Using electronic energy derivative information in automated potential energy surface construction for vibrational calculations

The availability of an accurate representation of the potential energy surface (PES) is an essential prerequisite in an anharmonic vibrational calculation. At the same time, the high dimensionality of the fully coupled PES and the adverse scaling properties with respect to the molecular size make the construction of an accurate PES a computationally demanding task. In the past few years, our group tested and developed a series of tools and techniques aimed at defining computationally efficient, black-box protocols for the construction of PESs for use in vibrational calculations. This includes the definition of an adaptive density-guided approach (ADGA) for the construction of PESs from an automatically generated set of evaluation points. Another separate aspect has been the exploration of the use of derivative information through modified Shepard (MS) interpolation/extrapolation procedures. With this article, we present an assembled machinery where these methods are embedded in an efficient way to provide both a general machinery as well as concrete computational protocols. In this framework we introduce and discuss the accuracy and computational efficiency of two methods, called ADGA[2gx3M] and ADGA[2hx3M], where the ADGA recipe is used (with MS interpolation) to automatically define modest sized grids for up to two-mode couplings, while MS extrapolation based on, respectively, gradients only and gradients and Hessians from the ADGA determined points provides access to sufficiently accurate three-mode couplings. The performance of the resulting potentials is investigated in vibrational coupled cluster (VCC) calculations. Three molecular systems serve as benchmarks: a trisubstituted methane (CHFClBr), methanimine (CH(2)NH), and oxazole (C(3)H(3)NO). Furthermore, methanimine and oxazole are addressed in accurate calculations aiming to reproduce experimental results.