rsspolymlp 0.0.10

A framework for random structure search using polynomial MLPs

A framework for random structure search (RSS) using polynomial MLPs

Citation of rsspolymlp

If you use rsspolymlp in your study, please cite the following articles.

“Efficient global crystal structure prediction using polynomial machine learning potential in the binary Al–Cu alloy system”, J. Ceram. Soc. Jpn. 131, 762 (2023)

@article{HayatoWakai202323053,
  title="{Efficient global crystal structure prediction using polynomial machine learning potential in the binary Al–Cu alloy system}",
  author={Hayato Wakai and Atsuto Seko and Isao Tanaka},
  journal={J. Ceram. Soc. Jpn.},
  volume={131},
  number={10},
  pages={762-766},
  year={2023},
  doi={10.2109/jcersj2.23053}
}

Installation

Required libraries and python modules

• python >= 3.9
• pypolymlp
• symfc
• spglib
• joblib

[Optional]

• matplotlib (if plotting RSS results)
• seaborn (if plotting RSS results)

How to install

• Install from conda-forge

Name	Downloads	Version	Platforms

conda create -n rsspolymlp
conda activate rsspolymlp
conda install -c conda-forge rsspolymlp

• Install from PyPI

conda create -n rsspolymlp
conda activate rsspolymlp
conda install -c conda-forge pypolymlp symfc spglib joblib
pip install rsspolymlp

Usage

Example Commands

Step 1–3: Execute for each combination of pressure (p), composition (c), and number of atoms (n).

rss-init-struct --elements Al Cu --atom_counts 4 4 --num_init_str 2000
rss-parallel --pot polymlp.yaml --pressure 0.0 --num_opt_str 1000
rss-uniq-struct

Steps 4–6: Execute after the above steps and analyze the results aggregated by (p, c) conditions.

rss-summarize --elements Al Cu --use_joblib --rss_paths <rss_directory>/*
rss-outlier --result_paths <summary_dir>/json/*
rss-phase-analysis --elements Al Cu --result_paths <summary_dir>/json/* --thresholds 10 30 50
# <rss_directory>: parent directory of RSS runs at the same pressure
# <summary_dir>: output directory from rss-summarize, storing RSS results

Workflow

The command-line interface of rsspolymlp is organized into 6 steps.

Steps 1–3 perform RSS using the polynomial MLP independently for each (p, c, n) conditions.

1. Generating initial structures (rss-init-struct)

   Random structures are generated under the specified conditions.

2. Performing parallel geometry optimization (rss-parallel)

   Each generated structure is optimized in parallel using polynomial MLPs.

3. Elimination of duplicate structures (rss-uniq-struct)

   This step processes the optimized structures. It includes:

   • Parsing optimization logs, filtering out failed or unconverged cases, and generating detailed computational summaries.
   • Removing duplicate structures and extracting unique optimized structures.

Steps 4–6 analyze the results aggregated over each (p, c) condition.

4. Identifying unique structures across atom numbers (rss-summarize)

   This step removes duplicate structures from the set of unique structures obtained at different atom counts n under the same pressure and composition conditions.

5. Outlier detection (rss-outlier)

   Identifying and filtering out anomalous structures based on energy values are performed.

6. Phase stability analysis (rss-phase-analysis)

   This step computes the relative or formation energies of structures obtained from the RSS and outputs the global minimum structures. It also identifies metastable structures near the convex hull based on a energy threshold (e.g., 30 meV/atom).

7. (Optional) Plotting RSS results (e.g., plot-binary)

   The energy distribution of structures obtained through this RSS workflow is visualized.

Additional information is here