8274b78f238537f55a2fdac5624687534650241ed111eeb7b3bab8b2371bbf9b

lib/python3.11/site-packages/mpmath/calculus/__init__.py

@@ -0,0 +1,6 @@
+from . import calculus
+# XXX: hack to set methods
+from . import approximation
+from . import differentiation
+from . import extrapolation
+from . import polynomials

lib/python3.11/site-packages/mpmath/calculus/approximation.py

@@ -0,0 +1,246 @@
+from ..libmp.backend import xrange
+from .calculus import defun
+
+#----------------------------------------------------------------------------#
+#                              Approximation methods                         #
+#----------------------------------------------------------------------------#
+
+# The Chebyshev approximation formula is given at:
+# http://mathworld.wolfram.com/ChebyshevApproximationFormula.html
+
+# The only major changes in the following code is that we return the
+# expanded polynomial coefficients instead of Chebyshev coefficients,
+# and that we automatically transform [a,b] -> [-1,1] and back
+# for convenience.
+
+# Coefficient in Chebyshev approximation
+def chebcoeff(ctx,f,a,b,j,N):
+    s = ctx.mpf(0)
+    h = ctx.mpf(0.5)
+    for k in range(1, N+1):
+        t = ctx.cospi((k-h)/N)
+        s += f(t*(b-a)*h + (b+a)*h) * ctx.cospi(j*(k-h)/N)
+    return 2*s/N
+
+# Generate Chebyshev polynomials T_n(ax+b) in expanded form
+def chebT(ctx, a=1, b=0):
+    Tb = [1]
+    yield Tb
+    Ta = [b, a]
+    while 1:
+        yield Ta
+        # Recurrence: T[n+1](ax+b) = 2*(ax+b)*T[n](ax+b) - T[n-1](ax+b)
+        Tmp = [0] + [2*a*t for t in Ta]
+        for i, c in enumerate(Ta): Tmp[i] += 2*b*c
+        for i, c in enumerate(Tb): Tmp[i] -= c
+        Ta, Tb = Tmp, Ta
+
+@defun
+def chebyfit(ctx, f, interval, N, error=False):
+    r"""
+    Computes a polynomial of degree `N-1` that approximates the
+    given function `f` on the interval `[a, b]`. With ``error=True``,
+    :func:`~mpmath.chebyfit` also returns an accurate estimate of the
+    maximum absolute error; that is, the maximum value of
+    `|f(x) - P(x)|` for `x \in [a, b]`.
+
+    :func:`~mpmath.chebyfit` uses the Chebyshev approximation formula,
+    which gives a nearly optimal solution: that is, the maximum
+    error of the approximating polynomial is very close to
+    the smallest possible for any polynomial of the same degree.
+
+    Chebyshev approximation is very useful if one needs repeated
+    evaluation of an expensive function, such as function defined
+    implicitly by an integral or a differential equation. (For
+    example, it could be used to turn a slow mpmath function
+    into a fast machine-precision version of the same.)
+
+    **Examples**
+
+    Here we use :func:`~mpmath.chebyfit` to generate a low-degree approximation
+    of `f(x) = \cos(x)`, valid on the interval `[1, 2]`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> poly, err = chebyfit(cos, [1, 2], 5, error=True)
+        >>> nprint(poly)
+        [0.00291682, 0.146166, -0.732491, 0.174141, 0.949553]
+        >>> nprint(err, 12)
+        1.61351758081e-5
+
+    The polynomial can be evaluated using ``polyval``::
+
+        >>> nprint(polyval(poly, 1.6), 12)
+        -0.0291858904138
+        >>> nprint(cos(1.6), 12)
+        -0.0291995223013
+
+    Sampling the true error at 1000 points shows that the error
+    estimate generated by ``chebyfit`` is remarkably good::
+
+        >>> error = lambda x: abs(cos(x) - polyval(poly, x))
+        >>> nprint(max([error(1+n/1000.) for n in range(1000)]), 12)
+        1.61349954245e-5
+
+    **Choice of degree**
+
+    The degree `N` can be set arbitrarily high, to obtain an
+    arbitrarily good approximation. As a rule of thumb, an
+    `N`-term Chebyshev approximation is good to `N/(b-a)` decimal
+    places on a unit interval (although this depends on how
+    well-behaved `f` is). The cost grows accordingly: ``chebyfit``
+    evaluates the function `(N^2)/2` times to compute the
+    coefficients and an additional `N` times to estimate the error.
+
+    **Possible issues**
+
+    One should be careful to use a sufficiently high working
+    precision both when calling ``chebyfit`` and when evaluating
+    the resulting polynomial, as the polynomial is sometimes
+    ill-conditioned. It is for example difficult to reach
+    15-digit accuracy when evaluating the polynomial using
+    machine precision floats, no matter the theoretical
+    accuracy of the polynomial. (The option to return the
+    coefficients in Chebyshev form should be made available
+    in the future.)
+
+    It is important to note the Chebyshev approximation works
+    poorly if `f` is not smooth. A function containing singularities,
+    rapid oscillation, etc can be approximated more effectively by
+    multiplying it by a weight function that cancels out the
+    nonsmooth features, or by dividing the interval into several
+    segments.
+    """
+    a, b = ctx._as_points(interval)
+    orig = ctx.prec
+    try:
+        ctx.prec = orig + int(N**0.5) + 20
+        c = [chebcoeff(ctx,f,a,b,k,N) for k in range(N)]
+        d = [ctx.zero] * N
+        d[0] = -c[0]/2
+        h = ctx.mpf(0.5)
+        T = chebT(ctx, ctx.mpf(2)/(b-a), ctx.mpf(-1)*(b+a)/(b-a))
+        for (k, Tk) in zip(range(N), T):
+            for i in range(len(Tk)):
+                d[i] += c[k]*Tk[i]
+        d = d[::-1]
+        # Estimate maximum error
+        err = ctx.zero
+        for k in range(N):
+            x = ctx.cos(ctx.pi*k/N) * (b-a)*h + (b+a)*h
+            err = max(err, abs(f(x) - ctx.polyval(d, x)))
+    finally:
+        ctx.prec = orig
+    if error:
+        return d, +err
+    else:
+        return d
+
+@defun
+def fourier(ctx, f, interval, N):
+    r"""
+    Computes the Fourier series of degree `N` of the given function
+    on the interval `[a, b]`. More precisely, :func:`~mpmath.fourier` returns
+    two lists `(c, s)` of coefficients (the cosine series and sine
+    series, respectively), such that
+
+    .. math ::
+
+        f(x) \sim \sum_{k=0}^N
+            c_k \cos(k m x) + s_k \sin(k m x)
+
+    where `m = 2 \pi / (b-a)`.
+
+    Note that many texts define the first coefficient as `2 c_0` instead
+    of `c_0`. The easiest way to evaluate the computed series correctly
+    is to pass it to :func:`~mpmath.fourierval`.
+
+    **Examples**
+
+    The function `f(x) = x` has a simple Fourier series on the standard
+    interval `[-\pi, \pi]`. The cosine coefficients are all zero (because
+    the function has odd symmetry), and the sine coefficients are
+    rational numbers::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> c, s = fourier(lambda x: x, [-pi, pi], 5)
+        >>> nprint(c)
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
+        >>> nprint(s)
+        [0.0, 2.0, -1.0, 0.666667, -0.5, 0.4]
+
+    This computes a Fourier series of a nonsymmetric function on
+    a nonstandard interval::
+
+        >>> I = [-1, 1.5]
+        >>> f = lambda x: x**2 - 4*x + 1
+        >>> cs = fourier(f, I, 4)
+        >>> nprint(cs[0])
+        [0.583333, 1.12479, -1.27552, 0.904708, -0.441296]
+        >>> nprint(cs[1])
+        [0.0, -2.6255, 0.580905, 0.219974, -0.540057]
+
+    It is instructive to plot a function along with its truncated
+    Fourier series::
+
+        >>> plot([f, lambda x: fourierval(cs, I, x)], I) #doctest: +SKIP
+
+    Fourier series generally converge slowly (and may not converge
+    pointwise). For example, if `f(x) = \cosh(x)`, a 10-term Fourier
+    series gives an `L^2` error corresponding to 2-digit accuracy::
+
+        >>> I = [-1, 1]
+        >>> cs = fourier(cosh, I, 9)
+        >>> g = lambda x: (cosh(x) - fourierval(cs, I, x))**2
+        >>> nprint(sqrt(quad(g, I)))
+        0.00467963
+
+    :func:`~mpmath.fourier` uses numerical quadrature. For nonsmooth functions,
+    the accuracy (and speed) can be improved by including all singular
+    points in the interval specification::
+
+        >>> nprint(fourier(abs, [-1, 1], 0), 10)
+        ([0.5000441648], [0.0])
+        >>> nprint(fourier(abs, [-1, 0, 1], 0), 10)
+        ([0.5], [0.0])
+
+    """
+    interval = ctx._as_points(interval)
+    a = interval[0]
+    b = interval[-1]
+    L = b-a
+    cos_series = []
+    sin_series = []
+    cutoff = ctx.eps*10
+    for n in xrange(N+1):
+        m = 2*n*ctx.pi/L
+        an = 2*ctx.quadgl(lambda t: f(t)*ctx.cos(m*t), interval)/L
+        bn = 2*ctx.quadgl(lambda t: f(t)*ctx.sin(m*t), interval)/L
+        if n == 0:
+            an /= 2
+        if abs(an) < cutoff: an = ctx.zero
+        if abs(bn) < cutoff: bn = ctx.zero
+        cos_series.append(an)
+        sin_series.append(bn)
+    return cos_series, sin_series
+
+@defun
+def fourierval(ctx, series, interval, x):
+    """
+    Evaluates a Fourier series (in the format computed by
+    by :func:`~mpmath.fourier` for the given interval) at the point `x`.
+
+    The series should be a pair `(c, s)` where `c` is the
+    cosine series and `s` is the sine series. The two lists
+    need not have the same length.
+    """
+    cs, ss = series
+    ab = ctx._as_points(interval)
+    a = interval[0]
+    b = interval[-1]
+    m = 2*ctx.pi/(ab[-1]-ab[0])
+    s = ctx.zero
+    s += ctx.fsum(cs[n]*ctx.cos(m*n*x) for n in xrange(len(cs)) if cs[n])
+    s += ctx.fsum(ss[n]*ctx.sin(m*n*x) for n in xrange(len(ss)) if ss[n])
+    return s

lib/python3.11/site-packages/mpmath/calculus/calculus.py

@@ -0,0 +1,6 @@
+class CalculusMethods(object):
+    pass
+
+def defun(f):
+    setattr(CalculusMethods, f.__name__, f)
+    return f

lib/python3.11/site-packages/mpmath/calculus/differentiation.py

@@ -0,0 +1,647 @@
+from ..libmp.backend import xrange
+from .calculus import defun
+
+try:
+    iteritems = dict.iteritems
+except AttributeError:
+    iteritems = dict.items
+
+#----------------------------------------------------------------------------#
+#                                Differentiation                             #
+#----------------------------------------------------------------------------#
+
+@defun
+def difference(ctx, s, n):
+    r"""
+    Given a sequence `(s_k)` containing at least `n+1` items, returns the
+    `n`-th forward difference,
+
+    .. math ::
+
+        \Delta^n = \sum_{k=0}^{\infty} (-1)^{k+n} {n \choose k} s_k.
+    """
+    n = int(n)
+    d = ctx.zero
+    b = (-1) ** (n & 1)
+    for k in xrange(n+1):
+        d += b * s[k]
+        b = (b * (k-n)) // (k+1)
+    return d
+
+def hsteps(ctx, f, x, n, prec, **options):
+    singular = options.get('singular')
+    addprec = options.get('addprec', 10)
+    direction = options.get('direction', 0)
+    workprec = (prec+2*addprec) * (n+1)
+    orig = ctx.prec
+    try:
+        ctx.prec = workprec
+        h = options.get('h')
+        if h is None:
+            if options.get('relative'):
+                hextramag = int(ctx.mag(x))
+            else:
+                hextramag = 0
+            h = ctx.ldexp(1, -prec-addprec-hextramag)
+        else:
+            h = ctx.convert(h)
+        # Directed: steps x, x+h, ... x+n*h
+        direction = options.get('direction', 0)
+        if direction:
+            h *= ctx.sign(direction)
+            steps = xrange(n+1)
+            norm = h
+        # Central: steps x-n*h, x-(n-2)*h ..., x, ..., x+(n-2)*h, x+n*h
+        else:
+            steps = xrange(-n, n+1, 2)
+            norm = (2*h)
+        # Perturb
+        if singular:
+            x += 0.5*h
+        values = [f(x+k*h) for k in steps]
+        return values, norm, workprec
+    finally:
+        ctx.prec = orig
+
+
+@defun
+def diff(ctx, f, x, n=1, **options):
+    r"""
+    Numerically computes the derivative of `f`, `f'(x)`, or generally for
+    an integer `n \ge 0`, the `n`-th derivative `f^{(n)}(x)`.
+    A few basic examples are::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> diff(lambda x: x**2 + x, 1.0)
+        3.0
+        >>> diff(lambda x: x**2 + x, 1.0, 2)
+        2.0
+        >>> diff(lambda x: x**2 + x, 1.0, 3)
+        0.0
+        >>> nprint([diff(exp, 3, n) for n in range(5)])   # exp'(x) = exp(x)
+        [20.0855, 20.0855, 20.0855, 20.0855, 20.0855]
+
+    Even more generally, given a tuple of arguments `(x_1, \ldots, x_k)`
+    and order `(n_1, \ldots, n_k)`, the partial derivative
+    `f^{(n_1,\ldots,n_k)}(x_1,\ldots,x_k)` is evaluated. For example::
+
+        >>> diff(lambda x,y: 3*x*y + 2*y - x, (0.25, 0.5), (0,1))
+        2.75
+        >>> diff(lambda x,y: 3*x*y + 2*y - x, (0.25, 0.5), (1,1))
+        3.0
+
+    **Options**
+
+    The following optional keyword arguments are recognized:
+
+    ``method``
+        Supported methods are ``'step'`` or ``'quad'``: derivatives may be
+        computed using either a finite difference with a small step
+        size `h` (default), or numerical quadrature.
+    ``direction``
+        Direction of finite difference: can be -1 for a left
+        difference, 0 for a central difference (default), or +1
+        for a right difference; more generally can be any complex number.
+    ``addprec``
+        Extra precision for `h` used to account for the function's
+        sensitivity to perturbations (default = 10).
+    ``relative``
+        Choose `h` relative to the magnitude of `x`, rather than an
+        absolute value; useful for large or tiny `x` (default = False).
+    ``h``
+        As an alternative to ``addprec`` and ``relative``, manually
+        select the step size `h`.
+    ``singular``
+        If True, evaluation exactly at the point `x` is avoided; this is
+        useful for differentiating functions with removable singularities.
+        Default = False.
+    ``radius``
+        Radius of integration contour (with ``method = 'quad'``).
+        Default = 0.25. A larger radius typically is faster and more
+        accurate, but it must be chosen so that `f` has no
+        singularities within the radius from the evaluation point.
+
+    A finite difference requires `n+1` function evaluations and must be
+    performed at `(n+1)` times the target precision. Accordingly, `f` must
+    support fast evaluation at high precision.
+
+    With integration, a larger number of function evaluations is
+    required, but not much extra precision is required. For high order
+    derivatives, this method may thus be faster if f is very expensive to
+    evaluate at high precision.
+
+    **Further examples**
+
+    The direction option is useful for computing left- or right-sided
+    derivatives of nonsmooth functions::
+
+        >>> diff(abs, 0, direction=0)
+        0.0
+        >>> diff(abs, 0, direction=1)
+        1.0
+        >>> diff(abs, 0, direction=-1)
+        -1.0
+
+    More generally, if the direction is nonzero, a right difference
+    is computed where the step size is multiplied by sign(direction).
+    For example, with direction=+j, the derivative from the positive
+    imaginary direction will be computed::
+
+        >>> diff(abs, 0, direction=j)
+        (0.0 - 1.0j)
+
+    With integration, the result may have a small imaginary part
+    even even if the result is purely real::
+
+        >>> diff(sqrt, 1, method='quad')    # doctest:+ELLIPSIS
+        (0.5 - 4.59...e-26j)
+        >>> chop(_)
+        0.5
+
+    Adding precision to obtain an accurate value::
+
+        >>> diff(cos, 1e-30)
+        0.0
+        >>> diff(cos, 1e-30, h=0.0001)
+        -9.99999998328279e-31
+        >>> diff(cos, 1e-30, addprec=100)
+        -1.0e-30
+
+    """
+    partial = False
+    try:
+        orders = list(n)
+        x = list(x)
+        partial = True
+    except TypeError:
+        pass
+    if partial:
+        x = [ctx.convert(_) for _ in x]
+        return _partial_diff(ctx, f, x, orders, options)
+    method = options.get('method', 'step')
+    if n == 0 and method != 'quad' and not options.get('singular'):
+        return f(ctx.convert(x))
+    prec = ctx.prec
+    try:
+        if method == 'step':
+            values, norm, workprec = hsteps(ctx, f, x, n, prec, **options)
+            ctx.prec = workprec
+            v = ctx.difference(values, n) / norm**n
+        elif method == 'quad':
+            ctx.prec += 10
+            radius = ctx.convert(options.get('radius', 0.25))
+            def g(t):
+                rei = radius*ctx.expj(t)
+                z = x + rei
+                return f(z) / rei**n
+            d = ctx.quadts(g, [0, 2*ctx.pi])
+            v = d * ctx.factorial(n) / (2*ctx.pi)
+        else:
+            raise ValueError("unknown method: %r" % method)
+    finally:
+        ctx.prec = prec
+    return +v
+
+def _partial_diff(ctx, f, xs, orders, options):
+    if not orders:
+        return f()
+    if not sum(orders):
+        return f(*xs)
+    i = 0
+    for i in range(len(orders)):
+        if orders[i]:
+            break
+    order = orders[i]
+    def fdiff_inner(*f_args):
+        def inner(t):
+            return f(*(f_args[:i] + (t,) + f_args[i+1:]))
+        return ctx.diff(inner, f_args[i], order, **options)
+    orders[i] = 0
+    return _partial_diff(ctx, fdiff_inner, xs, orders, options)
+
+@defun
+def diffs(ctx, f, x, n=None, **options):
+    r"""
+    Returns a generator that yields the sequence of derivatives
+
+    .. math ::
+
+        f(x), f'(x), f''(x), \ldots, f^{(k)}(x), \ldots
+
+    With ``method='step'``, :func:`~mpmath.diffs` uses only `O(k)`
+    function evaluations to generate the first `k` derivatives,
+    rather than the roughly `O(k^2)` evaluations
+    required if one calls :func:`~mpmath.diff` `k` separate times.
+
+    With `n < \infty`, the generator stops as soon as the
+    `n`-th derivative has been generated. If the exact number of
+    needed derivatives is known in advance, this is further
+    slightly more efficient.
+
+    Options are the same as for :func:`~mpmath.diff`.
+
+    **Examples**
+
+        >>> from mpmath import *
+        >>> mp.dps = 15
+        >>> nprint(list(diffs(cos, 1, 5)))
+        [0.540302, -0.841471, -0.540302, 0.841471, 0.540302, -0.841471]
+        >>> for i, d in zip(range(6), diffs(cos, 1)):
+        ...     print("%s %s" % (i, d))
+        ...
+        0 0.54030230586814
+        1 -0.841470984807897
+        2 -0.54030230586814
+        3 0.841470984807897
+        4 0.54030230586814
+        5 -0.841470984807897
+
+    """
+    if n is None:
+        n = ctx.inf
+    else:
+        n = int(n)
+    if options.get('method', 'step') != 'step':
+        k = 0
+        while k < n + 1:
+            yield ctx.diff(f, x, k, **options)
+            k += 1
+        return
+    singular = options.get('singular')
+    if singular:
+        yield ctx.diff(f, x, 0, singular=True)
+    else:
+        yield f(ctx.convert(x))
+    if n < 1:
+        return
+    if n == ctx.inf:
+        A, B = 1, 2
+    else:
+        A, B = 1, n+1
+    while 1:
+        callprec = ctx.prec
+        y, norm, workprec = hsteps(ctx, f, x, B, callprec, **options)
+        for k in xrange(A, B):
+            try:
+                ctx.prec = workprec
+                d = ctx.difference(y, k) / norm**k
+            finally:
+                ctx.prec = callprec
+            yield +d
+            if k >= n:
+                return
+        A, B = B, int(A*1.4+1)
+        B = min(B, n)
+
+def iterable_to_function(gen):
+    gen = iter(gen)
+    data = []
+    def f(k):
+        for i in xrange(len(data), k+1):
+            data.append(next(gen))
+        return data[k]
+    return f
+
+@defun
+def diffs_prod(ctx, factors):
+    r"""
+    Given a list of `N` iterables or generators yielding
+    `f_k(x), f'_k(x), f''_k(x), \ldots` for `k = 1, \ldots, N`,
+    generate `g(x), g'(x), g''(x), \ldots` where
+    `g(x) = f_1(x) f_2(x) \cdots f_N(x)`.
+
+    At high precision and for large orders, this is typically more efficient
+    than numerical differentiation if the derivatives of each `f_k(x)`
+    admit direct computation.
+
+    Note: This function does not increase the working precision internally,
+    so guard digits may have to be added externally for full accuracy.
+
+    **Examples**
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> f = lambda x: exp(x)*cos(x)*sin(x)
+        >>> u = diffs(f, 1)
+        >>> v = mp.diffs_prod([diffs(exp,1), diffs(cos,1), diffs(sin,1)])
+        >>> next(u); next(v)
+        1.23586333600241
+        1.23586333600241
+        >>> next(u); next(v)
+        0.104658952245596
+        0.104658952245596
+        >>> next(u); next(v)
+        -5.96999877552086
+        -5.96999877552086
+        >>> next(u); next(v)
+        -12.4632923122697
+        -12.4632923122697
+
+    """
+    N = len(factors)
+    if N == 1:
+        for c in factors[0]:
+            yield c
+    else:
+        u = iterable_to_function(ctx.diffs_prod(factors[:N//2]))
+        v = iterable_to_function(ctx.diffs_prod(factors[N//2:]))
+        n = 0
+        while 1:
+            #yield sum(binomial(n,k)*u(n-k)*v(k) for k in xrange(n+1))
+            s = u(n) * v(0)
+            a = 1
+            for k in xrange(1,n+1):
+                a = a * (n-k+1) // k
+                s += a * u(n-k) * v(k)
+            yield s
+            n += 1
+
+def dpoly(n, _cache={}):
+    """
+    nth differentiation polynomial for exp (Faa di Bruno's formula).
+
+    TODO: most exponents are zero, so maybe a sparse representation
+    would be better.
+    """
+    if n in _cache:
+        return _cache[n]
+    if not _cache:
+        _cache[0] = {(0,):1}
+    R = dpoly(n-1)
+    R = dict((c+(0,),v) for (c,v) in iteritems(R))
+    Ra = {}
+    for powers, count in iteritems(R):
+        powers1 = (powers[0]+1,) + powers[1:]
+        if powers1 in Ra:
+            Ra[powers1] += count
+        else:
+            Ra[powers1] = count
+    for powers, count in iteritems(R):
+        if not sum(powers):
+            continue
+        for k,p in enumerate(powers):
+            if p:
+                powers2 = powers[:k] + (p-1,powers[k+1]+1) + powers[k+2:]
+                if powers2 in Ra:
+                    Ra[powers2] += p*count
+                else:
+                    Ra[powers2] = p*count
+    _cache[n] = Ra
+    return _cache[n]
+
+@defun
+def diffs_exp(ctx, fdiffs):
+    r"""
+    Given an iterable or generator yielding `f(x), f'(x), f''(x), \ldots`
+    generate `g(x), g'(x), g''(x), \ldots` where `g(x) = \exp(f(x))`.
+
+    At high precision and for large orders, this is typically more efficient
+    than numerical differentiation if the derivatives of `f(x)`
+    admit direct computation.
+
+    Note: This function does not increase the working precision internally,
+    so guard digits may have to be added externally for full accuracy.
+
+    **Examples**
+
+    The derivatives of the gamma function can be computed using
+    logarithmic differentiation::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>>
+        >>> def diffs_loggamma(x):
+        ...     yield loggamma(x)
+        ...     i = 0
+        ...     while 1:
+        ...         yield psi(i,x)
+        ...         i += 1
+        ...
+        >>> u = diffs_exp(diffs_loggamma(3))
+        >>> v = diffs(gamma, 3)
+        >>> next(u); next(v)
+        2.0
+        2.0
+        >>> next(u); next(v)
+        1.84556867019693
+        1.84556867019693
+        >>> next(u); next(v)
+        2.49292999190269
+        2.49292999190269
+        >>> next(u); next(v)
+        3.44996501352367
+        3.44996501352367
+
+    """
+    fn = iterable_to_function(fdiffs)
+    f0 = ctx.exp(fn(0))
+    yield f0
+    i = 1
+    while 1:
+        s = ctx.mpf(0)
+        for powers, c in iteritems(dpoly(i)):
+            s += c*ctx.fprod(fn(k+1)**p for (k,p) in enumerate(powers) if p)
+        yield s * f0
+        i += 1
+
+@defun
+def differint(ctx, f, x, n=1, x0=0):
+    r"""
+    Calculates the Riemann-Liouville differintegral, or fractional
+    derivative, defined by
+
+    .. math ::
+
+        \,_{x_0}{\mathbb{D}}^n_xf(x) = \frac{1}{\Gamma(m-n)} \frac{d^m}{dx^m}
+        \int_{x_0}^{x}(x-t)^{m-n-1}f(t)dt
+
+    where `f` is a given (presumably well-behaved) function,
+    `x` is the evaluation point, `n` is the order, and `x_0` is
+    the reference point of integration (`m` is an arbitrary
+    parameter selected automatically).
+
+    With `n = 1`, this is just the standard derivative `f'(x)`; with `n = 2`,
+    the second derivative `f''(x)`, etc. With `n = -1`, it gives
+    `\int_{x_0}^x f(t) dt`, with `n = -2`
+    it gives `\int_{x_0}^x \left( \int_{x_0}^t f(u) du \right) dt`, etc.
+
+    As `n` is permitted to be any number, this operator generalizes
+    iterated differentiation and iterated integration to a single
+    operator with a continuous order parameter.
+
+    **Examples**
+
+    There is an exact formula for the fractional derivative of a
+    monomial `x^p`, which may be used as a reference. For example,
+    the following gives a half-derivative (order 0.5)::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> x = mpf(3); p = 2; n = 0.5
+        >>> differint(lambda t: t**p, x, n)
+        7.81764019044672
+        >>> gamma(p+1)/gamma(p-n+1) * x**(p-n)
+        7.81764019044672
+
+    Another useful test function is the exponential function, whose
+    integration / differentiation formula easy generalizes
+    to arbitrary order. Here we first compute a third derivative,
+    and then a triply nested integral. (The reference point `x_0`
+    is set to `-\infty` to avoid nonzero endpoint terms.)::
+
+        >>> differint(lambda x: exp(pi*x), -1.5, 3)
+        0.278538406900792
+        >>> exp(pi*-1.5) * pi**3
+        0.278538406900792
+        >>> differint(lambda x: exp(pi*x), 3.5, -3, -inf)
+        1922.50563031149
+        >>> exp(pi*3.5) / pi**3
+        1922.50563031149
+
+    However, for noninteger `n`, the differentiation formula for the
+    exponential function must be modified to give the same result as the
+    Riemann-Liouville differintegral::
+
+        >>> x = mpf(3.5)
+        >>> c = pi
+        >>> n = 1+2*j
+        >>> differint(lambda x: exp(c*x), x, n)
+        (-123295.005390743 + 140955.117867654j)
+        >>> x**(-n) * exp(c)**x * (x*c)**n * gammainc(-n, 0, x*c) / gamma(-n)
+        (-123295.005390743 + 140955.117867654j)
+
+
+    """
+    m = max(int(ctx.ceil(ctx.re(n)))+1, 1)
+    r = m-n-1
+    g = lambda x: ctx.quad(lambda t: (x-t)**r * f(t), [x0, x])
+    return ctx.diff(g, x, m) / ctx.gamma(m-n)
+
+@defun
+def diffun(ctx, f, n=1, **options):
+    r"""
+    Given a function `f`, returns a function `g(x)` that evaluates the nth
+    derivative `f^{(n)}(x)`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> cos2 = diffun(sin)
+        >>> sin2 = diffun(sin, 4)
+        >>> cos(1.3), cos2(1.3)
+        (0.267498828624587, 0.267498828624587)
+        >>> sin(1.3), sin2(1.3)
+        (0.963558185417193, 0.963558185417193)
+
+    The function `f` must support arbitrary precision evaluation.
+    See :func:`~mpmath.diff` for additional details and supported
+    keyword options.
+    """
+    if n == 0:
+        return f
+    def g(x):
+        return ctx.diff(f, x, n, **options)
+    return g
+
+@defun
+def taylor(ctx, f, x, n, **options):
+    r"""
+    Produces a degree-`n` Taylor polynomial around the point `x` of the
+    given function `f`. The coefficients are returned as a list.
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> nprint(chop(taylor(sin, 0, 5)))
+        [0.0, 1.0, 0.0, -0.166667, 0.0, 0.00833333]
+
+    The coefficients are computed using high-order numerical
+    differentiation. The function must be possible to evaluate
+    to arbitrary precision. See :func:`~mpmath.diff` for additional details
+    and supported keyword options.
+
+    Note that to evaluate the Taylor polynomial as an approximation
+    of `f`, e.g. with :func:`~mpmath.polyval`, the coefficients must be reversed,
+    and the point of the Taylor expansion must be subtracted from
+    the argument:
+
+        >>> p = taylor(exp, 2.0, 10)
+        >>> polyval(p[::-1], 2.5 - 2.0)
+        12.1824939606092
+        >>> exp(2.5)
+        12.1824939607035
+
+    """
+    gen = enumerate(ctx.diffs(f, x, n, **options))
+    if options.get("chop", True):
+        return [ctx.chop(d)/ctx.factorial(i) for i, d in gen]
+    else:
+        return [d/ctx.factorial(i) for i, d in gen]
+
+@defun
+def pade(ctx, a, L, M):
+    r"""
+    Computes a Pade approximation of degree `(L, M)` to a function.
+    Given at least `L+M+1` Taylor coefficients `a` approximating
+    a function `A(x)`, :func:`~mpmath.pade` returns coefficients of
+    polynomials `P, Q` satisfying
+
+    .. math ::
+
+        P = \sum_{k=0}^L p_k x^k
+
+        Q = \sum_{k=0}^M q_k x^k
+
+        Q_0 = 1
+
+        A(x) Q(x) = P(x) + O(x^{L+M+1})
+
+    `P(x)/Q(x)` can provide a good approximation to an analytic function
+    beyond the radius of convergence of its Taylor series (example
+    from G.A. Baker 'Essentials of Pade Approximants' Academic Press,
+    Ch.1A)::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> one = mpf(1)
+        >>> def f(x):
+        ...     return sqrt((one + 2*x)/(one + x))
+        ...
+        >>> a = taylor(f, 0, 6)
+        >>> p, q = pade(a, 3, 3)
+        >>> x = 10
+        >>> polyval(p[::-1], x)/polyval(q[::-1], x)
+        1.38169105566806
+        >>> f(x)
+        1.38169855941551
+
+    """
+    # To determine L+1 coefficients of P and M coefficients of Q
+    # L+M+1 coefficients of A must be provided
+    if len(a) < L+M+1:
+        raise ValueError("L+M+1 Coefficients should be provided")
+
+    if M == 0:
+        if L == 0:
+            return [ctx.one], [ctx.one]
+        else:
+            return a[:L+1], [ctx.one]
+
+    # Solve first
+    # a[L]*q[1] + ... + a[L-M+1]*q[M] = -a[L+1]
+    # ...
+    # a[L+M-1]*q[1] + ... + a[L]*q[M] = -a[L+M]
+    A = ctx.matrix(M)
+    for j in range(M):
+        for i in range(min(M, L+j+1)):
+            A[j, i] = a[L+j-i]
+    v = -ctx.matrix(a[(L+1):(L+M+1)])
+    x = ctx.lu_solve(A, v)
+    q = [ctx.one] + list(x)
+    # compute p
+    p = [0]*(L+1)
+    for i in range(L+1):
+        s = a[i]
+        for j in range(1, min(M,i) + 1):
+            s += q[j]*a[i-j]
+        p[i] = s
+    return p, q

