ConvergenceOccurred?

[Alireza Khorshidi]

Currently with training only energies, ConvergenceOccured is true if both energyconverged and maxresidconverged happens.

How do we want to extend it to forces? Two ways can be used:

1- energyconverged and forceconverged as before.

2- energyconverged, maxenergyresidconverged, forceconverged, and maxforceresidconverged.

Which way would you suggest to proceed?

[andrew_peterson]

I think we need to figure out a way to have the user able to cleanly specify which of these they want to be active. For example some users may only want to set a max force residual and don't care about energies.

I am not looking at the code right now, but a possibility is the user gives a dictionary of convergence criteria, something like:

convergence = {'energy_SSR': 0.01,
                          'energy_maxresidual': None,
                          'force_SSR': 0.04}

Any values they don't supply are assumed to be None. What do you think?

[Alireza Khorshidi]

Good idea! I added a similar convergence keyword in the commit 5644c7b. Of course, instead of SSR we have rmse.