Design and characterization of LaCo 1 − x Y x O 3 by machine learning

A physics-informed Gaussian Process Regression (GPR) model with monotonic baseline captured the decreasing trend of the magnetic excitation energy Δ 1 upon Y substitution with quantified uncertainty. Under 3-fold cross validation, the model achieved test RMSE = 7.21 ± 1.18 K and R 2 = 0.907 ± 0.049 (aggregate across all test points: RMSE = 7.28 K , R 2 = 0.935 ), while underestimating Δ 1 for LaCoO 3 (predicting 132 K vs. experimental 160 K ). At x = 0.05 , the model predicted Δ 1 = − 48.7 ± 7.5 K , indicating a transition toward a magnetic ground state; experimentally, the low characteristic energy ( ∼ 45 K ) is consistent with this trend. Uncertainty-aware screening (UCB, κ = 1.0 ) highlighted Y as a practical top candidate among feasible B-site trivalent substituents. • Physics-informed Gaussian process predicts Δ I trends with uncertainty. • Uncertainty-aware screening identifies Y 3+ as non-intuitive B-site dopant. • Y substitution reduces Δ I from 160 K to 45 K in LaCo 1 − x Y x O 3 . • Anisotropic lattice distortion ( c / a +0.43 %, V ±0.3 %) weakens crystal field. • Data-driven approach identifies rare-earth as viable B-site substituent.