Machine Learning for Potential Energy Surface Construction: A Benchmark Set

Outline<br>
=======<br>
<br>
We assembled data from Potential Energy Surface (PES) construction invoking the n-mode representation. The database contains all single point calculations (SPCs) we carried out on the CCSD(F12*)(T)/cc-pVDZ-F12 level of theory. However, we also were interested in different delta-learning schemes:<br>
<br>
- DZERO Learning the CCSD(F12*)(T) energy<br>
- DSCF Learning the difference of CCSD(F12*)(T) and SCF (HF)<br>
- DMP2 Learning the difference of CCSD(F12*)(T) and RI-MP2<br>
- DMP2F12 Learning the difference of CCSD(F12*)(T) and RI-MP2-F12<br>
<br>
And we use the listed names in the files as identifier.<br>
<br>
Furthermore, the data is partitioned according to the mode combination level (MCL) used in the n-mode expansion<br>
<br>
- 1M: Only 1-mode couplings are considered<br>
- 2M: 1-mode and 2-mode couplings are considered<br>
- 3M: 1-mode, 2-mode and 3-mode couplings are considered<br>
<br>
<br>
Grids<br>
========<br>
<br>
We use iteratively refined grids using the adaptive density guided approach [1] (ADGA) and static grids. For the ADGA computations the thresholds were chosen such that ADGA converges when the relative change in the potential times density from one iteration to the other is less than 1E-2 and the absolute change less than 1E-6 a.u.<br>
<br>
For the static grids, the number of SPCs along the distortion of the normal modes will be indicated. For example g16x8x4 means that 16 SPCs are carried out for the 1-mode part along each distortion, 8 for the 2-mode couplings and 4 for the 3-mode couplings.<br>
<br>
Additionally, the maximum physical dimension of the grid in terms of the turning points for a harmonic oscillator of the indicated quantum level will be given in the folder name as n<number> e.g. g16x8x4n10. Note that this information is not given for the iteratively refined grids since the boundaries are determined based on the ADGA thresholds.<br>
<br>
<b>References</b><br>
[1] M. Sparta, D. Toffoli, and O. Christiansen, Theoretical Chemistry Accounts 123, 413 (2009).
<br>
Molecules<br>
=========<br>
<br>
You should find data for these molecules:<br>
<br>
- ammonia<br>
- cis-HNO2<br>
- Cl2O<br>
- ClNO<br>
- ethene<br>
- F2O<br>
- H2CO<br>
- H2O2<br>
- H2S<br>
- H2S2<br>
- nitro<br>
- NOF<br>
- SOCl2<br>
- transCHFCHF<br>
- Water<br>