ecnn4klm aims to perform fast and large-scale simulations of localized spin dynamics in the Kondo lattice model using an equivariant convolutional neural network (ECNN). For details on the ECNN model, please refer to the paper: Y. Miyazaki, Mach. Learn.: Sci. Technol. 4 045006 (2023).

ecnn4klm is based on PyTorch and e3nn. To install, you need PyTorch version 1.8 or higher. I also highly recommend running it on a GPU to maximize performance.

pip install git+https://github.com/Miyazaki-Yu/ecnn4klm.git

You can easily try it out on your browser using Google Colab.

The models and the datasets for both the square lattice and the triangular lattice cases, which are discussed in the paper, can be found in the data/ directory.

@article{Miyazaki2023_ecnn4klm,
doi = {10.1088/2632-2153/acffa2},
url = {https://dx.doi.org/10.1088/2632-2153/acffa2},
year = {2023},
month = {oct},
publisher = {IOP Publishing},
volume = {4},
number = {4},
pages = {045006},
author = {Yu Miyazaki},
title = {Equivariant neural networks for spin dynamics simulations of itinerant magnets},
journal = {Machine Learning: Science and Technology},
}