cagrikymk / JAX-ReaxFF

JAX-ReaxFF: A Gradient Based Framework for Extremely Fast Optimization of Reactive Force Fields
GNU General Public License v3.0
53 stars 22 forks source link

How to generate the datasets and initial force filed parameters? #7

Open Eipgen opened 1 year ago

Eipgen commented 1 year ago

Hello developer, I have read your article and run your examples. But I'm so confused that if I have a new system without any known ReaxFF force filed file. Is there any tools can generate a initial force filed file with the new system?

cagrikymk commented 1 year ago

Hello, thank you for the interest!

I have some scripts that can generate the initial force field and the required parameters file but as ReaxFF parameters have physical meaning, the parameter ranges are quite important so I would like to share the scripts once I am sure I have the correct ranges for all of the parameters. Otherwise, it might just cripple the training. I will contact Dr. van Duin to verify the scripts first. Unlike the machine learning model, the starting force field and parameter ranges carry more importance for ReaxFF training and I would not like to share ill-prepared scripts.

Eipgen commented 1 year ago

Thank you very much. The script tool seems very useful. And also I find some tools like https://github.com/mertyigit/INDEEDopt. I'm going to try to see if it can meet my needs combined with jax-reaxff

cagrikymk commented 1 year ago

Also as a side note, you can find the existing force fields here: https://www.scm.com/doc/ReaxFF/Included_Forcefields.html

Eipgen commented 1 year ago

Thank you very much, I will try it.

Eipgen commented 1 year ago

I also confused that, is these some tools can generate the trainset.in and geo file for new systems automatically? I find it can be generated by the commerical software ADF.

geniusLihaotian commented 1 month ago

I also want to know how to get the trainset.in and geo files automatically. Have you found the answer, Eipgen?