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.

Hasten electromagnetic simulation by re scaling the dimensions

Mohammad Parvinnezhad Hokmabadi

Please login with a confirmed email address before reporting spam

Hi everyone,

I am simulating a electromagnetic structure in THz frequencies (0.1 - 2 THz).
When I am doing simulation in higher frequencies with smaller dimensions it runs fast but in lower frequencies like around 1 THz it runs very slow, because of corresponding larger dimensions that I need in lower frequencies.

Is there any way to speed up simulations in even lower frequencies with corresponding larger dimensions by for example normalizing the dimensions or something like that?

Because for example I think in mechanical student deal with larger dimensions around cm, while I am doing simulation with micrometer dimensions so they should use some method to reduce running time unless it will take a few weeks when mine takes 48 hours.

I appreciate if anyone could share his experience with me to help.

Regards,
Mohammad

3 Replies Last Post 9 giu 2012, 05:37 GMT-4
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 8 giu 2012, 14:46 GMT-4
Hi

it's not "just" the size of the models, it has also to do with the variations or gradients of the dependent variables.
In structural we have seldom waves, and if so these are of rather large wavelength compared to the structures (I'm thinking of bending waves and not compresion or acoustic waves first of all).
For RF you have the oscilaltory dependence of the EM wave, that in RF is to be small compared to the structure, and it's not the structure that is limiting but mostly the EM wavelength that must be resolved by 5-10 elements per periode.
Then depending on this ratio structue to EM wavelength size you might be resolving well both the structure and/or the wavelength, if you go to the regime where the wavelength and structure become similar you enter into the grey zone where neither ACDC nor RF are perfectly adapted, and you need either a very fine mesh to resolve the structure or to resolve the EM wave.

I believe that is the main reason why the convergence is slowing down, you probably need to play more with your mesh densities to adapt them better to both structure and EM waves to speed up the convergence, which in other words mean knowing the solution before you have solved it, something not necesarily obvious ;)

--
Good luck
Ivar
Hi it's not "just" the size of the models, it has also to do with the variations or gradients of the dependent variables. In structural we have seldom waves, and if so these are of rather large wavelength compared to the structures (I'm thinking of bending waves and not compresion or acoustic waves first of all). For RF you have the oscilaltory dependence of the EM wave, that in RF is to be small compared to the structure, and it's not the structure that is limiting but mostly the EM wavelength that must be resolved by 5-10 elements per periode. Then depending on this ratio structue to EM wavelength size you might be resolving well both the structure and/or the wavelength, if you go to the regime where the wavelength and structure become similar you enter into the grey zone where neither ACDC nor RF are perfectly adapted, and you need either a very fine mesh to resolve the structure or to resolve the EM wave. I believe that is the main reason why the convergence is slowing down, you probably need to play more with your mesh densities to adapt them better to both structure and EM waves to speed up the convergence, which in other words mean knowing the solution before you have solved it, something not necesarily obvious ;) -- Good luck Ivar

Mohammad Parvinnezhad Hokmabadi

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 8 giu 2012, 15:41 GMT-4
Thank you,
Actually, I am using coarser mesh in both higher and lower frequencies.
My smallest dimension is a rod of 2 micrometers width for both frequencies (1THz or 10 THz.)
Frequency step sizes is same for both 1 and 10 THz. because the results are almost same and I need this step size too see the results for both.
the length of rod is bigger in 1 THz than 10 THz which makes structure larger and it takes a longer time to be simulated.
As far as I understand, you mean that I should define mesh by myself and because wavelenght for 1 THz is 10 times larger than 10 THz so I can make mesh larger at rods by a factor like around 10. It may work.
Is that right?

Regards,
Mohammad
Thank you, Actually, I am using coarser mesh in both higher and lower frequencies. My smallest dimension is a rod of 2 micrometers width for both frequencies (1THz or 10 THz.) Frequency step sizes is same for both 1 and 10 THz. because the results are almost same and I need this step size too see the results for both. the length of rod is bigger in 1 THz than 10 THz which makes structure larger and it takes a longer time to be simulated. As far as I understand, you mean that I should define mesh by myself and because wavelenght for 1 THz is 10 times larger than 10 THz so I can make mesh larger at rods by a factor like around 10. It may work. Is that right? Regards, Mohammad

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 9 giu 2012, 05:37 GMT-4
Hi

the mesh size must be matched to the
a) structure or domains you want to resolve, and
b) the the variations of the dependent variables within these domains
and choose whatever is smallest.

For a 1THz (what I would call the FIR region) in air (of epsilonr=1) you have a wavelength of about 300 um if I got it right.
Therefore the mesh sizes should normally never be bigger than about <30-60 um in air, but significantly smaller in high epsilon_r material.
If you have objcts of 2um you mesh must be smaller than this too to resolve any gradients in these structures. On the other side 2um is very small for a 300 um wavelength you are in the "grey zone" between ACDC (long wavelength compared to object sizes) and RF (small wavelength compared to object sizes)

--
Good luck
Ivar
Hi the mesh size must be matched to the a) structure or domains you want to resolve, and b) the the variations of the dependent variables within these domains and choose whatever is smallest. For a 1THz (what I would call the FIR region) in air (of epsilonr=1) you have a wavelength of about 300 um if I got it right. Therefore the mesh sizes should normally never be bigger than about

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.