Exploration of elastic moduli of molecular crystals via database screening by pretrained neural network potential

Abstract

The elastic properties of molecular crystals are important for pharmaceutical and material applications, but calculating the elastic constant tensor through theoretical computations is computationally expensive. This study evaluated the feasibility of using neural network potentials, which are noted for their lower computational cost compared to theoretical calculations, in predicting the elastic moduli of molecular crystals. The calculated elastic moduli were sufficiently consistent with experimental values, outperforming Hartree–Fock calculations. It was found that this method is precise enough for predicting the Young's modulus in specific directions via nanoindentation measurements, and relationships between the magnitude of the Young's modulus and crystal structures were also discovered. Furthermore, the database screening of elastic moduli of molecular crystals using a pretrained neural network potential suggested crystals with large and small moduli.