Choosing the best iterative method / Achieve convergence

Numerical methods and mathematical models of Elmer
Post Reply
RaJa
Posts: 77
Joined: 22 Oct 2014, 09:48
Antispam: Yes

Choosing the best iterative method / Achieve convergence

Post by RaJa »

I usually use iterative methods to solve my FEM-problems. Not only because I lack the math libraries for direct solvers, but also because you can run iterative methods easily on multiple cores. However, finding the right solver settings is often more trial and error than actual knowing. Especially, the best settings for the method and the preconditioner to achieve convergence.

Is there any cookbock which methods fits which problem? Or what to do when the problem does not converge. I am assuming that there is no general rule, but lets speak from experience.

Maybe you can post your tricks?
How do you solve convergence problems?
Do you have a "standard" iterative method that you prefer?

Thanks,
Rainer
mzenker
Posts: 1999
Joined: 07 Dec 2009, 11:49
Location: Germany

Re: Choosing the best iterative method / Achieve convergence

Post by mzenker »

Hi Rainer,

I also have that problem from time to time, and I would also wish to have a "cookbook" guiding me through the dungeon of simulation settings...
There are some descriptions and hints in the Elmersolver manual, abnd some hints spread all over this forum. Also there are lecture notes by Adam Powell containing some interesting information, see viewtopic.php?f=3&t=914&p=3016#p3016.
But a all that is not really the "cookbook" we would like to have available.
If you find something, let us know... ;)

Matthias
Post Reply