Structural analysis of high-density silica glass by machine learning molecular dynamics simulation

Kobayashi, Keita 

The atomic arrangement in glass structures lacks periodicity, and the information obtained experimentally reflects an average structure. Therefore, to estimate the three-dimensional structure of glass materials, molecular dynamics simulations are effective. The results of molecular dynamics calculations strongly depend on the interatomic potential. We have created a machine learning potential (MLP) trained on first-principles calculation results for silica materials. This paper outlines our research of structural analysis of high-density silica glass using machine learning molecular dynamics (MLMD) with the MLP. The MLMD successfully reproduced the experimental data of silica glass. Furthermore, it was revealed that changes in the medium-range order structure in high-density silica glass are characterized by the deformation behavior of ring structures within the Si-O covalent bond network due to the compression.