oib12/nmdl-mp-20_atomic-dos

MP-20 with atomic DOS

MP-20 crystal structures with the atom-projected density of states (ADOS) of every atom attached, computed by our own DFT calculations and compressed to a 30-dimensional latent vector per atom. This is the mp-20 training set of the MatterGen-e- model atomic_dos_joint_model_mp_20.

Code: nmdl-mizo/mattergen-e- · Paper: npj Comput Mater 12, 332 (2026)

Columns

Column	Content
material_id	Materials Project ID
cif	Structure as a CIF string (written by pymatgen in P1)
atomic_dos	ADOS latents: a list of 30-element lists, one per site, in the site order of cif
formation_energy_per_atom, dft_band_gap, energy_above_hull, pretty_formula, elements, spacegroup.number	Taken unchanged from MP-20

Only atomic_dos comes from our calculations; the structures and the other properties are those of MP-20.

The atomic_dos values are not a spectrum — they are the latent vectors of the autoencoder mp_20_-15eV_to_5eV_30dim.pt in oib12/nmdl-mattergen-e_atomic-dos-autoencoder. Decode them with that checkpoint; the other one gives a plausible-looking but meaningless DOS.

Decoded, each latent is a spin-summed DOS projected onto one site, on 512 points covering E − E_F ∈ [−15, 5] eV (numpy.linspace(-15.0, 5.0, 512)).

Loading

import json
import pandas as pd
from pymatgen.core import Structure

df = pd.read_csv("train.csv")
row = df.iloc[0]
structure = Structure.from_str(row.cif, fmt="cif")
latents = json.loads(row.atomic_dos)  # len(latents) == len(structure), each of length 30

Source and license

MP-20 (Xie et al., 2022) is built from Materials Project structures with at most 20 atoms per unit cell (MP, CC BY 4.0). This dataset is released under CC BY 4.0 as well. Please cite MP-20 and the Materials Project (Jain et al., 2013) alongside the citation below.

Limitations

• 512 points compressed into 30 numbers: fine structure of the DOS is not recoverable.
• Spin-summed, atom-projected total DOS only — not spin- or orbital-resolved.

Citation

Okuda, I., Takahara, I. & Mizoguchi, T. Inverse materials design via joint generation of crystal structures and local electronic descriptors. npj Comput Mater 12, 332 (2026). https://doi.org/10.1038/s41524-026-02322-4

@article{Okuda2026MatterGenE,
  title   = {Inverse materials design via joint generation of crystal structures and local electronic descriptors},
  author  = {Okuda, Ibuki and Takahara, Izumi and Mizoguchi, Teruyasu},
  journal = {npj Computational Materials},
  volume  = {12},
  pages   = {332},
  year    = {2026},
  doi     = {10.1038/s41524-026-02322-4},
  url     = {https://doi.org/10.1038/s41524-026-02322-4},
}