Predicting glass structure by physics-informed machine learning

Abstract Machine learning (ML) is emerging as a powerful tool to predict the properties of materials, including glasses. Informing ML models with knowledge how glass composition affects short-range atomic structure has potential enhance ability composition-property extrapolate accurately outside their training sets. Here, we introduce an approach wherein statistical mechanics informs model that can non-linear composition-structure relations in oxide This combined offers improved prediction compared relying solely on physics or machine individually. Specifically, show both interpolates and extrapolates Na 2 O–SiO Importantly, able predictions its set, which evidenced by fact it series was kept fully hidden from during training.