On Thu, Dec 4, 2014 at 1:00 PM, Raphaël Couturier < [email protected]> wrote:

Hello Matt,

Thank you for all your work on PETSc. I would like to know if you or your colleagues have some examples of code with linear systems in PETSc that are difficult to solve with more of less all preconditioners and GMRES?

I am Cc'ing petsc-dev since better feedback might come from there. ex15 is the p-Laplacian, ex16 is large deformation elasticity (ex54 is infinitesimal). These are somewhat difficult, but most of the difficulty is getting MG to work right. These are likely candidates to be solved with FMG, neglecting Krylov. In the past, hard things to solve have generally been very non-normal matrices. We do not maintain a catalogue since we (I at least) have almost given up on black box solvers in favor of focusing on preconditioners for certain problem types. However, other people on here might have specific problems that they need solved. Thanks, Matt

In fact, I am working on a new method which solves, for example ex15 or ex54 of ksp/ksp/examples/tutorials with large instances (for example 200,000,000 components or more) with the mg preconditioner and for example 8192 cores, faster than GMRES. So I would like to know if it is possible to test other difficult examples?

I have submitted my work to a conference, and as soon as this work is accepted, I will be able to give more details on the method.

Best wishes,

Raphaël Couturier http://members.femto-st.fr/raphael-couturier/

-- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener