- Python 70.8%
- Julia 23.7%
- Shell 5.5%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| config | ||
| LICENSE | ||
| openmx_installation.md | ||
| README.md | ||
| requirements.txt | ||
OpenMX Workflow
A high-throughput workflow for OpenMX calculations using atomate. This workflow was also employed in our recent work: arXiv:2505.04862.
This repository is adapted from openmx-wf by @tsaie79.
The main extension is the post-processing of OpenMX outputs (.scfout and .out) to compute response functions such as:
- Frequency-dependent permittivity tensor
Prerequisites
This workflow assumes that users:
- Are familiar with OpenMX and have a compiled version 3.9 installed.
- Have Julia installed.
- Have the packages required by DeepH.
(Additionally, you will need HopTB.jl.) - Have access to a running MongoDB instance for job and data management.
Installation
To set up the environment, run:
pip install -r requirements.txt
This will install the necessary dependencies for the workflow, including two custom packages:
- pymatgen (custom fork) – provides a custom-defined basis set for passing into the
ASE.Atomsobject. - atomate (custom fork) – enables job submission and data storage.
Usage
-
Configure your workflow using the provided configuration files.
See openmx-wf and the atomate documentation for details. -
Once configured, run:
python config/wf_poscar_direct.py
This will:
- Convert your input into the required
openmx.datformat. - Insert the workflow into the FireWorks database (MongoDB).
From your designated launch site, execute the workflow using FireWorks:
qlaunch singleshot
or
qlaunch rapidfire -m 1
After the calculations are finished, all results will be stored in the MongoDB database.
You can easily explore and query the generated data using MongoDB Compass or any other MongoDB client.
Citing
If you use this workflow in your research, please cite:
@misc{hsu2025accuratepredictionsequentialtensor,
title={Accurate Prediction of Tensorial Spectra Using Equivariant Graph Neural Network},
author={Ting-Wei Hsu and Zhenyao Fang and Arun Bansil and Qimin Yan},
year={2025},
eprint={2505.04862},
archivePrefix={arXiv},
primaryClass={cond-mat.mtrl-sci},
url={https://arxiv.org/abs/2505.04862},
}