Predicting thermodynamic stability of magnesium alloys in machine learning

Density functional theory (DFT) have been widely used to screen thermodynamically stable material; however, its high computational cost limits its use. In this paper, we explore the use of DFT data from high-throughput calculations to create faster machine learning (ML) models that can be used to screen thermodynamically stable magnesium alloy materials. Our methods work by utilizing the kernel ridge regression (KRR) algorithm, as well as Deep Potential Molecular Dynamics (DeePMD) to train ML models for predicting the formation energy of magnesium alloys. The accuracy, stability, and generalization ability of the ML models created under both methods are evaluated in detail. Meanwhile, we have conducted in-depth comparative analysis of the two methods, which concluded that the accuracy of DeePMD model performs better and time efficiency of KRR model has more advantages. The results show that the best performing DeePMD model and KRR model achieve the RMSE of 0.43 meV/atom and 6.80 meV/atom, indicating that our methods provide a reliable idea for obtaining the formation energy of magnesium alloys.