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.