Hello.
I have a question that’s been on my mind, so I’m posting it here in the community for the first time.
For the Material Project, there are two versions: the Legacy site and the newly revamped site. I used the Legacy site first, and while using the recently updated site, I have some questions, so I’m reaching out.
How to export a primitive cell
On the Legacy site, there was a clear button to extract the primitive cell as a cif file, but on the New site, even when I change the crystal structure to a primitive cell and click “export as,” it seems to still give me the conventional cell structure. Additionally, the crystal structures of the cells downloaded from the Legacy site and the Renewal site look slightly different. Why is that?
Different Energy Above Hull
Of course, I understand that this issue may vary depending on the calculation process and the data that has been accumulated over time.
However, I’m generally looking to check Hull Energy as a criterion for finding a stable phase, but I’m confused because the Hull Energy differs between Legacy and New, making it hard to determine which material has a more stable phase.
For example, in the case of Co, the Energy Above Hull for “mp-54” and “mp-102” is different in Legacy and New.
If Energy Above Hull = 0 indicates a more stable state, then in Legacy, mp-54 shows a value of 0, making it appear more stable, while in New, mp-102 shows 0, indicating that it is more stable. I would appreciate it if you could explain the reason for these differences.
We’re currently working on adding features to the next-gen site that were present in the legacy site, including a structure download - @ruoxiyang can hopefully comment if I’m wrong. For now, you can download primitive structures through the API as follows:
from mp_api.client import MPRester
with MPRester("your_api_key") as mpr:
# if you only need a single structure:
structure = mpr.get_structure_by_material_id("mp-xxxxxx")
primitive_cell = structure.get_primitive_structure()
# write to file, can do something like "structure.cif", "POSCAR", etc.
primitive_cell.to("structure.cif")
If you want to download larger batches of structures, use mpr.materials.summary.search(material_ids=[<list of MP IDs>]).
Additionally, the crystal structures of the cells downloaded from the Legacy site and the Renewal site look slightly different. Why is that?
This also partly answers your second question, but MP is in the process of recomputing all structures in our database (relaxed structures + thermodynamic + electronic properties) with a higher level of theory (r2SCAN vs. the older PBE GGA). You can find details about this here and here.
The hulls in the next-gen site use a mixture of r2SCAN and PBE+U hulls (see the previous links), so they’ll differ from the legacy site. We recommend using the next-gen site for more accurate hulls/stabilities.
First off, since it looks like the group still adding features to the next-gen site, it seems that downloading primitive cells isn’t possible on the site, right?
I’ll try downloading it using the API.
Next, you mentioned that the results from the next-gen site are more accurate.
In the case of Co, it’s known to undergo a phase transition from HCP structure at low temperatures to FCC structure at high temperatures around 825K [1, 2].
I believe the Hull Energy calculation method is generally conducted at 0K,
so shouldn’t the HCP structure, mp-54, show a more stable result than the FCC structure, mp-102?
First off, since it looks like the group still adding features to the next-gen site, it seems that downloading primitive cells isn’t possible on the site, right?
I’ll try downloading it using the API.
Yes please use the API for now.
I believe the Hull Energy calculation method is generally conducted at 0K,
so shouldn’t the HCP structure, mp-54, show a more stable result than the FCC structure, mp-102?
You’ve found an interesting edge case where r2SCAN actually predicts hcp Co to be dynamically unstable, and fcc Co to be stable - see p. 024102-4 of this paper by Liu et al. I wasn’t aware of this previously.
There’s no single DFT method right now that offers universal accuracy. r2SCAN is generally more accurate but not in this specific case.
