Accurate Bernoulli numbers on the 41C, or "how close can you get"?

[Dieter]

(03-19-2014 07:32 PM)Ángel Martin Wrote:  ...the preferred approach is no doubt to use IPFC (based on the inverse error function) - much faster (really several orders of magnitude) and generally more accurate.

"Fast and accurate" always sounds very promising. ;-) Do you have any figures here? How fast is, for instance, qf(0,01) on an actual HP41? You say you used the inverse error function as implemented in the CUDA library. Do you have a link to this algorithm? I found some information on the library, but not on the actual algorithms.

Just for comparison: I currently use a method similar to the one posted above in combination with an optimized method for the CDF. In a regular FOCAL program, the CDF runs in about 3 - 12 seconds, the quantile function requires roughly 3 seconds more. Maybe you can estimate how fast this would run in MCODE.

Dieter