A crystal structure optimization tool built on top of ASE

Installation

After cloning this project, install lam-crystal-philately with common dependencies (including requirements for workflows) by

pip install .

To install additional dependencies for DP

pip install ".[dp]"

or mace

pip install ".[mace]"

Download the latest DP model for structure optimization by

wget https://bohrium-api.dp.tech/ds-dl/lam-crystal-model-01oe-v4.zip
unzip lam-crystal-model-01oe-v4.zip

Structure Optimization

Python API

from pathlib import Path
from lam_optimize.main import relax_run
from lam_optimize.relaxer import Relaxer

cif_folder_path = Path("/cifs")
relaxer = Relaxer("mace") # using default mace model
# relaxer = Relaxer(Path("dp.pth")) # using DP model
res_df = relax_run(
    cif_folder_path,
    relaxer
)

You should see something similar to this:

The resulting dataframe should contain the following columns:

To get the optimized structure (if converged), do the following:

from pymatgen.core import Structure

Structure.from_dict(df['final_structure'][0])

Commandline tool

To optimize structures using DP model

lam-opt relax -i examples/data -m <path-to-DP-model>

or using mace

lam-opt relax -i examples/data -t mace

To submit a workflow for optimizing structures on parallel

lam-opt submit examples/wf.json -i part0 part1 -m <path-to-DP-model>

where the arguments after -i should be a list of directories containing cifs.

Single Point Evaluation

from lam_optimize.main import single_point

single_point(Path(fpth), relaxer)

This returns the potential energy and forces for a given .cif structure.

Query crystal structures from OpenLAM Database

Set environmental variable BOHRIUM_ACCESS_KEY which is generated from https://bohrium.dp.tech/settings/user

export BOHRIUM_ACCESS_KEY=xxx

Query crystal structures from OpenLAM Database using Python API (The method query_by_page is deprecated! Use query_by_offset instead.)

from lam_optimize import CrystalStructure
data = CrystalStructure.query_by_offset()

The method query_by_offset accept following arguments as query conditions

formula: Optional[str] = None
min_energy: Optional[float] = None
max_energy: Optional[float] = None
min_submission_time: Optional[datetime.datetime] = None
max_submission_time: Optional[datetime.datetime] = None
offset: int = 0
limit: int = 10

The structure of the returned data is like

{'nextStartId': 18, 'items': [<lam_optimize.db.CrystalStructure object at 0x7fbd6832e520>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d04aaf0>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d11c610>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d11cd60>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d21a130>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d21a4c0>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d21a850>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d21abe0>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d21af70>, <lam_optimize.db.CrystalStructure object at 0x7fbd6d21d340>]}

Except for nextStartId (used as offset in the next query), items is a list of CrystalStructure objects

class CrystalStructure:
    formula: str
    structure: pymatgen.core.Structure
    energy: float
    submission_time: datetime.datetime

The method query merging paged results is also provided

structures = CrystalStructure.query(formula="Sr2YSbO6")

which returns a list of CrystalStructure objects.

NOTE: Calling non-paging method without query condition will be extremely slow.

deepmodeling / openlam