(613g) What Insights Can Machine Learning Provide Towards Multiblock Copolymer Self Assembly?

The physics and phase behavior of multiblock copolymers become increasingly complicated as the number of blocks increases. Correspondingly with increasing architectural complexity it becomes more and more difficult to predict both phase behavior and domain size. However machine learning models provide a potential solution to this problem by creating models capable of predicting the phase behavior and domain size based solely on sequence and the χ parameter using large amounts of polymer data. As a proof of concept here we present novel work using a neural network to predict the width of a family of block copolymer lamellae phases using the bead sequence as an input. We find that the machine learning model is not only capable of both predicting lamellae width as a function of sequence, but also that the machine learning model learns and preserves the underlying physics we expect from such block copolymers. This fact is demonstrated by the machine learning model preserving the expected behavior as a function of the different block fractions. The findings presented here demonstrate the versatility of such machine learning models and suggest their ability to provide insights into complex block copolymer physics.