Hi
first of all, COMSOL might find a solution with a coarse mesh, but it can be all wrong, if your mesh is too coarse. You need to have a mesh size that is sufficiently dense to catch the gradients of your dependent variables, and in case of waves that you have at least 5 or more mesh elements per period, in the material. Else you are under sampling and your results will be wrong (even if COMSOL sometimes might converge.
It's like aliasing in sound digital recording, a pure numerical sampling theory effect.
What is tricky is that often the gradient opf the dependent variables are very steep for given initial conditions, so often you need very fine mesh only very locally, and in time dependent , often only for the fist steps (typically for diffusion from a step concentration or thermal boundary limit. Then its worth to cheat a bit and introduce a >0 concentration or temperature locally close to the boundary with the initial conditions, it helps for the convergence. Often this is more critical for concentrations than for T as T is offset by 273°K most of the time, while the concentration is =0 as default initial value, and any c<0 makes people and COMSOL to scream ;)
--
Good luck
Ivar
first of all, COMSOL might find a solution with a coarse mesh, but it can be all wrong, if your mesh is too coarse. You need to have a mesh size that is sufficiently dense to catch the gradients of your dependent variables, and in case of waves that you have at least 5 or more mesh elements per period, in the material. Else you are under sampling and your results will be wrong (even if COMSOL sometimes might converge.
It's like aliasing in sound digital recording, a pure numerical sampling theory effect.
What is tricky is that often the gradient opf the dependent variables are very steep for given initial conditions, so often you need very fine mesh only very locally, and in time dependent , often only for the fist steps (typically for diffusion from a step concentration or thermal boundary limit. Then its worth to cheat a bit and introduce a >0 concentration or temperature locally close to the boundary with the initial conditions, it helps for the convergence. Often this is more critical for concentrations than for T as T is offset by 273°K most of the time, while the concentration is =0 as default initial value, and any c<0 makes people and COMSOL to scream ;)
--
Good luck
Ivar