def check_convergence(numer, denom, n):
"""
Returns (h, g, p) where
-- h is:
> 0 for convergence of rate 1/factorial(n)**h
< 0 for divergence of rate factorial(n)**(-h)
= 0 for geometric or polynomial convergence or divergence
-- abs(g) is:
> 1 for geometric convergence of rate 1/h**n
< 1 for geometric divergence of rate h**n
= 1 for polynomial convergence or divergence
(g < 0 indicates an alternating series)
-- p is:
> 1 for polynomial convergence of rate 1/n**h
<= 1 for polynomial divergence of rate n**(-h)
"""
npol = C.Poly(numer, n)
dpol = C.Poly(denom, n)
p = npol.degree()
q = dpol.degree()
rate = q - p
if rate:
return rate, None, None
constant = dpol.LC() / npol.LC()
if abs(constant) != 1:
return rate, constant, None
if npol.degree() == dpol.degree() == 0:
return rate, constant, 0
pc = npol.all_coeffs()[1]
qc = dpol.all_coeffs()[1]
return rate, constant, (qc - pc)/dpol.LC()
def evalf_log(expr, prec, options):
arg = expr.args[0]
workprec = prec + 10
xre, xim, xacc, _ = evalf(arg, workprec, options)
if xim:
# XXX: use get_abs etc instead
re = evalf_log(
C.log(C.Abs(arg, evaluate=False), evaluate=False), prec, options)
im = mpf_atan2(xim, xre or fzero, prec)
return re[0], im, re[2], prec
imaginary_term = (mpf_cmp(xre, fzero) < 0)
re = mpf_log(mpf_abs(xre), prec, rnd)
size = fastlog(re)
if prec - size > workprec:
# We actually need to compute 1+x accurately, not x
arg = C.Add(S.NegativeOne, arg, evaluate=False)
xre, xim, _, _ = evalf_add(arg, prec, options)
prec2 = workprec - fastlog(xre)
# xre is now x - 1 so we add 1 back here to calculate x
re = mpf_log(mpf_abs(mpf_add(xre, fone, prec2)), prec, rnd)
re_acc = prec
if imaginary_term:
return re, mpf_pi(prec), re_acc, prec
else:
return re, None, re_acc, None
def evalf_piecewise(expr, prec, options):
if 'subs' in options:
expr = expr.subs(evalf_subs(prec, options['subs']))
newopts = options.copy()
del newopts['subs']
if hasattr(expr, 'func'):
return evalf(expr, prec, newopts)
if type(expr) == float:
return evalf(C.Float(expr), prec, newopts)
if type(expr) == int:
return evalf(C.Integer(expr), prec, newopts)
# We still have undefined symbols
raise NotImplementedError
def extract_leading_order(self, symbols, point=None):
"""
Returns the leading term and it's order.
Examples
========
>>> from sympy.abc import x
>>> (x + 1 + 1/x**5).extract_leading_order(x)
((x**(-5), O(x**(-5))),)
>>> (1 + x).extract_leading_order(x)
((1, O(1)),)
>>> (x + x**2).extract_leading_order(x)
((x, O(x)),)
"""
lst = []
symbols = list(symbols if is_sequence(symbols) else [symbols])
if not point:
point = [0] * len(symbols)
seq = [(f, C.Order(f, *zip(symbols, point))) for f in self.args]
for ef, of in seq:
for e, o in lst:
if o.contains(of) and o != of:
of = None
break
if of is None:
continue
new_lst = [(ef, of)]
for e, o in lst:
if of.contains(o) and o != of:
continue
new_lst.append((e, o))
lst = new_lst
return tuple(lst)
def evalf_log(expr, prec, options):
arg = expr.args[0]
workprec = prec + 10
xre, xim, xacc, _ = evalf(arg, workprec, options)
if xim:
# XXX: use get_abs etc instead
re = evalf_log(
C.log(C.Abs(arg, evaluate=False), evaluate=False), prec, options)
im = mpf_atan2(xim, xre or fzero, prec)
return re[0], im, re[2], prec
imaginary_term = (mpf_cmp(xre, fzero) < 0)
re = mpf_log(mpf_abs(xre), prec, rnd)
size = fastlog(re)
if prec - size > workprec:
# We actually need to compute 1+x accurately, not x
arg = C.Add(S.NegativeOne, arg, evaluate=False)
xre, xim, _, _ = evalf_add(arg, prec, options)
prec2 = workprec - fastlog(xre)
# xre is now x - 1 so we add 1 back here to calculate x
re = mpf_log(mpf_abs(mpf_add(xre, fone, prec2)), prec, rnd)
re_acc = prec
if imaginary_term:
return re, mpf_pi(prec), re_acc, prec
else:
return re, None, re_acc, None
def __new__(cls, *args, **options):
args = list(map(_sympify, args))
args = [a for a in args if a is not cls.identity]
if not options.pop('evaluate', True):
return cls._from_args(args)
if len(args) == 0:
return cls.identity
if len(args) == 1:
return args[0]
c_part, nc_part, order_symbols = cls.flatten(args)
is_commutative = not nc_part
obj = cls._from_args(c_part + nc_part, is_commutative)
if order_symbols is not None:
return C.Order(obj, *order_symbols)
return obj
def _create_evalf_table():
global evalf_table
evalf_table = {
C.Symbol: evalf_symbol,
C.Dummy: evalf_symbol,
C.Float: lambda x, prec, options: (x._mpf_, None, prec, None),
C.Rational: lambda x, prec, options: (from_rational(x.p, x.q, prec), None, prec, None),
C.Integer: lambda x, prec, options: (from_int(x.p, prec), None, prec, None),
C.Zero: lambda x, prec, options: (None, None, prec, None),
C.One: lambda x, prec, options: (fone, None, prec, None),
C.Half: lambda x, prec, options: (fhalf, None, prec, None),
C.Pi: lambda x, prec, options: (mpf_pi(prec), None, prec, None),
C.Exp1: lambda x, prec, options: (mpf_e(prec), None, prec, None),
C.ImaginaryUnit: lambda x, prec, options: (None, fone, None, prec),
C.NegativeOne: lambda x, prec, options: (fnone, None, prec, None),
C.NaN : lambda x, prec, options: (fnan, None, prec, None),
C.exp: lambda x, prec, options: evalf_pow(C.Pow(S.Exp1, x.args[0],
evaluate=False), prec, options),
C.cos: evalf_trig,
C.sin: evalf_trig,
C.Add: evalf_add,
C.Mul: evalf_mul,
C.Pow: evalf_pow,
C.log: evalf_log,
C.atan: evalf_atan,
C.Abs: evalf_abs,
C.re: evalf_re,
C.im: evalf_im,
C.floor: evalf_floor,
C.ceiling: evalf_ceiling,
C.Integral: evalf_integral,
C.Sum: evalf_sum,
C.Product: evalf_prod,
C.Piecewise: evalf_piecewise,
C.bernoulli: evalf_bernoulli,
}
def evalf_sum(expr, prec, options):
if 'subs' in options:
expr = expr.subs(options['subs'])
func = expr.function
limits = expr.limits
if len(limits) != 1 or len(limits[0]) != 3:
raise NotImplementedError
if func is S.Zero:
return mpf(0), None, None, None
prec2 = prec + 10
try:
n, a, b = limits[0]
if b != S.Infinity or a != int(a):
raise NotImplementedError
# Use fast hypergeometric summation if possible
v = hypsum(func, n, int(a), prec2)
delta = prec - fastlog(v)
if fastlog(v) < -10:
v = hypsum(func, n, int(a), delta)
return v, None, min(prec, delta), None
except NotImplementedError:
# Euler-Maclaurin summation for general series
eps = C.Float(2.0)**(-prec)
for i in range(1, 5):
m = n = 2**i * prec
s, err = expr.euler_maclaurin(m=m,
n=n,
eps=eps,
eval_integral=False)
err = err.evalf()
if err <= eps:
break
err = fastlog(evalf(abs(err), 20, options)[0])
re, im, re_acc, im_acc = evalf(s, prec2, options)
if re_acc is None:
re_acc = -err
if im_acc is None:
im_acc = -err
return re, im, re_acc, im_acc
def eval(cls, arg):
if isinstance(arg, Number) or arg in (True, False):
return false if arg else true
if arg.is_Not:
return arg.args[0]
