Re: Improving sqrtm

[Mudit Sharma]

All,

Now that Carne has clarified that Higham's(sqrtm2 same as Matlab's) implementation
can be used for our work, I would like to go ahead with improving the current implementation

of sqrtm.cc.

The algorithms used in sqrtm2.m (https://github.com/RickOne16/matrix/blob/master/sqrtm2.m)

and the current version of sqrtm.cc (https://github.com/RickOne16/sqrtm/blob/master/sqrtm.cc) are same.

One can see on comparison that only the first output argument is exactly the same i.e. the matrix square root.

The second argument produced in each differ(which one should be preferred?) while
the third argument in sqrtm.cc is always '-1'(This needs to be fixed with condest from sqrtm2.m).

Any ideas how to fix sqrtm.cc?

Bests,

Mudit

On Tue, Feb 16, 2016 at 11:02 PM, Carnë Draug <address@hidden> wrote:

On 16 February 2016 at 17:11, Nicholas Jankowski <address@hidden> wrote:
> On Tue, Feb 16, 2016 at 11:45 AM, Mudit Sharma <address@hidden>
> wrote:
>>
>> Indeed that would be against the spirit of open-source.
>> I provided the sqrtm link so as to make it some informative for the
>> readers/maintainers.
>> Infact the improved sqrtm will be based on Higham's latest work on sqrtm
>> which
>> dates back to 1999
>> (www.maths.manchester.ac.uk/~higham/narep/narep336.ps.gz)
>> and Matlab still uses work previous than the latest algorithm proposed in
>> 1999.
>>
>> Higham's work on sqrtm in 1999 will be the basis for the improved sqrtm
>> and not on Matlab's
>> work.
>>
>> Bests,
>> Mudit
>>
>>
>> On Tue, Feb 16, 2016 at 9:31 PM, John W. Eaton <address@hidden> wrote:
>>>
>>> On 02/16/2016 03:10 AM, Mudit Sharma wrote:
>>>>
>>>> All,
>>>> I'm improving sqrtm in Octave and want to align it with
>>>> the sqrtm from Matlab which has 3 output arguments,
>>>> whereas the existing sqrtm.cc from octave has 2 .
>>>>
>>>> Here's the link for sqrtm from Matlab:
>>>> https://github.com/RickOne16/sqrtm/blob/master/sqrtm.m
>>>>
>>>> and this is octave's sqrtm:
>>>> https://github.com/RickOne16/sqrtm/blob/master/sqrtm.cc
>>>>
>>>> How do I change the octave's sqrtm code to align it with that
>>>> from Matlab which has 3 output arguments?
>>>
>>>
>>> One thing you can't do is use the code you've linked to as a basis for
>>> your changes.  That code clearly has a MathWorks copyright notice on it.  So
>>> now I'm not likely to accept any changes from you related to this function.
>>>
>>> If you want to contribute to the development of Octave it is critical
>>> that the work you do is your own, or that it is based on free software.
>>>
>>> jwe
>>>
>>
>
> Generally speaking, we should be able to pull most necessary information
> from public-facing help files, Matlab's public documentation, or user
> reports of behavior. The concern is copyright. Math is public domain.
> Software algorithms can be copyrighted. Copyright involves legal concerns.
> Even using copyrighted code as a reference before writing your own can be
> problematic. John's concern is valid, as he has to maintain integrity of the
> codebase.
>
> Nick J.

To everyone involved. I have attached the last implementation of sqrtm by Higham
before Mathworks.  I have received this file straight from Higham
himself last year.

Carnë