lib/python3.11/site-packages/mpmath/calculus/extrapolation.py

@@ -0,0 +1,2115 @@
+try:
+    from itertools import izip
+except ImportError:
+    izip = zip
+
+from ..libmp.backend import xrange
+from .calculus import defun
+
+try:
+    next = next
+except NameError:
+    next = lambda _: _.next()
+
+@defun
+def richardson(ctx, seq):
+    r"""
+    Given a list ``seq`` of the first `N` elements of a slowly convergent
+    infinite sequence, :func:`~mpmath.richardson` computes the `N`-term
+    Richardson extrapolate for the limit.
+
+    :func:`~mpmath.richardson` returns `(v, c)` where `v` is the estimated
+    limit and `c` is the magnitude of the largest weight used during the
+    computation. The weight provides an estimate of the precision
+    lost to cancellation. Due to cancellation effects, the sequence must
+    be typically be computed at a much higher precision than the target
+    accuracy of the extrapolation.
+
+    **Applicability and issues**
+
+    The `N`-step Richardson extrapolation algorithm used by
+    :func:`~mpmath.richardson` is described in [1].
+
+    Richardson extrapolation only works for a specific type of sequence,
+    namely one converging like partial sums of
+    `P(1)/Q(1) + P(2)/Q(2) + \ldots` where `P` and `Q` are polynomials.
+    When the sequence does not convergence at such a rate
+    :func:`~mpmath.richardson` generally produces garbage.
+
+    Richardson extrapolation has the advantage of being fast: the `N`-term
+    extrapolate requires only `O(N)` arithmetic operations, and usually
+    produces an estimate that is accurate to `O(N)` digits. Contrast with
+    the Shanks transformation (see :func:`~mpmath.shanks`), which requires
+    `O(N^2)` operations.
+
+    :func:`~mpmath.richardson` is unable to produce an estimate for the
+    approximation error. One way to estimate the error is to perform
+    two extrapolations with slightly different `N` and comparing the
+    results.
+
+    Richardson extrapolation does not work for oscillating sequences.
+    As a simple workaround, :func:`~mpmath.richardson` detects if the last
+    three elements do not differ monotonically, and in that case
+    applies extrapolation only to the even-index elements.
+
+    **Example**
+
+    Applying Richardson extrapolation to the Leibniz series for `\pi`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 30; mp.pretty = True
+        >>> S = [4*sum(mpf(-1)**n/(2*n+1) for n in range(m))
+        ...     for m in range(1,30)]
+        >>> v, c = richardson(S[:10])
+        >>> v
+        3.2126984126984126984126984127
+        >>> nprint([v-pi, c])
+        [0.0711058, 2.0]
+
+        >>> v, c = richardson(S[:30])
+        >>> v
+        3.14159265468624052829954206226
+        >>> nprint([v-pi, c])
+        [1.09645e-9, 20833.3]
+
+    **References**
+
+    1. [BenderOrszag]_ pp. 375-376
+
+    """
+    if len(seq) < 3:
+        raise ValueError("seq should be of minimum length 3")
+    if ctx.sign(seq[-1]-seq[-2]) != ctx.sign(seq[-2]-seq[-3]):
+        seq = seq[::2]
+    N = len(seq)//2-1
+    s = ctx.zero
+    # The general weight is c[k] = (N+k)**N * (-1)**(k+N) / k! / (N-k)!
+    # To avoid repeated factorials, we simplify the quotient
+    # of successive weights to obtain a recurrence relation
+    c = (-1)**N * N**N / ctx.mpf(ctx._ifac(N))
+    maxc = 1
+    for k in xrange(N+1):
+        s += c * seq[N+k]
+        maxc = max(abs(c), maxc)
+        c *= (k-N)*ctx.mpf(k+N+1)**N
+        c /= ((1+k)*ctx.mpf(k+N)**N)
+    return s, maxc
+
+@defun
+def shanks(ctx, seq, table=None, randomized=False):
+    r"""
+    Given a list ``seq`` of the first `N` elements of a slowly
+    convergent infinite sequence `(A_k)`, :func:`~mpmath.shanks` computes the iterated
+    Shanks transformation `S(A), S(S(A)), \ldots, S^{N/2}(A)`. The Shanks
+    transformation often provides strong convergence acceleration,
+    especially if the sequence is oscillating.
+
+    The iterated Shanks transformation is computed using the Wynn
+    epsilon algorithm (see [1]). :func:`~mpmath.shanks` returns the full
+    epsilon table generated by Wynn's algorithm, which can be read
+    off as follows:
+
+    * The table is a list of lists forming a lower triangular matrix,
+      where higher row and column indices correspond to more accurate
+      values.
+    * The columns with even index hold dummy entries (required for the
+      computation) and the columns with odd index hold the actual
+      extrapolates.
+    * The last element in the last row is typically the most
+      accurate estimate of the limit.
+    * The difference to the third last element in the last row
+      provides an estimate of the approximation error.
+    * The magnitude of the second last element provides an estimate
+      of the numerical accuracy lost to cancellation.
+
+    For convenience, so the extrapolation is stopped at an odd index
+    so that ``shanks(seq)[-1][-1]`` always gives an estimate of the
+    limit.
+
+    Optionally, an existing table can be passed to :func:`~mpmath.shanks`.
+    This can be used to efficiently extend a previous computation after
+    new elements have been appended to the sequence. The table will
+    then be updated in-place.
+
+    **The Shanks transformation**
+
+    The Shanks transformation is defined as follows (see [2]): given
+    the input sequence `(A_0, A_1, \ldots)`, the transformed sequence is
+    given by
+
+    .. math ::
+
+        S(A_k) = \frac{A_{k+1}A_{k-1}-A_k^2}{A_{k+1}+A_{k-1}-2 A_k}
+
+    The Shanks transformation gives the exact limit `A_{\infty}` in a
+    single step if `A_k = A + a q^k`. Note in particular that it
+    extrapolates the exact sum of a geometric series in a single step.
+
+    Applying the Shanks transformation once often improves convergence
+    substantially for an arbitrary sequence, but the optimal effect is
+    obtained by applying it iteratively:
+    `S(S(A_k)), S(S(S(A_k))), \ldots`.
+
+    Wynn's epsilon algorithm provides an efficient way to generate
+    the table of iterated Shanks transformations. It reduces the
+    computation of each element to essentially a single division, at
+    the cost of requiring dummy elements in the table. See [1] for
+    details.
+
+    **Precision issues**
+
+    Due to cancellation effects, the sequence must be typically be
+    computed at a much higher precision than the target accuracy
+    of the extrapolation.
+
+    If the Shanks transformation converges to the exact limit (such
+    as if the sequence is a geometric series), then a division by
+    zero occurs. By default, :func:`~mpmath.shanks` handles this case by
+    terminating the iteration and returning the table it has
+    generated so far. With *randomized=True*, it will instead
+    replace the zero by a pseudorandom number close to zero.
+    (TODO: find a better solution to this problem.)
+
+    **Examples**
+
+    We illustrate by applying Shanks transformation to the Leibniz
+    series for `\pi`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 50
+        >>> S = [4*sum(mpf(-1)**n/(2*n+1) for n in range(m))
+        ...     for m in range(1,30)]
+        >>>
+        >>> T = shanks(S[:7])
+        >>> for row in T:
+        ...     nprint(row)
+        ...
+        [-0.75]
+        [1.25, 3.16667]
+        [-1.75, 3.13333, -28.75]
+        [2.25, 3.14524, 82.25, 3.14234]
+        [-2.75, 3.13968, -177.75, 3.14139, -969.937]
+        [3.25, 3.14271, 327.25, 3.14166, 3515.06, 3.14161]
+
+    The extrapolated accuracy is about 4 digits, and about 4 digits
+    may have been lost due to cancellation::
+
+        >>> L = T[-1]
+        >>> nprint([abs(L[-1] - pi), abs(L[-1] - L[-3]), abs(L[-2])])
+        [2.22532e-5, 4.78309e-5, 3515.06]
+
+    Now we extend the computation::
+
+        >>> T = shanks(S[:25], T)
+        >>> L = T[-1]
+        >>> nprint([abs(L[-1] - pi), abs(L[-1] - L[-3]), abs(L[-2])])
+        [3.75527e-19, 1.48478e-19, 2.96014e+17]
+
+    The value for pi is now accurate to 18 digits. About 18 digits may
+    also have been lost to cancellation.
+
+    Here is an example with a geometric series, where the convergence
+    is immediate (the sum is exactly 1)::
+
+        >>> mp.dps = 15
+        >>> for row in shanks([0.5, 0.75, 0.875, 0.9375, 0.96875]):
+        ...     nprint(row)
+        [4.0]
+        [8.0, 1.0]
+
+    **References**
+
+    1. [GravesMorris]_
+
+    2. [BenderOrszag]_ pp. 368-375
+
+    """
+    if len(seq) < 2:
+        raise ValueError("seq should be of minimum length 2")
+    if table:
+        START = len(table)
+    else:
+        START = 0
+        table = []
+    STOP = len(seq) - 1
+    if STOP & 1:
+        STOP -= 1
+    one = ctx.one
+    eps = +ctx.eps
+    if randomized:
+        from random import Random
+        rnd = Random()
+        rnd.seed(START)
+    for i in xrange(START, STOP):
+        row = []
+        for j in xrange(i+1):
+            if j == 0:
+                a, b = 0, seq[i+1]-seq[i]
+            else:
+                if j == 1:
+                    a = seq[i]
+                else:
+                    a = table[i-1][j-2]
+                b = row[j-1] - table[i-1][j-1]
+            if not b:
+                if randomized:
+                    b = (1 + rnd.getrandbits(10))*eps
+                elif i & 1:
+                    return table[:-1]
+                else:
+                    return table
+            row.append(a + one/b)
+        table.append(row)
+    return table
+
+
+class levin_class:
+    # levin: Copyright 2013 Timo Hartmann (thartmann15 at gmail.com)
+    r"""
+    This interface implements Levin's (nonlinear) sequence transformation for
+    convergence acceleration and summation of divergent series. It performs
+    better than the Shanks/Wynn-epsilon algorithm for logarithmic convergent
+    or alternating divergent series.
+
+    Let *A* be the series we want to sum:
+
+    .. math ::
+
+        A = \sum_{k=0}^{\infty} a_k
+
+    Attention: all `a_k` must be non-zero!
+
+    Let `s_n` be the partial sums of this series:
+
+    .. math ::
+
+        s_n = \sum_{k=0}^n a_k.
+
+    **Methods**
+
+    Calling ``levin`` returns an object with the following methods.
+
+    ``update(...)`` works with the list of individual terms `a_k` of *A*, and
+    ``update_step(...)`` works with the list of partial sums `s_k` of *A*:
+
+    .. code ::
+
+        v, e = ...update([a_0, a_1,..., a_k])
+        v, e = ...update_psum([s_0, s_1,..., s_k])
+
+    ``step(...)`` works with the individual terms `a_k` and ``step_psum(...)``
+    works with the partial sums `s_k`:
+
+    .. code ::
+
+        v, e = ...step(a_k)
+        v, e = ...step_psum(s_k)
+
+    *v* is the current estimate for *A*, and *e* is an error estimate which is
+    simply the difference between the current estimate and the last estimate.
+    One should not mix ``update``, ``update_psum``, ``step`` and ``step_psum``.
+
+    **A word of caution**
+
+    One can only hope for good results (i.e. convergence acceleration or
+    resummation) if the `s_n` have some well defind asymptotic behavior for
+    large `n` and are not erratic or random. Furthermore one usually needs very
+    high working precision because of the numerical cancellation. If the working
+    precision is insufficient, levin may produce silently numerical garbage.
+    Furthermore even if the Levin-transformation converges, in the general case
+    there is no proof that the result is mathematically sound. Only for very
+    special classes of problems one can prove that the Levin-transformation
+    converges to the expected result (for example Stieltjes-type integrals).
+    Furthermore the Levin-transform is quite expensive (i.e. slow) in comparison
+    to Shanks/Wynn-epsilon, Richardson & co.
+    In summary one can say that the Levin-transformation is powerful but
+    unreliable and that it may need a copious amount of working precision.
+
+    The Levin transform has several variants differing in the choice of weights.
+    Some variants are better suited for the possible flavours of convergence
+    behaviour of *A* than other variants:
+
+    .. code ::
+
+       convergence behaviour   levin-u   levin-t   levin-v   shanks/wynn-epsilon
+
+       logarithmic               +         -         +           -
+       linear                    +         +         +           +
+       alternating divergent     +         +         +           +
+
+         "+" means the variant is suitable,"-" means the variant is not suitable;
+         for comparison the Shanks/Wynn-epsilon transform is listed, too.
+
+    The variant is controlled though the variant keyword (i.e. ``variant="u"``,
+    ``variant="t"`` or ``variant="v"``). Overall "u" is probably the best choice.
+
+    Finally it is possible to use the Sidi-S transform instead of the Levin transform
+    by using the keyword ``method='sidi'``. The Sidi-S transform works better than the
+    Levin transformation for some divergent series (see the examples).
+
+    Parameters:
+
+    .. code ::
+
+       method      "levin" or "sidi" chooses either the Levin or the Sidi-S transformation
+       variant     "u","t" or "v" chooses the weight variant.
+
+    The Levin transform is also accessible through the nsum interface.
+    ``method="l"`` or ``method="levin"`` select the normal Levin transform while
+    ``method="sidi"``
+    selects the Sidi-S transform. The variant is in both cases selected through the
+    levin_variant keyword. The stepsize in :func:`~mpmath.nsum` must not be chosen too large, otherwise
+    it will miss the point where the Levin transform converges resulting in numerical
+    overflow/garbage. For highly divergent series a copious amount of working precision
+    must be chosen.
+
+    **Examples**
+
+    First we sum the zeta function::
+
+        >>> from mpmath import mp
+        >>> mp.prec = 53
+        >>> eps = mp.mpf(mp.eps)
+        >>> with mp.extraprec(2 * mp.prec): # levin needs a high working precision
+        ...     L = mp.levin(method = "levin", variant = "u")
+        ...     S, s, n = [], 0, 1
+        ...     while 1:
+        ...         s += mp.one / (n * n)
+        ...         n += 1
+        ...         S.append(s)
+        ...         v, e = L.update_psum(S)
+        ...         if e < eps:
+        ...             break
+        ...         if n > 1000: raise RuntimeError("iteration limit exceeded")
+        >>> print(mp.chop(v - mp.pi ** 2 / 6))
+        0.0
+        >>> w = mp.nsum(lambda n: 1 / (n*n), [1, mp.inf], method = "levin", levin_variant = "u")
+        >>> print(mp.chop(v - w))
+        0.0
+
+    Now we sum the zeta function outside its range of convergence
+    (attention: This does not work at the negative integers!)::
+
+        >>> eps = mp.mpf(mp.eps)
+        >>> with mp.extraprec(2 * mp.prec): # levin needs a high working precision
+        ...     L = mp.levin(method = "levin", variant = "v")
+        ...     A, n = [], 1
+        ...     while 1:
+        ...         s = mp.mpf(n) ** (2 + 3j)
+        ...         n += 1
+        ...         A.append(s)
+        ...         v, e = L.update(A)
+        ...         if e < eps:
+        ...             break
+        ...         if n > 1000: raise RuntimeError("iteration limit exceeded")
+        >>> print(mp.chop(v - mp.zeta(-2-3j)))
+        0.0
+        >>> w = mp.nsum(lambda n: n ** (2 + 3j), [1, mp.inf], method = "levin", levin_variant = "v")
+        >>> print(mp.chop(v - w))
+        0.0
+
+    Now we sum the divergent asymptotic expansion of an integral related to the
+    exponential integral (see also [2] p.373). The Sidi-S transform works best here::
+
+        >>> z = mp.mpf(10)
+        >>> exact = mp.quad(lambda x: mp.exp(-x)/(1+x/z),[0,mp.inf])
+        >>> # exact = z * mp.exp(z) * mp.expint(1,z) # this is the symbolic expression for the integral
+        >>> eps = mp.mpf(mp.eps)
+        >>> with mp.extraprec(2 * mp.prec): # high working precisions are mandatory for divergent resummation
+        ...     L = mp.levin(method = "sidi", variant = "t")
+        ...     n = 0
+        ...     while 1:
+        ...         s = (-1)**n * mp.fac(n) * z ** (-n)
+        ...         v, e = L.step(s)
+        ...         n += 1
+        ...         if e < eps:
+        ...             break
+        ...         if n > 1000: raise RuntimeError("iteration limit exceeded")
+        >>> print(mp.chop(v - exact))
+        0.0
+        >>> w = mp.nsum(lambda n: (-1) ** n * mp.fac(n) * z ** (-n), [0, mp.inf], method = "sidi", levin_variant = "t")
+        >>> print(mp.chop(v - w))
+        0.0
+
+    Another highly divergent integral is also summable::
+
+        >>> z = mp.mpf(2)
+        >>> eps = mp.mpf(mp.eps)
+        >>> exact = mp.quad(lambda x: mp.exp( -x * x / 2 - z * x ** 4), [0,mp.inf]) * 2 / mp.sqrt(2 * mp.pi)
+        >>> # exact = mp.exp(mp.one / (32 * z)) * mp.besselk(mp.one / 4, mp.one / (32 * z)) / (4 * mp.sqrt(z * mp.pi)) # this is the symbolic expression for the integral
+        >>> with mp.extraprec(7 * mp.prec):  # we need copious amount of precision to sum this highly divergent series
+        ...     L = mp.levin(method = "levin", variant = "t")
+        ...     n, s = 0, 0
+        ...     while 1:
+        ...         s += (-z)**n * mp.fac(4 * n) / (mp.fac(n) * mp.fac(2 * n) * (4 ** n))
+        ...         n += 1
+        ...         v, e = L.step_psum(s)
+        ...         if e < eps:
+        ...             break
+        ...         if n > 1000: raise RuntimeError("iteration limit exceeded")
+        >>> print(mp.chop(v - exact))
+        0.0
+        >>> w = mp.nsum(lambda n: (-z)**n * mp.fac(4 * n) / (mp.fac(n) * mp.fac(2 * n) * (4 ** n)),
+        ...   [0, mp.inf], method = "levin", levin_variant = "t", workprec = 8*mp.prec, steps = [2] + [1 for x in xrange(1000)])
+        >>> print(mp.chop(v - w))
+        0.0
+
+    These examples run with 15-20 decimal digits precision. For higher precision the
+    working precision must be raised.
+
+    **Examples for nsum**
+
+    Here we calculate Euler's constant as the constant term in the Laurent
+    expansion of `\zeta(s)` at `s=1`. This sum converges extremly slowly because of
+    the logarithmic convergence behaviour of the Dirichlet series for zeta::
+
+        >>> mp.dps = 30
+        >>> z = mp.mpf(10) ** (-10)
+        >>> a = mp.nsum(lambda n: n**(-(1+z)), [1, mp.inf], method = "l") - 1 / z
+        >>> print(mp.chop(a - mp.euler, tol = 1e-10))
+        0.0
+
+    The Sidi-S transform performs excellently for the alternating series of `\log(2)`::
+
+        >>> a = mp.nsum(lambda n: (-1)**(n-1) / n, [1, mp.inf], method = "sidi")
+        >>> print(mp.chop(a - mp.log(2)))
+        0.0
+
+    Hypergeometric series can also be summed outside their range of convergence.
+    The stepsize in :func:`~mpmath.nsum` must not be chosen too large, otherwise it will miss the
+    point where the Levin transform converges resulting in numerical overflow/garbage::
+
+        >>> z = 2 + 1j
+        >>> exact = mp.hyp2f1(2 / mp.mpf(3), 4 / mp.mpf(3), 1 / mp.mpf(3), z)
+        >>> f = lambda n: mp.rf(2 / mp.mpf(3), n) * mp.rf(4 / mp.mpf(3), n) * z**n / (mp.rf(1 / mp.mpf(3), n) * mp.fac(n))
+        >>> v = mp.nsum(f, [0, mp.inf], method = "levin", steps = [10 for x in xrange(1000)])
+        >>> print(mp.chop(exact-v))
+        0.0
+
+    References:
+
+      [1] E.J. Weniger - "Nonlinear Sequence Transformations for the Acceleration of
+          Convergence and the Summation of Divergent Series" arXiv:math/0306302
+
+      [2] A. Sidi - "Pratical Extrapolation Methods"
+
+      [3] H.H.H. Homeier - "Scalar Levin-Type Sequence Transformations" arXiv:math/0005209
+
+    """
+
+    def __init__(self, method = "levin", variant = "u"):
+        self.variant = variant
+        self.n = 0
+        self.a0 = 0
+        self.theta = 1
+        self.A = []
+        self.B = []
+        self.last = 0
+        self.last_s = False
+
+        if method == "levin":
+            self.factor = self.factor_levin
+        elif method == "sidi":
+            self.factor = self.factor_sidi
+        else:
+            raise ValueError("levin: unknown method \"%s\"" % method)
+
+    def factor_levin(self, i):
+        # original levin
+        # [1] p.50,e.7.5-7 (with n-j replaced by i)
+        return (self.theta + i) * (self.theta + self.n - 1) ** (self.n - i - 2) / self.ctx.mpf(self.theta + self.n) ** (self.n - i - 1)
+
+    def factor_sidi(self, i):
+        # sidi analogon to levin (factorial series)
+        # [1] p.59,e.8.3-16 (with n-j replaced by i)
+        return (self.theta + self.n - 1) * (self.theta + self.n - 2) / self.ctx.mpf((self.theta + 2 * self.n - i - 2) * (self.theta + 2 * self.n - i - 3))
+
+    def run(self, s, a0, a1 = 0):
+        if self.variant=="t":
+            # levin t
+            w=a0
+        elif self.variant=="u":
+            # levin u
+            w=a0*(self.theta+self.n)
+        elif self.variant=="v":
+            # levin v
+            w=a0*a1/(a0-a1)
+        else:
+            assert False, "unknown variant"
+
+        if w==0:
+            raise ValueError("levin: zero weight")
+
+        self.A.append(s/w)
+        self.B.append(1/w)
+
+        for i in range(self.n-1,-1,-1):
+            if i==self.n-1:
+                f=1
+            else:
+                f=self.factor(i)
+
+            self.A[i]=self.A[i+1]-f*self.A[i]
+            self.B[i]=self.B[i+1]-f*self.B[i]
+
+        self.n+=1
+
+    ###########################################################################
+
+    def update_psum(self,S):
+        """
+        This routine applies the convergence acceleration to the list of partial sums.
+
+        A   = sum(a_k, k = 0..infinity)
+        s_n = sum(a_k, k = 0..n)
+
+        v, e = ...update_psum([s_0, s_1,..., s_k])
+
+        output:
+          v      current estimate of the series A
+          e      an error estimate which is simply the difference between the current
+                 estimate and the last estimate.
+        """
+
+        if self.variant!="v":
+            if self.n==0:
+                self.run(S[0],S[0])
+            while self.n<len(S):
+                self.run(S[self.n],S[self.n]-S[self.n-1])
+        else:
+            if len(S)==1:
+                self.last=0
+                return S[0],abs(S[0])
+
+            if self.n==0:
+                self.a1=S[1]-S[0]
+                self.run(S[0],S[0],self.a1)
+
+            while self.n<len(S)-1:
+                na1=S[self.n+1]-S[self.n]
+                self.run(S[self.n],self.a1,na1)
+                self.a1=na1
+
+        value=self.A[0]/self.B[0]
+        err=abs(value-self.last)
+        self.last=value
+
+        return value,err
+
+    def update(self,X):
+        """
+        This routine applies the convergence acceleration to the list of individual terms.
+
+        A = sum(a_k, k = 0..infinity)
+
+        v, e = ...update([a_0, a_1,..., a_k])
+
+        output:
+          v      current estimate of the series A
+          e      an error estimate which is simply the difference between the current
+                 estimate and the last estimate.
+        """
+
+        if self.variant!="v":
+            if self.n==0:
+                self.s=X[0]
+                self.run(self.s,X[0])
+            while self.n<len(X):
+                self.s+=X[self.n]
+                self.run(self.s,X[self.n])
+        else:
+            if len(X)==1:
+                self.last=0
+                return X[0],abs(X[0])
+
+            if self.n==0:
+                self.s=X[0]
+                self.run(self.s,X[0],X[1])
+
+            while self.n<len(X)-1:
+                self.s+=X[self.n]
+                self.run(self.s,X[self.n],X[self.n+1])
+
+        value=self.A[0]/self.B[0]
+        err=abs(value-self.last)
+        self.last=value
+
+        return value,err
+
+    ###########################################################################
+
+    def step_psum(self,s):
+        """
+        This routine applies the convergence acceleration to the partial sums.
+
+        A   = sum(a_k, k = 0..infinity)
+        s_n = sum(a_k, k = 0..n)
+
+        v, e = ...step_psum(s_k)
+
+        output:
+          v      current estimate of the series A
+          e      an error estimate which is simply the difference between the current
+                 estimate and the last estimate.
+        """
+
+        if self.variant!="v":
+            if self.n==0:
+                self.last_s=s
+                self.run(s,s)
+            else:
+                self.run(s,s-self.last_s)
+                self.last_s=s
+        else:
+            if isinstance(self.last_s,bool):
+                self.last_s=s
+                self.last_w=s
+                self.last=0
+                return s,abs(s)
+
+            na1=s-self.last_s
+            self.run(self.last_s,self.last_w,na1)
+            self.last_w=na1
+            self.last_s=s
+
+        value=self.A[0]/self.B[0]
