Closed cortner closed 4 months ago
It seems we have a similar situation here as with AtomsBase - a debate whether AtomsBase and AtomsCalculators should be purely interfaces, or whether they should also provide reference implementations.
We are such a small community that I think we should also provide reference implementations so that we minimize duplication or work. But it is obviously not my decision.
moved to AtomsCalculatorsUtilities
What should be in EmpiricalPotentials and what should be in AtomsCalculators?
For example, I would like to implement a new model that is specifies a site potential. In order to leverage the generic implementation available in EmpiricalPotentials I now need to depend on all of EmpiricalPotentials. It may be better to have to rely solely on AtomsCalculators which gives a more light-weight dependency, but also doesn't pre-assume the nature of my model (it is not an empirical potential but an ML potential).
I don't think there is a huge urgency to decide on this organizational question, since we are still in the experimentation phase. But it would be good to jot down some thoughts.
Maybe spinning out a package such as
InteratomicPotentialsCore
or something along those lines?