Title: High-$T_c$ ternary metal hydrides, YKH$_{12}$ and LaKH$_{12}$, discovered by machine learning

Abstract: The search for hydride compounds that exhibit high $T_c$ superconductivity has been extensively studied. Within the range of binary hydride compounds, the studies have been developed well including data-driven searches as a topic of interest. Toward the search for the ternary systems, the number of possible combinations grows rapidly, and hence the power of data-driven search gets more prominent. In this study, we constructed various regression models to predict $T_c$ for ternary hydride compounds and found the extreme gradient boosting (XGBoost) regression giving the best performance. The best performed regression predicts new promising candidates realizing higher $T_c$, for which we further identified their possible crystal structures. Confirming their lattice and thermodynamical stabilities, we finally predicted new ternary hydride superconductors, YKH$_{12}$ [$C2/m$ (No.12), $T_c$=143.2 K at 240 GPa] and LaKH$_{12}$ [$R\bar{3}m$ (No.166), $T_c$=99.2 K at 140 GPa] from first principles.