This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: qr method failed to converge
- To: jean-michel dot rouet at philips dot com
- Subject: Re: qr method failed to converge
- From: Brian Gough <bjg at network-theory dot co dot uk>
- Date: Mon, 29 Oct 2001 13:22:06 +0000 (GMT)
- Cc: gsl-discuss at sourceware dot cygnus dot com
- References: <OF428F1612.AE8DDFFE-ONC1256AF4.003BB73E@diamond.philips.com>
jean-michel.rouet@philips.com writes:
> Can anybody explain why such a hardcoded value "30" is used in the
> qr method ? and how it can we get rid of it (using an other
> hardcoded value such "40" as I did is not a good solution!)
The algorithm is from,
R.S. Martin, G. Peters and J.H. Wilkinson, ``The QR Algorithm for Real
Hessenberg Matrices'', Numerische Mathematik, 14 (1970), 219--231.
They do mention that the value of 30 in their code is adhoc. It might
have been used for single precision and so would need to be increased
now. I'd have to go back and study the paper to see what the rate of
convergence is and how that affects the number of iterations.
regards
Brian Gough