This project will enable predictive large-scale atomistic materials simulations that approach quantum-level accuracy by developing an automated framework for discovering and generating optimal fitted potentials with rigorous error bounds that do not require experimental input. The outcome will be a broad capability for predicting the complex physical and chemical responses of materials under a wide variety of conditions, dramatically expanding the range of applicability of atomistic materials simulations for critical NNSA and broader mission applications, leveraging high-performance computing resources at the exascale and beyond.