Modeling novel ionenes for electrochemical devices with first principles and machine learning methods

Polybenzimidazole-based ionenes are being developed for use in both alkaline anion-exchange membrane fuel cells and alkaline polymer electrolysers. The first part of this work explores the impact of the degree of methylation on the conformations and electronic structure properties of poly-(hexamethyl-p-terphenylbenzimidazolium) (HMT-PMBI), the materials of interest in this thesis. For this purpose, HMT-PMBI oligomers, from monomer to pentamer, are studied with density functional theory calculations. Next, molecular dynamics simulations are used to calculate the trajectory paths of all atoms of the fully methylated HMT-PMBI tetramer. Lastly, recurrent neural networks are explored as a means to accelerate the statistical sampling of molecular conformations of polymeric systems, thereby providing complementary tools for molecular dynamics simulations. It is demonstrated that these types of artificial neural networks can be learned from the distribution of the coordinates of atoms over molecular dynamics simulations. As shown, the trained multivariate time series model enables forecasting trajectory paths of atoms accurately and in much reduced time with over 96% accuracy.