Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Very big difference in the results between direct and iterative solvers.
Posted 30 lug 2015, 04:32 GMT-4 Low-Frequency Electromagnetics, Mesh, Studies & Solvers 2 Replies
Please login with a confirmed email address before reporting spam
Hello,
I am trying to solve a 3D Model using the MEF physics in the frequency Domain.
With a rather coarse mesh I am able to solve the model (DoF~1e6) with the direct solver (PARDISO) and the global results (resistive losses) are what I expect.
Now when I try to solve the same model with an iterative solver instead, the results vary very much from what I expect (factor~1e16) and the field distribution is totally wrong.
I already tried to lower the relative tolerance to 1e-6 and also tried different iterative solvers with and without preconditioning. But it didn't really help.
The reason I want to use a iterative solver is because I need to refine the mesh and the RAM on my PC is limited to 96GB. The direct solvers become very slow as soon as the RAM is fully used.
Do you guys have some tips/help ?
I am trying to solve a 3D Model using the MEF physics in the frequency Domain.
With a rather coarse mesh I am able to solve the model (DoF~1e6) with the direct solver (PARDISO) and the global results (resistive losses) are what I expect.
Now when I try to solve the same model with an iterative solver instead, the results vary very much from what I expect (factor~1e16) and the field distribution is totally wrong.
I already tried to lower the relative tolerance to 1e-6 and also tried different iterative solvers with and without preconditioning. But it didn't really help.
The reason I want to use a iterative solver is because I need to refine the mesh and the RAM on my PC is limited to 96GB. The direct solvers become very slow as soon as the RAM is fully used.
Do you guys have some tips/help ?
2 Replies Last Post 3 ago 2015, 04:25 GMT-4
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi Alexander,
The MEF physics requires gauge fixing when using a direct solver so I guess you used that? When using the default iterative solver, it is generally better not to use gauge fixing. Setting up iterative solvers manually requires a lot of expertise so if it did not work with the default iterative solver or if you got an unphysical solution I strongly recommend that you contact support. Or if you post the model here, I can have a look at it.
--
Magnus
The MEF physics requires gauge fixing when using a direct solver so I guess you used that? When using the default iterative solver, it is generally better not to use gauge fixing. Setting up iterative solvers manually requires a lot of expertise so if it did not work with the default iterative solver or if you got an unphysical solution I strongly recommend that you contact support. Or if you post the model here, I can have a look at it.
--
Magnus
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi Magnus,
first of all, thank you for answering my question.
We both already met on the COMSOL conference 2011 in Milan. We both took a look at my model of a 2d asynchronous motor model (coupled to a stator and rotor circuit) because the solving time was so huge. Maybe you remember ; )
On topic:
Yes, I use gauge fixing whenever I need the MEF physics interface.
I didn't know it should be restricted to direct solvers.
So, I tried to solve the model without gauge fixing with an iterative solver. Unfortunately the results are still not satisfying, but I have to take a closer look at this and do some tests with different solver settings etc.
Indeed, I already simulated and optimized a small transversal flux motor (3D, magnets as flux source) succesfully with the default solver settings (.mf physic interface, iterative solver).
The problem of my current model might be that I'm using a rather theoretical approach (Surface Currents as flux source) and that the default solver settings might be for more practical models...?
This is a very nice offer. Unfortunately this model is a part of my doctorate thesis and I'm not sure about sharing it before my thesis is published. I could think of sharing a simplified version.
Greetings
first of all, thank you for answering my question.
We both already met on the COMSOL conference 2011 in Milan. We both took a look at my model of a 2d asynchronous motor model (coupled to a stator and rotor circuit) because the solving time was so huge. Maybe you remember ; )
On topic:
The MEF physics requires gauge fixing when using a direct solver so I guess you used that?
Yes, I use gauge fixing whenever I need the MEF physics interface.
When using the default iterative solver, it is generally better not to use gauge fixing.
I didn't know it should be restricted to direct solvers.
So, I tried to solve the model without gauge fixing with an iterative solver. Unfortunately the results are still not satisfying, but I have to take a closer look at this and do some tests with different solver settings etc.
Setting up iterative solvers manually requires a lot of expertise so if it did not work with the default iterative solver or if you got an unphysical solution I strongly recommend that you contact support
Indeed, I already simulated and optimized a small transversal flux motor (3D, magnets as flux source) succesfully with the default solver settings (.mf physic interface, iterative solver).
The problem of my current model might be that I'm using a rather theoretical approach (Surface Currents as flux source) and that the default solver settings might be for more practical models...?
Or if you post the model here, I can have a look at it.
This is a very nice offer. Unfortunately this model is a part of my doctorate thesis and I'm not sure about sharing it before my thesis is published. I could think of sharing a simplified version.
Greetings
Hi Alexander,
The MEF physics requires gauge fixing when using a direct solver so I guess you used that? When using the default iterative solver, it is generally better not to use gauge fixing. Setting up iterative solvers manually requires a lot of expertise so if it did not work with the default iterative solver or if you got an unphysical solution I strongly recommend that you contact support. Or if you post the model here, I can have a look at it.
--
Magnus