Re: Need help for 'Improve logm, sqrtm, funm" project.

[Marco Caliari]

On Mon, 17 Oct 2016, nvs232 wrote:

> Marco Caliari-4 wrote
> > a previous student, Mudit Sharma, started to work on this project some
> > months ago. You can find his work here:
> >
> > https://github.com/RickOne16/matrix/
> >
> > There is a short history of his work here:
> >
> > http://lists.gnu.org/archive/html/octave-maintainers/2016-02/msg00121.html
> >
> > You could start by checking those codes, compare with expm.m, logm.m, and
> > sqrtm.m in Octave and try to understand what is working, what not, and
> > what can be improved.
> >
> > Regards,
> >
> > Marco
>
> Thanks Marco for your reply, in the other post Mudit made it clear that the
> project is still not finished. SO it feels a good point to start.
> Also I saw the github repo and the article. But I did not see the code in
> source code that I have. Is it accepted in octave core, or is it a future
> thing?

The code on github is not in Octave.

> Also Mudit on the other post mentioned that it need to be coneverted to it's
> C++ versions. Should I start porting expm_new to it's C++ version, or I
> should help from some other point?

I do not think it is necessary to translate expm_new in C++. Quoting a 
previous private email to Mudit

------------
you can go with sqrtm. If I remember it correctly, sqrtm in Octave (whose 
code is in libinterp/corefcn/sqrtm.cc) should already be aligned with the 
state of the art

www.maths.manchester.ac.uk/~higham/narep/narep336.ps.gz

On the other hand, Matlab's sqrtm allows three output arguments, while 
Octave's sqrtm only two. This should be fixed. So, the plan for sqrtm 
could be:

1) check that narep336 is still the state of the art (I mean, no more 
recent Higham's paper on sqrtm)

2) fix the output to align with Matlab

In sqrtm.cc you will found sqrtm_utri_inplace, which should be a fast way 
to do what you called sqrtm_tri in logm_new.m. Would it be possibile to 
use it in logm_new, without writing logm_new in C++? 
-------------

So, only some subfunctions might need a C++ version.

Regards,

Marco