Open Access

Table 1

Composition of the data set used for training the machine-learned interatomic potentials.

Label Number of points Ensemble Temperatures (K) Propagation method
31 H2O + H 240 NVT 300/150/50 GFN2-xTB
31 H2O + P 240 NVT 300/150/50 GFN2-xTB
31 H2O + PH 240 NVT 300/150/50 GFN2-xTB
46 H2O + H 600 NVT 300/150/50 GFN2-xTB
46 H2O + P 600 NVT 300/150/50 GFN2-xTB
46 H2O + PH 600 NVT 300/150/50 GFN2-xTB
74 H2O 800 NVT 100/500(1) GFN-FF
12 H2O + H Collision 500 NVE 100 (EH=1.5 eV) GFN2-xTB
12 H2O + P Collision 500 NVE 100 (EP=0.025 eV) GFN2-xTB
Long-range P(2) 400 N/A N/A N/A
20 H2O + P + H Reaction(3) 6927 NVE 30 HF-3c

Total 11 647 (11 249)(4)

Notes. We note that the propagation method is only used for sampling geometries. For these structures, energies and forces are later calculated at the PBE-D3BJ/def2-TZVP level of theory. (1)A spherical wall potential is applied to ensure the structural integrity of the cluster. (2)Interaction of P with the cluster at the distance of the cut-off radius (5.5 Å, required for binding energies). (3)The reaction part comprises reactions in weak, medium, and high binding sites at different P-H internuclear distances (3.0–4.5 Å). (4)398 structures have been removed from the training set (some distances between neighbouring atoms are larger than the selected cut-off radius).

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.