[Trilinos-Users] problem using amesos

[Massimiliano Martinelli]

On 28/3/11 18:25 , Thornquist, Heidi K wrote:
> Hi MM,
>
> I've used the Amesos KLU solver for several years in our circuit
> simulation code and have occasionally
> come across the singular matrix error (-22).  In many cases the matrix is
> truly close to singular, but
> as a sanity check I try to use SuperLU on the same matrices to see if KLU
> is just being picky.  Given the
> fact that your application is FE based, you might want to give SuperLU a
> shot.  It's easy to build and
> integrate into the Amesos interface.
>
> Thanks,
> Heidi
>
> ---
> Heidi K. Thornquist
> Senior Member of Technical Staff
> Electrical Systems Modeling
> Sandia National Laboratories
> P.O. Box 5800, MS 0316
> Albuquerque, NM  87185-0316
> Phone: (505)284-8426
> Fax:   (505)284-5451
>
>
>
>
>
> On 3/28/11 4:56 AM, "Massimiliano Martinelli"
> <massimiliano.martinelli at gmail.com>  wrote:
>
>> Hi,
>>
>> I'm trying to use Amesos (through Stratimikos) in my FE code but I have
>> a problem.
>>
>> Using AztecOO to solve the linear system, everything is working well,
>> but if I use Amesos I obtain the following error message:
>>
>> terminate called after throwing an instance of
>> 'Thyra::CatastrophicSolveFailure'
>>    what():
>> /Users/max/Downloads/trilinos-10.6.4-Source/packages/stratimikos/adapters/
>> amesos/src/Thyra_AmesosLinearOpWithSolveFactory.cpp:299:
>>
>> Throw number = 1
>>
>> Throw test that evaluated to true: 0!=err
>>
>> Error, NumericFactorization() on amesos solver of type 'Amesos_Klu'
>> returned error code -22!
>> make: *** [run_serial] Abort trap
>>
>>
>> Looking into Amesos implementation, code -22 means that my matrix is
>> numerically singular, but I know that my matrix is NOT
>> singular...(AztecOO give me the correct results)
>>
>> Have you any any idea why this strange behaviour?
>>
>> Thank you in advance.
>>
>> MM
>>
>> -- 
>> Massimiliano Martinelli, PhD.
>> IMATI-CNR Pavia
>> massimiliano.martinelli at gmail.com
>>
>>
>> _______________________________________________
>> Trilinos-Users mailing list
>> Trilinos-Users at software.sandia.gov
>> http://software.sandia.gov/mailman/listinfo/trilinos-users
>
I made some experiments and what I found is something strange: using 
Klu, if I set the parameter AddToDiag to a really tiny value (e.g. 
1.0e-300) I obtain the correct result!
I'm really not able to explain this kind of behaviour...

MM

-- 
Massimiliano Martinelli, PhD.
IMATI-CNR Pavia
massimiliano.martinelli at gmail.com