Can New Versions of COMSOL Improve Solver Efficiency and User-Friendliness?

[feynman1]

Do new versions show advantages in computing capability than old ones? For instance, could a solution not converged in an old version become converged in a new one?

 

[f95toli]

I think you might need to be bit more specific.
The reason for why a solution does not converge is usually because the mathematical problem is badly formulated, not because of a problem with the software you are using (unless of course there is a bug in the software, but that is very rare) .

That said, sometimes the lack of convergence is because you are using the "wrong" solver or the wrong type of mesh and newer versions of COMSOL have more options when it comes to the choice of solver and meshing algorithm, so it that sense a new version might be "better" for a given problem.

However, I would be very surprised if a model that did not converge on one version would suddenly converge on a newer version without any changes being made.

 

[feynman1]

yes, I mean in case a new version has a better solver, whether or not problems are more easily converged.

 

[f95toli]

Again, it depends on the problem. I don't think there is a such as thing as a "bad" solver in COMSOL; you just have to pick the right solver for your problem.
There are multiple solvers available in COMSOL plus a variety of meshing algorithms and different boundary conditions. if your problem falls neatly into one of the pre-definied categories you can usually use the default settings; but if not you need to think a bit about what you are trying to do and in some cases you need to consider which solver is best for you particular problem.
That said, in my experience most of the issues you come across are related to the meshing and the choice of boundary conditions. Again, this is not a problem with COMSOL as such; it is just the nature of FEM.

 

[feynman1]

usually solvers are chosen by COMSOL by default. most users have no knowledge of how to choose solvers. Will new versions make more foolproof or user friendly solvers?