Why POTCAR hash from mp_api doesn't match local POTCAR file hash: explained by md5_header_hash vs md5_computed_file_hash.

[hongyi-zhao]

See my following checking:

In [1]: from mp_api.client import MPRester
   ...: 
   ...: material_id = "mp-126"
   ...: with MPRester() as mpr:
   ...:     # 获取elastic数据来获取task id
   ...:     elasticity_doc = mpr.materials.elasticity.search(material_ids=[material_id])
   ...:     opt_id = elasticity_doc[0].fitting_data.optimization_task.string
   ...: 
   ...:     # 使用task id获取详细信息
   ...:     opt_doc = mpr.materials.tasks.search([opt_id],
   ...:         fields=["input", "calcs_reversed"])
   ...: 
   ...:     # 提取 potcar_spec
   ...:     potcar_spec = opt_doc[0].calcs_reversed[0].input.potcar_spec
   ...:     print(f"POTCAR specifications: {potcar_spec}")
   ...: 
Retrieving ElasticityDoc documents: 100%|████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 10618.49it/s]
Retrieving TaskDoc documents: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 12557.80it/s]
POTCAR specifications: [PotcarSpec(titel='PAW_PBE Pt 05Jan2001', hash='a604ea3c6a9cc23c739b762f625cf449', summary_stats=None)]

The potcar on my machine:

werner@x13dai-t:~/Desktop/pot$ ug 'PAW_PBE Pt 05Jan2001' -l | xargs md5sum
b1f52f99a6ad883e9383660a9fc1eebf  potpaw_PBE/Pt_ZORA/POTCAR
079fdad3df93dbceac18c56a69852e29  potpaw_PBE/Pt/POTCAR

As you can see, the hash doesn't match. So, what's the exact potcar version used by MP?

Regards,
Zhao

[tschaume]

That's a good question but it's better asked on our forum for the benefit of the whole community. Please search for existing explanations/solutions on the forum before posting your question there. Thanks!

[hongyi-zhao]

See here for the related discussion.

[hongyi-zhao]

I understand the issue now. There is no problem with potcar. The mp_api query shows the MD5 hash of the metadata defining the PotcarSingle, aka, PotcarSingle.md5_header_hash, not the MD5 hash of the entire PotcarSingle, aka, PotcarSingle.md5_computed_file_hash. The above two types of information are stored in pymatgen as below:

# The `MD5 hash of the metadata defining the PotcarSingle`:
werner@x13dai-t:~/Public/repo/github.com/materialsproject/pymatgen.git$ ug 'a604ea3c6a9cc23c739b762f625cf449' -l
tests/files/surfaces/ucell_entries.txt
docs/pymatgen.io.vasp.html
docs/searchindex.js
src/pymatgen/io/vasp/MITRelaxSet.yaml
src/pymatgen/io/vasp/vasp_potcar_pymatgen_hashes.json
tests/files/io/vasp/fixtures/grid_data_files/vasp_inputs_for_grid_check.json

# The `MD5 hash of the entire PotcarSingle`:
werner@x13dai-t:~/Public/repo/github.com/materialsproject/pymatgen.git$ ug 'b1f52f99a6ad883e9383660a9fc1eebf' -l
src/pymatgen/io/vasp/vasp_potcar_file_hashes.json
werner@x13dai-t:~/Public/repo/github.com/materialsproject/pymatgen.git$ ug '079fdad3df93dbceac18c56a69852e29' -l
src/pymatgen/io/vasp/vasp_potcar_file_hashes.json

The PotcarSpec used by the MP database and retrieved by mp_api can be generated as follows:

In [15]: from pymatgen.io.vasp import Potcar
    ...: from collections import namedtuple
    ...: import os
    ...: 
    ...: PotcarSpec = namedtuple('PotcarSpec', ['titel', 'hash', 'summary_stats'])
    ...: 
    ...: # Expand the ~ in the path
    ...: filepath = os.path.expanduser("~/Public/hpc/vasp/pot/pmg_potcar/POT_GGA_PAW_PBE/POTCAR.Pt.gz")
    ...: 
    ...: # Read POTCAR and create specs
    ...: potcar = Potcar.from_file(filepath)
    ...: 
    ...: specs = [
    ...:     PotcarSpec(
    ...:         titel=p.titel,
    ...:         hash=p.md5_header_hash,
    ...:         summary_stats=p._summary_stats
    ...:     )
    ...:     for p in potcar
    ...: ]
    ...: 
    ...: print(f"POTCAR specifications: {specs}")
POTCAR specifications: [PotcarSpec(titel='PAW_PBE Pt 05Jan2001', hash='a604ea3c6a9cc23c739b762f625cf449', summary_stats={'keywords': {'header': ['dexc', 'eatom', 'eaug', 'enmax', 'enmin', 'iunscr', 'lcor', 'lexch', 'lpaw', 'lultra', 'ndata', 'orbitaldescriptions', 'pomass', 'qcut', 'qgam', 'raug', 'rcloc', 'rcore', 'rdep', 'rmax', 'rpacor', 'rrkj', 'rwigs', 'step', 'titel', 'vrhfin', 'zval'], 'data': ['localpart', 'gradientcorrectionsusedforxc', 'corecharge-density(partial)', 'atomicpseudocharge-density', 'nonlocalpart', 'reciprocalspacepart', 'realspacepart', 'reciprocalspacepart', 'realspacepart', 'nonlocalpart', 'reciprocalspacepart', 'realspacepart', 'reciprocalspacepart', 'realspacepart', 'nonlocalpart', 'reciprocalspacepart', 'realspacepart', 'reciprocalspacepart', 'realspacepart', 'pawradialsets', '(5e20.12)', 'augmentationcharges(nonsperical)', 'uccopanciesinatom', 'grid', 'aepotential', 'corecharge-density', 'kineticenergy-density', 'pspotential', 'corecharge-density(pseudized)', 'pseudowavefunction', 'aewavefunction', 'pseudowavefunction', 'aewavefunction', 'pseudowavefunction', 'aewavefunction', 'pseudowavefunction', 'aewavefunction', 'pseudowavefunction', 'aewavefunction', 'pseudowavefunction', 'aewavefunction', 'endofdataset']}, 'stats': {'header': {'MEAN': 49.75002565203251, 'ABSMEAN': 49.81755410731707, 'VAR': 11430.679203375377, 'MIN': -4.114, 'MAX': 729.1171}, 'data': {'MEAN': 101824.1204566872, 'ABSMEAN': 101837.40806970962, 'VAR': 565262541893.0613, 'MIN': -175.707504926, 'MAX': 7968615.30025}}})]

The following code snippet can be used to compute the two hashes mentioned above:

In [31]: from pymatgen.io.vasp.inputs import PotcarSingle

In [32]: psingle = PotcarSingle.from_file('/home/werner/Desktop/pot/potpaw_PBE/Pt/POTCAR')
In [33]: psingle.md5_header_hash
Out[33]: 'a604ea3c6a9cc23c739b762f625cf449'

In [34]: psingle.md5_computed_file_hash
Out[34]: '079fdad3df93dbceac18c56a69852e29'

In [35]: psingle
Out[35]: PotcarSingle(symbol='Pt', functional='PBE', TITEL='PAW_PBE Pt 05Jan2001', VRHFIN='Pt: s1d9', n_valence_elec=10)

See here for the corresponding source code implementation. See here and here for the related discussion.

[tschaume]

Thanks for following up!