σ-hole interactions are a class of noncovalent interactions involving p-block elements. Comparable in strength and directionality to hydrogen bonds, σ-hole interactions present a useful tool for a wide range of applications.
The σ-hole arises when a polarisable p-block element is covalently bonded to an electron withdrawing group (EWG). The σ-bond to the EWG creates a charge anisotropy, resulting in a depletion of electron density along the extension of the σ-bond, the electropositive σ-hole.
Detailed searches of the CSD were constructed to investigate the geometric characteristics of σ-hole interactions to various nitrogen and oxygen lone-pair-possessing Lewis bases. Trends in the identity of the σ-hole donor atom and EWG were explored. The σ-hole acceptors also play a key role in the interactions and so were investigated on a functional group basis. Nitrogen Lewis bases are generally better σ-hole acceptors than oxygen Lewis bases, but oxygen Lewis bases show greater variety.
To predict the interaction energies for $\sigma$-hole interactions, the σ-Hole Approximation Tool (σHAT) was constructed. Based on a XGBoost regression machine learning model, and trained on geometric measurements and adapted molecular fingerprints of σ-hole complexes from a data set of high accuracy calculated σ-hole interactions. This model performs comparably to density functional theory (DFT) on the test set, with an RMSD of 1.03 kcal/mol, which is very close to the 1 kcal/mol "chemical accuracy" threshold.
This model was validated using cutouts derived from crystal structures in the CSD, which possess halogen bonds (a subset of σ-hole interactions). DFT calculated interaction energies for these cutouts were used to compare against the σHAT predicted interaction energies. The performance on this validation set showed an RMSD of 2.02 kcal/mol, but differed depending on the halogen bond donor atom, suggesting different character of the interactions.
