Discussion Forum

Hi Community,
I am a naive user for COMOSOL...I use COMSOL 3.4 on a basic system (8GB RAM, WinX64 OS..)...I was trying to do a transient analysis of an obliquely incident (theta=30[deg]) sinusiodal EM-wave in a 3-layered waveguide, using the basic PDE model-->Transient Analysis....

I was trying to simulate my COMSOL model, for a fixed number of time step (which is obviously less than the actual number of time steps, taken by the incident EM-wave). Even though I was doing all the required calculations properly using MATLAB, still my solution was not converging. I was encountering the"cryptic COMSOL error message" while playing with my solver settings.

I searched in the COMSOL community for the different error messages, but none of them helped me to get a proper solution. Fixing one error lead to another...Finally, after a long duration of time, I am able to solve the problem..
I have attached both the dimensional, as well as, non-dimensional model of the geometry (3-layered slab waveguide), along-with the MATLAB (.m) code ia apdf-file...One thing should be kept in mind that the calculation outputs of MATLAB and COMSOL might differ in certain situations as a "relative error % isassociated between the two independent platforms".

I got some of the answers to my question, which I believe, every COMSOL naive user has to face...I have jotted down, as many as I could remember...Hope it helps us all...

Relative tolerance & absolute tolerance should always be lesser than the residual error. Otherwise the system won't converge..For example, if you are facing an error, rediual is smaller than the absolute tolerance, then you need to decrease the relative and absolute tolerance values. In conjunction, you need to modify the tolerance values for the type of linear solver you are using. If it is Direct(UMFPACK) then in the Linear solver settings change the tolerance aswell...Same should be followed for GMRES with one additional step of adjusting the "conditioner" as well.

Sometimes solutions may not converge and the exception can be due to the "problem in the symmetry"...error: time stepping has found a symmetric matrix or whatever... In those cases, in the drop-down menu of Matrix symmetry (solver parameters tab), change it to "Symmetric" from "Automatic"...Even though "Direct(UMFPACK) does not take advantage of symmetric matrices",but your solution will cover the required number of time stepping..Most commonly, the LRU problem and the out of memory problem can be solved by the use of "proper meshing", "choosing the correct linear solver", "calculating the time in solver parameter" and "modifying the tolerence"...But the default tolerence value in COMSOL is set in such a way that you dont need to alter it always..Instead for a transient analysis, you shuold double check the times in solver manager (time steps)...

If you think that, all that I blabbered, has some meaning then you can refer into the solution...

Lastly, i would like to thanks "Igor " for his active participation in this community...
Good work and best of luck..

Dipta R Roy

SHM Systems Lab
Dept. of Aerospace Engineering
Indian Institute of Science
Bangalore, India