Clarification on energy scale & units in the tutorial si-traj.xyz

[seongboo]

Hi

While following the tutorial, I noticed that the total_energy values in the example trajectory file (si-traj.xyz in the tutorial; possibly named si-traj.xyz) look much larger than my first-principles results for Si. Concretely, the magnitude is more than 10× higher than the energies I computed from DFT.

To make sure I’m comparing apples to apples:

My DFT energies are in eV.

• My reference values are typically per atom (or sometimes reported as cohesive energy per atom).

• The tutorial file reports a field named total_energy (or energy, depending on the header).

Could you clarify the following?

1. Units: Is total_energy stored in eV, or another unit (e.g., Hartree, kcal/mol)?

2. Normalization: Is it per structure (sum over all atoms) or per atom?

Thanks in advance for clarifying the intended unit, normalization, and reference—this will help me prepare my Si dataset correctly for training!

[cw-tan]

Hi @seongboo

• The energy label in any dataset typically refers to the total energy of the whole structure (not normalized per atom).
• I don't know what the energy units are for that dataset, but in general NequIP works with whatever units as long as they are consistent (i.e. don't have energies in eV and forces in Ha/bohr). See https://nequip.readthedocs.io/en/latest/guide/reference/faq.html#units
• if your DFT code is compatible with ASE, I'd recommend using ase.io.read and ase.io.write to generate .xyz files: https://ase-lib.org/ase/io/io.html