Source code for mp_api.client.routes.materials.grain_boundary

from __future__ import annotations

import warnings
from collections import defaultdict

from emmet.core.grain_boundary import GBTypeEnum, GrainBoundaryDoc

from mp_api.client.core import BaseRester
from mp_api.client.core.utils import validate_ids


[docs] class GrainBoundaryRester(BaseRester[GrainBoundaryDoc]): suffix = "materials/grain_boundary" document_model = GrainBoundaryDoc # type: ignore primary_key = "task_id"
[docs] def search_grain_boundary_docs(self, *args, **kwargs): # pragma: no cover """Deprecated.""" warnings.warn( "MPRester.grain_boundary.search_grain_boundary_docs is deprecated. " "Please use MPRester.grain_boundary.search instead.", DeprecationWarning, stacklevel=2, ) return self.search(*args, **kwargs)
[docs] def search( self, chemsys: str | None = None, gb_plane: list[str] | None = None, gb_energy: tuple[float, float] | None = None, material_ids: list[str] | None = None, pretty_formula: str | None = None, rotation_axis: list[str] | None = None, rotation_angle: tuple[float, float] | None = None, separation_energy: tuple[float, float] | None = None, sigma: int | None = None, type: GBTypeEnum | None = None, num_chunks: int | None = None, chunk_size: int = 1000, all_fields: bool = True, fields: list[str] | None = None, ): """Query grain boundary docs using a variety of search criteria. Arguments: chemsys (str): Dash-delimited string of elements in the material. gb_plane(List[str]): The Miller index of grain boundary plane. e.g., [1, 1, 1] gb_energy (Tuple[float,float]): Minimum and maximum grain boundary energy in J/m³ to consider. material_ids (List[str]): List of Materials Project IDs to query with. pretty_formula (str): Formula of the material. rotation_angle (Tuple[float,float]): Minimum and maximum rotation angle in degrees to consider. rotation_axis(List[str]): The Miller index of rotation axis. e.g., [1, 0, 0], [1, 1, 0], and [1, 1, 1] sigma (int): Sigma value of grain boundary. separation_energy (Tuple[float,float]): Minimum and maximum work of separation energy in J/m³ to consider. sigma (int): Sigma value of the boundary. type (GBTypeEnum): Grain boundary type. num_chunks (int): Maximum number of chunks of data to yield. None will yield all possible. chunk_size (int): Number of data entries per chunk. all_fields (bool): Whether to return all fields in the document. Defaults to True. fields (List[str]): List of fields in GrainBoundaryDoc to return data for. Default is material_id and last_updated if all_fields is False. Returns: ([GrainBoundaryDoc]) List of grain boundary documents """ query_params = defaultdict(dict) # type: dict if material_ids: query_params.update({"task_ids": ",".join(validate_ids(material_ids))}) if gb_plane: query_params.update({"gb_plane": ",".join([str(n) for n in gb_plane])}) if gb_energy: query_params.update( {"gb_energy_min": gb_energy[0], "gb_energy_max": gb_energy[1]} ) if separation_energy: query_params.update( {"w_sep_min": separation_energy[0], "w_sep_max": separation_energy[1]} ) if rotation_angle: query_params.update( { "rotation_angle_min": rotation_angle[0], "rotation_angle_max": rotation_angle[1], } ) if rotation_axis: query_params.update( {"rotation_axis": ",".join([str(n) for n in rotation_axis])} ) if sigma: query_params.update({"sigma": sigma}) if type: query_params.update({"type": type.value}) if chemsys: query_params.update({"chemsys": chemsys}) if pretty_formula: query_params.update({"pretty_formula": pretty_formula}) query_params = { entry: query_params[entry] for entry in query_params if query_params[entry] is not None } return super()._search( num_chunks=num_chunks, chunk_size=chunk_size, all_fields=all_fields, fields=fields, **query_params, )