Interpretable descriptors enable prediction of hydrogen-based superconductors at moderate pressures

Jiawei Chen; Junhao Peng; Yanwei Liang; Renhai Wang; Huafeng Dong; Wei Zhang

Paper

Interpretable descriptors enable prediction of hydrogen-based superconductors at moderate pressures

Abstract

Room temperature superconductivity remains elusive, and hydrogen-base compounds despite remarkable transition temperatures(Tc) typically require extreme pressures that hinder application. To accelerate discovery under moderate pressures, an interpretable framework based on symbolic regression is developed to predict Tc in hydrogen-based superconductors. A key descriptor is an integrated density of states (IDOS) within 1 eV of the Fermi level (EF), which exhibits greater robustness than conventional single-point DOS features. The resulting analytic model links electronic-structure characteristics to superconducting performance, achieves high accuracy (RMSEtrain = 20.15 K), and generalizes well to external datasets. By relying solely on electronic structure calculations, the approach greatly accelerates materials screening. Guided by this model, four hydrogen-based candidates are identified and validated via calculation: Na2GaCuH6 with Tc =42.04 K at ambient pressure (exceeding MgB2), and NaCaH12, NaSrH12, and KSrH12 with Tc up to 162.35 K, 86.32 K, and 55.13 K at 100 GPa, 25 GPa, and 25 GPa, respectively. Beyond rapid screening, the interpretable form clarifies how hydrogen-projected electronic weight near EF and related features govern Tc in hydrides, offering a mechanism-aware route to stabilize high-Tc phases at reduced pressures.

Parse Issues:

(1x) Found unknown command: preprint