Abstract

Abstract: Computational methods have revolutionised material discovery, driving interest in designing new advanced materials, e.g., superconductors, which exhibit zero resistivity below a critical temperature (Tc). Reaching pressures in the 100 GPa range enables the synthesis of superhydrides, a new class of hydrogen-rich materials that exhibit remarkable properties such as superconductivity, hydrogen diffusion, and hydrogen storage. A striking example is LaH₁₀, which holds the record for the highest superconducting transition temperature at Tc = 250 K [1,2], and showcases high hydrogen diffusion at high temperatures [3]. However, stabilising such compounds at ambient or low pressure remains a major challenge for practical applications. This has driven the search beyond binary hydrides toward ternary superhydrides, aiming to discover new superconductors with high critical temperatures that can persist near ambient pressure.

The discovery of new superconducting hydrides traditionally relies on ab initio crystal structure prediction and experimental synthesis. However, ab initio calculations become computationally prohibitive, e.g. for assessing the thermodynamic stability of complex systems such as ternary superhydrides, due to the extremely vast chemical space. To overcome this limitation, our group has successfully developed tailored machine learning interatomic potentials – ephemeral data-driven potentials (EDDPs) – enabling a great acceleration in structure prediction [4]. This approach was used to build the convex-hull of a new predicted metastable ambient-pressure hydrogen-based superconductor Mg2IrH6 with a Tc of 160 K [5].

This talk will trace the hunt for high-pressure superconductors, highlight how EDD Ps revolutionize our high-throughput search workflow, and showcase some of our recent prediction of a new ternary hydride superconductor [6].

References [1] M. Somayazulu et al., Phys. Rev. Lett., vol. 122 (2019) 027001 [2] A.P. Drozdov et al., Nature, vol 569 (2019) 528-531 [3] M. Caussé et al., Phy. Rev. B, vol 107 (2023) L060301 [4] C. J. Pickard, Phys Rev B, vol 106 (2022) 014102 [5] K. Dolui et al., Phys. Rev. Lett., vol 132 (2024) 16600 [6] M. Caussé et al. arXiv (2025) https://doi.org/10.48550/arXiv.2512.19901