+        err=abs(value-self.last)
+        self.last=value
+
+        return value,err
+
+    def step(self,x):
+        """
+        This routine applies the convergence acceleration to the individual terms.
+
+        A = sum(a_k, k = 0..infinity)
+
+        v, e = ...step(a_k)
+
+        output:
+          v      current estimate of the series A
+          e      an error estimate which is simply the difference between the current
+                 estimate and the last estimate.
+        """
+
+        if self.variant!="v":
+            if self.n==0:
+                self.s=x
+                self.run(self.s,x)
+            else:
+                self.s+=x
+                self.run(self.s,x)
+        else:
+            if isinstance(self.last_s,bool):
+                self.last_s=x
+                self.s=0
+                self.last=0
+                return x,abs(x)
+
+            self.s+=self.last_s
+            self.run(self.s,self.last_s,x)
+            self.last_s=x
+
+        value=self.A[0]/self.B[0]
+        err=abs(value-self.last)
+        self.last=value
+
+        return value,err
+
+def levin(ctx, method = "levin", variant = "u"):
+    L = levin_class(method = method, variant = variant)
+    L.ctx = ctx
+    return L
+
+levin.__doc__ = levin_class.__doc__
+defun(levin)
+
+
+class cohen_alt_class:
+    # cohen_alt: Copyright 2013 Timo Hartmann (thartmann15 at gmail.com)
+    r"""
+    This interface implements the convergence acceleration of alternating series
+    as described in H. Cohen, F.R. Villegas, D. Zagier - "Convergence Acceleration
+    of Alternating Series". This series transformation works only well if the
+    individual terms of the series have an alternating sign. It belongs to the
+    class of linear series transformations (in contrast to the Shanks/Wynn-epsilon
+    or Levin transform). This series transformation is also able to sum some types
+    of divergent series. See the paper under which conditions this resummation is
+    mathematical sound.
+
+    Let *A* be the series we want to sum:
+
+    .. math ::
+
+        A = \sum_{k=0}^{\infty} a_k
+
+    Let `s_n` be the partial sums of this series:
+
+    .. math ::
+
+        s_n = \sum_{k=0}^n a_k.
+
+
+    **Interface**
+
+    Calling ``cohen_alt`` returns an object with the following methods.
+
+    Then ``update(...)`` works with the list of individual terms `a_k` and
+    ``update_psum(...)`` works with the list of partial sums `s_k`:
+
+    .. code ::
+
+        v, e = ...update([a_0, a_1,..., a_k])
+        v, e = ...update_psum([s_0, s_1,..., s_k])
+
+    *v* is the current estimate for *A*, and *e* is an error estimate which is
+    simply the difference between the current estimate and the last estimate.
+
+    **Examples**
+
+    Here we compute the alternating zeta function using ``update_psum``::
+
+        >>> from mpmath import mp
+        >>> AC = mp.cohen_alt()
+        >>> S, s, n = [], 0, 1
+        >>> while 1:
+        ...     s += -((-1) ** n) * mp.one / (n * n)
+        ...     n += 1
+        ...     S.append(s)
+        ...     v, e = AC.update_psum(S)
+        ...     if e < mp.eps:
+        ...         break
+        ...     if n > 1000: raise RuntimeError("iteration limit exceeded")
+        >>> print(mp.chop(v - mp.pi ** 2 / 12))
+        0.0
+
+    Here we compute the product `\prod_{n=1}^{\infty} \Gamma(1+1/(2n-1)) / \Gamma(1+1/(2n))`::
+
+        >>> A = []
+        >>> AC = mp.cohen_alt()
+        >>> n = 1
+        >>> while 1:
+        ...     A.append( mp.loggamma(1 + mp.one / (2 * n - 1)))
+        ...     A.append(-mp.loggamma(1 + mp.one / (2 * n)))
+        ...     n += 1
+        ...     v, e = AC.update(A)
+        ...     if e < mp.eps:
+        ...         break
+        ...     if n > 1000: raise RuntimeError("iteration limit exceeded")
+        >>> v = mp.exp(v)
+        >>> print(mp.chop(v - 1.06215090557106, tol = 1e-12))
+        0.0
+
+    ``cohen_alt`` is also accessible through the :func:`~mpmath.nsum` interface::
+
+        >>> v = mp.nsum(lambda n: (-1)**(n-1) / n, [1, mp.inf], method = "a")
+        >>> print(mp.chop(v - mp.log(2)))
+        0.0
+        >>> v = mp.nsum(lambda n: (-1)**n / (2 * n + 1), [0, mp.inf], method = "a")
+        >>> print(mp.chop(v - mp.pi / 4))
+        0.0
+        >>> v = mp.nsum(lambda n: (-1)**n * mp.log(n) * n, [1, mp.inf], method = "a")
+        >>> print(mp.chop(v - mp.diff(lambda s: mp.altzeta(s), -1)))
+        0.0
+
+    """
+
+    def __init__(self):
+        self.last=0
+
+    def update(self, A):
+        """
+        This routine applies the convergence acceleration to the list of individual terms.
+
+        A    = sum(a_k, k = 0..infinity)
+
+        v, e = ...update([a_0, a_1,..., a_k])
+
+        output:
+          v      current estimate of the series A
+          e      an error estimate which is simply the difference between the current
+                 estimate and the last estimate.
+        """
+
+        n = len(A)
+        d = (3 + self.ctx.sqrt(8)) ** n
+        d = (d + 1 / d) / 2
+        b = -self.ctx.one
+        c = -d
+        s = 0
+
+        for k in xrange(n):
+            c = b - c
+            if k % 2 == 0:
+                s = s + c * A[k]
+            else:
+                s = s - c * A[k]
+            b = 2 * (k + n) * (k - n) * b / ((2 * k + 1) * (k + self.ctx.one))
+
+        value = s / d
+
+        err = abs(value - self.last)
+        self.last = value
+
+        return value, err
+
+    def update_psum(self, S):
+        """
+        This routine applies the convergence acceleration to the list of partial sums.
+
+        A   = sum(a_k, k = 0..infinity)
+        s_n = sum(a_k ,k = 0..n)
+
+        v, e = ...update_psum([s_0, s_1,..., s_k])
+
+        output:
+          v      current estimate of the series A
+          e      an error estimate which is simply the difference between the current
+                 estimate and the last estimate.
+        """
+
+        n = len(S)
+        d = (3 + self.ctx.sqrt(8)) ** n
+        d = (d + 1 / d) / 2
+        b = self.ctx.one
+        s = 0
+
+        for k in xrange(n):
+            b = 2 * (n + k) * (n - k) * b / ((2 * k + 1) * (k + self.ctx.one))
+            s += b * S[k]
+
+        value = s / d
+
+        err = abs(value - self.last)
+        self.last = value
+
+        return value, err
+
+def cohen_alt(ctx):
+    L = cohen_alt_class()
+    L.ctx = ctx
+    return L
+
+cohen_alt.__doc__ = cohen_alt_class.__doc__
+defun(cohen_alt)
+
+
+@defun
+def sumap(ctx, f, interval, integral=None, error=False):
+    r"""
+    Evaluates an infinite series of an analytic summand *f* using the
+    Abel-Plana formula
+
+    .. math ::
+
+        \sum_{k=0}^{\infty} f(k) = \int_0^{\infty} f(t) dt + \frac{1}{2} f(0) +
+            i \int_0^{\infty} \frac{f(it)-f(-it)}{e^{2\pi t}-1} dt.
+
+    Unlike the Euler-Maclaurin formula (see :func:`~mpmath.sumem`),
+    the Abel-Plana formula does not require derivatives. However,
+    it only works when `|f(it)-f(-it)|` does not
+    increase too rapidly with `t`.
+
+    **Examples**
+
+    The Abel-Plana formula is particularly useful when the summand
+    decreases like a power of `k`; for example when the sum is a pure
+    zeta function::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> sumap(lambda k: 1/k**2.5, [1,inf])
+        1.34148725725091717975677
+        >>> zeta(2.5)
+        1.34148725725091717975677
+        >>> sumap(lambda k: 1/(k+1j)**(2.5+2.5j), [1,inf])
+        (-3.385361068546473342286084 - 0.7432082105196321803869551j)
+        >>> zeta(2.5+2.5j, 1+1j)
+        (-3.385361068546473342286084 - 0.7432082105196321803869551j)
+
+    If the series is alternating, numerical quadrature along the real
+    line is likely to give poor results, so it is better to evaluate
+    the first term symbolically whenever possible:
+
+        >>> n=3; z=-0.75
+        >>> I = expint(n,-log(z))
+        >>> chop(sumap(lambda k: z**k / k**n, [1,inf], integral=I))
+        -0.6917036036904594510141448
+        >>> polylog(n,z)
+        -0.6917036036904594510141448
+
+    """
+    prec = ctx.prec
+    try:
+        ctx.prec += 10
+        a, b = interval
+        if  b != ctx.inf:
+            raise ValueError("b should be equal to ctx.inf")
+        g = lambda x: f(x+a)
+        if integral is None:
+            i1, err1 = ctx.quad(g, [0,ctx.inf], error=True)
+        else:
+            i1, err1 = integral, 0
+        j = ctx.j
+        p = ctx.pi * 2
+        if ctx._is_real_type(i1):
+            h = lambda t: -2 * ctx.im(g(j*t)) / ctx.expm1(p*t)
+        else:
+            h = lambda t: j*(g(j*t)-g(-j*t)) / ctx.expm1(p*t)
+        i2, err2 = ctx.quad(h, [0,ctx.inf], error=True)
+        err = err1+err2
+        v = i1+i2+0.5*g(ctx.mpf(0))
+    finally:
+        ctx.prec = prec
+    if error:
+        return +v, err
+    return +v
+
+
+@defun
+def sumem(ctx, f, interval, tol=None, reject=10, integral=None,
+    adiffs=None, bdiffs=None, verbose=False, error=False,
+    _fast_abort=False):
+    r"""
+    Uses the Euler-Maclaurin formula to compute an approximation accurate
+    to within ``tol`` (which defaults to the present epsilon) of the sum
+
+    .. math ::
+
+        S = \sum_{k=a}^b f(k)
+
+    where `(a,b)` are given by ``interval`` and `a` or `b` may be
+    infinite. The approximation is
+
+    .. math ::
+
+        S \sim \int_a^b f(x) \,dx + \frac{f(a)+f(b)}{2} +
+        \sum_{k=1}^{\infty} \frac{B_{2k}}{(2k)!}
+        \left(f^{(2k-1)}(b)-f^{(2k-1)}(a)\right).
+
+    The last sum in the Euler-Maclaurin formula is not generally
+    convergent (a notable exception is if `f` is a polynomial, in
+    which case Euler-Maclaurin actually gives an exact result).
+
+    The summation is stopped as soon as the quotient between two
+    consecutive terms falls below *reject*. That is, by default
+    (*reject* = 10), the summation is continued as long as each
+    term adds at least one decimal.
+
+    Although not convergent, convergence to a given tolerance can
+    often be "forced" if `b = \infty` by summing up to `a+N` and then
+    applying the Euler-Maclaurin formula to the sum over the range
+    `(a+N+1, \ldots, \infty)`. This procedure is implemented by
+    :func:`~mpmath.nsum`.
+
+    By default numerical quadrature and differentiation is used.
+    If the symbolic values of the integral and endpoint derivatives
+    are known, it is more efficient to pass the value of the
+    integral explicitly as ``integral`` and the derivatives
+    explicitly as ``adiffs`` and ``bdiffs``. The derivatives
+    should be given as iterables that yield
+    `f(a), f'(a), f''(a), \ldots` (and the equivalent for `b`).
+
+    **Examples**
+
+    Summation of an infinite series, with automatic and symbolic
+    integral and derivative values (the second should be much faster)::
+
+        >>> from mpmath import *
+        >>> mp.dps = 50; mp.pretty = True
+        >>> sumem(lambda n: 1/n**2, [32, inf])
+        0.03174336652030209012658168043874142714132886413417
+        >>> I = mpf(1)/32
+        >>> D = adiffs=((-1)**n*fac(n+1)*32**(-2-n) for n in range(999))
+        >>> sumem(lambda n: 1/n**2, [32, inf], integral=I, adiffs=D)
+        0.03174336652030209012658168043874142714132886413417
+
+    An exact evaluation of a finite polynomial sum::
+
+        >>> sumem(lambda n: n**5-12*n**2+3*n, [-100000, 200000])
+        10500155000624963999742499550000.0
+        >>> print(sum(n**5-12*n**2+3*n for n in range(-100000, 200001)))
+        10500155000624963999742499550000
+
+    """
+    tol = tol or +ctx.eps
+    interval = ctx._as_points(interval)
+    a = ctx.convert(interval[0])
+    b = ctx.convert(interval[-1])
+    err = ctx.zero
+    prev = 0
+    M = 10000
+    if a == ctx.ninf: adiffs = (0 for n in xrange(M))
+    else:             adiffs = adiffs or ctx.diffs(f, a)
+    if b == ctx.inf:  bdiffs = (0 for n in xrange(M))
+    else:             bdiffs = bdiffs or ctx.diffs(f, b)
+    orig = ctx.prec
+    #verbose = 1
+    try:
+        ctx.prec += 10
+        s = ctx.zero
+        for k, (da, db) in enumerate(izip(adiffs, bdiffs)):
+            if k & 1:
+                term = (db-da) * ctx.bernoulli(k+1) / ctx.factorial(k+1)
+                mag = abs(term)
+                if verbose:
+                    print("term", k, "magnitude =", ctx.nstr(mag))
+                if k > 4 and mag < tol:
+                    s += term
+                    break
+                elif k > 4 and abs(prev) / mag < reject:
+                    err += mag
+                    if _fast_abort:
+                        return [s, (s, err)][error]
+                    if verbose:
+                        print("Failed to converge")
+                    break
+                else:
+                    s += term
+                prev = term
+        # Endpoint correction
+        if a != ctx.ninf: s += f(a)/2
+        if b != ctx.inf: s += f(b)/2
+        # Tail integral
+        if verbose:
+            print("Integrating f(x) from x = %s to %s" % (ctx.nstr(a), ctx.nstr(b)))
+        if integral:
+            s += integral
+        else:
+            integral, ierr = ctx.quad(f, interval, error=True)
+            if verbose:
+                print("Integration error:", ierr)
+            s += integral
+            err += ierr
+    finally:
+        ctx.prec = orig
+    if error:
+        return s, err
+    else:
+        return s
+
+@defun
+def adaptive_extrapolation(ctx, update, emfun, kwargs):
+    option = kwargs.get
+    if ctx._fixed_precision:
+        tol = option('tol', ctx.eps*2**10)
+    else:
+        tol = option('tol', ctx.eps/2**10)
+    verbose = option('verbose', False)
+    maxterms = option('maxterms', ctx.dps*10)
+    method = set(option('method', 'r+s').split('+'))
+    skip = option('skip', 0)
+    steps = iter(option('steps', xrange(10, 10**9, 10)))
+    strict = option('strict')
+    #steps = (10 for i in xrange(1000))
+    summer=[]
+    if 'd' in method or 'direct' in method:
+        TRY_RICHARDSON = TRY_SHANKS = TRY_EULER_MACLAURIN = False
+    else:
+        TRY_RICHARDSON = ('r' in method) or ('richardson' in method)
+        TRY_SHANKS = ('s' in method) or ('shanks' in method)
+        TRY_EULER_MACLAURIN = ('e' in method) or \
+            ('euler-maclaurin' in method)
+
+        def init_levin(m):
+            variant = kwargs.get("levin_variant", "u")
+            if isinstance(variant, str):
+                if variant == "all":
+                    variant = ["u", "v", "t"]
+                else:
+                    variant = [variant]
+            for s in variant:
+                L = levin_class(method = m, variant = s)
+                L.ctx = ctx
+                L.name = m + "(" + s + ")"
+                summer.append(L)
+
+        if ('l' in method) or ('levin' in method):
+            init_levin("levin")
+
+        if ('sidi' in method):
+            init_levin("sidi")
+
+        if ('a' in method) or ('alternating' in method):
+            L = cohen_alt_class()
+            L.ctx = ctx
+            L.name = "alternating"
+            summer.append(L)
+
+    last_richardson_value = 0
+    shanks_table = []
+    index = 0
+    step = 10
+    partial = []
+    best = ctx.zero
+    orig = ctx.prec
+    try:
+        if 'workprec' in kwargs:
+            ctx.prec = kwargs['workprec']
+        elif TRY_RICHARDSON or TRY_SHANKS or len(summer)!=0:
+            ctx.prec = (ctx.prec+10) * 4
+        else:
+            ctx.prec += 30
+        while 1:
+            if index >= maxterms:
+                break
+
+            # Get new batch of terms
+            try:
+                step = next(steps)
+            except StopIteration:
+                pass
+            if verbose:
+                print("-"*70)
+                print("Adding terms #%i-#%i" % (index, index+step))
+            update(partial, xrange(index, index+step))
+            index += step
+
+            # Check direct error
+            best = partial[-1]
+            error = abs(best - partial[-2])
+            if verbose:
+                print("Direct error: %s" % ctx.nstr(error))
+            if error <= tol:
+                return best
+
+            # Check each extrapolation method
+            if TRY_RICHARDSON:
+                value, maxc = ctx.richardson(partial)
+                # Convergence
+                richardson_error = abs(value - last_richardson_value)
+                if verbose:
+                    print("Richardson error: %s" % ctx.nstr(richardson_error))
+                # Convergence
+                if richardson_error <= tol:
+                    return value
+                last_richardson_value = value
+                # Unreliable due to cancellation
+                if ctx.eps*maxc > tol:
+                    if verbose:
+                        print("Ran out of precision for Richardson")
+                    TRY_RICHARDSON = False
+                if richardson_error < error:
+                    error = richardson_error
+                    best = value
+            if TRY_SHANKS:
+                shanks_table = ctx.shanks(partial, shanks_table, randomized=True)
+                row = shanks_table[-1]
+                if len(row) == 2:
+                    est1 = row[-1]
+                    shanks_error = 0
+                else:
+                    est1, maxc, est2 = row[-1], abs(row[-2]), row[-3]
+                    shanks_error = abs(est1-est2)
+                if verbose:
+                    print("Shanks error: %s" % ctx.nstr(shanks_error))
+                if shanks_error <= tol:
+                    return est1
+                if ctx.eps*maxc > tol:
+                    if verbose:
+                        print("Ran out of precision for Shanks")
+                    TRY_SHANKS = False
+                if shanks_error < error:
+                    error = shanks_error
+                    best = est1
+            for L in summer:
+                est, lerror = L.update_psum(partial)
+                if verbose:
+                    print("%s error: %s" % (L.name, ctx.nstr(lerror)))
+                if lerror <= tol:
+                    return est
+                if lerror < error:
+                    error = lerror
+                    best = est
+            if TRY_EULER_MACLAURIN:
+                if ctx.almosteq(ctx.mpc(ctx.sign(partial[-1]) / ctx.sign(partial[-2])), -1):
+                    if verbose:
+                        print ("NOT using Euler-Maclaurin: the series appears"
+                            " to be alternating, so numerical\n quadrature"
+                            " will most likely fail")
+                    TRY_EULER_MACLAURIN = False
+                else:
+                    value, em_error = emfun(index, tol)
+                    value += partial[-1]
+                    if verbose:
+                        print("Euler-Maclaurin error: %s" % ctx.nstr(em_error))
+                    if em_error <= tol:
+                        return value
+                    if em_error < error:
+                        best = value
+    finally:
+        ctx.prec = orig
+    if strict:
+        raise ctx.NoConvergence
+    if verbose:
+        print("Warning: failed to converge to target accuracy")
+    return best
+
+@defun
+def nsum(ctx, f, *intervals, **options):
+    r"""
+    Computes the sum
+
+    .. math :: S = \sum_{k=a}^b f(k)
+
+    where `(a, b)` = *interval*, and where `a = -\infty` and/or
+    `b = \infty` are allowed, or more generally
+
+    .. math :: S = \sum_{k_1=a_1}^{b_1} \cdots
+                   \sum_{k_n=a_n}^{b_n} f(k_1,\ldots,k_n)
+
+    if multiple intervals are given.
+
+    Two examples of infinite series that can be summed by :func:`~mpmath.nsum`,
+    where the first converges rapidly and the second converges slowly,
+    are::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> nsum(lambda n: 1/fac(n), [0, inf])
+        2.71828182845905
+        >>> nsum(lambda n: 1/n**2, [1, inf])
+        1.64493406684823
+
+    When appropriate, :func:`~mpmath.nsum` applies convergence acceleration to
+    accurately estimate the sums of slowly convergent series. If the series is
+    finite, :func:`~mpmath.nsum` currently does not attempt to perform any
+    extrapolation, and simply calls :func:`~mpmath.fsum`.
+
+    Multidimensional infinite series are reduced to a single-dimensional
+    series over expanding hypercubes; if both infinite and finite dimensions
+    are present, the finite ranges are moved innermost. For more advanced
+    control over the summation order, use nested calls to :func:`~mpmath.nsum`,
+    or manually rewrite the sum as a single-dimensional series.
+
+    **Options**
+
+    *tol*
+        Desired maximum final error. Defaults roughly to the
+        epsilon of the working precision.
+
+    *method*
+        Which summation algorithm to use (described below).
+        Default: ``'richardson+shanks'``.
+
+    *maxterms*
+        Cancel after at most this many terms. Default: 10*dps.
+
+    *steps*
+        An iterable giving the number of terms to add between
+        each extrapolation attempt. The default sequence is
+        [10, 20, 30, 40, ...]. For example, if you know that
+        approximately 100 terms will be required, efficiency might be
+        improved by setting this to [100, 10]. Then the first
+        extrapolation will be performed after 100 terms, the second
+        after 110, etc.
+
+    *verbose*
+        Print details about progress.
+
+    *ignore*
+        If enabled, any term that raises ``ArithmeticError``
+        or ``ValueError`` (e.g. through division by zero) is replaced
+        by a zero. This is convenient for lattice sums with
+        a singular term near the origin.
+
+    **Methods**
+
+    Unfortunately, an algorithm that can efficiently sum any infinite
+    series does not exist. :func:`~mpmath.nsum` implements several different
+    algorithms that each work well in different cases. The *method*
+    keyword argument selects a method.
+
+    The default method is ``'r+s'``, i.e. both Richardson extrapolation
+    and Shanks transformation is attempted. A slower method that
+    handles more cases is ``'r+s+e'``. For very high precision
+    summation, or if the summation needs to be fast (for example if
+    multiple sums need to be evaluated), it is a good idea to
+    investigate which one method works best and only use that.
+
+    ``'richardson'`` / ``'r'``:
+        Uses Richardson extrapolation. Provides useful extrapolation
+        when `f(k) \sim P(k)/Q(k)` or when `f(k) \sim (-1)^k P(k)/Q(k)`
+        for polynomials `P` and `Q`. See :func:`~mpmath.richardson` for
+        additional information.
+
+    ``'shanks'`` / ``'s'``:
+        Uses Shanks transformation. Typically provides useful
+        extrapolation when `f(k) \sim c^k` or when successive terms
+        alternate signs. Is able to sum some divergent series.
+        See :func:`~mpmath.shanks` for additional information.
+
+    ``'levin'`` / ``'l'``:
+        Uses the Levin transformation. It performs better than the Shanks
+        transformation for logarithmic convergent or alternating divergent
+        series. The ``'levin_variant'``-keyword selects the variant. Valid
+        choices are "u", "t", "v" and "all" whereby "all" uses all three
+        u,t and v simultanously (This is good for performance comparison in
+        conjunction with "verbose=True"). Instead of the Levin transform one can
+        also use the Sidi-S transform by selecting the method ``'sidi'``.
+        See :func:`~mpmath.levin` for additional details.
+
+    ``'alternating'`` / ``'a'``:
+        This is the convergence acceleration of alternating series developped
+        by Cohen, Villegras and Zagier.
+        See :func:`~mpmath.cohen_alt` for additional details.
+
+    ``'euler-maclaurin'`` / ``'e'``:
+        Uses the Euler-Maclaurin summation formula to approximate
+        the remainder sum by an integral. This requires high-order
+        numerical derivatives and numerical integration. The advantage
+        of this algorithm is that it works regardless of the
+        decay rate of `f`, as long as `f` is sufficiently smooth.
+        See :func:`~mpmath.sumem` for additional information.
+
+    ``'direct'`` / ``'d'``:
+        Does not perform any extrapolation. This can be used
+        (and should only be used for) rapidly convergent series.
+        The summation automatically stops when the terms
+        decrease below the target tolerance.
+
+    **Basic examples**
+
+    A finite sum::
+
+        >>> nsum(lambda k: 1/k, [1, 6])
+        2.45
+
+    Summation of a series going to negative infinity and a doubly
+    infinite series::
+
+        >>> nsum(lambda k: 1/k**2, [-inf, -1])
+        1.64493406684823
+        >>> nsum(lambda k: 1/(1+k**2), [-inf, inf])
+        3.15334809493716
+
+    :func:`~mpmath.nsum` handles sums of complex numbers::
+
+        >>> nsum(lambda k: (0.5+0.25j)**k, [0, inf])
+        (1.6 + 0.8j)
