# Refactor of DOS-like data objects
# towards replacing ase.dft.dos and ase.dft.pdos
from abc import ABCMeta, abstractmethod
import warnings
from typing import Any, Dict, Sequence, Tuple, TypeVar, Union
import numpy as np
from ase.utils.plotting import SimplePlottingAxes
# This import is for the benefit of type-checking / mypy
if False:
import matplotlib.axes
# For now we will be strict about Info and say it has to be str->str. Perhaps
# later we will allow other types that have reliable comparison operations.
Info = Dict[str, str]
[docs]class DOSData(metaclass=ABCMeta):
"""Abstract base class for a single series of DOS-like data
Only the 'info' is a mutable attribute; DOS data is set at init"""
def __init__(self,
info: Info = None) -> None:
if info is None:
self.info = {}
elif isinstance(info, dict):
self.info = info
else:
raise TypeError("Info must be a dict or None")
[docs] @abstractmethod
def get_energies(self) -> Sequence[float]:
"""Get energy data stored in this object"""
[docs] @abstractmethod
def get_weights(self) -> Sequence[float]:
"""Get DOS weights stored in this object"""
[docs] @abstractmethod
def copy(self) -> 'DOSData':
"""Returns a copy in which info dict can be safely mutated"""
def _sample(self,
energies: Sequence[float],
width: float = 0.1,
smearing: str = 'Gauss') -> np.ndarray:
"""Sample the DOS data at chosen points, with broadening
Note that no correction is made here for the sampling bin width; total
intensity will vary with sampling density.
Args:
energies: energy values for sampling
width: Width of broadening kernel
smearing: selection of broadening kernel (only "Gauss" is currently
supported)
Returns:
Weights sampled from a broadened DOS at values corresponding to x
"""
self._check_positive_width(width)
weights_grid = np.zeros(len(energies), float)
weights = self.get_weights()
energies = np.asarray(energies, float)
for i, raw_energy in enumerate(self.get_energies()):
delta = self._delta(energies, raw_energy, width, smearing=smearing)
weights_grid += weights[i] * delta
return weights_grid
def _almost_equals(self, other: Any) -> bool:
"""Compare with another DOSData for testing purposes"""
if not isinstance(other, type(self)):
return False
if self.info != other.info:
return False
if not np.allclose(self.get_weights(), other.get_weights()):
return False
return np.allclose(self.get_energies(), other.get_energies())
@staticmethod
def _delta(x: np.ndarray,
x0: float,
width: float,
smearing: str = 'Gauss') -> np.ndarray:
"""Return a delta-function centered at 'x0'.
This function is used with numpy broadcasting; if x is a row and x0 is
a column vector, the returned data will be a 2D array with each row
corresponding to a different delta center.
"""
if smearing.lower() == 'gauss':
x1 = -0.5 * ((x - x0) / width)**2
return np.exp(x1) / (np.sqrt(2 * np.pi) * width)
else:
msg = 'Requested smearing type not recognized. Got {}'.format(
smearing)
raise ValueError(msg)
@staticmethod
def _check_positive_width(width):
if width <= 0.0:
msg = 'Cannot add 0 or negative width smearing'
raise ValueError(msg)
[docs] def sample_grid(self,
npts: int,
xmin: float = None,
xmax: float = None,
padding: float = 3,
width: float = 0.1,
smearing: str = 'Gauss',
) -> 'GridDOSData':
"""Sample the DOS data on an evenly-spaced energy grid
Args:
npts: Number of sampled points
xmin: Minimum sampled x value; if unspecified, a default is chosen
xmax: Maximum sampled x value; if unspecified, a default is chosen
padding: If xmin/xmax is unspecified, default value will be padded
by padding * width to avoid cutting off peaks.
