Source code for pylbo.visualisation.modes.mode_data

from __future__ import annotations

import difflib
from typing import Union

import numpy as np
from pylbo.data_containers import LegolasDataSet
from pylbo.exceptions import BackgroundNotPresent
from pylbo.utilities.logger import pylboLogger
from pylbo.visualisation.utils import ef_name_to_latex, validate_ef_name


[docs] class ModeVisualisationData: """ Class that contains the data used for eigenmode visualisations. Parameters ---------- ds : ~pylbo.data_containers.LegolasDataSet The dataset containing the eigenfunctions and modes to visualise. omega : list[complex] The (approximate) eigenvalue(s) of the mode(s) to visualise. ef_name : str The name of the eigenfunction to visualise. use_real_part : bool Whether to use the real part of the eigenmode solution. complex_factor : complex A complex factor to multiply the eigenmode solution with. add_background : bool Whether to add the equilibrium background to the eigenmode solution. Attributes ---------- ds : ~pylbo.data_containers.LegolasDataSet The dataset containing the eigenfunctions and modes to visualise. omega : list[complex] The (approximate) eigenvalue(s) of the mode(s) to visualise. eigenfunction : list[np.ndarray] The eigenfunction of the mode(s) to visualise. use_real_part : bool Whether to use the real part of the eigenmode solution. complex_factor : complex The complex factor to multiply the eigenmode solution with. add_background : bool Whether to add the equilibrium background to the eigenmode solution. """ def __init__( self, ds: LegolasDataSet, omega: list[complex], ef_name: str = None, use_real_part: bool = True, complex_factor: complex = None, add_background: bool = False, ) -> None:
[docs] self.ds = ds
[docs] self.use_real_part = use_real_part
[docs] self.complex_factor = self._validate_complex_factor(complex_factor)
if add_background and not ds.has_background: raise BackgroundNotPresent(ds.datfile, "add background to solution")
[docs] self.add_background = add_background
[docs] self._print_bg_info = True
[docs] self._ef_name = None if ef_name is None else validate_ef_name(ds, ef_name)
[docs] self._ef_name_latex = None if ef_name is None else self.get_ef_name_latex()
[docs] self._all_efs = ds.get_eigenfunctions(ev_guesses=omega)
[docs] self.omega = [all_efs.get("eigenvalue") for all_efs in self._all_efs]
[docs] self.eigenfunction = [all_efs.get(self._ef_name) for all_efs in self._all_efs]
@property
[docs] def k2(self) -> float: """The k2 wave number of the eigenmode solution.""" return self.ds.parameters["k2"]
@property
[docs] def k3(self) -> float: """The k3 wave number of the eigenmode solution.""" return self.ds.parameters["k3"]
@property
[docs] def part_name(self) -> str: """ Returns the name of the part of the eigenmode solution to use, i.e. 'real' or 'imag'. """ return "real" if self.use_real_part else "imag"
[docs] def get_ef_name_latex(self) -> str: """Returns the latex representation of the eigenfunction name.""" return ef_name_to_latex( self._ef_name, geometry=self.ds.geometry, real_part=self.use_real_part )
[docs] def _validate_complex_factor(self, complex_factor: complex) -> complex: """ Validates the complex factor. Parameters ---------- complex_factor : complex The complex factor to validate. Returns ------- complex The complex factor if it is valid, otherwise 1. """ return complex_factor if complex_factor is not None else 1
[docs] def get_mode_solution( self, ef: np.ndarray, omega: complex, u2: Union[float, np.ndarray], u3: Union[float, np.ndarray], t: Union[float, np.ndarray], ) -> np.ndarray: """ Calculates the full eigenmode solution for given coordinates and time. If a complex factor was given, the eigenmode solution is multiplied with the complex factor. If :attr:`use_real_part` is True the real part of the eigenmode solution is returned, otherwise the complex part. Parameters ---------- ef : np.ndarray The eigenfunction to use. omega : complex The eigenvalue to use. u2 : Union[float, np.ndarray] The y coordinate(s) of the eigenmode solution. u3 : Union[float, np.ndarray] The z coordinate(s) of the eigenmode solution. t : Union[float, np.ndarray] The time(s) of the eigenmode solution. Returns ------- np.ndarray The real or imaginary part of the eigenmode solution for the given set of coordinate(s) and time(s). """ solution = ( self.complex_factor * ef * np.exp(1j * self.k2 * u2 + 1j * self.k3 * u3 - 1j * omega * t) ) return getattr(solution, self.part_name)
[docs] def get_background(self, shape: tuple[int, ...], name=None) -> np.ndarray: """ Returns the background of the eigenmode solution. Parameters ---------- shape : tuple[int, ...] The shape of the eigenmode solution. name : str The name of the background to use. If None, the background name will be inferred from the eigenfunction name. Returns ------- np.ndarray The background of the eigenmode solution, sampled on the eigenfunction grid and broadcasted to the same shape as the eigenmode solution. """ if name is None: name = self._get_background_name() bg = self.ds.equilibria[name] bg_sampled = self._sample_background_on_ef_grid(bg) if self._print_bg_info: pylboLogger.info(f"background {name} broadcasted to shape {shape}") return np.broadcast_to(bg_sampled, shape=reversed(shape)).transpose()
[docs] def _sample_background_on_ef_grid(self, bg: np.ndarray) -> np.ndarray: """ Samples the background array on the eigenfunction grid. Parameters ---------- bg : np.ndarray The background array with Gaussian grid spacing Returns ------- np.ndarray The background array with eigenfunction grid spacing """ if self._print_bg_info: pylboLogger.info( f"sampling background [{len(bg)}] on eigenfunction grid " f"[{len(self.ds.ef_grid)}]" ) return np.interp(self.ds.ef_grid, self.ds.grid_gauss, bg)
[docs] def _get_background_name(self) -> str: """ Returns the name of the background. Returns ------- str The closest match between the eigenfunction name and the equilibrium name. Raises ------ ValueError If the eigenfunction name is a magnetic vector potential component. """ if self._ef_name in ("a1", "a2", "a3"): raise ValueError( "Unable to add a background to the magnetic vector potential." ) (name,) = difflib.get_close_matches(self._ef_name, self.ds.eq_names, 1) if self._print_bg_info: pylboLogger.info( f"adding background for '{self._ef_name}', closest match is '{name}'" ) return name