search

BIMAU / transiflow

Implementations and continuation of some standard computational fluid dynamics problems using the finite volume method
https://transiflow.readthedocs.io/
Apache License 2.0
9 stars 12 forks source link

Continuation: smaller stepsize if solve goes bad? #4

Closed ipelupessy closed 2 years ago

ipelupessy commented 3 years ago

I think the interface.solve can go bad (ie takes too many iterations or doesn't converge) if the stepsize is too large? Is that correct? Maybe it is an idea to try: except around it and allow for a decrease in stepsize if this happens... Now the same thing will probably happen but very indirect (e.g. solveer iterations overflow -> the solve is bad -> hence the newton will not converge -> too many newton iterations -> stepsie decrease)

Sbte commented 3 years ago

This is usually a sign of an underlying problem, not something we can solve by adapting something.

Sbte commented 2 years ago

This should not be a problem in practice. The step size increases/decreases gradually, and will be adjusted accordingly in every step. If it goes from 3 Newton iterations to not converging at all, a smaller ds will not help.