[Fwd: Re: Interest in math functions and software]

[JohnT <jrt at worldlinc dot net>]

I looked around the NIST website and sent a note asking about reference
math functions and received this reply which I got the OK to forward. In
a comment below, I noted that the Gnu Scientific Library (GSL) also has
math functions that duplicate some glibc functions. Other libraries such
as Fortran function libraries going back 30 years or more implement
elementary real and complex functions, and some more complicated ones
intended to be in libc. Would borrowing from other GNU or public domain
libraries be a way to fix math problems in glibc?

Also, the math function initiative mentioned in Dr. Boisvert's response
seems like something that GNU-related entities such as Redhat and IBM
Linux development might want to participate in.

John T

Dear John,

Thanks for your interest.  We share your concern.

As a follow-on to our Digital Library of Mathematical Functions
project (dlmf.nist.gov) we are planning to start up an effort to
provide reference data for mathematical functions of the type you
suggest. Doing this an an absolutely reliable way is quite a
challenge, so this is a research project that will be carried out
over a number of years. Sorry that we can't solve your problem
immediately.

Please also note that many others are beginning to have similar
concerns. See http://www.nmconsortium.org/ for example.

Best wishes,

Dr. Ronald F. Boisvert
Chief, Mathematical and Computational Sciences Division
------------------------------------------+---------------------------
National Institute of Standards and Tech. | Voice +1-301-975-3812
100 Bureau Drive, Stop 8910               |   Fax +1-301-975-3553
Gaithersburg, MD 20899-8910 USA           | Email boisvert@nist.gov
------------------------------------------+---------------------------
       URL  http://math.nist.gov/mcsd/Staff/RBoisvert/
----------------------------------------------------------------------


JohnT wrote:
<snip>
>
> The same function (surely not the same code) is in the Gnu Scientific 
> Library as well as in libraries for other programming packages, but 
> where is a standard reference of sufficient precision to be a good 
> measure of how accurate any particular production library is? Does 
> NIST have any standard reference functions that gives accurate results 
> to maybe 256 or 512 bits, enough to provide definitive correct results 
> that production compilers and libraries can be compared to?

<snip>