Machine‐Learning‐Assisted Determination of the Global Zero‐Temperature Phase Diagram of Materials

• Published in: Advanced materials (Weinheim) Vol. 35; no. 22; pp. e2210788 - n/a
• Main Authors: Schmidt, Jonathan, Hoffmann, Noah, Wang, Hai‐Chen, Borlido, Pedro, Carriço, Pedro J. M. A., Cerqueira, Tiago F. T., Botti, Silvana, Marques, Miguel A. L.
• Format: Journal Article
• Language: English
• Published: Germany Wiley Subscription Services, Inc 01.06.2023
• Subjects: Apexes; Convexity; Crystals; Datasets; Extreme values; Graph neural networks; high‐throughput density functional theory calculations; Inhomogeneity; Machine learning; machine learning material science; material discovery; Materials science; Neural networks; Phase diagrams; Stability; Superconductivity; superhard materials; high-throughput density functional theory calculations; material discovery; superconductivity; superhard materials; machine learning material science
• ISSN: 0935-9648; 1521-4095; 1521-4095
• DOI: 10.1002/adma.202210788

Crystal‐graph attention neural networks have emerged recently as remarkable tools for the prediction of thermodynamic stability. The efficacy of their learning capabilities and their reliability is however subject to the quantity and quality of the data they are fed. Previous networks exhibit strong biases due to the inhomogeneity of the training data. Here a high‐quality dataset is engineered to provide a better balance across chemical and crystal‐symmetry space. Crystal‐graph neural networks trained with this dataset show unprecedented generalization accuracy. Such networks are applied to perform machine‐learning‐assisted high‐throughput searches of stable materials, spanning 1 billion candidates. In this way, the number of vertices of the global T = 0 K phase diagram is increased by 30% and find more than ≈150 000 compounds with a distance to the convex hull of stability of less than 50 meV atom−1. The discovered materials are then accessed for applications, identifying compounds with extreme values of a few properties, such as superconductivity, superhardness, and giant gap‐deformation potentials. By combining crystal graph attention networks with a newly created dataset, a compound space spanning 1 billion materials is scanned for new stable materials. As a result, the number of known theoretically synthesizable materials is increased by 30%. The resulting materials are then studied for extreme values of several properties.