Generative Artificial Intelligence Navigated Development of Solvents for Next Generation High‐Performance Magnesium Batteries

Le résumé fourni par la source

Abstract Traditional trial‐and‐error methods are inefficient and costly in discovering novel solvents for next‐generation magnesium (Mg) metal‐based batteries. Therefore, this work establishes a simple yet efficient screening criterion for solvents by integrating artificial intelligence techniques with a virtual molecular database, potentially revolutionizing the traditional solvent design pathway. A total of 823 solvents are generated using a self‐developed algorithm, and LUMO, ΔLUMO, ESP min , ESP max , and E b are identified to establish the screening criterion through the analysis with machine learning (ML) models. Eighteen candidate solvents are successfully identified, and two of which are subsequently selected and experimentally validated, i.e., C1COCOC1 and COCC(C)OC (abbreviated as “DOX” and “DMP”). Notably, neither of these solvents has been previously reported for use in Mg batteries. Experimental results indicate that the DOX solvent, when paired with the Mg boron‐based salt, i.e., Mg[B(hfip) 4 ] 2 , can significantly enhance the electrochemical performance. At a current density of 1.0 mAcm −2 , the average coulombic efficiency for Mg plating/stripping reaches 99.54 % after 5200 cycles. Furthermore, the Mg//Cu cell achieves a cumulative capacity exceeding 2000 mAhcm −2 , surpassing previously reported results. In summary, this work establishes a virtual molecular database and develops a streamlined screening methodology for Mg battery solvents based on their physicochemical properties, reducing the candidate pool from 823 to 18 and improving efficiency by nearly 50‐fold. This research paradigm is not limited to the development of Mg batteries and can be readily extended to the exploration of other battery systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.