Ligand Design for REE Recovery Using Artificial Intelligence

Abstract Rare earth elements (REEs) are designated as critical materials due to their urgent need and supply uncertainties. The scarcity of REE supply is exacerbated by environmentally damaging extraction practices, necessitating more efficient, sustainable extraction and separation methods. Traditional techniques such as solvent extraction are often constrained by their extensive organic solvent consumption and limited extraction efficiency, especially for e-waste feedstocks containing heterogeneous compositions. As an alternative to traditional methods, solid-phase extraction (SPE) technique streamlines REE extraction and separation by providing high chelating selectivity, fewer operational stages, and minimal secondary waste generation. This review explores recent advances in ligand design for REE recovery, especially carbonyl amide-based, phosphorus-containing, β-diketone- and acylpyrazolone-based, macrocyclic, and biologically derived ligands. Enhancing the surface density of more REE-selective ligands on solid-phase matrices provides a viable pathway to overcome the thermodynamic constraints of separating adjacent, chemically homologous REEs in urban mining processes. Direct applications of artificial intelligence (AI), particularly machine learning (ML), to experimentally validated REE recovery remain limited and are mainly focused on narrowly defined tasks, including metal–ligand stability-constant prediction, ligand ranking, extraction-equilibrium modeling, and process simulation. Rather than treating AI-assisted REE recovery as a mature field, this review critically distinguishes demonstrated REE-specific applications from transferable computational frameworks that remain to be validated. Particular attention is given to molecular descriptors, coordination geometry, thermodynamic properties, data quality, and validation requirements for predictive ligand design. By connecting established REE coordination chemistry with emerging AI methodologies, this review identifies current limitations and proposes a chemically grounded roadmap for future ligand discovery and process optimization.