On Fri, 2005-07-15 at 01:51 +0300, Evgeny Stambulchik wrote: > > Well, I mean - for (|a| > 10 || |b| > 10) the function (the original > variant of it) refuses to work at all. So I'd say it's better with the > patch ;-). Now, I didn't compare results for integer negative a or b > below 10 (by abs. value), but don't think a sum of a few (< 10) terms > can introduce any noticeable rounding errors.
I don't have time right now to look at this. It is not as trivial as it seems, since the series summation can become unstable for large arguments. One question is how to produce a meaningful error estimate. The patch might be ok, but it might not be consistent with the current logic. One has to check what each of those functions is doing inside the respective blocks. And test cases need to be generated for the newly exposed parameter regions. -- Gerard Jungman <[EMAIL PROTECTED]> Los Alamos National Laboratory _______________________________________________ Bug-gsl mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gsl
