Cartesian Tensor

Cartesian tensors are increasingly used to represent atomic environments in machine learning models for predicting molecular properties, such as interatomic potentials and forces. Current research focuses on developing efficient and accurate message-passing neural network architectures that leverage Cartesian tensor representations, often incorporating irreducible tensor decompositions to improve computational efficiency and equivariance to rotations. This approach offers a computationally advantageous alternative to spherical tensors, enabling faster and more accurate simulations of molecular systems, with applications ranging from materials science to drug discovery. The resulting models demonstrate competitive performance compared to existing methods, while requiring fewer parameters and computational resources.