# Simplify Relational objects.
if isinstance(arg, C.Equality):
return C.Unequality(*arg.args)
if isinstance(arg, C.Unequality):
return C.Equality(*arg.args)
if isinstance(arg, C.StrictLessThan):
return C.GreaterThan(*arg.args)
if isinstance(arg, C.StrictGreaterThan):
return C.LessThan(*arg.args)
if isinstance(arg, C.LessThan):
return C.StrictGreaterThan(*arg.args)
if isinstance(arg, C.GreaterThan):
return C.StrictLessThan(*arg.args)
def eval(cls, arg):
if isinstance(arg, Number) or arg in (True, False):
return false if arg else true
# apply De Morgan Rules
if arg.func is And:
return Or(*[Not(a) for a in arg.args])
if arg.func is Or:
return And(*[Not(a) for a in arg.args])
if arg.func is Not:
return arg.args[0]
# Simplify Relational objects.
if isinstance(arg, C.Equality):
return C.Unequality(*arg.args)
if isinstance(arg, C.Unequality):
return C.Equality(*arg.args)
if isinstance(arg, C.StrictLessThan):
return C.GreaterThan(*arg.args)
if isinstance(arg, C.StrictGreaterThan):
return C.LessThan(*arg.args)
if isinstance(arg, C.LessThan):
return C.StrictGreaterThan(*arg.args)
if isinstance(arg, C.GreaterThan):
return C.StrictLessThan(*arg.args)
def _matches_commutative(self, expr, repl_dict={}, old=False):
"""
Matches Add/Mul "pattern" to an expression "expr".
repl_dict ... a dictionary of (wild: expression) pairs, that get
returned with the results
This function is the main workhorse for Add/Mul.
For instance:
>>> from sympy import symbols, Wild, sin
>>> a = Wild("a")
>>> b = Wild("b")
>>> c = Wild("c")
>>> x, y, z = symbols("x y z")
>>> (a+sin(b)*c)._matches_commutative(x+sin(y)*z)
{a_: x, b_: y, c_: z}
In the example above, "a+sin(b)*c" is the pattern, and "x+sin(y)*z" is
the expression.
The repl_dict contains parts that were already matched. For example
here:
>>> (x+sin(b)*c)._matches_commutative(x+sin(y)*z, repl_dict={a: x})
{a_: x, b_: y, c_: z}
the only function of the repl_dict is to return it in the
result, e.g. if you omit it:
>>> (x+sin(b)*c)._matches_commutative(x+sin(y)*z)
{b_: y, c_: z}
the "a: x" is not returned in the result, but otherwise it is
equivalent.
"""
# handle simple patterns
if self == expr:
return repl_dict
d = self._matches_simple(expr, repl_dict)
if d is not None:
return d
# eliminate exact part from pattern: (2+a+w1+w2).matches(expr) -> (w1+w2).matches(expr-a-2)
from .function import WildFunction
from .symbol import Wild
wild_part = []
exact_part = []
for p in ordered(self.args):
if p.has(Wild, WildFunction) and (not expr.has(p)):
# not all Wild should stay Wilds, for example:
# (w2+w3).matches(w1) -> (w1+w3).matches(w1) -> w3.matches(0)
wild_part.append(p)
else:
exact_part.append(p)
if exact_part:
exact = self.func(*exact_part)
free = expr.free_symbols
if free and (exact.free_symbols - free):
# there are symbols in the exact part that are not
# in the expr; but if there are no free symbols, let
# the matching continue
return None
newpattern = self.func(*wild_part)
newexpr = self._combine_inverse(expr, exact)
if not old and (expr.is_Add or expr.is_Mul):
if newexpr.count_ops() > expr.count_ops():
return None
return newpattern.matches(newexpr, repl_dict)
# now to real work ;)
i = 0
saw = set()
while expr not in saw:
saw.add(expr)
expr_list = (self.identity,) + tuple(ordered(self.make_args(expr)))
for last_op in reversed(expr_list):
for w in reversed(wild_part):
d1 = w.matches(last_op, repl_dict)
if d1 is not None:
d2 = self.xreplace(d1).matches(expr, d1)
