Machine learning assisted screening of non-rare-earth elements for Mg alloys with low stacking fault energy

[Display omitted] •A machine learning model to predict stacking fault energies was constructed.•The model predictions were verified by density functional theory computations.•Two novel Mg alloys with high density of 〈c+a〉 dislocations were presented. Improved density of 〈c+a〉 dislocations is indispensable for enhancing the ductility of Mg alloys. More 〈c+a〉 dislocations can be activated by reducing the basal stacking fault energies (SFEs) through the addition of rare-earth (RE) elements, but they are rare and costly. Therefore, it is worthwhile to develop a screening criterion of RE elements free Mg alloys with low SFEs, especially when the space of candidate materials grows significantly with increasing number of alloying components. In the present work, a non-linear functional form with the identified most important atomic features (volume, first ionization energy, and bulk modulus) was established via machine learning (ML) to predict the values of SFEs computed by density functional theory (DFT). The ML model was then applied to estimating the SFEs of 300 ternary RE elements free Mg alloy systems. The predicted results of several promising candidates were successfully validated by additional laborious DFT computations. Out of them, two candidate alloys with novel compositions were fabricated and demonstrated to have high density of 〈c+a〉 dislocations. The proposed ML strategy shows broad applicability and potential in the rapid discovery of ductile multi-component Mg alloys without RE elements.