Open-Catalyst-Dataset

This repository hosts the adsorbate-catalyst input generation workflow used in the Open Catalyst Project.

Setup

The easiest way to install prerequisites is via conda. After installing conda, run the following commands:

• Create a new environment: conda create -n ocp python=3.9
• Activate the newly created environment: conda activate ocp
• Install specific versions of Pymatgen and ASE: pip install pymatgen==2023.5.10 ase==3.22.1
• Clone this repo and install with: pip install -e .

Workflow

The codebase supports the following workflow to generate adsorbate-catalyst input configurations.

1. Initialize a bulk:
   • By providing an atoms object, or
   • By bulk_id (e.g. mp-30), or
   • By its index in the database, or
   • By selecting randomly.
2. Initialize an adsorbate:
   • By providing an atoms object, or
   • By its SMILES string (e.g. *H), or
   • By its index in the database, or
   • By selecting randomly.
3. Enumerate slabs from the Bulk class.
   This internally uses pymatgen.core.surface.SlabGenerator and supports the following:
   • All slabs up to a specified miller index, or
   • A random slab among those enumerated by the previous method, or
   • A specific miller index.
4. Place the adsorbate on the slab.
   This broadly has two steps -- identifying a binding site on the surface of the slab, and orienting the adsorbate before placing it at that site. We use custom code inspired by pymatgen to do this. There are 3 modes: heuristic, random, and random_site_heuristic_placement.
   • Identifying a binding site: First, a Delaunay meshgrid is constructed with surface atoms as nodes. For heuristic, the sites considered are on the node (atop), between 2 nodes (bridge) and in the center of the triangle (hollow). For random and random_site_heuristic_placement, positions of the sites are uniformly randomly sampled along the Delaunay triangles.
   • Adsorbate orientation: For heuristic and random_site_heuristic_placement, the adsorbate is uniformly randomly rotated around the z direction, and provided a slight wobble around x and y, which amounts to randomized tilt within a certain cone around the north pole. For random, the adsorbate is uniformly randomly rotated about its center of mass along all directions.
   • Binding atom: The adsorbate database includes information about which atoms are expected to bind. For heuristic and random_site_heuristic_placement, the binding atom of the adsorbate is placed at the site, whereas for random the center of mass of the adsorbate is placed at the site.

Usage

Here is a simple example using the ocdata workflow to place CO on Cu (1,1,1):

bulk_src_id = "mp-30"
adsorbate_smiles = "*CO"

bulk = Bulk(bulk_src_id_from_db=bulk_src_id, bulk_db_path="your-path-here.pkl")
adsorbate = Adsorbate(adsorbate_smiles_from_db=adsorbate_smiles, adsorbate_db_path="your-path-here.pkl")
slabs = Slab.from_bulk_get_specific_millers(bulk=bulk, specific_millers=(1,1,1))

# Perform heuristic placements
heuristic_adslabs = AdsorbateSlabConfig(slabs[0], adsorbate, mode="heuristic")

# Perform random site, heuristic placements
random_adslabs = AdsorbateSlabConfig(slabs[0], adsorbate, mode="random_site_heuristic_placement", num_sites=100)

If you want to use a bulk and/or adsorbate that is not in the database here, you may supply your own ase.Atoms object:

bulk = Bulk(bulk_atoms=your_adsorbate_atoms)
adsorbate = Adsorbate(adsorbate_atoms=your_adsorbate_atoms)
slabs = Slab.from_bulk_get_all_slabs(bulk)

# Perform fully random placements
random_adslabs = AdsorbateSlabConfig(slabs[0], adsorbate, mode="random", num_sites=100)

If you would like to randomly choose a bulk, adsorbate, and slab:

bulk = Bulk()
adsorbate = Adsorbate()
slab = Slab.from_bulk_get_random_slab(bulk)

# Perform fully random placements
random_adslabs = AdsorbateSlabConfig(slab, adsorbate, mode="random", num_sites=100)

StructureGenerator API

We also provide a StructureGenerator helper class that wraps the core functionality described above for creating bulk/slab/adsorbate objects, and writing vasp input files and metadata for multiple placements of the adsorbate on the slab. There are a number of options to configure input generation to suit different usecases. We list a few examples here.

Command Line Args

Input files:

• --bulk_db (required): path to the bulk database file
• --adsorbate_db: path to the adsorbate database file - required if adsorbate placement is to be performed.
• --precomputed_slabs_dir: path to the precomputed slab directory, which saves cost/time if the slabs for each bulk have already been enumerated.

