Unveiling the Prediction of Lanthanide Binding Selectivity: Could 4f Electrons be Treated as the Core–Shell?

There has been a recent resurgence of interest in the lanthanide separation chemistry. With the advent of artificial intelligence, the high-throughput computation provides a powerful alternative to experimental screening of extractants. However, the reliable treatment of 4f electrons remains a major bottleneck due to their highly localized and strongly correlated nature. In this work, we systematically benchmark two prominent computational strategies including 4f-in-valence and 4f-in-core for calculating lanthanide binding selectivity. Within the 4f-in-valence framework, density functional theory (DFT) calculations yield irregular energetic trends across the lanthanide series. Multi-configurational and multi-reference methods can replicate the experimental selectivity trends, but they demand careful selection of the active space and the treatment of dynamic correlation, posing considerable computational difficulties. In contrast, the 4f-in-core approach, which excludes the chemically inert 4f orbitals from the valence space via 4f-in-core pseudopotentials (PPs), offers substantial improvements in computational efficiency. More importantly, the present results demonstrate that this scheme reliably reproduces key energetic trends, validating its applicability in non-redox lanthanide separation scenarios. This work establishes an efficient pathway by using 4f-in-core PPs for high-throughput computational screening of selective extractants.