Discussion Forum

Hi

could it be that you are hitting straight on a resonance, and your values grow out of bound for a "Double Float" ?

When I do modal sweeps, I always do my best to avoid direct vicinity of resonances, except if I have a lot of damping, which by default is seldom the case. This applies in all generality from Structural to ACDC to RF ...

--
Good luck
Ivar

Hi Ivar,

I have the same problem. I did the simulation of Photonic Crystal Fiber by using RF module. When I simulated my PCF model in frequency 2e13 Hz , it works. But, when I increased the frequency in 2e14 Hz, it did not work. What your opinion with this case?

Thank's for your attention.

Haris

Hi Ivar,

I mean that, when I increased the frequency in my simulation of Photonic Crystal Fiber, there is an error message

""Undefined value found.
- Detail: NaN_or_Inf_found_when_solving_linear_system_using_SOR
"
Do you have an idea for this problem?

Thank's

Haris

Hi

not really NaN is not a a number hence some undefined values, or underflow/overflow

Have you cheked that your emsh is compatible with the wavelength at those high frequencies ?

--
Good luck
Ivar

Hi Ivar,

I'm beginner, so i don't know How to check compatibility my mesh and frequency? Would you like share to me, How to check compatibiliy between mesh and frequency? If it is not compatible, Does the mesh must be changed?

thank's

Haris

Hi

I agree its difficult to learn all the underlaying hypothesis for all physics rapidly, it takes some time. First of all index the pdf files of the help and search in these with keywords, then take time to read (even acouple of times) the doc, particulrly the one related to your specific physics, i.e. RF.

Meshing is like digital sampling. most people have understood by now tha when digitising any time varying signal one must have enough samples to resolve correctly the highest frequency, otherwise one get aliasing. With FEm its the same. The mesh is sampling your domain and boundaries, on which are running functions that have gradients (slopes or even harmonic shapes) and these must be samples correctly. particularly in the time domain, when you can have very steep gradients along some boundaries, then the mesh will not correctly take into account these gradients and your results is wrong, or the solver gets lost, at best run for a looong time.

With RF the main issue is to say what is the wavelength of my EM wave in the different media (wavelength in air and glass are not the same because of the index of refraction of glass. So you need at least 5 ideally 10 mesh elements per spatial wave-length in all materials.

Now knowing your frequency and the medium (n) you can get the wavelength, find the shortest, and divide by 5 or 10: this is the max mesh size for RF ! you can also plot "h" over your meshed domains to see the actual values you have and compare to the desired one. You might end up with so many elements that you do not have enough RAM.
Then you must think how to reduce the size of your model, Use symmetry or perhaps a 2D-axi or 2DD instead of 3D etc

--
Good luck
Ivar

Hi Ivar,

Thank's Ivar, your explanation is very detail. It's reasonable, my meshing size isn't compatible for frequency 2e14 Hz. When I decrease the meshing size, it means that the number of elements increase and it's very hard for my computer. I think, I have to use the other technique.

Best regards
Haris

Hi Ivar,

I have simulated mode analysis of PCF (photonic crystal fiber), and I obtain the effective mode index is 1.459941 (6 decimals places after coma). I want to increase the accuracy of result, it means that I want to get the result 12 decimals places after coma. Do you know, how to do that?

Thank's
Haris

Hi Ivar,

I have simulated mode analysis of PCF (photonic crystal fiber) by using comsol 4.1, and I obtain the effective mode index is 1.459941 (6 decimals places after coma). I want to increase the accuracy of result, it means that I want to get the result 12 decimals places after coma. Do you know, how to do that?

Thank's
Haris

Hi

you can ask COMSOL to display all digits (check the doc, the preference and the full precision icon in the table view), and try to tweak your solver settings to continue solving to a higher resolution, but then you need also a more dense mesh, and be sure you have no irregularities.
Anyhow computing on todays PC's, even with double precision, gives only about 6 digits sure in good conditions, due to the way the binary system represents the "real" numbers.

Anyhow, I seldom trust more than 3 digits, a model is an approximation, no "true" value

--
Good luck
Ivar