+
+    The following sum converges very rapidly, so it is most
+    efficient to sum it by disabling convergence acceleration::
+
+        >>> mp.dps = 1000
+        >>> a = nsum(lambda k: -(-1)**k * k**2 / fac(2*k), [1, inf],
+        ...     method='direct')
+        >>> b = (cos(1)+sin(1))/4
+        >>> abs(a-b) < mpf('1e-998')
+        True
+
+    **Examples with Richardson extrapolation**
+
+    Richardson extrapolation works well for sums over rational
+    functions, as well as their alternating counterparts::
+
+        >>> mp.dps = 50
+        >>> nsum(lambda k: 1 / k**3, [1, inf],
+        ...     method='richardson')
+        1.2020569031595942853997381615114499907649862923405
+        >>> zeta(3)
+        1.2020569031595942853997381615114499907649862923405
+
+        >>> nsum(lambda n: (n + 3)/(n**3 + n**2), [1, inf],
+        ...     method='richardson')
+        2.9348022005446793094172454999380755676568497036204
+        >>> pi**2/2-2
+        2.9348022005446793094172454999380755676568497036204
+
+        >>> nsum(lambda k: (-1)**k / k**3, [1, inf],
+        ...     method='richardson')
+        -0.90154267736969571404980362113358749307373971925537
+        >>> -3*zeta(3)/4
+        -0.90154267736969571404980362113358749307373971925538
+
+    **Examples with Shanks transformation**
+
+    The Shanks transformation works well for geometric series
+    and typically provides excellent acceleration for Taylor
+    series near the border of their disk of convergence.
+    Here we apply it to a series for `\log(2)`, which can be
+    seen as the Taylor series for `\log(1+x)` with `x = 1`::
+
+        >>> nsum(lambda k: -(-1)**k/k, [1, inf],
+        ...     method='shanks')
+        0.69314718055994530941723212145817656807550013436025
+        >>> log(2)
+        0.69314718055994530941723212145817656807550013436025
+
+    Here we apply it to a slowly convergent geometric series::
+
+        >>> nsum(lambda k: mpf('0.995')**k, [0, inf],
+        ...     method='shanks')
+        200.0
+
+    Finally, Shanks' method works very well for alternating series
+    where `f(k) = (-1)^k g(k)`, and often does so regardless of
+    the exact decay rate of `g(k)`::
+
+        >>> mp.dps = 15
+        >>> nsum(lambda k: (-1)**(k+1) / k**1.5, [1, inf],
+        ...     method='shanks')
+        0.765147024625408
+        >>> (2-sqrt(2))*zeta(1.5)/2
+        0.765147024625408
+
+    The following slowly convergent alternating series has no known
+    closed-form value. Evaluating the sum a second time at higher
+    precision indicates that the value is probably correct::
+
+        >>> nsum(lambda k: (-1)**k / log(k), [2, inf],
+        ...     method='shanks')
+        0.924299897222939
+        >>> mp.dps = 30
+        >>> nsum(lambda k: (-1)**k / log(k), [2, inf],
+        ...     method='shanks')
+        0.92429989722293885595957018136
+
+    **Examples with Levin transformation**
+
+    The following example calculates Euler's constant as the constant term in
+    the Laurent expansion of zeta(s) at s=1. This sum converges extremly slow
+    because of the logarithmic convergence behaviour of the Dirichlet series
+    for zeta.
+
+      >>> mp.dps = 30
+      >>> z = mp.mpf(10) ** (-10)
+      >>> a = mp.nsum(lambda n: n**(-(1+z)), [1, mp.inf], method = "levin") - 1 / z
+      >>> print(mp.chop(a - mp.euler, tol = 1e-10))
+      0.0
+
+    Now we sum the zeta function outside its range of convergence
+    (attention: This does not work at the negative integers!):
+
+      >>> mp.dps = 15
+      >>> w = mp.nsum(lambda n: n ** (2 + 3j), [1, mp.inf], method = "levin", levin_variant = "v")
+      >>> print(mp.chop(w - mp.zeta(-2-3j)))
+      0.0
+
+    The next example resummates an asymptotic series expansion of an integral
+    related to the exponential integral.
+
+      >>> mp.dps = 15
+      >>> z = mp.mpf(10)
+      >>> # exact = mp.quad(lambda x: mp.exp(-x)/(1+x/z),[0,mp.inf])
+      >>> exact = z * mp.exp(z) * mp.expint(1,z) # this is the symbolic expression for the integral
+      >>> w = mp.nsum(lambda n: (-1) ** n * mp.fac(n) * z ** (-n), [0, mp.inf], method = "sidi", levin_variant = "t")
+      >>> print(mp.chop(w - exact))
+      0.0
+
+    Following highly divergent asymptotic expansion needs some care. Firstly we
+    need copious amount of working precision. Secondly the stepsize must not be
+    chosen to large, otherwise nsum may miss the point where the Levin transform
+    converges and reach the point where only numerical garbage is produced due to
+    numerical cancellation.
+
+      >>> mp.dps = 15
+      >>> z = mp.mpf(2)
+      >>> # exact = mp.quad(lambda x: mp.exp( -x * x / 2 - z * x ** 4), [0,mp.inf]) * 2 / mp.sqrt(2 * mp.pi)
+      >>> exact = mp.exp(mp.one / (32 * z)) * mp.besselk(mp.one / 4, mp.one / (32 * z)) / (4 * mp.sqrt(z * mp.pi)) # this is the symbolic expression for the integral
+      >>> w = mp.nsum(lambda n: (-z)**n * mp.fac(4 * n) / (mp.fac(n) * mp.fac(2 * n) * (4 ** n)),
+      ...   [0, mp.inf], method = "levin", levin_variant = "t", workprec = 8*mp.prec, steps = [2] + [1 for x in xrange(1000)])
+      >>> print(mp.chop(w - exact))
+      0.0
+
+    The hypergeoemtric function can also be summed outside its range of convergence:
+
+      >>> mp.dps = 15
+      >>> z = 2 + 1j
+      >>> exact = mp.hyp2f1(2 / mp.mpf(3), 4 / mp.mpf(3), 1 / mp.mpf(3), z)
+      >>> f = lambda n: mp.rf(2 / mp.mpf(3), n) * mp.rf(4 / mp.mpf(3), n) * z**n / (mp.rf(1 / mp.mpf(3), n) * mp.fac(n))
+      >>> v = mp.nsum(f, [0, mp.inf], method = "levin", steps = [10 for x in xrange(1000)])
+      >>> print(mp.chop(exact-v))
+      0.0
+
+    **Examples with Cohen's alternating series resummation**
+
+      The next example sums the alternating zeta function:
+
+      >>> v = mp.nsum(lambda n: (-1)**(n-1) / n, [1, mp.inf], method = "a")
+      >>> print(mp.chop(v - mp.log(2)))
+      0.0
+
+      The derivate of the alternating zeta function outside its range of
+      convergence:
+
+      >>> v = mp.nsum(lambda n: (-1)**n * mp.log(n) * n, [1, mp.inf], method = "a")
+      >>> print(mp.chop(v - mp.diff(lambda s: mp.altzeta(s), -1)))
+      0.0
+
+    **Examples with Euler-Maclaurin summation**
+
+    The sum in the following example has the wrong rate of convergence
+    for either Richardson or Shanks to be effective.
+
+        >>> f = lambda k: log(k)/k**2.5
+        >>> mp.dps = 15
+        >>> nsum(f, [1, inf], method='euler-maclaurin')
+        0.38734195032621
+        >>> -diff(zeta, 2.5)
+        0.38734195032621
+
+    Increasing ``steps`` improves speed at higher precision::
+
+        >>> mp.dps = 50
+        >>> nsum(f, [1, inf], method='euler-maclaurin', steps=[250])
+        0.38734195032620997271199237593105101319948228874688
+        >>> -diff(zeta, 2.5)
+        0.38734195032620997271199237593105101319948228874688
+
+    **Divergent series**
+
+    The Shanks transformation is able to sum some *divergent*
+    series. In particular, it is often able to sum Taylor series
+    beyond their radius of convergence (this is due to a relation
+    between the Shanks transformation and Pade approximations;
+    see :func:`~mpmath.pade` for an alternative way to evaluate divergent
+    Taylor series). Furthermore the Levin-transform examples above
+    contain some divergent series resummation.
+
+    Here we apply it to `\log(1+x)` far outside the region of
+    convergence::
+
+        >>> mp.dps = 50
+        >>> nsum(lambda k: -(-9)**k/k, [1, inf],
+        ...     method='shanks')
+        2.3025850929940456840179914546843642076011014886288
+        >>> log(10)
+        2.3025850929940456840179914546843642076011014886288
+
+    A particular type of divergent series that can be summed
+    using the Shanks transformation is geometric series.
+    The result is the same as using the closed-form formula
+    for an infinite geometric series::
+
+        >>> mp.dps = 15
+        >>> for n in range(-8, 8):
+        ...     if n == 1:
+        ...         continue
+        ...     print("%s %s %s" % (mpf(n), mpf(1)/(1-n),
+        ...         nsum(lambda k: n**k, [0, inf], method='shanks')))
+        ...
+        -8.0 0.111111111111111 0.111111111111111
+        -7.0 0.125 0.125
+        -6.0 0.142857142857143 0.142857142857143
+        -5.0 0.166666666666667 0.166666666666667
+        -4.0 0.2 0.2
+        -3.0 0.25 0.25
+        -2.0 0.333333333333333 0.333333333333333
+        -1.0 0.5 0.5
+        0.0 1.0 1.0
+        2.0 -1.0 -1.0
+        3.0 -0.5 -0.5
+        4.0 -0.333333333333333 -0.333333333333333
+        5.0 -0.25 -0.25
+        6.0 -0.2 -0.2
+        7.0 -0.166666666666667 -0.166666666666667
+
+    **Multidimensional sums**
+
+    Any combination of finite and infinite ranges is allowed for the
+    summation indices::
+
+        >>> mp.dps = 15
+        >>> nsum(lambda x,y: x+y, [2,3], [4,5])
+        28.0
+        >>> nsum(lambda x,y: x/2**y, [1,3], [1,inf])
+        6.0
+        >>> nsum(lambda x,y: y/2**x, [1,inf], [1,3])
+        6.0
+        >>> nsum(lambda x,y,z: z/(2**x*2**y), [1,inf], [1,inf], [3,4])
+        7.0
+        >>> nsum(lambda x,y,z: y/(2**x*2**z), [1,inf], [3,4], [1,inf])
+        7.0
+        >>> nsum(lambda x,y,z: x/(2**z*2**y), [3,4], [1,inf], [1,inf])
+        7.0
+
+    Some nice examples of double series with analytic solutions or
+    reductions to single-dimensional series (see [1])::
+
+        >>> nsum(lambda m, n: 1/2**(m*n), [1,inf], [1,inf])
+        1.60669515241529
+        >>> nsum(lambda n: 1/(2**n-1), [1,inf])
+        1.60669515241529
+
+        >>> nsum(lambda i,j: (-1)**(i+j)/(i**2+j**2), [1,inf], [1,inf])
+        0.278070510848213
+        >>> pi*(pi-3*ln2)/12
+        0.278070510848213
+
+        >>> nsum(lambda i,j: (-1)**(i+j)/(i+j)**2, [1,inf], [1,inf])
+        0.129319852864168
+        >>> altzeta(2) - altzeta(1)
+        0.129319852864168
+
+        >>> nsum(lambda i,j: (-1)**(i+j)/(i+j)**3, [1,inf], [1,inf])
+        0.0790756439455825
+        >>> altzeta(3) - altzeta(2)
+        0.0790756439455825
+
+        >>> nsum(lambda m,n: m**2*n/(3**m*(n*3**m+m*3**n)),
+        ...     [1,inf], [1,inf])
+        0.28125
+        >>> mpf(9)/32
+        0.28125
+
+        >>> nsum(lambda i,j: fac(i-1)*fac(j-1)/fac(i+j),
+        ...     [1,inf], [1,inf], workprec=400)
+        1.64493406684823
+        >>> zeta(2)
+        1.64493406684823
+
+    A hard example of a multidimensional sum is the Madelung constant
+    in three dimensions (see [2]). The defining sum converges very
+    slowly and only conditionally, so :func:`~mpmath.nsum` is lucky to
+    obtain an accurate value through convergence acceleration. The
+    second evaluation below uses a much more efficient, rapidly
+    convergent 2D sum::
+
+        >>> nsum(lambda x,y,z: (-1)**(x+y+z)/(x*x+y*y+z*z)**0.5,
+        ...     [-inf,inf], [-inf,inf], [-inf,inf], ignore=True)
+        -1.74756459463318
+        >>> nsum(lambda x,y: -12*pi*sech(0.5*pi * \
+        ...     sqrt((2*x+1)**2+(2*y+1)**2))**2, [0,inf], [0,inf])
+        -1.74756459463318
+
+    Another example of a lattice sum in 2D::
+
+        >>> nsum(lambda x,y: (-1)**(x+y) / (x**2+y**2), [-inf,inf],
+        ...     [-inf,inf], ignore=True)
+        -2.1775860903036
+        >>> -pi*ln2
+        -2.1775860903036
+
+    An example of an Eisenstein series::
+
+        >>> nsum(lambda m,n: (m+n*1j)**(-4), [-inf,inf], [-inf,inf],
+        ...     ignore=True)
+        (3.1512120021539 + 0.0j)
+
+    **References**
+
+    1. [Weisstein]_ http://mathworld.wolfram.com/DoubleSeries.html,
+    2. [Weisstein]_ http://mathworld.wolfram.com/MadelungConstants.html
+
+    """
+    infinite, g = standardize(ctx, f, intervals, options)
+    if not infinite:
+        return +g()
+
+    def update(partial_sums, indices):
+        if partial_sums:
+            psum = partial_sums[-1]
+        else:
+            psum = ctx.zero
+        for k in indices:
+            psum = psum + g(ctx.mpf(k))
+            partial_sums.append(psum)
+
+    prec = ctx.prec
+
+    def emfun(point, tol):
+        workprec = ctx.prec
+        ctx.prec = prec + 10
+        v = ctx.sumem(g, [point, ctx.inf], tol, error=1)
+        ctx.prec = workprec
+        return v
+
+    return +ctx.adaptive_extrapolation(update, emfun, options)
+
+
+def wrapsafe(f):
+    def g(*args):
+        try:
+            return f(*args)
+        except (ArithmeticError, ValueError):
+            return 0
+    return g
+
+def standardize(ctx, f, intervals, options):
+    if options.get("ignore"):
+        f = wrapsafe(f)
+    finite = []
+    infinite = []
+    for k, points in enumerate(intervals):
+        a, b = ctx._as_points(points)
+        if b < a:
+            return False, (lambda: ctx.zero)
+        if a == ctx.ninf or b == ctx.inf:
+            infinite.append((k, (a,b)))
+        else:
+            finite.append((k, (int(a), int(b))))
+    if finite:
+        f = fold_finite(ctx, f, finite)
+        if not infinite:
+            return False, lambda: f(*([0]*len(intervals)))
+    if infinite:
+        f = standardize_infinite(ctx, f, infinite)
+        f = fold_infinite(ctx, f, infinite)
+        args = [0] * len(intervals)
+        d = infinite[0][0]
+        def g(k):
+            args[d] = k
+            return f(*args)
+        return True, g
+
+# backwards compatible itertools.product
+def cartesian_product(args):
+    pools = map(tuple, args)
+    result = [[]]
+    for pool in pools:
+        result = [x+[y] for x in result for y in pool]
+    for prod in result:
+        yield tuple(prod)
+
+def fold_finite(ctx, f, intervals):
+    if not intervals:
+        return f
+    indices = [v[0] for v in intervals]
+    points = [v[1] for v in intervals]
+    ranges = [xrange(a, b+1) for (a,b) in points]
+    def g(*args):
+        args = list(args)
+        s = ctx.zero
+        for xs in cartesian_product(ranges):
+            for dim, x in zip(indices, xs):
+                args[dim] = ctx.mpf(x)
+            s += f(*args)
+        return s
+    #print "Folded finite", indices
+    return g
+
+# Standardize each interval to [0,inf]
+def standardize_infinite(ctx, f, intervals):
+    if not intervals:
+        return f
+    dim, [a,b] = intervals[-1]
+    if a == ctx.ninf:
+        if b == ctx.inf:
+            def g(*args):
+                args = list(args)
+                k = args[dim]
+                if k:
+                    s = f(*args)
+                    args[dim] = -k
+                    s += f(*args)
+                    return s
+                else:
+                    return f(*args)
+        else:
+            def g(*args):
+                args = list(args)
+                args[dim] = b - args[dim]
+                return f(*args)
+    else:
+        def g(*args):
+            args = list(args)
+            args[dim] += a
+            return f(*args)
+    #print "Standardized infinity along dimension", dim, a, b
+    return standardize_infinite(ctx, g, intervals[:-1])
+
+def fold_infinite(ctx, f, intervals):
+    if len(intervals) < 2:
+        return f
+    dim1 = intervals[-2][0]
+    dim2 = intervals[-1][0]
+    # Assume intervals are [0,inf] x [0,inf] x ...
+    def g(*args):
+        args = list(args)
+        #args.insert(dim2, None)
+        n = int(args[dim1])
+        s = ctx.zero
+        #y = ctx.mpf(n)
+        args[dim2] = ctx.mpf(n) #y
+        for x in xrange(n+1):
+            args[dim1] = ctx.mpf(x)
+            s += f(*args)
+        args[dim1] = ctx.mpf(n) #ctx.mpf(n)
+        for y in xrange(n):
+            args[dim2] = ctx.mpf(y)
+            s += f(*args)
+        return s
+    #print "Folded infinite from", len(intervals), "to", (len(intervals)-1)
+    return fold_infinite(ctx, g, intervals[:-1])
+
+@defun
+def nprod(ctx, f, interval, nsum=False, **kwargs):
+    r"""
+    Computes the product
+
+    .. math ::
+
+        P = \prod_{k=a}^b f(k)
+
+    where `(a, b)` = *interval*, and where `a = -\infty` and/or
+    `b = \infty` are allowed.
+
+    By default, :func:`~mpmath.nprod` uses the same extrapolation methods as
+    :func:`~mpmath.nsum`, except applied to the partial products rather than
+    partial sums, and the same keyword options as for :func:`~mpmath.nsum` are
+    supported. If ``nsum=True``, the product is instead computed via
+    :func:`~mpmath.nsum` as
+
+    .. math ::
+
+        P = \exp\left( \sum_{k=a}^b \log(f(k)) \right).
+
+    This is slower, but can sometimes yield better results. It is
+    also required (and used automatically) when Euler-Maclaurin
+    summation is requested.
+
+    **Examples**
+
+    A simple finite product::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> nprod(lambda k: k, [1, 4])
+        24.0
+
+    A large number of infinite products have known exact values,
+    and can therefore be used as a reference. Most of the following
+    examples are taken from MathWorld [1].
+
+    A few infinite products with simple values are::
+
+        >>> 2*nprod(lambda k: (4*k**2)/(4*k**2-1), [1, inf])
+        3.141592653589793238462643
+        >>> nprod(lambda k: (1+1/k)**2/(1+2/k), [1, inf])
+        2.0
+        >>> nprod(lambda k: (k**3-1)/(k**3+1), [2, inf])
+        0.6666666666666666666666667
+        >>> nprod(lambda k: (1-1/k**2), [2, inf])
+        0.5
+
+    Next, several more infinite products with more complicated
+    values::
+
+        >>> nprod(lambda k: exp(1/k**2), [1, inf]); exp(pi**2/6)
+        5.180668317897115748416626
+        5.180668317897115748416626
+
+        >>> nprod(lambda k: (k**2-1)/(k**2+1), [2, inf]); pi*csch(pi)
+        0.2720290549821331629502366
+        0.2720290549821331629502366
+
+        >>> nprod(lambda k: (k**4-1)/(k**4+1), [2, inf])
+        0.8480540493529003921296502
+        >>> pi*sinh(pi)/(cosh(sqrt(2)*pi)-cos(sqrt(2)*pi))
+        0.8480540493529003921296502
+
+        >>> nprod(lambda k: (1+1/k+1/k**2)**2/(1+2/k+3/k**2), [1, inf])
+        1.848936182858244485224927
+        >>> 3*sqrt(2)*cosh(pi*sqrt(3)/2)**2*csch(pi*sqrt(2))/pi
+        1.848936182858244485224927
+
+        >>> nprod(lambda k: (1-1/k**4), [2, inf]); sinh(pi)/(4*pi)
+        0.9190194775937444301739244
+        0.9190194775937444301739244
+
+        >>> nprod(lambda k: (1-1/k**6), [2, inf])
+        0.9826842777421925183244759
+        >>> (1+cosh(pi*sqrt(3)))/(12*pi**2)
+        0.9826842777421925183244759
+
+        >>> nprod(lambda k: (1+1/k**2), [2, inf]); sinh(pi)/(2*pi)
+        1.838038955187488860347849
+        1.838038955187488860347849
+
+        >>> nprod(lambda n: (1+1/n)**n * exp(1/(2*n)-1), [1, inf])
+        1.447255926890365298959138
+        >>> exp(1+euler/2)/sqrt(2*pi)
+        1.447255926890365298959138
+
+    The following two products are equivalent and can be evaluated in
+    terms of a Jacobi theta function. Pi can be replaced by any value
+    (as long as convergence is preserved)::
+
+        >>> nprod(lambda k: (1-pi**-k)/(1+pi**-k), [1, inf])
+        0.3838451207481672404778686
+        >>> nprod(lambda k: tanh(k*log(pi)/2), [1, inf])
+        0.3838451207481672404778686
+        >>> jtheta(4,0,1/pi)
+        0.3838451207481672404778686
+
+    This product does not have a known closed form value::
+
+        >>> nprod(lambda k: (1-1/2**k), [1, inf])
+        0.2887880950866024212788997
+
+    A product taken from `-\infty`::
+
+        >>> nprod(lambda k: 1-k**(-3), [-inf,-2])
+        0.8093965973662901095786805
+        >>> cosh(pi*sqrt(3)/2)/(3*pi)
+        0.8093965973662901095786805
+
+    A doubly infinite product::
+
+        >>> nprod(lambda k: exp(1/(1+k**2)), [-inf, inf])
+        23.41432688231864337420035
+        >>> exp(pi/tanh(pi))
+        23.41432688231864337420035
+
+    A product requiring the use of Euler-Maclaurin summation to compute
+    an accurate value::
+
+        >>> nprod(lambda k: (1-1/k**2.5), [2, inf], method='e')
+        0.696155111336231052898125
+
+    **References**
+
+    1. [Weisstein]_ http://mathworld.wolfram.com/InfiniteProduct.html
+
+    """
+    if nsum or ('e' in kwargs.get('method', '')):
+        orig = ctx.prec
+        try:
+            # TODO: we are evaluating log(1+eps) -> eps, which is
+            # inaccurate. This currently works because nsum greatly
+            # increases the working precision. But we should be
+            # more intelligent and handle the precision here.
+            ctx.prec += 10
+            v = ctx.nsum(lambda n: ctx.ln(f(n)), interval, **kwargs)
+        finally:
+            ctx.prec = orig
+        return +ctx.exp(v)
+
+    a, b = ctx._as_points(interval)
+    if a == ctx.ninf:
+        if b == ctx.inf:
+            return f(0) * ctx.nprod(lambda k: f(-k) * f(k), [1, ctx.inf], **kwargs)
+        return ctx.nprod(f, [-b, ctx.inf], **kwargs)
+    elif b != ctx.inf:
+        return ctx.fprod(f(ctx.mpf(k)) for k in xrange(int(a), int(b)+1))
+
+    a = int(a)
+
+    def update(partial_products, indices):
+        if partial_products:
+            pprod = partial_products[-1]
+        else:
+            pprod = ctx.one
+        for k in indices:
+            pprod = pprod * f(a + ctx.mpf(k))
+            partial_products.append(pprod)
+
+    return +ctx.adaptive_extrapolation(update, None, kwargs)
+
+
+@defun
+def limit(ctx, f, x, direction=1, exp=False, **kwargs):
+    r"""
+    Computes an estimate of the limit
+
+    .. math ::
+
+        \lim_{t \to x} f(t)
+
+    where `x` may be finite or infinite.
+
+    For finite `x`, :func:`~mpmath.limit` evaluates `f(x + d/n)` for
+    consecutive integer values of `n`, where the approach direction
+    `d` may be specified using the *direction* keyword argument.
+    For infinite `x`, :func:`~mpmath.limit` evaluates values of
+    `f(\mathrm{sign}(x) \cdot n)`.
+
+    If the approach to the limit is not sufficiently fast to give
+    an accurate estimate directly, :func:`~mpmath.limit` attempts to find
+    the limit using Richardson extrapolation or the Shanks
+    transformation. You can select between these methods using
+    the *method* keyword (see documentation of :func:`~mpmath.nsum` for
+    more information).
+
+    **Options**
+
+    The following options are available with essentially the
+    same meaning as for :func:`~mpmath.nsum`: *tol*, *method*, *maxterms*,
+    *steps*, *verbose*.
+
+    If the option *exp=True* is set, `f` will be
+    sampled at exponentially spaced points `n = 2^1, 2^2, 2^3, \ldots`
+    instead of the linearly spaced points `n = 1, 2, 3, \ldots`.
+    This can sometimes improve the rate of convergence so that
+    :func:`~mpmath.limit` may return a more accurate answer (and faster).
+    However, do note that this can only be used if `f`
+    supports fast and accurate evaluation for arguments that
+    are extremely close to the limit point (or if infinite,
+    very large arguments).
+
+    **Examples**
+
+    A basic evaluation of a removable singularity::
+
+        >>> from mpmath import *
+        >>> mp.dps = 30; mp.pretty = True
+        >>> limit(lambda x: (x-sin(x))/x**3, 0)
+        0.166666666666666666666666666667
+
+    Computing the exponential function using its limit definition::
+
+        >>> limit(lambda n: (1+3/n)**n, inf)
+        20.0855369231876677409285296546
+        >>> exp(3)
+        20.0855369231876677409285296546
+
+    A limit for `\pi`::
+
+        >>> f = lambda n: 2**(4*n+1)*fac(n)**4/(2*n+1)/fac(2*n)**2
+        >>> limit(f, inf)
+        3.14159265358979323846264338328
+
+    Calculating the coefficient in Stirling's formula::
+
+        >>> limit(lambda n: fac(n) / (sqrt(n)*(n/e)**n), inf)
+        2.50662827463100050241576528481
+        >>> sqrt(2*pi)
+        2.50662827463100050241576528481
+
+    Evaluating Euler's constant `\gamma` using the limit representation
+
+    .. math ::
+
+        \gamma = \lim_{n \rightarrow \infty } \left[ \left(
+        \sum_{k=1}^n \frac{1}{k} \right) - \log(n) \right]
+
+    (which converges notoriously slowly)::
+
+        >>> f = lambda n: sum([mpf(1)/k for k in range(1,int(n)+1)]) - log(n)
+        >>> limit(f, inf)
+        0.577215664901532860606512090082
+        >>> +euler
+        0.577215664901532860606512090082
+
+    With default settings, the following limit converges too slowly
+    to be evaluated accurately. Changing to exponential sampling
+    however gives a perfect result::
+
+        >>> f = lambda x: sqrt(x**3+x**2)/(sqrt(x**3)+x)
+        >>> limit(f, inf)
+        0.992831158558330281129249686491
+        >>> limit(f, inf, exp=True)
+        1.0
+
+    """
+
+    if ctx.isinf(x):
+        direction = ctx.sign(x)
+        g = lambda k: f(ctx.mpf(k+1)*direction)
+    else:
+        direction *= ctx.one
+        g = lambda k: f(x + direction/(k+1))
+    if exp:
+        h = g
+        g = lambda k: h(2**k)
+
+    def update(values, indices):
+        for k in indices:
+            values.append(g(k+1))
+
+    # XXX: steps used by nsum don't work well
+    if not 'steps' in kwargs:
+        kwargs['steps'] = [10]
+
+    return +ctx.adaptive_extrapolation(update, None, kwargs)

