On a regularized Levenberg-Marquardt method for solving nonlinear inverse problems (Q967578): Difference between revisions

Let \(F:D(D)\subset X\to Y\) be a nonlinear Fréchet differentiable operator between the Hilbert spaces \(X\) and \(Y\). Further, let \(x^+\) be a solution in \(D(F)\) of \(F(x)=y\) and \(y^\delta\) be a \(\delta\)-approximation of \(y\). Sufficient conditions are given so that the Levenberg-Marquardt method \[ x_{k+1}^\delta=x_k^\delta- (\alpha_kI+F'(x_k^\delta)^*F'(x_k^\delta))^{-1}F'(x_k^\delta)^*(F(x_k^\delta)-y^\delta),\quad k\geq 0 \] (where \(x_0^\delta\in D(F)\) is an initial guess of \(x^+\)) should converge to the solution \(x^+\) as \(\delta \to 0\). In addition, optimal convergence rates are derived under suitable source-wise representations.