Long-Range Equivariant Machine Learning Interatomic Potentials for Simulating Charge Transfer.

Machine learning interatomic potentials (MLIPs) provide a highly accurate and computationally efficient alternative to first-principles methods like Density Functional Theory (DFT). Typically, message passing neural networks (MPNNs) are employed to implement these MLIPs, utilizing local descriptor-based symmetry functions to capture atomic interactions. However, strictly local descriptor-based equivariant interatomic potentials are proven to be less accurate when employed in atomic systems where long-range interactions are significant such as systems with varying charge distributions among atoms of the same species. We incorporated long-range interactions into an equivariant neural network by constraining the predicted electronegativities within realistic bounds to ensure the model accurately reflects physical behavior. We evaluate our model on a range of benchmark datasets from the literature, including both periodic and non-periodic systems. Our model outperforms local descriptor-based interatomic potential in the prediction of energies and forces. We expect that this potential will enable more accurate and efficient molecular dynamics simulations of systems with long-range interactions.