Block scaling with optimal euclidean condition (Q795891): Difference between revisions

Let \({\mathbb{C}}^ n\) be the direct sum of the subspaces \(X_ 1,...,X_ r\) with dimensions \(n_ 1,...,n_ r\). Let \({\mathcal M}\) be the set of all matrices \(A=(A_ 1...A_ r)\), \(A_ i\in {\mathbb{C}}^{n\times n_ i}\), such that \(X_ i=Im(A_ I)\). Let \(\| A\|\) denote the Euclidean norm of \(A\in {\mathbb{C}}^{n\times n}\), and \(\kappa_ F(A)=\| A\| \cdot \| A^{-1}\|\) the Euclidean condition number of A. Then \[ \min \{\kappa_ F(A)| A\in {\mathcal M}\}=\sum^{r}_{i=1}\sum^{n_ i}_{j=1}1/\cos \theta^ i_ j, \] where \(\theta^ i_ i,...,\theta^ i_{n_ i}\) are the canonical angles between \(X_ i\) and \(\cap_{j\neq i}X_ j^{\perp}\).