Machine learning plays an increasingly important role in many areas of chemistry and materials science, being used to predict materials properties, accelerate simulations, design new structures, and predict syn
116,117,118. Molecular orbital interactions can in turn be used for improving the prediction performance2. Lastly, the mapping of atoms to non-Euclidean space such as in the proposed hyperbolic GNNs119can lead to gains in representational efficiency. For a more in-depth discussion of graph...
(distances and three-body angles are invariant to rotation), as opposed to vectors used in this work. Cormorant18uses an equivariant neural network for property prediction on small molecules. This method is demonstrated on potential energies of small molecules but not on atomic forces or systems ...
In this work, we present the Neural Equivariant Interatomic Potential (NequIP), a highly data-efficient deep learning approach for learning interatomic potentials from reference first-principles calculations. We show that the proposed method obtains high accuracy compared to existing ML-IP methods across...