BMach automates the choice of the Hubbard U parameter in DFT+U. It uses Gaussian process regression and physics-informed loss functions to fit U against a reference band structure from hybrid or quasiparticle (G0W0) calculations, so that a cheap DFT+U run reproduces the target electronic structure.
BMach: a Bayesian machine for optimizing Hubbard U parameters in DFT+U with machine learning
R. Das. arXiv preprint (2024). Preprint.
AI/MLPhysicsBayesian optimizationGaussian processesMachine learningDFT+U