Automated detection of spatial and dynamical heterogeneity of nano-domains in supercooled liquids via implementation of Machine Learning Algorithms

Understanding the physics of supercooled liquids near glassy transition remains one of the major challenges in condensed matter science. There has been long recognized that both dynamical and spatial structures in supercooled liquids are heterogeneous. As liquid is cooled far below its melting point fast, dynamics in some regions of the sample can be orders of magnitude faster than the dynamics in other regions only a few nanometers away. However, to identify such domain structures and the connection between structures and dynamics remains elusive. We developed a theoretical approach via implementation of Principle Component Analysis (PCA) and Gaussian Mixture (GM) clustering methods from Machine Learning (ML) algorithms to identify domain structures of a supercooled Kob-Andersen binary Lennard-Jones liquid. In our approach, raw features data are collected from the coordination numbers of particles smoothed using radial distribution function and are used as an order-parameter for training GM clustering after dimensionality reduction from the PCA. To transfer the knowledge from features space to real space, another GM clustering is performed using the Cartesian coordinates as an order-parameter with the initial values from GM in features space. Both GM clustering are performed iteratively until convergence. Final results show the appearance of aggregated clusters of nano-domains over sufficient long timescale with heterogeneous dynamics. More importantly, from these studies we consistently observe nano-domain structures as a function of temperature regardless of finite size effect and our approach can be broadly applied to more complex systems of interest.