The authors have developed a new local hybrid functional, CHYF, that is designed from first principles to be general and transferable across a wide range of applications in quantum physics, chemistry, and materials science. The functional shows excellent performance for a variety of properties, including thermochemistry, excitation energies, magnetic properties, and NMR parameters, while being numerically robust and requiring only small computational grids.
Geometric Graph Neural Networks leverage physical symmetries to model 3D atomic systems accurately.