Atomistic structure learning

• Published in: The Journal of chemical physics Vol. 151; no. 5
• Main Authors: Jørgensen, Mathias S., Mortensen, Henrik L., Meldgaard, Søren A., Kolsbjerg, Esben L., Jacobsen, Thomas L., Sørensen, Knud H., Hammer, Bjørk
• Format: Journal Article
• Language: English
• Published: Melville American Institute of Physics 07.08.2019
• Subjects: Algorithms; Artificial neural networks; Atomic interactions; Atomic properties; Chemical compounds; First principles; Grain boundaries; Graphene; Machine learning; Neural networks; Organic chemistry; Organic compounds; Physical chemistry; Physics; Quantum mechanics
• ISSN: 0021-9606; 1089-7690; 1520-9032; 1089-7690
• DOI: 10.1063/1.5108871

One endeavor of modern physical chemistry is to use bottom-up approaches to design materials and drugs with desired properties. Here, we introduce an atomistic structure learning algorithm (ASLA) that utilizes a convolutional neural network to build 2D structures and planar compounds atom by atom. The algorithm takes no prior data or knowledge on atomic interactions but inquires a first-principles quantum mechanical program for thermodynamical stability. Using reinforcement learning, the algorithm accumulates knowledge of chemical compound space for a given number and type of atoms and stores this in the neural network, ultimately learning the blueprint for the optimal structural arrangement of the atoms. ASLA is demonstrated to work on diverse problems, including grain boundaries in graphene sheets, organic compound formation, and a surface oxide structure.