Frank Heckenbach <frank@g-n-u.de> a dit :

Besides, as I noted, Emil evaluates the series only in a certain interval (|u| < 1/3, i.e. z < 1/9), so the number of terms needed to get the required accuracy is bounded by a constant (the size of the coefficients array). When you say "too high", do you mean this constant is too big?

Bigger than that of a Tchebiceff polynomial of same accuracy over same range.

Also, as I said, I can't see why there should be large rounding errors.

Mmm. I may have spoken too fast. Rounding errors are a plague in large degree polynomials of alternating signs, if terms (coeff * x^n) are of same order of magnitude. This should not occur for an absolutely convergent series, where the higher degree terms are small, if summed from higher to lower degree. Maurice -- Maurice Lombardi Laboratoire de Spectrometrie Physique, Universite Joseph Fourier de Grenoble, BP87 38402 Saint Martin d'Heres Cedex FRANCE Tel: 33 (0)4 76 51 47 51 Fax: 33 (0)4 76 63 54 95 mailto:Maurice.Lombardi@ujf-grenoble.fr