Datasets and scripts for the publication "Insights into Defect Cluster Formation in Non-Stoichiometric Wustite (Fe1-xO) at Elevated Temperatures: Accurate force field from Deep Learning"

All the datasets and scripts for the publication"Insights into Defect Cluster Formation in Non-Stoichiometric Wustite (Fe1-xO) at Elevated Temperatures: Accurate force field from Deep Learning".

This database contains high-fidelity datasets for non-stoichiometric wüstite (Fe₁₋ₓO), including atomic coordinates, energies, and forces generated through ab initio molecular dynamics (AIMD) and refined using Deep Potential (DP) training. The dataset encompasses bulk phases, vacancy structures, and surface orientations, enabling accurate modeling of defect clusters and thermodynamic properties. It supports machine-learning force field development, offering insights into defect formation and large-scale simulations of Fe₁₋ₓO systems at elevated temperatures.

Description of the File Structure of Fe1-xO_DeepMD_Code_Datasets_Analysis.zip:

1. `init` Folder  This folder contains the foundational datasets and inputs used for training and developing the machine-learning force field for Fe₁₋ₓO.  

1.1 `01.train_data` Subfolder This folder organizes data related to the initial training of the Deep Potential (DP) model.  - `dpmd_dataset`: Processed dataset ready for DeepMD training, containing atomic configurations, forces, and energies.- `dpmd_rawfiles`: Raw files from ab initio molecular dynamics (AIMD) simulations, serving as the source for generating training datasets.

1.2 `02.develop_data` SubfolderContains `.vasp` files representing structural data used to develop and refine the force field. The structures include bulk, vacancy, and surface configurations of Fe₁₋ₓO.  - Files labeled `bulk` represent bulk Fe₁₋ₓO systems with varying lattice constants.  - Files labeled `defect` represent Fe and O vacancy structures (single and double vacancies).  - Files labeled `surface` represent Fe₁₋ₓO surface structures in various crystallographic orientations.  

2. `run` FolderThis folder contains files and logs generated during iterative training and testing of the DP force field, as well as subfolders for each iteration of the training process.  

2.1 Iteration Folders (`iter.000000` to `iter.000024`):Each folder represents an iteration in the iterative refinement of the DP model, with three subfolders:  - `00.train`: Contains training data and outputs for the DP model during the current iteration.  - `01.model_devi`: Tracks deviations between DP predictions and ab initio results, guiding dataset selection for the next iteration.  - `02.fp`: Stores first-principles (FP) results from CP2K used to improve DP model accuracy.  

2.2 Other Key Files: - `cp2k.input`: Input file for CP2K, used for performing ab initio calculations on configurations during the iterative process.  - `dpdispatcher.log`: Log file tracking the progress of data dispatching and task execution.  - `dpgen.log`: Log file recording operations of DPGEN during dataset generation and force field development.  - `dpgen_nohup.sh`: Script for running DPGEN in the background.  - `machine_slurm_cp2k.json`: Configuration file specifying computing resources for CP2K simulations in a cluster environment.  - `param_cp2k.json`: Parameter file for CP2K calculations, defining simulation settings.  - `record.dpgen`: Record of iterative processes, including input parameters and outputs for each stage.

This organized structure ensures a systematic approach to dataset preparation, model training, and iterative refinement for developing accurate machine-learning potentials for Fe₁₋ₓO.

graph-compress.0330.pb is the final compressed DeepMD potential parameters.