Machine learning band gaps from the electron density
Javier Robledo Moreno, Johannes Flick, Antoine Georges
Abstract
The accurate estimation of band gaps of solid-state materials is of great relevance for modern optoelectronic, electronic, and photovoltaic applications. However, the precise $a\phantom{\rule{0}{0ex}}b$ $i\phantom{\rule{0}{0ex}}n\phantom{\rule{0}{0ex}}i\phantom{\rule{0}{0ex}}t\phantom{\rule{0}{0ex}}i\phantom{\rule{0}{0ex}}o$ computation of accurate band gaps is a resource intensive task. In this article, inspired by the Hohenberg-Kohn theorem of density-functional theory, the authors demonstrate the possibility of explicitly parametrizing the mapping between the electron density in the unit cell and the corresponding experimental band gap in real materials using deep neural networks. The proposed data-driven approach achieves accuracies comparable to state of the art $a\phantom{\rule{0}{0ex}}b$ $i\phantom{\rule{0}{0ex}}n\phantom{\rule{0}{0ex}}i\phantom{\rule{0}{0ex}}t\phantom{\rule{0}{0ex}}i\phantom{\rule{0}{0ex}}o$ approaches, at a much lower computational cost.