Involutions and angles between subspaces

From MaRDI portal





This paper develops a unified and geometrical framework for understanding involutory matrices \(S\), i.e., \(S^2=I\), through their block structure and its connection to principal angles between subspaces. The authors first give a self-contained treatment of principal angles via the singular values of products of orthogonal projections, leading to joint block diagonalizations that encode the exact relative position of two subspaces. Building on this, they prove that any involution \(S\) is unitarily equivalent (unitarily similar) to a direct sum of scalar blocks \(\pm I\) and \(2\times2\) blocks of the form\N\[\N\begin{pmatrix}0 & x_i^{-1} \\\Nx_i & 0\end{pmatrix}, \qquad 0<x_i\le 1,\N\]\Nand show that these parameters explicitly determine the principal angles \(\alpha_i\) between the \(+1\) and \(-1\) eigenspaces of \(S\) via formulas such as\N\[\N\alpha_i = 2\arctan x_i = \arcsin\!\left(\frac{2x_i}{1+x_i^2}\right).\N\]\NFrom this structural theorem, they derive several geometric and operator-theoretic consequences, including formulas for the Friedrichs angle (the smallest nonzero principal angle between two subspaces), a precise description of the numerical range of an involution as an ellipse with foci \(\pm 1\), and identities linking projections, angles, and Hilbert-Schmidt norms. In the final section, they study dilation problems, showing how contractions can be embedded into larger involutions with optimal or near-optimal operator norm bounds, and they pose open questions concerning minimal dilation constants, including in the real matrix setting.











This page was built for publication: Involutions and angles between subspaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6894391)