Development of Scalable and Generalizable Machine Learned Force Field for Polymers

This is a dataset of 344,654 clusters of ethylene glycol EG, diethylene glycol (EG)2, and triethylene glycol EG(3), along with DFT labels at the wB97X-D3BJ/def2-TZVPD level of theory. <br> The primary key for each datapoint is a string representing an index from 0 to 344,653 inclusive. Each datapoint is a dictionary containing the following keys: <br> * atomicNumbers: atomic number of each atom * charge: net charge for each system, in elementary charges (always 0 for this dataset) * datasetTitle: indicates the sampling step which produced this data, out of the following six options: * EG_run: initial cluster sampling of EG, (EG)2, and (EG)3 * decomp_EG_monomer: decomposition sampling for EG * decomp_EG_dimer: decomposition sampling for (EG)2 * decomp_EG_trimer: decomposition sampling for (EG)3 * active_one: first active learning round * active_two: second active learning round * elements: element for each atom (same info as atomicNumbers) * labels: for this dataset, always 'wB97X-D3BJ__def2-TZVPD', containing within the following keys: * atomizationEnergy: total energy with fitted per-atom atomic energies removed, in Hartree (H: -31.91939585331449, C: -6.840292406034887, O: -12.5395523076437) * dipoleMoment: DFT system dipole moment, in e-Angstroms * gradient: energy gradient dE/dx (negative of atomic forces), in Hartree/Angstrom * totalEnergy: total energy from wB97X-D3BJ/def2-TZVPD, in Hartree * xtbCharges: xtb partial charges, in elementary charge units * multiplicity: spin multiplicity for each system (always 1 for this dataset) * positions: xyz coordinates of atomic nuclei, in Angstroms <br> Authors: Shaswat Mohanty, James Stevenson, Andrea Browning, Leif Jacobson, Karl Leswing, Mathew D. Halls, Mohammad Atif Faiz Afzal