CompPhysVienna / n2p2

n2p2 - A Neural Network Potential Package
https://compphysvienna.github.io/n2p2/
GNU General Public License v3.0
214 stars 81 forks source link

Training without Potential Energies #162

Closed ThorrenKirschbaum closed 2 years ago

ThorrenKirschbaum commented 2 years ago

The question If I do not have total potential energies available that I can put into the training data file, can I just set them to zero and proceed with the training?

What did you already try? I have tried this and it seems to work, if I skip the normalization via nnp-norm (which would give me zeros and infinities). However, I would like to know whether it is mathematically sound to do this.

Many thanks in advance Thorren

Edit: I noted that this might not be a good idea overall. Please feel free to delete this post