Re: derivatives of matrix



I wrote:

>Aside from #853720, there appears to be one other bug
>manifested here. d(X.w)/dw = A, indeed diff(X.w,w) returns X.
>However, d(w'.X)/dw is X', yet diff(transpose(w).X,w))
>returns X, not transpose(X).

Furuya Gosei wrote:

> These are reasonable,but is to diff with vector w 
> to  diff(some,w))? norm2(x) is scalar,so usually we 
> must take gradient or "exterior derivative" to obtain 
> a special 1 form . [...]
> Maybe we must write a suitable and easyusable 
> differntial forms package.

I agree that a package for working with differential
forms is a good idea. However, I think that diff() 
should remain the interface to use that package.
I don't see any advantage to a package that returns
the correct answer for some problem, if diff() returns
a different (incorrect) answer.

Regards,
Robert Dodier

__________________________________
Do you Yahoo!?
Free Pop-Up Blocker - Get it now
http://companion.yahoo.com/