A Machine Learning Study on the Glass Transitions and Crystalline Orders
발표자
()
초록
내용
When systems in liquid phases are supercooled, the systems may avoid crystallization and undergo a glass transition. While the structures of the systems in glassy states are not very different from those in liquid states, glasses exhibit surprisingly high viscosity, thus becoming amorphous solids. Even though such glass transition is ubiquitous and has been observed in various materials and biological systems, there have been unanswered and challenging questions: is the glass transition a thermodynamic transition or a kinetic trap? Is there any universal structural order parameter like medium range crystalline order (MRCO)? In order to answer those questions, we perform molecular simulations for two-dimensional (2D) colloidal suspensions as a model system. We report in this talk that one can take advantage of the machine learning successfully to predict the glass states only from snapshots of 2D colloids, and also that there would be a universal structural order parameter beyond MRCO.