H2CO Dataset

The deposited data sets were used to compare three state-of-the art machine<br> learning (ML) approaches to obtain representations of potential energy<br> surfaces (PESs). The comparison is meant to be representative as it examines<br> a purely kernel-based approach (reproducing kernel Hilbert space plus<br> forces (RKHS+F))[1], a purely neural network based approach (PhysNet)[2]<br> and includes the FCHL representation [3] within kernel ridge regression.<br> Formaldehyde, H2CO, is used as a benchmark system. H2CO is a small molecule for which PESs can be calculated<br> at different levels of theory and, thus, suitable for an in-depth theoretical<br> study. Also, very high-level calculations have already been presented (see e.g.<br> Ref. [4]) and experimental reference data is available to compare with [5]. Using reference data calculated at three different levels of quantum chemical<br> theory (B3LYP/cc-pVDZ, MP2/aug-cc-pVTZ and CCSD(T)-F12/aug-cc-pVTZ-F12) ML<br> models are trained using the different ML methods. The performance of the<br> models is then examined by considering energy and force learning curves,<br> harmonic frequencies and IR spectra from finite-Temperature molecular dynamics<br> (MD) simulations. The data sets contain different geometries for the H2CO molecule generated using<br> the normal mode sampling approach [6] performed at different temperatures. Four<br> data sets are deposited: <br> i) "h2co_B3LYP_cc-pVDZ_4001.npz": 4001 geometries of H2CO generated using normal mode<br> sampling and calculated using ORCA [7] (B3LYP/cc-pVDZ).<br> ii) "h2co_mp2_avtz_4001.npz": 4001 geometries of H2CO generated using normal mode<br> sampling and calculated using MOLPRO 2019 [8] (MP2/aug-cc-pVTZ).<br> iii) "h2co_ccsdt_avtz_4001.npz": 4001 geometries of H2CO generated using normal mode<br> sampling and calculated using MOLPRO 2019 [8] (CCSD(T)-F12/aug-cc-pVTZ-F12).<br> iv) "h2co_ccsdt_avtz_2500_extrapol.npz": 2500 geometries of H2CO generated using normal mode<br> sampling and calculated using MOLPRO 2019 [8] (CCSD(T)-F12/aug-cc-pVTZ-F12). This sampling<br> was carried out at higher temperature (5000 K compared to 2000K) to test the extrapolation<br> ability of the ML methods. For more details, see http://arxiv.org/abs/2006.16752 ---------------------------------------------------------------------------------------<br> HOW TO CITE: When using this dataset, please cite the following paper:<br> Käser, S. and Koner, D. and Christensen, A. S. and von Lilienfeld, O. A. and Meuwly, M.<br> "ML Models of Vibrating H2CO: Comparing Reproducing Kernels, FCHL and PhysNet"<br> arXiv:2006.16752 and the digital object identifier (DOI):<br> Käser, S. and Koner, D. and Christensen, A. S. and von Lilienfeld, O. A. and Meuwly, M. (2020).<br> H2CO Dataset. Zenodo. http://doi.org/10.5281/zenodo.3923823 --------------------------------------------------------------------------------------- [1] Koner, D.; Meuwly, M. arXiv e-prints 2020, arXiv:2005.04667<br> [2] Unke, O. T.; Meuwly, M. J. Chem. Theory Comput. 2019, 15, 3678–3693<br> [3] Faber, F. A.; Christensen, A. S.; Huang, B.; von Lilienfeld, O. A. J. Chem. Phys. 2018, 148, 241717<br> [4] Zhang, X.; Zou, S.; Harding, L. B.; Bowman, J. M. J. Phys. Chem. A 2004, 108, 8980–8986<br> [5] Herndon, S. C.; Nelson Jr, D. D.; Li, Y.; Zahniser, M. S. J. Quant. Spectrosc. Radiat. Transf. 2005, 90, 207–216<br> [6] Smith, J. S.; Isayev, O.; Roitberg, A. E. Sci. Data 2017, 4, 170193<br> [7] Neese, F. The ORCA program system. WIREs Comput. Mol. Sci. 2012, 2, 73–78<br> [8] Werner, H.-J.; Knowles, P. J.; Knizia, G.; Manby, F. R.; Schütz, M.; et al. https://www.molpro.net