(611e) Bayesian Optimization of Molecular Structures: Data-Driven Sampling for Molecular Conformers

While most molecules with more than 4 atoms have multiple conformers as local minima, generating accurate ensembles requires of both efficient sampling and accurate potential energy surfaces. Despite many existing methods to find accurate molecular conformations, each year novel methods are published, indicating the level of difficulty in the problem. We will briefly review the history of the field, including previous data-driven approaches, as well as modern repositories, including both experimental and computational data. We will discuss recent efforts to benchmark force fields, semiempirical quantum chemical methods, density functional approximations, and higher-level quantum methods for conformer evaluation. Rankings and correlations between low-level methods are amazingly poor. Discussion will focus on new data-driven methods to efficiently sample the multi-dimensional problem and accurately provide energy evaluations.