Scalable electronic predictions are critical for soft materials design. Recently, the Electronic Coarse-Graining (ECG) method was introduced to renormalize all-atom quantum chemical (QC) predictions to coarse-grained (CG) molecular representations using deep neural networks (DNN). While DNN can learn complex representations that prove challenging for traditional kernel-based methods, they are susceptible to overfitting and the overconfidence of uncertainty estimations. Here, we develop ECG within the GPU-accelerated Deep Kernel Learning (DKL) framework to enable CG QC predictions of a conjugated oligomer using range-separated hybrid density functional theory. DKL-ECG provides accurate reproduction of QC electronic properties in conjunction with prediction uncertainties that facilitate efficient training over multiple temperature data sets via active learning. We show that while active learning algorithms enable efficient sampling of a more diverse configurational space relative to random sampling, the predictive accuracy of DKL-ECG models is effectively identical across all active learning methodologies employed. We attribute this result to the low conformational barriers of our test molecule and the redundant sampling of configurations induced by Boltzmann sampling, even for distinct temperature ensembles.
Coarse-Grained Density Functional Theory Predictions via Deep Kernel Learning
