Towards a general convergence theory for inexact Newton regularizations (Q849058): Difference between revisions

The problem is to solve approximately the nonlinear ill-posed problem \[ F(x)= y^\delta\tag{1} \] with inexact right-hand side \(y^\delta\) instead of the actual one \(y\), i.e. \(\| y- y^\delta\|\leq\delta\), where the nonnegative noise level \(\delta\) is assumed to be known. Here \(F: D(F)\to Y\) is a nonlinear operator between the real Hilbert spaces \(X\) and \(Y\), where \(D(F)\) denotes the domain of definition of \(F\). A general theory and unified analysis are developed for a class of inexact Newton-type regularization schemes for solving (1). Each method of this class consists of two components, the outer Newton iteration and the inner scheme providing the correction term by regularizing the local linearization. To present a common convergence analysis one compiles five features which not only guarantee convergence but are also shared by various inner regularization schemes, for instance, Landweber iteration, steepest descent iterationn, implicit iteration, iterated Tikhonov method, and method of conjugate gradient.