Validity of pre-computed phase diagrams for GGA/GGA+U/R2SCAN scheme

I’ve noticed that GGA/GGA+U/R2SCAN phase diagrams require to be constructed locally ^ref, because the correction to entries vary with the chemical systems. So, does it also implies the pre-computed phase diagrams, retrieved by mpr.get_phase_diagram_from_chemsys(‘x-x-x’, thermo_type = ThermoType.GGA_GGA_U_R2SCAN) method, are invalid? Are the pre-computed phase diagrams built with entries on the thermo data endpoint or having been mixing processed according to chemical systems?

Kai Pei

Hi @peikai, looks like you and @munrojm discussed mixing phase diagrams in this thread. If this is a follow-up, I’m tagging @munrojm in the hope that he can help.

Hi @tschaume, yes, it is! Thanks.

@peikai, the pre-computed GGA/GGA+U/R2SCAN phase diagrams have been created using proper application of the mixing scheme for the given chemical system. The canonical ComputedStructureEntry objects provided through the thermo endpoint with corrections are only from the “home” chemical system for that givevn material (i.e. Si-O for SiO2). That means if you want to compute the Si-O-P phase diagram yourself for example, you would need to re-apply the mixing scheme to the entries you pulled from the API. Unless you really need to/want do this, we recommend pulling the pre-computed phase diagram data.

– Jason

Hi @munrojm,
Great! And then, do they have comprehensive coverage on all chemical systems in the database? After all, there may be a lot of combinations of chemical systems for PD generation. This would determine the availability of pre-computed PDs in terms of the GGA/GGA+U/R2SCAN scheme.

I’ve noticed that there is a chemical system field at the Summary data endpoint in the new API, but however, I tested some chemical systems there but they have no corresponding pre-computed phase diagrams.

Kai Pei

@peikai, yes we should have complete coverage for chemical systems that are possible given the material entries in the database. Which chemical system are you querying with?

– Jason

Here are several chemical systems failed to retrieve pre-computed PDs in GGA/GGA+U/R2SCAN scheme:
C-Co-Li-Mn-O-P, C-Cr-Li-Mn-O-P, H-Li-Mn-O-P-V.

They come from Summary data endpoint. And PDs are retrieved in this way:

from emmet.core.thermo import ThermoType
from mp_api.client import MPRester

mpr.thermo.get_phase_diagram_from_chemsys('C-Co-Li-Mn-O-P', thermo_type = ThermoType.GGA_GGA_U_R2SCAN)`

Kai Pei

@peikai, I actually misspoke. Apologies for my previous answer. Currently, we only store pre-computed phase diagrams for chemical systems <= 5 elements. However, this is changing in the next build and we should have complete coverage.

For the chemical systems you list, you will have to pull entries, apply the corrections locally, and construct the phase diagrams yourself for now.

– Jason

Thanks @munrojm,
Well, what about chemical systems > 5 elements? I noticed drawbacks discussion of the R2SCAN/GGA(+U) mixing scheme when dealing with large chemical systems in its paper. But has any quantitative analysis in terms of element number of chemical system have been conducted, thus let >5 chemical systems be neglected?

I have not personally done that, so I am not sure. To be clear, we do compute phase diagrams for all chemical systems represented in the database and take energy above hull values from them. We just haven’t stored them for systems with elements > 5 before now. This has historically been due to infrastructure limitations that we have recently addressed.

– Jason

Get it. Would these pre-computed phase diagrams for >5 elements be published in due course recently? considering here is only one route to construct R2SCAN/GGA(+U) PD for them currently, which requires locally processing.

Kai Pei

@peikai, yes they should be available soon. I am aiming for the next database release.

– Jason

Great, and thanks!


1 Like

Thread closed due to inactivity, please open a new thread to address related issues.