Ab Initio Group Additive Values for Thermodynamic Carbenium Ion Property Prediction

Abstract

Carbenium ions are important intermediates in both zeolite and plasma chemistry. The construction of kinetic models for zeolite and plasma chemistry requires the incorporation of thermodynamic properties of these carbenium ions. In this way, thermodynamic equilibrium is incorporated resulting in more accurate and general kinetic models, which facilitate rational zeolite design and plasma process development. In this work, a consistent set of 46 group additive values (GAVs) and non-nearest neighbor interactions (NNIs) is determined for the standard enthalpy of formation, standard molar entropy, and heat capacity. The GAVs are regressed based on 165 quantum chemistry calculations with the CBS-QB3 composite method for carbenium ions as CBS-QB3 was proven a valid method for carbenium property prediction. The presented group additive approach gives the first GAVs and NNIs for entropy and heat capacity and 33 novel values for enthalpy prediction. The determined non-nearest neighbor interactions account for the effect of longer range hyperconjugation, resonance stabilization, and inductive stabilization, which are shown to be crucial for the carbenium stability. The presented group additivity scheme ensures an accurate prediction of all properties with a mean absolute error of 4.82 kJ/mol, 4.92 J/(mol·K), and 1.95 J/(mol·K) for the standard enthalpy of formation, standard molar entropy, and heat capacity, respectively. The reported GAVs and NNIs ensure the thermodynamic property prediction of paraffinic, olefinic, alkynic, aromatic, and five- and six-ring naphthenic cations.

Beyond group additivity: Transfer learning for molecular thermochemistry prediction

“…The quantum chemical data consists of thermochemical information of 330 acyclic uncharged [50]. The corrections for the number of optical isomers, and internal and external rotational symmetry were applied to obtain the total entropy.…”

“…As the inductive effect is group-dependent no accurate empirical relation can be proposed to account for this effect. Overall, the stabilization of alkyl chains neighboring the positively charged carbon atom is significantly underestimated as this value for the GAVs is taken from regular hydrocarbons [50] and considers no inductive stabilization. Hence, the stability of the C + -(C)2(H) is overestimated to compensate the lack to incorporate the inductive stabilization effect.…”

Self-Consistent Implementation of a Solvation Free Energy Framework to Predict the Salt Solubilities of Six Alkali Halides

“…These Gibbs free energy changes for six neutral substances in the gas phase can also be simulated using Gaussian16 software with the G4 , and CBS-QB3 , thermodynamic methods, which are compared with the experimental values. Both of these methods are quantum mechanical methods and have been demonstrated to yield satisfactory accuracy in predicting the thermodynamic properties of ionic systems. − The G4 method is an extension of the B3LYP method, whose values have an absolute deviation less than 10 kJ mol –1 comparing with the experimental values. The CBS-QB3 method is based on a series of approximations that include a combination of Hartree-Fock and post-Hartree-Fock methods, while it results in a larger absolute deviation than the G4 method.…”

A Subgraph Isomorphic Decision Tree to Predict Radical Thermochemistry with Bounded Uncertainty Estimation

“…Deep Neural Network (DNN)-based estimators for thermochemistry have gained much interest in recent years. [13][14][15] While transfer learning has shown promise as a mitigation strategy, 16 obtaining a reliable DNN-based estimator for radical thermochemistry can still be challenging due to the scarcity of high-quality thermochemical data for diverse sets of radical species. Moreover, DNN-based estimators often need extra efforts [17][18][19] to improve the model's interpretability.…”