Convergence of Whitney-AV Solver

Numerical methods and mathematical models of Elmer
Post Reply
carstenp
Posts: 37
Joined: 17 Oct 2012, 23:17
Antispam: Yes

Convergence of Whitney-AV Solver

Post by carstenp »

Hi everyone,

I was playing a little with the Whitney AV Solver in the last days and I am really impressed. However I noticed that convergence for a practical model size (> 10^6 elements) is hardly reachable without preconditioning (as expected for a curl curl matrix). Unfortunately, even with ILU I am not able to converge to a solution. Using the same model with a coarser mesh on the other hand works fine. So currently I am trying the TFQMR Solver without preconditioning. Has anybody a better suggestion or has experience with Elmers Whitney AV Solver and larger models and convergence issues?

carsten
Juha
Site Admin
Posts: 357
Joined: 21 Aug 2009, 15:11

Re: Convergence of Whitney-AV Solver

Post by Juha »

Hi,
If you haven't done so already, you could try using BiCGStabL instead of TFQMR, also
adding "BiCGStabL Polynomial Degree=4" might help(?) GMRES with the
"Linear System Gmres Restart=n" is also an option.

Using ILU might actually make things worse.
Regards, Juha
carstenp
Posts: 37
Joined: 17 Oct 2012, 23:17
Antispam: Yes

Re: Convergence of Whitney-AV Solver

Post by carstenp »

Hi Juha,

BiCGSTABl seems to do the trick. The simulation is still running but slowly converges. I post an update when it is done.

Thank you!

carsten
carstenp
Posts: 37
Joined: 17 Oct 2012, 23:17
Antispam: Yes

Re: Convergence of Whitney-AV Solver

Post by carstenp »

Hi Juha,

the simulation did converge so BiCGStabL works. Just one more questions: Might the convergence rate be further improved by a higher polynomial degree?

carsten
Post Reply