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

“New materials with the desired functional properties can now be designed expeditiously without experiments utilizing the power of quantum mechanical simulations of electrons around atoms. The state of the art technique for that is an iterative numerical procedure based on density functional theory (DFT). It is so computationally heavy that calculating the electron density for more than few hundred atoms (few nanometers) is arduous.
We will develop a deep neural network that can learn from millions of previously performed DFT simulations to predict electronic charge density around atomic structures without the expensive iterative process, with just the atomic structure as input. Linear scaling and parallelizable nature of this method will make large-scale electronic structure simulations feasible.
Success in this approach may lead to a paradigm shift in simulation-based research in physical, chemical and biological sciences, mechanical chemical and energy technology or medicine.”