Re: [boost] uBLAS vs MTL for sparse matrices

From: Joerg Walter (jhr.walter_at_[hidden])
Date: 2002-05-03 14:56:25

----- Original Message -----
From: "Paul C. Leopardi" <leopardi_at_[hidden]>
To: <boost_at_[hidden]>
Sent: Friday, May 03, 2002 4:38 PM
Subject: Re: [boost] uBLAS vs MTL for sparse matrices

> Joerg,
> Thanks for your reply. Comments below.
> > But these matrices aren't permutation matrices, i.e. the non zero matrix
> elements aren't always 1?
> No, they are not always 1.
> The matrices form an orthogonal basis for a Clifford algebra. eg.
> [1 0] [0 -1] [0 1]  [1  0]
> [0 1],[1  0],[1 0], [0 -1] form a basis for R_{1,1}
> See Ian R. Porteous, "Clifford algebras and the classical groups",
Cambridge
> University Press, 1995.

Without knowing anything about Clifford algebras (should I? ;-): it looks
like your basis could have enough special properties to justify a separate
matrix type (at least for performance reasons ;-). Thanks for the reference.

> > For O(n) complexity you better use sparse_vector_of_sparse_vector or
> compressed_matrix (both currently undocumented ;-(
> OK, I will try compressed_matrix, although I am worried that it is not
> documented. I may not be able to do any coding until June.

That didn't mean, that we do not intend to document it. May be this is
already done in June.

> > P.S.: do we need to consider a class permutation_matrix<>?
> Not on account of GluCat.

Ok.

> My own PS: a while ago, it was suggested that uBLAS and MTL be examined to
> see what the commonalities and differences are. Has this been done?

I didn't suggest it and I didn't do it.

> Is MTL3 still being developed?

I don't know.

> Will it have a uBLAS adaptor?

If the MTL3 interface is suited, we hopefully are able to help our users
with the migration.

Regards

Joerg