Reversible densification and cooperative atomic movement induced “compaction” in vitreous silica: a new sight from deep neural network interatomic potentials

Vitreous silica ( v -silica) is a challenging material to characterize due to its disordered structure and thermal-history-dependent properties, which are not fully captured by classical potential models. In this study, we trained deep neural network (DNN) potentials with ab initio precision to describe the structure, dynamics, thermal conductivity, and densification of v -silica, comparing the performance of two exchange correlation functionals, BLYP and AM05. Our results demonstrate that the cooperative atomic movement within inner-tetrahedral and inter-tetrahedral SiO 4 units plays a critical role in the volume-conservative “compaction” of v -silica, which in turn results in a “shrinking” of the vibrational density of state spectrum. We also found that the transition of coordination number is correlated with the minimum of longitudinal wave velocity. Moreover, our DNN model reveals that the long-range disorder changes linearly within the pressure range of 0–9 GPa.