Influence of the computational methods in the conformational assignment of experimental infrared spectra

Recent advancements in experimental techniques have provided spectra for single conformers for relatively large flexible molecules, creating an ideal environment to benchmark computational methods on their ability to identify the conformers responsible for the experimentally measured spectra. Here, we performed benchmarks of different functionals and basis sets using three molecules: diphenylalanine, delta-9-tetrahydrocannabinol, and the 3-O-Acetyl-2,4,6-tri-O-methyl-gluco-D-pyranosyl cation. For this task, a new type of spectral similarity score was introduced, the Logarithmic Convoluted Cosine Similarity (LCCS), which is able to quantify spectral differences in terms of both frequency and intensity mismatches. Our results show that, as long as hybrid functionals are selected, the most crucial factor for the correct conformer assignment is the basis set. In particular, polarization functions were found to play a crucial role. We analyzed three additional aspects of the problem: the scan of the potential energy surface to find candidate conformers, pre-optimization of candidate conformers, and the selection from the pool of conformers based on the energy. Our data indicate that for scanning the potential energy surface, the DFTB3 semi-empirical method is a good compromise between accuracy and low computational cost. For the pre-optimization, we found that GGA functionals and a small basis set with a polarization functional already achieve sufficient accuracy. Instead, in the pre-selection based on energy, hybrid functional should be preferred, and in any case, an energy window of at least 15 kJ/mol should be employed for the conformer selection. • Computational methods have been benchmarked on their ability to identify the conformers responsible for the experimentally measured spectra. • As long as hybrid functionals are used, the basis set, particularly the inclusion of polarization functions, is the most critical factor for determining the quality of spectrum predictions. • A new type of spectral similarity score, able to quantify spectral differences in terms of both frequency and intensity mismatches, has been introduced. • Three additional aspects of the problem have been analyzed: the scan of the potential energy surface to find candidate conformers, pre-optimization of candidate conformers, and the selection from the pool of conformers based on the energy. • for scanning the potential energy surface, the DFTB3 semi-empirical method is a good compromise. For the pre-optimization, GGA functionals already achieve sufficient accuracy.