Repository logo
 

Research Data for: Transferability of datasets between machine-learned interaction potential algorithms


Loading...
Thumbnail Image

Change log

Description

Data associated with a research project into how far training data from one Machine-learned Interaction Potential (MLIP) can be used by another.

Repository includes molecular configurations labelled with DFT energies and forces, generated through a variety of approaches, that were used as training data for multiple MLIPs. The MLIP models themselves are also provided, together with a number of scripts used to train them, generate new training configurations through active learning, and validate them through molecular simulations.

Finally, we include the data required to reproduce the figures in the associated article. The trajectories from which they were derived are not provided but could easily be reproduced with the scripts presented. The original trajectories can be made available on reasonable request.

Please also refer to the Readme files located in subfolders to explain further the contents.

Version

Software / Usage instructions

To reproduce this data the following packages would be required: Lammps: https://github.com/lammps/lammps.git DeePMD-kit: https://github.com/deepmodeling/deepmd-kit.git MACE: https://github.com/ACEsuit/mace.git

Publisher

Rights and licensing

Except where otherwised noted, this item's license is described as https://creativecommons.org/licenses/by/4.0/
Sponsorship
European Commission Horizon 2020 (H2020) Research Infrastructures (RI) (957189)
European Commission Horizon 2020 (H2020) ERC (835073)