Machine Learning to Predict Ionic Liquid Propellant Properties

The efficiency of ion emission from electrospray ionization sources is fundamentally constrained by the binding energy between the ionic liquid (IL) cation and anion. High binding energies favor the ejection of clusters and droplets, reducing the achievable specific impulse, while low binding energies promote the formation of a pure-ion beam. Density functional theory (DFT) provides accurate ion-pair energetics but is too computationally expensive for large-scale screening of IL propellants. In this work, we develop a machine learning (ML) framework capable of predicting DFT-calculated ion-pair binding energies directly from the SMILES representations of the constituent ions. Dimensionality-reduced features obtained via principal component analysis (PCA) were used to train two regressors: Ridge regression and kernel Ridge regression. Kernel Ridge regression achieved the better performance, with a coefficient of determination (R²) of 0.4, and a Spearman’s rank correlation coefficient of 0.9 with ten principal components. An R² value of 0.4 is still considered weak, which can be attributed to the size of the training set (only 20 data points). While there is room for improvement in the model’s predictive power, these results demonstrate that descriptor-based ML models can reliably rank IL binding energies and serve as efficient surrogates for first-principles screening of potential ionic liquid propellants for electrospray applications.