Quasi-localized vibrations arise from liquid-like structures in glasses: Insights from deep neural networks

Abstract

Glasses possess material properties that differ significantly from those of crystals. These properties are closely linked to quasi-localized vibrations (QLVs) present in glasses, thus understanding that QLVs are crucial for comprehending the material properties of glasses. However, the connection between QLVs and amorphous structures of glasses remains poorly understood. To address this issue, we employ a deep neural network approach to characterize the amorphous structures. By training the neural network using data on the structures from both liquid and solid (glass) samples, the network is able to identify the liquid-like and solid-like regions within a sample. An examination of the spatial correlations of the QLVs with these liquid-like and solid-like structural distributions reveals that the QLVs originate from the liquid-like regions. Our findings indicate that, in glasses, the QLVs emerging from the liquid-like structures contribute to glassy phenomena such as the dynamic heterogeneity of supercooled liquids and the non-affine elasticity of glasses.