Deep Learning of Electron Density for Predicting Energies: The Case of Boron Clusters

Determination and prediction of atomic cluster structures is crucial in nanoclusters/materials research. The molecular structure greatly influences, if not determines the properties of nanoclusters. While traditional quantum chemical calculations are time-consuming, structure prediction approaches are computationally expensive. In this chapter, we introduce a convolutional neural network that is able to evaluate, with reasonable accuracy the electronic energies for the ground state of nanoclusters using the promolecule electron density. This model, which is applied to the pure neutral boron clusters as an aid in their structure prediction, can be utilized for both regression and classification purposes.