(660e) Development of High-Accuracy Semiempirical Models with a Minimal Training Set

We have developed high accuracy semiempirical models for organic materials by leveraging a machine-learned force field based on Chebyshev polynomials. The benefits of our approach are: (1) many-body interactions can be corrected for in a systematic and rapidly tunable process, and (2) high-level quantum accuracy for a broad range of compounds can be achieved with less than 1% of data required for similar approach that couples semiempirical model with neural network. Though the model was trained only for gas phase configurations, it shows great transferability to condensed phases. The results for solid carbon phases, molecular crystals, and explosive materials will be discussed.