PNNL

Abstract

CH-p and p-p interactions have been established as vital noncovalent interactions in biomolecules despite being relatively weak on their own. However, describing dispersion and higher order multipolar interactions typically requires a high-level (and high scaling) method such as CCSD(T) which scales as N7, rendering calculations of large molecules infeasible. As the simplest system exhibiting these noncovalent interactions, the benzene dimer is a prime candidate to benchmark theoretical methods to determine which lower scaling methods accurately describe these interactions. MP2 was used to explore the potential energy surface of the CH-p interactions of the benzene dimer, resulting in T-shaped and tilted T-shaped conformers with binding energies within 0.2 kcal/mol of each other. CCSD(T) binding energies at the complete basis set (CBS) have been calculated for the parallel displaced (PD; p-p interaction), T-shaped (T(C2v); CH-p interaction), and tilted T-shaped (TT(Cs); CH-p interaction) benzene dimers yielding values of -2.65±0.02, -2.74±0.03, and -2.83±0.01 kcal/mol, respectively. These benchmarks indicate that the tilted T-shaped dimer is the absolute minimum of the benzene dimer. Further, these benchmark values reveal the inaccuracies of the MP2 level of theory on the CH-p interactions, overestimating the differences between the T-shaped and tilted T-shaped conformers as well of incorrectly predicting the relative stabilities of the CH-p and p-p interactions. The CCSD(T)/CBS benchmark values were then used to assess the performance of spin biased MP2 methods (SCS-MP2, SOS-MP2, SCS-MI-MP2) and 11 DFT functionals from different rungs of Jacob’s ladder. While the spin biased MP2 methods improved the quantitative description of these interactions (relative to MP2), they incorrectly estimated the relative stabilities of the conformers. However, the spin-biased methods (SCS-MP2 and SCS-MI-MP2, in particular) produced geometries with RCOM values within 0.02 Å of the CCSD(T) geometries, suggesting that these methods be used as a cost-effective alternative to produce high-level structures. Many DFT functionals including TPSS-D3, PBE-D3, B3LYP-D3, B97-D, and PBE0-D3 provided a good description of the CH-p and p-p binding energies yielding values within 5% of the CCSD(T)/CBS limits. In addition, these functionals (save PBE-D3) correctly predict the relative stability of the PD and T(C2v) dimers. Despite this impressive performance, a challenge that remains is ensuring the transferability of these functionals to larger biomolecules (thus requiring high-level benchmarks of these systems).