You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
436 lines
13 KiB
436 lines
13 KiB
"""Low-level linear systems solver. """
|
|
|
|
|
|
from sympy.utilities.exceptions import sympy_deprecation_warning
|
|
from sympy.utilities.iterables import connected_components
|
|
|
|
from sympy.core.sympify import sympify
|
|
from sympy.core.numbers import Integer, Rational
|
|
from sympy.matrices.dense import MutableDenseMatrix
|
|
from sympy.polys.domains import ZZ, QQ
|
|
|
|
from sympy.polys.domains import EX
|
|
from sympy.polys.rings import sring
|
|
from sympy.polys.polyerrors import NotInvertible
|
|
from sympy.polys.domainmatrix import DomainMatrix
|
|
|
|
|
|
class PolyNonlinearError(Exception):
|
|
"""Raised by solve_lin_sys for nonlinear equations"""
|
|
pass
|
|
|
|
|
|
class RawMatrix(MutableDenseMatrix):
|
|
"""
|
|
.. deprecated:: 1.9
|
|
|
|
This class fundamentally is broken by design. Use ``DomainMatrix`` if
|
|
you want a matrix over the polys domains or ``Matrix`` for a matrix
|
|
with ``Expr`` elements. The ``RawMatrix`` class will be removed/broken
|
|
in future in order to reestablish the invariant that the elements of a
|
|
Matrix should be of type ``Expr``.
|
|
|
|
"""
|
|
_sympify = staticmethod(lambda x: x) # type: ignore
|
|
|
|
def __init__(self, *args, **kwargs):
|
|
sympy_deprecation_warning(
|
|
"""
|
|
The RawMatrix class is deprecated. Use either DomainMatrix or
|
|
Matrix instead.
|
|
""",
|
|
deprecated_since_version="1.9",
|
|
active_deprecations_target="deprecated-rawmatrix",
|
|
)
|
|
|
|
domain = ZZ
|
|
for i in range(self.rows):
|
|
for j in range(self.cols):
|
|
val = self[i,j]
|
|
if getattr(val, 'is_Poly', False):
|
|
K = val.domain[val.gens]
|
|
val_sympy = val.as_expr()
|
|
elif hasattr(val, 'parent'):
|
|
K = val.parent()
|
|
val_sympy = K.to_sympy(val)
|
|
elif isinstance(val, (int, Integer)):
|
|
K = ZZ
|
|
val_sympy = sympify(val)
|
|
elif isinstance(val, Rational):
|
|
K = QQ
|
|
val_sympy = val
|
|
else:
|
|
for K in ZZ, QQ:
|
|
if K.of_type(val):
|
|
val_sympy = K.to_sympy(val)
|
|
break
|
|
else:
|
|
raise TypeError
|
|
domain = domain.unify(K)
|
|
self[i,j] = val_sympy
|
|
self.ring = domain
|
|
|
|
|
|
def eqs_to_matrix(eqs_coeffs, eqs_rhs, gens, domain):
|
|
"""Get matrix from linear equations in dict format.
|
|
|
|
Explanation
|
|
===========
|
|
|
|
Get the matrix representation of a system of linear equations represented
|
|
as dicts with low-level DomainElement coefficients. This is an
|
|
*internal* function that is used by solve_lin_sys.
|
|
|
|
Parameters
|
|
==========
|
|
|
|
eqs_coeffs: list[dict[Symbol, DomainElement]]
|
|
The left hand sides of the equations as dicts mapping from symbols to
|
|
coefficients where the coefficients are instances of
|
|
DomainElement.
|
|
eqs_rhs: list[DomainElements]
|
|
The right hand sides of the equations as instances of
|
|
DomainElement.
|
|
gens: list[Symbol]
|
|
The unknowns in the system of equations.
|
|
domain: Domain
|
|
The domain for coefficients of both lhs and rhs.
|
|
|
|
Returns
|
|
=======
|
|
|
|
The augmented matrix representation of the system as a DomainMatrix.
