[petsc-users] snes + matrix free + coloring

Zou (Non-US), Ling ling.zou at inl.gov
Fri Mar 7 17:45:25 CST 2014


Hi Jed,

I was searching a topic on preconditioning in the mail list and found this
email.


On Wed, Oct 10, 2012 at 2:55 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> Then you aren't using a matrix-free method at all.


In case of using SNESDefaultComputeJacobianColor to compute Jacobian matrix
and mf_operator command, is this exactly what you are describing ('aren't
using a matrix-free method at all')?

The SNESDefaultComputeJacobianColor I found in lines 149-155 from
http://www.mcs.anl.gov/petsc/petsc-2.3.3/src/snes/examples/tutorials/ex5.c.html

Best,

Ling




> Just don't call SNESSetJacobian and it will be done for you. Are you using
> a DM? If not, you will also need to create a MatFDColoring. The easiest way
> to do that, if you don't have a structured mesh, is to assemble the nonzero
> pattern of your operator.
>
>
> On Wed, Oct 10, 2012 at 3:44 PM, <panourg at mech.upatras.gr> wrote:
>
>> Is exactly the same.
>>
>> Kostas
>>
>>
>>
>>
>>
>>
>> > On Wed, Oct 10, 2012 at 3:32 PM, <panourg at mech.upatras.gr> wrote:
>> >
>> >> Dear all
>> >>
>> >> I have a non-linear system which i want to solve it with matrix free
>> >> method. However i must form the whole proconditioning matrix using
>> >> finite
>> >> differences and coloring method.
>> >>
>> >
>> > How is the preconditioning matrix different from the actual operator?
>> >
>> >
>> >>
>> >> I can set the function F() of non-linear system (i take the notation
>> F()
>> >> from documentation).
>> >>
>> >> Could you send me exactly the commands and the order of it which i must
>> >> use or an apt example for that?
>> >>
>> >> I have looked for that in internet but i could not find any suitable
>> >> example.
>> >>
>> >> Kostas
>> >>
>> >> Thanks
>> >>
>> >>
>> >
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140307/20b3a53c/attachment.html>


More information about the petsc-users mailing list