width: Width of broadening kernel
smearing: selection of broadening kernel (only 'Gauss' is
implemented)
Returns:
(energy values, sampled DOS)
"""
if xmin is None:
xmin = min(self.get_energies()) - (padding * width)
if xmax is None:
xmax = max(self.get_energies()) + (padding * width)
energies_grid = np.linspace(xmin, xmax, npts)
weights_grid = self._sample(energies_grid, width=width,
smearing=smearing)
return GridDOSData(energies_grid, weights_grid, info=self.info.copy())
[docs] def plot(self,
npts: int = 1000,
xmin: float = None,
xmax: float = None,
width: float = 0.1,
smearing: str = 'Gauss',
ax: 'matplotlib.axes.Axes' = None,
show: bool = False,
filename: str = None,
mplargs: dict = None) -> 'matplotlib.axes.Axes':
"""Simple 1-D plot of DOS data, resampled onto a grid
If the special key 'label' is present in self.info, this will be set
as the label for the plotted line (unless overruled in mplargs). The
label is only seen if a legend is added to the plot (i.e. by calling
``ax.legend()``).
Args:
npts, xmin, xmax: output data range, as passed to self.sample_grid
width: Width of broadening kernel for self.sample_grid()
smearing: selection of broadening kernel for self.sample_grid()
ax: existing Matplotlib axes object. If not provided, a new figure
with one set of axes will be created using Pyplot
show: show the figure on-screen
filename: if a path is given, save the figure to this file
mplargs: additional arguments to pass to matplotlib plot command
(e.g. {'linewidth': 2} for a thicker line).
Returns:
Plotting axes. If "ax" was set, this is the same object.
"""
if mplargs is None:
mplargs = {}
if 'label' not in mplargs:
mplargs.update({'label': self.label_from_info(self.info)})
return self.sample_grid(npts, xmin=xmin, xmax=xmax,
width=width,
smearing=smearing
).plot(ax=ax, xmin=xmin, xmax=xmax,
show=show, filename=filename,
mplargs=mplargs)
[docs] @staticmethod
def label_from_info(info: Dict[str, str]):
"""Generate an automatic legend label from info dict"""
if 'label' in info:
return info['label']
else:
return '; '.join(map(lambda x: '{}: {}'.format(x[0], x[1]),
info.items()))
[docs]class GeneralDOSData(DOSData):
"""Base class for a single series of DOS-like data
Only the 'info' is a mutable attribute; DOS data is set at init
This is the base class for DOSData objects that accept/set seperate
"energies" and "weights" sequences of equal length at init.
"""
def __init__(self,
energies: Union[Sequence[float], np.ndarray],
weights: Union[Sequence[float], np.ndarray],
info: Info = None) -> None:
super().__init__(info=info)
n_entries = len(energies)
if len(weights) != n_entries:
raise ValueError("Energies and weights must be the same length")
# Internally store the data as a np array with two rows; energy, weight
self._data = np.empty((2, n_entries), dtype=float, order='C')
self._data[0, :] = energies
self._data[1, :] = weights
[docs] def get_energies(self) -> np.ndarray:
return self._data[0, :].copy()
[docs] def get_weights(self) -> np.ndarray:
return self._data[1, :].copy()
D = TypeVar('D', bound='GeneralDOSData')
[docs] def copy(self: D) -> D: # noqa F821
return type(self)(self.get_energies(), self.get_weights(),
info=self.info.copy())
[docs]class RawDOSData(GeneralDOSData):
"""A collection of weighted delta functions which sum to form a DOS
This is an appropriate data container for density-of-states (DOS) or
spectral data where the energy data values not form a known regular
grid. The data may be plotted or resampled for further analysis using the
sample_grid() and plot() methods. Multiple weights at the same
energy value will *only* be combined in output data, and data stored in
RawDOSData is never resampled. A plot_deltas() function is also provided
which plots the raw data.
Metadata may be stored in the info dict, in which keys and values must be
strings. This data is used for selecting and combining multiple DOSData
objects in a DOSCollection object.
