[CP2K-user] Machine Learning Force Fields

Nicklas Österbacka nicklas.... at gmail.com
Fri Jun 18 12:59:26 UTC 2021

n2p2 <https://compphysvienna.github.io/n2p2> implements 
Behler-Parinello-style neural network potentials. There is also a plugin 
for the MD package QUIP that implements Bartók's kernel-based Gaussian 
Approximation Potential <https://libatoms.github.io/GAP/>.

There are plenty more, but those two should give you something to test 
things out with and are code-agnostic. You do have to prepare the data set 
for training, however. Writing a script to do so should not be particularly 

Good luck,
fredag 18 juni 2021 kl. 14:33:38 UTC+2 skrev aw... at gmail.com:

> Dear All,
> Is there any straighforward way of generating machine learning force 
> fields using a CP2K trajectory? I noticed that most of the tools 
> available online use VASP or QUANTUM ESPRESSO. I am looking for a tool 
> which allows me to construct a force field by training on both energy and 
> forces, but the tool needs to be useful for a complete newbie. Any ideas ? 
> Best wishes,
> Ana
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210618/e8335c20/attachment.htm>

More information about the CP2K-user mailing list