The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies

The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies Scott Kirklin James E. Saal Bryce Meredig Alexander Thompson Jeff W. Doak Muratahan Aykol Stefan Rühl Chris Wolverton

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies 2015 npj Computational Materials 10.1038/npjcompumats.2015.10 kirklin2015-the-open-quantum-materials-databas kirklin2015-the-open-quantum-materials-databas.txt, the project's staged copy 08f4a10fd1c0a770877fe0673e24332f1825560f cc-by materials-science

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them nearly 300,000 DFT total-energy calculations of inorganic compounds, and how accurate they are when checked against 1,670 experimental formation energies Quotation withheld because this record is not cleared for redistribution. how much of the gap between calculated and measured formation energies belongs to the calculation Quotation withheld because this record is not cleared for redistribution. an openly downloadable database of DFT energies, and the largest comparison of DFT with experimental formation energies to date - with the experimental measurements also compared against each other Quotation withheld because this record is not cleared for redistribution. method Quotation withheld because this record is not cleared for redistribution. a reference standard whose internal disagreement is nearly as large as the error it is being used to measure Quotation withheld because this record is not cleared for redistribution. simulation simulation Quotation withheld because this record is not cleared for redistribution. the apparent DFT error is 0.096 eV/atom. The authors then measured how far different experimental determinations of the SAME formation energy sit from each other and got 0.082. Almost the whole gap is the width of the reference. THE CONTROL IS THAT THE STANDARD WAS ASKED HOW WELL IT AGREES WITH ITSELF, and almost nobody asks measure the ruler before blaming the thing measured Quotation withheld because this record is not cleared for redistribution. on the 75 intermetallic compounds shared by the two thermochemical databases, the databases differ by 0.082 eV/atom and OQMD sits 0.071 from experiment. A model cannot be scored below the noise floor of its reference, and here it is already under it the calculation closer to experiment than experiment is to itself Quotation withheld because this record is not cleared for redistribution. the entire OQMD is downloadable without conditions or limitations, and it is computed from crystal structures obtained through a partnership with the ICSD. THE OPENNESS IS OF THE CALCULATION, NOT OF THE MEASUREMENT IT STARTED FROM - which is exactly the position the corpus itself is in an open derivative of a closed source Quotation withheld because this record is not cleared for redistribution. the elements whose predicted ground state is wrong are the ones held together by van der Waals forces, which the semi-local functionals omit by construction. The error is attributed to a missing term in the theory rather than to a fitting failure, and the paper says which elements and why a known blind spot, named by physics rather than by residual Quotation withheld because this record is not cleared for redistribution. having compared what is known, the same stability analysis over ICSD compounds and hypothetical structures yields about 3,200 compounds predicted to exist that have never been characterised. The negative space of the experimental record, read off the same calculation prediction as the leftover of a comparison Quotation withheld because this record is not cleared for redistribution. the ground-state structure of a whole group of elements cannot be determined reliably with the semi-local functionals used, because van der Waals interactions are missing Quotation withheld because this record is not cleared for redistribution. for noble gas and molecular solids the change in predicted stabilities cannot be substantiated, since van der Waals contributes a significant portion of the energy Quotation withheld because this record is not cleared for redistribution. where two experimental databases disagree there is no principled way to say which measurement is wrong, so the residual cannot be fully assigned Quotation withheld because this record is not cleared for redistribution. the structures are obtained from the ICSD by partnership; the openness of the derived data set does not extend to the source Quotation withheld because this record is not cleared for redistribution. inter-rater reliability logic from measurement theory - asking how far two independent determinations of the same quantity sit from each other before scoring a third against either Quotation withheld because this record is not cleared for redistribution. the deviation between different experimental measurements of the same formation energy, computed wherever multiple sources exist, before any error is attributed to the calculation eliminates: the whole model-versus-experiment residual being read as model error - the reference's own spread is 0.082 against an apparent 0.096 · outcome: held Quotation withheld because this record is not cleared for redistribution. the two experimental thermochemical databases compared on the 75 intermetallic compounds they have in common eliminates: a single compilation standing for 'experiment' - two of them disagree by more than the calculation differs from either · outcome: held Quotation withheld because this record is not cleared for redistribution. elements whose predicted ground state is wrong grouped by physical cause - missing van der Waals interactions, and f-electrons - rather than reported as residual eliminates: a list of failures with no account of them; each group is attached to a known limitation of the functional · outcome: held Quotation withheld because this record is not cleared for redistribution. higher-cutoff calculations run on the noble gas and molecular solids to test whether the discrepancy is a convergence artefact eliminates: nothing here - the recomputation reproduces the experimental ground states for He, F and Ar, but the change cannot be substantiated because van der Waals is still missing · outcome: partial Quotation withheld because this record is not cleared for redistribution. the entire data set published for download without conditions or limitations, so the comparison can be repeated by anyone eliminates: a comparison nobody can repeat - the whole data set is published without conditions, so the numbers can be checked by anyone