From: SourceForge.net <no...@so...> - 2012-11-14 23:38:15
|
Bugs item #3587295, was opened at 2012-11-14 15:38 Message generated for change (Tracker Item Submitted) made by rtoy You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=104933&aid=3587295&group_id=4933 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: Lisp Core - Floating point Group: None Status: Open Resolution: None Priority: 5 Private: No Submitted By: Raymond Toy (rtoy) Assigned to: Nobody/Anonymous (nobody) Summary: erf inaccurate for small bigfloat values Initial Comment: erf(1b-20) -> 1.387778780781446b-17 The correct answer is closer to 1.128379167095513b-20. This is caused by bfloat-erf computing erf using 1-gamma_incomplete(1/2,x^2)/sqrt(%pi). For small x, gamma_incomplete(1/2,x^2) is very close to sqrt(%pi), so we lose lots of precision using this formula. For small x, we should just use the taylor series. ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=104933&aid=3587295&group_id=4933 |