Re: [ublas] Solving linear equations of the form Ax = b ?

From: Michael Stevens (mail_at_[hidden])
Date: 2006-04-14 04:30:30

On Thursday, 13. April 2006 19:35, John Maddock wrote:

> It'll be a few days now before I get test this out, but in the mean time I
> have a few more questions for you:
>
> 1) Is the interface to these functions likely to change in future?
> 2) Is this going to become an official part of ublas and get documented?
> 3) Are folks using this? How reliable is it?  I ask because there are no
> test cases for it in the ublas regression test suite, and I normally work
> on the assumption that "if it's not tested it doesn't work" :-)

John,
Questions regarding "lu.hpp" often comes up so I guess we should do something 
about it!

My feeling is we should move "lu.hpp" to a "contrib" sub-directory under 
"libs/numeric/ublas" with and appropriate README. That we are not advertising 
a stable interface or documentation. Simply supplying useful code that people 
can copy and use in there own projects.

Personally my feeling with regard to 2) is that it should not become part of 
uBLAS. It is not a BLAS function. If factorisations are going to be added 
they should be structured in seperate library with appropriate documentation, 
tests etc

With regard to question 3) I think your assumption would be correct. The good 
thing about the factorisation is that you know what the inverse should be so 
you can potentially check the result. Depending on the numerics of your 
problem a simple inverse sanity check may be good enough.

Michael

Michael

-- 
___________________________________
Michael Stevens Systems Engineering
34128 Kassel, Germany
Phone/Fax: [phone]
Navigation Systems, Estimation  and
                 Bayesian Filtering
    http://bayesclasses.sf.net
___________________________________