Fast exploration of potential energy surfaces with a joint venture of quantum chemistry, evolutionary algorithms and unsupervised learning

Contemporary molecular spectroscopy allows the study of flexible molecules, whose conformational behavior is ruled by flat potential energy surfaces (PESs) involving a large number of energy minima with comparable stability. Under such circumstances assignment and interpretation of the spectral signatures can strongly benefit from quantum chemical computations, which face, however, several difficulties. In particular, the mandatory characterization of all the relevant energy minima leads to a huge increase in the number of accurate quantum chemical computations (which may even hamper the feasibility of a study) and the intricate couplings among several soft degrees of freedom can defy simple heuristic approaches and chemical intuition. From this point of view, the exploration of flat PESs is akin to other optimization problems and can be tackled with suitable metaheuristics, which can drive QC computations by reducing the number of necessary calculations and providing effective routes to sample the most relevant regions of the PES. Unfortunately, in spite of the significant reduction of the number of QC calculations, a brute-force approach based on state-of-the-art methods remains infeasible. This problem can be solved effectively by multi-level strategies combining methods of different accuracy in the first PES exploration, refinement of the structures of the most important stationary points and computation of spectroscopic parameters. Building on previous experience, in this contribution we introduce new improvements in an evolutionary algorithm based method using curvilinear coordinates for both intra- and inter-molecular interactions. Two test cases will be analyzed in detail, namely aspartic acid in the gas-phase and the silver cation in aqueous solution. Comparison between fully a priori computed spectroscopic parameters and the experimental counterparts will provide an unbiased validation of the proposed strategy.