BMach treats the choice of U (and of hybrid-functional parameters) as an optimization problem over a Gaussian-process surrogate, so that DFT-based workflows reach the accuracy of much more expensive methods with far fewer calculations.
BMach, Bayesian optimization of Hubbard U
A Bayesian framework that calibrates Hubbard U and hybrid-functional parameters at a small fraction of the cost of direct high-level methods.
Earlier work · 2023–2024
Related publications
AI/MLPhysicsBayesian optimizationGaussian processesMachine learningDFT+UHybrid functionals