Estimation of the Pitzer Parameters for 1-1, 2-1, 1-2, 3-1, and 1-3 Single Electrolytes at 25 °C

Abstract

Abstract The Pitzer model is widely used in process models for determining the thermodynamic properties of concentrated aqueous electrolytes. The main drawback is the need for experimentally determined parameters for each ion pair in the system. The estimation of unknown Pitzer parameters of electrolytes is thus an attractive prospect. This paper investigates machine learning-based models for the task of predicting the Pitzer parameters of binary electrolytes based on the properties of the constituent ions. A kernel ridge regression model was found to give the best results. Parameters important for the prediction task were found to be the ionic charges, radii, and approximate values for the chemical hardness. Pitzer parameters derived from the model generally gave similar or better predictions for mean activity coefficients than non-electrolyte specific terms or a previously published expression for the estimation of Pitzer parameters. However, performance was poor for ions showing strong association, for four-coordinated complex ions, and organic ions. Nevertheless, this paper shows that there is systematic behavior in the Pitzer parameters that makes it possible to estimate these parameters if enough training data is available.