Closed dominicrufa closed 3 months ago
@peastman, it's not obvious to me how to place it in anipotential
. the documentation here seems to suggest that the EnergyShifter
adds another multiple of the self interaction energy onto the pre-computed energy, unless i am reading it incorrectly.
I think it's already included:
>>> model = torchani.models.ANI1ccx(periodic_table_index=True)
>>> print(model.energy_shifter.self_energies)
tensor([ -0.5992, -38.0375, -54.6745, -75.1604], dtype=torch.float64)
would it be possible to add
EnergyShifter
to the currenttorchani
workflow in order to remove the self energy of molecules treated with ML? Energies at present are very large/small and there is concern that this could leading to numerical instability with finite precision.