[Trilinos-Users] AztecOO::Iterate error code -3: loss of precision
Denis Davydov
davydden at gmail.com
Mon May 18 04:26:56 EDT 2015
Dear Simone and Mike,
Thanks for your prompt reply.
So far it seems it is related to poor-conditioned system as
with direct solvers I can go further and results look reasonable.
I also notice that the AMG preconditioner is more robust but not enough
to go through the complete hp-refinement cycle.
Kind regards,
Denis
> On 15 May 2015, at 22:00, Deparis Simone <simone.deparis at gmail.com> wrote:
>
> Other 2 possibilities for CG breakdown are
> 1- your preconditioner changes between iterates, for example in case you have inner iterations
> 2- your system (matrix or preconditioner) is not symmetric.
>
> GMRES would cure 2), Flexible GMRES 1)
> They are both available in Belos,
>
> Best regards
> Simone
>
>
>> On 15 May 2015, at 21:49, Heroux, Mike <MHeroux at CSBSJU.EDU> wrote:
>>
>> Denis,
>>
>> This message is generated when the iterative method detects numerical
>> breakdown. Although it can in principle come from coding errors,
>> especially if you have logic errors in formulating your linear system that
>> might make the matrix or RHS values incorrect, it is more likely a symptom
>> of a naturally poorly conditioned system. This is almost certainly the
>> case if you have other systems for which the solver works.
>>
>> In these situations, you can experiment with different preconditioners
>> that might be more robust. Also, you could consider GMRES in this case,
>> perhaps without restart. It can be more robust in these situations.
>>
>> Mike
>>
>>
>> On 5/15/15, 1:55 PM, "Denis Davydov" <davydden at gmail.com> wrote:
>>
>>> Dear all,
>>>
>>> I am getting this error when solving a poisson equation discretized with
>>> hp-FEM and Lagrange basis functions.
>>> The solver is CG.
>>> Any idea what could that be?
>>>
>>> p.s. the error appears only for a very particular combination of h and p
>>> refinement,
>>> perhaps it could be related to a particularly bad condition number of the
>>> matrix?
>>>
>>> Kind regards,
>>> Denis
>>>
>>> _______________________________________________
>>> Trilinos-Users mailing list
>>> Trilinos-Users at trilinos.org
>>> https://trilinos.org/mailman/listinfo/trilinos-users
>>
>> _______________________________________________
>> Trilinos-Users mailing list
>> Trilinos-Users at trilinos.org
>> https://trilinos.org/mailman/listinfo/trilinos-users
>
More information about the Trilinos-Users
mailing list