# -*- coding: utf-8 -*-
"""Compound tensor algebra operations."""
# Copyright (C) 2008-2016 Martin Sandve Alnæs
#
# This file is part of UFL (https://www.fenicsproject.org)
#
# SPDX-License-Identifier: LGPL-3.0-or-later
from ufl.log import error
from ufl.core.expr import ufl_err_str
from ufl.core.ufl_type import ufl_type
from ufl.constantvalue import Zero
from ufl.algebra import Operator, Conj
from ufl.precedence import parstr
from ufl.sorting import sorted_expr
from ufl.index_combination_utils import merge_nonoverlapping_indices
# Algebraic operations on tensors:
# FloatValues:
# dot(a,b) = a*b
# inner(a,b) = a*b
# outer(a,b) = a*b
# Vectors:
# dot(u,v) = u_i v_i
# inner(u,v) = u_i v_i
# outer(u,v) = A | A_ij = u_i v_j
# Matrices:
# dot(A,B) = C | C_ij = A_{ik} B_{kj}
# inner(A,B) = A_{ij} B_{ij}
# outer(A,B) = C | C_ijkl = A_ij B_kl
# Combined:
# dot(A,u) = v | v_i = A_{ik} u_k
# inner(A,u) = v | v_i = A_{ik} u_k
# outer(A,u) = C | C_ijk = B_ij u_k
# dot(u,B) = v | v_i = u_k B_{ki}
# inner(u,B) = v | v_i = u_k B_{ki}
# outer(u,B) = C | C_ijk = u_i B_jk
#
# Argument requirements:
# dot(x,y): last index of x has same dimension as first index of y
# inner(x,y): shape of x equals the shape of y
# --- Classes representing compound tensor algebra operations ---
[docs]@ufl_type(is_abstract=True)
class CompoundTensorOperator(Operator):
__slots__ = ()
def __init__(self, operands):
Operator.__init__(self, operands)
# TODO: Use this and make Sum handle scalars only?
# This would simplify some algorithms. The only
# problem is we can't use + in many algorithms because
# this type should be expanded by expand_compounds.
# class TensorSum(CompoundTensorOperator):
# "Sum of nonscalar expressions."
# pass
# TODO: Use this similarly to TensorSum?
# This would simplify some algorithms. The only
# problem is we can't use / in many algorithms because
# this type should be expanded by expand_compounds.
# class TensorDivision(CompoundTensorOperator):
# "Division of nonscalar expression with a scalar expression."
# pass
# TODO: Use this similarly to TensorSum?
# This would simplify some algorithms. The only
# problem is we can't use * in many algorithms because
# this type should be expanded by expand_compounds.
# class MatrixProduct(CompoundTensorOperator):
# "Product of a matrix with a matrix or vector."
# pass
# TODO: Use this similarly to TensorSum?
# This would simplify some algorithms. The only
# problem is we can't use abs in many algorithms because
# this type should be expanded by expand_compounds.
# class TensorAbs(CompoundTensorOperator):
# "Absolute value of nonscalar expression."
# pass
[docs]@ufl_type(is_shaping=True, num_ops=1, inherit_indices_from_operand=0)
class Transposed(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
if isinstance(A, Zero):
a, b = A.ufl_shape
return Zero((b, a), A.ufl_free_indices, A.ufl_index_dimensions)
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
if len(A.ufl_shape) != 2:
error("Transposed is only defined for rank 2 tensors.")
@property
def ufl_shape(self):
s = self.ufl_operands[0].ufl_shape
return (s[1], s[0])
def __str__(self):
return "%s^T" % parstr(self.ufl_operands[0], self)
[docs]@ufl_type(num_ops=2)
class Outer(CompoundTensorOperator):
__slots__ = ("ufl_free_indices", "ufl_index_dimensions")
def __new__(cls, a, b):
ash, bsh = a.ufl_shape, b.ufl_shape
if isinstance(a, Zero) or isinstance(b, Zero):
fi, fid = merge_nonoverlapping_indices(a, b)
return Zero(ash + bsh, fi, fid)
if ash == () or bsh == ():
return Conj(a) * b
return CompoundTensorOperator.__new__(cls)
def __init__(self, a, b):
CompoundTensorOperator.__init__(self, (a, b))
fi, fid = merge_nonoverlapping_indices(a, b)
self.ufl_free_indices = fi
self.ufl_index_dimensions = fid
@property
def ufl_shape(self):
return self.ufl_operands[0].ufl_shape + self.ufl_operands[1].ufl_shape
def __str__(self):
return "%s (X) %s" % (parstr(self.ufl_operands[0], self),
parstr(self.ufl_operands[1], self))
[docs]@ufl_type(num_ops=2)
class Inner(CompoundTensorOperator):
__slots__ = ("ufl_free_indices", "ufl_index_dimensions")
def __new__(cls, a, b):
# Checks
ash, bsh = a.ufl_shape, b.ufl_shape
if ash != bsh:
error("Shapes do not match: %s and %s." % (ufl_err_str(a), ufl_err_str(b)))
# Simplification
if isinstance(a, Zero) or isinstance(b, Zero):
fi, fid = merge_nonoverlapping_indices(a, b)
return Zero((), fi, fid)
elif ash == ():
return a * Conj(b)
