This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: Landau distribution
- To: Dave Morrison <dave at bnl dot gov>
- Subject: Re: Landau distribution
- From: Brian Gough <bjg at network-theory dot co dot uk>
- Date: Thu, 15 Mar 2001 20:18:25 +0000 (GMT)
- Cc: GSL mailing list <gsl-discuss at sourceware dot cygnus dot com>
- References: <3AB0E229.6527EBE0@bnl.gov>
> OK, now the more general issue. Schorr's technique is fairly fast, but
> not astoundingly precise. In his paper he claims it's good to a few
> significant digits in the middle of the approximation interval, but only
> good to about 5% or so at the end of that range. That's not exactly the
> sort of stuff numerical analysts write home about. However, it's fine
> for describing many physical situations, and that's how it's typically
> used. The question then is GSL interested in having some distributions
> that are "good enough" for some, but not all, purposes. This case would
> satisfy many physicists but few mathematicians (I think).
If the approximation is the only game in town then it's better to use
that rather than not have a useful function. The limitations should
just be noted in the documentation.
For special functions an error estimate is part of the computed answer
so the function can accommodate a region where the approximation is
less good.