The successive reduction of data subsets: A new method for estimating parameters of thermodynamic models

The successive reduction method allows the fitting of models to large sets of data using microcomputers of moderate memory size. It relies on the assumption that the information brought by a set of data is summarized by the maximum likelihood parameters and their inverse variance—covariance matrix. The parameters are estimated from the last data subset, taking account of the information brought by the former ones. The successive reduction method is compared with the classical 'global' method. They give exactly the same results for linear models. With a non-linear model, equivalent results are obtained when this model is consistent with the data. Applications are presented, relating to parameter estimation for generalized equations of state, to vapour—liquid equilibrium and excess enthalpy data reduction using the NRTL and UNIQUAC models.