PETSc+Hypre
 

Hi

Is there any one having experience of using HYPRE as a preconditioner for PETSc KSP solvers?

Please guide me through installation, ...

Best Regards

Soheyl

Re: PETSc+Hypre
 

Yes, I just installed PETSc-2.2.1 and Hypre on our cluster and the process is straight forward. I assume you have a BLAS a LAPACK and some sort of MPI already installed on your system, than adding PETSc to it is fairly simple, since PETSc comes with a very advanced configure Python script.

just use something like:

./config/configure.py --download-hypre=ifneeded

in the PETSc source dir and of course check all the othe roptions you have with this script.

If you don't have BLAS,LAPACK and some MPI than PETSc can also take care of that but I never tried that, I prefer to build them manual, specially BLAS, where I use the ATLAS package to optimize the BLAS lib for the system, also easy to install

hope I could help

alexice

Boomeramg Options
 

Hi Thanks alot, because I finally set up PETSc with hypre support.

I want to use the Boomeramg (The Algebraic Multigrid Preconditioner) of Hypre to solve a problem in CFD (fully implicit N.S.). Do you have any experience using Boomeramg? What options are useful to experience?

Best Regards

Soheyl

Re: Boomeramg Options
 

Sorry to say, I do not use the Boomeramg in Hyre. I use the Pilut for my problems, which are glaciers.

I hope somebody out there can help.

anyway good luck with your model

alexice

Re: Boomeramg Options
 

Does anyone has previous experiences in using PETSc to solve Poisson equation (or incompressible NS) with multigrid.

petsc and multigrid
 

Does anyone has previous experiences in using PETSc to solve 3D FEA with multigrid? Whether or not the multigrid is petsc' or external software?

AMG for implicit Navier-Stokes
 

If you are solving compressible flow, multigrid might work fine, but for incompressible flow, your matrix will be indefinite. No black-box preconditioners (short of a direct solve) work well for indefinite problems. You will have to exploit the problem structure, but multigrid will be very useful for the definite parts inside your preconditioner. I often find ML (-pc_type ml) to be better for vector-valued problems, but it's very problem specific. See http://www.cfd-online.com/Forums/mai...tml#post210577 for more comment on this.

If you are solving 3D problems using BoomerAMG and having memory issues, try these options http://lists.mcs.anl.gov/pipermail/p...ry/000001.html

How do you utilize hypre as a PC of PETSc KSP? Is it necessary to modify original program, or only utilize PETSc's runtime option?

Configure PETSc with (or give path to your already installed version) and run with It is really rare that you have to modify code when you change preconditioners (i.e. the preconditioner must require extra information, most third-party preconditioners are purely algebraic).

My problem lies in moving flapping airfoil, incompressible flow using fractional step and structured grid. i've managed to use PETSc and hypre to solve my problem.

For the momentum eqn, I use KSP BCGS. For the poisson eqn, it's faster to use BoomerAMG under PETSc. Or better still, if it's structured grid, you can also use hypre standalone and use its solvers like SMG etc. It's even faster.

Hope it helps.

Thank you!
Beause I need solve eigenvalue,I will try SLEPc package!Hope you can help me! Thank you again!

Sorry for being pedantic, but you must be using BCGS with some preconditioner for the momentum equation and some Krylov method with BoomerAMG for the Poisson equation. Naming one without the other isn't very useful. If you insist on naming just one, name the preconditioner.

If you have a structured grid, you can use PETSc's DMMG which may be faster than SMG and is definitely more flexible.

AMG using Hypre
 

Dear Soheil,

Please tell us about your experience of using hypre for AMG. Did you use it for momentum equation as well? I thaught you use it for CFD.

Kind Regards,

Farshad