Eike Rathke schrieb:
On Monday, 2008-03-10 18:15:29 +0100, Regina Henschel wrote:
significant digits. I looked around for some better coefficient and
found them in Apache Library
Are we allowed to use this values?
You mean to use only the coefficient values?
Yes, I mean the pure values. Problem is, that although Paul Godfrey
shows a way to calculate such coefficients, you need to calculate a lot
of tables with high precision to get _good_ coefficients.
I have no idea whether
coefficient values could be copyrighted or fall under the same license
the source code is licensed, and of course IANAL, so I can't give any
Do you know someone to ask?
The result would be excellent. If you compare in a Calc cell the values
calculated with GAMMALN with the directly calculated values (which is
possible for GAMMALN(n) and GAMMALN(n+0.5) with positive integer n) you
will get TRUE.
Sounds nice, but ...
Btw, do you happen to know the Cephes library?
No, I didn't.
It also has gamma (didn't test, could you take a look at it?) among
other nice algorithms, and is
said to be numerically accurate, several other software products use it
as well. See http://www.moshier.net/#Cephes
I have downloaded it and have a quick glance on it. It divides the
problem in two cases, one is approximated with a polynom, the other with
Stirlings formula. I will need some time to understand in detail, how it
works, before I will be able to adapt it to OOo. The algorithm is more
complex than the solution with the Lanczos approximation.
Can you point me to an ready implementation to test it, before I start
to become acquainted with it?
I currently favour that library, it looks like it could solve our
numerical stability problems, and maybe more. Unfortunately the license
is somewhat unclear, respectively there's just a "may be used freely"
statement. I'll get in contact with Steve and see whether we can arrange
something for the legal department.
BTW: I have come across the gamma function when looking at issue 86294.
Unfortunately a precise gamma function doesn't solve it. Is 86294
another candidate for issue 18704?
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]