Machine learning of atomic forces from quantum mechanics: An approach based on pairwise interatomic forces

In this work, we explore a machine learning approach based on pairwise interatomic forces to directly predict atomic in molecule quantum chemical accuracy. The scheme, denoted pairF-Net, uses neural network designed and trained Cartesian as linear combination of set force components an basis; these exhibit chemically intuitive profiles implicitly maintain rotational translational invariance. Application the model small systems indicates accuracy reconstructed ~1 kcal mol−1 Å−1 from reference values obtained via density functional theory. pairF-Net scheme predicts at mechanical level but fraction cost, enabling molecular dynamics simulations model.