[petsc-users] KSPSetComputeRHS and KSPSetComputeOperators

[iwaddington .]

On Thu, Nov 21, 2013 at 11:50 PM, iwaddington . <iwaddington at gmail.com>wrote:

> But still regarding the issue of setting the matrix that will be used to
> build the preconditioner, what would be the way to do it in the function
> that I described you before , without the use of matcopy to make it equal
> to the system matrix set with MatSetValuesStencil? My idea is to use the
> conjugate gradient method. Sorry to bother so much, but I am learning petsc
> all by myself.
>
>
>
> On Thu, Nov 21, 2013 at 11:27 PM, Matthew Knepley <knepley at gmail.com>wrote:
>
>> On Thu, Nov 21, 2013 at 7:24 PM, iwaddington . <iwaddington at gmail.com>wrote:
>>
>>> My idea is to use MatSetValuesStencil to set the system matrix, because
>>> I am dealing with a dmda, and regarding the prototype, I know that what is
>>> written in the web page is different, but when you look further in the
>>> src/ksp/ksp/interface/itfunc.c<http://www.mcs.anl.gov/petsc/petsc-current/src/ksp/ksp/interface/itfunc.c.html#KSPSetComputeOperators>and also the example files you see that my prototype is in fact the right
>>> one.
>>>
>>
>> I will fix the docs. Yes,  just call MatSetValuesStencil.
>>
>>    Matt
>>
>>
>>> On Thu, Nov 21, 2013 at 11:15 PM, Matthew Knepley <knepley at gmail.com>wrote:
>>>
>>>> On Thu, Nov 21, 2013 at 7:08 PM, iwaddington . <iwaddington at gmail.com>wrote:
>>>>
>>>>> Thanks Matthew, but how exactly do I do that in a user defined
>>>>> function of the form used by KSPSetComputeOperators:
>>>>>
>>>>
>>>> There are a few problems here:
>>>>
>>>> 1) Your prototype is wrong:
>>>>
>>>>
>>>> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/KSP/KSPSetComputeOperators.html
>>>>
>>>> 2) Regardless, people generally use the matrix that is passed in. Thus
>>>> you call
>>>>
>>>>   MatSetValues(*J, ...)
>>>>
>>>>     Why are you not doing this?
>>>>
>>>>  3) It is customary if you have the same system and preconditioner
>>>> matrix to assemble into the preconditioner
>>>>
>>>>    Matt
>>>>
>>>>
>>>>> #undef __FUNCT__
>>>>> #define __FUNCT__ "ComputeMatrix"
>>>>> PetscErrorCode ComputeMatrix(KSP ksp, Mat J,Mat jac,MatStructure *str,
>>>>> void *ctx)
>>>>> {
>>>>>
>>>>> /* variables */
>>>>>
>>>>>   PetscFunctionBeginUser;
>>>>>
>>>>> /* Setting the system linear matrix J*/
>>>>>
>>>>> /* Now I want to set jac=J which is the preconditioning matrix in this
>>>>> part , my first idea was to use matcopy, but I would like something more
>>>>> efficient */
>>>>>
>>>>>  PetscFunctionReturn(0);
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Thu, Nov 21, 2013 at 10:26 PM, Matthew Knepley <knepley at gmail.com>wrote:
>>>>>
>>>>>> On Thu, Nov 21, 2013 at 6:21 PM, iwaddington . <iwaddington at gmail.com
>>>>>> > wrote:
>>>>>>
>>>>>>> Sorry Matthew, but I didn't understand what you said, because it is
>>>>>>> not the case of passing arguments to a function, I have to write the
>>>>>>> function which I will later pass to KSPSetComputeOperators, and in
>>>>>>> the function I have to set the system linear matrix and preconditioning
>>>>>>> matrix.
>>>>>>>
>>>>>>
>>>>>> You can return the same matrix for the system and preconditioner.
>>>>>>
>>>>>>    Matt
>>>>>>
>>>>>>
>>>>>>> On Thu, Nov 21, 2013 at 10:00 PM, Matthew Knepley <knepley at gmail.com
>>>>>>> > wrote:
>>>>>>>
>>>>>>>> On Thu, Nov 21, 2013 at 5:44 PM, iwaddington . <
>>>>>>>> iwaddington at gmail.com> wrote:
>>>>>>>>
>>>>>>>>> HI everybody, I have a doubt concerning how the function
>>>>>>>>> KSPSetComputeRHS works, because according to its prototype one of its
>>>>>>>>> arguments is a pointer to a function that takes a vector object as
>>>>>>>>> argument, not a pointer towards a vector object, so how is it able to set
>>>>>>>>> the rhs if the passage of the vector is done by value and not by reference ?
>>>>>>>>>
>>>>>>>>
>>>>>>>> The values of the Vec can be set, rather than changing the object
>>>>>>>> pointer itself.
>>>>>>>>
>>>>>>>>
>>>>>>>>> Another thing is the functioning of KSPSetComputeOperators. I want
>>>>>>>>> to set the linear system matrix and I want the preconditioning matrix to be
>>>>>>>>> equal, so can I just set the linar system matrix in the function that is
>>>>>>>>> argument of KSPSetComputeOperators, or do I need to make a matcopy in it ?
>>>>>>>>>
>>>>>>>>
>>>>>>>> No, just give the same argument twice.
>>>>>>>>
>>>>>>>>    Matt
>>>>>>>>
>>>>>>>> --
>>>>>>>> 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
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> 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
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> 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
>>>>
>>>
>>>
>>
>>
>> --
>> 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
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20131122/2d4acb71/attachment-0001.html>