def test_rewrite_MaxMin_as_Piecewise():
from sympy import symbols, Piecewise
x, y, z, a, b = symbols('x y z a b', real=True)
vx, vy, va = symbols('vx vy va')
assert Max(a, b).rewrite(Piecewise) == Piecewise((a, a >= b), (b, True))
assert Max(x, y, z).rewrite(Piecewise) == Piecewise(
(x, (x >= y) & (x >= z)), (y, y >= z), (z, True))
assert Max(x, y, a, b).rewrite(Piecewise) == Piecewise(
(a, (a >= b) & (a >= x) & (a >= y)), (b, (b >= x) & (b >= y)),
(x, x >= y), (y, True))
assert Min(a, b).rewrite(Piecewise) == Piecewise((a, a <= b), (b, True))
assert Min(x, y, z).rewrite(Piecewise) == Piecewise(
(x, (x <= y) & (x <= z)), (y, y <= z), (z, True))
assert Min(x, y, a, b).rewrite(Piecewise) == Piecewise(
(a, (a <= b) & (a <= x) & (a <= y)), (b, (b <= x) & (b <= y)),
(x, x <= y), (y, True))
# Piecewise rewriting of Min/Max does also takes place for not explicitly real arguments
assert Max(vx, vy).rewrite(Piecewise) == Piecewise((vx, vx >= vy),
(vy, True))
assert Min(va, vx, vy).rewrite(Piecewise) == Piecewise(
(va, (va <= vx) & (va <= vy)), (vx, vx <= vy), (vy, True))
def _eval_imageset(self, f):
# Cut out 0, perform image, add back in image of 0
if self.contains(0) == True:
return imageset(f, self - FiniteSet(0)) + imageset(f, FiniteSet(0))
from sympy.functions.elementary.miscellaneous import Min, Max
# TODO: manage left_open and right_open better in case of
# non-comparable left/right (e.g. Interval(x, y))
_left, _right = f(self.left), f(self.right)
left, right = Min(_left, _right), Max(_left, _right)
if _right == left: # switch happened
left_open, right_open = self.right_open, self.left_open
else:
left_open, right_open = self.left_open, self.right_open
return Interval(left, right, left_open, right_open)
def test_Max():
from sympy.abc import x, y, z
n = Symbol('n', negative=True)
n_ = Symbol('n_', negative=True)
nn = Symbol('nn', nonnegative=True)
nn_ = Symbol('nn_', nonnegative=True)
p = Symbol('p', positive=True)
p_ = Symbol('p_', positive=True)
np = Symbol('np', nonpositive=True)
np_ = Symbol('np_', nonpositive=True)
assert Max(5, 4) == 5
# lists
raises(ValueError, lambda: Max())
assert Max(x, y) == Max(y, x)
assert Max(x, y, z) == Max(z, y, x)
assert Max(x, Max(y, z)) == Max(z, y, x)
assert Max(x, Min(y, oo)) == Max(x, y)
assert Max(n, -oo, n_, p, 2) == Max(p, 2)
assert Max(n, -oo, n_, p) == p
assert Max(2, x, p, n, -oo, S.NegativeInfinity, n_, p, 2) == Max(2, x, p)
assert Max(0, x, 1, y) == Max(1, x, y)
assert Max(x, x + 1, x - 1) == 1 + x
assert Max(1000, 100, -100, x, p, n) == Max(p, x, 1000)
assert Max(cos(x), sin(x)) == Max(sin(x), cos(x))
assert Max(cos(x), sin(x)).subs(x, 1) == sin(1)
assert Max(cos(x), sin(x)).subs(x, S(1)/2) == cos(S(1)/2)
raises(ValueError, lambda: Max(cos(x), sin(x)).subs(x, I))
raises(ValueError, lambda: Max(I))
raises(ValueError, lambda: Max(I, x))
raises(ValueError, lambda: Max(S.ComplexInfinity, 1))
# interesting:
# Max(n, -oo, n_, p, 2) == Max(p, 2)
# True
# Max(n, -oo, n_, p, 1000) == Max(p, 1000)
# False
from sympy.functions.special.delta_functions import Heaviside
assert Max(1,x).diff(x) == Heaviside(x-1)
assert Max(x,1).diff(x) == Heaviside(x-1)
assert Max(x**2, 1+x, 1).diff(x) == 2*x*Heaviside(x**2 - Max(1,x+1)) \
+ Heaviside(x - Max(1,x**2) + 1)
a, b = Symbol('a', real=True), Symbol('b', real=True)
# a and b are both real, Max(a, b) should be real
assert Max(a, b).is_real
def test_issue_18770():
numpy = import_module('numpy')
if not numpy:
skip("numpy not installed.")
from sympy.functions.elementary.miscellaneous import (Max, Min)
from sympy.utilities.lambdify import lambdify
expr1 = Min(0.1 * x + 3, x + 1, 0.5 * x + 1)
func = lambdify(x, expr1, "numpy")
assert (func(numpy.linspace(0, 3, 3)) == [1.0, 1.75, 2.5]).all()
assert func(4) == 3
expr1 = Max(x**2, x**3)
func = lambdify(x, expr1, "numpy")
assert (func(numpy.linspace(-1, 2, 4)) == [1, 0, 1, 8]).all()
assert func(4) == 64
def __mul__(self, other):
if isinstance(other, Expr):
if isinstance(other, AccumBounds):
return AccumBounds(Min(Mul(self.min, other.min),
Mul(self.min, other.max),
Mul(self.max, other.min),
Mul(self.max, other.max)),
Max(Mul(self.min, other.min),
Mul(self.min, other.max),
Mul(self.max, other.min),
Mul(self.max, other.max)))
if other is S.Infinity:
if self.min.is_zero:
return AccumBounds(0, oo)
if self.max.is_zero:
return AccumBounds(-oo, 0)
if other is S.NegativeInfinity:
if self.min.is_zero:
return AccumBounds(-oo, 0)
if self.max.is_zero:
return AccumBounds(0, oo)
if other.is_real:
if other.is_zero:
if self == AccumBounds(-oo, oo):
return AccumBounds(-oo, oo)
if self.max is S.Infinity:
return AccumBounds(0, oo)
if self.min is S.NegativeInfinity:
return AccumBounds(-oo, 0)
return S.Zero
if other.is_positive:
return AccumBounds(
Mul(self.min, other),
Mul(self.max, other))
elif other.is_negative:
return AccumBounds(
Mul(self.max, other),
Mul(self.min, other))
if isinstance(other, Order):
return other
return Mul(self, other, evaluate=False)
return NotImplemented
def __mul__(self, other):
if self.args == (-oo, oo):
return self
if isinstance(other, Expr):
if isinstance(other, AccumBounds):
if other.args == (-oo, oo):
return other
v = set()
for a in self.args:
vi = other * a
for i in vi.args or (vi, ):
v.add(i)
return AccumBounds(Min(*v), Max(*v))
if other is S.Infinity:
if self.min.is_zero:
return AccumBounds(0, oo)
if self.max.is_zero:
return AccumBounds(-oo, 0)
if other is S.NegativeInfinity:
if self.min.is_zero:
return AccumBounds(-oo, 0)
if self.max.is_zero:
return AccumBounds(0, oo)
if other.is_extended_real:
if other.is_zero:
if self.max is S.Infinity:
return AccumBounds(0, oo)
if self.min is S.NegativeInfinity:
return AccumBounds(-oo, 0)
return S.Zero
if other.is_extended_positive:
return AccumBounds(Mul(self.min, other),
Mul(self.max, other))
elif other.is_extended_negative:
return AccumBounds(Mul(self.max, other),
Mul(self.min, other))
if isinstance(other, Order):
return other
return Mul(self, other, evaluate=False)
return NotImplemented
def minor(self):
"""Shorter axis of the ellipse (if it can be determined) else vradius.
