R-K type Landweber method for nonlinear ill-posed problems (Q2370588)

The paper concerns a regularization method for solving a nonlinear ill-posed problem \(F(x)=y\) in a Hilbert space. The Runge-Kutta (R-K) method of order 2 is applied to the continuous analog of the Landweber method: \(\dot{x}^\delta(t)=F'(x^\delta(t))^*[y^\delta-F(x^\delta(t))]\), \(0<t\leq T\), \(x^\delta(0)=x_0\). Here \(y^\delta\) are available noisy data with \(\| y-y^\delta\| \leq\delta\). This method, called by the authors the Runge-Kutta type Landweber method, gives a sequence of regularized solutions \(\{x^\delta_k\}\), where \(k\) plays the role of regularization parameter. Applying a generalized discrepancy principle as the stopping rule (\(k\leq k_*(\delta)\)) the authors prove the convergence when \(\delta\rightarrow 0\). Moreover, a convergence rate is obtained under a source condition and usual additional assumptions on \(F\). Finally, numerical results of the considered method applied to the convolution equation are compared with the corresponding results for the Landweber method.