Extrapolation warning occurs when using my NNP in interface-LAMMPS.

[ANCHENMA]

Hi,

I used my own dataset (300K, 1000K, density + 10%) to train the high-dimensional neural network potential of bulk water with RuNNer, and then used a randomly built 64-molecule water box to run 1ns NVT in lammps interface of n2p2, but there were extrapolation warnings (449EWPERSTEP) at 71.5ps. After that, there were more and more extrapolation warnings(428371EWPERSTEP at 73 ps), and the water molecular dispersed, the hydrogen and oxygen atoms gathered together respectively .

What should I do next, pick configurations with nnp-comp2?

md_lmp.txt input_nn.txt log_lammps.txt.txt

I would be very grateful indeed for any help you could give me. Zhaoan Chen

[ANCHENMA]

Initial data set was obtained from MD simulation in vasp of 64 water molecules in the NVT ensemble with a length of 8ps

[philippmisof]

Usually this is a sign of an insufficient training data set. Constructing a good data set is a non-trivial task. To get a feeling, look at the data set for bulk water at HDNNP training data set for H2O and its corresponding README.pdf

[ANCHENMA]

Thanks for your help，I solved the question.