When RawDOSData objects are combined with the addition operator::
big_dos = raw_dos_1 + raw_dos_2
the energy and weights data is *concatenated* (i.e. combined without
sorting or replacement) and the new info dictionary consists of the
*intersection* of the inputs: only key-value pairs that were common to both
of the input objects will be retained in the new combined object. For
example::
(RawDOSData([x1], [y1], info={'symbol': 'O', 'index': '1'})
+ RawDOSData([x2], [y2], info={'symbol': 'O', 'index': '2'}))
will yield the equivalent of::
RawDOSData([x1, x2], [y1, y2], info={'symbol': 'O'})
"""
def __add__(self, other: 'RawDOSData') -> 'RawDOSData':
if not isinstance(other, RawDOSData):
raise TypeError("RawDOSData can only be combined with other "
"RawDOSData objects")
# Take intersection of metadata (i.e. only common entries are retained)
new_info = dict(set(self.info.items()) & set(other.info.items()))
# Concatenate the energy/weight data
new_data = np.concatenate((self._data, other._data), axis=1)
new_object = RawDOSData([], [], info=new_info)
new_object._data = new_data
return new_object
[docs] def plot_deltas(self,
ax: 'matplotlib.axes.Axes' = None,
show: bool = False,
filename: str = None,
mplargs: dict = None) -> 'matplotlib.axes.Axes':
"""Simple plot of sparse DOS data as a set of delta functions
Items at the same x-value can overlap and will not be summed together
Args:
ax: existing Matplotlib axes object. If not provided, a new figure
with one set of axes will be created using Pyplot
show: show the figure on-screen
filename: if a path is given, save the figure to this file
mplargs: additional arguments to pass to matplotlib Axes.vlines
command (e.g. {'linewidth': 2} for a thicker line).
Returns:
Plotting axes. If "ax" was set, this is the same object.
"""
if mplargs is None:
mplargs = {}
with SimplePlottingAxes(ax=ax, show=show, filename=filename) as ax:
ax.vlines(self.get_energies(), 0, self.get_weights(), **mplargs)
return ax
[docs]class GridDOSData(GeneralDOSData):
"""A collection of regularly-sampled data which represents a DOS
This is an appropriate data container for density-of-states (DOS) or
spectral data where the intensity values form a regular grid. This
is generally the result of sampling or integrating into discrete
bins, rather than a collection of unique states. The data may be
plotted or resampled for further analysis using the sample_grid()
and plot() methods.
Metadata may be stored in the info dict, in which keys and values must be
strings. This data is used for selecting and combining multiple DOSData
objects in a DOSCollection object.
When RawDOSData objects are combined with the addition operator::
big_dos = raw_dos_1 + raw_dos_2
the weights data is *summed* (requiring a consistent energy grid) and the
new info dictionary consists of the *intersection* of the inputs: only
key-value pairs that were common to both of the input objects will be
retained in the new combined object. For example::
(GridDOSData([0.1, 0.2, 0.3], [y1, y2, y3],
info={'symbol': 'O', 'index': '1'})
+ GridDOSData([0.1, 0.2, 0.3], [y4, y5, y6],
info={'symbol': 'O', 'index': '2'}))
will yield the equivalent of::
GridDOSData([0.1, 0.2, 0.3], [y1+y4, y2+y5, y3+y6], info={'symbol': 'O'})
"""
def __init__(self,
energies: Sequence[float],
weights: Sequence[float],
info: Info = None) -> None:
n_entries = len(energies)
if not np.allclose(energies,
np.linspace(energies[0], energies[-1], n_entries)):
raise ValueError("Energies must be an evenly-spaced 1-D grid")
if len(weights) != n_entries:
raise ValueError("Energies and weights must be the same length")
super().__init__(energies, weights, info=info)
self.sigma_cutoff = 3
def _check_spacing(self, width) -> float:
current_spacing = self._data[0, 1] - self._data[0, 0]
if width < (2 * current_spacing):
warnings.warn(
"The broadening width is small compared to the original "
"sampling density. The results are unlikely to be smooth.")
