On Sat, Feb 25, 2012 at 12:18, Matthew Knepley <[email protected]> wrote:

This is incorrect. I want to use PetscSection because DM just cannot do arbitrary data layout. Can we please stop making examples with vertex unknowns on structures meshes? They have nothing left to tell us.

This has nothing to do with DM, that's DMDA. I now realize a weakness in my PF transformation suggestion: it doesn't work with vector elements (Raviart-Thomas or Nedelec) or staggered grids. I don't know a way to include that with the ability to transform from conservative to primitive variables. This is unfortunate because many important preconditioning methods operate in different variables than the nonlinear system is defined in.