This is the mail archive of the
gsl-discuss@sourceware.org
mailing list for the GSL project.
Re: multidimensional numerical integration
- From: Brian Gough <bjg at gnu dot org>
- To: "Kevin Jackman" <jackmank at gmail dot com>
- Cc: gsl-discuss at sourceware dot org
- Date: Tue, 23 Oct 2007 23:10:38 +0100
- Subject: Re: multidimensional numerical integration
- References: <8d19b8a10710221159x23d93074tfea410a9e3bddaeb@mail.gmail.com>
- Reply-to: gsl-discuss at sourceware dot org
At Mon, 22 Oct 2007 12:59:16 -0600,
Kevin Jackman wrote:
> Q: I realize "epsabs" and "epsrel" control the desired accuracy of numerical
> integration, but how? Mainly, how do they each affect the GSL routines, and
> how do they differ from each other? How about on a Quadpack level? I think
> the GSL manual is kind of vague on how these parameters individually
> control numerical integration convergence. (No hard feelings, ok?)
Hello,
There is a definition of epsabs and epsrel in the introductory section
of the Integration chapter but maybe it is not clear enough.
The original book on quadpack has a complete explanation of them. It
is worth getting for serious use, the GSL routines are a faithful
implementation of the algorithms in the book.
I think the book also has a section on error bounds for computing
multidimensional integrals recursively.
--
Brian Gough
GNU Scientific Library -
http://www.gnu.org/software/gsl/