VIB5 database with accurate ab initio quantum chemical molecular potential energy surfaces
We report global potential energy surfaces (PESs) database VIB5 of 5 molecules of astrophysical interest which was used to produce rovibrational spectra approaching spectroscopic accuracy and contains state-of-the-art, high-level energies and energy gradients. This database can be used to develop and test various ML models.
Accurate, global ab initio quantum chemical (QC) molecular potential energy surfaces (PESs) are indispensable in the simulations of rovibrational spectra which is of great significance in astrophysical studies. Such PESs were generated over the years for many molecules of astrophysical interest, including 5 small molecules, CH3Cl, CH4, SiH4, CH3F, and NaOH. Unfortunately, the raw data containing geometries, theoretical best estimates of energies (TBEs), which were used to generate accurate rovibrational spectra of these 5 molecules, and constituent terms of TBEs, are either not reported or scattered across the literature. Thus, here we report the collection of these data for altogether more than 300 thousand geometries.
Since now the attention of many researchers is turned to overcome the high computational cost of constructing PESs with traditional quantum chemical methods by machine learning (ML) methods, this database is of potential interest for the ML community, especially, because our database contains a higher energy range than many popular PES databases with data from low-temperature molecular dynamics. The PES of one of the molecules, CH3Cl, was already used in ML studies: in the development of self-correcting ML, structure-based sampling, and hierarchical ML. The latter approach especially benefits from the availability of the constituent terms of final energies. In addition, for testing ML approaches, it is beneficial to provide energy gradients. To this end, we have additionally calculated energies and energy gradients at MP2/cc-pVTZ and the CCSD(T)/cc-pVQZ levels of theory, and provide the Hartree-Fock energies calculated with corresponding basis sets.
Finally, our database comes with a convenient data-extraction script that can be used to pull the required information in a suitable format.
- Lina Zhang, Shuang Zhang, Alec Owens*, Sergei N. Yurchenko, Pavlo O. Dral*, VIB5 database with accurate ab initio quantum chemical molecular potential energy surfaces, Sci. Data, 2022, 9, 84. DOI: 10.1038/s41597-022-01185-w.