Source code for pybamm.expression_tree.broadcasts

#
# Unary operator classes and methods
#
from __future__ import annotations

import numpy as np
from scipy.sparse import csr_matrix
from typing import cast

import pybamm
from pybamm.type_definitions import (
    DomainType,
    AuxiliaryDomainType,
    DomainsType,
    Numeric,
)


[docs] class Broadcast(pybamm.SpatialOperator): """ A node in the expression tree representing a broadcasting operator. Broadcasts a child to a specified domain. After discretisation, this will evaluate to an array of the right shape for the specified domain. For an example of broadcasts in action, see `this example notebook <https://github.com/pybamm-team/PyBaMM/blob/develop/docs/source/examples/notebooks/expression_tree/broadcasts.ipynb>`_ Parameters ---------- child : :class:`Symbol` child node domains : iterable of str Domain(s) of the symbol after broadcasting name : str name of the node """ def __init__( self, child: pybamm.Symbol, domains: dict[str, list[str] | str], name: str | None = None, ): if name is None: name = "broadcast" super().__init__(name, child, domains=domains) @property def broadcasts_to_nodes(self): if self.broadcast_type.endswith("nodes"): return True else: return False def _sympy_operator(self, child: pybamm.Symbol): """Override :meth:`pybamm.UnaryOperator._sympy_operator`""" return child def _diff(self, variable): """See :meth:`pybamm.Symbol._diff()`.""" # Differentiate the child and broadcast the result in the same way return self._unary_new_copy(self.child.diff(variable))
[docs] def reduce_one_dimension(self): # pragma: no cover """Reduce the broadcast by one dimension.""" raise NotImplementedError
[docs] def to_json(self): raise NotImplementedError( "pybamm.Broadcast: Serialisation is only implemented for discretised models" )
@classmethod def _from_json(cls, snippet): raise NotImplementedError( "pybamm.Broadcast: Please use a discretised model when reading in from JSON" ) def _unary_new_copy(self, child: pybamm.Symbol, perform_simplifications=True): """See :meth:`pybamm.UnaryOperator._unary_new_copy()`.""" return self.__class__(child, self.broadcast_domain)
[docs] class PrimaryBroadcast(Broadcast): """ A node in the expression tree representing a primary broadcasting operator. Broadcasts in a `primary` dimension only. That is, makes explicit copies of the symbol in the domain specified by `broadcast_domain`. This should be used for broadcasting from a "larger" scale to a "smaller" scale, for example broadcasting temperature T(x) from the electrode to the particles, or broadcasting current collector current i(y, z) from the current collector to the electrodes. Parameters ---------- child : :class:`Symbol`, numeric child node broadcast_domain : iterable of str Primary domain for broadcast. This will become the domain of the symbol name : str name of the node """ def __init__( self, child: Numeric | pybamm.Symbol, broadcast_domain: list[str] | str, name: str | None = None, ): # Convert child to scalar if it is a number if isinstance(child, (float, int, np.number)): child = pybamm.Scalar(child) # cast child to Symbol for mypy child = cast(pybamm.Symbol, child) # Convert domain to list if it's a string if isinstance(broadcast_domain, str): broadcast_domain = [broadcast_domain] # perform some basic checks and set attributes domains = self.check_and_set_domains(child, broadcast_domain) self.broadcast_domain = broadcast_domain self.broadcast_type = "primary to nodes" super().__init__(child, domains, name=name)
[docs] def check_and_set_domains(self, child: pybamm.Symbol, broadcast_domain: list[str]): """See :meth:`Broadcast.check_and_set_domains`""" # Can only do primary broadcast from current collector to electrode, # particle-size or particle or from electrode to particle-size or particle. # Note e.g. current collector to particle *is* allowed if broadcast_domain == []: raise pybamm.DomainError("Cannot Broadcast an object into empty domain.") if child.domain == []: pass elif child.domain == ["current collector"] and not ( broadcast_domain[0] in [ "negative electrode", "separator", "positive electrode", ] or "particle" in broadcast_domain[0] ): raise pybamm.DomainError( """Primary broadcast from current collector domain must be to electrode or separator or particle or particle size domains""" ) elif ( child.domain[0] in [ "negative electrode", "separator", "positive electrode", ] and "particle" not in broadcast_domain[0] ): raise pybamm.DomainError( """Primary broadcast from electrode or separator must be to particle or particle size domains""" ) elif child.domain[0] in [ "negative particle size", "positive particle size", ] and broadcast_domain[0] not in ["negative particle", "positive particle"]: raise pybamm.DomainError( """Primary broadcast from particle size domain must be to particle domain""" ) elif child.domain[0] in ["negative particle", "positive particle"]: raise pybamm.DomainError("Cannot do primary broadcast from particle domain") domains = { "primary": broadcast_domain, "secondary": child.domain, "tertiary": child.domains["secondary"], "quaternary": child.domains["tertiary"], } return domains
