A systematic approach for modelling the affinity coefficient in the Dubinin-Radushkevich equation

An unbiased evaluation of predictive models for the affinity coefficient of the Dubinin-Radushkevich equation was performed. The first step involved the selection of a minimum number of representative and chemically diverse organic compounds, the training set. This set, isopropylamine, heptane, dichloromethane, 2-chloro-2-methylpropane, 2-butanone, 1-chloropentane, acetonitrile, and benzene, covering five compounds classes, was selected with the help of PCA and statistical design. Secondly, experimental affinity coefficients of the training set compounds were determined from adsorption isotherms on Norit RI activated carbon. In a third step, 45 physico-chemical properties were assembled for the training set compounds. A model was developed, based on PLS analysis, which correlates the measured affinity coefficients and the physico-chemical properties. Finally the model was validated by comparing model predictions of the affinity coefficients with literature data for an external validation set of 40 compounds. It was found that the predictive power of this model (RMS error=0.090) is better than using traditional methods based on parachor, molar polarizability or molar volume. The proposed new model for the affinity coefficient is based on three parameters only, the molecular weight and VdW volume of the compound and the calculated energy of interaction between the compound and a graphite model surface. (C) 2002 Elsevier Science Ltd. All rights reserved.