Characterising a protic ionic liquid library with applied machine learning algorithms

Abstract

The ability to synthesise large libraries of ionic liquids (ILs) is one of their greatest selling points. This allows for systematic alteration of ion structure to tailor them for applications, and to study structure–property trends of IL physicochemical, thermal and solvation properties. The influence of Coulombic, Van der Waal’s and H-bond cohesive forces as well as Pauli repulsive forces on these properties can be understood via the development of large libraries of ILs. However, due to the complexity of ILs, the ion stoichiometry, water content, ion reactivity (pKa) and ion clustering have all been reported as contributing parameters. Here we report on the physicochemical, thermal and liquid nanostructure properties of 52 stoichiometric ion combinations from a series of 10 cations and 11 anions. Of these, 9 have not been reported before, 3 have been reported with no characterisation and 17 have undergone partial characterisation. Here we discuss the importance of reproducible synthesis methods and show structure–property relationships consistent across ion series, including changing alkyl chain length on the cation or anion, multiple chains on the cation and the presence of hydroxyl groups on the cation or anion. Relatively new data analysis methods applied in the IL field of machine learning algorithms, including multiple linear regression, random forest and k-nearest neighbour algorithms were applied to the data, providing a quantitative summation of individual structural moiety contributions. We highlight the importance of the full characterisation of ILs as well as detailed reporting of their preparation for experimentation including water content and pH at 10 wt%.