Using machine learning to characterize active matter phases: implications for active dynamics on optical substrates

Abstract

Active matter or self-propelled particles arise across a variety of soft matter systems such as self-motile colloids and artificial swimmers, biological systems such as bacteria, and robotic systems. Active matter exhibits a wide variety of phenomena that do not occur in its equilibrium counterparts. One of the biggest issues in active systems is how to characterize and understand whether they have distinct phases or phase transitions. Here, we use machine learning (ML) in the form of principal component analysis (PCA) to study active matter phases for a collection of interacting run-and-tumble disks. One of the most interesting phenomena exhibited by active particles is motility induced phase separation. Using ML, we find evidence of the existence of multiple regimes within the motility induced phase separated state. We discuss future directions in which ML methods could be used to characterize active matter on ordered substrates created by optical means. We also describe how ML approaches could be used as a tool to characterize more complex active matter systems, to optimize rules for motion, or to create optimal substrates for specific applications.