Python client library for the Open Catalyst API
Project description
fairchem-demo-ocpapi
Python library for programmatic use of the Open Catalyst Demo. Users unfamiliar with the Open Catalyst Demo are encouraged to read more about it before continuing.
Installation
Ensure you have Python 3.9.1 or newer, and install fairchem-demo-ocpapi
using:
pip install fairchem-demo-ocpapi
Quickstart
The following examples are used to search for *OH binding sites on Pt surfaces. They use the find_adsorbate_binding_sites
function, which is a high-level workflow on top of other methods included in this library. Once familiar with this routine, users are encouraged to learn about lower-level methods and features that support more advanced use cases.
Note about async methods
This package relies heavily on asyncio. The examples throughout this document can be copied to a python repl launched with:
$ python -m asyncio
Alternatively, an async function can be run in a script by wrapping it with asyncio.run():
import asyncio
from fairchem.demo.ocpapi import find_adsorbate_binding_sites
asyncio.run(find_adsorbate_binding_sites(...))
Search over all surfaces
from fairchem.demo.ocpapi import find_adsorbate_binding_sites
results = await find_adsorbate_binding_sites(
adsorbate="*OH",
bulk="mp-126",
)
Users will be prompted to select one or more surfaces that should be relaxed.
Input to this function includes:
- The name of the adsorbate to place
- A unique ID of the bulk structure from which surfaces will be generated
This function will perform the following steps:
- Enumerate surfaces of the bulk material
- On each surface, enumerate initial guesses for adorbate binding sites
- Run local force-based relaxations of each adsorbate placement
In addition, this handles:
- Retrying failed calls to the Open Catalyst Demo API
- Retrying submission of relaxations when they are rate limited
This should take 2-10 minutes to finish while tens to hundreds (depending on the number of surfaces that are selected) of individual adsorbate placements are relaxed on unique surfaces of Pt. Each of the objects in the returned list includes (among other details):
- Information about the surface being searched, including its structure and Miller indices
- The initial positions of the adsorbate before relaxation
- The final structure after relaxation
- The predicted energy of the final structure
- The predicted force on each atom in the final structure
Supported bulks and adsorbates
A finite set of bulk materials and adsorbates can be referenced by ID throughout the OCP API. The lists of supported values can be viewed in two ways.
- Visit the UI at https://open-catalyst.metademolab.com/demo and explore the lists in Step 1 and Step 3.
- Use the low-level client that ships with this library:
from fairchem.demo.ocpapi import Client
client = Client()
bulks = await client.get_bulks()
print({b.src_id: b.formula for b in bulks.bulks_supported})
adsorbates = await client.get_adsorbates()
print(adsorbates.adsorbates_supported)
Persisting results
Results should be saved whenever possible in order to avoid expensive recomputation.
Assuming results
was generated with the find_adsorbate_binding_sites
method used above, it is an AdsorbateBindingSites
object. This can be saved to file with:
with open("results.json", "w") as f:
f.write(results.to_json())
Similarly, results can be read back from file to an AdsorbateBindingSites
object with:
from fairchem.demo.ocpapi import AdsorbateBindingSites
with open("results.json", "r") as f:
results = AdsorbateBindingSites.from_json(f.read())
Viewing results in the web UI
Relaxation results can be viewed in a web UI. For example, https://open-catalyst.metademolab.com/results/7eaa0d63-83aa-473f-ac84-423ffd0c67f5 shows the results of relaxing *OH on a Pt (1,1,1) surface; the uuid, "7eaa0d63-83aa-473f-ac84-423ffd0c67f5", is referred to as the system_id
.
