Re: [boost] Re: ublas: hermitean matrices

From: boost (boost_at_[hidden])
Date: 2001-12-05 10:45:56

Salut,

>
> That is conceivable. Do these border cases justify the introduction
> of a new feature? We then would have to invent something I'd like to
> call return type deduction,  i.e. implement and use a couple of
> traits classes, which describe for example, that the product of a
> hermitean with a real (or a complex with zero imaginary part?) is a
> hermitean and the product of a hermitean with a complex (with nonzero
> imaginary part?) is a general matrix (not to forget the corresponding
> promotion rules).

I would vote for it, but I agree that a proper design is non trivial.
I'm planning to adapt my sparse matrix diagonalization routine
for hermitean matrices to ublas. There I've implemented the Davidson
algorithm, which needs in addition to the scalar product of the hermitean 
matrix with a vector, the diagonal elements of the hermitean matrix.
Since the eigenvalues of the hermitean matrix are real, a generic 
implementation is cleaner using this feature.

Best wishes,
Peter