[WP34s] Regularized incomplete Beta function

[Dieter]

(05-05-2014 09:49 PM)Paul Dale Wrote:  Changing the convergence check will pretty much force an extra iteration to detect the convergence. Thus it will cause a slow down. So an increase to the maximum to 6 seems reasonable "just to be sure".

Changing the convergence check from 1E-14 to 1E-24 equals roughly half an iteration (on average). ;-) I am not sure if I understand this correctly, but do you say that CNVG? 00 which usually tests a relative error < 1E–14 automatically switches to 1E–24 if run in XROM? I think that 1E–14 is on the safe side.

Quote:No, I don't have an example that increases accuracy. I'm leaning towards the minimum of thirty digits being sufficient in DP, since we don't guarantee accuracy there.

Yes, that makes sense since this is what can be achieved with 34-digit user code. The one or other last digit may be off. For my earlier humble contributions to the 34s project I tried to get 32 digits +/– 1 ULP.

Dieter