lib/python3.11/site-packages/mpmath/calculus/inverselaplace.py

@@ -0,0 +1,973 @@
+# contributed to mpmath by Kristopher L. Kuhlman, February 2017
+# contributed to mpmath by Guillermo Navas-Palencia, February 2022
+
+class InverseLaplaceTransform(object):
+    r"""
+    Inverse Laplace transform methods are implemented using this
+    class, in order to simplify the code and provide a common
+    infrastructure.
+
+    Implement a custom inverse Laplace transform algorithm by
+    subclassing :class:`InverseLaplaceTransform` and implementing the
+    appropriate methods. The subclass can then be used by
+    :func:`~mpmath.invertlaplace` by passing it as the *method*
+    argument.
+    """
+
+    def __init__(self, ctx):
+        self.ctx = ctx
+
+    def calc_laplace_parameter(self, t, **kwargs):
+        r"""
+        Determine the vector of Laplace parameter values needed for an
+        algorithm, this will depend on the choice of algorithm (de
+        Hoog is default), the algorithm-specific parameters passed (or
+        default ones), and desired time.
+        """
+        raise NotImplementedError
+
+    def calc_time_domain_solution(self, fp):
+        r"""
+        Compute the time domain solution, after computing the
+        Laplace-space function evaluations at the abscissa required
+        for the algorithm. Abscissa computed for one algorithm are
+        typically not useful for another algorithm.
+        """
+        raise NotImplementedError
+
+
+class FixedTalbot(InverseLaplaceTransform):
+
+    def calc_laplace_parameter(self, t, **kwargs):
+        r"""The "fixed" Talbot method deforms the Bromwich contour towards
+        `-\infty` in the shape of a parabola. Traditionally the Talbot
+        algorithm has adjustable parameters, but the "fixed" version
+        does not. The `r` parameter could be passed in as a parameter,
+        if you want to override the default given by (Abate & Valko,
+        2004).
+
+        The Laplace parameter is sampled along a parabola opening
+        along the negative imaginary axis, with the base of the
+        parabola along the real axis at
+        `p=\frac{r}{t_\mathrm{max}}`. As the number of terms used in
+        the approximation (degree) grows, the abscissa required for
+        function evaluation tend towards `-\infty`, requiring high
+        precision to prevent overflow.  If any poles, branch cuts or
+        other singularities exist such that the deformed Bromwich
+        contour lies to the left of the singularity, the method will
+        fail.
+
+        **Optional arguments**
+
+        :class:`~mpmath.calculus.inverselaplace.FixedTalbot.calc_laplace_parameter`
+        recognizes the following keywords
+
+        *tmax*
+            maximum time associated with vector of times
+            (typically just the time requested)
+        *degree*
+            integer order of approximation (M = number of terms)
+        *r*
+            abscissa for `p_0` (otherwise computed using rule
+            of thumb `2M/5`)
+
+        The working precision will be increased according to a rule of
+        thumb. If 'degree' is not specified, the working precision and
+        degree are chosen to hopefully achieve the dps of the calling
+        context. If 'degree' is specified, the working precision is
+        chosen to achieve maximum resulting precision for the
+        specified degree.
+
+        .. math ::
+
+            p_0=\frac{r}{t}
+
+        .. math ::
+
+            p_i=\frac{i r \pi}{Mt_\mathrm{max}}\left[\cot\left(
+            \frac{i\pi}{M}\right) + j \right] \qquad 1\le i <M
+
+        where `j=\sqrt{-1}`, `r=2M/5`, and `t_\mathrm{max}` is the
+        maximum specified time.
+
+        """
+
+        # required
+        # ------------------------------
+        # time of desired approximation
+        self.t = self.ctx.convert(t)
+
+        # optional
+        # ------------------------------
+        # maximum time desired (used for scaling) default is requested
+        # time.
+        self.tmax = self.ctx.convert(kwargs.get('tmax', self.t))
+
+        # empirical relationships used here based on a linear fit of
+        # requested and delivered dps for exponentially decaying time
+        # functions for requested dps up to 512.
+
+        if 'degree' in kwargs:
+            self.degree = kwargs['degree']
+            self.dps_goal = self.degree
+        else:
+            self.dps_goal = int(1.72*self.ctx.dps)
+            self.degree = max(12, int(1.38*self.dps_goal))
+
+        M = self.degree
+
+        # this is adjusting the dps of the calling context hopefully
+        # the caller doesn't monkey around with it between calling
+        # this routine and calc_time_domain_solution()
+        self.dps_orig = self.ctx.dps
+        self.ctx.dps = self.dps_goal
+
+        # Abate & Valko rule of thumb for r parameter
+        self.r = kwargs.get('r', self.ctx.fraction(2, 5)*M)
+
+        self.theta = self.ctx.linspace(0.0, self.ctx.pi, M+1)
+
+        self.cot_theta = self.ctx.matrix(M, 1)
+        self.cot_theta[0] = 0  # not used
+
+        # all but time-dependent part of p
+        self.delta = self.ctx.matrix(M, 1)
+        self.delta[0] = self.r
+
+        for i in range(1, M):
+            self.cot_theta[i] = self.ctx.cot(self.theta[i])
+            self.delta[i] = self.r*self.theta[i]*(self.cot_theta[i] + 1j)
+
+        self.p = self.ctx.matrix(M, 1)
+        self.p = self.delta/self.tmax
+
+        # NB: p is complex (mpc)
+
+    def calc_time_domain_solution(self, fp, t, manual_prec=False):
+        r"""The fixed Talbot time-domain solution is computed from the
+        Laplace-space function evaluations using
+
+        .. math ::
+
+            f(t,M)=\frac{2}{5t}\sum_{k=0}^{M-1}\Re \left[
+            \gamma_k \bar{f}(p_k)\right]
+
+        where
+
+        .. math ::
+
+            \gamma_0 = \frac{1}{2}e^{r}\bar{f}(p_0)
+
+        .. math ::
+
+            \gamma_k = e^{tp_k}\left\lbrace 1 + \frac{jk\pi}{M}\left[1 +
+            \cot \left( \frac{k \pi}{M} \right)^2 \right] - j\cot\left(
+            \frac{k \pi}{M}\right)\right \rbrace \qquad 1\le k<M.
+
+        Again, `j=\sqrt{-1}`.
+
+        Before calling this function, call
+        :class:`~mpmath.calculus.inverselaplace.FixedTalbot.calc_laplace_parameter`
+        to set the parameters and compute the required coefficients.
+
+        **References**
+
+        1. Abate, J., P. Valko (2004). Multi-precision Laplace
+           transform inversion. *International Journal for Numerical
+           Methods in Engineering* 60:979-993,
+           http://dx.doi.org/10.1002/nme.995
+        2. Talbot, A. (1979). The accurate numerical inversion of
+           Laplace transforms. *IMA Journal of Applied Mathematics*
+           23(1):97, http://dx.doi.org/10.1093/imamat/23.1.97
+        """
+
+        # required
+        # ------------------------------
+        self.t = self.ctx.convert(t)
+
+        # assume fp was computed from p matrix returned from
+        # calc_laplace_parameter(), so is already a list or matrix of
+        # mpmath 'mpc' types
+
+        # these were computed in previous call to
+        # calc_laplace_parameter()
+        theta = self.theta
+        delta = self.delta
+        M = self.degree
+        p = self.p
+        r = self.r
+
+        ans = self.ctx.matrix(M, 1)
+        ans[0] = self.ctx.exp(delta[0])*fp[0]/2
+
+        for i in range(1, M):
+            ans[i] = self.ctx.exp(delta[i])*fp[i]*(
+                1 + 1j*theta[i]*(1 + self.cot_theta[i]**2) -
+                1j*self.cot_theta[i])
+
+        result = self.ctx.fraction(2, 5)*self.ctx.fsum(ans)/self.t
+
+        # setting dps back to value when calc_laplace_parameter was
+        # called, unless flag is set.
+        if not manual_prec:
+            self.ctx.dps = self.dps_orig
+
+        return result.real
+
+
+# ****************************************
+
+class Stehfest(InverseLaplaceTransform):
+
+    def calc_laplace_parameter(self, t, **kwargs):
+        r"""
+        The Gaver-Stehfest method is a discrete approximation of the
+        Widder-Post inversion algorithm, rather than a direct
+        approximation of the Bromwich contour integral.
+
+        The method abscissa along the real axis, and therefore has
+        issues inverting oscillatory functions (which have poles in
+        pairs away from the real axis).
+
+        The working precision will be increased according to a rule of
+        thumb. If 'degree' is not specified, the working precision and
+        degree are chosen to hopefully achieve the dps of the calling
+        context. If 'degree' is specified, the working precision is
+        chosen to achieve maximum resulting precision for the
+        specified degree.
+
+        .. math ::
+
+            p_k = \frac{k \log 2}{t} \qquad 1 \le k \le M
+        """
+
+        # required
+        # ------------------------------
+        # time of desired approximation
+        self.t = self.ctx.convert(t)
+
+        # optional
+        # ------------------------------
+
+        # empirical relationships used here based on a linear fit of
+        # requested and delivered dps for exponentially decaying time
+        # functions for requested dps up to 512.
+
+        if 'degree' in kwargs:
+            self.degree = kwargs['degree']
+            self.dps_goal = int(1.38*self.degree)
+        else:
+            self.dps_goal = int(2.93*self.ctx.dps)
+            self.degree = max(16, self.dps_goal)
+
+        # _coeff routine requires even degree
+        if self.degree % 2 > 0:
+            self.degree += 1
+
+        M = self.degree
+
+        # this is adjusting the dps of the calling context
+        # hopefully the caller doesn't monkey around with it
+        # between calling this routine and calc_time_domain_solution()
+        self.dps_orig = self.ctx.dps
+        self.ctx.dps = self.dps_goal
+
+        self.V = self._coeff()
+        self.p = self.ctx.matrix(self.ctx.arange(1, M+1))*self.ctx.ln2/self.t
+
+        # NB: p is real (mpf)
+
+    def _coeff(self):
+        r"""Salzer summation weights (aka, "Stehfest coefficients")
+        only depend on the approximation order (M) and the precision"""
+
+        M = self.degree
+        M2 = int(M/2)  # checked earlier that M is even
+
+        V = self.ctx.matrix(M, 1)
+
+        # Salzer summation weights
+        # get very large in magnitude and oscillate in sign,
+        # if the precision is not high enough, there will be
+        # catastrophic cancellation
+        for k in range(1, M+1):
+            z = self.ctx.matrix(min(k, M2)+1, 1)
+            for j in range(int((k+1)/2), min(k, M2)+1):
+                z[j] = (self.ctx.power(j, M2)*self.ctx.fac(2*j)/
+                        (self.ctx.fac(M2-j)*self.ctx.fac(j)*
+                         self.ctx.fac(j-1)*self.ctx.fac(k-j)*
+                         self.ctx.fac(2*j-k)))
+            V[k-1] = self.ctx.power(-1, k+M2)*self.ctx.fsum(z)
+
+        return V
+
+    def calc_time_domain_solution(self, fp, t, manual_prec=False):
+        r"""Compute time-domain Stehfest algorithm solution.
+
+        .. math ::
+
+            f(t,M) = \frac{\log 2}{t} \sum_{k=1}^{M} V_k \bar{f}\left(
+            p_k \right)
+
+        where
+
+        .. math ::
+
+            V_k = (-1)^{k + N/2} \sum^{\min(k,N/2)}_{i=\lfloor(k+1)/2 \rfloor}
+            \frac{i^{\frac{N}{2}}(2i)!}{\left(\frac{N}{2}-i \right)! \, i! \,
+            \left(i-1 \right)! \, \left(k-i\right)! \, \left(2i-k \right)!}
+
+        As the degree increases, the abscissa (`p_k`) only increase
+        linearly towards `\infty`, but the Stehfest coefficients
+        (`V_k`) alternate in sign and increase rapidly in sign,
+        requiring high precision to prevent overflow or loss of
+        significance when evaluating the sum.
+
+        **References**
+
+        1. Widder, D. (1941). *The Laplace Transform*. Princeton.
+        2. Stehfest, H. (1970). Algorithm 368: numerical inversion of
+           Laplace transforms. *Communications of the ACM* 13(1):47-49,
+           http://dx.doi.org/10.1145/361953.361969
+
+        """
+
+        # required
+        self.t = self.ctx.convert(t)
+
+        # assume fp was computed from p matrix returned from
+        # calc_laplace_parameter(), so is already
+        # a list or matrix of mpmath 'mpf' types
+
+        result = self.ctx.fdot(self.V, fp)*self.ctx.ln2/self.t
+
+        # setting dps back to value when calc_laplace_parameter was called
+        if not manual_prec:
+            self.ctx.dps = self.dps_orig
+
+        # ignore any small imaginary part
+        return result.real
+
+
+# ****************************************
+
+class deHoog(InverseLaplaceTransform):
+
+    def calc_laplace_parameter(self, t, **kwargs):
+        r"""the de Hoog, Knight & Stokes algorithm is an
+        accelerated form of the Fourier series numerical
+        inverse Laplace transform algorithms.
+
+        .. math ::
+
+            p_k = \gamma + \frac{jk}{T} \qquad 0 \le k < 2M+1
+
+        where
+
+        .. math ::
+
+            \gamma = \alpha - \frac{\log \mathrm{tol}}{2T},
+
+        `j=\sqrt{-1}`, `T = 2t_\mathrm{max}` is a scaled time,
+        `\alpha=10^{-\mathrm{dps\_goal}}` is the real part of the
+        rightmost pole or singularity, which is chosen based on the
+        desired accuracy (assuming the rightmost singularity is 0),
+        and `\mathrm{tol}=10\alpha` is the desired tolerance, which is
+        chosen in relation to `\alpha`.`
+
+        When increasing the degree, the abscissa increase towards
+        `j\infty`, but more slowly than the fixed Talbot
+        algorithm. The de Hoog et al. algorithm typically does better
+        with oscillatory functions of time, and less well-behaved
+        functions. The method tends to be slower than the Talbot and
+        Stehfest algorithsm, especially so at very high precision
+        (e.g., `>500` digits precision).
+
+        """
+
+        # required
+        # ------------------------------
+        self.t = self.ctx.convert(t)
+
+        # optional
+        # ------------------------------
+        self.tmax = kwargs.get('tmax', self.t)
+
+        # empirical relationships used here based on a linear fit of
+        # requested and delivered dps for exponentially decaying time
+        # functions for requested dps up to 512.
+
+        if 'degree' in kwargs:
+            self.degree = kwargs['degree']
+            self.dps_goal = int(1.38*self.degree)
+        else:
+            self.dps_goal = int(self.ctx.dps*1.36)
+            self.degree = max(10, self.dps_goal)
+
+        # 2*M+1 terms in approximation
+        M = self.degree
+
+        # adjust alpha component of abscissa of convergence for higher
+        # precision
+        tmp = self.ctx.power(10.0, -self.dps_goal)
+        self.alpha = self.ctx.convert(kwargs.get('alpha', tmp))
+
+        # desired tolerance (here simply related to alpha)
+        self.tol = self.ctx.convert(kwargs.get('tol', self.alpha*10.0))
+        self.np = 2*self.degree+1  # number of terms in approximation
+
+        # this is adjusting the dps of the calling context
+        # hopefully the caller doesn't monkey around with it
+        # between calling this routine and calc_time_domain_solution()
+        self.dps_orig = self.ctx.dps
+        self.ctx.dps = self.dps_goal
+
+        # scaling factor (likely tun-able, but 2 is typical)
+        self.scale = kwargs.get('scale', 2)
+        self.T = self.ctx.convert(kwargs.get('T', self.scale*self.tmax))
+
+        self.p = self.ctx.matrix(2*M+1, 1)
+        self.gamma = self.alpha - self.ctx.log(self.tol)/(self.scale*self.T)
+        self.p = (self.gamma + self.ctx.pi*
+                  self.ctx.matrix(self.ctx.arange(self.np))/self.T*1j)
+
+        # NB: p is complex (mpc)
+
+    def calc_time_domain_solution(self, fp, t, manual_prec=False):
+        r"""Calculate time-domain solution for
+        de Hoog, Knight & Stokes algorithm.
+
+        The un-accelerated Fourier series approach is:
+
+        .. math ::
+
+            f(t,2M+1) = \frac{e^{\gamma t}}{T} \sum_{k=0}^{2M}{}^{'}
+            \Re\left[\bar{f}\left( p_k \right)
+            e^{i\pi t/T} \right],
+
+        where the prime on the summation indicates the first term is halved.
+
+        This simplistic approach requires so many function evaluations
+        that it is not practical. Non-linear acceleration is
+        accomplished via Pade-approximation and an analytic expression
+        for the remainder of the continued fraction. See the original
+        paper (reference 2 below) a detailed description of the
+        numerical approach.
+
+        **References**
+
+        1. Davies, B. (2005). *Integral Transforms and their
+           Applications*, Third Edition. Springer.
+        2. de Hoog, F., J. Knight, A. Stokes (1982). An improved
+           method for numerical inversion of Laplace transforms. *SIAM
+           Journal of Scientific and Statistical Computing* 3:357-366,
+           http://dx.doi.org/10.1137/0903022
+
+        """
+
+        M = self.degree
+        np = self.np
+        T = self.T
+
+        self.t = self.ctx.convert(t)
+
+        # would it be useful to try re-using
+        # space between e&q and A&B?
+        e = self.ctx.zeros(np, M+1)
+        q = self.ctx.matrix(2*M, M)
+        d = self.ctx.matrix(np, 1)
+        A = self.ctx.zeros(np+1, 1)
+        B = self.ctx.ones(np+1, 1)
+
+        # initialize Q-D table
+        e[:, 0] = 0.0 + 0j
+        q[0, 0] = fp[1]/(fp[0]/2)
+        for i in range(1, 2*M):
+            q[i, 0] = fp[i+1]/fp[i]
+
+        # rhombus rule for filling triangular Q-D table (e & q)
+        for r in range(1, M+1):
+            # start with e, column 1, 0:2*M-2
+            mr = 2*(M-r) + 1
+            e[0:mr, r] = q[1:mr+1, r-1] - q[0:mr, r-1] + e[1:mr+1, r-1]
+            if not r == M:
+                rq = r+1
+                mr = 2*(M-rq)+1 + 2
+                for i in range(mr):
+                    q[i, rq-1] = q[i+1, rq-2]*e[i+1, rq-1]/e[i, rq-1]
+
+        # build up continued fraction coefficients (d)
+        d[0] = fp[0]/2
+        for r in range(1, M+1):
+            d[2*r-1] = -q[0, r-1]  # even terms
+            d[2*r]   = -e[0, r]    # odd terms
+
+        # seed A and B for recurrence
+        A[0] = 0.0 + 0.0j
+        A[1] = d[0]
+        B[0:2] = 1.0 + 0.0j
+
+        # base of the power series
+        z = self.ctx.expjpi(self.t/T)  # i*pi is already in fcn
+
+        # coefficients of Pade approximation (A & B)
+        # using recurrence for all but last term
+        for i in range(1, 2*M):
+            A[i+1] = A[i] + d[i]*A[i-1]*z
+            B[i+1] = B[i] + d[i]*B[i-1]*z
+
+        # "improved remainder" to continued fraction
+        brem = (1 + (d[2*M-1] - d[2*M])*z)/2
+        # powm1(x,y) computes x^y - 1 more accurately near zero
+        rem = brem*self.ctx.powm1(1 + d[2*M]*z/brem,
+                                  self.ctx.fraction(1, 2))
+
+        # last term of recurrence using new remainder
+        A[np] = A[2*M] + rem*A[2*M-1]
+        B[np] = B[2*M] + rem*B[2*M-1]
+
+        # diagonal Pade approximation
+        # F=A/B represents accelerated trapezoid rule
+        result = self.ctx.exp(self.gamma*self.t)/T*(A[np]/B[np]).real
+
+        # setting dps back to value when calc_laplace_parameter was called
+        if not manual_prec:
+            self.ctx.dps = self.dps_orig
+
+        return result
+
+
+# ****************************************
+
+class Cohen(InverseLaplaceTransform):
+
+    def calc_laplace_parameter(self, t, **kwargs):
+        r"""The Cohen algorithm accelerates the convergence of the nearly
+        alternating series resulting from the application of the trapezoidal
+        rule to the Bromwich contour inversion integral.
+
+        .. math ::
+
+            p_k = \frac{\gamma}{2 t} + \frac{\pi i k}{t} \qquad 0 \le k < M
+
+        where
+
+        .. math ::
+
+            \gamma = \frac{2}{3} (d + \log(10) + \log(2 t)),
+
+        `d = \mathrm{dps\_goal}`, which is chosen based on the desired
+        accuracy using the method developed in [1] to improve numerical
+        stability. The Cohen algorithm shows robustness similar to the de Hoog
+        et al. algorithm, but it is faster than the fixed Talbot algorithm.
+
+        **Optional arguments**
+
+        *degree*
+            integer order of the approximation (M = number of terms)
+        *alpha*
+            abscissa for `p_0` (controls the discretization error)
+
+        The working precision will be increased according to a rule of
+        thumb. If 'degree' is not specified, the working precision and
+        degree are chosen to hopefully achieve the dps of the calling
+        context. If 'degree' is specified, the working precision is
+        chosen to achieve maximum resulting precision for the
+        specified degree.
+
+        **References**
+
+        1. P. Glasserman, J. Ruiz-Mata (2006). Computing the credit loss
+        distribution in the Gaussian copula model: a comparison of methods.
+        *Journal of Credit Risk* 2(4):33-66, 10.21314/JCR.2006.057
+
+        """
+        self.t = self.ctx.convert(t)
+
+        if 'degree' in kwargs:
+            self.degree = kwargs['degree']
+            self.dps_goal = int(1.5 * self.degree)
+        else:
+            self.dps_goal = int(self.ctx.dps * 1.74)
+            self.degree = max(22, int(1.31 * self.dps_goal))
+
+        M = self.degree + 1
+
+        # this is adjusting the dps of the calling context hopefully
+        # the caller doesn't monkey around with it between calling
+        # this routine and calc_time_domain_solution()
+        self.dps_orig = self.ctx.dps
+        self.ctx.dps = self.dps_goal
+
+        ttwo = 2 * self.t
+        tmp = self.ctx.dps * self.ctx.log(10) + self.ctx.log(ttwo)
+        tmp = self.ctx.fraction(2, 3) * tmp
+        self.alpha = self.ctx.convert(kwargs.get('alpha', tmp))
+
+        # all but time-dependent part of p
+        a_t = self.alpha / ttwo
+        p_t = self.ctx.pi * 1j / self.t
+
+        self.p = self.ctx.matrix(M, 1)
+        self.p[0] = a_t
+
+        for i in range(1, M):
+            self.p[i] = a_t + i * p_t
+
+    def calc_time_domain_solution(self, fp, t, manual_prec=False):
+        r"""Calculate time-domain solution for Cohen algorithm.
+
+        The accelerated nearly alternating series is:
+
+        .. math ::
+
+            f(t, M) = \frac{e^{\gamma / 2}}{t} \left[\frac{1}{2}
+            \Re\left(\bar{f}\left(\frac{\gamma}{2t}\right) \right) -
+            \sum_{k=0}^{M-1}\frac{c_{M,k}}{d_M}\Re\left(\bar{f}
+            \left(\frac{\gamma + 2(k+1) \pi i}{2t}\right)\right)\right],
+
+        where coefficients `\frac{c_{M, k}}{d_M}` are described in [1].
+
+        1. H. Cohen, F. Rodriguez Villegas, D. Zagier (2000). Convergence
+        acceleration of alternating series. *Experiment. Math* 9(1):3-12
+
+        """
+        self.t = self.ctx.convert(t)
+
+        n = self.degree
+        M = n + 1
+
+        A = self.ctx.matrix(M, 1)
+        for i in range(M):
+            A[i] = fp[i].real
+
+        d = (3 + self.ctx.sqrt(8)) ** n
+        d = (d + 1 / d) / 2
+        b = -self.ctx.one
+        c = -d
+        s = 0
+
+        for k in range(n):
+            c = b - c
+            s = s + c * A[k + 1]
+            b = 2 * (k + n) * (k - n) * b / ((2 * k + 1) * (k + self.ctx.one))
+
+        result = self.ctx.exp(self.alpha / 2) / self.t * (A[0] / 2 - s / d)
+
+        # setting dps back to value when calc_laplace_parameter was
+        # called, unless flag is set.
+        if not manual_prec:
+            self.ctx.dps = self.dps_orig
+
+        return result
+
+
+# ****************************************
+
+class LaplaceTransformInversionMethods(object):
+    def __init__(ctx, *args, **kwargs):
+        ctx._fixed_talbot = FixedTalbot(ctx)
+        ctx._stehfest = Stehfest(ctx)
+        ctx._de_hoog = deHoog(ctx)
+        ctx._cohen = Cohen(ctx)
+
+    def invertlaplace(ctx, f, t, **kwargs):
+        r"""Computes the numerical inverse Laplace transform for a
+        Laplace-space function at a given time.  The function being
+        evaluated is assumed to be a real-valued function of time.
+
+        The user must supply a Laplace-space function `\bar{f}(p)`,
+        and a desired time at which to estimate the time-domain
+        solution `f(t)`.
+
+        A few basic examples of Laplace-space functions with known
+        inverses (see references [1,2]) :
+
+        .. math ::
+
+            \mathcal{L}\left\lbrace f(t) \right\rbrace=\bar{f}(p)
+
+        .. math ::
+
+            \mathcal{L}^{-1}\left\lbrace \bar{f}(p) \right\rbrace = f(t)
+
+        .. math ::
+
+            \bar{f}(p) = \frac{1}{(p+1)^2}
+
+        .. math ::
+
+            f(t) = t e^{-t}
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> tt = [0.001, 0.01, 0.1, 1, 10]
+        >>> fp = lambda p: 1/(p+1)**2
+        >>> ft = lambda t: t*exp(-t)
+        >>> ft(tt[0]),ft(tt[0])-invertlaplace(fp,tt[0],method='talbot')
+        (0.000999000499833375, 8.57923043561212e-20)
+        >>> ft(tt[1]),ft(tt[1])-invertlaplace(fp,tt[1],method='talbot')
+        (0.00990049833749168, 3.27007646698047e-19)
+        >>> ft(tt[2]),ft(tt[2])-invertlaplace(fp,tt[2],method='talbot')
+        (0.090483741803596, -1.75215800052168e-18)
+        >>> ft(tt[3]),ft(tt[3])-invertlaplace(fp,tt[3],method='talbot')
+        (0.367879441171442, 1.2428864009344e-17)
+        >>> ft(tt[4]),ft(tt[4])-invertlaplace(fp,tt[4],method='talbot')
+        (0.000453999297624849, 4.04513489306658e-20)
+
+        The methods also work for higher precision:
+
+        >>> mp.dps = 100; mp.pretty = True
+        >>> nstr(ft(tt[0]),15),nstr(ft(tt[0])-invertlaplace(fp,tt[0],method='talbot'),15)
+        ('0.000999000499833375', '-4.96868310693356e-105')
+        >>> nstr(ft(tt[1]),15),nstr(ft(tt[1])-invertlaplace(fp,tt[1],method='talbot'),15)
+        ('0.00990049833749168', '1.23032291513122e-104')
+
+        .. math ::
+
+            \bar{f}(p) = \frac{1}{p^2+1}
+
+        .. math ::
+
+            f(t) = \mathrm{J}_0(t)
+
+        >>> mp.dps = 15; mp.pretty = True
+        >>> fp = lambda p: 1/sqrt(p*p + 1)
+        >>> ft = lambda t: besselj(0,t)
+        >>> ft(tt[0]),ft(tt[0])-invertlaplace(fp,tt[0],method='dehoog')
+        (0.999999750000016, -6.09717765032273e-18)
+        >>> ft(tt[1]),ft(tt[1])-invertlaplace(fp,tt[1],method='dehoog')
+        (0.99997500015625, -5.61756281076169e-17)
+
+        .. math ::
+
+            \bar{f}(p) = \frac{\log p}{p}
+
+        .. math ::
+
+            f(t) = -\gamma -\log t
+
+        >>> mp.dps = 15; mp.pretty = True
+        >>> fp = lambda p: log(p)/p
+        >>> ft = lambda t: -euler-log(t)
+        >>> ft(tt[0]),ft(tt[0])-invertlaplace(fp,tt[0],method='stehfest')
+        (6.3305396140806, -1.92126634837863e-16)
+        >>> ft(tt[1]),ft(tt[1])-invertlaplace(fp,tt[1],method='stehfest')
+        (4.02795452108656, -4.81486093200704e-16)
+
+        **Options**
+
+        :func:`~mpmath.invertlaplace` recognizes the following optional
+        keywords valid for all methods:
+
+        *method*
+            Chooses numerical inverse Laplace transform algorithm
+            (described below).
+        *degree*
+            Number of terms used in the approximation
+
+        **Algorithms**
+
+        Mpmath implements four numerical inverse Laplace transform
+        algorithms, attributed to: Talbot, Stehfest, and de Hoog,
+        Knight and Stokes. These can be selected by using
+        *method='talbot'*, *method='stehfest'*, *method='dehoog'* or
+        *method='cohen'* or by passing the classes *method=FixedTalbot*,
+        *method=Stehfest*, *method=deHoog*, or *method=Cohen*. The functions
+        :func:`~mpmath.invlaptalbot`, :func:`~mpmath.invlapstehfest`,
+        :func:`~mpmath.invlapdehoog`, and :func:`~mpmath.invlapcohen`
+        are also available as shortcuts.
+
+        All four algorithms implement a heuristic balance between the
+        requested precision and the precision used internally for the
+        calculations. This has been tuned for a typical exponentially
+        decaying function and precision up to few hundred decimal
+        digits.
+
+        The Laplace transform converts the variable time (i.e., along
+        a line) into a parameter given by the right half of the
+        complex `p`-plane.  Singularities, poles, and branch cuts in
+        the complex `p`-plane contain all the information regarding
+        the time behavior of the corresponding function. Any numerical
+        method must therefore sample `p`-plane "close enough" to the
+        singularities to accurately characterize them, while not
+        getting too close to have catastrophic cancellation, overflow,
+        or underflow issues. Most significantly, if one or more of the
+        singularities in the `p`-plane is not on the left side of the
+        Bromwich contour, its effects will be left out of the computed
+        solution, and the answer will be completely wrong.
+
+        *Talbot*
+
+        The fixed Talbot method is high accuracy and fast, but the
+        method can catastrophically fail for certain classes of time-domain
+        behavior, including a Heaviside step function for positive
+        time (e.g., `H(t-2)`), or some oscillatory behaviors. The
+        Talbot method usually has adjustable parameters, but the
+        "fixed" variety implemented here does not. This method
+        deforms the Bromwich integral contour in the shape of a
+        parabola towards `-\infty`, which leads to problems
+        when the solution has a decaying exponential in it (e.g., a
+        Heaviside step function is equivalent to multiplying by a
+        decaying exponential in Laplace space).
+
+        *Stehfest*
+
+        The Stehfest algorithm only uses abscissa along the real axis
+        of the complex `p`-plane to estimate the time-domain
+        function. Oscillatory time-domain functions have poles away
+        from the real axis, so this method does not work well with
+        oscillatory functions, especially high-frequency ones. This
+        method also depends on summation of terms in a series that
+        grows very large, and will have catastrophic cancellation
+        during summation if the working precision is too low.
+
+        *de Hoog et al.*
+
+        The de Hoog, Knight, and Stokes method is essentially a
+        Fourier-series quadrature-type approximation to the Bromwich
+        contour integral, with non-linear series acceleration and an
+        analytical expression for the remainder term. This method is
+        typically one of the most robust. This method also involves the
+        greatest amount of overhead, so it is typically the slowest of the
+        four methods at high precision.
+
+        *Cohen*
+
+        The Cohen method is a trapezoidal rule approximation to the Bromwich
+        contour integral, with linear acceleration for alternating
+        series. This method is as robust as the de Hoog et al method and the
+        fastest of the four methods at high precision, and is therefore the
+        default method.
+
+        **Singularities**
+
+        All numerical inverse Laplace transform methods have problems
+        at large time when the Laplace-space function has poles,
+        singularities, or branch cuts to the right of the origin in
+        the complex plane. For simple poles in `\bar{f}(p)` at the
+        `p`-plane origin, the time function is constant in time (e.g.,
+        `\mathcal{L}\left\lbrace 1 \right\rbrace=1/p` has a pole at
+        `p=0`). A pole in `\bar{f}(p)` to the left of the origin is a
+        decreasing function of time (e.g., `\mathcal{L}\left\lbrace
+        e^{-t/2} \right\rbrace=1/(p+1/2)` has a pole at `p=-1/2`), and
+        a pole to the right of the origin leads to an increasing
+        function in time (e.g., `\mathcal{L}\left\lbrace t e^{t/4}
+        \right\rbrace = 1/(p-1/4)^2` has a pole at `p=1/4`).  When
+        singularities occur off the real `p` axis, the time-domain
+        function is oscillatory. For example `\mathcal{L}\left\lbrace
+        \mathrm{J}_0(t) \right\rbrace=1/\sqrt{p^2+1}` has a branch cut
+        starting at `p=j=\sqrt{-1}` and is a decaying oscillatory
+        function, This range of behaviors is illustrated in Duffy [3]
+        Figure 4.10.4, p. 228.
+
+        In general as `p \rightarrow \infty` `t \rightarrow 0` and
+        vice-versa. All numerical inverse Laplace transform methods
+        require their abscissa to shift closer to the origin for
+        larger times. If the abscissa shift left of the rightmost
+        singularity in the Laplace domain, the answer will be
+        completely wrong (the effect of singularities to the right of
+        the Bromwich contour are not included in the results).
+
+        For example, the following exponentially growing function has
+        a pole at `p=3`:
+
+        .. math ::
+
+            \bar{f}(p)=\frac{1}{p^2-9}
+
+        .. math ::
+
+            f(t)=\frac{1}{3}\sinh 3t
+
+        >>> mp.dps = 15; mp.pretty = True
+        >>> fp = lambda p: 1/(p*p-9)
+        >>> ft = lambda t: sinh(3*t)/3
+        >>> tt = [0.01,0.1,1.0,10.0]
+        >>> ft(tt[0]),invertlaplace(fp,tt[0],method='talbot')
+        (0.0100015000675014, 0.0100015000675014)
+        >>> ft(tt[1]),invertlaplace(fp,tt[1],method='talbot')
+        (0.101506764482381, 0.101506764482381)
+        >>> ft(tt[2]),invertlaplace(fp,tt[2],method='talbot')
+        (3.33929164246997, 3.33929164246997)
+        >>> ft(tt[3]),invertlaplace(fp,tt[3],method='talbot')
+        (1781079096920.74, -1.61331069624091e-14)
+
+        **References**
+
+        1. [DLMF]_ section 1.14 (http://dlmf.nist.gov/1.14T4)
+        2. Cohen, A.M. (2007). Numerical Methods for Laplace Transform
+           Inversion, Springer.
+        3. Duffy, D.G. (1998). Advanced Engineering Mathematics, CRC Press.
+
+        **Numerical Inverse Laplace Transform Reviews**
+
+        1. Bellman, R., R.E. Kalaba, J.A. Lockett (1966). *Numerical
+           inversion of the Laplace transform: Applications to Biology,
+           Economics, Engineering, and Physics*. Elsevier.
+        2. Davies, B., B. Martin (1979). Numerical inversion of the
+           Laplace transform: a survey and comparison of methods. *Journal
+           of Computational Physics* 33:1-32,
+           http://dx.doi.org/10.1016/0021-9991(79)90025-1
+        3. Duffy, D.G. (1993). On the numerical inversion of Laplace
+           transforms: Comparison of three new methods on characteristic
+           problems from applications. *ACM Transactions on Mathematical
+           Software* 19(3):333-359, http://dx.doi.org/10.1145/155743.155788
+        4. Kuhlman, K.L., (2013). Review of Inverse Laplace Transform
+           Algorithms for Laplace-Space Numerical Approaches, *Numerical
+           Algorithms*, 63(2):339-355.
+           http://dx.doi.org/10.1007/s11075-012-9625-3
+
+        """
+
+        rule = kwargs.get('method', 'cohen')
+        if type(rule) is str:
+            lrule = rule.lower()
+            if lrule == 'talbot':
+                rule = ctx._fixed_talbot
+            elif lrule == 'stehfest':
+                rule = ctx._stehfest
+            elif lrule == 'dehoog':
+                rule = ctx._de_hoog
+            elif rule == 'cohen':
+                rule = ctx._cohen
+            else:
+                raise ValueError("unknown invlap algorithm: %s" % rule)
+        else:
+            rule = rule(ctx)
+
+        # determine the vector of Laplace-space parameter
+        # needed for the requested method and desired time
+        rule.calc_laplace_parameter(t, **kwargs)
+
+        # compute the Laplace-space function evalutations
+        # at the required abscissa.
+        fp = [f(p) for p in rule.p]
+
+        # compute the time-domain solution from the
+        # Laplace-space function evaluations
+        return rule.calc_time_domain_solution(fp, t)
+
+    # shortcuts for the above function for specific methods
+    def invlaptalbot(ctx, *args, **kwargs):
+        kwargs['method'] = 'talbot'
+        return ctx.invertlaplace(*args, **kwargs)
+
+    def invlapstehfest(ctx, *args, **kwargs):
+        kwargs['method'] = 'stehfest'
+        return ctx.invertlaplace(*args, **kwargs)
+
+    def invlapdehoog(ctx, *args, **kwargs):
+        kwargs['method'] = 'dehoog'
+        return ctx.invertlaplace(*args, **kwargs)
+
+    def invlapcohen(ctx, *args, **kwargs):
+        kwargs['method'] = 'cohen'
+        return ctx.invertlaplace(*args, **kwargs)
+
+
+# ****************************************
+
+if __name__ == '__main__':
+    import doctest
+    doctest.testmod()

