Electron Affinity of Actinide (IV) Carboxylate Complexes From MN12-L Density-Functional Calculations and Explainable...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Actinide(IV) carboxylate complexes are crucial in nuclear fuel cycles and radiochemistry.
• Understanding their electronic properties, particularly electron affinity (EA), is vital for predicting reactivity and stability.
• Relativistic effects are significant for heavy elements like actinides, influencing their electronic structure.

Purpose of the Study:

• To computationally determine the vertical electron affinity (EA) of 14 homologous actinide(IV) carboxylate complexes.
• To investigate the electronic structure and bonding characteristics influencing the EA across the actinide series.
• To develop and validate an explainable machine learning model for predicting actinide EA.

Main Methods:

• Relativistic density-functional theory (DFT) calculations using the MN12-L functional and appropriate basis sets/effective core potentials.
• Systematic exploration of all physically accessible spin multiplicities for each actinide complex.
• Application of explainable machine learning (ML) with SHapley additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• A monotonic increase in vertical electron affinity (EA) was observed across the actinide series (Th-Lr), from 5.53 eV to 10.80 eV for propionates.
• The lowest unoccupied molecular orbital (LUMO) composition shifts from predominantly metal 5f-orbitals to significant carboxylate π* character at the end of the series (Lr).
• A nine-feature ML model achieved high predictive accuracy (R² = +0.70) and identified 5f-electron count and atomic spin-orbit constant as dominant predictors.

Conclusions:

• The 5f-electron count is the primary physical factor governing the reduction propensity of actinide(IV) carboxylate complexes.
• The developed DFT-ML approach provides a reliable quantitative reference for actinide carboxylate EAs.
• Excellent cross-ligand transferability of the ML model suggests broad applicability for similar actinide systems.