Totally wrong dynamics with i-PI

[mcourtiel]

Hi,

I have a problem when I run a dynamic with i-PI using the NNP I have built.
I trained it on the water monomer, with 160 000 structures of the monomer extracted from a dynamic at 10K. I use 80% for the training and 20% for the validation and then testing.

The RMSE seems good with 3e-05 eV for the energies and 3e-03 eV/A for the forces. (On the curves you can see the mobile average)

The problem is when I run a dynamic with i-PI using this potential (that seems good to me), the two H atoms totally go away from the O one so this seems really strange. Does anyone have any idea where this error may have come from?

300000_input.pdf
i-pi_input_xml.pdf