> I see a lot of classes, but none is giving me either an
> inverse of a big
> *sparse* matrix or a corresponding solution  not to mention other
> expectations/doubts I wrote about :(
I think vnl_lsqr is the class you are looking for. From the documentation
http://www.isbe.man.ac.uk/public_vxl_doc/vxl/vnl/html/class_vnl_lsqr.html
"vnl_lsqr implements an algorithm for large, sparse linear systems and
sparse, linear least squares"
As I undestand it, linear least squares applied to the equation Ax=b will
give you a least squares solution for x, which is as good as you are likely
to get for a massive sparse matrix A.
> I'm sorry, but there is no clear answer there to "questions
> like this".
The place for an overview of numerical algorithms is a book like "Numerical
Recipes." There is no point in us replicating this sort of information in
VXL's documentation. However, the search engine and documentation do provide
a means of finding algorithms. vnl_lsqr comes 4th on a search for "vnl
sparse". Ok, the vnl_lsqr documentation doesn't mention the word "inverse"
(because it doesn't actually do an inverse,) but this is exactly the process
I used to answer your question.
I readily acknowledge that VXL's documentation is far from perfect. If you
can think of some better documentation (either for the vnl chapter in the
book, or the doxygen output,) we would be grateful. Improvements to the VXL
code or documentation are always welcome.
Ian.
