Assertion for "force_consistent=True"

[Georg Kastlunger]

We should add an assertion or exception if the user loads trained amp files, where the force_consistent=True keyword has been used, but does not explicitly give the keyword when restarting and viceversa.

This could be done in two ways:

• Add an assertion somewhere in AMP
• Change the call from ase.results so it creates an error when the keyword is given (or not) and the respective energy is not available.

[andrew_peterson]

Or more specifically, the error would be raised at the point when the user calls calc.get_potential_energy(force_consistent=False).

[Alireza Khorshidi]

force_consistent is no where inside the Amp code, and I am not sure if it is accessible. I wonder if we should add a new method like

def get_potential_energy(self, atoms, force_consistent=False):
    self.update(atoms)
    if force_consistent:
       ....
    else:
        raise RuntimeError('set force_consistent=True')

from https://github.com/qsnake/ase/blob/master/ase/calculators/general.py.

[andrew_peterson]

It's been a while since we discussed this, but I think the idea is that users could choose to train with force-consistent or 0-Kelvin energy. That is, we might add a force_consistent keyword to Amp.train. Then we need to have the force_consistent keyword in energy and force call methods, and raise an error if it doesn't match what was chosen during training. Does that make sense?

[Alireza Khorshidi]

The first part seems simple, we just add a force_consistent keyword to Amp.train and propagate it to the model/__init__.py and add it to image.get_potential_energy.

However, what I am confused about is the second part where we need to access force_consistent keyword fed by the user when calling

calc = Amp(...)
image.set_calculator(calc)
image.get_potential_energy(force_consistent=True)

(and then compare it with the value saved in calc.amp). Let's discuss it more in person, if what I said is not clear.

[andrew_peterson]

The force consistent keyword is already there through the ASE calculator. See this link.

So we just need to work within this framework, I would think.

Note I have a patch in progress for ASE that will make that force_consistent keyword work better. Perhaps we just want to put this issue on hold until after v0.5 since we're not using it at the moment.