Zr–O Ab Initio Training Data Created by Molecular Dynamics, Contour Exploration, and Dimer Searches

    These density functional theory calculations span a diverse set of structures in the Zr–O system which was used as machine-learned interatomic potential (MLIP) training data. This data set was used to benchmark different structural evolution methods (molecular dynamics, contour exploration, and dimer searches) for the quality and accuracy of MLIPs trained on them. The data is provided in the .traj format from ASE. Along with data set used in our publication, we provide a large set of extra unused data and a small Python script example for parsing the data set. The set contains 120,068 structures which contain a total of 3,154,158 atoms.

For more details, please see our paper:
Michael J Waters and James M Rondinelli,  J. Phys.: Condens. Matter 34 385901(2022) (https://dx.doi.org/10.1088/1361-648X/ac7f73)