4. Tests

Our continuous integration (CI) tests are implemented using GitHub Actions. This involves two main tests for consistency and validation:

1. Consistency of linear fits. Here we check that our simple examples examples/Ta_Linear_JCP2014 produces the same fitting coefficients from a previous standard in 20May21_Standard.

2. Proper implementation of neural network forces. This test involves calculating forces via finite difference and comparing to the analytical forces calculated via automatic differentation that we fit to. This ensures that neural network forces are properly coded. Finite difference and analytical forces in an interatomic potential should agree within a small amount such as 1e-3 force units.