Series expansion of weighted pseudoinverse matrices and iterative methods for calculating weighted pseudoinverse matrices and weighted normal pseudosolutions (Q2501327)

The authors propose and analyze the expansion of weighted pseudoinverse matrices with positive definite and degenerate weights into matrix power series with negative exponents and arbitrary positive parameters for changing (accelerating) the convergence rate. To derive formulas for the expansion of weighted pseudoinverse matrices into matrix power series a representation of weighted pseudoinverse matrices in terms of the coefficients of characteristic polynomials of symmetrizable matrices is needed. To this end, a statement about symmetrizability of matrices inverse to symmetrizable ones whose definition does not include a condition for the rank of matrices is established. On the basis of the expansion of weighted pseudoinverse matrices with positive definite and degenerate weights into matrix power series with negative exponents and arbitrary positive parameters some iterative methods are constructed for calculating weighted pseudoinverse matrices. Further on, methods for solving constrained least-squares problems as well as problems of calculating \(Lp\)-pseudosolution and \(Lg\)-pseudosolution (a coupled normal pseudosolution) using the iterative methods obtained here for calculating weighted normal pseudosolutions are developed. A condition, for which ML-weighted pseudoinvere matrices coincide with weighted pseudoinverse matrices with degenerate weights are presented. Finally, it is shown that \(L\)- and \(Lg\)-pseudosolution and a coupled normal pseudosolution are also determined by the sum of weighted normal pseudosolution and the ordinary normal pseudosolution.