Machine learning and first-principles calculations for the prediction and analysis of superconductivity in Mg–B–N systems

Abstract

Light-element compound superconductors have attracted more and more attention due to their unique electronic properties and broad applicability. In this work, we conducted an efficient search for potential Mg–B–N superconducting materials. By combining element substitution with structural perturbation, we synthesized a dataset of 1 115 435 hypothetical Mg–B–N materials. We found several promising candidates within the Mg–B–N dataset through first-principles calculations combined with machine learning (ML). Under ambient pressure, materials with specific layered crystal structures, including I4mm-Mg₂BN, Cm-Mg₂BN, Cmmm-MgB₂N, and R3m-Mg₂BN, exhibited superconductivity with the superconducting transition temperature (T_c) of 31 K, 19 K, 11 K, and 4.5 K, respectively. We comprehensively analyzed the structural stability, electronic properties, and superconducting behavior of these materials. Phonon dispersion analysis has highlighted the critical role of lattice vibrations in electron–phonon coupling (EPC), where the low-frequency vibrations of Mg and B atoms are essential for facilitating electron pairing and the onset of superconductivity. This systematic approach, which combines precise first-principles calculations with efficient machine learning techniques, accelerates the discovery of superconductors.