# sort operands for unique representation,
# must be independent of various counts etc.
# as explained in cmp_expr
if (a, b) != tuple(sorted_expr((a, b))):
return Conj(Inner(b, a))
return CompoundTensorOperator.__new__(cls)
def __init__(self, a, b):
CompoundTensorOperator.__init__(self, (a, b))
fi, fid = merge_nonoverlapping_indices(a, b)
self.ufl_free_indices = fi
self.ufl_index_dimensions = fid
ufl_shape = ()
def __str__(self):
return "%s : %s" % (parstr(self.ufl_operands[0], self),
parstr(self.ufl_operands[1], self))
[docs]@ufl_type(num_ops=2)
class Dot(CompoundTensorOperator):
__slots__ = ("ufl_free_indices", "ufl_index_dimensions")
def __new__(cls, a, b):
ash = a.ufl_shape
bsh = b.ufl_shape
ar, br = len(ash), len(bsh)
scalar = (ar == 0 and br == 0)
# Checks
if not ((ar >= 1 and br >= 1) or scalar):
error("Dot product requires non-scalar arguments, "
"got arguments with ranks %d and %d." % (ar, br))
if not (scalar or ash[-1] == bsh[0]):
error("Dimension mismatch in dot product.")
# Simplification
if isinstance(a, Zero) or isinstance(b, Zero):
shape = ash[:-1] + bsh[1:]
fi, fid = merge_nonoverlapping_indices(a, b)
return Zero(shape, fi, fid)
elif scalar: # TODO: Move this to def dot()?
return a * b
return CompoundTensorOperator.__new__(cls)
def __init__(self, a, b):
CompoundTensorOperator.__init__(self, (a, b))
fi, fid = merge_nonoverlapping_indices(a, b)
self.ufl_free_indices = fi
self.ufl_index_dimensions = fid
@property
def ufl_shape(self):
return self.ufl_operands[0].ufl_shape[:-1] + self.ufl_operands[1].ufl_shape[1:]
def __str__(self):
return "%s . %s" % (parstr(self.ufl_operands[0], self),
parstr(self.ufl_operands[1], self))
[docs]@ufl_type(num_ops=2)
class Cross(CompoundTensorOperator):
__slots__ = ("ufl_free_indices", "ufl_index_dimensions")
def __new__(cls, a, b):
ash = a.ufl_shape
bsh = b.ufl_shape
# Checks
if not (len(ash) == 1 and ash == bsh):
error("Cross product requires arguments of rank 1, got %s and %s." % (
ufl_err_str(a), ufl_err_str(b)))
# Simplification
if isinstance(a, Zero) or isinstance(b, Zero):
fi, fid = merge_nonoverlapping_indices(a, b)
return Zero(ash, fi, fid)
return CompoundTensorOperator.__new__(cls)
def __init__(self, a, b):
CompoundTensorOperator.__init__(self, (a, b))
fi, fid = merge_nonoverlapping_indices(a, b)
self.ufl_free_indices = fi
self.ufl_index_dimensions = fid
ufl_shape = (3,)
def __str__(self):
return "%s x %s" % (parstr(self.ufl_operands[0], self),
parstr(self.ufl_operands[1], self))
[docs]@ufl_type(num_ops=1, inherit_indices_from_operand=0)
class Trace(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
# Checks
if len(A.ufl_shape) != 2:
error("Trace of tensor with rank != 2 is undefined.")
