Machine Learning-based approach for Tailor-Made design of ionic Liquids: Application to CO2 capture

In this article, we present a machine learning-based approach for the tailor-made design of ionic liquids (ILs) promising toward the desired target applications. Our computational framework combines multi-player Monte Carlo tree search and recurrent neural network, within a parallel scheme of generating and testing multiple ILs simultaneously, to improve the efficiency of searching optimal structures. For two cases of CO2 capture from 1) flue gas (CO2/N2) and 2) from syngas (CO2/H2), target-specific ILs were generated in our computational platform according to objective function values that combine three requirements of high CO2 solubility, absorption selectivity of IL for CO2, and easiness of subsequent desorption. Our results showed that high-performance ILs can be designed with great efficiency using our algorithm. Furthermore, topological data analysis on newly designed ILs demonstrated that our algorithm allows us to explore materials space widely to find highperforming ILs with good diversity.

Automated summary

This study introduces a machine learning-based approach for designing tailored ionic liquids (ILs) for specific target applications, using a combination of multi-player Monte Carlo tree search and recurrent neural network. The results show that high-performance ILs can be efficiently designed using this algorithm. Topological data analysis also demonstrates that the algorithm allows for wide exploration of materials space to find high-performing ILs with good diversity.