(569bh) Prediction of Relative SN2 Rate Coefficients in Different Solvents

Ionic reactions are common in pharmaceutical development, biochemistry, chemical manufacturing, environmental science, and many other applications of chemistry. The ability to predict rate coefficients in different solvents is important for these applications, as it would help enable solvent screening for optimizing reaction rates, minimizing the need for costly laboratory experiments.

Relative rate coefficients (i.e. the ratio of rate coefficients for a given reaction between two solvents) have recently been calculated using the solvation model COSMO-RS with mean absolute errors (MAEs) within 1 order of magnitude. However, the number of studied reactions in recent years is very small, spanning just 15 non-ionic reactions and three S_N2 reactions.

In this work, we introduce a digitized compilation of S_N2 rate data in solvents, spanning 55 ionic reactions involving 16 electrophiles and 16 nucleophiles across 15 solvents, including four relative rates of binary solvent mixtures. We use this data to benchmark the COSMO-RS method for predicting relative rate coefficients of SN2 reactions. We assess four different quantum-chemical levels used for geometry optimization: semi-empirical GFN2-xTB method in the gas phase; DFT in the gas phase; GFN2-xTB using the analytical linearized Poisson–Boltzmann (ALPB) implicit solvation model, and DFT using the CPCM implicit solvation model.

Our results indicate that the COSMO-RS computed solvation energies are not sensitive to the geometries, and the MAE of the computed relative rates fall between 0.9 to 1.0 log10 units. We also observe systematic error, with overprediction and underprediction depending on the type of nucleophile. We discuss potential applications for using this method to generate synthetic data that can be used for machine learning, as well as the limitations of the quantum-chemical method.