(747b) Intelligent, Autonomous Exploration of Self Assembly Simulation Parameter Space

Computational “experiments” using coarse-grained models are immensely popular and have helped us understand the behavior of systems with length and time scales ranging from those of atoms to proteins. For exploratory studies in particular, researchers often spend large amounts of computational resources simulating not-so-interesting behavior of these models, such as large regions of parameter space that form trivial structures or kinetically arrested states. It is usually difficult to know which areas in parameter space should be sampled or how systems should be annealed in order to yield interesting or desirable phase behavior. Without the guidance of prior simulations, this leads to an iterative “trial-and-error” approach in many exploratory studies. Here we discuss strategies to incorporate machine learning into the experimental design loop in order to optimize computational time spent simulating interesting phenomena in the area of self assembly.