# Simplification
if isinstance(A, Zero):
return Zero((), A.ufl_free_indices, A.ufl_index_dimensions)
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
ufl_shape = ()
def __str__(self):
return "tr(%s)" % self.ufl_operands[0]
[docs]@ufl_type(is_scalar=True, num_ops=1)
class Determinant(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
sh = A.ufl_shape
r = len(sh)
Afi = A.ufl_free_indices
# Checks
if r not in (0, 2):
error("Determinant of tensor with rank != 2 is undefined.")
if r == 2 and sh[0] != sh[1]:
error("Cannot take determinant of rectangular rank 2 tensor.")
if Afi:
error("Not expecting free indices in determinant.")
# Simplification
if isinstance(A, Zero):
return Zero((), Afi, A.ufl_index_dimensions)
if r == 0:
return A
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
def __str__(self):
return "det(%s)" % self.ufl_operands[0]
# TODO: Drop Inverse and represent it as product of Determinant and
# Cofactor?
[docs]@ufl_type(is_index_free=True, num_ops=1)
class Inverse(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
sh = A.ufl_shape
r = len(sh)
# Checks
if A.ufl_free_indices:
error("Not expecting free indices in Inverse.")
if isinstance(A, Zero):
error("Division by zero!")
# Simplification
if r == 0:
return 1 / A
# More checks
if r != 2:
error("Inverse of tensor with rank != 2 is undefined.")
if sh[0] != sh[1]:
error("Cannot take inverse of rectangular matrix with dimensions %s." % (sh,))
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
@property
def ufl_shape(self):
return self.ufl_operands[0].ufl_shape
def __str__(self):
return "%s^-1" % parstr(self.ufl_operands[0], self)
[docs]@ufl_type(is_index_free=True, num_ops=1)
class Cofactor(CompoundTensorOperator):
__slots__ = ()
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
# Checks
sh = A.ufl_shape
if len(sh) != 2:
error("Cofactor of tensor with rank != 2 is undefined.")
if sh[0] != sh[1]:
error("Cannot take cofactor of rectangular matrix with dimensions %s." % (sh,))
if A.ufl_free_indices:
error("Not expecting free indices in Cofactor.")
if isinstance(A, Zero):
error("Cannot take cofactor of zero matrix.")
@property
def ufl_shape(self):
return self.ufl_operands[0].ufl_shape
def __str__(self):
return "cofactor(%s)" % self.ufl_operands[0]
[docs]@ufl_type(num_ops=1, inherit_shape_from_operand=0, inherit_indices_from_operand=0)
class Deviatoric(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
sh = A.ufl_shape
# Checks
if len(sh) != 2:
error("Deviatoric part of tensor with rank != 2 is undefined.")
if sh[0] != sh[1]:
error("Cannot take deviatoric part of rectangular matrix with dimensions %s." % (sh,))
if A.ufl_free_indices:
error("Not expecting free indices in Deviatoric.")
# Simplification
if isinstance(A, Zero):
return Zero(A.ufl_shape, A.ufl_free_indices, A.ufl_index_dimensions)
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
def __str__(self):
return "dev(%s)" % self.ufl_operands[0]
[docs]@ufl_type(num_ops=1, inherit_shape_from_operand=0, inherit_indices_from_operand=0)
class Skew(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
sh = A.ufl_shape
Afi = A.ufl_free_indices
# Checks
if len(sh) != 2:
error("Skew symmetric part of tensor with rank != 2 is undefined.")
if sh[0] != sh[1]:
error("Cannot take skew part of rectangular matrix with dimensions %s." % (sh,))
if Afi:
error("Not expecting free indices in Skew.")
# Simplification
if isinstance(A, Zero):
return Zero(A.ufl_shape, Afi, A.ufl_index_dimensions)
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
def __str__(self):
return "skew(%s)" % self.ufl_operands[0]
[docs]@ufl_type(num_ops=1, inherit_shape_from_operand=0, inherit_indices_from_operand=0)
class Sym(CompoundTensorOperator):
__slots__ = ()
def __new__(cls, A):
sh = A.ufl_shape
Afi = A.ufl_free_indices
# Checks
if len(sh) != 2:
error("Symmetric part of tensor with rank != 2 is undefined.")
if sh[0] != sh[1]:
error("Cannot take symmetric part of rectangular matrix with dimensions %s." % (sh,))
if Afi:
error("Not expecting free indices in Sym.")
# Simplification
if isinstance(A, Zero):
return Zero(A.ufl_shape, Afi, A.ufl_index_dimensions)
return CompoundTensorOperator.__new__(cls)
def __init__(self, A):
CompoundTensorOperator.__init__(self, (A,))
def __str__(self):
return "sym(%s)" % self.ufl_operands[0]