Machine-Learning-Enabled Ligand Screening for Cs/Sr Crystallizing Separation.

Area of Science:

• Nuclear Chemistry
• Materials Science
• Computational Chemistry

Background:

• Sustainable nuclear energy relies on efficient spent nuclear fuel reprocessing.
• Current separation technologies face challenges, necessitating advanced materials.
• Selective coordination of metal cations in liquid waste is crucial.

Purpose of the Study:

• To develop a high-throughput screening framework for identifying effective ligands for nuclear waste separation.
• To improve the efficiency of discovering next-generation separation materials.
• To address challenges in spent nuclear fuel reprocessing.

Main Methods:

• Utilized a computational framework incorporating aqueous solubility, pKa, and coordination bond length.
• Employed machine learning models with graph convolution and transfer learning for property prediction.
• Implemented a "computational funnel" to identify suitable ligands for Cesium/Strontium separation.

Main Results:

• Successfully identified potential ligands for Cesium/Strontium crystallizing separation.
• Machine learning models accurately predicted key chemical characteristics of ligands.
• Selected top-ranking ligands that are non-toxic and low-cost for experimental validation.

Conclusions:

• The proposed framework significantly enhances the efficiency of ligand discovery for nuclear waste reprocessing.
• Computational screening accelerates the identification of promising materials for critical separation processes.
• This approach supports the development of sustainable nuclear energy by improving fuel cycle technologies.