Iteratively regularized gradient method with a posteriori stopping rule for 2D inverse gravimetry problem (Q2496949)

The inverse problem \( F(x)=f; \) \( F: H_1 \to H_2\) is considered. Here \(F\) is a nonlinear Fréchet differentiable operator on a pair of real Hilbert spaces \(H_1\) and \(H_2\), the element \(f \in H_2\) is given by its \(\delta\)-approximation: \(\| f-f_{\delta} \| \leq \delta.\) In order to solve the above equation numerically, the iteratively regularized gradient method \[ x_0 \in H_1; \;\;x_{n+1}=x_n + \mu_n \{ {F'}^*(x_n) (F(x_n)-f_{\delta})+ \alpha_n (x_n-x_0) \} \] is combined with a generalized discrepancy principle \[ \| F(x_N)-f_{\delta} \| ^2 < \tau \delta \leq \| F(x_n)-f_{\delta} \| ^2. \] Here, \( 0 \leq n < N,\) \(\tau >1,\) \(\mu_n >0\) is an a priori prescribed step size and \(\alpha_n\) is a regularization parameter at the \(n\)th iteration. A convergence theorem is obtained under the source type condition and examples of regularization parameters are given. The proposed algorithm is tested on the 2D inverse gravimetry problem reduced to a nonlinear integral equation of the first kind. The numerical results are presented and some practical recommendations on the choice of parameters are given.