Revisiting Boron Cluster Dihydrogen Bonding: Improved Reference Energies and Benchmark of Localized Coupled Cluster and <scp>DFT</scp> Methods

ABSTRACT Dihydrogen bonds involving boron clusters have gained significant attention in chemistry, biology, and medicine. The reference interaction energies of diH‐norm and diH‐linear datasets (Fanfrlík et al., ChemPhysChem 2020 , 21 , 2599), which include such bonds both in bent and linear configurations, have been re‐evaluated at the gold standard. This re‐evaluation utilizes explicitly correlated MP2‐F12 near the complete basis set limit, CCSD(F12*)/VQZ‐F12, and a (T) correction from conventional CCSD(T)/haV{T,Q}Z calculations. Relative to the updated reference data, Fanfrlík's original reference shows very small root mean square deviations, resulting from significant error compensation between the [CCSD‐MP2] and (T) high‐level corrections. The new reference data were subsequently used to assess the performance of various localized‐orbital coupled‐cluster methods, such as PNO‐LCCSD(T), DLPNO‐CCSD(T 1 ), and LNO‐CCSD(T), at their respective “Normal”, “Tight”, “vTight”, and “vvTight” accuracy settings. For these methods, gradually increasing the basis set size and tightening the accuracy cutoffs consistently improves their performance. For the dihydrogen bonds in the diH‐norm set, two cost‐effective and accurate methods—PNO‐LCCSD(T)/Tight/haV{T,Q}Z and DLPNO‐CCSD(T 1 )/TightPNO/CPS(6,7)/haV{T,Q}Z—yields RMS errors of just 0.010 and 0.027 kcal·mol −1 , respectively. For the diH‐linear set, the best‐performing low‐cost options are LNO‐CCSD(T)/vTight/haV{T,Q}Z and PNO‐LCCSD(T)/Tight/haV{T,Q}Z, with RMS errors of 0.013 and 0.016 kcal·mol −1 , respectively. None of the explicitly correlated localized coupled‐cluster methods tested surpass the accuracy of these efficient low‐cost methods. Among the more affordable density functional methods, r 2 SCAN‐3c stands out for delivering excellent accuracy at minimal cost. For hybrid and double‐hybrid functionals, the incorporation of range separation significantly enhances performance.