return current_spacing
def _sample(self,
energies: Sequence[float],
width: float = 0.1,
smearing: str = 'Gauss') -> np.ndarray:
current_spacing = self._check_spacing(width)
return super()._sample(energies=energies,
width=width, smearing=smearing
) * current_spacing
def __add__(self, other: 'GridDOSData') -> 'GridDOSData':
# This method uses direct access to the mutable energy and weights data
# (self._data) to avoid redundant copying operations. The __init__
# method of GridDOSData will write this to a new array, so on this
# occasion it is safe to pass references to the mutable data.
if not isinstance(other, GridDOSData):
raise TypeError("GridDOSData can only be combined with other "
"GridDOSData objects")
if len(self._data[0, :]) != len(other.get_energies()):
raise ValueError("Cannot add GridDOSData objects with different-"
"length energy grids.")
if not np.allclose(self._data[0, :], other.get_energies()):
raise ValueError("Cannot add GridDOSData objects with different "
"energy grids.")
# Take intersection of metadata (i.e. only common entries are retained)
new_info = dict(set(self.info.items()) & set(other.info.items()))
# Concatenate the energy/weight data
new_weights = self._data[1, :] + other.get_weights()
new_object = GridDOSData(self._data[0, :], new_weights,
info=new_info)
return new_object
@staticmethod
def _interpret_smearing_args(npts: int,
width: float = None,
default_npts: int = 1000,
default_width: float = 0.1
) -> Tuple[int, Union[float, None]]:
"""Figure out what the user intended: resample if width provided"""
if width is not None:
if npts:
return (npts, float(width))
else:
return (default_npts, float(width))
else:
if npts:
return (npts, default_width)
else:
return (0, None)
[docs] def plot(self,
npts: int = 0,
xmin: float = None,
xmax: float = None,
width: float = None,
smearing: str = 'Gauss',
ax: 'matplotlib.axes.Axes' = None,
show: bool = False,
filename: str = None,
mplargs: dict = None) -> 'matplotlib.axes.Axes':
"""Simple 1-D plot of DOS data
Data will be resampled onto a grid with `npts` points unless `npts` is
set to zero, in which case:
- no resampling takes place
- `width` and `smearing` are ignored
- `xmin` and `xmax` affect the axis limits of the plot, not the
underlying data.
If the special key 'label' is present in self.info, this will be set
as the label for the plotted line (unless overruled in mplargs). The
label is only seen if a legend is added to the plot (i.e. by calling
``ax.legend()``).
Args:
npts, xmin, xmax: output data range, as passed to self.sample_grid
width: Width of broadening kernel, passed to self.sample_grid().
If no npts was set but width is set, npts will be set to 1000.
smearing: selection of broadening kernel for self.sample_grid()
ax: existing Matplotlib axes object. If not provided, a new figure
with one set of axes will be created using Pyplot
show: show the figure on-screen
filename: if a path is given, save the figure to this file
mplargs: additional arguments to pass to matplotlib plot command
(e.g. {'linewidth': 2} for a thicker line).
Returns:
Plotting axes. If "ax" was set, this is the same object.
"""
npts, width = self._interpret_smearing_args(npts, width)
if mplargs is None:
mplargs = {}
if 'label' not in mplargs:
mplargs.update({'label': self.label_from_info(self.info)})
if npts:
assert isinstance(width, float)
dos = self.sample_grid(npts, xmin=xmin,
xmax=xmax, width=width,
smearing=smearing)
else:
dos = self
energies, intensity = dos.get_energies(), dos.get_weights()
with SimplePlottingAxes(ax=ax, show=show, filename=filename) as ax:
ax.plot(energies, intensity, **mplargs)
ax.set_xlim(left=xmin, right=xmax)
return ax