(Partial) reproduction of the results outlined in R. Drautz's 2019 paper.
Dataset acquired from pACE [1].
Cu_FHIaims_dataset_overview.ipynb contains exploratory data analysis of the dataset, and preparation of the dataset for the ACE fitting model.
To train the model, you define an input configuration input.yaml (currently in model/) and train it using the TensorFlow backend provided as part of the pyace package. input.yaml defines the model parameters (cutoffs, basis order, training weights).
Run the command pacemaker input.yaml in the working directory to begin training. When the model is done training, it outputs output_potential.yace that can be used for further steps.
fitting_data_info.pckl.gzip contains the metadata information for the model train run.
The model is currently incompletely trained, but you can look at current_extended_potential.yaml for the current progress. current_extended_potential.yaml is an incremental checkpoint of output_potential.yace.
Alternatively, you can look at toy_model/ for a working example of a smaller functional model.
Currently Incomplete
To export to LAMMPS to test on custom pyace outputted yaml to a yace format readable by LAMMPS (through the pair_style pace interface):
- Build LAMMPS with ACE
- Run the command:
pace_yaml2yace current_extended_potential.yaml,
Where you can replace current_extended_potential with your (completely trained) potential file.
- Create an input file (e.g.
input.in) for LAMMPS, including your model in the file as:
pair_style pace
pair_coeff * * potential.pace Cu
- Run the model via
lmp -in input.in, and you're done!
- Lysogorskiy, Y., Oord, C.v.d., Bochkarev, A. et al. Performant implementation of the atomic cluster expansion (PACE) and application to copper and silicon. npj Comput Mater 7, 97 (2021). https://doi.org/10.1038/s41524-021-00559-9
- Drautz, R. (2019). Atomic cluster expansion for accurate and transferable interatomic potentials. Physical Review B, 99(1), 014104. https://doi.org/10.1103/PhysRevB.99.014104