Practical fast polynomial multiplication

Abstract

The "fast" polynomial multiplication algorithms for dense univariate polynomials are those which are asymptotically faster than the classical 0(N 2) method. These "fast" algorithms suffer from a common defect that the size of the problem at which they start to be better than the classical method is quite large; so large, in fact that it is impractical to use them in an algebraic manipulation system. A number of techniques are discussed here for improving these fast algorithms. The combination of the best of these improvements results in a Hybrid Mixed Basis FFT multiplication algorithm which has a cross-over point at degree 25 and is generally faster than a basic FFT algorithm, while retaining the desirable O(N log N) timing function of an FFT approach. The application of these methods to multivariate polynomials is also discussed. The use is advocated of the Kronecker Trick to speed up a fast algorithm. This results in a method which has a cross-over point at degree 5 for bivariate polynomials. Both theoretical and empirical computing times are presented for all algorithms discussed.