Re: half-assed rcond

[David Bateman]

Søren Hauberg wrote:
> The choice to use 'det', was at best random, so I'll be sure to use
> 'inv' from now on. The reason I posted, was mostly to hear if Octave
> should have an implementation of 'rcond' based on such a simple
> implementation.
If it was implemented we should do it as an oct-file and probably use
the solve methods, as lower cost than the inversion of the matrix.
Ideally the matrix classes would separate the matrix factorization from
the back-substitution, though for certain matrix type LAPACK doesn't
separate the two operations and so that not that easy. There a solve
with a dummy RHS is probably the way to go.

We'd also have to re-write the sparse tri-diagonal solvers slightly so
that they call the banded solvers if the condition number is requested
as in the way they are currently written there is no means of
calculating the condition number..

Regards
David

-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary