A universal model for predicting the stability of complexes with rare earths and trivalent actinides.

Despite significant interest in rare earth and actinide complexes, experimental characterization is challenging due to element scarcity, cost, and safety requirements, while accurate quantum chemical modeling remains computationally intractable for relevant system sizes. We overcome these barriers by introducing a novel machine learning architecture. Trained on available experimental data spanning multiple f-block elements, our model achieves significantly improved predictive accuracy for complex stability. Analysis of its applicability domain and fragment contributions reveals critical structural determinants of stability.