Please login with a confirmed email address before reporting spam
Posted:
8 years ago
14 mars 2016, 16:29 UTC−4
If you are using MW field simulation, then use GMRES solver which is much faster than BiCGtab. Check for mesh also. Enable mesh control in EM wave subtab.
If you are using MW field simulation, then use GMRES solver which is much faster than BiCGtab. Check for mesh also. Enable mesh control in EM wave subtab.
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
14 mars 2016, 20:14 UTC−4
Thanks your answer Birla. Actually I tried GMRES solver but it also does not converge. Regarding mesh, I used physics-controlled mesh then checked element quality and it seems okay. I guess the problem may be about geometry therefore I still working on it.
Thanks your answer Birla. Actually I tried GMRES solver but it also does not converge. Regarding mesh, I used physics-controlled mesh then checked element quality and it seems okay. I guess the problem may be about geometry therefore I still working on it.
Robert Koslover
Certified Consultant
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
14 mars 2016, 22:47 UTC−4
I've had similar problems with BiCGStab. It seems to help to use at least quadratic-order elements (if you have enough memory) and to adjust your geometry and meshing to ensure high element quality. Alternatively, consider using linear-order elements and PARDISO, if you have enough memory.
I've had similar problems with BiCGStab. It seems to help to use at least quadratic-order elements (if you have enough memory) and to adjust your geometry and meshing to ensure high element quality. Alternatively, consider using linear-order elements and PARDISO, if you have enough memory.
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
15 mars 2016, 09:36 UTC−4
Hi Koslover,
Yes you are right direct solver can overcome this issue however, unfortunately the memory ( 8 GB) is not able to handle number of degree of freedom for very high mesh quality that is why I keep it about 0.77 in average .
Actually there is a trivial trick to make BiCGStab converge that if you rearrange factor in error estimate low enough such as 50 instead of 400( default) it converges but it is not a precise solution.
Anyway I'll try with 16GB memory then I'll update news here. Thank you.
Hi Koslover,
Yes you are right direct solver can overcome this issue however, unfortunately the memory ( 8 GB) is not able to handle number of degree of freedom for very high mesh quality that is why I keep it about 0.77 in average .
Actually there is a trivial trick to make BiCGStab converge that if you rearrange factor in error estimate low enough such as 50 instead of 400( default) it converges but it is not a precise solution.
Anyway I'll try with 16GB memory then I'll update news here. Thank you.
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
17 mars 2016, 07:11 UTC−4
*updates
for memory issue, 16 GB is neither enough. Therefore I reduced all sizes then solver convergences.
But another problem is occured such that when I use 3D cartesian PML with size of width is .5[um], again I face convergence problem for BiCGStab.
Instead, If I use only PEC and no interior port, light propogates through the waveguide but not well guided.
*updates
for memory issue, 16 GB is neither enough. Therefore I reduced all sizes then solver convergences.
But another problem is occured such that when I use 3D cartesian PML with size of width is .5[um], again I face convergence problem for BiCGStab.
Instead, If I use only PEC and no interior port, light propogates through the waveguide but not well guided.