Returns
=======
minor : number or expression
See Also
========
hradius, vradius, major
Examples
========
>>> from sympy import Point, Ellipse, Symbol
>>> p1 = Point(0, 0)
>>> e1 = Ellipse(p1, 3, 1)
>>> e1.minor
1
>>> a = Symbol('a')
>>> b = Symbol('b')
>>> Ellipse(p1, a, b).minor
b
>>> Ellipse(p1, b, a).minor
a
>>> m = Symbol('m')
>>> M = m + 1
>>> Ellipse(p1, m, M).minor
m
"""
rv = Min(*self.args[1:3])
if rv.func is Min:
return self.vradius
return rv
def __rdiv__(self, other):
if isinstance(other, Expr):
if other.is_real:
if other.is_zero:
return S.Zero
if S.Zero in self:
if self.min == S.Zero:
if other.is_positive:
return AccumBounds(Mul(other, 1/self.max), oo)
if other.is_negative:
return AccumBounds(-oo, Mul(other, 1/self.max))
if self.max == S.Zero:
if other.is_positive:
return AccumBounds(-oo, Mul(other, 1/self.min))
if other.is_negative:
return AccumBounds(Mul(other, 1/self.min), oo)
return AccumBounds(-oo, oo)
else:
return AccumBounds(Min(other/self.min, other/self.max),
Max(other/self.min, other/self.max))
return Mul(other, 1/self, evaluate=False)
else:
return NotImplemented
def __rtruediv__(self, other):
if isinstance(other, Expr):
if other.is_extended_real:
if other.is_zero:
return S.Zero
if (self.min.is_extended_nonpositive
and self.max.is_extended_nonnegative):
if self.min.is_zero:
if other.is_extended_positive:
return AccumBounds(Mul(other, 1 / self.max), oo)
if other.is_extended_negative:
return AccumBounds(-oo, Mul(other, 1 / self.max))
if self.max.is_zero:
if other.is_extended_positive:
return AccumBounds(-oo, Mul(other, 1 / self.min))
if other.is_extended_negative:
return AccumBounds(Mul(other, 1 / self.min), oo)
return AccumBounds(-oo, oo)
else:
return AccumBounds(Min(other / self.min, other / self.max),
Max(other / self.min, other / self.max))
return Mul(other, 1 / self, evaluate=False)
else:
return NotImplemented
def test_instantiation_evaluation():
from sympy.abc import v, w, x, y, z
assert Min(1, Max(2, x)) == 1
assert Max(3, Min(2, x)) == 3
assert Min(Max(x, y), Max(x, z)) == Max(x, Min(y, z))
assert set(Min(Max(w, x), Max(y, z)).args) == set([Max(w, x), Max(y, z)])
assert Min(Max(x, y), Max(x, z), w) == Min(w, Max(x, Min(y, z)))
A, B = Min, Max
for i in range(2):
assert A(x, B(x, y)) == x
assert A(x, B(y, A(x, w, z))) == A(x, B(y, A(w, z)))
A, B = B, A
assert Min(w, Max(x, y), Max(v, x, z)) == Min(w, Max(x, Min(y, Max(v, z))))
def test_issue_12638():
from sympy.abc import a, b, c, d
assert Min(a, b, c, Max(a, b)) == Min(a, b, c)
assert Min(a, b, Max(a, b, c)) == Min(a, b)
assert Min(a, b, Max(a, c)) == Min(a, b)
def test_Min():
from sympy.abc import x, y, z
n = Symbol('n', negative=True)
n_ = Symbol('n_', negative=True)
nn = Symbol('nn', nonnegative=True)
nn_ = Symbol('nn_', nonnegative=True)
p = Symbol('p', positive=True)
p_ = Symbol('p_', positive=True)
np = Symbol('np', nonpositive=True)
np_ = Symbol('np_', nonpositive=True)
r = Symbol('r', real=True)
assert Min(5, 4) == 4
assert Min(-oo, -oo) == -oo
assert Min(-oo, n) == -oo
assert Min(n, -oo) == -oo
assert Min(-oo, np) == -oo
assert Min(np, -oo) == -oo
assert Min(-oo, 0) == -oo
assert Min(0, -oo) == -oo
assert Min(-oo, nn) == -oo
assert Min(nn, -oo) == -oo
assert Min(-oo, p) == -oo
assert Min(p, -oo) == -oo
assert Min(-oo, oo) == -oo
assert Min(oo, -oo) == -oo
assert Min(n, n) == n
assert Min(n, np) == Min(n, np)
assert Min(np, n) == Min(np, n)
assert Min(n, 0) == n
assert Min(0, n) == n
assert Min(n, nn) == n
assert Min(nn, n) == n
assert Min(n, p) == n
assert Min(p, n) == n
assert Min(n, oo) == n
assert Min(oo, n) == n
assert Min(np, np) == np
assert Min(np, 0) == np
assert Min(0, np) == np
assert Min(np, nn) == np
assert Min(nn, np) == np
assert Min(np, p) == np
assert Min(p, np) == np
assert Min(np, oo) == np
assert Min(oo, np) == np
assert Min(0, 0) == 0
assert Min(0, nn) == 0
assert Min(nn, 0) == 0
assert Min(0, p) == 0
assert Min(p, 0) == 0
assert Min(0, oo) == 0
assert Min(oo, 0) == 0
assert Min(nn, nn) == nn
assert Min(nn, p) == Min(nn, p)
assert Min(p, nn) == Min(p, nn)
assert Min(nn, oo) == nn
assert Min(oo, nn) == nn
assert Min(p, p) == p
assert Min(p, oo) == p
assert Min(oo, p) == p
assert Min(oo, oo) == oo
assert Min(n, n_).func is Min
assert Min(nn, nn_).func is Min
assert Min(np, np_).func is Min
assert Min(p, p_).func is Min
# lists
raises(ValueError, lambda: Min())
assert Min(x, y) == Min(y, x)
assert Min(x, y, z) == Min(z, y, x)
assert Min(x, Min(y, z)) == Min(z, y, x)
assert Min(x, Max(y, -oo)) == Min(x, y)
assert Min(p, oo, n, p, p, p_) == n
assert Min(p_, n_, p) == n_
assert Min(n, oo, -7, p, p, 2) == Min(n, -7)
assert Min(2, x, p, n, oo, n_, p, 2, -2, -2) == Min(-2, x, n, n_)
assert Min(0, x, 1, y) == Min(0, x, y)
assert Min(1000, 100, -100, x, p, n) == Min(n, x, -100)
assert Min(cos(x), sin(x)) == Min(cos(x), sin(x))
assert Min(cos(x), sin(x)).subs(x, 1) == cos(1)
assert Min(cos(x), sin(x)).subs(x, S(1) / 2) == sin(S(1) / 2)
raises(ValueError, lambda: Min(cos(x), sin(x)).subs(x, I))
raises(ValueError, lambda: Min(I))
raises(ValueError, lambda: Min(I, x))
raises(ValueError, lambda: Min(S.ComplexInfinity, x))
assert Min(1, x).diff(x) == Heaviside(1 - x)
assert Min(x, 1).diff(x) == Heaviside(1 - x)
assert Min(0, -x, 1 - 2*x).diff(x) == -Heaviside(x + Min(0, -2*x + 1)) \
- 2*Heaviside(2*x + Min(0, -x) - 1)
# issue 7619
f = Function('f')
assert Min(1, 2 * Min(f(1), 2)) # doesn't fail
# issue 7233
e = Min(0, x)
assert e.evalf == e.n
assert e.n().args == (0, x)
# issue 8643
m = Min(n, p_, n_, r)
assert m.is_positive is False
assert m.is_nonnegative is False
assert m.is_negative is True
m = Min(p, p_)
assert m.is_positive is True
assert m.is_nonnegative is True
assert m.is_negative is False
m = Min(p, nn_, p_)
assert m.is_positive is None
assert m.is_nonnegative is True
assert m.is_negative is False
m = Min(nn, p, r)
assert m.is_positive is None
assert m.is_nonnegative is None
assert m.is_negative is None
def test_Max():
from sympy.abc import x, y, z
n = Symbol('n', negative=True)
n_ = Symbol('n_', negative=True)
nn = Symbol('nn', nonnegative=True)
nn_ = Symbol('nn_', nonnegative=True)
p = Symbol('p', positive=True)
p_ = Symbol('p_', positive=True)
np = Symbol('np', nonpositive=True)
np_ = Symbol('np_', nonpositive=True)
r = Symbol('r', real=True)
assert Max(5, 4) == 5
# lists
raises(ValueError, lambda: Max())
assert Max(x, y) == Max(y, x)
assert Max(x, y, z) == Max(z, y, x)
assert Max(x, Max(y, z)) == Max(z, y, x)
assert Max(x, Min(y, oo)) == Max(x, y)
assert Max(n, -oo, n_, p, 2) == Max(p, 2)
assert Max(n, -oo, n_, p) == p
assert Max(2, x, p, n, -oo, S.NegativeInfinity, n_, p, 2) == Max(2, x, p)
