Re: comments on ublas

From: walter_at_[hidden]
Date: 2001-11-20 12:39:47

--- In boost_at_y..., Peter Schmitteckert (boost) <boost_at_s...> wrote:
> 
> Dear All,
> 
> I'm currently reading the documentation of ublas
> and would like to send some comments.
> 
> -> doc/expression.htm
>    "A Scalar Expression is an expression convertible to a builtin 
type."
>    I'd like to  use scalar types / matrix elements which are 
(currently?) 
>    not a builtin C++ types, like interval<..> or 
MyTypeWhichBehavesLikeScalar.

This is a documentation error. Is the following formulation ok for 
you?
"A Scalar Expression is an expression convertible to a scalar."

BTW, if one intends to use something like MyOwnFloatingPointType, 
at least type_traits and promote_traits in traits.h have to be 
extended.

> -> value type
>    In my own matrix package I have defined a norm_type, since the 
type
>    of the norm of a complex matrix is real. 
>    In addition, one should take into account,
>    that the type of an eigen value of an hermitean is again real.
>    This gets interesting once one tries to write a generic eigen 
value
>    routine. E.g., I have written a jacobi algorithm which works for 
arbitrary
>    hermitean matrices that avoids uneccessary complex operations 
and works
>    even for more complicated 'scalar types'.

type_traits already contains a norm_type used in norm_1, norm_2 and 
norm_inf. eigen_value_type(?) is out of scope for ublas.

> -> op(A) * op(B)
>    in addition to op= A, A^T, A^H i would like to have A^c,
>    i.e. complex conjugated Matrices, since I need to evaluate
>    A * B^c.

What's wrong with prod (A, conj (B))?

> Once I manage to get the sources I'll continue.

That's fine.

Regards

Joerg