Machine Learning-Guided Discovery of Kagome Superconductors YRu3B2 and LuRu3B2

Abstract

We report the experimental discovery of bulk superconductivity in two kagome lattice compounds, YRu$_3$B$_2$ and LuRu$_3$B$_2$, which were predicted through machine learning-accelerated high-throughput screening combined with first principles calculations. These materials crystallize in the hexagonal CeCo$_3$B$_2$-type structure with planar kagome networks formed by Ru atoms. We observe superconducting critical temperatures of $T_{c} = 0.81$~K for YRu$_3$B$_2$ and $T_{c} = 0.95$~K for LuRu$_3$B$_2$, confirmed through magnetization and specific heat measurements. Both compounds exhibit nearly 100\% superconducting volume fractions, demonstrating bulk superconductivity. Compared with LaRu$_3$Si$_2$, YRu$_3$B$_2$ and LuRu$_3$B$_2$ show a more dispersive Ru local $d_{x^2-y^2}$ quasi-flat band (and thus a reduced DOS at $E_F$) together with an overall hardening of the phonon spectrum, both of which lower the electron-phonon coupling (EPC) constant $λ$. Meanwhile, the dominant real-space EPC between Ru local $d_{x^2-y^2}$ states and the low-frequency Ru in-plane local $x$ branch remains nearly unchanged, indicating that the reduction of $λ$ originates from the $d_{x^2-y^2}$ DOS reduction and the overall phonon hardening. Superfluid weight calculations show that conventional contributions dominate over quantum geometric effects due to the dispersive nature of bands near the Fermi level. This work demonstrates the effectiveness of integrating machine learning screening, first principles theory, and experimental synthesis for accelerating the discovery of new superconducting materials.

Source-reported materials — not catalogue approval

Similar papers