Authors: Enze Chen (Stanford University) and Timofey Frolov (Lawrence Livermore National Laboratory)
Version: 0.2024.07.22

An algorithm for performing grand canonical optimization (GCO) of interfacial structure (e.g., grain boundaries) in crystalline materials. It automates sampling of slab translations and reconstructions along with vacancy generation and finite temperature molecular dynamics (MD). The algorithm repeatedly samples different structures in two phases:

1. Structure generation and manipulation is largely handled using the Atomic Simulation Environment (ASE).
2. Molecular dynamics and static relaxations are currently performed using LAMMPS, although in principle other energy evaluation methods (e.g., density functional theory in VASP) may be used.

• Python (3.6+)
• NumPy (1.23.0)
• ASE (3.22.1)
• LAMMPS (stable)

Optional

• pandas (1.5.3)
• Matplotlib (3.5.3)

Assuming the above libraries are installed, clone the repo and make the appropriate modifications in params.yaml (see file for detailed comments), including the path to the LAMMPS binary on your system. If you wish, you can supply your own slabs for the bicrystal configuration as POSCAR_LOWER and POSCAR_UPPER (in the POSCAR file format). Then call:

python main.py

If you don't have LAMMPS or just want to test the script, you can run it with the -d flag. See the .examples folder for a SLURM submission script for parallel execution (preferred).

• main.py: Script to launch everything.
• params.yaml: Simulation parameters; you'll want to edit this.
• core: Main classes (Bicrystal, Simulation, etc.)
• utility: Main helper functions (utils.py, unique.py, etc.)
• simul_files: Files for simulations (LAMMPS input files, etc.)
• best: All relaxed structures are stored here. The naming convention is: lammps_Egb_n_X-SHIFT_Y-SHIFT_X-REPS_Y-REPS_TEMP_STEPS

Duplicate files are periodically deleted by calling clear_best() in utils/unique.py. The default method cleans about 1-3% of files on average. Use the -e flag for more aggressive cleaning (>50%). Use the -s flag to save the processed results to CSV from a pandas DataFrame.

Results can be visualized by running python utils/plot_gco.py and it generates a GCO plot of $E_{\mathrm{gb}}$ vs. $n$. The .examples folder has this plot for several boundaries. By default executing this file will save both the results (CSV) and the figure (PNG) to the same folder as the GRIP output files.

• Add parallelism for other job schedulers besides SLURM.
• Create more flexible workflow classes for Monte Carlo moves, energy minimization options, etc.
• Extend the code to be compatible with the parallel build of LAMMPS.
• Extend the compositional DOF to work with multi-component systems.
• Add in Bayesian optimization to narrow down simulation parameters.
• Incorporate ways to identify the GB atoms.
• Optimize the saving of files to reduce the memory footprint (sometimes > 1 GB).

If you encounter any errors or have a suggestion, feel free to raise an Issue or Pull Request. We'll try to respond as soon as possible!

If you use GRIP in your work, we would appreciate a citation to the original manuscript:

Enze Chen, Tae Wook Heo, Brandon C. Wood, Mark Asta, and Timofey Frolov. "Grand canonically optimized grain boundary phases in hexagonal close-packed titanium." Nature Communications, 15, 7049, 2024.

or in BibTeX format:

@article{chen_2024_grip,
    author = {Chen, Enze and Heo, Tae Wook and Wood, Brandon C. and Asta, Mark and Frolov, Timofey},
    title = {Grand canonically optimized grain boundary phases in hexagonal close-packed titanium},
    year = {2024},
    journal = {Nature Communications},
    volume = {15},
    number = {1},
    pages = {7049},
    doi = {10.1038/s41467-024-51330-9},
}

GRIP is distributed under the terms of the MIT license. All new contributions must be made under this license.

SPDX-License-Identifier: MIT

LLNL-CODE-XXXXXX