OQMD

Structures and energies of the Open Quantum Materials Database, preprocessed for machine learning.

File	Structures	Atoms	Content
all.hdf5	1,407,269	14,266,404	one structure per entry of OQMD v1.8
unique.hdf5	1,259,553	13,089,068	same, without the entries marked as duplicates in OQMD
all_index.parquet, unique_index.parquet			OQMD entry_id, calculation_id and structure_id of each row of the file above
oqmd.tar.gz, dump.py	1,021,320	7,117,087	older export (2024) in another layout, kept unchanged, see the end of this page

Structures have 1 to 312 atoms (median 6).

How structures are selected

Source: the SQL dump of OQMD v1.8 (qmdb__v1_8__022026.sql.gz, February 2026), 1,407,285 entries.

For each entry, one calculation is selected:

• its output structure must be consistent: the number of atoms declared for the structure equals its number of rows in the atoms table (47,859 of 8,374,305 structures are not);
• the calculation is converged, its label starts with static or standard, and does not end with lda;
• among those, the one with the lowest energy per atom is kept (lowest calculation id when equal).

1,407,269 entries have such a calculation. Nothing else is filtered in all.hdf5.

unique.hdf5 groups each entry with the entry it duplicates according to OQMD (entries.duplicate_of_id) and keeps the entry with the lowest energy per atom of each group; when energies are exactly equal the original entry is kept. 147,716 entries are removed this way. Duplicates that OQMD does not mark are not removed.

Labels

Field	Source	Meaning
total_energy	calculations.energy_pa × number of atoms	DFT total energy of the cell in eV
formation_energy_per_atom	formation_energies.delta_e (fit standard)	formation energy in eV/atom
energy_above_hull	formation_energies.stability (fit standard)	distance to the convex hull in eV/atom, stored as in OQMD: negative for phases below the hull
band_gap	calculations.band_gap	band gap in eV, NaN for 3,155 structures of all.hdf5

Labels are not cleaned. all.hdf5 contains unphysical values from failed calculations: 559 structures with a positive total energy (up to +704 eV/atom), 668 with a stability above 5 eV/atom, and one negative band gap. Filter them for training.

	all.hdf5	unique.hdf5
stability ≤ 0	97,228	96,246
stability ≤ 0.1 eV/atom	458,906	402,102

Format

Flat concatenated arrays:

Dataset	Shape	Type	Content
lattice	(n_struct, 3, 3)	float32	lattice vectors as rows, in angstrom
num_atoms	(n_struct,)	int64	number of atoms of each structure
ptr	(n_struct,)	int64	index of the first atom of each structure in x and z
x	(n_atoms, 3)	float32	fractional coordinates in [0, 1)
z	(n_atoms,)	int64	atomic numbers
total_energy, formation_energy_per_atom, energy_above_hull, band_gap	(n_struct,)	float32	labels, NaN when unknown

The lattice is rebuilt from cell lengths and angles, so it is in the standard orientation (first vector along x). Fractional coordinates are wrapped into the cell. The batch array is an artefact of the writer and is not used when reading. Structures are ordered by OQMD entry id; row i of an index file describes structure i.

Usage

from pycrystalgen.data import HDF5Dataset

dataset = HDF5Dataset.from_hub("materials-toolkits/oqmd", "unique.hdf5")
structures = dataset[:32]  # batched Structures

Processing and checks

Built with pycrystalgen (scripts/build_oqmd.py): the dump is read as a stream without a MySQL server, the tables are extracted, structures are written to CIF then preprocessed to HDF5.

Every structure of both files was compared with the tables of the dump: number of atoms, elements, fractional coordinates, lattice and the four labels.

Older export (oqmd.tar.gz)

oqmd.tar.gz holds a 2024 export produced by dump.py from an earlier version of the database, in the layout of materials-toolkit (data.hdf5, batching.json). It contains 1,021,320 structures with an energy per atom and an energy above the hull recomputed by materials-toolkit (not OQMD's stability). The selection rules above follow the ones of dump.py. 1,007,099 of its entries are also in all.hdf5; 1,005,090 of them have the same selected structure.

References

• J. E. Saal, S. Kirklin, M. Aykol, B. Meredig and C. Wolverton, Materials Design and Discovery with High-Throughput Density Functional Theory: The Open Quantum Materials Database (OQMD), JOM 65, 1501 (2013).
• S. Kirklin et al., The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies, npj Computational Materials 1, 15010 (2015).