❓ [QUESTION] Colab tutorial

[knc6]

Thanks for developing nequip and sharing the colab tutorial, which is very helpful.

The notebook mentions "We see that the model has converged to an energy accuarcy < 1meV/atom and a force accuracy of approx. 40 meV/A within 5 minutes and trained on only 100 samples."

1. Its unclear to me which number is referred to in this statement. Is it the e_mae e/N_mae?

2. Is there any config file to reproduce the publication results from the original nequip paper? Perhaps this yaml file?

[Linux-cpp-lisp]

Hi @knc6 ,

Thanks for your interest in our code!

1. e/N_mae is the per-atom error (i.e. energy units/atom). Quoting the notebook:

   The units of the reference data are in kcal/mol and A. If you're more familiar with eV, remember 1 kcal/mol is chemical accuracy and is approximately 43 meV

   You can see that 0.0225 kcal/mol * 43 meV / (kcal/mol) is about 1 meV.

2. See https://github.com/mir-group/nequip-input-files. But as that README says, new applications (and not exact reproducing) should use newer versions of the code and take configs/example.yaml as a starting point.