Multivariate regression of several different quantum chemical descriptors was used to build a model for the reactivity of nucleophilic aromatic substitution reactions, i.e. S_NAr reactions, through predictingthe molecular reaction site’s Gibb’s free activation energy (ΔG^‡). The datasets used for training provided data of ΔG^‡ for several differing halide leaving groups including chloride, bromide, and fluoride. A set of descriptors were tested for the different leaving groups revealing that dissimilar leaving groups are more dependent on certain descriptors than others, meaning each model has to be tailored for the specific leaving groups.

Excellent correlations (R² = 0.93) were achieved between the predicted ΔG‡ and the experimental ΔG^‡.The ability of the model to predict regioselectivity in aromatic compounds with multiple leaving groups was tested and successfully predicted the correct regioselectivity through the calculation of ΔΔG^‡ in each case tested. However, the model’s validity outside of the training dataset was put into doubt through low R² values when the model was tested with several external datasets. An unknown factor arose which is speculated to be because of how differing nucleophiles and solvents affect the ΔG^‡. One of these tests yielded excellent correlations (R² = 0.9525) which could be because of similarities between solvents and nucleophiles between the training dataset but a similar factor between predicted ΔG^‡ and the experimental ΔG^‡ could still be observed.