Rational Design of Perovskite Anode Materials in Solid Oxide Electrolysis Cells: Machine Learning-Assisted Prediction of Thermal Expansion Coefficients

Perovskite oxides have been recognized as promising electrode materials for solid oxide electrolysis cells. In this work, the phonon dispersion and thermal expansion coefficients (TECs) of LaBO₃ (B = Sc–Cu), LaFe_0.5B_0.5O₃ (B = Cr, Mn, Co, Ni), and LaCrO_3−δ and LaFeO_3−δ (δ = 0, 0.25, 0.5) have been studied by performing density functional perturbation theory calculations under the quasi-harmonic approximation. The calculated TECs agree well with experimental data, and LaFe_0.5Co_0.5O₃ and LaFe_0.5Ni_0.5O₃ have TECs close to those of the electrolyte yttria-stabilized zirconia (YSZ). A machine learning model is then developed to enable high-throughput screening of potential perovskite anode materials, where the data set is established based on our calculated TECs and experimentally reported values. The SHAP analysis indicates dominant factors governing the TEC include the cation radius and Mulliken electronegativity of the B-site element and the crystal gamma angle. Finally, 507 candidates compatible with YSZ are identified from 13,095 perovskite oxides.