Symmetry-adapted graph neural networks for constructing molecular dynamics force fields

Molecular dynamics is a powerful simulation tool to explore material properties. Most realistic systems are too large be simulated using first-principles molecular dynamics. Classical has lower computational cost but requires accurate force fields achieve chemical accuracy. In this work, we develop symmetry-adapted graph neural network framework called the (MDGNN) construct automatically for simulations both molecules and crystals. This architecture consistently preserves translation, rotation, permutation invariance in simulations. We also propose new feature engineering method that includes high-order terms of interatomic distances demonstrate MDGNN accurately reproduces results classical addition, constructed by proposed model have good transferability. The thus an efficient promising option performing large-scale with high