Re: [Boost-bugs] [Boost C++ Libraries] #12066: Accuracy of modified Bessel functions

Subject: Re: [Boost-bugs] [Boost C++ Libraries] #12066: Accuracy of modified Bessel functions
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2016-03-14 08:09:46

#12066: Accuracy of modified Bessel functions
------------------------------------------+--------------------------
  Reporter:  Pavel Holoborodko <pavel@…>  |      Owner:  johnmaddock
      Type:  Library Submissions          |     Status:  new
 Milestone:  To Be Determined             |  Component:  math
   Version:  Boost 1.61.0                 |   Severity:  Optimization
Resolution:                               |   Keywords:
------------------------------------------+--------------------------

Comment (by Pavel Holoborodko <pavel@…>):

 @"...none of our input test values have more than 23-bit precision."

 Probably this affects the accuracy tests. I use 50K uniformly distributed
 test points (Mersenne-Twister) generated in full double precision. Then
 compute function value with 34 digits accuracy ("true" value) and compare
 with double precision counterpart.

 I think usage of extended precision is vital for measuring actual error of
 float/double precision routines, especially on magnitudes comparable to
 machine epsilon.

 I will try to prepare PR in near future (anyone more experienced with
 boost is very welcome to do this, as coefficients are publicly available).

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/12066#comment:5>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.