assert Max(0, x, 1, y) == Max(1, x, y)
assert Max(r, r + 1, r - 1) == 1 + r
assert Max(1000, 100, -100, x, p, n) == Max(p, x, 1000)
assert Max(cos(x), sin(x)) == Max(sin(x), cos(x))
assert Max(cos(x), sin(x)).subs(x, 1) == sin(1)
assert Max(cos(x), sin(x)).subs(x, S(1) / 2) == cos(S(1) / 2)
raises(ValueError, lambda: Max(cos(x), sin(x)).subs(x, I))
raises(ValueError, lambda: Max(I))
raises(ValueError, lambda: Max(I, x))
raises(ValueError, lambda: Max(S.ComplexInfinity, 1))
assert Max(n, -oo, n_, p, 2) == Max(p, 2)
assert Max(n, -oo, n_, p, 1000) == Max(p, 1000)
assert Max(1, x).diff(x) == Heaviside(x - 1)
assert Max(x, 1).diff(x) == Heaviside(x - 1)
assert Max(x**2, 1 + x, 1).diff(x) == \
2*x*Heaviside(x**2 - Max(1, x + 1)) \
+ Heaviside(x - Max(1, x**2) + 1)
e = Max(0, x)
assert e.evalf == e.n
assert e.n().args == (0, x)
# issue 8643
m = Max(p, p_, n, r)
assert m.is_positive is True
assert m.is_nonnegative is True
assert m.is_negative is False
m = Max(n, n_)
assert m.is_positive is False
assert m.is_nonnegative is False
assert m.is_negative is True
m = Max(n, n_, r)
assert m.is_positive is None
assert m.is_nonnegative is None
assert m.is_negative is None
m = Max(n, nn, r)
assert m.is_positive is None
assert m.is_nonnegative is True
assert m.is_negative is False
def test_sympy__functions__elementary__miscellaneous__Min():
from sympy.functions.elementary.miscellaneous import Min
assert _test_args(Min(x, 2))
def test_issue_21399():
from sympy.abc import a, b, c
assert Max(Min(a, b), Min(a, b, c)) == Min(a, b)
def _eval_imageset(self, f):
from sympy.functions.elementary.miscellaneous import Min, Max
from sympy.solvers import solve
from sympy.core.function import diff
from sympy.series import limit
from sympy.calculus.singularities import singularities
# TODO: handle piecewise defined functions
# TODO: handle functions with infinitely many solutions (eg, sin, tan)
# TODO: handle multivariate functions
expr = f.expr
if len(expr.free_symbols) > 1 or len(f.variables) != 1:
return
var = f.variables[0]
if not self.start.is_comparable or not self.end.is_comparable:
return
try:
sing = [
x for x in singularities(expr, var) if x.is_real and x in self
]
except NotImplementedError:
return
if self.left_open:
_start = limit(expr, var, self.start, dir="+")
elif self.start not in sing:
_start = f(self.start)
if self.right_open:
_end = limit(expr, var, self.end, dir="-")
elif self.end not in sing:
_end = f(self.end)
if len(sing) == 0:
solns = solve(diff(expr, var), var)
extr = [_start, _end
] + [f(x) for x in solns if x.is_real and x in self]
start, end = Min(*extr), Max(*extr)
left_open, right_open = False, False
if _start <= _end:
# the minimum or maximum value can occur simultaneously
# on both the edge of the interval and in some interior
# point
if start == _start and start not in solns:
left_open = self.left_open
if end == _end and end not in solns:
right_open = self.right_open
else:
if start == _end and start not in solns:
left_open = self.right_open
if end == _start and end not in solns:
right_open = self.left_open
return Interval(start, end, left_open, right_open)
else:
return imageset(f, Interval(self.start, sing[0],
self.left_open, True)) + \
Union(*[imageset(f, Interval(sing[i], sing[i + 1]), True, True)
for i in range(1, len(sing) - 1)]) + \
imageset(f, Interval(sing[-1], self.end, True, self.right_open))
def _print_Min(self, expr):
from sympy.functions.elementary.miscellaneous import Min
if len(expr.args) == 1:
return self._print(expr.args[0])
return "%smin(%s, %s)" % (self._ns, self._print(expr.args[0]),
self._print(Min(*expr.args[1:])))
def _intervals(self, sym):
"""Return a list of unique tuples, (a, b, e, i), where a and b
are the lower and upper bounds in which the expression e of
argument i in self is defined and a < b (when involving
numbers) or a <= b when involving symbols.
If there are any relationals not involving sym, or any
relational cannot be solved for sym, NotImplementedError is
raised. The calling routine should have removed such
relationals before calling this routine.
The evaluated conditions will be returned as ranges.
Discontinuous ranges will be returned separately with
identical expressions. The first condition that evaluates to
True will be returned as the last tuple with a, b = -oo, oo.
"""
from sympy.solvers.inequalities import _solve_inequality
from sympy.logic.boolalg import to_cnf, distribute_or_over_and
assert isinstance(self, Piecewise)
def _solve_relational(r):
if sym not in r.free_symbols:
nonsymfail(r)
rv = _solve_inequality(r, sym)
if isinstance(rv, Relational):
free = rv.args[1].free_symbols
if rv.args[0] != sym or sym in free:
raise NotImplementedError(
filldedent('''
Unable to solve relational
%s for %s.''' % (r, sym)))
if rv.rel_op == '==':
# this equality has been affirmed to have the form
# Eq(sym, rhs) where rhs is sym-free; it represents
# a zero-width interval which will be ignored
# whether it is an isolated condition or contained
# within an And or an Or
rv = S.false
elif rv.rel_op == '!=':
try:
rv = Or(sym < rv.rhs, sym > rv.rhs)
except TypeError:
# e.g. x != I ==> all real x satisfy
rv = S.true
elif rv == (S.NegativeInfinity < sym) & (sym < S.Infinity):
rv = S.true
return rv
def nonsymfail(cond):
raise NotImplementedError(
filldedent('''
A condition not involving
%s appeared: %s''' % (sym, cond)))
# make self canonical wrt Relationals
reps = dict([(r, _solve_relational(r))
for r in self.atoms(Relational)])
# process args individually so if any evaluate, their position
# in the original Piecewise will be known
args = [i.xreplace(reps) for i in self.args]
# precondition args
expr_cond = []
default = idefault = None
for i, (expr, cond) in enumerate(args):
if cond is S.false:
continue
elif cond is S.true:
default = expr
idefault = i
break
cond = to_cnf(cond)
if isinstance(cond, And):
cond = distribute_or_over_and(cond)
if isinstance(cond, Or):
expr_cond.extend([(i, expr, o) for o in cond.args
if not isinstance(o, Equality)])
elif cond is not S.false:
expr_cond.append((i, expr, cond))
# determine intervals represented by conditions
int_expr = []
for iarg, expr, cond in expr_cond:
if isinstance(cond, And):
lower = S.NegativeInfinity
upper = S.Infinity
for cond2 in cond.args:
if isinstance(cond2, Equality):
lower = upper # ignore
break
elif cond2.lts == sym:
upper = Min(cond2.gts, upper)
elif cond2.gts == sym:
lower = Max(cond2.lts, lower)
else:
nonsymfail(cond2) # should never get here
elif isinstance(cond, Relational):
lower, upper = cond.lts, cond.gts # part 1: initialize with givens
if cond.lts == sym: # part 1a: expand the side ...
