A hybrid machine-learning/theory approach to dynamics in supercooled liquids

The three-dimensional glass transition is an infamous example of an emergent collective phenomenon in many-body systems that is stubbornly resistant to microscopic understanding using traditional statistical physics approaches. Establishing the connection between microscopic properties and the glass transition requires reducing vast quantities of microscopic information to a few relevant microscopic variables and their distributions. I will demonstrate how machine learning, designed for dimensional reduction, can be combined with theory to provide a natural way forward when standard statistical physics tools fail. We have harnessed machine learning to identify a useful microscopic structural quantity for the glass transition and have used it to build a new theoretical model for glassy dynamics. At the heart of the problem is the question of whether spatial correlations of structure (correlation) or dynamical facilitation (causation) is more important for the slowing down. We show that time-reversal-invariance, needed for the system to remain in thermal equilibrium, demands that correlation and causation are linked together.