Question about isotope treatment (H/D/T) in i-PI + VASP RPMD and potential training

[Mia Amelia]

Dear i-PI developers,

I am currently using i-PI coupled with VASP to perform RPMD (Ring Polymer Molecular Dynamics) simulations of hydrogen isotope diffusion (H, D, T) in a metallic system.
My goal is to use first-principles data from these simulations to train a machine-learning potential, which will then be applied to study larger systems and longer time scales.

Since VASP, as a DFT code, determines the potential energy surface based on electronic density rather than nuclear mass, I would like to clarify the following:

When generating the training dataset from first-principles simulations, do I need to treat H, D, and T as different atomic species and train separate potentials for each isotope,
or is it sufficient to train only one potential (e.g., for H) and simply change the atomic mass when using the trained potential for subsequent i-PI simulations of D or T?

Any guidance or best practices on this point would be greatly appreciated.
Thank you very much for your time and for the excellent tool you have developed.

Best regards,
Mia Amelia

[Mariana Rossi]

Hi Mia,

You don’t need to train something separate to each isotope as you guessed.

All the best,

Mariana

[Mariana Rossi]

You should do dynamics with different isotopes to get different configurations but the dft calculations are all H