A method is developed for predicting the sorption parameters of polymers (the infinite dilution solubility coefficient of gases S and the Henry law equilibrium constant k D of double sorption models) of most light gases in glassy polymers on the basis of a computer model of a polymer chain segment. From the molecular dynamic model of conformations of a short (about 200–600 atoms) segment of the polymer chain, the dependences of a number of geometric indices, such as the accessible surface area and the accessible surface area with a positive partial charge, on the probe radius (in fact, on the effective radius of penetrant or sorbate) are calculated. The linear approximation coefficients of these dependences are used as explanatory variables in multiple linear regression. Significant variables and their weights in regression are determined on the basis of experimental measurements from the Database of Physicochemical Properties of Polymers of the Laboratory of Membrane Gas Separation at the Institute of Petrochemical Synthesis, Russian Academy of Sciences. By using a more accurate algorithm for calculating the areas of macromolecules, the shortcomings of the first version of this method are eliminated. For S, the coefficient of regression determination (adjusted R 2 ) on the test sample is equal to 0.8 (this corresponds to an average relative error of 77% in predicting S, which is considered good accuracy), and for k D , the coefficient of determination is 0.81 and the average relative error is 47%, which is also a good result considering large (up to decimal order) errors of the experimental measurements of S and k D .
