Bad convergence using BoomerAMG for system of equations and InterpType doesn't work

[sahar-shpitz]

Hello,

I'm trying to use BoomerAMG as a preconditioner to GMRES to solve the linear system resulting from compressible 2D flow. There are 4 unknowns (density, velocity x, velocity y, and energy), so the matrix is a 4x4 block matrix. I did use the following functions:

HYPRE_BoomerAMGSetNumFunctions(precond, 4);
HYPRE_BoomerAMGSetDofFunc(recond, dof_func);
HYPRE_BoomerAMGSetNodal(precond, 1);

However, I get bad convergence or no convergence at all (depending on how low I set HYPRE_BoomerAMGSetRelaxWt), while the regular Jacobi relaxation function that I wrote does converge. I suspect that the relaxation type set by the function HYPRE_BoomerAMGSetRelaxType does not use a 'block' relaxation method, with a small 4x4 dense matrix inversion. Is that true? If so, can I do something about it to get good convergence? Is there something that I'm missing?

I also tried HYPRE_BoomerAMGSetInterpType(precond, 10), but it doesn't work for some reason. It allocates all the memory on my computer and then crashes.

Thanks in advance,

Sahar

[victorapm]

Hi Sahar,

BoomerAMG alone is not suitable to your problem. One option may be MGR, which targets multiphysics problems. Is your code open-source?

[sahar-shpitz]

Hi Victor,

Thank you for your reply.

The code is a research code and is not open source at the moment. Would it help to print out A and b?

I can't find much documentation on MGR. Are there any examples or explanations on how to use it?