Quantitative structure-property relationship modelling of thermal decomposition temperatures of ionic liquids

• Published in: Journal of molecular liquids 2016-11, Vol.223, p.60-67
• Main Authors: Venkatraman, Vishwesh, Alsberg, Bjørn Kåre
• Format: Article
• Language: English
• Subjects: Ionic liquids; Machine learning; QSPR; Thermal decomposition
• ISSN: 0167-7322; 1873-3166
• DOI: 10.1016/j.molliq.2016.08.023

The thermal decomposition point for ionic liquids (ILs) is an essential property that imposes an upper operating limit for many applications. Since the decomposition of ILs can lead to unwanted byproducts, it is desirable to improve their thermophysical properties and create more application specific compounds. With a view to rapidly estimate these properties of interest, approaches based on quantitative structure-property relationship (QSPR) models have been relatively successful but somewhat restricted to small datasets with limited diversity. Here, we investigate the effectiveness of a wide range of electronic, thermodynamic and geometrical descriptors derived from semi-empirical PM6 calculations to estimate the thermal decomposition temperatures of 995 diverse ILs comprising 461 cations and 119 anions. Of the two regression schemes used: partial least squares and random forests, the latter yielded slightly improved performances (Rcv2=0.81,Rtest2=0.77). Analysis based on variable importance indicates that the anion-specific descriptors such as nucleophilicity and size significantly influence the thermal stabilities which are in agreement with experimental observations. •Machine learning predicts the thermal decomposition temperature of ionic liquids.•Data set was based on very diverse 995 ionic liquids (461 cations and 119 anions).•Semi-empirical (PM6) electronic, thermodynamic and geometrical descriptors used.•Anion-specific descriptors (e.g. nucleophilicity and size) are important.