Database version 2026.09.28 is going live today (Oct 19, 2026) at about the same time as this post, see the release notes for a summary. As always, please be sure to update the mp-api Python client to ensure compatibility (current stable version is 0.47.0).
Thermo Data Corrections:
This release was solely focused on incorporating an important correction to the code used to generate the mixing scheme thermo data + phase diagrams (effects GGA_GGA+U_R2SCAN data in the thermo collection, phase diagrams, and summary collection). For emphasis, this has only effected data with thermo_type=GGA_GGA+U_R2SCAN. The GGA_GGA+U only and r2SCAN only thermodynamic hulls are unaffected.
A big thank you to @kavanase for finding the root cause, submitting the patch, and looping the MP staff into the issue. Briefly paraphrasing Seán’s findings: the mixing scheme code relied on filtering and sorting a chemical system’s Entrys based on hashed f-string representations of the Entrys. CPython randomizes string hashing per process, so this initial sorting step was non-deterministic. Definitely check out Seán’s pymatgen PR for the full explanation. This also addressed a number of open issues in pymatgen related to the mixing scheme code: pymatgen #3113, pymatgen #4261
Along with this I have prepared a brief dashboard summarizing the effects that had on the thermodynamic data for this database release (hopefully discourse handles the attachment well). The dashboard should be directly viewable in any browser after downloading the attached html document:
gga_gga_u_r2scan_overview_dashboard.html (6.4 MB)
Thankfully only a small percentage of materials were effected, most of which already marked as unstable in the thermo data. I have also attached the notebook used to generate this overview for reproducibility (+ further exploration for the curious):
mixing_scheme_changes_db_v2028_09_28.ipynb (40.3 KB)
mp-api Updates:
Seán also contributed patches to the mp-api Python client related to unit cell re-scaling and mixed entry handling: mp-api #1121, mp-api #1126, both of which are relevant to the mixing scheme data corrections.
The mp-api now also better communicates what is happening during data retrieval/data re-shaping. A quiet context manager is now also included with mp-api if you find that the client is too verbose:
>>> from mp_api.client import MPRester, quiet
>>> mpr = MPRester()
>>> entries = mpr.get_entries_in_chemsys( "Li-Fe-P-O", additional_criteria={"thermo_types": ["GGA_GGA+U_R2SCAN"]}, compatible_only=False)
MPRestWarning Mixed GGA(+U)/r2SCAN entries can only be placed on a common energy scale with `compatible_only = True`, so these uncorrected entries are not suitable for constructing a phase diagram. Either use `compatible_only = True`, or request a single functional with `additional_criteria = {"thermo_types":
["GGA_GGA+U"]}`.
<stdin>:1
✓ Retrieved 918 ThermoDoc documents in 2.6s
✓ Retrieved 55 ThermoDoc documents in 0.3s
>>> with quiet():
... entries = mpr.get_entries_in_chemsys( "Li-Fe-P-O", additional_criteria={"thermo_types": ["GGA_GGA+U_R2SCAN"]}, compatible_only=False,)
...
>>>
There are additional ways to configure the way the mp-api client will communicate with you (or in the context of your application using the client), see the module doc strings for additional config examples: mp_api.client.core._display