Discussion Forum

Hi

I do not believe there is any other log file concerning the model solver than what is stored under the solver sub-node

For frequency sweep solving the most common error is no damping in the system and you get an numerical overflow at the resonance frequency if you (or the solver stepping) hits the resonance too well.
Such errors can be improved, i.e. doing a scan "up" and a scan "down" in frequency around the main resonance. At least you get the start of the slope.
Another way is to manually play with the stepper resolution of the solver around the resonance (identified by some eigenfrequency analysis, if this makes sense) by defining the solver steps manually and enforcing an intermediate or even strict stepping mode.

The best way is to add some damping, but what is reasonable damping ? this is often an unknown parameter, but in all cases there are damping

--
Good luck
Ivar

Thanks Ivar,
I do not think it is due to resonance phenomenon. It happens in other frequencies too and I have not performed parametric study. It is a complex model and most probably there is some thing wrong in the model. That's why I was expecting to have more details to find the source of the error.

Hi

Well I do not know about any other place than the log section of the solver, on some solver nodes there is a log: normal verbose, but it hardly adds more to the log file.

If you have a complex model then indeed it's more difficult to find. We cannot help out here without more info on the model, but often "support" has some good ideas, if you can send them the model.

If you have many physics its often worth to try to solve just 1 first (or one by one) and then add more and more physics

--
Good luck
Ivar

Hi Ivar,
I have already started debugging step by step. But it would be nice to identify the source of errors immediately.

Hi

I too find the error messages rather cryptic, but I have also noticed that COMSOL has managed to give more and more details over the years and in the newer versions.

But solver errors are not easy to backtrack to "a" particular issue, its often a very subtle combination of issues, each taken individually are OK, but once combined it fails. So my experience is that you have to go the "hard" way and analyse your model carefully step by step.

One very good debugging tool is the *Plot While Solving* feature in the solver node, but often you must turn on also the lowest level node Plot options to catch all solver step results and not only the main stored steps. You can combine several plots into one windows by using a "deformation" sub node with a fixed value (see the Postprocessing videos on their web site). What might take some tries though, is to identify which of the many dependent variables to track (or combinations thereof).

--
Good luck
Ivar