Accelerating the discovery of ionic liquids for zinc electrochemical applications

This paper discusses the combined use of deep neural networks (DNN), and variational autoencoders (VAE) for predicting key ionic liquids (IL) properties, and for generating new anion and cation pairs for potential electrolytes. Predictive models were trained using datasets of IL molecular fingerprints to forecast critical properties such as melting and decomposition temperatures. A VAE was effectively utilized to generate chemically plausible combinations of anions and cations, aiming to discover new IL electrolytes suitable for operation at room temperature. Transfer learning further refined these models, enhancing prediction accuracy specifically for zinc electrochemical applications. The resulting predictive models showed strong performance, achieving R² values exceeding 0.97 for both melting point and decomposition temperature predictions across multiple datasets. Several of the newly proposed ILs were validated through existing literature, underscoring their practical relevance. This combined approach efficiently bridges computational predictions with practical electrolyte development, enhancing the development of zinc battery technologies.