Advancing Rare-Earth Separation by Machine Learning.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Lanthanides are essential for many technologies but require efficient separation.
• Current rare-earth separation methods, like solvent extraction, rely on inefficient trial-and-error ligand discovery.

Purpose of the Study:

• To develop a predictive model for high-throughput screening of ligands for enhanced rare-earth separation.
• To accelerate the discovery of novel ligands for solvent extraction of lanthanides.

Main Methods:

• Utilized deep neural networks trained on experimental data.
• Employed a combined ligand representation using physicochemical descriptors and atomic extended-connectivity fingerprints.
• Validated the model by synthesizing and testing new ligands.

Main Results:

• The deep neural network model accurately predicts distribution coefficients for lanthanide ion solvent extraction.
• The combined ligand representation significantly improved model accuracy.
• Predicted values for newly synthesized ligands closely matched experimental measurements.

Conclusions:

• Machine learning, specifically deep neural networks, offers a powerful tool for accelerating ligand discovery in rare-earth separations.
• This approach enables high-throughput screening, overcoming the limitations of traditional methods.
• The developed model paves the way for discovering advanced ligands for critical material applications.