Accelerating the discovery of novel materials using a machine learning guided framework

Alert
Current COVID-19 Community Level: LOW (Updated Feb. 3) CDC COVID-19 County Check
La-Si-P compound
130 La-Si-P compounds within 100 meV/atom from the convex hull are discovered. La5SiP3 and La2SiP phases are only 2 and 10 meV/atom respectively above the convex hull. These two structures are dynamically stable at least up to 1000 K and would be synthesizable. 

Scientific Achievement

Novel low-energy La-Si-P ternary compounds are efficiently predicted by an ML-guided framework which dramatically (~100 times) accelerates the pace of materials discovery.

Significance and Impact

The ML-guided framework can be applied to any systems that are not feasible by traditional algorithms or high throughput ab initio calculations.

Research Details

  • Effectively integrates ML tools and databases with ab initio calculations and experiment.
  • Adapt deep ML model for fast prediction of composition-structure-energy relationship in complex ternary systems.
  • The approach identifies a small set of promising compositions/structure by ML for further refining by ab initio calculations, thus accelerating (~100 times) the discovery.
 
H. J. Sun, C. Zhang, W. Xia, L. Tang, R. Wang, G. Akopov, N. W. Hewage, K.-M. Ho, K. Kovnir, C. Z. Wang, Inorganic Chemistry 61, 16699 (2022). https://doi.org/10.1021/acs.inorgchem.2c02431