The glass transition of CuZr metallic glasses in the perspective of machine learning

Abstract

During the cooling process, the liquid phase may transition to a glassy state. This is a continuous process, and pinpointing the exact location of the glass transition can be challenging. In this study, we simulated the glass forming process of CuZr liquids using classical molecular dynamics simulation. We demonstrate that by slicing the 3D configuration which generated by molecular dynamics simulation into 2D images and utilizing machine learning (ML) techniques, we can precisely identify the glass transition based solely on these 2D images. We introduced a new order parameter, Confidence Index for Glass (CIG), to characterize the structural changes during the glass transition. Our findings show that CIG can effectively distinguish between the structures of glass and liquid in CuZr alloys with various compositions and cooling rates, and the glass transition becomes much more evident from the perspective of CIG compared to other parameters like volume or potential energy. Our results demonstrate the significant potential of AI tools in advancing research in the field of physics.