Electronic-Structure Informatics Guided Identification of Actinide/Lanthanide Selectivity Factors.

Area of Science:

• Computational Chemistry
• Materials Science
• Inorganic Chemistry

Background:

• Actinide/lanthanide (An/Ln) selectivity is crucial for nuclear fuel reprocessing and waste management.
• Limited experimental data and complex electronic structures of open-shell f-elements hinder traditional computational approaches.
• Machine learning (ML) has faced challenges in this domain due to data scarcity and electronic complexity.

Purpose of the Study:

• To develop an interpretable electronic-structure informatics strategy for identifying ligand features that govern An/Ln selectivity.
• To enable accurate prediction of metal-ligand complex formation free energies and An/Ln selectivity differences.
• To provide a mechanistic understanding of selectivity drivers for actionable ligand design.

Main Methods:

• Systematic benchmarking of density functional theory (DFT) calculations for terpyridine-type ligands with Am3+, Cm3+, Eu3+, and Gd3+.
• Construction of compact electronic descriptors and application of Gaussian process regression (GPR) for predictive modeling.
• Utilizing Shapley Additive exPlanations (SHAP) to interpret model predictions and identify key governing features.

Main Results:

• A quantitative ML model predicting binding free energies (MAE = 17.5 kJ mol-1) and An/Ln selectivity differences (MAE = 2.6 kJ mol-1).
• Identification of metal ionization potentials and inner-site charge patterns as key drivers of An/Ln selectivity.
• Derivation of actionable design rules favoring electron-deficient azines and π-accepting terminal donors for enhanced An binding.

Conclusions:

• The developed ML framework provides a reproducible and mechanistically grounded approach for predicting intrinsic An/Ln selectivity.
• The electronic design variables derived are consistent with experimental trends and offer a practical basis for designing ligands for extraction processes.
• This interpretable strategy overcomes limitations of previous ML applications in An/Ln chemistry, facilitating targeted ligand development.