MatCreateMPIAIJ

[Xiaoxu Wang]

Hi, Thank you for your help already. Can anybody tell me how to set the 
size of local matrix? Can I set it arbitrarily?

Xiaoxu

Xiaoxu Wang wrote:

>
> The error is at line 1023 of "plog.c".
> ---------------
> An unhandled exception of type 'System.NullReferenceException' 
> occurred in Unknown Module.
>
> Additional information: Object reference not set to an instance of an 
> object.
> ----------------
>
> The matrix I am trying to initialize is the stiffness matrix K of 
> finite element methods. Later the dynamical system is solved as  (I + 
> Dt*K) U(t+1) = U(t) + Dt.F(t) by calling KSPSolve. Any idea how to set 
> the number of local rows?  I am running the code on PC now, but want 
> to keep MatCreateMPIAIJ in it. 
> Xiaoxu
> ----- Original Message ----- From: "Barry Smith" <bsmith at mcs.anl.gov>
> To: <petsc-users at mcs.anl.gov>
> Sent: Tuesday, October 03, 2006 8:32 AM
> Subject: Re: MatCreateMPIAIJ
>
>
>>
>>
>> On Mon, 2 Oct 2006, Xiaoxu Wang wrote:
>>
>>> Hi,
>>>
>>>      I am a beginner on PETSC. This is my question about 
>>> MatCreateMPIAIJ. I
>>> have a sparse matrix and only know the total number of nonzeros each 
>>> row. The
>>> number of local rows and the number of local columns has been set to
>>> PETSC_DECIDE. Therefore I don't know the size of each portion and it 
>>> is hard
>>> to tell d_nnz and o_nnz. I use the total number of nunzeros as d_nnz 
>>> and get
>>> an allocation error.
>>
>>
>>
>>   It would be helpful to know the exact error message:
>>
>>>  How to deal  with  this problem?  Should  I  explicitly
>>> specify the number of local rows and the number of local columns and 
>>> calculate
>>> d_nnz and o_nnz? Or let PETSC decide the number of local rows and 
>>> the number
>>> of local columns, and  then use another function to get these 
>>> values? Any
>>> other better solution?
>>
>>
>>
>>   You should specifically set the number of local rows and columns 
>> and then
>> compute the values of d_nnz and o_nnz. How you determine d_nnz and 
>> o_nnz is
>> problem dependent and depends on what type of discretization you are 
>> using:
>> finite differences, elements etc.
>>
>>
>>   Barry
>>
>>>
>>> Thanks a lot,
>>> Xiaoxu
>>>
>>>
>>
>>
>