lower = S.NegativeInfinity # e.g. x <= 0 ---> -oo <= 0
elif cond.gts == sym: # part 1a: ... that can be expanded
upper = S.Infinity # e.g. x >= 0 ---> oo >= 0
else:
nonsymfail(cond)
else:
raise NotImplementedError('unrecognized condition: %s' % cond)
lower, upper = lower, Max(lower, upper)
if (lower >= upper) is not S.true:
int_expr.append((lower, upper, expr, iarg))
if default is not None:
int_expr.append(
(S.NegativeInfinity, S.Infinity, default, idefault))
return list(uniq(int_expr))
def __pow__(self, other):
from sympy.functions.elementary.miscellaneous import real_root
if isinstance(other, Expr):
if other is S.Infinity:
if self.min.is_nonnegative:
if self.max < 1:
return S.Zero
if self.min > 1:
return S.Infinity
return AccumBounds(0, oo)
elif self.max.is_negative:
if self.min > -1:
return S.Zero
if self.max < -1:
return FiniteSet(-oo, oo)
return AccumBounds(-oo, oo)
else:
if self.min > -1:
if self.max < 1:
return S.Zero
return AccumBounds(0, oo)
return AccumBounds(-oo, oo)
if other is S.NegativeInfinity:
return (1/self)**oo
if other.is_real and other.is_number:
if other.is_zero:
return S.One
if other.is_Integer:
if self.min.is_positive:
return AccumBounds(Min(self.min**other, self.max**other),
Max(self.min**other, self.max**other))
elif self.max.is_negative:
return AccumBounds(Min(self.max**other, self.min**other),
Max(self.max**other, self.min**other))
if other % 2 == 0:
if other.is_negative:
if self.min.is_zero:
return AccumBounds(self.max**other, oo)
if self.max.is_zero:
return AccumBounds(self.min**other, oo)
return AccumBounds(0, oo)
return AccumBounds(S.Zero,
Max(self.min**other, self.max**other))
else:
if other.is_negative:
if self.min.is_zero:
return AccumBounds(self.max**other, oo)
if self.max.is_zero:
return AccumBounds(-oo, self.min**other)
return AccumBounds(-oo, oo)
return AccumBounds(self.min**other, self.max**other)
num, den = other.as_numer_denom()
if num == S(1):
if den % 2 == 0:
if S.Zero in self:
if self.min.is_negative:
return AccumBounds(0, real_root(self.max, den))
return AccumBounds(real_root(self.min, den),
real_root(self.max, den))
num_pow = self**num
return num_pow**(1/den)
return Pow(self, other, evaluate=False)
return NotImplemented
def test_Min():
from sympy.abc import x, y, z
n = Symbol('n', negative=True)
n_ = Symbol('n_', negative=True)
nn = Symbol('nn', nonnegative=True)
nn_ = Symbol('nn_', nonnegative=True)
p = Symbol('p', positive=True)
p_ = Symbol('p_', positive=True)
np = Symbol('np', nonpositive=True)
np_ = Symbol('np_', nonpositive=True)
r = Symbol('r', real=True)
assert Min(5, 4) == 4
assert Min(-oo, -oo) == -oo
assert Min(-oo, n) == -oo
assert Min(n, -oo) == -oo
assert Min(-oo, np) == -oo
assert Min(np, -oo) == -oo
assert Min(-oo, 0) == -oo
assert Min(0, -oo) == -oo
assert Min(-oo, nn) == -oo
assert Min(nn, -oo) == -oo
assert Min(-oo, p) == -oo
assert Min(p, -oo) == -oo
assert Min(-oo, oo) == -oo
assert Min(oo, -oo) == -oo
assert Min(n, n) == n
assert Min(n, np) == Min(n, np)
assert Min(np, n) == Min(np, n)
assert Min(n, 0) == n
assert Min(0, n) == n
assert Min(n, nn) == n
assert Min(nn, n) == n
assert Min(n, p) == n
assert Min(p, n) == n
assert Min(n, oo) == n
assert Min(oo, n) == n
assert Min(np, np) == np
assert Min(np, 0) == np
assert Min(0, np) == np
assert Min(np, nn) == np
assert Min(nn, np) == np
assert Min(np, p) == np
assert Min(p, np) == np
assert Min(np, oo) == np
assert Min(oo, np) == np
assert Min(0, 0) == 0
assert Min(0, nn) == 0
assert Min(nn, 0) == 0
assert Min(0, p) == 0
assert Min(p, 0) == 0
assert Min(0, oo) == 0
assert Min(oo, 0) == 0
assert Min(nn, nn) == nn
assert Min(nn, p) == Min(nn, p)
assert Min(p, nn) == Min(p, nn)
assert Min(nn, oo) == nn
assert Min(oo, nn) == nn
assert Min(p, p) == p
assert Min(p, oo) == p
assert Min(oo, p) == p
assert Min(oo, oo) == oo
assert Min(n, n_).func is Min
assert Min(nn, nn_).func is Min
assert Min(np, np_).func is Min
assert Min(p, p_).func is Min
# lists
raises(ValueError, lambda: Min())
assert Min(x, y) == Min(y, x)
assert Min(x, y, z) == Min(z, y, x)
assert Min(x, Min(y, z)) == Min(z, y, x)
assert Min(x, Max(y, -oo)) == Min(x, y)
assert Min(p, oo, n, p, p, p_) == n
assert Min(p_, n_, p) == n_
assert Min(n, oo, -7, p, p, 2) == Min(n, -7)
assert Min(2, x, p, n, oo, n_, p, 2, -2, -2) == Min(-2, x, n, n_)
assert Min(0, x, 1, y) == Min(0, x, y)
assert Min(1000, 100, -100, x, p, n) == Min(n, x, -100)
assert Min(cos(x), sin(x)) == Min(cos(x), sin(x))
assert Min(cos(x), sin(x)).subs(x, 1) == cos(1)
assert Min(cos(x), sin(x)).subs(x, S(1)/2) == sin(S(1)/2)
raises(ValueError, lambda: Min(cos(x), sin(x)).subs(x, I))
raises(ValueError, lambda: Min(I))
raises(ValueError, lambda: Min(I, x))
raises(ValueError, lambda: Min(S.ComplexInfinity, x))
assert Min(1, x).diff(x) == Heaviside(1 - x)
assert Min(x, 1).diff(x) == Heaviside(1 - x)
assert Min(0, -x, 1 - 2*x).diff(x) == -Heaviside(x + Min(0, -2*x + 1)) \
- 2*Heaviside(2*x + Min(0, -x) - 1)
a, b = Symbol('a', real=True), Symbol('b', real=True)
# a and b are both real, Min(a, b) should be real
assert Min(a, b).is_real
# issue 7619
f = Function('f')
assert Min(1, 2*Min(f(1), 2)) # doesn't fail
# issue 7233
e = Min(0, x)
assert e.evalf == e.n
assert e.n().args == (0, x)
# issue 8643
m = Min(n, p_, n_, r)
assert m.is_positive is False
assert m.is_nonnegative is False
assert m.is_negative is True
m = Min(p, p_)
assert m.is_positive is True
assert m.is_nonnegative is True
assert m.is_negative is False
m = Min(p, nn_, p_)
assert m.is_positive is None
assert m.is_nonnegative is True
assert m.is_negative is False
m = Min(nn, p, r)
assert m.is_positive is None
assert m.is_nonnegative is None
assert m.is_negative is None
def sum_limit(r, w, n0, di, i0, index=0):
return slice(
di, Max(0, -(i0 // r[index])),
Min(-1 // r[index] + w.shape[index], (n0 - i0 - 1) // r[index]) + 1)
def _inf(self):
# We use Min so that sup is meaningful in combination with symbolic
# interval end points.
from sympy.functions.elementary.miscellaneous import Min
return Min(*[set.inf for set in self.args])
def _sort_expr_cond(self, sym, a, b, targetcond=None):
"""Determine what intervals the expr, cond pairs affect.
1) If cond is True, then log it as default
1.1) Currently if cond can't be evaluated, throw NotImplementedError.
2) For each inequality, if previous cond defines part of the interval
update the new conds interval.
