The QuasiGCD(a, b, z) command returns a univariate numeric polynomial g with a positive float eta such that g is a quasi-GCD with precision eta for the input polynomials (a,b). (See [3,2] for a definition of a quasi-GCD in the sense of Schoenhage.)

This quasi-GCD g is derived from the stable algorithm of [2] as follows. The algorithm of [2] computes a numerical pseudo remainder sequence (ai,bi) for (a,b) in a weakly stable way, accepting only the pairs that are well-conditioned (because the others produce instability). The maximum index i for which (ai,bi) is accepted yields the quasi-GCD g=ai provided the 1-norm of bi is small enough in a sense precised in [2]. The value of eta depends in particular  on the value of bi and on the 1-norm of the residual error at the last accepted step.

If the problem is poorly conditioned, the QuasiGCD(a, b, z) command may return FAIL (rather than a meaningless answer). Here, ill-conditioning is a function of the parameter tau. Its default value is the cubic root of the current value of the Digits variable. Decreasing the value of tau yields a more reliable solution. Increasing the value of tau reduces the risk of failure.

Beckermann, B., and Labahn, G. "A fast and numerically stable Euclidean-like algorithm for detecting relatively prime numerical polynomials." Journal of Symbolic Computation. Vol. 26, (1998): 691-714.

Beckermann, B., and Labahn, G. "When are two numerical polynomials relatively prime?" Journal of Symbolic Computation. Vol. 26, (1998): 677-689.

Schoenhage, A. "Quasi-GCD computations" Journal of Complexity. Vol. 1, (1985): 118-137.