lib/python3.11/site-packages/mpmath/calculus/odes.py

@@ -0,0 +1,288 @@
+from bisect import bisect
+from ..libmp.backend import xrange
+
+class ODEMethods(object):
+    pass
+
+def ode_taylor(ctx, derivs, x0, y0, tol_prec, n):
+    h = tol = ctx.ldexp(1, -tol_prec)
+    dim = len(y0)
+    xs = [x0]
+    ys = [y0]
+    x = x0
+    y = y0
+    orig = ctx.prec
+    try:
+        ctx.prec = orig*(1+n)
+        # Use n steps with Euler's method to get
+        # evaluation points for derivatives
+        for i in range(n):
+            fxy = derivs(x, y)
+            y = [y[i]+h*fxy[i] for i in xrange(len(y))]
+            x += h
+            xs.append(x)
+            ys.append(y)
+        # Compute derivatives
+        ser = [[] for d in range(dim)]
+        for j in range(n+1):
+            s = [0]*dim
+            b = (-1) ** (j & 1)
+            k = 1
+            for i in range(j+1):
+                for d in range(dim):
+                    s[d] += b * ys[i][d]
+                b = (b * (j-k+1)) // (-k)
+                k += 1
+            scale = h**(-j) / ctx.fac(j)
+            for d in range(dim):
+                s[d] = s[d] * scale
+                ser[d].append(s[d])
+    finally:
+        ctx.prec = orig
+    # Estimate radius for which we can get full accuracy.
+    # XXX: do this right for zeros
+    radius = ctx.one
+    for ts in ser:
+        if ts[-1]:
+            radius = min(radius, ctx.nthroot(tol/abs(ts[-1]), n))
+    radius /= 2  # XXX
+    return ser, x0+radius
+
+def odefun(ctx, F, x0, y0, tol=None, degree=None, method='taylor', verbose=False):
+    r"""
+    Returns a function `y(x) = [y_0(x), y_1(x), \ldots, y_n(x)]`
+    that is a numerical solution of the `n+1`-dimensional first-order
+    ordinary differential equation (ODE) system
+
+    .. math ::
+
+        y_0'(x) = F_0(x, [y_0(x), y_1(x), \ldots, y_n(x)])
+
+        y_1'(x) = F_1(x, [y_0(x), y_1(x), \ldots, y_n(x)])
+
+        \vdots
+
+        y_n'(x) = F_n(x, [y_0(x), y_1(x), \ldots, y_n(x)])
+
+    The derivatives are specified by the vector-valued function
+    *F* that evaluates
+    `[y_0', \ldots, y_n'] = F(x, [y_0, \ldots, y_n])`.
+    The initial point `x_0` is specified by the scalar argument *x0*,
+    and the initial value `y(x_0) =  [y_0(x_0), \ldots, y_n(x_0)]` is
+    specified by the vector argument *y0*.
+
+    For convenience, if the system is one-dimensional, you may optionally
+    provide just a scalar value for *y0*. In this case, *F* should accept
+    a scalar *y* argument and return a scalar. The solution function
+    *y* will return scalar values instead of length-1 vectors.
+
+    Evaluation of the solution function `y(x)` is permitted
+    for any `x \ge x_0`.
+
+    A high-order ODE can be solved by transforming it into first-order
+    vector form. This transformation is described in standard texts
+    on ODEs. Examples will also be given below.
+
+    **Options, speed and accuracy**
+
+    By default, :func:`~mpmath.odefun` uses a high-order Taylor series
+    method. For reasonably well-behaved problems, the solution will
+    be fully accurate to within the working precision. Note that
+    *F* must be possible to evaluate to very high precision
+    for the generation of Taylor series to work.
+
+    To get a faster but less accurate solution, you can set a large
+    value for *tol* (which defaults roughly to *eps*). If you just
+    want to plot the solution or perform a basic simulation,
+    *tol = 0.01* is likely sufficient.
+
+    The *degree* argument controls the degree of the solver (with
+    *method='taylor'*, this is the degree of the Taylor series
+    expansion). A higher degree means that a longer step can be taken
+    before a new local solution must be generated from *F*,
+    meaning that fewer steps are required to get from `x_0` to a given
+    `x_1`. On the other hand, a higher degree also means that each
+    local solution becomes more expensive (i.e., more evaluations of
+    *F* are required per step, and at higher precision).
+
+    The optimal setting therefore involves a tradeoff. Generally,
+    decreasing the *degree* for Taylor series is likely to give faster
+    solution at low precision, while increasing is likely to be better
+    at higher precision.
+
+    The function
+    object returned by :func:`~mpmath.odefun` caches the solutions at all step
+    points and uses polynomial interpolation between step points.
+    Therefore, once `y(x_1)` has been evaluated for some `x_1`,
+    `y(x)` can be evaluated very quickly for any `x_0 \le x \le x_1`.
+    and continuing the evaluation up to `x_2 > x_1` is also fast.
+
+    **Examples of first-order ODEs**
+
+    We will solve the standard test problem `y'(x) = y(x), y(0) = 1`
+    which has explicit solution `y(x) = \exp(x)`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> f = odefun(lambda x, y: y, 0, 1)
+        >>> for x in [0, 1, 2.5]:
+        ...     print((f(x), exp(x)))
+        ...
+        (1.0, 1.0)
+        (2.71828182845905, 2.71828182845905)
+        (12.1824939607035, 12.1824939607035)
+
+    The solution with high precision::
+
+        >>> mp.dps = 50
+        >>> f = odefun(lambda x, y: y, 0, 1)
+        >>> f(1)
+        2.7182818284590452353602874713526624977572470937
+        >>> exp(1)
+        2.7182818284590452353602874713526624977572470937
+
+    Using the more general vectorized form, the test problem
+    can be input as (note that *f* returns a 1-element vector)::
+
+        >>> mp.dps = 15
+        >>> f = odefun(lambda x, y: [y[0]], 0, [1])
+        >>> f(1)
+        [2.71828182845905]
+
+    :func:`~mpmath.odefun` can solve nonlinear ODEs, which are generally
+    impossible (and at best difficult) to solve analytically. As
+    an example of a nonlinear ODE, we will solve `y'(x) = x \sin(y(x))`
+    for `y(0) = \pi/2`. An exact solution happens to be known
+    for this problem, and is given by
+    `y(x) = 2 \tan^{-1}\left(\exp\left(x^2/2\right)\right)`::
+
+        >>> f = odefun(lambda x, y: x*sin(y), 0, pi/2)
+        >>> for x in [2, 5, 10]:
+        ...     print((f(x), 2*atan(exp(mpf(x)**2/2))))
+        ...
+        (2.87255666284091, 2.87255666284091)
+        (3.14158520028345, 3.14158520028345)
+        (3.14159265358979, 3.14159265358979)
+
+    If `F` is independent of `y`, an ODE can be solved using direct
+    integration. We can therefore obtain a reference solution with
+    :func:`~mpmath.quad`::
+
+        >>> f = lambda x: (1+x**2)/(1+x**3)
+        >>> g = odefun(lambda x, y: f(x), pi, 0)
+        >>> g(2*pi)
+        0.72128263801696
+        >>> quad(f, [pi, 2*pi])
+        0.72128263801696
+
+    **Examples of second-order ODEs**
+
+    We will solve the harmonic oscillator equation `y''(x) + y(x) = 0`.
+    To do this, we introduce the helper functions `y_0 = y, y_1 = y_0'`
+    whereby the original equation can be written as `y_1' + y_0' = 0`. Put
+    together, we get the first-order, two-dimensional vector ODE
+
+    .. math ::
+
+        \begin{cases}
+        y_0' = y_1 \\
+        y_1' = -y_0
+        \end{cases}
+
+    To get a well-defined IVP, we need two initial values. With
+    `y(0) = y_0(0) = 1` and `-y'(0) = y_1(0) = 0`, the problem will of
+    course be solved by `y(x) = y_0(x) = \cos(x)` and
+    `-y'(x) = y_1(x) = \sin(x)`. We check this::
+
+        >>> f = odefun(lambda x, y: [-y[1], y[0]], 0, [1, 0])
+        >>> for x in [0, 1, 2.5, 10]:
+        ...     nprint(f(x), 15)
+        ...     nprint([cos(x), sin(x)], 15)
+        ...     print("---")
+        ...
+        [1.0, 0.0]
+        [1.0, 0.0]
+        ---
+        [0.54030230586814, 0.841470984807897]
+        [0.54030230586814, 0.841470984807897]
+        ---
+        [-0.801143615546934, 0.598472144103957]
+        [-0.801143615546934, 0.598472144103957]
+        ---
+        [-0.839071529076452, -0.54402111088937]
+        [-0.839071529076452, -0.54402111088937]
+        ---
+
+    Note that we get both the sine and the cosine solutions
+    simultaneously.
+
+    **TODO**
+
+    * Better automatic choice of degree and step size
+    * Make determination of Taylor series convergence radius
+      more robust
+    * Allow solution for `x < x_0`
+    * Allow solution for complex `x`
+    * Test for difficult (ill-conditioned) problems
+    * Implement Runge-Kutta and other algorithms
+
+    """
+    if tol:
+        tol_prec = int(-ctx.log(tol, 2))+10
+    else:
+        tol_prec = ctx.prec+10
+    degree = degree or (3 + int(3*ctx.dps/2.))
+    workprec = ctx.prec + 40
+    try:
+        len(y0)
+        return_vector = True
+    except TypeError:
+        F_ = F
+        F = lambda x, y: [F_(x, y[0])]
+        y0 = [y0]
+        return_vector = False
+    ser, xb = ode_taylor(ctx, F, x0, y0, tol_prec, degree)
+    series_boundaries = [x0, xb]
+    series_data = [(ser, x0, xb)]
+    # We will be working with vectors of Taylor series
+    def mpolyval(ser, a):
+        return [ctx.polyval(s[::-1], a) for s in ser]
+    # Find nearest expansion point; compute if necessary
+    def get_series(x):
+        if x < x0:
+            raise ValueError
+        n = bisect(series_boundaries, x)
+        if n < len(series_boundaries):
+            return series_data[n-1]
+        while 1:
+            ser, xa, xb = series_data[-1]
+            if verbose:
+                print("Computing Taylor series for [%f, %f]" % (xa, xb))
+            y = mpolyval(ser, xb-xa)
+            xa = xb
+            ser, xb = ode_taylor(ctx, F, xb, y, tol_prec, degree)
+            series_boundaries.append(xb)
+            series_data.append((ser, xa, xb))
+            if x <= xb:
+                return series_data[-1]
+    # Evaluation function
+    def interpolant(x):
+        x = ctx.convert(x)
+        orig = ctx.prec
+        try:
+            ctx.prec = workprec
+            ser, xa, xb = get_series(x)
+            y = mpolyval(ser, x-xa)
+        finally:
+            ctx.prec = orig
+        if return_vector:
+            return [+yk for yk in y]
+        else:
+            return +y[0]
+    return interpolant
+
+ODEMethods.odefun = odefun
+
+if __name__ == "__main__":
+    import doctest
+    doctest.testmod()

