Machine Learning Predicts Two New Kagome-Lattice Superconductors, YRu3B2 and LuRu3B2, Before Either Was Synthesized
A machine-learning screening pipeline predicted two new superconductors, YRu3B2 and LuRu3B2, which Rice University then synthesized and confirmed.
Overview
A machine-learning screening pipeline has helped physicists identify two superconductors that did not exist in any laboratory before the algorithm flagged them as promising candidates. The compounds, YRu3B2 and LuRu3B2, were predicted computationally and then synthesized and experimentally confirmed by a team at Rice University, according to Aalto University, whose Professor Päivi Törmä leads the international SuperC consortium behind the work. The findings were published in Physical Review Research on June 17, 2026, according to EurekAlert!.
What We Know
Both new compounds are built around a kagome lattice, an atomic arrangement that gets its name from a hexagonal Japanese basket-weaving pattern. According to Phys.org, “YRu3B2 and LuRu3B2 gain their superconductivity from electrons forming flat bands in a kagome lattice, named after a hexagonal Japanese basket-weaving pattern.”
According to the research paper posted on arXiv, the team measured superconducting critical temperatures of Tc = 0.81 K for YRu3B2 and Tc = 0.95 K for LuRu3B2, confirmed through magnetization and specific-heat measurements. The paper describes the discovery as the result of “machine learning-accelerated high-throughput screening combined with first principles calculations,” which flagged the two compounds as candidates before either was made in a lab.
Once the computational predictions pointed to viable candidates, a Rice University team led by Professor Emilia Morosan synthesized the materials by chemically combining their constituent elements, then verified experimentally that both were indeed superconducting, according to Aalto University.
Törmä framed the discovery against the historically slow, trial-and-error pace of superconductor research. “Over the decades researchers have recognised over 7,000 superconductors, but mostly serendipitously,” she explained. Describing the new approach, she said: “Our method uses machine-learning-based pre-screening followed by targeted calculations on the promising candidates. This approach will greatly speed up superconductor discovery in the future. With machine learning, we may be able to push the number of materials we can process into the billions.”
The work comes out of the SuperC consortium, which Törmä and a group of physicists formed in 2023 as what Aalto University describes as “the first coordinated global collaboration to find new superconductors,” with a stated aim of finding a room-temperature superconductor by 2033. Törmä tied that long-term goal to the discovery of YRu3B2 and LuRu3B2: “Superconductive materials that can operate at room temperature would forever change the way we consume energy,” she said, according to Aalto University.
ScienceDaily confirmed the paper appeared in Physical Review Research, volume 8, issue 2, and that Törmä at Aalto University and Morosan at Rice University led the theoretical and experimental sides of the collaboration, respectively.
What We Don’t Know
Both YRu3B2 and LuRu3B2 become superconducting only below 1 kelvin, far colder than liquid nitrogen temperatures, let alone room temperature, so neither compound is close to practical use on its own. The published material does not specify which additional elemental combinations the SuperC consortium’s algorithm has flagged next, nor how many candidate materials the screening pipeline has evaluated to date beyond the description that machine learning could eventually let researchers process material candidates “into the billions.” It also remains to be seen how quickly the pre-screening method can be applied to the wider search for higher-temperature superconductors, since YRu3B2 and LuRu3B2 serve primarily as a proof of concept for the discovery pipeline itself.
Analysis
The significance of the result lies less in the two compounds themselves — both are low-temperature superconductors of limited immediate application — than in the demonstration that a machine-learning-guided pipeline can predict an undiscovered superconductor accurately enough that a synthesis lab can go on to make and verify it. Of the more than 7,000 superconductors identified across decades of research, the vast majority were found through direct experimentation rather than prediction. A validated computational shortcut that can be pointed at a much larger space of elemental combinations would mark a meaningful change in how the field searches for new candidates, including the higher-temperature materials that remain the field’s long-sought goal.