Machine Learning and Polymer Self-Consistent Field Theory in Two Spatial Dimensions

Numerical simulations based on self-consistent field theory (SCFT) are a powerful tool to study the energetics and structures of polymer phases [2, 3, 4]. However, the high cost of these direct computations, involving the repeated solution of multiple modified diffusion (Fokker-Planck) equations [2, 5, 6], hinders the scalability of SCFT for its application in polymer phase discovery. Machine learning (ML), which obtains an approximate input-to-output map from data, can substantially reduce (after training) the computational cost of evaluating quantities of interest. Consequently, there has been increasing interest to combine ML with traditional polymer SCFT simulations to speed up the exploration of parameter space. Most of the work to date has been focused on separate stages of SCFT computation and/or downstream tasks.