Hi, Toby!

  Thanks for the wish to help me!

  In the previous discussion I've stated that when I solve the same
  PDE (heat equation) with deal.II's CG solver and with PETSC's CG
  solver solution behaves quite different. Particularly with deal.II
  it works and with Petsc - it blows up quite fast (to get the
  solution in this case I have to take time step 1 order of magnitude
  smaller). So the situation is that deal.II's solver works "better"
  then PETSC. But - it is not parallel one, thus I cannot use
  deal.II's solver in parallel code.

  Idea here - to make PETSC work at least with the same precision as
  deal.II's solver does. I am sure that one have only to tune it
  somehow, or via wrapper class if it provides some parameters to
  change, or digging a bit down to actually change the wrapper class.

  I cannot answer your questions because I have no clue about solvers
  and wrapper classes, I use them as a black box.

Regards,
Slawa

Monday, July 13, 2009, 15:54, you wrote:



TDY> Slawa,

>> >   The question arises - is it possible to tune PETSC CG solver
>> > somehow using parameters for wrapper class (if there are any) to
>> > behave in the same way as deal.II's CG solver does?

TDY> Far from a PETSc expert myself, but, I would be willing to give this a
TDY> try in spare time.

TDY> Counter questions arise:

TDY> (1) What is different between the default deal.ii CG solver and the
TDY> default PETScWrapper solver?

TDY> (2) Why is this important? How would modification of what we have
TDY> affect the PETScWrappers? 

TDY> (3) Do you mean to make PETScWrappers default to deal.ii standards? 

TDY> I could agree with (3) if I understood (1)...   :-)

TDY> I do spend alot of time digging into the code of PETSc and SLEPc. If
TDY> you really think this would be a useful task and can provide some
TDY> information for me on my email address; and we can take a look.

TDY> Best,
TDY>         Toby




-- 
Best regards,                                       16.07.2009
Svyatoslav Gladkov, Kharkov, UKRAINE                15:06

_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to