From: David Knezevic <dknezevic@se...>  20110906 17:21:48

On 09/06/2011 12:19 PM, Roy Stogner wrote: > On Tue, 6 Sep 2011, John Peterson wrote: > >> On Tue, Sep 6, 2011 at 3:14 AM, Martin Hess<hessm@...> wrote: >> >>> My background is that I want to use Reduced Basis functionality, on problems resulting >>> from electromagnetics. So I am using an external solver for generating system matrices >>> with nedelec and RTelements, and would like to import the matrices to libmesh, to do >>> parametric model order reduction via reduced basis. >> A trickier part might be using the SparseMatrix/LinearSolver stuff >> without an underlying mesh and finite element discretization. > SparseMatrix and LinearSolver should work just fine without an > underlying mesh, but using the rbOOmit code for the Reduced Basis > functionality is probably out of the question without the underlying > Mesh and System, isn't it? hmm, well the key would be to get RBConstruction to work properly, and RBConstruction just uses the underlying mesh in order to preassemble the necessary (parameter independent) sparse matrices. Assembly for a given parameter then just involves taking a linear combination of the preassembled sparse matrices. If you can read the matrices in instead of assembling them, then I think things should work OK (though there may be some "gotchas" hidden in there somewhere...) Dave 