Discussion Forum

Yuki,

I changed to the direct solver and added Gauge Fixing to the physics. Then it solves in Comsol 5.5. I don't have 4.n versions available here any more. So it is not an issue of the coil feed. If you need to use the iterative solver (e.g. for memory reasons) then you may need to tweak the mesh.

Cheers Edgar

Dear Edgar,

Thank you for your quick reply. I had thought I can't get some replies, so I'm happy!

I use direct solver and added Gauge Fixing for A- Field. But, the out of memory error happened just like you said. Then I coarsened mesh. And an another error happened "Failed to find a solution. The relative error (1.9e+005) is greater than the relative tolerance. Returned solution is not converged. - Feature: Steady Solver 1 (sol3/s1)" What does this means?

Is my understanding correct with that "Gauge Fixing" means Gauge Fixing for A-Field? I haven't be able to understand about effect of "Gauge Fixing for A-Field". (It is written only about 2D model at the help.) Should I apply it to all domain? What value should I assign to ψ0? What means a check box?

I'm sorry I have many questions. Maybe, I can use comsol 5 a few weeks later. So, Could you send your mph file? For reference.

appendix : I tried changing the direct solver to PARDISO. Then I got the following error: "Singular matrix"

Dear Yuki,

you modified exactly the way I did. Gauge Fixing for A-field takes care of the fact that the A-field is not unique. Mostly you can just use the default configuration in all domains.

I played a little with your iterative setup and with a little mesh refinement on the coil domain it converges. It took 450 iterations and almost 2 hours on my machine, but needs less than 13 GB then. Maybe you can tweak that a little. The screen shot shows the only modification I did to the mesh. I didn't touch the solver. This is the model you posted first.

Cheers Edgar

Dear Edgar,

Thank you for your teaching. I tried to change the mesh like your screen shot. But It did not converge. I will try other meshes.

Is my understanding correct with that, In iterative solvers, finer meshes tend to converge?

And, is there any other effective way to converge?

My purpose is to compute more complex 3D models. However, this relatively simple model does not converge. Do I need to change something fundamentally to compute complex models? For example, machines, calculation methods, modeling methods, etc.

Thanks,

Every model is different and may require different approaches. Particularly as a beginner you must try many options. It is often useful to find out a good approach on a smaller version of the model and then ramp it up, once the you have good mesh and solver settings.

More memory gives you more options.

In your case you may also try to implement it in mf. You loose the capacitive effects by this, but if you are only interested in inductance that doesn't matter.