Extending the examples above, the URLs to visualize the results of relaxations on each Pt surface can be obtained with:
urls = [slab.ui_url for slab in results.slabs]
Advanced usage
Changing the model type
The API currently supports two models:
equiformer_v2_31M_s2ef_all_md
(default): https://arxiv.org/abs/2306.12059gemnet_oc_base_s2ef_all_md
: https://arxiv.org/abs/2204.02782
A specific model type can be requested with:
from fairchem.demo.ocpapi import find_adsorbate_binding_sites
results = await find_adsorbate_binding_sites(
adsorbate="*OH",
bulk="mp-126",
model="gemnet_oc_base_s2ef_all_md",
)
Skip relaxation approval prompts
Calls to find_adsorbate_binding_sites()
will, by default, show the user all pending relaxations and ask for approval before they are submitted. In order to run the relaxations automatically without manual approval, adslab_filter
can be set to a function that automatically approves any or all adsorbate/slab (adslab) configurations.
Run relaxations for all slabs that are generated:
from fairchem.demo.ocpapi import find_adsorbate_binding_sites, keep_all_slabs
results = await find_adsorbate_binding_sites(
adsorbate="*OH",
bulk="mp-126",
adslab_filter=keep_all_slabs(),
)
Run relaxations only for slabs with Miller Indices in the input set:
from fairchem.demo.ocpapi import (
find_adsorbate_binding_sites,
keep_slabs_with_miller_indices,
)
results = await find_adsorbate_binding_sites(
adsorbate="*OH",
bulk="mp-126",
adslab_filter=keep_slabs_with_miller_indices([(1, 0, 0), (1, 1, 1)]),
)
Converting to ase.Atoms objects
Important! The to_ase_atoms()
method described below will fail with an import error if ase is not installed.
Two classes have support for generating ase.Atoms objects:
ocpapi.Atoms.to_ase_atoms()
: Adds unit cell, atomic positions, and other structural information to the returnedase.Atoms
object.ocpapi.AdsorbateSlabRelaxationResult.to_ase_atoms()
: Adds the same structure information to thease.Atoms
object. Also adds the predicted forces and energy of the relaxed structure, which can be accessed with thease.Atoms.get_potential_energy()
andase.Atoms.get_forces()
methods.
For example, the following would generate an ase.Atoms
object for the first relaxed adsorbate configuration on the first slab generated for *OH binding on Pt:
from fairchem.demo.ocpapi import find_adsorbate_binding_sites
results = await find_adsorbate_binding_sites(
adsorbate="*OH",
bulk="mp-126",
)
ase_atoms = results.slabs[0].configs[0].to_ase_atoms()
Converting to other structure formats
From an ase.Atoms
object (see previous section), is is possible to write to other structure formats. Extending the example above, the ase_atoms
object could be written to a VASP POSCAR file with:
from ase.io import write
write("POSCAR", ase_atoms, "vasp")
License
fairchem-demo-ocpapi
is released under the MIT License.
Citing fairchem-demo-ocpapi
If you use fairchem-demo-ocpapi
in your research , please consider citing the AdsorbML paper (in addition to the relevant datasets / models used):
@article{lan2023adsorbml,
title={{AdsorbML}: a leap in efficiency for adsorption energy calculations using generalizable machine learning potentials},
author={Lan*, Janice and Palizhati*, Aini and Shuaibi*, Muhammed and Wood*, Brandon M and Wander, Brook and Das, Abhishek and Uyttendaele, Matt and Zitnick, C Lawrence and Ulissi, Zachary W},
journal={npj Computational Materials},
year={2023},
}
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for fairchem_demo_ocpapi-0.1.0.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2f3a42c65529d97ea9e2f19d5ea90b1c9df72fc620c7c8c84cb4b657dad133b3 |
|
MD5 | 6cf6a9001d85a3cd5f31945130137dda |
|
BLAKE2b-256 | 7ac00c696b950a0eb47e17b17ab4467f569ba8acc50fc4907f45ed9752ea7814 |
Hashes for fairchem_demo_ocpapi-0.1.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 6b69922c5bfe1d051bc0e3b9635b498e3e8eb1e93d00d83831193eadcd17934c |
|
MD5 | 6a3d8d640029d664d4a7c41b73c8b611 |
|
BLAKE2b-256 | 16b8f214fc357ff33f875061cb9aff0be45fcd796463645930037e74f76a6c12 |