if d2 is not None:
return d2
if i == 0:
if self.is_Mul:
# make e**i look like Mul
if expr.is_Pow and expr.exp.is_Integer:
if expr.exp > 0:
expr = C.Mul(*[expr.base, expr.base**(expr.exp - 1)], evaluate=False)
else:
expr = C.Mul(*[1/expr.base, expr.base**(expr.exp + 1)], evaluate=False)
i += 1
continue
elif self.is_Add:
# make i*e look like Add
c, e = expr.as_coeff_Mul()
if abs(c) > 1:
if c > 0:
expr = C.Add(*[e, (c - 1)*e], evaluate=False)
else:
expr = C.Add(*[-e, (c + 1)*e], evaluate=False)
i += 1
continue
# try collection on non-Wild symbols
from sympy.simplify.simplify import collect
was = expr
did = set()
for w in reversed(wild_part):
c, w = w.as_coeff_mul(Wild)
free = c.free_symbols - did
if free:
did.update(free)
expr = collect(expr, free)
if expr != was:
i += 0
continue
break # if we didn't continue, there is nothing more to do
return
def do_integral(expr, prec, options):
func = expr.args[0]
x, xlow, xhigh = expr.args[1]
if xlow == xhigh:
xlow = xhigh = 0
elif x not in func.free_symbols:
# only the difference in limits matters in this case
# so if there is a symbol in common that will cancel
# out when taking the difference, then use that
# difference
if xhigh.free_symbols & xlow.free_symbols:
diff = xhigh - xlow
if not diff.free_symbols:
xlow, xhigh = 0, diff
oldmaxprec = options.get('maxprec', DEFAULT_MAXPREC)
options['maxprec'] = min(oldmaxprec, 2*prec)
with workprec(prec + 5):
xlow = as_mpmath(xlow, prec + 15, options)
xhigh = as_mpmath(xhigh, prec + 15, options)
# Integration is like summation, and we can phone home from
# the integrand function to update accuracy summation style
# Note that this accuracy is inaccurate, since it fails
# to account for the variable quadrature weights,
# but it is better than nothing
have_part = [False, False]
max_real_term = [MINUS_INF]
max_imag_term = [MINUS_INF]
def f(t):
re, im, re_acc, im_acc = evalf(func, mp.prec, {'subs': {x: t}})
have_part[0] = re or have_part[0]
have_part[1] = im or have_part[1]
max_real_term[0] = max(max_real_term[0], fastlog(re))
max_imag_term[0] = max(max_imag_term[0], fastlog(im))
if im:
return mpc(re or fzero, im)
return mpf(re or fzero)
if options.get('quad') == 'osc':
A = C.Wild('A', exclude=[x])
B = C.Wild('B', exclude=[x])
D = C.Wild('D')
m = func.match(C.cos(A*x + B)*D)
if not m:
m = func.match(C.sin(A*x + B)*D)
if not m:
raise ValueError("An integrand of the form sin(A*x+B)*f(x) "
"or cos(A*x+B)*f(x) is required for oscillatory quadrature")
period = as_mpmath(2*S.Pi/m[A], prec + 15, options)
result = quadosc(f, [xlow, xhigh], period=period)
# XXX: quadosc does not do error detection yet
quadrature_error = MINUS_INF
else:
result, quadrature_error = quadts(f, [xlow, xhigh], error=1)
quadrature_error = fastlog(quadrature_error._mpf_)
options['maxprec'] = oldmaxprec
if have_part[0]:
re = result.real._mpf_
if re == fzero:
re, re_acc = scaled_zero(
min(-prec, -max_real_term[0], -quadrature_error))
re = scaled_zero(re) # handled ok in evalf_integral
else:
re_acc = -max(max_real_term[0] - fastlog(re) -
prec, quadrature_error)
else:
re, re_acc = None, None
if have_part[1]:
im = result.imag._mpf_
if im == fzero:
im, im_acc = scaled_zero(
min(-prec, -max_imag_term[0], -quadrature_error))
im = scaled_zero(im) # handled ok in evalf_integral
else:
im_acc = -max(max_imag_term[0] - fastlog(im) -
prec, quadrature_error)
else:
im, im_acc = None, None
result = re, im, re_acc, im_acc
