path integral molecular dynamics using i-PI

[lnnbig]

Can we train an NNP using a training set generated from classical (DFT) MD, then use this NNP for path integral MD using lammps+i-PI?
Does the training set has to be from path integral rather than classical MD?

[njzjz]

Can we train an NNP using a training set generated from classical (DFT) MD, then use this NNP for path integral MD

Yes, I have a paper about it: https://doi.org/10.1021/acs.jctc.2c00151