No abstract

No abstract

Many important constants, such as e and Apéry's constant ζ(3), can be approximated by a truncated hypergeometric series. The evaluation of such series to high precision has traditionally been done by binary splitting followed by fixed-point division. However, the numerator and the denominator computed by binary splitting usually contain a very large common factor. In this paper, we apply standard computer algebra techniques including modular computation and rational reconstruction to overcome the shortcomings of the binary splitting method. The space complexity of our algorithm is the same as a bound on the size of the reduced numerator and denominator of the series approximation. Moreover, if the predicted bound is small, the time complexity is better than the standard binary splitting approach. Our approach allows a series to be evaluated to a higher precision without additional memory. We show that when our algorithm is applied to compute ζ(3), the memory requirement is significantly reduced compared to the binary splitting approach.

We give a modular algorithm to perform row reduction of a matrix of Ore polynomials with coefficients in Z [t]. Both the transformation matrix and the transformed matrix are computed. The algorithm can be used for finding the rank and left nullspace of such matrices. In the special case of shift polynomials, we obtain algorithms for computing a weak Popov form and for computing a greatest common right divisor (GCRD) and a least common left multiple (LCLM) of matrices of shift polynomials. Our algorithms improve on existing fraction-free algorithms and can be viewed as generalizations of the work of Li and Nemes on GCRDs and LCLMs of Ore polynomials. We define lucky homomorphisms, determine the appropriate normalization, as well as bound the number of homomorphic images required. Our algorithm is output-sensitive, such that the number of homomorphic images required depends on the size of the output. Furthermore, there is no need to verify the result by trial division or multiplication. When our algorithm is used to compute a GCRD and a LCLM of shift polynomials, we obtain a new output-sensitive modular algorithm.