[petsc-users] Values of a column in a parallel matrix

[Barry Smith]

> On Mar 20, 2017, at 11:04 PM, Sanjay Govindjee <s_g at berkeley.edu> wrote:
> 
> If you zero the row and column as suggested, you can get what you want by building the
> RHS as you construct your matrix (i.e. do in while assembling your matrix and RHS).

   Yes, and often this is good way to proceed. But I don't think you HAVE to do it this way, I think MatZeroRowsColums() is an alternative.

   Barry

> 
> On 3/20/17 8:48 PM, Daralagodu Dattatreya Jois, Sathwik Bharadw wrote:
>> Hey Barry, 
>> I am already using MatZeroRows. I was actually stuck precisely while applying non zero Dirchlet boundary conditions. But to account for non zero value traditionally we subtract the corresponding column with the right hand side vector. If we zero the rows and columns we can only have value zero for solutions at the boundary. 
>> 
>> Get Outlook for Android
>> 
>> From: Barry Smith <bsmith at mcs.anl.gov>
>> Sent: Monday, March 20, 2017 11:33:20 PM
>> To: Daralagodu Dattatreya Jois, Sathwik Bharadw
>> Cc: petsc-users at mcs.anl.gov
>> Subject: Re: [petsc-users] Values of a column in a parallel matrix
>>  
>> 
>> > On Mar 20, 2017, at 6:07 PM, Daralagodu Dattatreya Jois, Sathwik Bharadw <sdaralagodudatta at wpi.edu> wrote:
>> > 
>> > Hey all,
>> > 
>> >  I am using  AIJ matrix to solve Laplace problem in finite element framework. To apply Neumann boundary conditions I need to obtain values of first and last few columns and subtract it with the corresponding right hand side vector. I understand that MatGetColumnVector andMatGetValues are not collective. Is there any other alternative petsc calls to achieve this in parallel? 
>> 
>>   Hmm, I think you mean non-zero Dirichlet boundary conditions. In that case the recommended approach is calling MatZeroRows() or if you want to preserve symmetry MatZeroRowsColumns(). There are also MatZeroRowsLocal() and MatZeroRowsColumnsLocal() and a few other variants.
>> 
>>   Barry
>> 
>> 
>> >   
>> > Thanks,
>> > Sathwik 
>> 
> 
>