Hi Elmer,
I'm currently trying to use non direct methods to resolve the stokes linear system, since I start simulating 3D problems. As it noticed in ElmerSolver Manual, the choice of an iterative solver coupled with a preconditioning strategy is problem dependent, so you have to apply the try/error method.
I'm simulating a 3D ice cap flowing down under the gravity. The basal surface is either in contact with a rigid substratum (I use here a Dirichlet condition) or floating above the sea (a neumann BC is used here). The top surface undergoes a zero pressure.
When I use the condition "Noslip wall BC" for both lateral walls, a good "iterative" strategy is the Biconjugate Gradient Stabilized (BiCGStab) associated with ILU0 preconditioner.
But when I don't use any friction for the lateral walls, BiCGStab option works well with ILU5 or ILU6, depending on the size of the linear system. I'm happy it works well in that latter condition, but the computation cost is important with this preconditioner, and I wonder if I couldn't use more efficient options.
A non exhaustive .sif file and the ice geometry are attached at the end of the post
I saw it is also existing some Multilevel methods to solve large linear systems. That is a way I didn't explore yet. There is a lot of various problems solved with Elmer, but I only found a few questions about non direct methods in the forum. So this post is mainly to collect some clues about the non direct method to use, related to the above described problem.
Cheers all
lionel