def _evaluate_for_shape(self): """ Returns a vector of NaNs to represent the shape of a Broadcast. See :meth:`pybamm.Symbol.evaluate_for_shape_using_domain()` """ child_eval = self.children[0].evaluate_for_shape() vec = pybamm.evaluate_for_shape_using_domain(self.domains["primary"]) return np.outer(child_eval, vec).reshape(-1, 1)
[docs] def reduce_one_dimension(self): """Reduce the broadcast by one dimension.""" return self.orphans[0]
[docs] class PrimaryBroadcastToEdges(PrimaryBroadcast): """A primary broadcast onto the edges of the domain.""" def __init__( self, child: Numeric | pybamm.Symbol, broadcast_domain: list[str] | str, name: str | None = None, ): name = name or "broadcast to edges" super().__init__(child, broadcast_domain, name) self.broadcast_type = "primary to edges" def _evaluates_on_edges(self, dimension): return True
[docs] class SecondaryBroadcast(Broadcast): """ A node in the expression tree representing a secondary broadcasting operator. Broadcasts in a `secondary` dimension only. That is, makes explicit copies of the symbol in the domain specified by `broadcast_domain`. This should be used for broadcasting from a "smaller" scale to a "larger" scale, for example broadcasting SPM particle concentrations c_s(r) from the particles to the electrodes. Note that this wouldn't be used to broadcast particle concentrations in the DFN, since these already depend on both x and r. Parameters ---------- child : :class:`Symbol` child node broadcast_domain : iterable of str Secondary domain for broadcast. This will become the secondary domain of the symbol, shifting the child's `secondary` and `tertiary` (if present) over by one position. name : str name of the node """ def __init__( self, child: pybamm.Symbol, broadcast_domain: list[str] | str, name: str | None = None, ): # Convert domain to list if it's a string if isinstance(broadcast_domain, str): broadcast_domain = [broadcast_domain] # perform some basic checks and set attributes domains = self.check_and_set_domains(child, broadcast_domain) self.broadcast_domain = broadcast_domain self.broadcast_type = "secondary to nodes" super().__init__(child, domains, name=name)
[docs] def check_and_set_domains(self, child: pybamm.Symbol, broadcast_domain: list[str]): """See :meth:`Broadcast.check_and_set_domains`""" if child.domain == []: raise TypeError( "Cannot take SecondaryBroadcast of an object with empty domain. " "Use PrimaryBroadcast instead." ) # Can only do secondary broadcast from particle to electrode or current # collector or from electrode to current collector if child.domain[0] in [ "negative particle", "positive particle", ] and broadcast_domain[0] not in [ "negative particle size", "positive particle size", "negative electrode", "separator", "positive electrode", "current collector", ]: raise pybamm.DomainError( """Secondary broadcast from particle domain must be to particle-size, electrode, separator, or current collector domains""" ) if child.domain[0] in [ "negative particle size", "positive particle size", ] and broadcast_domain[0] not in [ "negative electrode", "separator", "positive electrode", "current collector", ]: raise pybamm.DomainError( """Secondary broadcast from particle size domain must be to electrode or separator or current collector domains""" ) elif child.domain[0] in [ "negative electrode", "separator", "positive electrode", ] and broadcast_domain != ["current collector"]: raise pybamm.DomainError( """Secondary broadcast from electrode or separator must be to current collector domains""" ) elif child.domain == ["current collector"]: raise pybamm.DomainError( "Cannot do secondary broadcast from current collector domain" ) # Domain stays the same as child domain and broadcast domain is secondary # domain # Child's secondary domain becomes tertiary domain, tertiary becomes quaternary domains = { "primary": child.domains["primary"], "secondary": broadcast_domain, "tertiary": child.domains["secondary"], "quaternary": child.domains["tertiary"], } return domains
def _evaluate_for_shape(self): """ Returns a vector of NaNs to represent the shape of a Broadcast. See :meth:`pybamm.Symbol.evaluate_for_shape_using_domain()` """ child_eval = self.children[0].evaluate_for_shape() vec = pybamm.evaluate_for_shape_using_domain(self.domains["secondary"]) return np.outer(vec, child_eval).reshape(-1, 1)
