FEAP User Forum

FEAP => Parallel FEAP => Topic started by: Sebastian Castro on May 19, 2014, 09:13:20 AM

Title: Error solving in parallel
Post by: Sebastian Castro on May 19, 2014, 09:13:20 AM

Hello everyone,

I have a problem using parFEAP when I use two type of elements (user elements). Every time I run parFEAP I get this message

Code: [Select]

NO CONVERGENCE REASON:  Indefinite Preconditioner or

Code: [Select]

NO CONVERGENCE REASON:  Indefinite Matrix, depending if I use conjugate gradient or minimal residual method.

The weird thing is that if I use parFEAP in meshes with only one kind of element, it works fine. And weirder than before, if I use both kind of elements in the serial version of FEAP I get the right solution (using the same parameters!). Unfortunately I need to do it in parallel in order to use a finer mesh.

I tried with the following options

Code: [Select]

$PETSC_DIR/$PETSC_ARCH/bin/mpirun -np nproc $FEAPHOME8_4/parfeap/feap -ksp_type cg -pc_type jacobiand

Code: [Select]

$PETSC_DIR/$PETSC_ARCH/bin/mpirun -np nproc $FEAPHOME8_4/parfeap/feap -ksp_type minres -pc_type jacobi
Should I use other preconditioner? I tried GAMG but I couldn't make it run

I hope you can help me.

Thanks

Title: Re: Error solving in parallel
Post by: Prof. R.L. Taylor on May 19, 2014, 10:05:48 AM

Can you describe your element formulations?  Do both use the same degree of freedoms? 

Title: Re: Error solving in parallel
Post by: Sebastian Castro on May 19, 2014, 11:05:02 AM

Yes, both of them have the same degree of freedom (action potential). The first one is a line element for the Aliev-Panfilov model and the second one is the same model but in a tetrahedron element.

Title: Re: Error solving in parallel
Post by: Sebastian Castro on May 19, 2014, 08:00:42 PM

Hi

I just notices that I didn't try GMRES... and it works!

You can forget this thread, haha

Thank you