lib/python3.11/site-packages/mpmath/calculus/optimization.py

@@ -0,0 +1,1102 @@
+from __future__ import print_function
+
+from copy import copy
+
+from ..libmp.backend import xrange
+
+class OptimizationMethods(object):
+    def __init__(ctx):
+        pass
+
+##############
+# 1D-SOLVERS #
+##############
+
+class Newton:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Needs starting points x0 close to the root.
+
+    Pro:
+
+    * converges fast
+    * sometimes more robust than secant with bad second starting point
+
+    Contra:
+
+    * converges slowly for multiple roots
+    * needs first derivative
+    * 2 function evaluations per iteration
+    """
+    maxsteps = 20
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if len(x0) == 1:
+            self.x0 = x0[0]
+        else:
+            raise ValueError('expected 1 starting point, got %i' % len(x0))
+        self.f = f
+        if not 'df' in kwargs:
+            def df(x):
+                return self.ctx.diff(f, x)
+        else:
+            df = kwargs['df']
+        self.df = df
+
+    def __iter__(self):
+        f = self.f
+        df = self.df
+        x0 = self.x0
+        while True:
+            x1 = x0 - f(x0) / df(x0)
+            error = abs(x1 - x0)
+            x0 = x1
+            yield (x1, error)
+
+class Secant:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Needs starting points x0 and x1 close to the root.
+    x1 defaults to x0 + 0.25.
+
+    Pro:
+
+    * converges fast
+
+    Contra:
+
+    * converges slowly for multiple roots
+    """
+    maxsteps = 30
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if len(x0) == 1:
+            self.x0 = x0[0]
+            self.x1 = self.x0 + 0.25
+        elif len(x0) == 2:
+            self.x0 = x0[0]
+            self.x1 = x0[1]
+        else:
+            raise ValueError('expected 1 or 2 starting points, got %i' % len(x0))
+        self.f = f
+
+    def __iter__(self):
+        f = self.f
+        x0 = self.x0
+        x1 = self.x1
+        f0 = f(x0)
+        while True:
+            f1 = f(x1)
+            l = x1 - x0
+            if not l:
+                break
+            s = (f1 - f0) / l
+            if not s:
+                break
+            x0, x1 = x1, x1 - f1/s
+            f0 = f1
+            yield x1, abs(l)
+
+class MNewton:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Needs starting point x0 close to the root.
+    Uses modified Newton's method that converges fast regardless of the
+    multiplicity of the root.
+
+    Pro:
+
+    * converges fast for multiple roots
+
+    Contra:
+
+    * needs first and second derivative of f
+    * 3 function evaluations per iteration
+    """
+    maxsteps = 20
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if not len(x0) == 1:
+            raise ValueError('expected 1 starting point, got %i' % len(x0))
+        self.x0 = x0[0]
+        self.f = f
+        if not 'df' in kwargs:
+            def df(x):
+                return self.ctx.diff(f, x)
+        else:
+            df = kwargs['df']
+        self.df = df
+        if not 'd2f' in kwargs:
+            def d2f(x):
+                return self.ctx.diff(df, x)
+        else:
+            d2f = kwargs['df']
+        self.d2f = d2f
+
+    def __iter__(self):
+        x = self.x0
+        f = self.f
+        df = self.df
+        d2f = self.d2f
+        while True:
+            prevx = x
+            fx = f(x)
+            if fx == 0:
+                break
+            dfx = df(x)
+            d2fx = d2f(x)
+            # x = x - F(x)/F'(x) with F(x) = f(x)/f'(x)
+            x -= fx / (dfx - fx * d2fx / dfx)
+            error = abs(x - prevx)
+            yield x, error
+
+class Halley:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Needs a starting point x0 close to the root.
+    Uses Halley's method with cubic convergence rate.
+
+    Pro:
+
+    * converges even faster the Newton's method
+    * useful when computing with *many* digits
+
+    Contra:
+
+    * needs first and second derivative of f
+    * 3 function evaluations per iteration
+    * converges slowly for multiple roots
+    """
+
+    maxsteps = 20
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if not len(x0) == 1:
+            raise ValueError('expected 1 starting point, got %i' % len(x0))
+        self.x0 = x0[0]
+        self.f = f
+        if not 'df' in kwargs:
+            def df(x):
+                return self.ctx.diff(f, x)
+        else:
+            df = kwargs['df']
+        self.df = df
+        if not 'd2f' in kwargs:
+            def d2f(x):
+                return self.ctx.diff(df, x)
+        else:
+            d2f = kwargs['df']
+        self.d2f = d2f
+
+    def __iter__(self):
+        x = self.x0
+        f = self.f
+        df = self.df
+        d2f = self.d2f
+        while True:
+            prevx = x
+            fx = f(x)
+            dfx = df(x)
+            d2fx = d2f(x)
+            x -=  2*fx*dfx / (2*dfx**2 - fx*d2fx)
+            error = abs(x - prevx)
+            yield x, error
+
+class Muller:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Needs starting points x0, x1 and x2 close to the root.
+    x1 defaults to x0 + 0.25; x2 to x1 + 0.25.
+    Uses Muller's method that converges towards complex roots.
+
+    Pro:
+
+    * converges fast (somewhat faster than secant)
+    * can find complex roots
+
+    Contra:
+
+    * converges slowly for multiple roots
+    * may have complex values for real starting points and real roots
+
+    http://en.wikipedia.org/wiki/Muller's_method
+    """
+    maxsteps = 30
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if len(x0) == 1:
+            self.x0 = x0[0]
+            self.x1 = self.x0 + 0.25
+            self.x2 = self.x1 + 0.25
+        elif len(x0) == 2:
+            self.x0 = x0[0]
+            self.x1 = x0[1]
+            self.x2 = self.x1 + 0.25
+        elif len(x0) == 3:
+            self.x0 = x0[0]
+            self.x1 = x0[1]
+            self.x2 = x0[2]
+        else:
+            raise ValueError('expected 1, 2 or 3 starting points, got %i'
+                             % len(x0))
+        self.f = f
+        self.verbose = kwargs['verbose']
+
+    def __iter__(self):
+        f = self.f
+        x0 = self.x0
+        x1 = self.x1
+        x2 = self.x2
+        fx0 = f(x0)
+        fx1 = f(x1)
+        fx2 = f(x2)
+        while True:
+            # TODO: maybe refactoring with function for divided differences
+            # calculate divided differences
+            fx2x1 = (fx1 - fx2) / (x1 - x2)
+            fx2x0 = (fx0 - fx2) / (x0 - x2)
+            fx1x0 = (fx0 - fx1) / (x0 - x1)
+            w = fx2x1 + fx2x0 - fx1x0
+            fx2x1x0 = (fx1x0 - fx2x1) / (x0 - x2)
+            if w == 0 and fx2x1x0 == 0:
+                if self.verbose:
+                    print('canceled with')
+                    print('x0 =', x0, ', x1 =', x1, 'and x2 =', x2)
+                break
+            x0 = x1
+            fx0 = fx1
+            x1 = x2
+            fx1 = fx2
+            # denominator should be as large as possible => choose sign
+            r = self.ctx.sqrt(w**2 - 4*fx2*fx2x1x0)
+            if abs(w - r) > abs(w + r):
+                r = -r
+            x2 -= 2*fx2 / (w + r)
+            fx2 = f(x2)
+            error = abs(x2 - x1)
+            yield x2, error
+
+# TODO: consider raising a ValueError when there's no sign change in a and b
+class Bisection:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Uses bisection method to find a root of f in [a, b].
+    Might fail for multiple roots (needs sign change).
+
+    Pro:
+
+    * robust and reliable
+
+    Contra:
+
+    * converges slowly
+    * needs sign change
+    """
+    maxsteps = 100
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if len(x0) != 2:
+            raise ValueError('expected interval of 2 points, got %i' % len(x0))
+        self.f = f
+        self.a = x0[0]
+        self.b = x0[1]
+
+    def __iter__(self):
+        f = self.f
+        a = self.a
+        b = self.b
+        l = b - a
+        fb = f(b)
+        while True:
+            m = self.ctx.ldexp(a + b, -1)
+            fm = f(m)
+            sign = fm * fb
+            if sign < 0:
+                a = m
+            elif sign > 0:
+                b = m
+                fb = fm
+            else:
+                yield m, self.ctx.zero
+            l /= 2
+            yield (a + b)/2, abs(l)
+
+def _getm(method):
+    """
+    Return a function to calculate m for Illinois-like methods.
+    """
+    if method == 'illinois':
+        def getm(fz, fb):
+            return 0.5
+    elif method == 'pegasus':
+        def getm(fz, fb):
+            return fb/(fb + fz)
+    elif method == 'anderson':
+        def getm(fz, fb):
+            m = 1 - fz/fb
+            if m > 0:
+                return m
+            else:
+                return 0.5
+    else:
+        raise ValueError("method '%s' not recognized" % method)
+    return getm
+
+class Illinois:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Uses Illinois method or similar to find a root of f in [a, b].
+    Might fail for multiple roots (needs sign change).
+    Combines bisect with secant (improved regula falsi).
+
+    The only difference between the methods is the scaling factor m, which is
+    used to ensure convergence (you can choose one using the 'method' keyword):
+
+    Illinois method ('illinois'):
+        m = 0.5
+
+    Pegasus method ('pegasus'):
+        m = fb/(fb + fz)
+
+    Anderson-Bjoerk method ('anderson'):
+        m = 1 - fz/fb if positive else 0.5
+
+    Pro:
+
+    * converges very fast
+
+    Contra:
+
+    * has problems with multiple roots
+    * needs sign change
+    """
+    maxsteps = 30
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if len(x0) != 2:
+            raise ValueError('expected interval of 2 points, got %i' % len(x0))
+        self.a = x0[0]
+        self.b = x0[1]
+        self.f = f
+        self.tol = kwargs['tol']
+        self.verbose = kwargs['verbose']
+        self.method = kwargs.get('method', 'illinois')
+        self.getm = _getm(self.method)
+        if self.verbose:
+            print('using %s method' % self.method)
+
+    def __iter__(self):
+        method = self.method
+        f = self.f
+        a = self.a
+        b = self.b
+        fa = f(a)
+        fb = f(b)
+        m = None
+        while True:
+            l = b - a
+            if l == 0:
+                break
+            s = (fb - fa) / l
+            z = a - fa/s
+            fz = f(z)
+            if abs(fz) < self.tol:
+                # TODO: better condition (when f is very flat)
+                if self.verbose:
+                    print('canceled with z =', z)
+                yield z, l
+                break
+            if fz * fb < 0: # root in [z, b]
+                a = b
+                fa = fb
+                b = z
+                fb = fz
+            else: # root in [a, z]
+                m = self.getm(fz, fb)
+                b = z
+                fb = fz
+                fa = m*fa # scale down to ensure convergence
+            if self.verbose and m and not method == 'illinois':
+                print('m:', m)
+            yield (a + b)/2, abs(l)
+
+def Pegasus(*args, **kwargs):
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Uses Pegasus method to find a root of f in [a, b].
+    Wrapper for illinois to use method='pegasus'.
+    """
+    kwargs['method'] = 'pegasus'
+    return Illinois(*args, **kwargs)
+
+def Anderson(*args, **kwargs):
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Uses Anderson-Bjoerk method to find a root of f in [a, b].
+    Wrapper for illinois to use method='pegasus'.
+    """
+    kwargs['method'] = 'anderson'
+    return Illinois(*args, **kwargs)
+
+# TODO: check whether it's possible to combine it with Illinois stuff
+class Ridder:
+    """
+    1d-solver generating pairs of approximative root and error.
+
+    Ridders' method to find a root of f in [a, b].
+    Is told to perform as well as Brent's method while being simpler.
+
+    Pro:
+
+    * very fast
+    * simpler than Brent's method
+
+    Contra:
+
+    * two function evaluations per step
+    * has problems with multiple roots
+    * needs sign change
+
+    http://en.wikipedia.org/wiki/Ridders'_method
+    """
+    maxsteps = 30
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        self.f = f
+        if len(x0) != 2:
+            raise ValueError('expected interval of 2 points, got %i' % len(x0))
+        self.x1 = x0[0]
+        self.x2 = x0[1]
+        self.verbose = kwargs['verbose']
+        self.tol = kwargs['tol']
+
+    def __iter__(self):
+        ctx = self.ctx
+        f = self.f
+        x1 = self.x1
+        fx1 = f(x1)
+        x2 = self.x2
+        fx2 = f(x2)
+        while True:
+            x3 = 0.5*(x1 + x2)
+            fx3 = f(x3)
+            x4 = x3 + (x3 - x1) * ctx.sign(fx1 - fx2) * fx3 / ctx.sqrt(fx3**2 - fx1*fx2)
+            fx4 = f(x4)
+            if abs(fx4) < self.tol:
+                # TODO: better condition (when f is very flat)
+                if self.verbose:
+                    print('canceled with f(x4) =', fx4)
+                yield x4, abs(x1 - x2)
+                break
+            if fx4 * fx2 < 0: # root in [x4, x2]
+                x1 = x4
+                fx1 = fx4
+            else: # root in [x1, x4]
+                x2 = x4
+                fx2 = fx4
+            error = abs(x1 - x2)
+            yield (x1 + x2)/2, error
+
+class ANewton:
+    """
+    EXPERIMENTAL 1d-solver generating pairs of approximative root and error.
+
+    Uses Newton's method modified to use Steffensens method when convergence is
+    slow. (I.e. for multiple roots.)
+    """
+    maxsteps = 20
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        if not len(x0) == 1:
+            raise ValueError('expected 1 starting point, got %i' % len(x0))
+        self.x0 = x0[0]
+        self.f = f
+        if not 'df' in kwargs:
+            def df(x):
+                return self.ctx.diff(f, x)
+        else:
+            df = kwargs['df']
+        self.df = df
+        def phi(x):
+            return x - f(x) / df(x)
+        self.phi = phi
+        self.verbose = kwargs['verbose']
+
+    def __iter__(self):
+        x0 = self.x0
+        f = self.f
+        df = self.df
+        phi = self.phi
+        error = 0
+        counter = 0
+        while True:
+            prevx = x0
+            try:
+                x0 = phi(x0)
+            except ZeroDivisionError:
+                if self.verbose:
+                    print('ZeroDivisionError: canceled with x =', x0)
+                break
+            preverror = error
+            error = abs(prevx - x0)
+            # TODO: decide not to use convergence acceleration
+            if error and abs(error - preverror) / error < 1:
+                if self.verbose:
+                    print('converging slowly')
+                counter += 1
+            if counter >= 3:
+                # accelerate convergence
+                phi = steffensen(phi)
+                counter = 0
+                if self.verbose:
+                    print('accelerating convergence')
+            yield x0, error
+
+# TODO: add Brent
+
+############################
+# MULTIDIMENSIONAL SOLVERS #
+############################
+
+def jacobian(ctx, f, x):
+    """
+    Calculate the Jacobian matrix of a function at the point x0.
+
+    This is the first derivative of a vectorial function:
+
+        f : R^m -> R^n with m >= n
+    """
+    x = ctx.matrix(x)
+    h = ctx.sqrt(ctx.eps)
+    fx = ctx.matrix(f(*x))
+    m = len(fx)
+    n = len(x)
+    J = ctx.matrix(m, n)
+    for j in xrange(n):
+        xj = x.copy()
+        xj[j] += h
+        Jj = (ctx.matrix(f(*xj)) - fx) / h
+        for i in xrange(m):
+            J[i,j] = Jj[i]
+    return J
+
+# TODO: test with user-specified jacobian matrix
+class MDNewton:
+    """
+    Find the root of a vector function numerically using Newton's method.
+
+    f is a vector function representing a nonlinear equation system.
+
+    x0 is the starting point close to the root.
+
+    J is a function returning the Jacobian matrix for a point.
+
+    Supports overdetermined systems.
+
+    Use the 'norm' keyword to specify which norm to use. Defaults to max-norm.
+    The function to calculate the Jacobian matrix can be given using the
+    keyword 'J'. Otherwise it will be calculated numerically.
+
+    Please note that this method converges only locally. Especially for high-
+    dimensional systems it is not trivial to find a good starting point being
+    close enough to the root.
+
+    It is recommended to use a faster, low-precision solver from SciPy [1] or
+    OpenOpt [2] to get an initial guess. Afterwards you can use this method for
+    root-polishing to any precision.
+
+    [1] http://scipy.org
+
+    [2] http://openopt.org/Welcome
+    """
+    maxsteps = 10
+
+    def __init__(self, ctx, f, x0, **kwargs):
+        self.ctx = ctx
+        self.f = f
+        if isinstance(x0, (tuple, list)):
+            x0 = ctx.matrix(x0)
+        assert x0.cols == 1, 'need a vector'
+        self.x0 = x0
+        if 'J' in kwargs:
+            self.J = kwargs['J']
+        else:
+            def J(*x):
+                return ctx.jacobian(f, x)
+            self.J = J
+        self.norm = kwargs['norm']
+        self.verbose = kwargs['verbose']
+
+    def __iter__(self):
+        f = self.f
+        x0 = self.x0
+        norm = self.norm
+        J = self.J
+        fx = self.ctx.matrix(f(*x0))
+        fxnorm = norm(fx)
+        cancel = False
+        while not cancel:
+            # get direction of descent
+            fxn = -fx
+            Jx = J(*x0)
+            s = self.ctx.lu_solve(Jx, fxn)
+            if self.verbose:
+                print('Jx:')
+                print(Jx)
+                print('s:', s)
+            # damping step size TODO: better strategy (hard task)
+            l = self.ctx.one
+            x1 = x0 + s
+            while True:
+                if x1 == x0:
+                    if self.verbose:
+                        print("canceled, won't get more excact")
+                    cancel = True
+                    break
+                fx = self.ctx.matrix(f(*x1))
+                newnorm = norm(fx)
+                if newnorm < fxnorm:
+                    # new x accepted
+                    fxnorm = newnorm
+                    x0 = x1
+                    break
+                l /= 2
+                x1 = x0 + l*s
+            yield (x0, fxnorm)
+
+#############
+# UTILITIES #
+#############
+
+str2solver = {'newton':Newton, 'secant':Secant, 'mnewton':MNewton,
+              'halley':Halley, 'muller':Muller, 'bisect':Bisection,
+              'illinois':Illinois, 'pegasus':Pegasus, 'anderson':Anderson,
+              'ridder':Ridder, 'anewton':ANewton, 'mdnewton':MDNewton}
+
+def findroot(ctx, f, x0, solver='secant', tol=None, verbose=False, verify=True, **kwargs):
+    r"""
+    Find an approximate solution to `f(x) = 0`, using *x0* as starting point or
+    interval for *x*.
+
+    Multidimensional overdetermined systems are supported.
+    You can specify them using a function or a list of functions.
+
+    Mathematically speaking, this function returns `x` such that
+    `|f(x)|^2 \leq \mathrm{tol}` is true within the current working precision.
+    If the computed value does not meet this criterion, an exception is raised.
+    This exception can be disabled with *verify=False*.
+
+    For interval arithmetic (``iv.findroot()``), please note that
+    the returned interval ``x`` is not guaranteed to contain `f(x)=0`!
+    It is only some `x` for which `|f(x)|^2 \leq \mathrm{tol}` certainly holds
+    regardless of numerical error. This may be improved in the future.
+
+    **Arguments**
+
+    *f*
+        one dimensional function
+    *x0*
+        starting point, several starting points or interval (depends on solver)
+    *tol*
+        the returned solution has an error smaller than this
+    *verbose*
+        print additional information for each iteration if true
+    *verify*
+        verify the solution and raise a ValueError if `|f(x)|^2 > \mathrm{tol}`
+    *solver*
+        a generator for *f* and *x0* returning approximative solution and error
+    *maxsteps*
+        after how many steps the solver will cancel
+    *df*
+        first derivative of *f* (used by some solvers)
+    *d2f*
+        second derivative of *f* (used by some solvers)
+    *multidimensional*
+        force multidimensional solving
+    *J*
+        Jacobian matrix of *f* (used by multidimensional solvers)
+    *norm*
+        used vector norm (used by multidimensional solvers)
+
+    solver has to be callable with ``(f, x0, **kwargs)`` and return an generator
+    yielding pairs of approximative solution and estimated error (which is
+    expected to be positive).
+    You can use the following string aliases:
+    'secant', 'mnewton', 'halley', 'muller', 'illinois', 'pegasus', 'anderson',
+    'ridder', 'anewton', 'bisect'
+
+    See mpmath.calculus.optimization for their documentation.
+
+    **Examples**
+
+    The function :func:`~mpmath.findroot` locates a root of a given function using the
+    secant method by default. A simple example use of the secant method is to
+    compute `\pi` as the root of `\sin x` closest to `x_0 = 3`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 30; mp.pretty = True
+        >>> findroot(sin, 3)
+        3.14159265358979323846264338328
+
+    The secant method can be used to find complex roots of analytic functions,
+    although it must in that case generally be given a nonreal starting value
+    (or else it will never leave the real line)::
+
+        >>> mp.dps = 15
+        >>> findroot(lambda x: x**3 + 2*x + 1, j)
+        (0.226698825758202 + 1.46771150871022j)
+
+    A nice application is to compute nontrivial roots of the Riemann zeta
+    function with many digits (good initial values are needed for convergence)::
+
+        >>> mp.dps = 30
+        >>> findroot(zeta, 0.5+14j)
+        (0.5 + 14.1347251417346937904572519836j)
+
+    The secant method can also be used as an optimization algorithm, by passing
+    it a derivative of a function. The following example locates the positive
+    minimum of the gamma function::
+
+        >>> mp.dps = 20
+        >>> findroot(lambda x: diff(gamma, x), 1)
+        1.4616321449683623413
+
+    Finally, a useful application is to compute inverse functions, such as the
+    Lambert W function which is the inverse of `w e^w`, given the first
+    term of the solution's asymptotic expansion as the initial value. In basic
+    cases, this gives identical results to mpmath's built-in ``lambertw``
+    function::
+
+        >>> def lambert(x):
+        ...     return findroot(lambda w: w*exp(w) - x, log(1+x))
+        ...
+        >>> mp.dps = 15
+        >>> lambert(1); lambertw(1)
+        0.567143290409784
+        0.567143290409784
+        >>> lambert(1000); lambert(1000)
+        5.2496028524016
+        5.2496028524016
+
+    Multidimensional functions are also supported::
+
+        >>> f = [lambda x1, x2: x1**2 + x2,
+        ...      lambda x1, x2: 5*x1**2 - 3*x1 + 2*x2 - 3]
+        >>> findroot(f, (0, 0))
+        [-0.618033988749895]
+        [-0.381966011250105]
+        >>> findroot(f, (10, 10))
+        [ 1.61803398874989]
+        [-2.61803398874989]
+
+    You can verify this by solving the system manually.
+
+    Please note that the following (more general) syntax also works::
+
+        >>> def f(x1, x2):
+        ...     return x1**2 + x2, 5*x1**2 - 3*x1 + 2*x2 - 3
+        ...
+        >>> findroot(f, (0, 0))
+        [-0.618033988749895]
+        [-0.381966011250105]
+
+
+    **Multiple roots**
+
+    For multiple roots all methods of the Newtonian family (including secant)
+    converge slowly. Consider this example::
+
+        >>> f = lambda x: (x - 1)**99
+        >>> findroot(f, 0.9, verify=False)
+        0.918073542444929
+
+    Even for a very close starting point the secant method converges very
+    slowly. Use ``verbose=True`` to illustrate this.
+
+    It is possible to modify Newton's method to make it converge regardless of
+    the root's multiplicity::
+
+        >>> findroot(f, -10, solver='mnewton')
+        1.0
+
+    This variant uses the first and second derivative of the function, which is
+    not very efficient.
+
+    Alternatively you can use an experimental Newtonian solver that keeps track
+    of the speed of convergence and accelerates it using Steffensen's method if
+    necessary::
+
+        >>> findroot(f, -10, solver='anewton', verbose=True)
+        x:     -9.88888888888888888889
+        error: 0.111111111111111111111
+        converging slowly
+        x:     -9.77890011223344556678
+        error: 0.10998877665544332211
+        converging slowly
+        x:     -9.67002233332199662166
+        error: 0.108877778911448945119
+        converging slowly
+        accelerating convergence
+        x:     -9.5622443299551077669
+        error: 0.107778003366888854764
+        converging slowly
+        x:     0.99999999999999999214
+        error: 10.562244329955107759
+        x:     1.0
+        error: 7.8598304758094664213e-18
+        ZeroDivisionError: canceled with x = 1.0
+        1.0
+
+    **Complex roots**
+
+    For complex roots it's recommended to use Muller's method as it converges
+    even for real starting points very fast::
+
+        >>> findroot(lambda x: x**4 + x + 1, (0, 1, 2), solver='muller')
+        (0.727136084491197 + 0.934099289460529j)
+
+
+    **Intersection methods**
+
+    When you need to find a root in a known interval, it's highly recommended to
+    use an intersection-based solver like ``'anderson'`` or ``'ridder'``.
+    Usually they converge faster and more reliable. They have however problems
+    with multiple roots and usually need a sign change to find a root::
+
+        >>> findroot(lambda x: x**3, (-1, 1), solver='anderson')
+        0.0
+
+    Be careful with symmetric functions::
+
+        >>> findroot(lambda x: x**2, (-1, 1), solver='anderson') #doctest:+ELLIPSIS
+        Traceback (most recent call last):
+          ...
+        ZeroDivisionError
+
+    It fails even for better starting points, because there is no sign change::
+
+        >>> findroot(lambda x: x**2, (-1, .5), solver='anderson')
+        Traceback (most recent call last):
+          ...
+        ValueError: Could not find root within given tolerance. (1.0 > 2.16840434497100886801e-19)
+        Try another starting point or tweak arguments.
+
+    """
+    prec = ctx.prec
+    try:
+        ctx.prec += 20
+
+        # initialize arguments
+        if tol is None:
+            tol = ctx.eps * 2**10
+
+        kwargs['verbose'] = kwargs.get('verbose', verbose)
+
+        if 'd1f' in kwargs:
+            kwargs['df'] = kwargs['d1f']
+
+        kwargs['tol'] = tol
+        if isinstance(x0, (list, tuple)):
+            x0 = [ctx.convert(x) for x in x0]
+        else:
+            x0 = [ctx.convert(x0)]
+
+        if isinstance(solver, str):
+            try:
+                solver = str2solver[solver]
+            except KeyError:
+                raise ValueError('could not recognize solver')
+
+        # accept list of functions
+        if isinstance(f, (list, tuple)):
+            f2 = copy(f)
+            def tmp(*args):
+                return [fn(*args) for fn in f2]
+            f = tmp
+
+        # detect multidimensional functions
+        try:
+            fx = f(*x0)
+            multidimensional = isinstance(fx, (list, tuple, ctx.matrix))
+        except TypeError:
+            fx = f(x0[0])
+            multidimensional = False
+        if 'multidimensional' in kwargs:
+            multidimensional = kwargs['multidimensional']
+        if multidimensional:
+            # only one multidimensional solver available at the moment
+            solver = MDNewton
+            if not 'norm' in kwargs:
+                norm = lambda x: ctx.norm(x, 'inf')
+                kwargs['norm'] = norm
+            else:
+                norm = kwargs['norm']
+        else:
+            norm = abs
+
+        # happily return starting point if it's a root
+        if norm(fx) == 0:
+            if multidimensional:
+                return ctx.matrix(x0)
+            else:
+                return x0[0]
+
+        # use solver
+        iterations = solver(ctx, f, x0, **kwargs)
+        if 'maxsteps' in kwargs:
+            maxsteps = kwargs['maxsteps']
+        else:
+            maxsteps = iterations.maxsteps
+        i = 0
+        for x, error in iterations:
+            if verbose:
+                print('x:    ', x)
+                print('error:', error)
+            i += 1
+            if error < tol * max(1, norm(x)) or i >= maxsteps:
+                break
+        else:
+            if not i:
+                raise ValueError('Could not find root using the given solver.\n'
+                                 'Try another starting point or tweak arguments.')
+        if not isinstance(x, (list, tuple, ctx.matrix)):
+            xl = [x]
+        else:
+            xl = x
+        if verify and norm(f(*xl))**2 > tol: # TODO: better condition?
+            raise ValueError('Could not find root within given tolerance. '
+                             '(%s > %s)\n'
+                             'Try another starting point or tweak arguments.'
+                             % (norm(f(*xl))**2, tol))
+        return x
+    finally:
+        ctx.prec = prec
+
+
+def multiplicity(ctx, f, root, tol=None, maxsteps=10, **kwargs):
+    """
+    Return the multiplicity of a given root of f.
+
+    Internally, numerical derivatives are used. This might be inefficient for
+    higher order derviatives. Due to this, ``multiplicity`` cancels after
+    evaluating 10 derivatives by default. You can be specify the n-th derivative
+    using the dnf keyword.
+
+    >>> from mpmath import *
+    >>> multiplicity(lambda x: sin(x) - 1, pi/2)
+    2
+
+    """
+    if tol is None:
+        tol = ctx.eps ** 0.8
+    kwargs['d0f'] = f
+    for i in xrange(maxsteps):
+        dfstr = 'd' + str(i) + 'f'
+        if dfstr in kwargs:
+            df = kwargs[dfstr]
+        else:
+            df = lambda x: ctx.diff(f, x, i)
+        if not abs(df(root)) < tol:
+            break
+    return i
+
+def steffensen(f):
+    """
+    linear convergent function -> quadratic convergent function
+
+    Steffensen's method for quadratic convergence of a linear converging
+    sequence.
+    Don not use it for higher rates of convergence.
+    It may even work for divergent sequences.
+
+    Definition:
+    F(x) = (x*f(f(x)) - f(x)**2) / (f(f(x)) - 2*f(x) + x)
+
+    Example
+    .......
+
+    You can use Steffensen's method to accelerate a fixpoint iteration of linear
+    (or less) convergence.
+
+    x* is a fixpoint of the iteration x_{k+1} = phi(x_k) if x* = phi(x*). For
+    phi(x) = x**2 there are two fixpoints: 0 and 1.
+
+    Let's try Steffensen's method:
+
+    >>> f = lambda x: x**2
+    >>> from mpmath.calculus.optimization import steffensen
+    >>> F = steffensen(f)
+    >>> for x in [0.5, 0.9, 2.0]:
+    ...     fx = Fx = x
+    ...     for i in xrange(9):
+    ...         try:
+    ...             fx = f(fx)
+    ...         except OverflowError:
+    ...             pass
+    ...         try:
+    ...             Fx = F(Fx)
+    ...         except ZeroDivisionError:
+    ...             pass
+    ...         print('%20g  %20g' % (fx, Fx))
+                    0.25                  -0.5
+                  0.0625                   0.1
+              0.00390625            -0.0011236
+             1.52588e-05           1.41691e-09
+             2.32831e-10          -2.84465e-27
+             5.42101e-20           2.30189e-80
+             2.93874e-39          -1.2197e-239
+             8.63617e-78                     0
+            7.45834e-155                     0
+                    0.81               1.02676
+                  0.6561               1.00134
+                0.430467                     1
+                0.185302                     1
+               0.0343368                     1
+              0.00117902                     1
+             1.39008e-06                     1
+             1.93233e-12                     1
+             3.73392e-24                     1
+                       4                   1.6
+                      16                1.2962
+                     256               1.10194
+                   65536               1.01659
+             4.29497e+09               1.00053
+             1.84467e+19                     1
+             3.40282e+38                     1
+             1.15792e+77                     1
+            1.34078e+154                     1
+
+    Unmodified, the iteration converges only towards 0. Modified it converges
+    not only much faster, it converges even to the repelling fixpoint 1.
+    """
+    def F(x):
+        fx = f(x)
+        ffx = f(fx)
+        return (x*ffx - fx**2) / (ffx - 2*fx + x)
+    return F
+
+OptimizationMethods.jacobian = jacobian
+OptimizationMethods.findroot = findroot
+OptimizationMethods.multiplicity = multiplicity
+
+if __name__ == '__main__':
+    import doctest
+    doctest.testmod()

lib/python3.11/site-packages/mpmath/calculus/polynomials.py

@@ -0,0 +1,213 @@
+from ..libmp.backend import xrange
+from .calculus import defun
+
+#----------------------------------------------------------------------------#
+#                                Polynomials                                 #
+#----------------------------------------------------------------------------#
+
+# XXX: extra precision
+@defun
+def polyval(ctx, coeffs, x, derivative=False):
+    r"""
+    Given coefficients `[c_n, \ldots, c_2, c_1, c_0]` and a number `x`,
+    :func:`~mpmath.polyval` evaluates the polynomial
+
+    .. math ::
+
+        P(x) = c_n x^n + \ldots + c_2 x^2 + c_1 x + c_0.
+
+    If *derivative=True* is set, :func:`~mpmath.polyval` simultaneously
+    evaluates `P(x)` with the derivative, `P'(x)`, and returns the
+    tuple `(P(x), P'(x))`.
+
+        >>> from mpmath import *
+        >>> mp.pretty = True
+        >>> polyval([3, 0, 2], 0.5)
+        2.75
+        >>> polyval([3, 0, 2], 0.5, derivative=True)
+        (2.75, 3.0)
+
+    The coefficients and the evaluation point may be any combination
+    of real or complex numbers.
+    """
+    if not coeffs:
+        return ctx.zero
+    p = ctx.convert(coeffs[0])
+    q = ctx.zero
+    for c in coeffs[1:]:
+        if derivative:
+            q = p + x*q
+        p = c + x*p
+    if derivative:
+        return p, q
+    else:
+        return p
+
+@defun
+def polyroots(ctx, coeffs, maxsteps=50, cleanup=True, extraprec=10,
+        error=False, roots_init=None):
+    """
+    Computes all roots (real or complex) of a given polynomial.
+
+    The roots are returned as a sorted list, where real roots appear first
+    followed by complex conjugate roots as adjacent elements. The polynomial
+    should be given as a list of coefficients, in the format used by
+    :func:`~mpmath.polyval`. The leading coefficient must be nonzero.
+
+    With *error=True*, :func:`~mpmath.polyroots` returns a tuple *(roots, err)*
+    where *err* is an estimate of the maximum error among the computed roots.
+
+    **Examples**
+
+    Finding the three real roots of `x^3 - x^2 - 14x + 24`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 15; mp.pretty = True
+        >>> nprint(polyroots([1,-1,-14,24]), 4)
+        [-4.0, 2.0, 3.0]
+
+    Finding the two complex conjugate roots of `4x^2 + 3x + 2`, with an
+    error estimate::
+
+        >>> roots, err = polyroots([4,3,2], error=True)
+        >>> for r in roots:
+        ...     print(r)
+        ...
+        (-0.375 + 0.59947894041409j)
+        (-0.375 - 0.59947894041409j)
+        >>>
+        >>> err
+        2.22044604925031e-16
+        >>>
+        >>> polyval([4,3,2], roots[0])
+        (2.22044604925031e-16 + 0.0j)
+        >>> polyval([4,3,2], roots[1])
+        (2.22044604925031e-16 + 0.0j)
+
+    The following example computes all the 5th roots of unity; that is,
+    the roots of `x^5 - 1`::
+
+        >>> mp.dps = 20
+        >>> for r in polyroots([1, 0, 0, 0, 0, -1]):
+        ...     print(r)
+        ...
+        1.0
+        (-0.8090169943749474241 + 0.58778525229247312917j)
+        (-0.8090169943749474241 - 0.58778525229247312917j)
+        (0.3090169943749474241 + 0.95105651629515357212j)
+        (0.3090169943749474241 - 0.95105651629515357212j)
+
+    **Precision and conditioning**
+
+    The roots are computed to the current working precision accuracy. If this
+    accuracy cannot be achieved in ``maxsteps`` steps, then a
+    ``NoConvergence`` exception is raised. The algorithm internally is using
+    the current working precision extended by ``extraprec``. If
+    ``NoConvergence`` was raised, that is caused either by not having enough
+    extra precision to achieve convergence (in which case increasing
+    ``extraprec`` should fix the problem) or too low ``maxsteps`` (in which
+    case increasing ``maxsteps`` should fix the problem), or a combination of
+    both.
+
+    The user should always do a convergence study with regards to
+    ``extraprec`` to ensure accurate results. It is possible to get
+    convergence to a wrong answer with too low ``extraprec``.
+
+    Provided there are no repeated roots, :func:`~mpmath.polyroots` can
+    typically compute all roots of an arbitrary polynomial to high precision::
+
+        >>> mp.dps = 60
+        >>> for r in polyroots([1, 0, -10, 0, 1]):
+        ...     print(r)
+        ...
+        -3.14626436994197234232913506571557044551247712918732870123249
+        -0.317837245195782244725757617296174288373133378433432554879127
+        0.317837245195782244725757617296174288373133378433432554879127
+        3.14626436994197234232913506571557044551247712918732870123249
+        >>>
+        >>> sqrt(3) + sqrt(2)
+        3.14626436994197234232913506571557044551247712918732870123249
+        >>> sqrt(3) - sqrt(2)
+        0.317837245195782244725757617296174288373133378433432554879127
+
+    **Algorithm**
+
+    :func:`~mpmath.polyroots` implements the Durand-Kerner method [1], which
+    uses complex arithmetic to locate all roots simultaneously.
+    The Durand-Kerner method can be viewed as approximately performing
+    simultaneous Newton iteration for all the roots. In particular,
+    the convergence to simple roots is quadratic, just like Newton's
+    method.
+
+    Although all roots are internally calculated using complex arithmetic, any
+    root found to have an imaginary part smaller than the estimated numerical
+    error is truncated to a real number (small real parts are also chopped).
+    Real roots are placed first in the returned list, sorted by value. The
+    remaining complex roots are sorted by their real parts so that conjugate
+    roots end up next to each other.
+
+    **References**
+
+    1. http://en.wikipedia.org/wiki/Durand-Kerner_method
+
+    """
+    if len(coeffs) <= 1:
+        if not coeffs or not coeffs[0]:
+            raise ValueError("Input to polyroots must not be the zero polynomial")
+        # Constant polynomial with no roots
+        return []
+
+    orig = ctx.prec
+    tol = +ctx.eps
+    with ctx.extraprec(extraprec):
+        deg = len(coeffs) - 1
+        # Must be monic
+        lead = ctx.convert(coeffs[0])
+        if lead == 1:
+            coeffs = [ctx.convert(c) for c in coeffs]
+        else:
+            coeffs = [c/lead for c in coeffs]
+        f = lambda x: ctx.polyval(coeffs, x)
+        if roots_init is None:
+            roots = [ctx.mpc((0.4+0.9j)**n) for n in xrange(deg)]
+        else:
+            roots = [None]*deg;
+            deg_init = min(deg, len(roots_init))
+            roots[:deg_init] = list(roots_init[:deg_init])
+            roots[deg_init:] = [ctx.mpc((0.4+0.9j)**n) for n
+                                in xrange(deg_init,deg)]
+        err = [ctx.one for n in xrange(deg)]
+        # Durand-Kerner iteration until convergence
+        for step in xrange(maxsteps):
+            if abs(max(err)) < tol:
+                break
+            for i in xrange(deg):
+                p = roots[i]
+                x = f(p)
+                for j in range(deg):
+                    if i != j:
+                        try:
+                            x /= (p-roots[j])
+                        except ZeroDivisionError:
+                            continue
+                roots[i] = p - x
+                err[i] = abs(x)
+        if abs(max(err)) >= tol:
+            raise ctx.NoConvergence("Didn't converge in maxsteps=%d steps." \
+                    % maxsteps)
+        # Remove small real or imaginary parts
+        if cleanup:
+            for i in xrange(deg):
+                if abs(roots[i]) < tol:
+                    roots[i] = ctx.zero
+                elif abs(ctx._im(roots[i])) < tol:
+                    roots[i] = roots[i].real
+                elif abs(ctx._re(roots[i])) < tol:
+                    roots[i] = roots[i].imag * 1j
+        roots.sort(key=lambda x: (abs(ctx._im(x)), ctx._re(x)))
+    if error:
+        err = max(err)
+        err = max(err, ctx.ldexp(1, -orig+1))
+        return [+r for r in roots], +err
+    else:
+        return [+r for r in roots]

