Re: logm, sqrtm, funm

[Philip Nienhuis]

"Tommy Guy" wrote:
> funm and logm do have the same problems.  Using a Schur decomposition
> followed by an algorithm to calculate the function of the upper
> triangular algorithm is probably what the competition is using, but it
> is also an openly published algorithm with good results.  sqrtm (in
> Octave source) has been converted to use the Schur decomposition
> method and implemented in c++.  This provides a guide to anyone
> wanting to try to fix logm.

Repeating the same code for logm, sqrtm, <whatever>m, .... looks a bit silly
to me. IMO it's better to write a core-octave funm (or fix funm in
octave-forge) so that many more matrix functions become available.
FWIW, last time I looked (quite a while ago BTW) the competition's logm was
invoking funm too. After a quick check on-line a few minutes ago I now see
that they use the method mentioned in the very same literature reference you
mentioned earlier.
BTW complex input matrices may need special treatment - is this covered in
the sqrtm c++ source?

Is (a sort of) eval or feval possible in .oct-files? I suppose not, but if
it is, generalizing sqrtm to funm might be straightforward. Note: I'm just
guessing here.

Philip

-- 
View this message in context: 
http://n4.nabble.com/logm-sqrtm-funm-tp2018829p2019421.html
Sent from the Octave - General mailing list archive at Nabble.com.