Hi,
I am solving a high-frequency 3D acoustic problem with millions of DOFs. Documentation suggests to use a GMRES solver with GMG as a preconditioner. However, it gives very little details on the parameters to select. The default COMSOL parameters do not seem to do a great job at solving my problem. Could someone having experience with GMRES solvers tell me what values to input for:
Number of iterations before restart:
Factor in error estimate:
Also, in the GMG preconditioner, I notice that when the number of iterations is too small (e.g. 2), the convergence rate is very slow and when the number of iterations is too large (e.g. 10), then COMSOL indicates that the problem is ill-conditioned. I usually select 5 iterations, a V-cycle, and 3 multigrid levels. Does it seem reasonable? Should I check or clear the "assemble on all levels" box? Could someone tell me how to tune this to have optimum preconditioning and so fastest convergence?
Thanks!
Marcel
I am solving a high-frequency 3D acoustic problem with millions of DOFs. Documentation suggests to use a GMRES solver with GMG as a preconditioner. However, it gives very little details on the parameters to select. The default COMSOL parameters do not seem to do a great job at solving my problem. Could someone having experience with GMRES solvers tell me what values to input for:
Number of iterations before restart:
Factor in error estimate:
Also, in the GMG preconditioner, I notice that when the number of iterations is too small (e.g. 2), the convergence rate is very slow and when the number of iterations is too large (e.g. 10), then COMSOL indicates that the problem is ill-conditioned. I usually select 5 iterations, a V-cycle, and 3 multigrid levels. Does it seem reasonable? Should I check or clear the "assemble on all levels" box? Could someone tell me how to tune this to have optimum preconditioning and so fastest convergence?
Thanks!
Marcel