(386f) Machine Learning Predictions of Electronic Couplings for Charge Transport Calculations of P3HT

Improving the inexpensive generation of electricity with organic semiconductors requires an understanding of how charges move through these materials. Here we aim to predict charge mobility in films of poly-(3-hexylthiophene) with varying degrees of molecular order. We lower the computational cost of these predictions by training machines to determine electronic couplings between polymer chromophores, which is a bottleneck in this prediction pipeline. We evaluate five machine learning techniques and find random forests accurately predict couplings 390 times faster than performing quantum chemical calculations from scratch. Using these couplings as inputs to kinetic Monte Carlo simulations of charge transport we find zero-field hole mobilities within 5% of prior work. We conclude with discussions of the training data that are needed to inform these accurate predictions and preliminary applications of this approach to more complex organic semiconductors.