabilizing properties of the gradient method for unstable approximate schemes
• Gradient method for linear approximate schemes

This page was built for publication: Regularization of linear approximate schemes by the gradient descent

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2706411)