Accelerated discovery of Magnesium-based amorphous alloys through a property-driven active learning strategy

Magnesium (Mg)-based amorphous alloys hold significant potential for applications in the automotive, aerospace, and biomedical industries. However, they are limited by their smaller size compared to other amorphous alloys. A higher reduced glass transition temperature ( T rg ) is associated with larger sizes in Mg-based amorphous alloys. Yet, due to the vast chemical space involved, designing Mg-based amorphous alloys with higher T rg using traditional ‘trial and error’ method is a challenging endeavor. In this work, we developed a property-driven active learning strategy to customize Mg-based amorphous alloys with enhanced T rg . After just two iterations, we successfully tailored four amorphous alloys with high T rg values. Under identical experimental conditions, two of these alloys exhibited T rg values surpassing that of Mg 0·65 Ag 0·19 Cu 0.06 Gd 0.1 , the alloy with best T rg value in the reported references. SHAP analysis revealed that T rg tends to be higher when the Ag atomic ratio exceeds 0.045, the Cu atomic ratio is below 0.18, the Ni atomic ratio is below 0.025, and the Mg atomic ratio is below 0.665. Our work offers a reliable strategy for designing Mg-based amorphous alloys with higher T rg and provides valuable insights for the rational design of these alloys.