Leveraging machine learning and experimental screening of sulfur sources to mediate interlayer engineering of...

Lithium ion sieves are promising materials for lithium extraction from liquids, but achieving high adsorption remains challenging. In this study, machine learning (ML) was employed to predict non-metal element-doped lithium ion sieves. A comparative evaluation of six ML models indicated that XGBoost performed the best, and the predictions indicated that lithium-ion sieve exhibits optimal performance after S doping. Based on model predictions, various sulfur-containing sources were screened, and experimentally anhydrous Li₂S was ultimately identified as the optimal sulfur source to mediate the interlayer spacing of the (111) crystal plane of the cubic Li₄Ti₅O₁₂ (denoted as LTOS). The adsorbent showed a Li⁺ adsorption capacity of 28.08 mg g^-1, exhibiting outstanding selectivity toward heteroatoms in artificial brine. The structural stability of S doping H₄Ti₅O₁₂ (denoted as HTOS) and its adsorption energy for Li⁺ were evaluated by density functional theory (DFT), confirming the superior properties of sulfur doping. Furthermore, the computational results were experimentally validated, and the experimental data showed excellent agreement with the predictions (R² = 0.851). The results indicate that the combination of ML with targeted experimental screening can efficiently guide the development of advanced lithium-ion sieve materials.