Machine Learning Identifies Strong Electronic Contacts in Semiconducting Polymer Melts

In semiconducting polymers, charge transport depends on the electronic coupling between neighboring molecules and is governed by their local structures. Here, we identify the local arrangements that promote strong electronic contacts and assess how thermal annealing affects the probability of such contacts. We use molecular dynamics with virtual site coarse graining to simulate the bulk morphology of a semicrystalline donor polymer (P3HT) and then evaluate electronic coupling between monomers on neighboring chains. To avoid brute-force electronic calculations and tedious manual identification of strongly coupled pairs, we apply feature selection in machine learning to identify the most important configurations and develop a predictive model to predict electronic coupling from the most important geometric features of the local arrangement. We find that the key geometric features for strong contacts promoting hole transport closely relate to coherent overlap between HOMO wavefunctions on nearby moieties. Strong contacts in amorphous P3HT are rare but become more common with slow cooling, which leads to the formation of crystalline regions in which π-stacked configurations have more coherent overlap and thus stronger electronic couplings.