Iterative methods with different rates of convergence for calculating weighted pseudoinverse matrices and weighted normal pseudosolutions with positive definite weights (Q2574237)

The paper is devoted to the development and analysis of iterative methods for approximating weighted pseudoinverse matrices and weighted normal pseudosolutions with positive definite weights, with different rates of convergence. The properties of symmetrizable and weighted pseudoinverse matrices are under consideration. An expansion of weighted pseudoinverse matrices into infinite matrix power products is presented and analyzed. Expansions of such matrices into infinite matrix power products are obtained, including symmetric and symmetrizable matrices and inverse matrices. Expansions of two types are constructed: with the parameters calculated based on weighted singular numbers and without parameters. A number of matrix identities are obtained. Next, iterative processes of various orders of the rate of convergence for calculation of weighted pseudoinverse matrices with positive definite weights are proposed and analyzed. Emphasis is on the rate of convergence of iterative processes, which is determined based on an estimate of the decreasing error of the problem solution, depending on the fixed number of steps of the iterative process. Iterative processes with a quadratic rate of convergence for calculation of weighted normal pseudosolutions with positive definite weights are proposed and analyzed. The solution errors as a function of the number of steps of the iterative process are estimated. Iterative processes for calculation of weighted normal pseudosolutions are used for iterative solution of the least squares problems with constraints and for obtaining an L-pseudosolution (an Lg-pseudosolution, or a coupled normal pseudosolution).