Re: Making some minor contributions to GSL

[linas at austin dot ibm dot com]

On Tue, Jan 13, 2004 at 03:56:00PM -0700, Gerard Jungman wrote:
> On Tue, 2004-01-13 at 11:17, linas@austin.ibm.com wrote:
> > 
> > Polygamma functions, Abramowitz & Stegun chapter 6.4 -- these
> > are derivatives of the gamma (factorial) function.   The algo
> > I have is fairly straightforward ... and is accurate to about
> > 1e-14 or so for "small" orders and arguments.  (The "multiplication
> > theorm" breaks down above order 30 or 40 or so, i.e. for arguments
> > greater than a 100 or 200 or so. I'm not sure why.)
> 
> I have never made a study of robustness of the GSL
> implementation, in gsl_sf_psi_n(). There are only

p.s.  I forgot to mention that the problems alluded to above 
only affect my implementation, and they're not as bad as I thought,
the errors are only 1.0e-14, I thought they were worse.  The gsl
implementation passes my tests with better than 1e-15 accuracy.

So other than the negative argument issue, there is no problem.

--linas