- eg x < 1, x < 3 -> [oo,1],[1,3] instead of [oo,1],[oo,3]
3) Sort the intervals to make it easier to find correct exprs
Under normal use, we return the expr,cond pairs in increasing order
along the real axis corresponding to the symbol sym. If targetcond
is given, we return a list of (lowerbound, upperbound) pairs for
this condition."""
default = None
int_expr = []
expr_cond = []
or_cond = False
or_intervals = []
for expr, cond in self.args:
if isinstance(cond, Or):
for cond2 in sorted(cond.args, key=default_sort_key):
expr_cond.append((expr, cond2))
else:
expr_cond.append((expr, cond))
if cond is True:
break
for expr, cond in expr_cond:
if cond is True:
default = expr
break
elif isinstance(cond, Equality):
continue
elif isinstance(cond, And):
lower = S.NegativeInfinity
upper = S.Infinity
for cond2 in cond.args:
if cond2.lts.has(sym):
upper = Min(cond2.gts, upper)
elif cond2.gts.has(sym):
lower = Max(cond2.lts, lower)
else:
lower, upper = cond.lts, cond.gts # part 1: initialize with givens
if cond.lts.has(sym): # part 1a: expand the side ...
lower = S.NegativeInfinity # e.g. x <= 0 ---> -oo <= 0
elif cond.gts.has(sym): # part 1a: ... that can be expanded
upper = S.Infinity # e.g. x >= 0 ---> oo >= 0
else:
raise NotImplementedError(
"Unable to handle interval evaluation of expression.")
# part 1b: Reduce (-)infinity to what was passed in.
lower, upper = Max(a, lower), Min(b, upper)
for n in xrange(len(int_expr)):
# Part 2: remove any interval overlap. For any conflicts, the
# iterval already there wins, and the incoming interval updates
# its bounds accordingly.
if self.__eval_cond(lower < int_expr[n][1]) and \
self.__eval_cond(lower >= int_expr[n][0]):
lower = int_expr[n][1]
elif len(int_expr[n][1].free_symbols) and \
self.__eval_cond(lower >= int_expr[n][0]):
if self.__eval_cond(lower == int_expr[n][0]):
lower = int_expr[n][1]
else:
int_expr[n][1] = Min(lower, int_expr[n][1])
elif len(int_expr[n][1].free_symbols) and \
lower < int_expr[n][0] is not True:
upper = Min(upper, int_expr[n][0])
elif self.__eval_cond(upper > int_expr[n][0]) and \
self.__eval_cond(upper <= int_expr[n][1]):
upper = int_expr[n][0]
elif len(int_expr[n][0].free_symbols) and \
self.__eval_cond(upper < int_expr[n][1]):
int_expr[n][0] = Max(upper, int_expr[n][0])
if self.__eval_cond(
lower >= upper) is not True: # Is it still an interval?
int_expr.append([lower, upper, expr])
if cond is targetcond:
return [(lower, upper)]
elif isinstance(targetcond, Or) and cond in targetcond.args:
or_cond = Or(or_cond, cond)
or_intervals.append((lower, upper))
if or_cond == targetcond:
or_intervals.sort(key=lambda x: x[0])
return or_intervals
int_expr.sort(key=lambda x: x[1].sort_key()
if x[1].is_number else S.NegativeInfinity.sort_key())
int_expr.sort(key=lambda x: x[0].sort_key()
if x[0].is_number else S.Infinity.sort_key())
from sympy.functions.elementary.miscellaneous import MinMaxBase
for n in xrange(len(int_expr)):
if len(int_expr[n][0].free_symbols) or len(
int_expr[n][1].free_symbols):
if isinstance(int_expr[n][1], Min) or int_expr[n][1] == b:
newval = Min(*int_expr[n][:-1])
if n > 0 and int_expr[n][0] == int_expr[n - 1][1]:
int_expr[n - 1][1] = newval
int_expr[n][0] = newval
else:
newval = Max(*int_expr[n][:-1])
if n < len(int_expr) - 1 and int_expr[n][1] == int_expr[
n + 1][0]:
int_expr[n + 1][0] = newval
int_expr[n][1] = newval
# Add holes to list of intervals if there is a default value,
# otherwise raise a ValueError.
holes = []
curr_low = a
for int_a, int_b, expr in int_expr:
if (curr_low < int_a) is True:
holes.append([curr_low, Min(b, int_a), default])
elif (curr_low >= int_a) is not True:
holes.append([curr_low, Min(b, int_a), default])
curr_low = Min(b, int_b)
if (curr_low < b) is True:
holes.append([Min(b, curr_low), b, default])
elif (curr_low >= b) is not True:
holes.append([Min(b, curr_low), b, default])
if holes and default is not None:
int_expr.extend(holes)
if targetcond is True:
return [(h[0], h[1]) for h in holes]
elif holes and default is None:
raise ValueError("Called interval evaluation over piecewise "
"function on undefined intervals %s" %
", ".join([str((h[0], h[1])) for h in holes]))
return int_expr
def _inf(self):
from sympy.functions.elementary.miscellaneous import Min
return Min(*self)
def _LUdecomposition_Simple(M, iszerofunc=_iszero, simpfunc=None,
rankcheck=False):
r"""Compute the PLU decomposition of the matrix.
Parameters
==========
rankcheck : bool, optional
Determines if this function should detect the rank
deficiency of the matrixis and should raise a
``ValueError``.
iszerofunc : function, optional
A function which determines if a given expression is zero.
The function should be a callable that takes a single
sympy expression and returns a 3-valued boolean value
``True``, ``False``, or ``None``.
It is internally used by the pivot searching algorithm.
See the notes section for a more information about the
pivot searching algorithm.
simpfunc : function or None, optional
A function that simplifies the input.
If this is specified as a function, this function should be
a callable that takes a single sympy expression and returns
an another sympy expression that is algebraically
equivalent.
If ``None``, it indicates that the pivot search algorithm
should not attempt to simplify any candidate pivots.
It is internally used by the pivot searching algorithm.
See the notes section for a more information about the
pivot searching algorithm.
Returns
=======
(lu, row_swaps) : (Matrix, list)
If the original matrix is a $m, n$ matrix:
*lu* is a $m, n$ matrix, which contains result of the
decomposition in a compresed form. See the notes section
to see how the matrix is compressed.
*row_swaps* is a $m$-element list where each element is a
pair of row exchange indices.
``A = (L*U).permute_backward(perm)``, and the row
permutation matrix $P$ from the formula $P A = L U$ can be
computed by ``P=eye(A.row).permute_forward(perm)``.
Raises
======
ValueError
Raised if ``rankcheck=True`` and the matrix is found to
be rank deficient during the computation.
Notes
=====
About the PLU decomposition:
PLU decomposition is a generalization of a LU decomposition
which can be extended for rank-deficient matrices.
It can further be generalized for non-square matrices, and this
is the notation that SymPy is using.
PLU decomposition is a decomposition of a $m, n$ matrix $A$ in
the form of $P A = L U$ where
* $L$ is a $m, m$ lower triangular matrix with unit diagonal
entries.
* $U$ is a $m, n$ upper triangular matrix.
* $P$ is a $m, m$ permutation matrix.
