Deep Learning Boosted Langevin Field-Theoretic Simulation of Polymers

Abstract

Physicists' community recently adopted machine learning for various research tasks, but performing polymer simulation using deep learning (DL) is still a relatively unexplored subject. For the widespread use of DL in a polymer simulation, accuracy of the DL prediction and the neural net training time are the two key issues that must be overcome. Field-theoretic simulations are promising tools in polymer field theory that can account for the compositional fluctuation effect, and among them, Langevin field-theoretic simulation (L-FTS) is known to be fast and free from the instability issue at low invariant polymerization index. However, it is still a computationally expensive tool, and it may take weeks to accurately calculate ensemble averages of thermodynamic quantities. In this presentation, we introduce a DNN that can be successively applied to determine the partial saddle point of the pressure field. Major deep learning (DL) models for semantic segmentation in computer vision are adopted to construct the optimal DNN architecture. Our model utilizing atrous convolutions in parallel is robust to the simulation parameter changes and can be reused after single training. Our DNN can achieve speedup of factor 6 or more compared to the conventional method without affecting accuracy. Open-source code for our deep Langevin FTS (DL-FTS) enables an easy and rapid Python scripting of SCFT and L-FTS incorporated with CPU or GPU parallelization and DL.