On Fri, Aug 17, 2012 at 3:10 AM, Thomas Witkowski <thomas.witkowski at tu-dresden.de <mailto:thomas.witkowski at tu-dresden.de>> wrote: > >> I want to solve some (weakly) coupled system of equations of >> the following form: >> >> A B u >> . = ..... >> 0 C v >> >> >> so, C is the discrete Laplacian and A and B are some more >> complicated operators (I make use of linear finite elements). >> All boundary conditions are periodic, so the unknown v is >> determined only up to a constant. A and B contain both the >> identity operator, so u is fixed. Now I want to solve the >> system on the whole (there are reasons to do it in this way!) >> and I must provide information about the nullspace to the >> solver. When I am right, to provide the correct nullspace I >> must solve one equation with A. Is there any way in PETSc to >> circumvent the problem? >> >> >> If I understand you correctly, your null space vector is (0 I). I >> use the same null space for SNES ex62. > (0 I) cannot be an element of the null space, as multiplying it > with the matrix results in a non-zero vector. Or am I totally > wrong about null spaces of matrices? > > > Maybe you could as your question again. I am not understanding what > you want. > I want to solve the block triangular system as described above. My problem is, that it has a one dimensional null space, but I'm not able to define it. My question is: does anyone can give me an advice how to EITHER compute the null space explicitly OR how to solve the system in such a way that the null space is considered by the solver. The only constraint is that I cannot split the system of equations into two independent solve for both variables. I know that from this description its not clear why there is this constraint, but it would take too long to describe it.
Thomas -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120817/e2c69fa5/attachment-0001.html>