So, for a square matrix, the decomposition would look like:
.. math::
L = \begin{bmatrix}
1 & 0 & 0 & \cdots & 0 \\
L_{1, 0} & 1 & 0 & \cdots & 0 \\
L_{2, 0} & L_{2, 1} & 1 & \cdots & 0 \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
L_{n-1, 0} & L_{n-1, 1} & L_{n-1, 2} & \cdots & 1
\end{bmatrix}
.. math::
U = \begin{bmatrix}
U_{0, 0} & U_{0, 1} & U_{0, 2} & \cdots & U_{0, n-1} \\
0 & U_{1, 1} & U_{1, 2} & \cdots & U_{1, n-1} \\
0 & 0 & U_{2, 2} & \cdots & U_{2, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
0 & 0 & 0 & \cdots & U_{n-1, n-1}
\end{bmatrix}
And for a matrix with more rows than the columns,
the decomposition would look like:
.. math::
L = \begin{bmatrix}
1 & 0 & 0 & \cdots & 0 & 0 & \cdots & 0 \\
L_{1, 0} & 1 & 0 & \cdots & 0 & 0 & \cdots & 0 \\
L_{2, 0} & L_{2, 1} & 1 & \cdots & 0 & 0 & \cdots & 0 \\
\vdots & \vdots & \vdots & \ddots & \vdots & \vdots & \ddots
& \vdots \\
L_{n-1, 0} & L_{n-1, 1} & L_{n-1, 2} & \cdots & 1 & 0
& \cdots & 0 \\
L_{n, 0} & L_{n, 1} & L_{n, 2} & \cdots & L_{n, n-1} & 1
& \cdots & 0 \\
\vdots & \vdots & \vdots & \ddots & \vdots & \vdots
& \ddots & \vdots \\
L_{m-1, 0} & L_{m-1, 1} & L_{m-1, 2} & \cdots & L_{m-1, n-1}
& 0 & \cdots & 1 \\
\end{bmatrix}
.. math::
U = \begin{bmatrix}
U_{0, 0} & U_{0, 1} & U_{0, 2} & \cdots & U_{0, n-1} \\
0 & U_{1, 1} & U_{1, 2} & \cdots & U_{1, n-1} \\
0 & 0 & U_{2, 2} & \cdots & U_{2, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
0 & 0 & 0 & \cdots & U_{n-1, n-1} \\
0 & 0 & 0 & \cdots & 0 \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
0 & 0 & 0 & \cdots & 0
\end{bmatrix}
Finally, for a matrix with more columns than the rows, the
decomposition would look like:
.. math::
L = \begin{bmatrix}
1 & 0 & 0 & \cdots & 0 \\
L_{1, 0} & 1 & 0 & \cdots & 0 \\
L_{2, 0} & L_{2, 1} & 1 & \cdots & 0 \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
L_{m-1, 0} & L_{m-1, 1} & L_{m-1, 2} & \cdots & 1
\end{bmatrix}
.. math::
U = \begin{bmatrix}
U_{0, 0} & U_{0, 1} & U_{0, 2} & \cdots & U_{0, m-1}
& \cdots & U_{0, n-1} \\
0 & U_{1, 1} & U_{1, 2} & \cdots & U_{1, m-1}
& \cdots & U_{1, n-1} \\
0 & 0 & U_{2, 2} & \cdots & U_{2, m-1}
& \cdots & U_{2, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots
& \cdots & \vdots \\
0 & 0 & 0 & \cdots & U_{m-1, m-1}
& \cdots & U_{m-1, n-1} \\
\end{bmatrix}
About the compressed LU storage:
The results of the decomposition are often stored in compressed
forms rather than returning $L$ and $U$ matrices individually.
It may be less intiuitive, but it is commonly used for a lot of
numeric libraries because of the efficiency.
The storage matrix is defined as following for this specific
method:
* The subdiagonal elements of $L$ are stored in the subdiagonal
portion of $LU$, that is $LU_{i, j} = L_{i, j}$ whenever
$i > j$.
* The elements on the diagonal of $L$ are all 1, and are not
explicitly stored.
* $U$ is stored in the upper triangular portion of $LU$, that is
$LU_{i, j} = U_{i, j}$ whenever $i <= j$.
* For a case of $m > n$, the right side of the $L$ matrix is
trivial to store.
* For a case of $m < n$, the below side of the $U$ matrix is
trivial to store.
So, for a square matrix, the compressed output matrix would be:
.. math::
LU = \begin{bmatrix}
U_{0, 0} & U_{0, 1} & U_{0, 2} & \cdots & U_{0, n-1} \\
L_{1, 0} & U_{1, 1} & U_{1, 2} & \cdots & U_{1, n-1} \\
L_{2, 0} & L_{2, 1} & U_{2, 2} & \cdots & U_{2, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
L_{n-1, 0} & L_{n-1, 1} & L_{n-1, 2} & \cdots & U_{n-1, n-1}
\end{bmatrix}
For a matrix with more rows than the columns, the compressed
output matrix would be:
.. math::
LU = \begin{bmatrix}
U_{0, 0} & U_{0, 1} & U_{0, 2} & \cdots & U_{0, n-1} \\
L_{1, 0} & U_{1, 1} & U_{1, 2} & \cdots & U_{1, n-1} \\
L_{2, 0} & L_{2, 1} & U_{2, 2} & \cdots & U_{2, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
L_{n-1, 0} & L_{n-1, 1} & L_{n-1, 2} & \cdots
& U_{n-1, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots \\
L_{m-1, 0} & L_{m-1, 1} & L_{m-1, 2} & \cdots
& L_{m-1, n-1} \\
\end{bmatrix}
For a matrix with more columns than the rows, the compressed
output matrix would be:
.. math::
LU = \begin{bmatrix}
U_{0, 0} & U_{0, 1} & U_{0, 2} & \cdots & U_{0, m-1}
& \cdots & U_{0, n-1} \\
L_{1, 0} & U_{1, 1} & U_{1, 2} & \cdots & U_{1, m-1}
& \cdots & U_{1, n-1} \\
L_{2, 0} & L_{2, 1} & U_{2, 2} & \cdots & U_{2, m-1}
& \cdots & U_{2, n-1} \\
\vdots & \vdots & \vdots & \ddots & \vdots
& \cdots & \vdots \\
L_{m-1, 0} & L_{m-1, 1} & L_{m-1, 2} & \cdots & U_{m-1, m-1}
& \cdots & U_{m-1, n-1} \\
\end{bmatrix}
About the pivot searching algorithm:
When a matrix contains symbolic entries, the pivot search algorithm
differs from the case where every entry can be categorized as zero or
nonzero.
The algorithm searches column by column through the submatrix whose
top left entry coincides with the pivot position.
If it exists, the pivot is the first entry in the current search
column that iszerofunc guarantees is nonzero.
If no such candidate exists, then each candidate pivot is simplified
if simpfunc is not None.
The search is repeated, with the difference that a candidate may be
the pivot if ``iszerofunc()`` cannot guarantee that it is nonzero.
In the second search the pivot is the first candidate that
iszerofunc can guarantee is nonzero.
If no such candidate exists, then the pivot is the first candidate
for which iszerofunc returns None.
If no such candidate exists, then the search is repeated in the next
column to the right.
The pivot search algorithm differs from the one in ``rref()``, which
relies on ``_find_reasonable_pivot()``.
Future versions of ``LUdecomposition_simple()`` may use
``_find_reasonable_pivot()``.
See Also
========
sympy.matrices.matrices.MatrixBase.LUdecomposition
LUdecompositionFF
LUsolve
"""
if rankcheck:
# https://github.com/sympy/sympy/issues/9796
pass
if M.rows == 0 or M.cols == 0:
# Define LU decomposition of a matrix with no entries as a matrix
# of the same dimensions with all zero entries.
return M.zeros(M.rows, M.cols), []
dps = _get_intermediate_simp()
lu = M.as_mutable()
row_swaps = []
pivot_col = 0
for pivot_row in range(0, lu.rows - 1):
# Search for pivot. Prefer entry that iszeropivot determines
# is nonzero, over entry that iszeropivot cannot guarantee
# is zero.
# XXX ``_find_reasonable_pivot`` uses slow zero testing. Blocked by bug #10279
# Future versions of LUdecomposition_simple can pass iszerofunc and simpfunc
# to _find_reasonable_pivot().
# In pass 3 of _find_reasonable_pivot(), the predicate in ``if x.equals(S.Zero):``
# calls sympy.simplify(), and not the simplification function passed in via
# the keyword argument simpfunc.
iszeropivot = True
while pivot_col != M.cols and iszeropivot:
sub_col = (lu[r, pivot_col] for r in range(pivot_row, M.rows))
pivot_row_offset, pivot_value, is_assumed_non_zero, ind_simplified_pairs =\
_find_reasonable_pivot_naive(sub_col, iszerofunc, simpfunc)
iszeropivot = pivot_value is None
if iszeropivot:
# All candidate pivots in this column are zero.
# Proceed to next column.
pivot_col += 1
if rankcheck and pivot_col != pivot_row:
# All entries including and below the pivot position are
# zero, which indicates that the rank of the matrix is
# strictly less than min(num rows, num cols)
# Mimic behavior of previous implementation, by throwing a
# ValueError.
raise ValueError("Rank of matrix is strictly less than"
" number of rows or columns."
