Replication Data for: HDNNP for Co3O4

The data files contain the raw input files for training the neural network, the required input files with the settings for the RuNNer code, and the resulting weights and scaling data used to perform the MD simulations reported in the paper.

More Details:
The dataset files are formatted for the RuNNer package (version 1.3) with details accessible in its documentation. The input.data file includes reference structures with atomic coordinates (in Bohr), forces (Hartree/Bohr), and energies (Hartree). Structures are defined between "begin" and "end" keywords, with flexible line ordering. The input.nn file serves as the main control file, allowing arbitrary keyword order and containing customizable settings for the RuNNer settings. The opt.weights.008.out and opt.weights.027.out files stored neural network weights for Oxygen and Cobalt, respectively. The scaling.data file, generated during fitting, records symmetry function statistics and is essential for energy predictions.