Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

TT: Fachverband Tiefe Temperaturen

TT 34: Superconductivity: Theory

TT 34.13: Talk

Wednesday, March 19, 2025, 18:15–18:30, H36

Describing superconductivity through interpretable artificial intelligence — •Herzain I. Rivera-Arrieta, Lucas Foppa, and Matthias Scheffler — The NOMAD Laboratory at the Fritz Haber Institute of the Max Planck Society, Berlin, Germany

Superconductivity is governed by an intricate interplay among electronic structure, lattice vibrations, and pressure effects, among many other phenomena [1]. Thus, a (single) physical model might not be enough to describe superconductivity. Interpretable artificial intelligence (AI) can provide valuable insights into the underlying mechanisms driving superconductivity, e.g., in conventional superconductors. Herein, we compile a dataset containing approximately 1,000 materials [2] and a diverse range of compositional, structural, electronic, and phonon-related properties. Then, we employ the symbolic-regression SISSO and subgroup discovery AI approaches [3, 4], to identify the few, key physicochemical parameters correlated with a superconductor’s critical temperature. This approach is a step towards identifying the “materials genes” [5] of superconductivity.

[1] X. Gui, B. Lv, and W. Xie, Chem. Rev., 121, 2966 (2021).

[2] K. Choudhary, and K. Garrity, Npj. Comut. Mater., 8, 244 (2022).

[3] R. Ouyang, et al., Phys. Rev. Mat., 2, 083802 (2018).

[4] S. Wrobel, 1st Europ. Symp. on Princ. of Data Min. and Knowl. Discov., 19, 78 (1997).

[5] L. Foppa, et al., MRS bulletin, 46, 1016 (2021).

Keywords: Superconductivity; Artificial Intelligence; Symbolic regression; Subgroup discovery