" Pass keyword argument"
" rankcheck=False to compute"
" the LU decomposition of this matrix.")
candidate_pivot_row = None if pivot_row_offset is None else pivot_row + pivot_row_offset
if candidate_pivot_row is None and iszeropivot:
# If candidate_pivot_row is None and iszeropivot is True
# after pivot search has completed, then the submatrix
# below and to the right of (pivot_row, pivot_col) is
# all zeros, indicating that Gaussian elimination is
# complete.
return lu, row_swaps
# Update entries simplified during pivot search.
for offset, val in ind_simplified_pairs:
lu[pivot_row + offset, pivot_col] = val
if pivot_row != candidate_pivot_row:
# Row swap book keeping:
# Record which rows were swapped.
# Update stored portion of L factor by multiplying L on the
# left and right with the current permutation.
# Swap rows of U.
row_swaps.append([pivot_row, candidate_pivot_row])
# Update L.
lu[pivot_row, 0:pivot_row], lu[candidate_pivot_row, 0:pivot_row] = \
lu[candidate_pivot_row, 0:pivot_row], lu[pivot_row, 0:pivot_row]
# Swap pivot row of U with candidate pivot row.
lu[pivot_row, pivot_col:lu.cols], lu[candidate_pivot_row, pivot_col:lu.cols] = \
lu[candidate_pivot_row, pivot_col:lu.cols], lu[pivot_row, pivot_col:lu.cols]
# Introduce zeros below the pivot by adding a multiple of the
# pivot row to a row under it, and store the result in the
# row under it.
# Only entries in the target row whose index is greater than
# start_col may be nonzero.
start_col = pivot_col + 1
for row in range(pivot_row + 1, lu.rows):
# Store factors of L in the subcolumn below
# (pivot_row, pivot_row).
lu[row, pivot_row] = \
dps(lu[row, pivot_col]/lu[pivot_row, pivot_col])
# Form the linear combination of the pivot row and the current
# row below the pivot row that zeros the entries below the pivot.
# Employing slicing instead of a loop here raises
# NotImplementedError: Cannot add Zero to MutableSparseMatrix
# in sympy/matrices/tests/test_sparse.py.
# c = pivot_row + 1 if pivot_row == pivot_col else pivot_col
for c in range(start_col, lu.cols):
lu[row, c] = dps(lu[row, c] - lu[row, pivot_row]*lu[pivot_row, c])
if pivot_row != pivot_col:
# matrix rank < min(num rows, num cols),
# so factors of L are not stored directly below the pivot.
# These entries are zero by construction, so don't bother
# computing them.
for row in range(pivot_row + 1, lu.rows):
lu[row, pivot_col] = M.zero
pivot_col += 1
if pivot_col == lu.cols:
# All candidate pivots are zero implies that Gaussian
# elimination is complete.
return lu, row_swaps
if rankcheck:
if iszerofunc(
lu[Min(lu.rows, lu.cols) - 1, Min(lu.rows, lu.cols) - 1]):
raise ValueError("Rank of matrix is strictly less than"
" number of rows or columns."
" Pass keyword argument"
" rankcheck=False to compute"
" the LU decomposition of this matrix.")
return lu, row_swaps
def __pow__(self, other):
if isinstance(other, Expr):
if other is S.Infinity:
if self.min.is_extended_nonnegative:
if self.max < 1:
return S.Zero
if self.min > 1:
return S.Infinity
return AccumBounds(0, oo)
elif self.max.is_extended_negative:
if self.min > -1:
return S.Zero
if self.max < -1:
return zoo
return S.NaN
else:
if self.min > -1:
if self.max < 1:
return S.Zero
return AccumBounds(0, oo)
return AccumBounds(-oo, oo)
if other is S.NegativeInfinity:
return (1 / self)**oo
# generically true
if (self.max - self.min).is_nonnegative:
# well defined
if self.min.is_nonnegative:
# no 0 to worry about
if other.is_nonnegative:
# no infinity to worry about
return self.func(self.min**other, self.max**other)
if other.is_zero:
return S.One # x**0 = 1
if other.is_Integer or other.is_integer:
if self.min.is_extended_positive:
return AccumBounds(Min(self.min**other, self.max**other),
Max(self.min**other, self.max**other))
elif self.max.is_extended_negative:
return AccumBounds(Min(self.max**other, self.min**other),
Max(self.max**other, self.min**other))
if other % 2 == 0:
if other.is_extended_negative:
if self.min.is_zero:
return AccumBounds(self.max**other, oo)
if self.max.is_zero:
return AccumBounds(self.min**other, oo)
return AccumBounds(0, oo)
return AccumBounds(S.Zero,
Max(self.min**other, self.max**other))
elif other % 2 == 1:
if other.is_extended_negative:
if self.min.is_zero:
return AccumBounds(self.max**other, oo)
if self.max.is_zero:
return AccumBounds(-oo, self.min**other)
return AccumBounds(-oo, oo)
return AccumBounds(self.min**other, self.max**other)
# non-integer exponent
# 0**neg or neg**frac yields complex
if (other.is_number or other.is_rational) and (
self.min.is_extended_nonnegative or
(other.is_extended_nonnegative
and self.min.is_extended_nonnegative)):
num, den = other.as_numer_denom()
if num is S.One:
return AccumBounds(*[i**(1 / den) for i in self.args])
elif den is not S.One: # e.g. if other is not Float
return (self**num)**(1 / den) # ok for non-negative base
if isinstance(other, AccumBounds):
if (self.min.is_extended_positive
or self.min.is_extended_nonnegative
and other.min.is_extended_nonnegative):
p = [self**i for i in other.args]
if not any(i.is_Pow for i in p):
a = [j for i in p for j in i.args or (i, )]
try:
return self.func(min(a), max(a))
except TypeError: # can't sort
pass
return Pow(self, other, evaluate=False)
return NotImplemented
def test_Min_Max():
# see gh-10375
assert lambdify((x, y, z), Min(x, y, z))(1, 2, 3) == 1
assert lambdify((x, y, z), Max(x, y, z))(1, 2, 3) == 3
def limit_seq(expr, n=None, trials=5):
"""Finds the limit of a sequence as index ``n`` tends to infinity.
Parameters
==========
expr : Expr
SymPy expression for the ``n-th`` term of the sequence
n : Symbol, optional
The index of the sequence, an integer that tends to positive
infinity. If None, inferred from the expression unless it has
multiple symbols.
trials: int, optional
The algorithm is highly recursive. ``trials`` is a safeguard from
infinite recursion in case the limit is not easily computed by the
algorithm. Try increasing ``trials`` if the algorithm returns ``None``.
Admissible Terms
================
The algorithm is designed for sequences built from rational functions,
indefinite sums, and indefinite products over an indeterminate n. Terms of
alternating sign are also allowed, but more complex oscillatory behavior is
not supported.
Examples
========
>>> from sympy import limit_seq, Sum, binomial
>>> from sympy.abc import n, k, m
>>> limit_seq((5*n**3 + 3*n**2 + 4) / (3*n**3 + 4*n - 5), n)
5/3
>>> limit_seq(binomial(2*n, n) / Sum(binomial(2*k, k), (k, 1, n)), n)
3/4
>>> limit_seq(Sum(k**2 * Sum(2**m/m, (m, 1, k)), (k, 1, n)) / (2**n*n), n)
4
See Also
========
sympy.series.limitseq.dominant
References
==========
.. [1] Computing Limits of Sequences - Manuel Kauers
"""
from sympy.concrete.summations import Sum
if n is None:
free = expr.free_symbols
if len(free) == 1:
n = free.pop()
elif not free:
return expr
else:
raise ValueError("Expression has more than one variable. "
"Please specify a variable.")
