Abstract
We study two-dimensional (2D) magnetic materials using state-of-the-art machine learning models that use a graph-theory framework. We find that representing materials as graphs allows us to better learn structure-property relationships by leveraging both the chemical properties of the constituent atoms and the connectivity between those atoms. Graph neural network models are capable of predicting global properties of crystal structure (i.e., graphwise properties) and local properties of the constituent atoms (i.e., nodewise properties). We embed physical constraints into our model by simultaneously making predictions of local and global properties. In particular, we use the atomistic line graph neural network (ALIGNN) architecture. We train the ALIGNN model on data comprising local and global magnetic moments of 314 2D structures of the form $\mathrm{Cr}{A}^{\text{ii}}{B}^{\text{i}}{B}^{\text{ii}}{X}_{6}$, based on monolayer ${\mathrm{Cr}}_{2}{\mathrm{Ge}}_{2}{\mathrm{Te}}_{6}$, calculated from first principles. By learning the relationships between both local and global magnetic properties, we demonstrate an improvement over models that only consider global magnetic properties.
Funding Information
  • National Science Foundation (ACI-1548562, 2044842)
  • Office of Science (DE-AC02-06CH11357)
  • National Institute of Standards and Technology

This publication has 48 references indexed in Scilit: