Modelling photoactive soft materials typically requires combining classical and quantum mechanical simulation protocols to describe the impacts of molecular conformation on electronic structure. Due to the strongly interacting nature of many conjugated moieties, the sampling of conformational space over mesoscopic length and time scales can be particularly demanding. Coarse-grained (CG) models can allow access to longer length and time scales, but must be combined with atomistic back-mapping to derive electronic structure; this process is computationally intensive and not guaranteed to be consistent with the spatial resolution of the coarse-graining. A supervised machine learning technique, denoted Artificial Neural Network Electronic Coarse-Graining (ANN-ECG), is described that directly maps conformationally dependent electronic structure, averaged over the decimated degrees of freedom, to CG pseudoatom configurations, thereby accelerating simulations via the elimination of back-mapping and repeated quantum-chemical calculations. We demonstrate how ANN-ECG can be used to identify computationally optimal levels of coarse-graining, and demonstrate applications of the method to polymeric systems and field-based coarse-grained variables.

Coarse-grained electronic structure in organic semiconductors via supervised machine learning

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.