[oevermann]

I'm having two questions here:
1) is the implemented algebraic multigrid solver (AMG) really limited to symmetric matrices? In other words: which AMG exactly is implemented (reference)?. In other projects I am using the (very fast) AMG of the hypre package (www.llnl.gov/CASC/linear_solvers) which does not have this limitation. For better performance it might be usefull to have an AMG preconditioned BICGSTAB for the non-symmetric problems.
2) is there a coupled solver comparable to the solver in CFX?

[hjasak]

Heya,

1) Yes. The solver is an agglomerative version of AMG (have a look at Klauss Stuben's references) with scaling proposed by Joe Maruszewski. The scaling is wrong for non symmetric matrices (in my opinion) and I have therefore disabled it. Additionally, the AMG implementation I did does not allow pre-smoothing sweeps on the basis of cost/benefit (you would have to re-calculate the residual before the restriction).

AMG is only really good for elliptis problems; for normal transport equations, CG-based solvers converge in a couple of iterations and the benefit from another solver would be minimal.

2) Currently, no. I am playing with block matrices and block solution but for various reasons this is not a priority. If I can get myself another PhD student to work on this, we might get it sooner rather than later.

Enjoy,

Hrv

[oevermann]

Thanks for the info! I agree with your point concerning AMG, and I normally use a BICGSTAB
preconditioned by an AMG.

Regards

M