[petsc-dev] MatNest and FieldSplit

Sun Apr 14 15:54:43 CDT 2019

I think I figured out what my problem is _exactly_.
The Mat inside the MATNEST on which I’m using a PCFIELDSPLIT is unassembled before the first KSPSolve, except for the last field.
Matt, you nailed it, when I call KSPSetFromOptions on the global PCFIELDSPLIT, then KSPSetUp explicitly on the inner PCFIELDSPLIT, the matrices associated to all fields are created here: https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line656 <https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line656>
Then, whatever I do with the matrix from the last field, currently trying MatCopy(myAssembledS, generatedS), before the first KSPSolve is reset by https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line688 <https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line688> and my solver fails…

So pretty simple question, how do I set a “pmat” for my last assembled field so that https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line646 <https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line646> and https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line686 <https://www.mcs.anl.gov/petsc/petsc-current/src/ksp/pc/impls/fieldsplit/fieldsplit.c.html#line686> will return a non null pmat.

This may sound really trivial but I’m lost in limbo right now. When everything is not wrapped inside an outer PCFIELDSPLIT, everything just work.

Thanks,
Pierre

> On 25 Mar 2019, at 6:57 PM, Pierre Jolivet via petsc-dev <petsc-dev at mcs.anl.gov> wrote:
> 
> Thanks, this makes (slightly) more sense to me know.
> For some reason my application is still not acting properly but I must be screwing somewhere else the nested FieldSplit…
> 
> Thank you,
> Pierre
> 
>> On 24 Mar 2019, at 11:42 PM, Dave May via petsc-dev <petsc-dev at mcs.anl.gov <mailto:petsc-dev at mcs.anl.gov>> wrote:
>> 
>> Matt is right.
>> 
>> When you defined the operator S, you basically invalidate the operator N (in the sense that they are no longer consistent). Hence when you use KSP nest to solve your problem your A matrix looks like 
>>   A = diag[1, 2, 4, 0, 8]
>> but the B matrix you have defined looks like
>>   B = diag[1, 2, 4, 0.00001]
>> 
>> The only way to obtain the correct answer with your code is thus to use the option
>> -ksp_type preonly
>> 
>> Thanks
>> Dave
>> 
>> 
>> 
>> On Sun, 24 Mar 2019 at 22:09, Mark Adams via petsc-dev <petsc-dev at mcs.anl.gov <mailto:petsc-dev at mcs.anl.gov>> wrote:
>> I think he is saying that this line seems to have no effect (and the comment is hence wrong):
>> 
>> KSPSetOperators(subksp[nsplits - 1], S, S);
>> // J2 = [[4, 0] ; [0, 0.00001]]
>> 
>> J2 is a 2x2 but this block has been changed into two single equation fields. Does this KSPSetOperators supposed to copy this 1x1 S matrix into the (1,1) block of the "J2", or do some sort of correct mixing internally, to get what he wants?
>> 
>> BTW, this line does not seem necessary to me so maybe I'm missing something.
>> 
>> KSPSetOperators(sub, J2, J2);
>> 
>> 
>> On Sun, Mar 24, 2019 at 4:33 PM Matthew Knepley via petsc-dev <petsc-dev at mcs.anl.gov <mailto:petsc-dev at mcs.anl.gov>> wrote:
>> On Sun, Mar 24, 2019 at 10:21 AM Pierre Jolivet <pierre.jolivet at enseeiht.fr <mailto:pierre.jolivet at enseeiht.fr>> wrote:
>> It’s a 4x4 matrix.
>> The first 2x2 diagonal matrix is a field.
>> The second 2x2 diagonal matrix is another field.
>> In the second field, the first diagonal coefficient is a subfield.
>> In the second field, the second diagonal coefficient is another subfield.
>> I’m changing the operators from the second subfield (last diagonal coefficient of the matrix).
>> When I solve a system with the complete matrix (2 fields), I get a different “partial solution" than when I solve the “partial system” on just the second field (with the two subfields in which I modified the operators from the second one).
>> 
>> I may understand waht you are doing.
>> Fieldsplit calls MatGetSubMatrix() which can copy values, depending on the implementation,
>> so changing values in the original matrix may or may not change it in the PC.
>>  
>>    Matt
>> 
>> I don’t know if this makes more or less sense… sorry :\
>> Thanks,
>> Pierre
>> 
>>> On 24 Mar 2019, at 8:42 PM, Matthew Knepley <knepley at gmail.com <mailto:knepley at gmail.com>> wrote:
>>> 
>>> On Sat, Mar 23, 2019 at 9:12 PM Pierre Jolivet via petsc-dev <petsc-dev at mcs.anl.gov <mailto:petsc-dev at mcs.anl.gov>> wrote:
>>> I’m trying to figure out why both solutions are not consistent in the following example.
>>> Is what I’m doing complete nonsense?
>>> 
>>> The code does not make clear what you are asking. I can see its a nested fieldsplit.
>>> 
>>>   Thanks,
>>> 
>>>      Matt
>>>  
>>> Thanks in advance for your help,
>>> Pierre
>>> 
>>> 
>>> 
>>> -- 
>>> 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
>>> 
>>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
>> 
>> 
>> 
>> -- 
>> 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
>> 
>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20190414/b8c48f15/attachment.html>