Hello everyone, So firstly, I'm using version 8.3.0. If this is a real bug but has been addressed in a newer version please let me know, but I have not found any info about it in the change logs up to version 8.4.2 or the user group.
I'm using a block matrix to solve a steady, variable density Navier-Stokes system in parallel (an extension of step-32). Rather than approximate the Schur complement term as the pressure mass matrix, I'm approximating it as the matrix BG, where B is the continuity matrix sub-block and G is the pressure gradient sub-block. So the full Schur complement term is B A^-1 G ~ BG. B and G are non-square matrices, and BG should be a square matrix that has the same size as the pressure mass matrix. In serial, this works just fine and everything goes smoothly. In parallel, if I export B and G and multiply them (I used Mathematica), I also get the correct BG matrix. However, when I use TrilinosWrappers::SparseMatrix::mmult in parallel, something goes wrong and the resulting matrix is incorrect (not even square). I've attached Mathematica's view of the correct BG matrix from mmult in serial and also the BG matrix from mmult using 2 processors in parallel. Any help would be greatly appreciated. Cheers, Ben Shields -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
BG_serial.pdf
Description: Adobe PDF document
BG_parallel.pdf
Description: Adobe PDF document
