May 14, 2025 Dataset Open Access

Boron data set for machine learning applications

Fiedler, Lenz; Cangi, Attila

Boron data set for machine learning applications

This dataset contains DFT inputs, outputs, LDOS data and bispectrum descriptor vectors for an α-rhombohedral boron cell of 144 atoms at room temperature and ambient mass density. All simulations have been performed at an LDOS converged k-grid of 4x4x4 k-points.

This dataset contains one .zip file for each of its five type of data (bispectrum descriptors, LDOS, DFT inputs, DFT outputs and trained models).

Authors:

- Fiedler, Lenz (HZDR / CASUS)
- Cangi, Attila (HZDR / CASUS)

Affiliations:

HZDR - Helmholtz-Zentrum Dresden-Rossendorf
CASUS - Center for Advanced Systems Understanding

Dataset description

- Total size: 26 GB
- System: B144
- Temperature(s): 298K
- Mass density(ies): 2.483 gcc
- Crystal Structure: amorphous (material mp-160 in the materials project)
- Number of atomic snapshots: 15
- Contents:
    - ideal crystal structure: no
    - MD trajectory: no
    - Atomic positions: no
    - DFT inputs: yes
    - DFT outputs (energies): yes
    - SNAP vectors: yes
        - dimensions: 108x108x35x94 (last dimension: first three entries are x,y,z coordinates, data size is 91)
        - units: a.u.
    - LDOS vectors: yes
        - dimensions: 108x108x35x241
        - units: 1/(eV*Angstrom^3)
    - trained networks: yes

Dataset structure

A .zip file is included for each for each of its five type of data:

- ldos.zip: holds the LDOS vectors (one HDF5 file per snapshot)
- bispectrum.zip: holds the bispectrum fingerprint vectors  (one HDF5 file per snapshot)
- dft_outputs: holds the outputs from the DFT calculations, i.e. energies and simulation parameters in a .json format (one per snapshot)
- dft_inputs: holds the inputs for the DFT calculations, in the form of a QE input file (one per snapshot)
- models: holds five trained NN models for the data set