Software

Open-source software for machine learning and electronic-structure theory

Our methods are only useful if other people can run them. We develop and maintain open-source software that brings machine learning into electronic-structure workflows, with a focus on reproducibility, scalability on HPC systems, and documentation that lets new users get started in minutes rather than weeks.

Both projects below are archived on Zenodo so that every release is citable with a DOI.

MALA — Materials Learning Algorithms

MALA is a framework for scalable machine learning of electronic-structure properties, trained on first-principles (DFT) data. Instead of solving the Kohn–Sham equations from scratch at every molecular-dynamics step, MALA predicts the electronic structure — density of states, band energies, total energy contributions — with a neural network, making large-scale atomistic simulations tractable at near-DFT accuracy.

   
Description Scalable machine learning for electronic structure calculations
Repository https://github.com/mala-project/mala
Documentation https://mala-project.github.io/mala/
Archived releases DOI 10.5281/zenodo.5557254
Language / stack Python, PyTorch, MPI (with some Fortran components)
License BSD-3-Clause

Highlights

  • Trains on data produced by your own DFT code (e.g. Quantum ESPRESSO, Vienna Ab initio Simulation Package, S/PHI/nX) — no proprietary training set required.
  • Scales from a laptop to leadership-class HPC systems through MPI-parallel data preprocessing and distributed training.
  • Provides ready-to-use workflows for total energy, density of states, and band-structure prediction inside molecular dynamics.
  • Ships as a Python package with a documented API, tutorials, and an extensive test suite.

Installation

git clone https://github.com/mala-project/mala.git
cd mala
pip install -e .

Citing MALA

If you use MALA in your work, please cite the paper and the Zenodo archive:

A. Cangi, L. Fiedler, B. Brzoza, K. Shah, T. J. Callow, D. Kotik, S. Schmerler, M. C. Barry, J. M. Goff, A. Rohskopf, D. J. Vogel, N. Modine, A. P. Thompson, S. Rajamanickam, Materials Learning Algorithms (MALA): Scalable machine learning for electronic structure calculations in large-scale atomistic simulations, Comput. Phys. Commun. 314, 109654 (2025). doi:10.1016/j.cpc.2025.109654

@article{Cangi2025Materials,
  title   = {Materials Learning Algorithms (MALA): Scalable machine learning for electronic
             structure calculations in large-scale atomistic simulations},
  author  = {Cangi, Attila and Fiedler, Lenz and Brzoza, Bartosz and Shah, Karan and
             Callow, Timothy J. and Kotik, Daniel and Schmerler, Steve and Barry, Matthew C. and
             Goff, James and Rohskopf, Andrew and Vogel, Dayton J. and Modine, Normand A. and
             Thompson, Aidan P. and Rajamanickam, Sivasankaran},
  journal = {Computer Physics Communications},
  volume  = {314},
  pages   = {109654},
  year    = {2025},
  doi     = {10.1016/j.cpc.2025.109654}
}

atoMEC — average-atom modelling for warm dense matter

atoMEC is a Python package that solves the average-atom model — a density-functional description of a single atom embedded in a plasma — to obtain equation-of-state, electronic-structure, and ionization properties of warm dense matter. It is designed as a readable, extensible reference implementation: the physics is documented, the numerical choices are explicit, and every model input can be inspected and modified.

   
Description Average-atom Python code for warm dense matter
Repository https://github.com/atomec-project/atoMEC
Documentation https://atomec-project.github.io/atoMEC
Archived releases DOI 10.5281/zenodo.5205718
Language / stack Python (NumPy, SciPy)
License Open source — see the repository for the current license

Highlights

  • Computes warm-dense-matter properties — ionization states, electron densities, equations of state — from a self-consistent average-atom model.
  • Supports multiple exchange–correlation functionals and bound-state boundary conditions, so models can be compared on equal footing.
  • Pure Python, easy to install, and simple to extend: suitable for teaching as well as production calculations.

Installation

pip install atoMEC

Citing atoMEC

T. J. Callow, D. Kotik, E. Kraisler, A. Cangi, atoMEC: An open-source average-atom Python code, Proceedings of the 21st Python in Science Conference (SciPy 2022). doi:10.25080/majora-212e5952-006 Archive DOI: 10.5281/zenodo.5205718

Contributing

We welcome contributions — bug reports, documentation fixes, and new features. Both projects use GitHub issues and pull requests; please open an issue before starting on a larger change so we can agree on the design. We are also happy to support new users getting started; see Join Us or write to a.cangi@hzdr.de.