×

Recommended by

Indexed by

Properties of α-brass nanoparticles. 1. Neural network potential energy surface

Jan Weinreich1*, Martín Leandro Paleico1*, Anton Roemer1*

1 Institut für Physikalische Chemie, University of Göttingen, Germany

* Corresponding authors emails: jan.weinreich@univie.ac.at, martin.paleico@mpibpc.mpg.de, anton.roemer@stud.uni-goettingen.de
DOI10.24435/materialscloud:94-aq [version v1]

Publication date: Sep 26, 2021

How to cite this record

Jan Weinreich, Martín Leandro Paleico, Anton Roemer, Properties of α-brass nanoparticles. 1. Neural network potential energy surface, Materials Cloud Archive 2021.153 (2021), doi: 10.24435/materialscloud:94-aq.

Description

**Data for Properties of α-Brass Nanoparticles. 1. Neural Network Potential Energy Surface** Jan Weinreich, Anton Römer, Martín Leandro Paleico, and Jörg Behler 53 841 reference structures of alpha brass (less 40 % Zn) with following split - 4009 brass clusters - 8492 molten brass bulk structures - 8964 copper slabs, and 16 878 brass slabs - 5377 copper bulk structures - 10 121 brass bulk structures have been included. 53 841 total energies and 8 903 340 force components. The ranges of values for the energies and force components to be fitted have a width of about 2 eV/atom and 15 eV/Å, respectively. However, some structures may have slightly higher Zn content as discussed in Fig 3 (https://arxiv.org/abs/2001.10906) The archive contains an easily usable npz file as well as the original input.data file used to fit the potential energy surface. Additionally a Jupyter notebook describes in great detail how the data was converted to the npz format and how to read the data e.g. for subsequent use with python. In addition an example VASP calculation was added to provide detailed information about how the reference data was calculated with DFT. DETAILS: - DFT PB VASP-5.3 target accuracy of the total energy few meV/atom - Convergence tests with respect to the number of k-points showed that in order to fulfill this criterion a k-point grid of 12 × 12 × 12 is needed for a conventional four-atom copper fcc unit cell with a lattice constant of about 3.63 Å along with a plane wave cutoff energy of 500 eV and projector augmented wave potentials.(61,63) - Larger systems have been calculated using an adapted k-point grid corresponding to the same k-point density. The Γ-point centered k-point grids have been constructed employing the Monkhorst–Pack scheme. For surface calculations, 4–14 layer slabs with a total vacuum thickness of at least 8 Å have been used. - In the case of cluster calculations, which have also been treated in a periodic setup, the periodic images of the clusters have been separated by at least 8 Å in all three spatial directions. The convergence of very large clusters with diameters of d ≈ 22 Å, which have been used to include specific atomic environments in the data set, has been extensively tested and we found that using the Γ-point only is sufficient to reach the required convergence level.

Materials Cloud sections using this data

No Explore or Discover sections associated with this archive record.

Files

File name Size Description
brass_DFT_data.zip
MD5md5:f88a44e5d69b02f30d68aeadfaeae878
221.6 MiB Brass structures, energies and forces from DFT used to fit a potential energy surface using a neural network potential & example DFT calculation with VASP input used

License

Files and data are licensed under the terms of the following license: Creative Commons Attribution 4.0 International.
Metadata, except for email addresses, are licensed under the Creative Commons Attribution Share-Alike 4.0 International license.

External references

Journal reference (you may also read the preprint version: https://arxiv.org/abs/2001.10906)
Journal reference (you may also read the arxiv version: https://arxiv.org/abs/2103.14130)

Keywords

database machine learning neural networks DFT alloys brass potential energy surfaces

Version history:

2021.153 (version v1) [This version] Sep 26, 2021 DOI10.24435/materialscloud:94-aq