Some matrix properties preserved by generalized matrix functions (Q2302609): Difference between revisions

Let \(A\) be an \(m\times n\) complex matrix with singular value decomposition \(A=U\Lambda V^{H}\), where \(U\) and \(V\) are unitary and \(\Lambda\) is diagonal with nonnegative real entries in decreasing order. Let \(\sigma_{1}\geq\sigma_{2}\geq \dots\) be the singular values of \(A\). Suppose that \(A\) has rank \(r\). Then exactly \(r\) singular values are nonzero and \(A=U_{r}\Lambda_{r}V_{r}^{H}\), where \(U_{r}\) and \(V_{r}\) consist of the first \(r\) columns of \(U\) and \(V\), respectively, and \(\Lambda_{r}\) is the principal \(r\times r\) submatrix of \(\Lambda\). In this case, for every function \(f:\mathbb{R}\rightarrow\mathbb{R}\) we can define \(f^{\diamondsuit}:\mathbb{C}^{m\times n}\rightarrow\mathbb{C}^{m\times n}\) by \(f^{\diamondsuit}(A):=U_{r}f(\Lambda_{r})V_{r}^{H}\) where \(f(\Lambda_{r}):=\operatorname{diag}(f(\sigma_{1}),\dots,f(\sigma_{r}))\). The authors call \(f^{\mathbb{\diamondsuit}}\)a generalized matrix function (a slightly different definition is given in [\textit{J. B. Hawkins} and \textit{A. Ben-Israel}, Linear Multilinear Algebra 1, 163--171 (1973; Zbl 0291.15004)]). Evidently, generalized matrix functions are very different from the usual polynomial or analytic functions defined on a ring of square matrices over \(\mathbb{C}\), and the authors' task is to explore similarities and differences. For example, when \(m=n\) we have: if \(A\) is normal then \(f^{\diamondsuit}(A)\) is normal; and, if \(A\) is a block circulant matrix which has \(d^{2}\) blocks each of which is a \((n/d)\times(n/d)\) circulant, then \(f^{\diamondsuit}(A)\) is also block circulant. Moreover, under suitable conditions the mapping \(A\longmapsto f^{\diamondsuit}(A)\) maps the algebras of Hermitian and skew-Hermitian matrices into themselves, and similarly maps classical groups such as the unitary group and symplectic group into themselves.