return result
def do_integral(expr, prec, options):
func = expr.args[0]
x, xlow, xhigh = expr.args[1]
if xlow == xhigh:
xlow = xhigh = 0
elif x not in func.free_symbols:
# only the difference in limits matters in this case
# so if there is a symbol in common that will cancel
# out when taking the difference, then use that
# difference
if xhigh.free_symbols & xlow.free_symbols:
diff = xhigh - xlow
if not diff.free_symbols:
xlow, xhigh = 0, diff
orig = mp.prec
oldmaxprec = options.get('maxprec', DEFAULT_MAXPREC)
options['maxprec'] = min(oldmaxprec, 2*prec)
try:
mp.prec = prec + 5
xlow = as_mpmath(xlow, prec + 15, options)
xhigh = as_mpmath(xhigh, prec + 15, options)
# Integration is like summation, and we can phone home from
# the integrand function to update accuracy summation style
# Note that this accuracy is inaccurate, since it fails
# to account for the variable quadrature weights,
# but it is better than nothing
have_part = [False, False]
max_real_term = [MINUS_INF]
max_imag_term = [MINUS_INF]
def f(t):
re, im, re_acc, im_acc = evalf(func, mp.prec, {'subs': {x: t}})
have_part[0] = re or have_part[0]
have_part[1] = im or have_part[1]
max_real_term[0] = max(max_real_term[0], fastlog(re))
max_imag_term[0] = max(max_imag_term[0], fastlog(im))
if im:
return mpc(re or fzero, im)
return mpf(re or fzero)
if options.get('quad') == 'osc':
A = C.Wild('A', exclude=[x])
B = C.Wild('B', exclude=[x])
D = C.Wild('D')
m = func.match(C.cos(A*x + B)*D)
if not m:
m = func.match(C.sin(A*x + B)*D)
if not m:
raise ValueError("An integrand of the form sin(A*x+B)*f(x) "
"or cos(A*x+B)*f(x) is required for oscillatory quadrature")
period = as_mpmath(2*S.Pi/m[A], prec + 15, options)
result = quadosc(f, [xlow, xhigh], period=period)
# XXX: quadosc does not do error detection yet
quadrature_error = MINUS_INF
else:
result, quadrature_error = quadts(f, [xlow, xhigh], error=1)
quadrature_error = fastlog(quadrature_error._mpf_)
finally:
options['maxprec'] = oldmaxprec
mp.prec = orig
if have_part[0]:
re = result.real._mpf_
if re == fzero:
re, re_acc = scaled_zero(
min(-prec, -max_real_term[0], -quadrature_error))
re = scaled_zero(re) # handled ok in evalf_integral
else:
re_acc = -max(max_real_term[0] - fastlog(re) -
prec, quadrature_error)
else:
re, re_acc = None, None
if have_part[1]:
im = result.imag._mpf_
if im == fzero:
im, im_acc = scaled_zero(
min(-prec, -max_imag_term[0], -quadrature_error))
im = scaled_zero(im) # handled ok in evalf_integral
else:
im_acc = -max(max_imag_term[0] - fastlog(im) -
prec, quadrature_error)
else:
im, im_acc = None, None
result = re, im, re_acc, im_acc
return result
def subs(self, *args, **kwargs):
"""
Substitutes old for new in an expression after sympifying args.
`args` is either:
- two arguments, e.g. foo.subs(old, new)
- one iterable argument, e.g. foo.subs(iterable). The iterable may be
o an iterable container with (old, new) pairs. In this case the
replacements are processed in the order given with successive
patterns possibly affecting replacements already made.
o a dict or set whose key/value items correspond to old/new pairs.
In this case the old/new pairs will be sorted by op count and in
case of a tie, by number of args and the default_sort_key. The
resulting sorted list is then processed as an iterable container
(see previous).
If the keyword ``simultaneous`` is True, the subexpressions will not be
evaluated until all the substitutions have been made.