elif n not in expr.free_symbols:
return expr
expr = expr.rewrite(fibonacci, S.GoldenRatio)
expr = expr.rewrite(factorial, subfactorial, gamma)
n_ = Dummy("n", integer=True, positive=True)
n1 = Dummy("n", odd=True, positive=True)
n2 = Dummy("n", even=True, positive=True)
# If there is a negative term raised to a power involving n, or a
# trigonometric function, then consider even and odd n separately.
powers = (p.as_base_exp() for p in expr.atoms(Pow))
if (any(b.is_negative and e.has(n) for b, e in powers)
or expr.has(cos, sin)):
L1 = _limit_seq(expr.xreplace({n: n1}), n1, trials)
if L1 is not None:
L2 = _limit_seq(expr.xreplace({n: n2}), n2, trials)
if L1 != L2:
if L1.is_comparable and L2.is_comparable:
return AccumulationBounds(Min(L1, L2), Max(L1, L2))
else:
return None
else:
L1 = _limit_seq(expr.xreplace({n: n_}), n_, trials)
if L1 is not None:
return L1
else:
if expr.is_Add:
limits = [limit_seq(term, n, trials) for term in expr.args]
if any(result is None for result in limits):
return None
else:
return Add(*limits)
# Maybe the absolute value is easier to deal with (though not if
# it has a Sum). If it tends to 0, the limit is 0.
elif not expr.has(Sum):
lim = _limit_seq(Abs(expr.xreplace({n: n_})), n_, trials)
if lim is not None and lim.is_zero:
return S.Zero
def _set_function(f, x): # noqa:F811
from sympy.functions.elementary.miscellaneous import Min, Max
from sympy.solvers.solveset import solveset
from sympy.core.function import diff, Lambda
from sympy.series import limit
from sympy.calculus.singularities import singularities
from sympy.sets import Complement
# TODO: handle functions with infinitely many solutions (eg, sin, tan)
# TODO: handle multivariate functions
expr = f.expr
if len(expr.free_symbols) > 1 or len(f.variables) != 1:
return
var = f.variables[0]
if not var.is_real:
if expr.subs(var, Dummy(real=True)).is_real is False:
return
if expr.is_Piecewise:
result = S.EmptySet
domain_set = x
for (p_expr, p_cond) in expr.args:
if p_cond is true:
intrvl = domain_set
else:
intrvl = p_cond.as_set()
intrvl = Intersection(domain_set, intrvl)
if p_expr.is_Number:
image = FiniteSet(p_expr)
else:
image = imageset(Lambda(var, p_expr), intrvl)
result = Union(result, image)
# remove the part which has been `imaged`
domain_set = Complement(domain_set, intrvl)
if domain_set is S.EmptySet:
break
return result
if not x.start.is_comparable or not x.end.is_comparable:
return
try:
sing = [i for i in singularities(expr, var) if i.is_real and i in x]
except NotImplementedError:
return
if x.left_open:
_start = limit(expr, var, x.start, dir="+")
elif x.start not in sing:
_start = f(x.start)
if x.right_open:
_end = limit(expr, var, x.end, dir="-")
elif x.end not in sing:
_end = f(x.end)
if len(sing) == 0:
solns = list(solveset(diff(expr, var), var))
extr = [_start, _end] + [f(i) for i in solns if i.is_real and i in x]
start, end = Min(*extr), Max(*extr)
left_open, right_open = False, False
if _start <= _end:
# the minimum or maximum value can occur simultaneously
# on both the edge of the interval and in some interior
# point
if start == _start and start not in solns:
left_open = x.left_open
if end == _end and end not in solns:
right_open = x.right_open
else:
if start == _end and start not in solns:
left_open = x.right_open
if end == _start and end not in solns:
right_open = x.left_open
return Interval(start, end, left_open, right_open)
else:
return imageset(f, Interval(x.start, sing[0],
x.left_open, True)) + \
Union(*[imageset(f, Interval(sing[i], sing[i + 1], True, True))
for i in range(0, len(sing) - 1)]) + \
imageset(f, Interval(sing[-1], x.end, True, x.right_open))
def _rsolve_hypergeometric(f, x, P, Q, k, m):
"""Recursive wrapper to rsolve_hypergeometric.
Returns a Tuple of (formula, series independent terms,
maximum power of x in independent terms) if successful
otherwise ``None``.
See :func:`rsolve_hypergeometric` for details.
"""
from sympy.polys import lcm, roots
from sympy.integrals import integrate
# transformation - c
proots, qroots = roots(P, k), roots(Q, k)
all_roots = dict(proots)
all_roots.update(qroots)
scale = lcm(
[r.as_numer_denom()[1] for r, t in all_roots.items() if r.is_rational])
f, P, Q, m = _transformation_c(f, x, P, Q, k, m, scale)
# transformation - a
qroots = roots(Q, k)
if qroots:
k_min = Min(*qroots.keys())
else:
k_min = S.Zero
shift = k_min + m
f, P, Q, m = _transformation_a(f, x, P, Q, k, m, shift)
l = (x * f).limit(x, 0)
if not isinstance(l, Limit) and l != 0: # Ideally should only be l != 0
return None
qroots = roots(Q, k)
if qroots:
k_max = Max(*qroots.keys())
else:
k_max = S.Zero
ind, mp = S.Zero, -oo
for i in range(k_max + m + 1):
r = f.diff(x, i).limit(x, 0) / factorial(i)
if r.is_finite is False:
old_f = f
f, P, Q, m = _transformation_a(f, x, P, Q, k, m, i)
f, P, Q, m = _transformation_e(f, x, P, Q, k, m)
sol, ind, mp = _rsolve_hypergeometric(f, x, P, Q, k, m)
sol = _apply_integrate(sol, x, k)
sol = _apply_shift(sol, i)
ind = integrate(ind, x)
ind += (old_f - ind).limit(x, 0) # constant of integration
mp += 1
return sol, ind, mp
elif r:
ind += r * x**(i + shift)
pow_x = Rational((i + shift), scale)
if pow_x > mp:
mp = pow_x # maximum power of x
ind = ind.subs(x, x**(1 / scale))
sol = _compute_formula(f, x, P, Q, k, m, k_max)
sol = _apply_shift(sol, shift)
sol = _apply_scale(sol, scale)
return sol, ind, mp
def _eval_interval(self, sym, a, b):
"""Evaluates the function along the sym in a given interval ab"""
# FIXME: Currently complex intervals are not supported. A possible
# replacement algorithm, discussed in issue 2128, can be found in the
# following papers;
# http://portal.acm.org/citation.cfm?id=281649
# http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.70.4127&rep=rep1&type=pdf
if a is None or b is None:
# In this case, it is just simple substitution
return super(Piecewise, self)._eval_interval(sym, a, b)
mul = 1
if (a == b) is True:
return S.Zero
elif (a > b) is True:
a, b, mul = b, a, -1
elif (a <= b) is not True:
newargs = []
for e, c in self.args:
intervals = self._sort_expr_cond(sym, S.NegativeInfinity,
S.Infinity, c)
values = []
for lower, upper in intervals:
if (a < lower) is True:
mid = lower
rep = b
val = e.subs(sym, b) - e.subs(sym, mid)
val += self._eval_interval(sym, a, mid)
elif (a > upper) is True:
mid = upper
rep = b
val = e.subs(sym, b) - e.subs(sym, mid)
val += self._eval_interval(sym, a, mid)
elif (a >= lower) is True and (a <= upper) is True:
rep = b
val = e.subs(sym, b) - e.subs(sym, a)
elif (b < lower) is True:
mid = lower
rep = a
val = e.subs(sym, mid) - e.subs(sym, a)
val += self._eval_interval(sym, mid, b)
elif (b > upper) is True:
mid = upper
rep = a
val = e.subs(sym, mid) - e.subs(sym, a)
val += self._eval_interval(sym, mid, b)
elif ((b >= lower) is True) and ((b <= upper) is True):
rep = a
val = e.subs(sym, b) - e.subs(sym, a)
else:
raise NotImplementedError(
"""The evaluation of a Piecewise interval when both the lower
and the upper limit are symbolic is not yet implemented."""
)
values.append(val)
if len(set(values)) == 1:
try:
c = c.subs(sym, rep)
except AttributeError:
pass
e = values[0]
newargs.append((e, c))
else:
for i in range(len(values)):
newargs.append(
(values[i], (c is True and i == len(values) - 1)
or And(rep >= intervals[i][0],
rep <= intervals[i][1])))
return Piecewise(*newargs)
# Determine what intervals the expr,cond pairs affect.
int_expr = self._sort_expr_cond(sym, a, b)
# Finally run through the intervals and sum the evaluation.
ret_fun = 0
for int_a, int_b, expr in int_expr:
ret_fun += expr._eval_interval(sym, Max(a, int_a), Min(b, int_b))
return mul * ret_fun