Part 3 - Datasets of all active learning cycles of constructing a HDNNP for the first step of the Strecker synthesis

The construction of a high-dimensional neural network potential (HDNNP) able to determine the free energy profile of the first part of the Strecker synthesis was done in an active learning process. This required fitting HDNNPs, the propagation of molecular dynamics (MD) simulation employing umbrella sampling, identifying new structures and the computation of energies and forces with e.g. a DFT method.

Finally, a reliable HDNNP and a converged free energy profile is obtained. The four datasets of the dataverse contain separately 1) the input files for FHI-aims and MD simulations, 2) the obtained HDNNPs, 3) the obtained new configurations and 4) the final HDNNP and free energy profile. This dataset contains the sub datasets adding up to the full dataset constructed in the active learning process.