Inferring the Isotropic-nematic Phase Transition with Generative Machine Learning

Abstract

Contemporary work implies generative machine learning models are capable of learning the phase behavior in condensed matter systems such as the Ising model. In this Letter, we utilize a score-based modeling procedure called Thermodynamic Maps to describe the isotropic-nematic phase transition in a melt of $N=343$ calamitic Gay-Berne ellipsoids. When trained on samples generated by molecular dynamics simulation from a single temperature on either side of the phase transition, we demonstrate this generative machine learning approach infers information regarding the critical behavior and estimates effectively the nematic order parameter at sampled temperatures between the two training temperatures. These results demonstrate score-based models' ability to learn the physics of a non-trivial liquid crystalline phase transition driven by anisotropic interactions both entropic and energetic in nature.