Benchmarking of electronic structure methods for non-covalent interactions
Location: Bat. Bienvenue (former IFSTTAR building), U. Gustave Eiffel 14-20 Bd Newton, 77420 Champs-sur-Marne, France
The accurate description of molecular non-covalent interactions is a fundamental piece of the puzzle for the development of predictive computational models in chemistry. The in silico design of new catalysts, screening of drug molecules, materials simulation, all depend on the quality of the underlying model, the balance between strong and weak chemical forces. Even in an age of rapid progression in machine learning tools, our development in these areas will be ultimately limited by the quality of the data available. Furthermore, one needs to consider the level of resolution needed to achieve chemical predictions. A 1:9 product ratio in a kinetically controlled chemical reaction can be translated to little more than 1 kJ/mol difference in activation barriers. Conformer predictions can only be useful if the accuracy is below kT. This lays out a clear challenge for the future: the development of ever more accurate computational protocols for the description of individual molecular interactions.
In the pursuit of this objective, collaboration is key. The simulation of chemical processes at this resolution level is not possible by one group alone. It requires a wide network for exchange on the different accuracy bottlenecks simulation faces. It also calls for a dialogue between experiment and theory, as only the former can ultimately provide the necessary reference data. Albeit the insight provided by the comparison of different theoretical approaches in the calculation of electronic energies, real-world challenges go well beyond this. The impact of different approaches to the calculation of zero-point energy effects is still a matter of intense discussion. The case only worsens when extending to finite temperature, where entropy effects also need to be included.
In 2016, a long-term benchmarking challenge was set in motion, starting with the methanol solvation preference of 2,5-dimethylfuran, which was shown to be subtle, but experimentally tractable. Furan offers a planar scaffold with many potential modification options via alkyl and other substitution. Methanol can also be easily modified by alkylation. The methanol-furan contact pair is small enough to leave some hope for future anharmonic treatments of zero point energy, but large enough to render this a major challenge for relative energy predictions on the sub-kJ/mol scale. The results of the challenge provided helpful clues in the design of future test systems.
The systems considered in two later challenges were the dimers of furan/methanol, 2-methylfuran/methanol and 2,5-dimethylfuran/methanol. The challenge consisted of the (blind) theoretical calculation of the preference between the competing O-O or O-π docking conformations. The goal was to present the best possible predictions for the energy difference (at 0K) between these two modes of binding. This has evolved to a rather broad list of challenges for theory, encompassing energetic preferences, prediction of IR and microwave spectra, among other properties. The exchange from the first phase has also motivated the pursuit of further experiments which have ultimately led to benchmarking criteria under the kJ/mol range. Several theory groups have successfully participated, providing comparative data on a fair range of methodologies, including DFT, DFT-SAPT, coupled cluster theory, explicit correlation, harmonic and anharmonic vibrational calculations.
Majdi Hochlaf (Université Gustave Eiffel) - Organiser
Ricardo Mata (University of Göttingen) - Organiser
Daniel Obenchain (University of Göttingen) - Organiser