who doubts them · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. OQMD OQMD Quotation withheld because this record is not cleared for redistribution. ICSD ICSD Quotation withheld because this record is not cleared for redistribution. DFT DFT Quotation withheld because this record is not cleared for redistribution. formation energies formation energies Quotation withheld because this record is not cleared for redistribution. experimental uncertainties experimental uncertainties Quotation withheld because this record is not cleared for redistribution. databases databases Quotation withheld because this record is not cleared for redistribution. van der Waals van der Waals Quotation withheld because this record is not cleared for redistribution. experiment experiment Quotation withheld because this record is not cleared for redistribution. AFLOWLIB AFLOWLIB Quotation withheld because this record is not cleared for redistribution. the DFT error - AS APPARENT: 0.096 eV/atom mean absolute error against experimental formation energies Quotation withheld because this record is not cleared for redistribution. the DFT error - AFTER THE REFERENCE IS CHECKED: a significant fraction of it attributable to experimental uncertainty, because different experimental measurements of the same quantity differ by 0.082 eV/atom Quotation withheld because this record is not cleared for redistribution. agreement - BETWEEN TWO EXPERIMENTAL DATABASES: 0.082 eV/atom on the 75 intermetallic compounds they share Quotation withheld because this record is not cleared for redistribution. agreement - BETWEEN OQMD AND EXPERIMENT: 0.071 eV/atom on the same class, which is closer than the two experiments are to each other Quotation withheld because this record is not cleared for redistribution. `open` - AS OQMD MEANS IT: the entire data set downloadable without any conditions or limitations Quotation withheld because this record is not cleared for redistribution. `open` - AS THE SOURCE IS: the structures come from the ICSD by partnership, and that database is not the thing being given away Quotation withheld because this record is not cleared for redistribution. qualifies → the practice of reporting DFT accuracy as a single error against experiment, by showing the reference disagrees with itself by nearly as much Quotation withheld because this record is not cleared for redistribution. critiques → the two experimental thermochemical databases, which differ by 0.082 eV/atom on the compounds they share Quotation withheld because this record is not cleared for redistribution. distinguishes-from → the Materials Project, the Computational Materials Repository and AFLOWLIB, the neighbouring high-throughput efforts Quotation withheld because this record is not cleared for redistribution. predicts → the existence of about 3,200 compounds that have not been experimentally characterised Quotation withheld because this record is not cleared for redistribution. DFT —distinct-from→ experiment DFT experiment differs in: whether a residual is model error or reference spread · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): whether a residual is model error or reference spread · the relation stated: where two experiments disagree, the calculation is often much closer to one of them than they are to each other, so the residual cannot simply be read as the calculation's error Audit 2026-09-28: DOUBTFUL - read against its span and kept, but DFT is closer to one experiment than the other; distinct-from is loose. Quotation withheld because this record is not cleared for redistribution. the apparent DFT error against experiment is 0.096 eV/atom and different experimental measurements of the same quantity differ by 0.082 the apparent DFT error against experiment is 0.096 eV/atom and different experimental measurements of the same quantity differ by 0.082 Quotation withheld because this record is not cleared for redistribution. on the 75 intermetallic compounds two experimental databases share they disagree by 0.082 eV/atom while OQMD is 0.071 from experiment on the 75 intermetallic compounds two experimental databases share they disagree by 0.082 eV/atom while OQMD is 0.071 from experiment Quotation withheld because this record is not cleared for redistribution. where experiments disagree, DFT is often significantly closer to one experiment than the other where experiments disagree, DFT is often significantly closer to one experiment than the other Quotation withheld because this record is not cleared for redistribution. a significant fraction of the DFT-experiment error may be attributed to experimental uncertainties a significant fraction of the DFT-experiment error may be attributed to experimental uncertainties Quotation withheld because this record is not cleared for redistribution. elemental ground states held together by van der Waals forces cannot be determined reliably with semi-local functionals elemental ground states held together by van der Waals forces cannot be determined reliably with semi-local functionals Quotation withheld because this record is not cleared for redistribution. about 3,200 compounds are predicted to exist that have not been experimentally characterised about 3,200 compounds are predicted to exist that have not been experimentally characterised Quotation withheld because this record is not cleared for redistribution. the entire database is released for download without any conditions or limitations the entire database is released for download without any conditions or limitations Quotation withheld because this record is not cleared for redistribution. a high-throughput database of nearly 300,000 DFT total-energy calculations of ICSD compounds and decorations of common crystal structures, released in full without restrictions OQMD Quotation withheld because this record is not cleared for redistribution. eV/atom - the apparent mean absolute error between experiment and the calculations 0.096 Quotation withheld because this record is not cleared for redistribution. eV/atom - how far different EXPERIMENTAL measurements of the same quantity sit from each other 0.082 Quotation withheld because this record is not cleared for redistribution. eV/atom - OQMD against experiment on intermetallics, closer than the two experimental databases are to each other 0.071 Quotation withheld because this record is not cleared for redistribution. MAE MAE Quotation withheld because this record is not cleared for redistribution. formation energies formation energies Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1038/npjcompumats.2015.10.