Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

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.

Model convergence dependent on mesh quality, confused!

Please login with a confirmed email address before reporting spam

Hi all,

My COMSOL model only solves when the mesh size is "above" the predefined Normal size. When I increase the mesh resolution to "Fine" or higher, the step BiCGStab in the progress tab stalls at around 96%, and continues iterations forever until infinity trying to converge.

I'll describe the model briefly in this paragraph, and include an attachment of it. I'm simulating a very small, simple, conductive, flattened-conical tip (sliced in half vertically for symmetry) that radiates microwave (~6GHz) energy into the surrounding region via a coaxial lumped port. I'm using the Electromagnetic Waves, Frequency Domain physics module. The model behaves exactly as desired when the lower-limit of mesh quality is met, but endlessly iterates otherwise. I've tested varying the geometry and size of the tip to no avail. It seems that the minimum element size I can reach before causing non-convergence is about 3e-7 meters. (My model size is on the order of 1e-6 meters).

I'm so confused; there's obviously nothing wrong with the model setup/geometry/physics, because it solves for lower-quality meshes. Is this a computer memory problem? (I've got 8GB). Is there a lower-limit to element size / calculation in the [EMW, Frequency Domain] module that I've reached?

Thanks for the help.

Jon Michelson


4 Replies Last Post 10 juil. 2014, 10:32 UTC−4

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 8 juil. 2014, 03:19 UTC−4
Maybe you can indeed check the memory usage, if you start using swap the simulation speed will go down orders of magnitude.
Maybe you can indeed check the memory usage, if you start using swap the simulation speed will go down orders of magnitude.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 8 juil. 2014, 10:38 UTC−4
Thanks for the advice, Pieter.

I'm not sure it's a memory issue though; I've solved much denser meshes, with many more DOF, with ease before. And I'm not currently using swap, so the simulation speed shouldn't be drastically reduced. And it's not even that the speed is reduced either -- the simulation is actually failing to converge, not just taking a long time. The iteration count on BiCGStab continuously increases while percentage completion and convergence show no progress.

I've been searching around the internet and I can't even find much helpful information on using/implementing swap files either... would you know how I could use one, if I decided to give it a shot?
Thanks for the advice, Pieter. I'm not sure it's a memory issue though; I've solved much denser meshes, with many more DOF, with ease before. And I'm not currently using swap, so the simulation speed shouldn't be drastically reduced. And it's not even that the speed is reduced either -- the simulation is actually failing to converge, not just taking a long time. The iteration count on BiCGStab continuously increases while percentage completion and convergence show no progress. I've been searching around the internet and I can't even find much helpful information on using/implementing swap files either... would you know how I could use one, if I decided to give it a shot?

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 9 juil. 2014, 04:02 UTC−4
swap is on your hard disk which is much slower than your RAM so do not use it if you have enough RAM. I have never really understood about simulations not converging. Maybe use a different solver? Or ask the COMSOL helpdesk.
swap is on your hard disk which is much slower than your RAM so do not use it if you have enough RAM. I have never really understood about simulations not converging. Maybe use a different solver? Or ask the COMSOL helpdesk.

Robert Koslover Certified Consultant

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 10 juil. 2014, 10:32 UTC−4
This might help: If you don't have enough RAM to use a direct solver, try using GMRES along with the SSOR preconditioner. Also, if you aren't already doing so, setting the discretization to linear (the default is quadratic) will save a lot of memory.
This might help: If you don't have enough RAM to use a direct solver, try using GMRES along with the SSOR preconditioner. Also, if you aren't already doing so, setting the discretization to linear (the default is quadratic) will save a lot of memory.

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.