Supplementary data for the paper: "Dataset Merging Strategy for Constructing a Machine-Learning Force Field of CeO2 Surfaces with an Oxygen Vacancy"

This Zenodo record contains supplementary computational data for the paper “Dataset Merging Strategy for Constructing a Machine-Learning Force Field of CeO2 Surfaces with an Oxygen Vacancy”.

ChemRxiv DOI:https://doi.org/10.26434/chemrxiv.15003015/v1

The deposited data include VASP machine-learned force field (MLFF) training databases and force-field files, initial structures used for on-the-fly MLFF generation, validation structures used to evaluate the trained MLFFs, and nudged elastic band (NEB) paths discussed in the manuscript.

The uploaded files are organized into four directories:

01_dataset_and_force_field.zip/This directory contains VASP MLFF-related files obtained at the FF3000 stage of on-the-fly training. The ML_ABN files contain the collected training databases generated during on-the-fly MLFF training, while the ML_FFN files are the binary VASP MLFF files generated from the corresponding training data.

02_training_structures.zip/This directory contains the initial structures used for the on-the-fly MLFF training simulations. The file names indicate the type of facet, surface termination, slab thickness, and oxygen-vacancy condition.

03_validation_structures.zip/This directory contains initial structures used to validate the accuracy of the trained MLFFs. The validation structures are grouped into medium, large, and huge model sizes. These structures were used to compare MLFF-predicted energies with the corresponding reference calculations.

04_neb_paths_and_energies.zip/This directory contains optimized NEB paths discussed in the manuscript. The data include DFT-optimized NEB paths, MLFF-optimized NEB paths, MLFF single-point energies evaluated along DFT-optimized paths, and lattice-vector information required to convert the xyz files into periodic POSCAR files.

File naming conventions:- 100-Ce: Ce-terminated CeO2 (100) facet.- 100-O: O-terminated CeO2 (100) facet.- 110: CeO2 (110) facet.- 111: CeO2 (111) facet.- 221: CeO2 (221) facet.- 331: CeO2 (331) facet.- 2L, 3L: number of layers, corresponding to the slab thickness labels used in the manuscript.- 0V, 1V: number of oxygen vacancies in the model.- merge_...: MLFF trained using a merged training database.- dft-neb: NEB path optimized using DFT.- mlff-neb: NEB path optimized using MLFF.- mlff-singlepoint: MLFF single-point energies evaluated for a path optimized by DFT.

Software:The MLFF files were generated and used with VASP 6.4.2.

Notes on VASP MLFF files:- ML_ABN contains the training database generated during on-the-fly MLFF training.- ML_FFN is the binary force-field file produced by VASP.

Notes on NEB xyz files:The xyz files contain the structures and energies of images along the NEB paths. Because these xyz files do not include periodic boundary conditions, the corresponding lattice vectors are provided separately in the lattice_vectors_... files.