Parallel variable selection of molecular dynamics clusters as a tool for calculation of spectroscopic properties

Clusters of a solute and a few solvent molecules obtained from molecular dynamics (MD) are a powerful tool to study solvation effects by advanced quantum chemical (QC) methods. For spectroscopic properties strongly dependent on the solvation, however, a large number of clusters are needed for a good convergence. In this work, a parallel variable selection (PVS) method is proposed that in some cases efficiently reduces the number of clusters needed for the averaging. The mass, charge, or atomic density MD distributions are used as a secondary variable to preselect the most probable cluster geometries used for averaging of solute spectral properties. When applied to nuclear magnetic resonance chemical shift of a model alcohol, the method allowed one to significantly reduce the total computational time, by a factor of 10. Even larger savings were achieved for the modeling of Raman and Raman optical activity spectra of (S)-lactamide molecule dissolved in water. The results thus suggest that the PVS method can be generally used for simulations of spectroscopic properties of solvated molecules and makes multiscale MD/QC computations more affordable.