Temperature dependence of infrared dielectric responses of polar amorphous solids by machine-learning molecular dynamics

The theoretical explanation and understanding of the temperature dependence of the infrared optical response of polar amorphous solids remains not yet established. In this study, we target this challenge via Green–Kubo molecular dynamics based on a machine-learning potential with dynamic charges, using amorphous silicon nitride (a-Si3N4) and amorphous silicon dioxide (a-SiO2) as prototype materials. Negligibly and weakly temperature-dependent infrared dielectric functions of a-Si3N4 and a-SiO2 are predicted, respectively, and show good agreement with the experimental results. Complementary analyses of characteristic vibrational spectral features, dynamic Born effective charges, and local structural distributions provide insight into the contrasting temperature dependence of the two materials. These findings thus provide significant physical insight into the temperature dependence of the infrared optical response in polar amorphous solids.