Classification of linear operators in a finite-dimensional unitary space (Q761519)

The matrix \(A=(A_{ij})\) of (in general) rectangular blocks \(A_{ij}\) is said to be equivalent to \(B=(S_ i^{-1}A_{ij}R_ j)\) if \(S_ i\), \(R_ j\) are unitary matrices and a block is ''distinguished'' if \(S_ i=R_ j\). The ''unitary problem'' is to determine a canonical form for a given block matrix under equivalence. The tools needed to achieve this are the well-known results: (1) If A is a rectangular matrix there exist unitary matrices U and V so that \(UAV=a_ 1E_ 1\oplus a_ 2E_ 2\oplus...\oplus a_ kE_ k\oplus 0\) where \(a_ 1>a_ 2>...>a_ k\) are positive real numbers (the singular values of A) and \(E_ 1,...,E_ k\) are identity matrices; (2) Every square matrix is unitarily similar to a block triangular matrix \((F_{ij})\) where \(F_{ij}=0\) if \(i<j\), \(F_{ij}=\lambda_ iE_ i\) where \(E_ i\) is an identity matrix, \(1\leq i\leq k\), \(\lambda_ 1\geq \lambda_ 2\geq...\geq \lambda_ k\) (lexicographic ordering of complex numbers) and \(F_{i+1,i}\) has linearly independent rows if \(\lambda_ i=\lambda_{i+1}.\) The author applies these results in a systematic fashion to the blocks of \(A=(A_{ij})\) and arrives at a canonical form \(A^{\infty}\), which is ''irreducible''. He proves that if A and B are equivalent then \(A^{\infty}=B^{\infty}\). With each canonical matrix he associates a ''scheme'' of labels on the elements and he describes all canonical matrices with a given ''scheme''. The author says his algorithm is much simpler than that of \textit{J. Brenner} [Acta Math. 86, 297-308 (1951; Zbl 0045.297)] and \textit{H. Radjavi} [Trans. Am. Math. Soc. 104, 363-373 (1962; Zbl 0103.251)].