|
|
|
|
Examples
|
|
========
|
|
|
|
>>> from sympy import symbols, ZZ
|
|
>>> from sympy.polys.solvers import eqs_to_matrix
|
|
>>> x, y = symbols('x, y')
|
|
>>> eqs_coeff = [{x:ZZ(1), y:ZZ(1)}, {x:ZZ(1), y:ZZ(-1)}]
|
|
>>> eqs_rhs = [ZZ(0), ZZ(-1)]
|
|
>>> eqs_to_matrix(eqs_coeff, eqs_rhs, [x, y], ZZ)
|
|
DomainMatrix([[1, 1, 0], [1, -1, 1]], (2, 3), ZZ)
|
|
|
|
See also
|
|
========
|
|
|
|
solve_lin_sys: Uses :func:`~eqs_to_matrix` internally
|
|
"""
|
|
sym2index = {x: n for n, x in enumerate(gens)}
|
|
nrows = len(eqs_coeffs)
|
|
ncols = len(gens) + 1
|
|
rows = [[domain.zero] * ncols for _ in range(nrows)]
|
|
for row, eq_coeff, eq_rhs in zip(rows, eqs_coeffs, eqs_rhs):
|
|
for sym, coeff in eq_coeff.items():
|
|
row[sym2index[sym]] = domain.convert(coeff)
|
|
row[-1] = -domain.convert(eq_rhs)
|
|
|
|
return DomainMatrix(rows, (nrows, ncols), domain)
|
|
|
|
|
|
def sympy_eqs_to_ring(eqs, symbols):
|
|
"""Convert a system of equations from Expr to a PolyRing
|
|
|
|
Explanation
|
|
===========
|
|
|
|
High-level functions like ``solve`` expect Expr as inputs but can use
|
|
``solve_lin_sys`` internally. This function converts equations from
|
|
``Expr`` to the low-level poly types used by the ``solve_lin_sys``
|
|
function.
|
|
|
|
Parameters
|
|
==========
|
|
|
|
eqs: List of Expr
|
|
A list of equations as Expr instances
|
|
symbols: List of Symbol
|
|
A list of the symbols that are the unknowns in the system of
|
|
equations.
|
|
|
|
Returns
|
|
=======
|
|
|
|
Tuple[List[PolyElement], Ring]: The equations as PolyElement instances
|
|
and the ring of polynomials within which each equation is represented.
|
|
|
|
Examples
|
|
========
|
|
|
|
>>> from sympy import symbols
|
|
>>> from sympy.polys.solvers import sympy_eqs_to_ring
|
|
>>> a, x, y = symbols('a, x, y')
|
|
>>> eqs = [x-y, x+a*y]
|
|
>>> eqs_ring, ring = sympy_eqs_to_ring(eqs, [x, y])
|
|
>>> eqs_ring
|
|
[x - y, x + a*y]
|
|
>>> type(eqs_ring[0])
|
|
<class 'sympy.polys.rings.PolyElement'>
|
|
>>> ring
|
|
ZZ(a)[x,y]
|
|
|
|
With the equations in this form they can be passed to ``solve_lin_sys``:
|
|
|
|
>>> from sympy.polys.solvers import solve_lin_sys
|
|
>>> solve_lin_sys(eqs_ring, ring)
|
|
{y: 0, x: 0}
|
|
"""
|
|
try:
|
|
K, eqs_K = sring(eqs, symbols, field=True, extension=True)
|
|
except NotInvertible:
|
|
# https://github.com/sympy/sympy/issues/18874
|
|
K, eqs_K = sring(eqs, symbols, domain=EX)
|
|
return eqs_K, K.to_domain()
|
|
|
|
|
|
def solve_lin_sys(eqs, ring, _raw=True):
|
|
"""Solve a system of linear equations from a PolynomialRing
|
|
|
|
Explanation
|
|
===========
|
|
|
|
Solves a system of linear equations given as PolyElement instances of a
|
|
PolynomialRing. The basic arithmetic is carried out using instance of
|
|
DomainElement which is more efficient than :class:`~sympy.core.expr.Expr`
|
|
for the most common inputs.
|
|
|
|
While this is a public function it is intended primarily for internal use
|
|
so its interface is not necessarily convenient. Users are suggested to use
|
|
the :func:`sympy.solvers.solveset.linsolve` function (which uses this
|
|
function internally) instead.
|
|
|
|
Parameters
|
|
==========
|
|
|
|
eqs: list[PolyElement]
|
|
The linear equations to be solved as elements of a
|
|
PolynomialRing (assumed equal to zero).
|
|
ring: PolynomialRing
|
|
The polynomial ring from which eqs are drawn. The generators of this
|
|
ring are the unknowns to be solved for and the domain of the ring is
|
|
the domain of the coefficients of the system of equations.
|
|
_raw: bool
|
|
If *_raw* is False, the keys and values in the returned dictionary
|
|
will be of type Expr (and the unit of the field will be removed from
|
|
the keys) otherwise the low-level polys types will be returned, e.g.
