Machine Learning Improves Stability Prediction of High-Entropy Alloys

Strategic Insights -

Research from National Tsing Hua University has investigated the use of machine learning to predict the stability of high-entropy alloys (HEAs). The study demonstrates the importance of considering both microstructure-based and electronic-structure-based descriptors in achieving accurate correlation of dopant stability in HEAs. The researchers employed a machine-learning interatomic potential and Monte Carlo simulations to verify the short-range order in the HEA, achieving improved accuracy with the addition of volume and electronic-structure-based descriptors.

Key Takeaways:

  • The study focused on the C- or N-doped VNbMoTaWTiAl (BCC) HEA, using density functional theory (DFT) calculations to predict stability.
  • Six types of microstructure-based descriptors and seven types of electronic-structure-based local-environment descriptors were considered to predict the stability of the HEA.
  • The best correlation between the descriptor and the doping energy was found for 1NN with coefficient of determination (R^2) values of 51 or 61% obtained using the LOOCV approach for C or N doping, respectively.
  • The addition of volume and electronic-structure-based descriptors to the linear regression model improved the accuracy, with R^2 values of 72, 76, 75, and 80% for C and N doping, respectively.
  • The study demonstrated the importance of considering both microstructure-based and electronic-structure-based descriptors to achieve accurate correlation of dopant stability in HEAs.
  • The combined approaches used in this study could be further applied to other materials systems, research fields, and applications.

Statistics:

  • 51 or 61%: Coefficient of determination (R^2) values obtained using the LOOCV approach for C or N doping, respectively.
  • 72 and 76%: R^2 values for C and N doping, respectively, after adding volume descriptors into the linear regression model with the 1NN descriptor.
  • 75 and 80%: R^2 values for C and N doping, respectively, after further adding the electronic-structure-based EP descriptor.

Sources:

  • "Optimising descriptors to correlate stability of C- or N-doped high-entropy alloys: a combined DFT and machine-learning regression study" by Chih-Heng Lee, Jyh-Wei Lee, and Hsin-Yi Tiffany Chen (Faraday Discussions, 2025)