Applying Kernel Ridge Regression to Predict Dielectric Properties of Ceramic Oxides

As renewable energy sources become more prevalent, the demand for energy storage devices with large energy storage capacity and high power density has also increased. These ultrahigh capacitance energy storage devices require materials with the rare combination of a large dielectric constant and high dielectric strength. We use kernel ridge regression to predict new ultrahigh capacitance materials, focusing our search on oxides with the general formula ABO_n (where 1 ≤ n ≤ 4) and primarily using compositional information to train the model. We systematically vary the training data in order to investigate the interplay between oxide composition and dielectric behavior before using our top-performing model to predict the dielectric constant and band gap for new compositions. After identifying several promising materials, we use DFT calculations to validate our predictions.