Bulk / Slab / Adsorbate specification

Option 1: provide indices. All three must be provided to generate adsorbate-slab configurations, otherwise only slab enumeration will be performed.

• --adsorbate_index: index of the desired adsorbate in the database file.
• --bulk_index: index of the desired bulk
• --surface_index: index of the desired surface

Option 2: provide a set of indices (one of the following)

• --indices_file: a file containing strings with the following format f"{adsorbate_idx}_{bulk_idx}_{surface_idx}". This will enumerate slabs as well as adsorbate-slab configurations.
• --bulk_indices_file: a file containing bulk indices. This will only do slab enumeration.

Slab enumeration

• --max_miller: the max miller index of slabs to be generated (i.e. 1, 2, or 3)

Adsorbate Placement

• --seed: random seed for sampling/random sites generation.
• --heuristic_placements: to be provided if heuristic placements are desired.
• --random_placements: to be provided if random sites are desired. You may do both heuristic and random placements in the same run.
• --full_random_rotations: to be provided in addition to --random_placements if fully random placements are desired, as opposed to small wobbles around x/y axis.
• --random_sites: the number of sites per slab, which should be provided if --random_placements are used.
• --num_augmentations: the number of random adsorbate configurations per site (defaults to 1).

Multiprocessing, when given a file of indices

• --chunks: for multi-node processing, number of chunks to split inputs across.
• --chunk_index: for multi-node processing, index of chunk to process.
• --workers: number of workers for multiprocessing within one job

Outputs

• --output_dir: directory to save outputs
• --no_vasp: for VASP input files, only write POSCAR and do not write INCAR, KPOINTS, or POTCAR
• --verbose: if detailed info should be logged

Usage

python structure_generator.py \
  --bulk_db databases/pkls/bulks.pkl \
  --adsorbate_db databases/pkls/adsorbates.pkl  \
  --output_dir outputs/ \
  --adsorbate_index 0 \
  --bulk_index 0 \
  --surface_index 0 \
  --heuristic_placements

python structure_generator.py \
  --bulk_db databases/pkls/bulks.pkl \
  --adsorbate_db databases/pkls/adsorbates.pkl  \
  --indices_file your_index_file.txt \
  --seed 0 \
  --random_placements \
  --random_sites 100

Databases for bulks and adsorbates

Bulks

A database of bulk materials taken from existing databases (i.e. Materials Project) and relaxed with consistent RPBE settings may be found in ocdata/databases/pkls/bulks.pkl. To preview what bulks are available, view the corresponding mapping between indices and bulks (bulk id and composition): https://dl.fbaipublicfiles.com/opencatalystproject/data/input_generation/mapping_bulks_2021sep20.txt

Adsorbates

A database of adsorbates may be found in ocdata/databases/pkls/adsorbates.pkl. Alternatively, it may be downloaded using the following link: The latest version is https://dl.fbaipublicfiles.com/opencatalystproject/data/input_generation/adsorbate_db_2021apr28.pkl (MD5 checksum: 975e00a62c7b634b245102e42167b3fb). To preview what adsorbates are available, view the corresponding mapping between indices and adsorbates (SMILES): https://dl.fbaipublicfiles.com/opencatalystproject/data/input_generation/mapping_adsorbates_2020may12.txt

Previous snapshots of the codebase

• OC20 was generated with an older version of the bulks and this repository. If you would like to exactly reproduce that work, see README_legacy_OC20.md.
• OC22 was generated from the OC22_dataset branch of this repository.

License

ocdata is released under the MIT license.

Citation

If you use this codebase in your work, please consider citing:

@article{ocp_dataset,
    author = {Chanussot*, Lowik and Das*, Abhishek and Goyal*, Siddharth and Lavril*, Thibaut and Shuaibi*, Muhammed and Riviere, Morgane and Tran, Kevin and Heras-Domingo, Javier and Ho, Caleb and Hu, Weihua and Palizhati, Aini and Sriram, Anuroop and Wood, Brandon and Yoon, Junwoong and Parikh, Devi and Zitnick, C. Lawrence and Ulissi, Zachary},
    title = {Open Catalyst 2020 (OC20) Dataset and Community Challenges},
    journal = {ACS Catalysis},
    year = {2021},
    doi = {10.1021/acscatal.0c04525},
}

The Open Catalyst 2020 (OC20) and Open Catalyst 2022 (OC22) datasets are licensed under a Creative Commons Attribution 4.0 License.