Transformations to rank structures by unitary similarity (Q556881)

Let \(A\) be a complex \(n\times n\) matrix. A Krylov matrix of \(A\) is a matrix of the form \(K:=\left[ v,Av,...,A^{n-1}v\right] \) for some column vector \(v\). If \(A\) is nonderogatory, then \(v\) can be chosen so that \(K\) is nonsingular; in this case \(K^{-1}AK\) is equal to the Frobenius (companion) matrix for \(A\). A matrix \(L=\left[ l_{ij}\right] \) is called strictly \((-k)\)-banded if \(l_{ij}=0\) whenever \(j-i>k\) and \(l_{ij}\neq0\) whenever \(j-i=k\). A typical theorem proved in this paper is of the following form. Let \(A\) be a nonderogatory matrix, let \(L\) be a nonsingular strictly \((-k)\)-banded matrix for some \(k\geq0\), and let \(K\) be a nonsingular Krylov matrix of \(A\). Consider the QR-factorization \(KL=QR\) (\(Q\) unitary and \(R\) upper triangular). Then \(Q^{-1}AQ\) is the sum of a \((-1)\)-banded matrix and a matrix of rank at most \(k\).