Parallel iterative regularization methods for solving systems of ill-posed equations (Q1026330)

In some situations of science and engineering such as the well-known convex feasibility problems, we are led to solve a system of operator equations \(A_i(x)=0\) for \(i=1,2,\dots, N\), where each equation may be nonlinear and ill-posed. Of course this problem can be considered as a general ill-posed equation \(\mathcal{A}(x)=0\) with \(\mathcal{A}=(A_1, \dots, A_N)\). Under the \(c^{-1}\)-inverse strongly monotone assumptions on the operator \(A_i\), which can be considered as a generalization of the linear self-adjoint nonnegative compact operator, this paper proposes to solve this coupled system by parallel iterative regularization methods. By constituting the iterative algorithm \[ A_i(x_n^i)+(\frac{\alpha_n}{N}+\gamma_n)x_n^i=\gamma_nx_n, i=1,2,\dots, N, \quad x_{n+1}=\frac{1}{N}\sum_{i=1}^Nx_n^i, n=0,1,\dots, \] with suitably chosen relaxed parameters \(\alpha_n\to 0\) and \(\gamma_n\to\infty\), the authors prove that the iteration solution will approach to the minimal normal solution of the original problem. For finding the solution \(x_n^i\) to each regularizing equation, a parallel explicit iterative regularization method is proposed with a convergence proof. Finally the authors apply this iterative scheme to a system of linear first kind Fredholm integral equations to get some numerical results. Roughly speaking, the scheme proposed in this paper can be considered as some modification on the Lavrent'ev regularization for the equation system. It is an interesting problem to consider the numerical behavior of this iterative scheme for some nonlinear system.