Tim Odenthal wrote:
Hi!
I need to solve the classical linear equation A*x=b with x the unknown vector and A a very large, sparse and symmetrical matrix. I want to try the conjugate gradient iterative method to solve the problem (unless somebody knows a definitely much better one). Now I've seen that uBLAS provides the symmetric_adaptor class - should I use this one in combination with one of the sparse matrix classes? Does this reduce the memory usage as compared to only using a sparse matrix?
Many thanks in advance
Tim Odenthal
Hello Tim, I don't think the symmetric adaptor will do much with your algorithm. The algorithm in question just needs to know whether the matrix is symmetric, and whether it should use the upper or lower part of the matrix. You could give the MUMPS part of Boost.Numeric_bindings a shot. For Mumps, see here http://mumps.enseeiht.fr/. For the "stable" numeric_bindings, see here http://svn.boost.org/svn/boost/sandbox/numeric_bindings-v1/ Support for MUMPS in the development version of the bindings is expected within a fortnight. In general, for more in-depth answers to numerics and ublas, you could also consider posting to the ublas-specific mailing list. HtH, Cheers, Rutger