(622c) Machine Learning-Enabled Prediction of Electronic Properties of Radical Polymers at Coarse-Grained Resolutions

Non-conjugated radical polymers are conducting polymers with non-conjugated backbones bearing pendant stable radical sites. These polymers promise sustainable and tunable materials for applications in, for example, all-organic energy storage devices. However, a fundamental understanding of how the chemical structure impacts charge transport in these materials is still lacking. To address this, efficient computational schemes that take into account both electronic properties and mesoscale morphological features are required.

We present an efficient computational scheme that uses supervised machine learning (ML) to predict electronic-structure information relating to charge transport at coarse-grained (CG) resolutions. ML models are trained on data obtained from quantum chemical calculations on all-atom conformations sampled from condensed-phase simulations. Predictions of several electronic properties are investigated as a function of CG resolution. Trained ML models then enable electronic property predictions directly from CG polymer morphologies. We evaluate both systematic and building-block coarse-graining techniques. We validate the approach by comparing to state-of-the-art methodologies that require backmapping to atomistic resolution and subsequent quantum chemical calculations. The ML-assisted scheme shows potential for drastically accelerating multiscale computational workflows that connect electronic properties to the mesoscale, thereby enabling high-throughput modeling aimed at the understanding and design of radical polymers and other soft electronic materials.