[WP34s] Regularized incomplete Beta function

[Dieter]

(05-05-2014 10:16 PM)Paul Dale Wrote:  CNVG? 00 will always check against 1E-14. Depending on mode, CNVG? 03 chooses either 1E-14 or 1E-32 in user code but 1E-24 or 1E-32 in XROM.

I now did the following test: After the quantile has been computed, a relative error of 1E–12 was added back to the result and the iteration restarted. Some code was added to display the relative error of the current approximation. As expected, the next iteration showed 1E–12. The following one mostly displayed a plain zero (no correction, result is considered exact) or an error near 1...5E–34. In two cases a correction near 1...3E–33 was applied.

In other words, even if the last correction was 1E–12, it looks like we can expect 32...34 valid digits. So I assume a test against 1E–14 should be on the safe side.

Dieter