Examples
========
>>> from sympy import pi, exp
>>> from sympy.abc import x, y
>>> (1 + x*y).subs(x, pi)
pi*y + 1
>>> (1 + x*y).subs({x:pi, y:2})
1 + 2*pi
>>> (1 + x*y).subs([(x, pi), (y, 2)])
1 + 2*pi
>>> reps = [(y, x**2), (x, 2)]
>>> (x + y).subs(reps)
6
>>> (x + y).subs(reversed(reps))
x**2 + 2
>>> (x**2 + x**4).subs(x**2, y)
y**2 + y
To replace only the x**2 but not the x**4, use xreplace:
>>> (x**2 + x**4).xreplace({x**2: y})
x**4 + y
To delay evaluation until all substitutions have been made,
set the keyword ``simultaneous`` to True:
>>> (x/y).subs([(x, 0), (y, 0)])
0
>>> (x/y).subs([(x, 0), (y, 0)], simultaneous=True)
nan
This has the added feature of not allowing subsequent substitutions
to affect those already made:
>>> ((x + y)/y).subs({x + y: y, y: x + y})
1
>>> ((x + y)/y).subs({x + y: y, y: x + y}, simultaneous=True)
y/(x + y)
In order to obtain a canonical result, unordered iterables are
sorted by count_op length, number of arguments and by the
default_sort_key to break any ties. All other iterables are left
unsorted.
>>> from sympy import sqrt, sin, cos, exp
>>> from sympy.abc import a, b, c, d, e
>>> A = (sqrt(sin(2*x)), a)
>>> B = (sin(2*x), b)
>>> C = (cos(2*x), c)
>>> D = (x, d)
>>> E = (exp(x), e)
>>> expr = sqrt(sin(2*x))*sin(exp(x)*x)*cos(2*x) + sin(2*x)
>>> expr.subs(dict([A,B,C,D,E]))
a*c*sin(d*e) + b
See Also
========
replace: replacement capable of doing wildcard-like matching,
parsing of match, and conditional replacements
xreplace: exact node replacement in expr tree; also capable of
using matching rules
"""
from sympy.core.containers import Dict
from sympy.utilities import default_sort_key
unordered = False
if len(args) == 1:
sequence = args[0]
if isinstance(sequence, set):
unordered = True
elif isinstance(sequence, (Dict, dict)):
unordered = True
sequence = sequence.items()
elif not iterable(sequence):
from sympy.utilities.misc import filldedent
raise ValueError(
filldedent("""
When a single argument is passed to subs
it should be a dictionary of old: new pairs or an iterable
of (old, new) tuples."""))
elif len(args) == 2:
sequence = [args]
else:
raise ValueError("subs accepts either 1 or 2 arguments")
sequence = list(sequence)
for i in range(len(sequence)):
o, n = sequence[i]
so, sn = sympify(o), sympify(n)
if not isinstance(so, Basic):
if type(o) is str:
so = C.Symbol(o)
sequence[i] = (so, sn)
if _aresame(so, sn):
sequence[i] = None
continue
sequence = filter(None, sequence)
if unordered:
sequence = dict(sequence)
if not all(k.is_Atom for k in sequence):
d = {}
for o, n in sequence.iteritems():
try:
ops = o.count_ops(), len(o.args)
except TypeError:
ops = (0, 0)
d.setdefault(ops, []).append((o, n))
newseq = []
for k in sorted(d.keys(), reverse=True):
newseq.extend(
sorted([v[0] for v in d[k]], key=default_sort_key))
sequence = [(k, sequence[k]) for k in newseq]
del newseq, d
else:
sequence = sorted([(k, v) for (k, v) in sequence.iteritems()],
key=default_sort_key)
if kwargs.pop('simultaneous',
False): # XXX should this be the default for dict subs?
reps = {}
rv = self
for old, new in sequence:
d = C.Dummy()
rv = rv._subs(old, d)
reps[d] = new
if not isinstance(rv, Basic):
break
return rv.xreplace(reps)
else:
rv = self
for old, new in sequence:
rv = rv._subs(old, new)
if not isinstance(rv, Basic):
break
return rv
def hypsum(expr, n, start, prec):
"""
Sum a rapidly convergent infinite hypergeometric series with
given general term, e.g. e = hypsum(1/factorial(n), n). The
quotient between successive terms must be a quotient of integer
polynomials.
