Machine learning accurate exchange and correlation functionals of the electronic density*

Here, we review recent efforts to use machine learning (ML) methods for the creation of density functionals. We showcase our own framework, NeuralXC, which is based on a projection of the electron density onto localized atomic orbitals and a functional parametrized by neural networks. The functionals thus created are designed to lift the accuracy of a baseline method towards that provided by more accurate reference calculations, all while maintaining their efficiency. We show that a meaningful representation of the physical information contained in the training data is learned, making the functionals transferable across systems. Challenges on the path to a truly universal ML-functional are outlined and possible future approaches are discussed.

Dick, Sebastian, and Marivi Fernandez-Serra. "Machine learning accurate exchange and correlation functionals of the electronic density." Nature communications 11.1 (2020): 1-10.