Tridiagonalization of complex matrices and a problem of Longstaff (Q1111648)

Given a positive integer k, a square matrix \(B=(b_{ij})\) is called (2k- 1)-diagonal if \(b_{ij}=0\) for \(| i-j| >k-1\). A complex \(n\times n\) square matrix is called (2k-1)-diagonalizable if it is unitarily similar to a (2k-1)-diagonal matrix. The author proves that, for \(n>2k- 1+\sqrt{2k^ 2-2k+1},\) there exist \(n\times n\) complex matrices which are not (2k-1)-diagonalizable. This answers in the negative a question posed for \(k=2\) by \textit{W. E. Longstaff} [ibid. 109, 153-163 (1988; reviewed above)]. Thus, if there is a classification of the indecomposable under unitary similarity complex square matrices in terms of multidiagonal matrices, then there must exist canonical matrices with an arbitrarily large number of diagonals.