On the local convergence of an iterative approach for inverse singular value problems (Q2508946): Difference between revisions

Consider the following inverse singular value problem: Given \(n\) real \(m \times n\) matrices \((A_ i)_{1 \leq i \leq n } ,\quad m \geq n,\) and \(n\) real numbers \(\sigma_1^* > \sigma_2^* > \cdots > \sigma_n^* > 0;\) find a vector \(c = (c_ i)_{1 \leq i \leq n}\) such that the singular values \(\sigma_ i(c),\quad 1 \leq i \leq n,\) of the matrix \(A(c) = \sum_{1 \leq i \leq n }\;c_ i A_ i\) are precisely \(\sigma_ i^*,\quad 1 \leq i \leq n.\) Let \(\Sigma_* = \text{diag}(\sigma_1^* ,\sigma_2^*, \cdots, \sigma_n^* ) \in \mathbb{R}^{m \times n},\) and \({\mathcal O}(n)\;\) denotes the set of all orthogonal matrices in \(\mathbb{R}^{m \times n}.\) With : \[ {\mathcal A}=\{ A(c);\;c \in \mathbb{R}^n\},\;M(\Sigma_* ) = \{ U\Sigma_*V^T;\;U,V \in {\mathcal O}(n)\}; \] the current iterate \(c^k\) , and \(X_k = U_k\Sigma_*V_k^T \in M(\Sigma_* ),\) the authors prove that there exists: \[ \varepsilon > 0;\;{\tilde c} \in \mathbb{R}^n;\;{\tilde U},{\tilde V} \in {\mathcal O}(m), \] such that, if \(\| c^{\circ} - c^*\| \leq \varepsilon,\) the iterates \(c^k,\quad U_k,\quad V_k,\quad U_k^T A(c^k) V_k\) converge to \[ {\tilde c},\quad {\tilde U},\quad {\tilde V},\quad {\tilde U}^T A({\tilde c}) {\tilde V} = \Sigma_*; \] and correct a paper of \textit{M. T. Chu} [SIAM J. Numer. Anal. 29, No.~3, 885--903 (1992; Zbl 0757.65041)].