A note on variational representation for singular values of matrix. (Q1399821): Difference between revisions

The authors consider the variational property of the singular values of a matrix and point out that a theorem in the book of \textit{G. H. Golub} and \textit{C. F. Van Loan} [Matrix Computations, John Hopkins University Press, Baltimore, MD,1989; Zbl 0733.65016)] is incomplete. They prove the following variational representation for the singular values of a matrix. Let \(A \in \mathcal{R}^{m{\times}n}\) and let \(\sigma_i(A)\) denote the \(i\)th largest singular value of \(A\). Then for \(k= 1 : \min\{m,n\}\) \[ \sigma_k(A) = \max_{\dim(S)=k} \min_{\begin{matrix} x\in S \\ x\neq 0 \end{matrix}}\!\max_{\begin{matrix} y\in \mathbb{R}^m \\ y\neq0 \end{matrix}} \frac{y^TAx}{\| x\|_2\| y\|_2} \] \[ = \max_{\dim(T)=k} \min_{\begin{matrix} y\in T \\ y\neq 0 \end{matrix}}\!\max_{\begin{matrix} x\in \mathbb{R}^n \\ x\neq0 \end{matrix}} \frac{y^TAx}{\| x\|_2\| y\|_2}. \]