From help-octave-request at bevo dot che dot wisc dot edu Wed Dec 17 14:33:23 2003
Subject: Re: Octave QR factorization
From: taltman at lbl dot gov
To: A S Hodel <hodelas at eng dot auburn dot edu>
cc: Bart Vandewoestyne <Bart dot Vandewoestyne at pandora dot be>,    <help-octave@bevo.che.wisc.edu>
Date: Wed, 17 Dec 2003 20:32:52 +0000 (UTC)

On Dec 17, 2003 at 2:19pm, A S Hodel wrote:

hodela >norm(Q*R - A)/norm(A), which will (typically) be around m*n*1e-15, 
hodela >where [m,n] = size(A).

Yes, this is EXACTLY what I do in my algorithms, where I use iterative
improvement techniques. The key thing is that you need to analyze your
input matrices, and the iterative calculations that you're applying to
them, to come up with a upper-bound for the minimal amount of the
tolerance ( the expression you wrote above ) that will be acceptibly
small. 

And use:

eps

to check your install's epsilon, or minimal exact (i.e.,
full-precision ) representation of the non-zero difference between two
values.

---

~T



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------