Deep Learning Enhanced Density Functional Theory (DFT) Simulation for Millions of Atoms

Deep Learning Enhanced Density Functional Theory (DFT) based electronic structure simulation for Millions of Atoms (DeepDFT). We are developing a deep learning platform that can learn from millions of previously performed quantum mechanical DFT simulations to replace state of the art - an expensive iterative process, with linear scaling and parallelizable method which will make large-scale electronic structure simulations feasible. Training with a dataset of trillions of data points with very large feature vectors. Defining new fingerprint operators and functional architecture. Testing models with hundreds of millions of parameters.