A fast and stable algorithm for splitting polynomials (Q679271)

A numerical algorithm to factorize a degree \(n\) polynomial \(f\) may be described as follows. First pack the roots into clusters near zero and infinity, by means of Graeffe iterations, then factor the transformed polynomial \(G_f\) into two factors \(G_g\) and \(G_h\), where \(G_g\) has all its roots close to zero (respectively, \(G_h\) has all its roots close to infinity). This problem is called splitting. Next, one may return from Graeffe's transformations to obtain \(f=gh\) and, finally, \(g\) and \(h\) may be recursively factorized until one obtains degree 1 or 2 polynomials. The authors develop, analyze and implement an algorithm to perform the splitting of a polynomial whose complexity is in \(O(n \log \varepsilon^{-1})^{1+\delta}\) floating point operations with \(O(\log \varepsilon^{-1})^{1+\delta}\) bits of precision, improving upon previous known algorithms. The paper finishes by presenting tables with numerical results.