Weighted singular value decomposition of matrices with singular weights based on weighted orthogonal transformations (Q747296)

The paper deals with a singular value decomposition and different weighted variants. In the introduction, a short history of the weighted singular value decomposition, weighted pseudoinverses, pseudosolutions, etc. are recalled. The next section contains many notations, starting from \(\mathbb{R}^n\), but most of the notations used later in the paper are quite complicated. There are for example matrices \(H_{EE}^{+p}, (H^p)_{EE}^{+}\). In my opinion, the notations are too complicated. There are also different norms introduced. The authors define for example the ``matrix norm for a square matrix'' as \[ ||A||_H= [ \lambda_{\max}(H_{EE}^{+1/2}A^{\mathrm{T}} HAH^{+1/2}_{EE})]^{1/2}, \] where \(H\) is a symmetric positive semidefinite matrix that satisfies the conditions \(\mathrm{rk}(HA)= \mathrm{rk}(AH) =\mathrm{rk} (A)\) and \(\mathrm{rk}\) stands for the rank. There are also many other definitions: matrices left- or right-symmetrizable by symmetric positive semidefinite matrices \(M\) and \(N\), \(H\)-weighted orthogonal matrices with weight \(H\), \(H\)-weighted pseudoorthogonal matrices with weight \(H\), etc. In the main part of the paper, the authors discuss two variants of the weighted singular value decomposition and the weighted pseudoinversion with singular weights, Variants I and II. Theorem 1 gives a list of assumptions necessary to obtain for an arbitrary rectangular matrix the (probably weighted) singular value decomposition of the form \(A=U\Sigma W^{\text{T}}N\). The formula is fully explained and the theorem is proved. Based on Theorem 1, the existence of the decomposition of a weighted pseudoinverse of the matrix \(A\) is derived and proved. In Variant II, the similar weighted singular decomposition and the weighted pseudoinverse of an arbitrary rectangular matrix is obtained and proved. In the last part of the paper, the authors give us for an arbitrary rectangular matrix \(A\) the decomposition of weighted pseudoinverses into matrix power series. Similarly, based on Variant II, the authors get a decomposition of weighted pseudoinverses with singular weights into the matrix power product.