|
|
PolyElement: PythonRational.
|
|
|
|
Returns
|
|
=======
|
|
|
|
``None`` if the system has no solution.
|
|
|
|
dict[Symbol, Expr] if _raw=False
|
|
|
|
dict[Symbol, DomainElement] if _raw=True.
|
|
|
|
Examples
|
|
========
|
|
|
|
>>> from sympy import symbols
|
|
>>> from sympy.polys.solvers import solve_lin_sys, sympy_eqs_to_ring
|
|
>>> x, y = symbols('x, y')
|
|
>>> eqs = [x - y, x + y - 2]
|
|
>>> eqs_ring, ring = sympy_eqs_to_ring(eqs, [x, y])
|
|
>>> solve_lin_sys(eqs_ring, ring)
|
|
{y: 1, x: 1}
|
|
|
|
Passing ``_raw=False`` returns the same result except that the keys are
|
|
``Expr`` rather than low-level poly types.
|
|
|
|
>>> solve_lin_sys(eqs_ring, ring, _raw=False)
|
|
{x: 1, y: 1}
|
|
|
|
See also
|
|
========
|
|
|
|
sympy_eqs_to_ring: prepares the inputs to ``solve_lin_sys``.
|
|
linsolve: ``linsolve`` uses ``solve_lin_sys`` internally.
|
|
sympy.solvers.solvers.solve: ``solve`` uses ``solve_lin_sys`` internally.
|
|
"""
|
|
as_expr = not _raw
|
|
|
|
assert ring.domain.is_Field
|
|
|
|
eqs_dict = [dict(eq) for eq in eqs]
|
|
|
|
one_monom = ring.one.monoms()[0]
|
|
zero = ring.domain.zero
|
|
|
|
eqs_rhs = []
|
|
eqs_coeffs = []
|
|
for eq_dict in eqs_dict:
|
|
eq_rhs = eq_dict.pop(one_monom, zero)
|
|
eq_coeffs = {}
|
|
for monom, coeff in eq_dict.items():
|
|
if sum(monom) != 1:
|
|
msg = "Nonlinear term encountered in solve_lin_sys"
|
|
raise PolyNonlinearError(msg)
|
|
eq_coeffs[ring.gens[monom.index(1)]] = coeff
|
|
if not eq_coeffs:
|
|
if not eq_rhs:
|
|
continue
|
|
else:
|
|
return None
|
|
eqs_rhs.append(eq_rhs)
|
|
eqs_coeffs.append(eq_coeffs)
|
|
|
|
result = _solve_lin_sys(eqs_coeffs, eqs_rhs, ring)
|
|
|
|
if result is not None and as_expr:
|
|
|
|
def to_sympy(x):
|
|
as_expr = getattr(x, 'as_expr', None)
|
|
if as_expr:
|
|
return as_expr()
|
|
else:
|
|
return ring.domain.to_sympy(x)
|
|
|
|
tresult = {to_sympy(sym): to_sympy(val) for sym, val in result.items()}
|
|
|
|
# Remove 1.0x
|
|
result = {}
|
|
for k, v in tresult.items():
|
|
if k.is_Mul:
|
|
c, s = k.as_coeff_Mul()
|
|
result[s] = v/c
|
|
else:
|
|
result[k] = v
|
|
|
|
return result
|
|
|
|
|
|
def _solve_lin_sys(eqs_coeffs, eqs_rhs, ring):
|
|
"""Solve a linear system from dict of PolynomialRing coefficients
|
|
|
|
Explanation
|
|
===========
|
|
|
|
This is an **internal** function used by :func:`solve_lin_sys` after the
|
|
equations have been preprocessed. The role of this function is to split
|
|
the system into connected components and pass those to
|
|
:func:`_solve_lin_sys_component`.
|
|
|
|
Examples
|
|
========
|
|
|
|
Setup a system for $x-y=0$ and $x+y=2$ and solve:
|
|
|
|
>>> from sympy import symbols, sring
|
|
>>> from sympy.polys.solvers import _solve_lin_sys
|
|
>>> x, y = symbols('x, y')
|
|
>>> R, (xr, yr) = sring([x, y], [x, y])
|
|
>>> eqs = [{xr:R.one, yr:-R.one}, {xr:R.one, yr:R.one}]
|
|
>>> eqs_rhs = [R.zero, -2*R.one]
|
|
>>> _solve_lin_sys(eqs, eqs_rhs, R)
|
|
{y: 1, x: 1}
|
|
|
|
See also
|
|
========
|
|
|
|
solve_lin_sys: This function is used internally by :func:`solve_lin_sys`.