"""
from sympy import hypersimp, lambdify
if prec == float('inf'):
raise NotImplementedError('does not support inf prec')
if start:
expr = expr.subs(n, n + start)
hs = hypersimp(expr, n)
if hs is None:
raise NotImplementedError("a hypergeometric series is required")
num, den = hs.as_numer_denom()
func1 = lambdify(n, num)
func2 = lambdify(n, den)
h, g, p = check_convergence(num, den, n)
if h < 0:
raise ValueError("Sum diverges like (n!)^%i" % (-h))
term = expr.subs(n, 0)
if not term.is_Rational:
raise NotImplementedError(
"Non rational term functionality is not implemented.")
# Direct summation if geometric or faster
if h > 0 or (h == 0 and abs(g) > 1):
term = (MPZ(term.p) << prec) // term.q
s = term
k = 1
while abs(term) > 5:
term *= MPZ(func1(k - 1))
term //= MPZ(func2(k - 1))
s += term
k += 1
return from_man_exp(s, -prec)
else:
alt = g < 0
if abs(g) < 1:
raise ValueError("Sum diverges like (%i)^n" % abs(1 / g))
if p < 1 or (p == 1 and not alt):
raise ValueError("Sum diverges like n^%i" % (-p))
# We have polynomial convergence: use Richardson extrapolation
vold = None
ndig = prec_to_dps(prec)
while True:
# Need to use at least quad precision because a lot of cancellation
# might occur in the extrapolation process; we check the answer to
# make sure that the desired precision has been reached, too.
prec2 = 4 * prec
term0 = (MPZ(term.p) << prec2) // term.q
def summand(k, _term=[term0]):
if k:
k = int(k)
_term[0] *= MPZ(func1(k - 1))
_term[0] //= MPZ(func2(k - 1))
return make_mpf(from_man_exp(_term[0], -prec2))
with workprec(prec):
v = nsum(summand, [0, mpmath_inf], method='richardson')
vf = C.Float(v, ndig)
if vold is not None and vold == vf:
break
prec += prec # double precision each time
vold = vf
return v._mpf_
def _eval_product(self, term, limits):
from sympy.concrete.delta import deltaproduct, _has_simple_delta
from sympy.concrete.summations import summation
from sympy.functions import KroneckerDelta
(k, a, n) = limits
if k not in term.free_symbols:
return term**(n - a + 1)
if a == n:
return term.subs(k, a)
if term.has(KroneckerDelta) and _has_simple_delta(term, limits[0]):
return deltaproduct(term, limits)
dif = n - a
if dif.is_Integer:
return Mul(*[term.subs(k, a + i) for i in xrange(dif + 1)])
elif term.is_polynomial(k):
poly = term.as_poly(k)
A = B = Q = S.One
all_roots = roots(poly)
M = 0
for r, m in all_roots.items():
M += m
A *= C.RisingFactorial(a - r, n - a + 1)**m
Q *= (n - r)**m
if M < poly.degree():
arg = quo(poly, Q.as_poly(k))
B = self.func(arg, (k, a, n)).doit()
return poly.LC()**(n - a + 1) * A * B
elif term.is_Add:
p, q = term.as_numer_denom()
p = self._eval_product(p, (k, a, n))
q = self._eval_product(q, (k, a, n))
return p / q
elif term.is_Mul:
exclude, include = [], []
for t in term.args:
p = self._eval_product(t, (k, a, n))
if p is not None:
exclude.append(p)
else:
include.append(t)
if not exclude:
return None
else:
arg = term._new_rawargs(*include)
A = Mul(*exclude)
B = self.func(arg, (k, a, n)).doit()
return A * B
elif term.is_Pow:
if not term.base.has(k):
s = summation(term.exp, (k, a, n))
return term.base**s
elif not term.exp.has(k):
p = self._eval_product(term.base, (k, a, n))
if p is not None:
return p**term.exp
elif isinstance(term, Product):
evaluated = term.doit()
f = self._eval_product(evaluated, limits)
if f is None:
return self.func(evaluated, limits)
else:
return f