[docs] def reduce_one_dimension(self): """Reduce the broadcast by one dimension.""" return self.orphans[0]
[docs] class SecondaryBroadcastToEdges(SecondaryBroadcast): """A secondary broadcast onto the edges of a domain.""" def __init__( self, child: pybamm.Symbol, broadcast_domain: list[str] | str, name: str | None = None, ): name = name or "broadcast to edges" super().__init__(child, broadcast_domain, name) self.broadcast_type = "secondary to edges" def _evaluates_on_edges(self, dimension): return True
class TertiaryBroadcast(Broadcast): """ A node in the expression tree representing a tertiary broadcasting operator. Broadcasts in a `tertiary` dimension only. That is, makes explicit copies of the symbol in the domain specified by `broadcast_domain`. This is used, e.g., for broadcasting particle concentrations c_s(r,R) in the MPM, which have a `primary` and `secondary` domain, to the electrode x, which is added as a `tertiary` domain. Note: the symbol for broadcast must already have a non-empty `secondary` domain. Parameters ---------- child : :class:`Symbol` child node broadcast_domain : iterable of str The domain for broadcast. This will become the tertiary domain of the symbol. The `tertiary` domain of the child, if present, is shifted by one to the `quaternary` domain of the symbol. name : str name of the node """ def __init__( self, child: pybamm.Symbol, broadcast_domain: list[str] | str, name: str | None = None, ): # Convert domain to list if it's a string if isinstance(broadcast_domain, str): broadcast_domain = [broadcast_domain] # perform some basic checks and set attributes domains = self.check_and_set_domains(child, broadcast_domain) self.broadcast_domain = broadcast_domain self.broadcast_type = "tertiary to nodes" super().__init__(child, domains, name=name) def check_and_set_domains( self, child: pybamm.Symbol, broadcast_domain: list[str] | str ): """See :meth:`Broadcast.check_and_set_domains`""" if child.domains["secondary"] == []: raise TypeError( """Cannot take TertiaryBroadcast of an object without a secondary domain. Use SecondaryBroadcast instead.""" ) # Can only do tertiary broadcast to a "higher dimension" than the # secondary domain of child if child.domains["secondary"][0] in [ "negative particle size", "positive particle size", ] and broadcast_domain[0] not in [ "negative electrode", "separator", "positive electrode", "current collector", ]: raise pybamm.DomainError( """Tertiary broadcast from a symbol with particle size secondary domain must be to electrode, separator or current collector""" ) if child.domains["secondary"][0] in [ "negative electrode", "separator", "positive electrode", ] and broadcast_domain != ["current collector"]: raise pybamm.DomainError( """Tertiary broadcast from a symbol with an electrode or separator secondary domain must be to current collector""" ) if child.domains["secondary"] == ["current collector"]: raise pybamm.DomainError( """Cannot do tertiary broadcast for symbol with a current collector secondary domain""" ) # Primary and secondary domains stay the same as child's, # and broadcast domain is tertiary domains = { "primary": child.domains["primary"], "secondary": child.domains["secondary"], "tertiary": broadcast_domain, "quaternary": child.domains["tertiary"], } return domains def _evaluate_for_shape(self): """ Returns a vector of NaNs to represent the shape of a Broadcast. See :meth:`pybamm.Symbol.evaluate_for_shape_using_domain()` """ child_eval = self.children[0].evaluate_for_shape() vec = pybamm.evaluate_for_shape_using_domain(self.domains["tertiary"]) return np.outer(vec, child_eval).reshape(-1, 1) def reduce_one_dimension(self): """Reduce the broadcast by one dimension.""" raise NotImplementedError class TertiaryBroadcastToEdges(TertiaryBroadcast): """A tertiary broadcast onto the edges of a domain.""" def __init__( self, child: pybamm.Symbol, broadcast_domain: list[str] | str, name: str | None = None, ): name = name or "broadcast to edges" super().__init__(child, broadcast_domain, name) self.broadcast_type = "tertiary to edges" def _evaluates_on_edges(self, dimension): return True
[docs] class FullBroadcast(Broadcast): """A class for full broadcasts.""" def __init__( self, child_input: Numeric | pybamm.Symbol, broadcast_domain: DomainType = None, auxiliary_domains: AuxiliaryDomainType = None, broadcast_domains: DomainsType = None, name: str | None = None, ): # Convert child to scalar if it is a number if isinstance(child_input, (float, int, np.number)): child: pybamm.Scalar = pybamm.Scalar(child_input) else: child: pybamm.Symbol = child_input # type: ignore[no-redef] if isinstance(auxiliary_domains, str): auxiliary_domains = {"secondary": auxiliary_domains} broadcast_domains = self.read_domain_or_domains( broadcast_domain, auxiliary_domains, broadcast_domains ) # perform some basic checks and set attributes domains = self.check_and_set_domains(child, broadcast_domains) self.broadcast_domain = broadcast_domains["primary"] self.broadcast_type = "full to nodes" super().__init__(child, domains, name=name)
