When faced with 100K DOF comsols defeault solver is a 1 or 2 level multigrid with GMRES as iterative and PARADISO as linear solver.
Some papers on CFD and mass trasfer (such as the one by Birgit Stute, 2013) report that solves lik the BiCGStab work much better.
I'm really curious to hear your ideas and thoughts on what changes you made to the default solver (or how you made a custom one) and how this improved the convergence or computatinal time of the model.
I know that each problem is different and that a solution strategy cannot by blindly copied but perhaps your methods can help me on my way.
Your ideas and thoughts are highly appreciated.
Regards,
Victor
[Edit: In the first sentence it should read: Segregated GMRES solver with multigrid preconditioner, SCGS for u/p and SORline for kappa/epsilon)]
Some papers on CFD and mass trasfer (such as the one by Birgit Stute, 2013) report that solves lik the BiCGStab work much better.
I'm really curious to hear your ideas and thoughts on what changes you made to the default solver (or how you made a custom one) and how this improved the convergence or computatinal time of the model.
I know that each problem is different and that a solution strategy cannot by blindly copied but perhaps your methods can help me on my way.
Your ideas and thoughts are highly appreciated.
Regards,
Victor
[Edit: In the first sentence it should read: Segregated GMRES solver with multigrid preconditioner, SCGS for u/p and SORline for kappa/epsilon)]