MACE-MDP: Dipole and Polarizability Models for Organic Systems

April 24, 2026 · View on GitHub

MACE-MDP is a machine-learning model for predicting molecular dipole moments and fully anisotropic polarizability tensors for organic systems. The model is based on an E(3)-equivariant message-passing neural network architecture (MACE) and is trained on the SPICE-α dataset.

MACE-MDP enables efficient prediction of dielectric response properties that are central to infrared (IR) and Raman spectroscopy, electrostatics, and molecular response to external electric fields. When combined with machine-learning interatomic potentials (MLIPs), the model allows rapid, first-principles–accurate vibrational spectroscopy calculations across molecules, clusters, and condensed-phase systems.

MACE-MDP provides trained MACE models and ready-to-run tutorials for:

  • Dipole (e A˚e\ \text{Å}) | Polarizability (e A˚2/Ve\ \text{Å}^2/\text{V})
  • IR spectra
  • Raman spectra

What you need to use MACE-MDP

This repository already contains everything required for inference and tutorial use:

  • model file: models/MACE-MDP.model
  • tutorials and scripts: examples/
  • example structure set: examples/mini_database_IR-R-7193_wB97MD3.xyz

Installation

Clone the repository, enter it, and install the pinned dependencies with pip:

git clone https://github.com/Nilsgoe/MACE-MDP.git
cd MACE-MDP
pip install .

This installs the same pinned dependencies listed in reproducibility/requirements.txt.

For newer MACE versions, clone the MACE repository and follow the installation instructions there.

Tutorials

The repository includes example notebooks demonstrating how to use MACE-MDP:

  • examples/IR/IR_tutorial.ipynb
  • examples/Raman/Raman_tutorial.ipynb
  • examples/Dipole_Polarizability/Dipoles_Polarizability_tutorial.ipynb

These tutorials are configured to use:

  • XYZ file: examples/mini_database_IR-R-7193_wB97MD3.xyz
  • model path: models/MACE-MDP.model

Repository layout

  • models/ – trained MACE-MDP model(s)
  • examples/ – notebooks, scripts, and example inputs
  • reproducibility/ – smoke test and environment instructions

Reproducibility smoke test

Run a minimal CPU check on the first bundled example structure:

python reproducibility/run_minimal_inference.py

Code availability

The code, trained model, examples, and smoke test are available in this repository. MACE-MDP was trained on SPICE-alpha, available at https://zenodo.org/records/19205036; that record also includes the IR-R-7193 and R-3B69 test sets. Additional scripts for the paper plots and dataset-related workflows are available at https://github.com/Nilsgoe/Benchmark-MACE-MDP. See CODE_AVAILABILITY.md and LICENSE.md for details.

Citation

If you use MACE-MDP in your research, please cite:

Gönnheimer, N.; Reuter, K.; Kapil, V.; Margraf, J. T. MACE-MDP: A General Dipole and Polarizability Model for Organic Molecules and Materials. ChemRxiv (2026). https://chemrxiv.org/doi/full/10.26434/chemrxiv.15000716

License

Copyright [MACE-MDP] is © 2026, [Nils Gönnheimer]

MACE-MDP is distributed under the Academic Software License v1.0 (ASL) for academic non-commercial use.

See LICENSE.md for details.