This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: gsl_diff_central
- From: Andrew Steiner <stein at physics dot umn dot edu>
- To: Brian Gough <bjg at network-theory dot co dot uk>
- Cc: gsl-discuss at sources dot redhat dot com
- Date: Fri, 20 Jun 2003 10:04:00 -0500 (CDT)
- Subject: Re: gsl_diff_central
Brian,
Certainly it is clear that neither the present version, nor my
suggested changes represent the state of the art in numerical
differentiation. I'd love to see something better. :) However, the changes
that I suggest at least correct what I deem to be serious problems with
the present version. Not only does the present code fail to deal with
arguments which are not of order unity, but also underestimates the third
derivative for certain functions (simple trigonometric functions for
example) and likely underestimates the error for these functions.
Argh! I just need 8 days in the week...
Andrew
On Fri, 20 Jun 2003, Brian Gough wrote:
> Andrew Steiner writes:
> > I had some ideas for changes to gsl_diff_central in response to my
> > earlier complaint that it fails for large values of "x" (BUG #13). I
> > wanted to post them so that someone could check and make sure that I
> > haven't missed something. If people like it, then I'll make a proper
> > patch or whatever.
> >
>
> Before making any patch could you look at the literature and see what
> the current state of the art is in numerical differentation -- there
> is a list of relevant journals in the GSL design document (available
> from the website).
>
> Brian
>