2d Materials and Electronic Interactions Density Functional Theory and Benchmarking Open Databases for Machine Learning

Abstract

Two-dimensional (2D) materials, with their unique blend of electronic, mechanical, optical and newlinethermal properties, are rapidly gaining prominence in a multitude of technological applications. newlineModifications on 2D materials such as heteroatom doping/decoration, vacancy formation, edge newlinedoping, intercalation, bilayer twist angles etc. have been known to alter the electronic and charge newlineenvironment of 2D materials. Such modifications have been shown to enhance vital properties newlinesuch as catalytic activity, gas sensing, electronic conductivity, and magnetism of materials. Two newlinesuch instances of different metal adatoms enhancing the binding of hydrogen gas on 2D hexagonal newlineboron nitride (h-BN) and elevating the electronic transmission along the edges of 2D MoS2 were newlinestudied using ab initio density functional theory (DFT) studies. In the first case, decoration of rare newlineearth metal, Ce creates conducive geometrical and charge sites on h-BN surface for adsorbing newlinehigher number of hydrogen molecules. Using quantum transport calculations, Au doping along the newlineetched edges of MoS2 was shown to open transmission channels for electron transport. Thus, by newlinemodifying various 2D materials with heteroatoms, we can hope to design novel materials for newlinevarious applications. However, identifying or designing the right 2D materials for a targeted newlineapplication is a non-trivial task because of a huge number of combinatorial possibilities. While newlineexperiments on 2D materials and their various modifications are resource and time intensive, so newlineare physics based first principles calculations. Unlike 3D bulk materials, the unavailability of large newlinedatabases for 2D materials poses a unique challenge for most advanced machine learning (ML) newlineprotocols. We employ the recently developed crystal graph convolutional neural network newline(CGCNN) to benchmark some of the major open databases of 2D materials for predicting both newlinetheoretical and experimental properties of 2D materials.