Discovery of metastable P–Si–Sn ternary compounds aided by machine learning

Abstract The crystalline structures of energetically favorable P–Si–Sn ternary compounds containing two group 14 elements Si and Sn are explored using a combined machine learning approach and first-principles calculations. While no binary Si-Sn compounds have been reported in the literature, several energetically favorable metastable P–Si–Sn ternary compounds have been predicted from this study. The crystalline structures and electronic properties of five low-energy metastable compounds, namely, Pnma -P 3 SiSn, Cmc 2 1 -P 5 SiSn 2 , Pnma -P 2 SiSn, P 2 1 / c -P 14 Si 8 Sn 3 , and P 2 1 2 1 2 1 -P 4 SiSn, are investigated. In these compounds, Si and Sn atoms are separated by P atoms, forming distinct structural motifs, such as Si-centered tetrahedra (SiP 4 ), Sn-centered octahedra (SnP 6 ), and P chains. The electronic properties vary among these metastable compounds: Pnma -P 3 SiSn and Cmc 2 1 -P 5 SiSn 2 exhibit metallic behavior, whereas Pnma -P 2 SiSn, P 2 1 / c -P 14 Si 8 Sn3, and P 2 1 2 1 2 1 -P 4 SiSn are semiconductors. These compounds were dynamically stable at ambient conditions, indicating the possibility of experimental synthesis. The results demonstrate the effectiveness of combining machine learning and first-principles approaches to discover and characterize novel materials, thereby providing valuable insights into the advancement of materials science and technology.