Polynomial division and its computational complexity

Abstract (i) First we show that all the known algorithms for polynomial division can be represented as algorithms for triangular Toeplitz matrix inversion. In spite of the apparent difference of the algorithms of these two classes, their strong equivalence is demonstrated. (ii) Then we accelerate parallel division of two polynomials with integer coefficients of degrees at most m by a factor of log m comparing with the parallel version of the algorithm of Sieveking and Kung. The result relies on the analysis of the recent algorithm of D. Bini adjusted to the division of polynomials over integers. (Some known parallel algorithms attain the same parallel time but use ≥m times more processors.) (iii) Finally the authors' new algorithm improves the estimates for sequential time complexity of division with a remainder of two integer polynomials by a factor of log m, m being the degree of the dividend. Under the parallel model, it attains Boolean logarithmic time, which is asymptotically optimum. The algorithm exploits the reduction of the problem to integer division; the polynomial remainder and quotient are recovered from integer remainder and quotient via binary segmentation. (iv) The latter approach is also extended to the sequential evaluation of the gcd of two polynomials over integers.