← Back to team overview

dolfin team mailing list archive

Re: [Question #105253]: Lu.factorize solver by UMFPACK

 

Question #105253 on DOLFIN changed:
https://answers.launchpad.net/dolfin/+question/105253

Murtazo Nazarov gave more information on the question:
Murtazo Nazarov wrote:
> Question #105253 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/105253
>
>     Status: Answered => Open
>
> Murtazo Nazarov is still having a problem:
> Anders Logg wrote:
>   
>> On Wed, Apr 28, 2010 at 04:40:07AM -0000, Murtazo Nazarov wrote:
>>   
>>     
>>> Question #105253 on DOLFIN changed:
>>> https://answers.launchpad.net/dolfin/+question/105253
>>>
>>>     Status: Expired => Open
>>>
>>> Murtazo Nazarov is still having a problem:
>>> I was satisfied with hypre, but now even hypre stopped working in my mass matrix solver:
>>> https://answers.launchpad.net/dolfin/+question/108259
>>>
>>> It seems no one uses this solvers. So, it would be nice to have a test
>>> benchmark in dolfin/demo which solves different types of the linear
>>> systems by different preconditions.
>>>
>>> Murtazo
>>>     
>>>       
>> Hypre works fine for me. Do you say it breaks for mass matrix only?
>> What about Poisson?
>>
>> --
>> Anders
>>   
>>     
>
> If I solve the mass matrix without applying Dirichlet bc, then there is 
> now problem. But, when I apply Dirichlet/Slip bc and then solve it 
> crashes. However before the last upgrade there was no problem.
>
> I check again now: I have dolfin from one month before: everything works 
> but Lu.factorize;
> I upgraded dolfin couple of days ago: hypre does not work for the mass 
> matrix after applying dirichlet bc, and lu.factorize still does not work.
>
>
> I am checking now for poisson.
>
> murtazo
>
>   

I check Hypre for poisson. It works well.

I realized another strange thing: I solve compressible flow around 
different objects: cylinder and wing with the same bcs and solver. Hypre 
works for wing, but for cylinder non of the Krylov solvers work. Lu 
works fine.

Hypre works well in older solver for both problems.

murtazo

-- 
You received this question notification because you are a member of
DOLFIN Team, which is an answer contact for DOLFIN.



References