|
|
"""
|
|
V = ring.gens
|
|
E = []
|
|
for eq_coeffs in eqs_coeffs:
|
|
syms = list(eq_coeffs)
|
|
E.extend(zip(syms[:-1], syms[1:]))
|
|
G = V, E
|
|
|
|
components = connected_components(G)
|
|
|
|
sym2comp = {}
|
|
for n, component in enumerate(components):
|
|
for sym in component:
|
|
sym2comp[sym] = n
|
|
|
|
subsystems = [([], []) for _ in range(len(components))]
|
|
for eq_coeff, eq_rhs in zip(eqs_coeffs, eqs_rhs):
|
|
sym = next(iter(eq_coeff), None)
|
|
sub_coeff, sub_rhs = subsystems[sym2comp[sym]]
|
|
sub_coeff.append(eq_coeff)
|
|
sub_rhs.append(eq_rhs)
|
|
|
|
sol = {}
|
|
for subsystem in subsystems:
|
|
subsol = _solve_lin_sys_component(subsystem[0], subsystem[1], ring)
|
|
if subsol is None:
|
|
return None
|
|
sol.update(subsol)
|
|
|
|
return sol
|
|
|
|
|
|
def _solve_lin_sys_component(eqs_coeffs, eqs_rhs, ring):
|
|
"""Solve a linear system from dict of PolynomialRing coefficients
|
|
|
|
Explanation
|
|
===========
|
|
|
|
This is an **internal** function used by :func:`solve_lin_sys` after the
|
|
equations have been preprocessed. After :func:`_solve_lin_sys` splits the
|
|
system into connected components this function is called for each
|
|
component. The system of equations is solved using Gauss-Jordan
|
|
elimination with division followed by back-substitution.
|
|
|
|
Examples
|
|
========
|
|
|
|
Setup a system for $x-y=0$ and $x+y=2$ and solve:
|
|
|
|
>>> from sympy import symbols, sring
|
|
>>> from sympy.polys.solvers import _solve_lin_sys_component
|
|
>>> x, y = symbols('x, y')
|
|
>>> R, (xr, yr) = sring([x, y], [x, y])
|
|
>>> eqs = [{xr:R.one, yr:-R.one}, {xr:R.one, yr:R.one}]
|
|
>>> eqs_rhs = [R.zero, -2*R.one]
|
|
>>> _solve_lin_sys_component(eqs, eqs_rhs, R)
|
|
{y: 1, x: 1}
|
|
|
|
See also
|
|
========
|
|
|
|
solve_lin_sys: This function is used internally by :func:`solve_lin_sys`.
|
|
"""
|
|
|
|
# transform from equations to matrix form
|
|
matrix = eqs_to_matrix(eqs_coeffs, eqs_rhs, ring.gens, ring.domain)
|
|
|
|
# convert to a field for rref
|
|
if not matrix.domain.is_Field:
|
|
matrix = matrix.to_field()
|
|
|
|
# solve by row-reduction
|
|
echelon, pivots = matrix.rref()
|
|
|
|
# construct the returnable form of the solutions
|
|
keys = ring.gens
|
|
|
|
if pivots and pivots[-1] == len(keys):
|
|
return None
|
|
|
|
if len(pivots) == len(keys):
|
|
sol = []
|
|
for s in [row[-1] for row in echelon.rep.to_ddm()]:
|
|
a = s
|
|
sol.append(a)
|
|
sols = dict(zip(keys, sol))
|
|
else:
|
|
sols = {}
|
|
g = ring.gens
|
|
# Extract ground domain coefficients and convert to the ring:
|
|
if hasattr(ring, 'ring'):
|
|
convert = ring.ring.ground_new
|
|
else:
|
|
convert = ring.ground_new
|
|
echelon = echelon.rep.to_ddm()
|
|
vals_set = {v for row in echelon for v in row}
|
|
vals_map = {v: convert(v) for v in vals_set}
|
|
echelon = [[vals_map[eij] for eij in ei] for ei in echelon]
|
|
for i, p in enumerate(pivots):
|
|
v = echelon[i][-1] - sum(echelon[i][j]*g[j] for j in range(p+1, len(g)) if echelon[i][j])
|
|
sols[keys[p]] = v
|
|
|
|
return sols
|