A study of the relationship between the electronic structure and the thermodynamic properties of individual compounds

Abstract

464 The study of reaction mechanisms, development of chemical technology, search for biologically active substances, and other topical physicochemical investi� gations require knowledge of the quantitative charac� teristics of compounds. These include key thermody� namic properties, such as the entropy, enthalpy, free energy, and heat capacity. Calculation of the thermo� dynamic properties of substances is associated with certain theoretical difficulties, is extremely resource� intensive, and usually does not provide the desired accuracy for the required quantity. This is especially the case for intermediates. Therefore, this field is dominated by phenomenological models, the most popular of which are additive models (additivegroup approach and macroincrementation). These models require parameterization and are not free from sys� tematic errors, a feature that requires the introduction of corrective procedures, for example, corrections for nonvalence interactions. Construction of models for calculating the thermo� dynamic properties with acceptable error for sub� stances consisting of large polyatomic molecules is a challenging task. Most of them are based on the assumption that the structure of a fragment is related to the contribution from this fragment to the given property of the molecule (quantitative structure- property relationship, QSPR). This assumption is adopted as a postulate, since it is impossible to theo� retically predict the contribution from a fragment based on its type or the relative arrangement of its atoms ("structure" descriptor). In such models, the introduction of descriptors involves certain arbitrari� ness, a circumstance that prevents ensuring the stated accuracy in predicting the properties. Development of a phenomenological model (1, 2) begins with the selection of the experimentally mea� sured properties of reference substances, followed by the partition of the samples into two sets, training and testing. The model is parameterized using the training set, whereas the testing set is employed to estimate the errors of the respective quantitative correlations, and, if required, corrective procedures are introduced into the model. The emergence of significant errors and failures outside the training set requires a changeover from searching for possible empirical relationships in the QSPR approach to constructing phenomenologi� cal models based on rigorous theorems and postulates of quantum mechanics. The use of such methods enables, on the one hand, to specify the confidence interval for the values obtained (based on the approxi� mations made in the derivation of the relationships) and, on the other, to determine the limits of applica� bility of the model.