Factorization of analytic functions by means of Koenig's theorem and Toeplitz computations (Q5944727): Difference between revisions

Let \(f(z)\) be an analytic function such that the series \(f(z)=\sum_{i=0}^{\infty} a_{i}z^{i} \) is convergent for \(|x|<R\). Let \(\xi_{i}, i=1,2,\ldots\), be the zeros of \(f(z)\) such that \(|\xi_{i+1}|\geq |\xi_{i}|\)for any \(i\) and \(|\xi_{n+1}|> |\xi_{n}|\) for a given \(n>0\). Define \(r(z)=\prod_{i=1}^{n}(z-\xi_{i})\) and denote by \(\mathbf{v}\) the vector of the first \(n\) coefficients of \(r(z)\). Consider the infinite matrix \(A=(a_{i-j+n})_{i,j=1,2,\ldots}\), defined by the coefficients \(a_{0}, a_{1}, \ldots\), and by the integer \(n\), where we assume \(a_{i}=0\) for \(i<0\). By providing a matrix version of Koenig's theorem, the authors reduce the problem of evaluating the coefficients of the factor \(r(z)\) to that of approximating the first \(n\) entries in the first column of the inverse Toeplitz matrix A, which is in block Hessenberg form, if \(f(z)\) is a power series, or a band matrix when \(f(z)\) is a polynomial. It is also shown that the factorization problem can be reduced to the solution of a matrix equation \(\Phi(X)=0\) for a \(n \times n\) matrix \(X\), where \(\Phi\) is a matrix power series whose coefficients are Toeplitz matrices. Based on the cyclic reduction technique, an algorithm for generating a sequence of vectors \(\mathbf{v}^{2^j}\) that converges quadratically to the vector \(\mathbf{v}\) and has low computation cost, is designed. This algorithm has been tested on seven test polynomial and power series.