The diagonal is typically not used in mixed / block methods. So from that perspective you can remove it, I think.
Kent On 21 May 2014 12:44, Garth N. Wells <[email protected]> wrote: > Does the option to keep the diagonal for non-square matrices in assembly > (i.e., no not let the linear algebra backend optimise it away) make much > sense? It causes all sorts of problems with local (process-wise) indexing > of dofs because the ownership ranges of the row and column maps differ. It > would be handy to remove support for preserving the diagonal for non-square > matrices. > > Garth > > _______________________________________________ > fenics mailing list > [email protected] > http://fenicsproject.org/mailman/listinfo/fenics >
_______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
