Zero cancellation for general rational matrix functions (Q734912)

A general rational matrix function (\textit{rmf}) is characterized by its structural elements: finite and infinite poles and zeros together with their partial multiplicities and the minimal indices of polynomial basis of the null space to the left and right. The problem of elimination (cancelling or dislocation) of some of the structural elements of an \textit{rmf} \(R(\lambda)\) by multiplication with an invertible \textit{rmf} \(R_{\ell}(\lambda)\), i. e., \(R_{\ell}(\lambda)R(\lambda)= \widehat{R}_{\ell}(\lambda)\), where \(\widehat{R}_{\ell}(\lambda)\) has only part of structural elements of the original \(R(\lambda)\), is significant for many applications in linear system theory. In this paper, previous results on zero and pole cancellation are extended. The given \textit{rmf} are completely general, possibly improper or polynomial and without any restriction on its structural elements. The class of invertible factors of minimal McMillan degree that are able to cancel any desired zeros of \(R(\lambda)\), possibly including those at infinity, are characterized. \(J\)-symmetries with respect to both the imaginary axis and the unit circle are considered and necessary and sufficient solvability conditions are given using the so called centered type of realization. All proposed solutions are based preferably on constant unitary transformations, numerically -- sound solutions of pencil equations and numerical reliability of the underlying algorithms for eigenvalue assignment. Finally, some relevant numerical examples and algorithms for the construction of solutions are given.