Michael Ashby1,Difan Zhang1,Roger Rousseau2,Vassiliki-Alexandra Glezakou2
Pacific Northwest National Laboratory1,Oak Ridge National Laboratory2
Michael Ashby1,Difan Zhang1,Roger Rousseau2,Vassiliki-Alexandra Glezakou2
Pacific Northwest National Laboratory1,Oak Ridge National Laboratory2
Ionic liquids have a wide range of applications in electrochemistry, engineering, and chemical processes as solvents, catalysts, or reagents. To understand the properties of ionic liquids, Nuclear Magnetic Resonance (NMR) is commonly used because it provides a unique set of chemical shifts that can uniquely characterize an ionic liquid compound. However, due to the versatile categories and complex environments of ionic liquids, it is time-consuming and costly to measure the experimental NMR data for ionic liquids. Although computational methods such as density functional theory calculations provide an alternative way to obtain NMR data, they are limited to small molecular models and are computationally expensive, especially for complex systems. Here, we propose a machine learning approach to evaluate the NMR chemical shifts of ionic liquids. A neural network is trained on an ionic liquid dataset established on experiments and ab initio calculations of ionic liquids. The network can help us to evaluate the NMR data of ionic liquids more quickly and efficiently, as well as predict the NMR data of unknown ionic liquids to accelerate the understanding of their structures and properties. We believe such an approach could be applied beyond the scope of ionic liquids as well as other energy materials,such as optimizing electrolyte solutions to extend the life of flow batteries at a cheaper cost.