Data supporting "Learning Pairwise Interaction for Extrapolative and Interpretable Machine Learning Interatomic Potentials with Physics-informed Neural Network"

<b>Data supporting "Learning Pairwise Interaction for Extrapolative and Interpretable Machine Learning Interatomic Potentials with Physics-informed Neural Network"</b><br>This archive contains training data used for models in the manuscript, which are created from the authors or modified/subsampled from public datasets. All molecular geometries are saved as extended xyz (extxyz) format.<b>Dataset file description</b><b>Ethanol</b>`ethanol/ethanol_dissociation.xyz`: The geometries for calculating PES of ethanol dissociation.`ethanol/ethanol_optimized.xyz`: The optimized geometry used for computing pairwise interaction energies and intrinsic bond strength index (IBSI).<b>Extrapolation data</b>The files listed in `extrapolation_data` are the dataset for <b>Figure 2</b> of the manuscript, which are splitted to train and test dataset as described in the manuscript.<b>S</b>_{<strong>N</strong>}<b>2 reaction</b><br>`sn2_reaction/sn2_reaction_I_Br.xyz`: The subset of publicly available S_N2 reaction of methyl halide[1], only containing I and Br as halogen species.`sn2_reaction/sn2_I_Br_subsampled.xyz`: Normal mode sampled (NMS) images &amp; AIMD trajectory of the reaction CH₃I + Br^- &lt;=&gt; CH₃Br + I^-.`sn2_reaction/sn2_I_Br_noTS.xyz`: NMS-only dataset for the same reaction.<b>Liquid water</b>Classical MD trajectory of liquid water. The original dataset is created from the authors of [2]. We subsampled the dataset with 500 geometries.<br>`water/water_0.5k_train.xyz`: 450 geometries for training the model`water/water_0.5k_val.xyz`: 50 geometries for validation steps.<b>References</b>[1] Unke, O. T.; Meuwly, M. PhysNet: A neural network for predict-ing energies, forces, dipole moments, and partial charges. Journal of chemical theory and computation 2019, 15 (6), 3678-3693[2] Fu, X.; Wu, Z.; Wang, W.; Xie, T.; Keten, S.; Gomez-Bombarelli, R.; Jaakkola, T. Forces are not Enough: Benchmark and Critical Evaluation for Machine Learning Force Fields with Molecular Simulations. 2023; https://arxiv.org/abs/2210.07237.<br>