[docs] def check_and_set_domains(self, child: pybamm.Symbol, broadcast_domains: dict): """See :meth:`Broadcast.check_and_set_domains`""" if broadcast_domains["primary"] == []: raise pybamm.DomainError( """Cannot do full broadcast to an empty primary domain""" ) # Variables on the current collector can only be broadcast to 'primary' if child.domain == ["current collector"]: raise pybamm.DomainError( "Cannot do full broadcast from current collector domain" ) return broadcast_domains
def _unary_new_copy(self, child, perform_simplifications=True): """See :meth:`pybamm.UnaryOperator._unary_new_copy()`.""" return self.__class__(child, broadcast_domains=self.domains) def _evaluate_for_shape(self): """ Returns a vector of NaNs to represent the shape of a Broadcast. See :meth:`pybamm.Symbol.evaluate_for_shape_using_domain()` """ child_eval = self.children[0].evaluate_for_shape() vec = pybamm.evaluate_for_shape_using_domain(self.domains) return child_eval * vec
[docs] def reduce_one_dimension(self): """Reduce the broadcast by one dimension.""" if self.domains["secondary"] == []: return self.orphans[0] elif self.domains["tertiary"] == []: return PrimaryBroadcast(self.orphans[0], self.domains["secondary"]) else: domains = { "primary": self.domains["secondary"], "secondary": self.domains["tertiary"], "tertiary": self.domains["quaternary"], } return FullBroadcast(self.orphans[0], broadcast_domains=domains)
[docs] class FullBroadcastToEdges(FullBroadcast): """ A full broadcast onto the edges of a domain (edges of primary dimension, nodes of other dimensions) """ def __init__( self, child: Numeric | pybamm.Symbol, broadcast_domain: DomainType = None, auxiliary_domains: AuxiliaryDomainType = None, broadcast_domains: DomainsType = None, name: str | None = None, ): name = name or "broadcast to edges" super().__init__( child, broadcast_domain, auxiliary_domains, broadcast_domains, name ) self.broadcast_type = "full to edges" def _evaluates_on_edges(self, dimension): return True
[docs] def reduce_one_dimension(self): """Reduce the broadcast by one dimension.""" if self.domains["secondary"] == []: return self.orphans[0] elif self.domains["tertiary"] == []: return PrimaryBroadcastToEdges(self.orphans[0], self.domains["secondary"]) else: return FullBroadcastToEdges( self.orphans[0], broadcast_domains={ "primary": self.domains["secondary"], "secondary": self.domains["tertiary"], }, )
[docs] def full_like(symbols: tuple[pybamm.Symbol, ...], fill_value: float) -> pybamm.Symbol: """ Returns an array with the same shape and domains as the sum of the input symbols, with a constant value given by `fill_value`. Parameters ---------- symbols : :class:`Symbol` Symbols whose shape to copy fill_value : number Value to assign """ # Make a symbol that combines all the children, to get the right domain # that takes all the child symbols into account sum_symbol = symbols[0] for sym in symbols[1:]: sum_symbol += sym # Just return scalar if symbol shape is scalar if sum_symbol.evaluates_to_number(): return pybamm.Scalar(fill_value) try: shape = sum_symbol.shape # return dense array, except for a matrix of zeros if shape[1] != 1 and fill_value == 0: entries = csr_matrix(shape) else: entries = fill_value * np.ones(shape) # use vector or matrix if shape[1] == 1: return pybamm.Vector(entries, domains=sum_symbol.domains) else: return pybamm.Matrix(entries, domains=sum_symbol.domains) except NotImplementedError: if ( sum_symbol.shape_for_testing == (1, 1) or sum_symbol.shape_for_testing == (1,) or sum_symbol.domain == [] ): return pybamm.Scalar(fill_value) if sum_symbol.evaluates_on_edges("primary"): return FullBroadcastToEdges( fill_value, broadcast_domains=sum_symbol.domains ) else: return FullBroadcast(fill_value, broadcast_domains=sum_symbol.domains)
[docs] def zeros_like(*symbols: pybamm.Symbol): """ Returns an array with the same shape and domains as the sum of the input symbols, with each entry equal to zero. Parameters ---------- symbols : :class:`Symbol` Symbols whose shape to copy """ return full_like(symbols, 0)
[docs] def ones_like(*symbols: pybamm.Symbol): """ Returns an array with the same shape and domains as the sum of the input symbols, with each entry equal to one. Parameters ---------- symbols : :class:`Symbol` Symbols whose shape to copy """ return full_like(symbols, 1)