lib/python3.11/site-packages/mpmath/calculus/quadrature.py

@@ -0,0 +1,1115 @@
+import math
+
+from ..libmp.backend import xrange
+
+class QuadratureRule(object):
+    """
+    Quadrature rules are implemented using this class, in order to
+    simplify the code and provide a common infrastructure
+    for tasks such as error estimation and node caching.
+
+    You can implement a custom quadrature rule by subclassing
+    :class:`QuadratureRule` and implementing the appropriate
+    methods. The subclass can then be used by :func:`~mpmath.quad` by
+    passing it as the *method* argument.
+
+    :class:`QuadratureRule` instances are supposed to be singletons.
+    :class:`QuadratureRule` therefore implements instance caching
+    in :func:`~mpmath.__new__`.
+    """
+
+    def __init__(self, ctx):
+        self.ctx = ctx
+        self.standard_cache = {}
+        self.transformed_cache = {}
+        self.interval_count = {}
+
+    def clear(self):
+        """
+        Delete cached node data.
+        """
+        self.standard_cache = {}
+        self.transformed_cache = {}
+        self.interval_count = {}
+
+    def calc_nodes(self, degree, prec, verbose=False):
+        r"""
+        Compute nodes for the standard interval `[-1, 1]`. Subclasses
+        should probably implement only this method, and use
+        :func:`~mpmath.get_nodes` method to retrieve the nodes.
+        """
+        raise NotImplementedError
+
+    def get_nodes(self, a, b, degree, prec, verbose=False):
+        """
+        Return nodes for given interval, degree and precision. The
+        nodes are retrieved from a cache if already computed;
+        otherwise they are computed by calling :func:`~mpmath.calc_nodes`
+        and are then cached.
+
+        Subclasses should probably not implement this method,
+        but just implement :func:`~mpmath.calc_nodes` for the actual
+        node computation.
+        """
+        key = (a, b, degree, prec)
+        if key in self.transformed_cache:
+            return self.transformed_cache[key]
+        orig = self.ctx.prec
+        try:
+            self.ctx.prec = prec+20
+            # Get nodes on standard interval
+            if (degree, prec) in self.standard_cache:
+                nodes = self.standard_cache[degree, prec]
+            else:
+                nodes = self.calc_nodes(degree, prec, verbose)
+                self.standard_cache[degree, prec] = nodes
+            # Transform to general interval
+            nodes = self.transform_nodes(nodes, a, b, verbose)
+            if key in self.interval_count:
+                self.transformed_cache[key] = nodes
+            else:
+                self.interval_count[key] = True
+        finally:
+            self.ctx.prec = orig
+        return nodes
+
+    def transform_nodes(self, nodes, a, b, verbose=False):
+        r"""
+        Rescale standardized nodes (for `[-1, 1]`) to a general
+        interval `[a, b]`. For a finite interval, a simple linear
+        change of variables is used. Otherwise, the following
+        transformations are used:
+
+        .. math ::
+
+            \lbrack a, \infty \rbrack : t = \frac{1}{x} + (a-1)
+
+            \lbrack -\infty, b \rbrack : t = (b+1) - \frac{1}{x}
+
+            \lbrack -\infty, \infty \rbrack : t = \frac{x}{\sqrt{1-x^2}}
+
+        """
+        ctx = self.ctx
+        a = ctx.convert(a)
+        b = ctx.convert(b)
+        one = ctx.one
+        if (a, b) == (-one, one):
+            return nodes
+        half = ctx.mpf(0.5)
+        new_nodes = []
+        if ctx.isinf(a) or ctx.isinf(b):
+            if (a, b) == (ctx.ninf, ctx.inf):
+                p05 = -half
+                for x, w in nodes:
+                    x2 = x*x
+                    px1 = one-x2
+                    spx1 = px1**p05
+                    x = x*spx1
+                    w *= spx1/px1
+                    new_nodes.append((x, w))
+            elif a == ctx.ninf:
+                b1 = b+1
+                for x, w in nodes:
+                    u = 2/(x+one)
+                    x = b1-u
+                    w *= half*u**2
+                    new_nodes.append((x, w))
+            elif b == ctx.inf:
+                a1 = a-1
+                for x, w in nodes:
+                    u = 2/(x+one)
+                    x = a1+u
+                    w *= half*u**2
+                    new_nodes.append((x, w))
+            elif a == ctx.inf or b == ctx.ninf:
+                return [(x,-w) for (x,w) in self.transform_nodes(nodes, b, a, verbose)]
+            else:
+                raise NotImplementedError
+        else:
+            # Simple linear change of variables
+            C = (b-a)/2
+            D = (b+a)/2
+            for x, w in nodes:
+                new_nodes.append((D+C*x, C*w))
+        return new_nodes
+
+    def guess_degree(self, prec):
+        """
+        Given a desired precision `p` in bits, estimate the degree `m`
+        of the quadrature required to accomplish full accuracy for
+        typical integrals. By default, :func:`~mpmath.quad` will perform up
+        to `m` iterations. The value of `m` should be a slight
+        overestimate, so that "slightly bad" integrals can be dealt
+        with automatically using a few extra iterations. On the
+        other hand, it should not be too big, so :func:`~mpmath.quad` can
+        quit within a reasonable amount of time when it is given
+        an "unsolvable" integral.
+
+        The default formula used by :func:`~mpmath.guess_degree` is tuned
+        for both :class:`TanhSinh` and :class:`GaussLegendre`.
+        The output is roughly as follows:
+
+            +---------+---------+
+            | `p`     | `m`     |
+            +=========+=========+
+            | 50      | 6       |
+            +---------+---------+
+            | 100     | 7       |
+            +---------+---------+
+            | 500     | 10      |
+            +---------+---------+
+            | 3000    | 12      |
+            +---------+---------+
+
+        This formula is based purely on a limited amount of
+        experimentation and will sometimes be wrong.
+        """
+        # Expected degree
+        # XXX: use mag
+        g = int(4 + max(0, self.ctx.log(prec/30.0, 2)))
+        # Reasonable "worst case"
+        g += 2
+        return g
+
+    def estimate_error(self, results, prec, epsilon):
+        r"""
+        Given results from integrations `[I_1, I_2, \ldots, I_k]` done
+        with a quadrature of rule of degree `1, 2, \ldots, k`, estimate
+        the error of `I_k`.
+
+        For `k = 2`, we estimate  `|I_{\infty}-I_2|` as `|I_2-I_1|`.
+
+        For `k > 2`, we extrapolate `|I_{\infty}-I_k| \approx |I_{k+1}-I_k|`
+        from `|I_k-I_{k-1}|` and `|I_k-I_{k-2}|` under the assumption
+        that each degree increment roughly doubles the accuracy of
+        the quadrature rule (this is true for both :class:`TanhSinh`
+        and :class:`GaussLegendre`). The extrapolation formula is given
+        by Borwein, Bailey & Girgensohn. Although not very conservative,
+        this method seems to be very robust in practice.
+        """
+        if len(results) == 2:
+            return abs(results[0]-results[1])
+        try:
+            if results[-1] == results[-2] == results[-3]:
+                return self.ctx.zero
+            D1 = self.ctx.log(abs(results[-1]-results[-2]), 10)
+            D2 = self.ctx.log(abs(results[-1]-results[-3]), 10)
+        except ValueError:
+            return epsilon
+        D3 = -prec
+        D4 = min(0, max(D1**2/D2, 2*D1, D3))
+        return self.ctx.mpf(10) ** int(D4)
+
+    def summation(self, f, points, prec, epsilon, max_degree, verbose=False):
+        """
+        Main integration function. Computes the 1D integral over
+        the interval specified by *points*. For each subinterval,
+        performs quadrature of degree from 1 up to *max_degree*
+        until :func:`~mpmath.estimate_error` signals convergence.
+
+        :func:`~mpmath.summation` transforms each subintegration to
+        the standard interval and then calls :func:`~mpmath.sum_next`.
+        """
+        ctx = self.ctx
+        I = total_err = ctx.zero
+        for i in xrange(len(points)-1):
+            a, b = points[i], points[i+1]
+            if a == b:
+                continue
+            # XXX: we could use a single variable transformation,
+            # but this is not good in practice. We get better accuracy
+            # by having 0 as an endpoint.
+            if (a, b) == (ctx.ninf, ctx.inf):
+                _f = f
+                f = lambda x: _f(-x) + _f(x)
+                a, b = (ctx.zero, ctx.inf)
+            results = []
+            err = ctx.zero
+            for degree in xrange(1, max_degree+1):
+                nodes = self.get_nodes(a, b, degree, prec, verbose)
+                if verbose:
+                    print("Integrating from %s to %s (degree %s of %s)" % \
+                        (ctx.nstr(a), ctx.nstr(b), degree, max_degree))
+                result = self.sum_next(f, nodes, degree, prec, results, verbose)
+                results.append(result)
+                if degree > 1:
+                    err = self.estimate_error(results, prec, epsilon)
+                    if verbose:
+                        print("Estimated error:", ctx.nstr(err), " epsilon:", ctx.nstr(epsilon), " result: ", ctx.nstr(result))
+                    if err <= epsilon:
+                        break
+            I += results[-1]
+            total_err += err
+        if total_err > epsilon:
+            if verbose:
+                print("Failed to reach full accuracy. Estimated error:", ctx.nstr(total_err))
+        return I, total_err
+
+    def sum_next(self, f, nodes, degree, prec, previous, verbose=False):
+        r"""
+        Evaluates the step sum `\sum w_k f(x_k)` where the *nodes* list
+        contains the `(w_k, x_k)` pairs.
+
+        :func:`~mpmath.summation` will supply the list *results* of
+        values computed by :func:`~mpmath.sum_next` at previous degrees, in
+        case the quadrature rule is able to reuse them.
+        """
+        return self.ctx.fdot((w, f(x)) for (x,w) in nodes)
+
+
+class TanhSinh(QuadratureRule):
+    r"""
+    This class implements "tanh-sinh" or "doubly exponential"
+    quadrature. This quadrature rule is based on the Euler-Maclaurin
+    integral formula. By performing a change of variables involving
+    nested exponentials / hyperbolic functions (hence the name), the
+    derivatives at the endpoints vanish rapidly. Since the error term
+    in the Euler-Maclaurin formula depends on the derivatives at the
+    endpoints, a simple step sum becomes extremely accurate. In
+    practice, this means that doubling the number of evaluation
+    points roughly doubles the number of accurate digits.
+
+    Comparison to Gauss-Legendre:
+      * Initial computation of nodes is usually faster
+      * Handles endpoint singularities better
+      * Handles infinite integration intervals better
+      * Is slower for smooth integrands once nodes have been computed
+
+    The implementation of the tanh-sinh algorithm is based on the
+    description given in Borwein, Bailey & Girgensohn, "Experimentation
+    in Mathematics - Computational Paths to Discovery", A K Peters,
+    2003, pages 312-313. In the present implementation, a few
+    improvements have been made:
+
+      * A more efficient scheme is used to compute nodes (exploiting
+        recurrence for the exponential function)
+      * The nodes are computed successively instead of all at once
+
+    **References**
+
+    * [Bailey]_
+    * http://users.cs.dal.ca/~jborwein/tanh-sinh.pdf
+
+    """
+
+    def sum_next(self, f, nodes, degree, prec, previous, verbose=False):
+        """
+        Step sum for tanh-sinh quadrature of degree `m`. We exploit the
+        fact that half of the abscissas at degree `m` are precisely the
+        abscissas from degree `m-1`. Thus reusing the result from
+        the previous level allows a 2x speedup.
+        """
+        h = self.ctx.mpf(2)**(-degree)
+        # Abscissas overlap, so reusing saves half of the time
+        if previous:
+            S = previous[-1]/(h*2)
+        else:
+            S = self.ctx.zero
+        S += self.ctx.fdot((w,f(x)) for (x,w) in nodes)
+        return h*S
+
+    def calc_nodes(self, degree, prec, verbose=False):
+        r"""
+        The abscissas and weights for tanh-sinh quadrature of degree
+        `m` are given by
+
+        .. math::
+
+            x_k = \tanh(\pi/2 \sinh(t_k))
+
+            w_k = \pi/2 \cosh(t_k) / \cosh(\pi/2 \sinh(t_k))^2
+
+        where `t_k = t_0 + hk` for a step length `h \sim 2^{-m}`. The
+        list of nodes is actually infinite, but the weights die off so
+        rapidly that only a few are needed.
+        """
+        ctx = self.ctx
+        nodes = []
+
+        extra = 20
+        ctx.prec += extra
+        tol = ctx.ldexp(1, -prec-10)
+        pi4 = ctx.pi/4
+
+        # For simplicity, we work in steps h = 1/2^n, with the first point
+        # offset so that we can reuse the sum from the previous degree
+
+        # We define degree 1 to include the "degree 0" steps, including
+        # the point x = 0. (It doesn't work well otherwise; not sure why.)
+        t0 = ctx.ldexp(1, -degree)
+        if degree == 1:
+            #nodes.append((mpf(0), pi4))
+            #nodes.append((-mpf(0), pi4))
+            nodes.append((ctx.zero, ctx.pi/2))
+            h = t0
+        else:
+            h = t0*2
+
+        # Since h is fixed, we can compute the next exponential
+        # by simply multiplying by exp(h)
+        expt0 = ctx.exp(t0)
+        a = pi4 * expt0
+        b = pi4 / expt0
+        udelta = ctx.exp(h)
+        urdelta = 1/udelta
+
+        for k in xrange(0, 20*2**degree+1):
+            # Reference implementation:
+            # t = t0 + k*h
+            # x = tanh(pi/2 * sinh(t))
+            # w = pi/2 * cosh(t) / cosh(pi/2 * sinh(t))**2
+
+            # Fast implementation. Note that c = exp(pi/2 * sinh(t))
+            c = ctx.exp(a-b)
+            d = 1/c
+            co = (c+d)/2
+            si = (c-d)/2
+            x = si / co
+            w = (a+b) / co**2
+            diff = abs(x-1)
+            if diff <= tol:
+                break
+
+            nodes.append((x, w))
+            nodes.append((-x, w))
+
+            a *= udelta
+            b *= urdelta
+
+            if verbose and k % 300 == 150:
+                # Note: the number displayed is rather arbitrary. Should
+                # figure out how to print something that looks more like a
+                # percentage
+                print("Calculating nodes:", ctx.nstr(-ctx.log(diff, 10) / prec))
+
+        ctx.prec -= extra
+        return nodes
+
+
+class GaussLegendre(QuadratureRule):
+    r"""
+    This class implements Gauss-Legendre quadrature, which is
+    exceptionally efficient for polynomials and polynomial-like (i.e.
+    very smooth) integrands.
+
+    The abscissas and weights are given by roots and values of
+    Legendre polynomials, which are the orthogonal polynomials
+    on `[-1, 1]` with respect to the unit weight
+    (see :func:`~mpmath.legendre`).
+
+    In this implementation, we take the "degree" `m` of the quadrature
+    to denote a Gauss-Legendre rule of degree `3 \cdot 2^m` (following
+    Borwein, Bailey & Girgensohn). This way we get quadratic, rather
+    than linear, convergence as the degree is incremented.
+
+    Comparison to tanh-sinh quadrature:
+      * Is faster for smooth integrands once nodes have been computed
+      * Initial computation of nodes is usually slower
+      * Handles endpoint singularities worse
+      * Handles infinite integration intervals worse
+
+    """
+
+    def calc_nodes(self, degree, prec, verbose=False):
+        r"""
+        Calculates the abscissas and weights for Gauss-Legendre
+        quadrature of degree of given degree (actually `3 \cdot 2^m`).
+        """
+        ctx = self.ctx
+        # It is important that the epsilon is set lower than the
+        # "real" epsilon
+        epsilon = ctx.ldexp(1, -prec-8)
+        # Fairly high precision might be required for accurate
+        # evaluation of the roots
+        orig = ctx.prec
+        ctx.prec = int(prec*1.5)
+        if degree == 1:
+            x = ctx.sqrt(ctx.mpf(3)/5)
+            w = ctx.mpf(5)/9
+            nodes = [(-x,w),(ctx.zero,ctx.mpf(8)/9),(x,w)]
+            ctx.prec = orig
+            return nodes
+        nodes = []
+        n = 3*2**(degree-1)
+        upto = n//2 + 1
+        for j in xrange(1, upto):
+            # Asymptotic formula for the roots
+            r = ctx.mpf(math.cos(math.pi*(j-0.25)/(n+0.5)))
+            # Newton iteration
+            while 1:
+                t1, t2 = 1, 0
+                # Evaluates the Legendre polynomial using its defining
+                # recurrence relation
+                for j1 in xrange(1,n+1):
+                    t3, t2, t1 = t2, t1, ((2*j1-1)*r*t1 - (j1-1)*t2)/j1
+                t4 = n*(r*t1-t2)/(r**2-1)
+                a = t1/t4
+                r = r - a
+                if abs(a) < epsilon:
+                    break
+            x = r
+            w = 2/((1-r**2)*t4**2)
+            if verbose  and j % 30 == 15:
+                print("Computing nodes (%i of %i)" % (j, upto))
+            nodes.append((x, w))
+            nodes.append((-x, w))
+        ctx.prec = orig
+        return nodes
+
+class QuadratureMethods(object):
+
+    def __init__(ctx, *args, **kwargs):
+        ctx._gauss_legendre = GaussLegendre(ctx)
+        ctx._tanh_sinh = TanhSinh(ctx)
+
+    def quad(ctx, f, *points, **kwargs):
+        r"""
+        Computes a single, double or triple integral over a given
+        1D interval, 2D rectangle, or 3D cuboid. A basic example::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = True
+            >>> quad(sin, [0, pi])
+            2.0
+
+        A basic 2D integral::
+
+            >>> f = lambda x, y: cos(x+y/2)
+            >>> quad(f, [-pi/2, pi/2], [0, pi])
+            4.0
+
+        **Interval format**
+
+        The integration range for each dimension may be specified
+        using a list or tuple. Arguments are interpreted as follows:
+
+        ``quad(f, [x1, x2])`` -- calculates
+        `\int_{x_1}^{x_2} f(x) \, dx`
+
+        ``quad(f, [x1, x2], [y1, y2])`` -- calculates
+        `\int_{x_1}^{x_2} \int_{y_1}^{y_2} f(x,y) \, dy \, dx`
+
+        ``quad(f, [x1, x2], [y1, y2], [z1, z2])`` -- calculates
+        `\int_{x_1}^{x_2} \int_{y_1}^{y_2} \int_{z_1}^{z_2} f(x,y,z)
+        \, dz \, dy \, dx`
+
+        Endpoints may be finite or infinite. An interval descriptor
+        may also contain more than two points. In this
+        case, the integration is split into subintervals, between
+        each pair of consecutive points. This is useful for
+        dealing with mid-interval discontinuities, or integrating
+        over large intervals where the function is irregular or
+        oscillates.
+
+        **Options**
+
+        :func:`~mpmath.quad` recognizes the following keyword arguments:
+
+        *method*
+            Chooses integration algorithm (described below).
+        *error*
+            If set to true, :func:`~mpmath.quad` returns `(v, e)` where `v` is the
+            integral and `e` is the estimated error.
+        *maxdegree*
+            Maximum degree of the quadrature rule to try before
+            quitting.
+        *verbose*
+            Print details about progress.
+
+        **Algorithms**
+
+        Mpmath presently implements two integration algorithms: tanh-sinh
+        quadrature and Gauss-Legendre quadrature. These can be selected
+        using *method='tanh-sinh'* or *method='gauss-legendre'* or by
+        passing the classes *method=TanhSinh*, *method=GaussLegendre*.
+        The functions :func:`~mpmath.quadts` and :func:`~mpmath.quadgl` are also available
+        as shortcuts.
+
+        Both algorithms have the property that doubling the number of
+        evaluation points roughly doubles the accuracy, so both are ideal
+        for high precision quadrature (hundreds or thousands of digits).
+
+        At high precision, computing the nodes and weights for the
+        integration can be expensive (more expensive than computing the
+        function values). To make repeated integrations fast, nodes
+        are automatically cached.
+
+        The advantages of the tanh-sinh algorithm are that it tends to
+        handle endpoint singularities well, and that the nodes are cheap
+        to compute on the first run. For these reasons, it is used by
+        :func:`~mpmath.quad` as the default algorithm.
+
+        Gauss-Legendre quadrature often requires fewer function
+        evaluations, and is therefore often faster for repeated use, but
+        the algorithm does not handle endpoint singularities as well and
+        the nodes are more expensive to compute. Gauss-Legendre quadrature
+        can be a better choice if the integrand is smooth and repeated
+        integrations are required (e.g. for multiple integrals).
+
+        See the documentation for :class:`TanhSinh` and
+        :class:`GaussLegendre` for additional details.
+
+        **Examples of 1D integrals**
+
+        Intervals may be infinite or half-infinite. The following two
+        examples evaluate the limits of the inverse tangent function
+        (`\int 1/(1+x^2) = \tan^{-1} x`), and the Gaussian integral
+        `\int_{\infty}^{\infty} \exp(-x^2)\,dx = \sqrt{\pi}`::
+
+            >>> mp.dps = 15
+            >>> quad(lambda x: 2/(x**2+1), [0, inf])
+            3.14159265358979
+            >>> quad(lambda x: exp(-x**2), [-inf, inf])**2
+            3.14159265358979
+
+        Integrals can typically be resolved to high precision.
+        The following computes 50 digits of `\pi` by integrating the
+        area of the half-circle defined by `x^2 + y^2 \le 1`,
+        `-1 \le x \le 1`, `y \ge 0`::
+
+            >>> mp.dps = 50
+            >>> 2*quad(lambda x: sqrt(1-x**2), [-1, 1])
+            3.1415926535897932384626433832795028841971693993751
+
+        One can just as well compute 1000 digits (output truncated)::
+
+            >>> mp.dps = 1000
+            >>> 2*quad(lambda x: sqrt(1-x**2), [-1, 1])  #doctest:+ELLIPSIS
+            3.141592653589793238462643383279502884...216420199
+
+        Complex integrals are supported. The following computes
+        a residue at `z = 0` by integrating counterclockwise along the
+        diamond-shaped path from `1` to `+i` to `-1` to `-i` to `1`::
+
+            >>> mp.dps = 15
+            >>> chop(quad(lambda z: 1/z, [1,j,-1,-j,1]))
+            (0.0 + 6.28318530717959j)
+
+        **Examples of 2D and 3D integrals**
+
+        Here are several nice examples of analytically solvable
+        2D integrals (taken from MathWorld [1]) that can be evaluated
+        to high precision fairly rapidly by :func:`~mpmath.quad`::
+
+            >>> mp.dps = 30
+            >>> f = lambda x, y: (x-1)/((1-x*y)*log(x*y))
+            >>> quad(f, [0, 1], [0, 1])
+            0.577215664901532860606512090082
+            >>> +euler
+            0.577215664901532860606512090082
+
+            >>> f = lambda x, y: 1/sqrt(1+x**2+y**2)
+            >>> quad(f, [-1, 1], [-1, 1])
+            3.17343648530607134219175646705
+            >>> 4*log(2+sqrt(3))-2*pi/3
+            3.17343648530607134219175646705
+
+            >>> f = lambda x, y: 1/(1-x**2 * y**2)
+            >>> quad(f, [0, 1], [0, 1])
+            1.23370055013616982735431137498
+            >>> pi**2 / 8
+            1.23370055013616982735431137498
+
+            >>> quad(lambda x, y: 1/(1-x*y), [0, 1], [0, 1])
+            1.64493406684822643647241516665
+            >>> pi**2 / 6
+            1.64493406684822643647241516665
+
+        Multiple integrals may be done over infinite ranges::
+
+            >>> mp.dps = 15
+            >>> print(quad(lambda x,y: exp(-x-y), [0, inf], [1, inf]))
+            0.367879441171442
+            >>> print(1/e)
+            0.367879441171442
+
+        For nonrectangular areas, one can call :func:`~mpmath.quad` recursively.
+        For example, we can replicate the earlier example of calculating
+        `\pi` by integrating over the unit-circle, and actually use double
+        quadrature to actually measure the area circle::
+
+            >>> f = lambda x: quad(lambda y: 1, [-sqrt(1-x**2), sqrt(1-x**2)])
+            >>> quad(f, [-1, 1])
+            3.14159265358979
+
+        Here is a simple triple integral::
+
+            >>> mp.dps = 15
+            >>> f = lambda x,y,z: x*y/(1+z)
+            >>> quad(f, [0,1], [0,1], [1,2], method='gauss-legendre')
+            0.101366277027041
+            >>> (log(3)-log(2))/4
+            0.101366277027041
+
+        **Singularities**
+
+        Both tanh-sinh and Gauss-Legendre quadrature are designed to
+        integrate smooth (infinitely differentiable) functions. Neither
+        algorithm copes well with mid-interval singularities (such as
+        mid-interval discontinuities in `f(x)` or `f'(x)`).
+        The best solution is to split the integral into parts::
+
+            >>> mp.dps = 15
+            >>> quad(lambda x: abs(sin(x)), [0, 2*pi])   # Bad
+            3.99900894176779
+            >>> quad(lambda x: abs(sin(x)), [0, pi, 2*pi])  # Good
+            4.0
+
+        The tanh-sinh rule often works well for integrands having a
+        singularity at one or both endpoints::
+
+            >>> mp.dps = 15
+            >>> quad(log, [0, 1], method='tanh-sinh')  # Good
+            -1.0
+            >>> quad(log, [0, 1], method='gauss-legendre')  # Bad
+            -0.999932197413801
+
+        However, the result may still be inaccurate for some functions::
+
+            >>> quad(lambda x: 1/sqrt(x), [0, 1], method='tanh-sinh')
+            1.99999999946942
+
+        This problem is not due to the quadrature rule per se, but to
+        numerical amplification of errors in the nodes. The problem can be
+        circumvented by temporarily increasing the precision::
+
+            >>> mp.dps = 30
+            >>> a = quad(lambda x: 1/sqrt(x), [0, 1], method='tanh-sinh')
+            >>> mp.dps = 15
+            >>> +a
+            2.0
+
+        **Highly variable functions**
+
+        For functions that are smooth (in the sense of being infinitely
+        differentiable) but contain sharp mid-interval peaks or many
+        "bumps", :func:`~mpmath.quad` may fail to provide full accuracy. For
+        example, with default settings, :func:`~mpmath.quad` is able to integrate
+        `\sin(x)` accurately over an interval of length 100 but not over
+        length 1000::
+
+            >>> quad(sin, [0, 100]); 1-cos(100)   # Good
+            0.137681127712316
+            0.137681127712316
+            >>> quad(sin, [0, 1000]); 1-cos(1000)   # Bad
+            -37.8587612408485
+            0.437620923709297
+
+        One solution is to break the integration into 10 intervals of
+        length 100::
+
+            >>> quad(sin, linspace(0, 1000, 10))   # Good
+            0.437620923709297
+
+        Another is to increase the degree of the quadrature::
+
+            >>> quad(sin, [0, 1000], maxdegree=10)   # Also good
+            0.437620923709297
+
+        Whether splitting the interval or increasing the degree is
+        more efficient differs from case to case. Another example is the
+        function `1/(1+x^2)`, which has a sharp peak centered around
+        `x = 0`::
+
+            >>> f = lambda x: 1/(1+x**2)
+            >>> quad(f, [-100, 100])   # Bad
+            3.64804647105268
+            >>> quad(f, [-100, 100], maxdegree=10)   # Good
+            3.12159332021646
+            >>> quad(f, [-100, 0, 100])   # Also good
+            3.12159332021646
+
+        **References**
+
+        1. http://mathworld.wolfram.com/DoubleIntegral.html
+
+        """
+        rule = kwargs.get('method', 'tanh-sinh')
+        if type(rule) is str:
+            if rule == 'tanh-sinh':
+                rule = ctx._tanh_sinh
+            elif rule == 'gauss-legendre':
+                rule = ctx._gauss_legendre
+            else:
+                raise ValueError("unknown quadrature rule: %s" % rule)
+        else:
+            rule = rule(ctx)
+        verbose = kwargs.get('verbose')
+        dim = len(points)
+        orig = prec = ctx.prec
+        epsilon = ctx.eps/8
+        m = kwargs.get('maxdegree') or rule.guess_degree(prec)
+        points = [ctx._as_points(p) for p in points]
+        try:
+            ctx.prec += 20
+            if dim == 1:
+                v, err = rule.summation(f, points[0], prec, epsilon, m, verbose)
+            elif dim == 2:
+                v, err = rule.summation(lambda x: \
+                        rule.summation(lambda y: f(x,y), \
+                        points[1], prec, epsilon, m)[0],
+                    points[0], prec, epsilon, m, verbose)
+            elif dim == 3:
+                v, err = rule.summation(lambda x: \
+                        rule.summation(lambda y: \
+                            rule.summation(lambda z: f(x,y,z), \
+                            points[2], prec, epsilon, m)[0],
+                        points[1], prec, epsilon, m)[0],
+                    points[0], prec, epsilon, m, verbose)
+            else:
+                raise NotImplementedError("quadrature must have dim 1, 2 or 3")
+        finally:
+            ctx.prec = orig
+        if kwargs.get("error"):
+            return +v, err
+        return +v
+
+    def quadts(ctx, *args, **kwargs):
+        """
+        Performs tanh-sinh quadrature. The call
+
+            quadts(func, *points, ...)
+
+        is simply a shortcut for:
+
+            quad(func, *points, ..., method=TanhSinh)
+
+        For example, a single integral and a double integral:
+
+            quadts(lambda x: exp(cos(x)), [0, 1])
+            quadts(lambda x, y: exp(cos(x+y)), [0, 1], [0, 1])
+
+        See the documentation for quad for information about how points
+        arguments and keyword arguments are parsed.
+
+        See documentation for TanhSinh for algorithmic information about
+        tanh-sinh quadrature.
+        """
+        kwargs['method'] = 'tanh-sinh'
+        return ctx.quad(*args, **kwargs)
+
+    def quadgl(ctx, *args, **kwargs):
+        """
+        Performs Gauss-Legendre quadrature. The call
+
+            quadgl(func, *points, ...)
+
+        is simply a shortcut for:
+
+            quad(func, *points, ..., method=GaussLegendre)
+
+        For example, a single integral and a double integral:
+
+            quadgl(lambda x: exp(cos(x)), [0, 1])
+            quadgl(lambda x, y: exp(cos(x+y)), [0, 1], [0, 1])
+
+        See the documentation for quad for information about how points
+        arguments and keyword arguments are parsed.
+
+        See documentation for TanhSinh for algorithmic information about
+        tanh-sinh quadrature.
+        """
+        kwargs['method'] = 'gauss-legendre'
+        return ctx.quad(*args, **kwargs)
+
+    def quadosc(ctx, f, interval, omega=None, period=None, zeros=None):
+        r"""
+        Calculates
+
+        .. math ::
+
+            I = \int_a^b f(x) dx
+
+        where at least one of `a` and `b` is infinite and where
+        `f(x) = g(x) \cos(\omega x  + \phi)` for some slowly
+        decreasing function `g(x)`. With proper input, :func:`~mpmath.quadosc`
+        can also handle oscillatory integrals where the oscillation
+        rate is different from a pure sine or cosine wave.
+
+        In the standard case when `|a| < \infty, b = \infty`,
+        :func:`~mpmath.quadosc` works by evaluating the infinite series
+
+        .. math ::
+
+            I = \int_a^{x_1} f(x) dx +
+            \sum_{k=1}^{\infty} \int_{x_k}^{x_{k+1}} f(x) dx
+
+        where `x_k` are consecutive zeros (alternatively
+        some other periodic reference point) of `f(x)`.
+        Accordingly, :func:`~mpmath.quadosc` requires information about the
+        zeros of `f(x)`. For a periodic function, you can specify
+        the zeros by either providing the angular frequency `\omega`
+        (*omega*) or the *period* `2 \pi/\omega`. In general, you can
+        specify the `n`-th zero by providing the *zeros* arguments.
+        Below is an example of each::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = True
+            >>> f = lambda x: sin(3*x)/(x**2+1)
+            >>> quadosc(f, [0,inf], omega=3)
+            0.37833007080198
+            >>> quadosc(f, [0,inf], period=2*pi/3)
+            0.37833007080198
+            >>> quadosc(f, [0,inf], zeros=lambda n: pi*n/3)
+            0.37833007080198
+            >>> (ei(3)*exp(-3)-exp(3)*ei(-3))/2  # Computed by Mathematica
+            0.37833007080198
+
+        Note that *zeros* was specified to multiply `n` by the
+        *half-period*, not the full period. In theory, it does not matter
+        whether each partial integral is done over a half period or a full
+        period. However, if done over half-periods, the infinite series
+        passed to :func:`~mpmath.nsum` becomes an *alternating series* and this
+        typically makes the extrapolation much more efficient.
+
+        Here is an example of an integration over the entire real line,
+        and a half-infinite integration starting at `-\infty`::
+
+            >>> quadosc(lambda x: cos(x)/(1+x**2), [-inf, inf], omega=1)
+            1.15572734979092
+            >>> pi/e
+            1.15572734979092
+            >>> quadosc(lambda x: cos(x)/x**2, [-inf, -1], period=2*pi)
+            -0.0844109505595739
+            >>> cos(1)+si(1)-pi/2
+            -0.0844109505595738
+
+        Of course, the integrand may contain a complex exponential just as
+        well as a real sine or cosine::
+
+            >>> quadosc(lambda x: exp(3*j*x)/(1+x**2), [-inf,inf], omega=3)
+            (0.156410688228254 + 0.0j)
+            >>> pi/e**3
+            0.156410688228254
+            >>> quadosc(lambda x: exp(3*j*x)/(2+x+x**2), [-inf,inf], omega=3)
+            (0.00317486988463794 - 0.0447701735209082j)
+            >>> 2*pi/sqrt(7)/exp(3*(j+sqrt(7))/2)
+            (0.00317486988463794 - 0.0447701735209082j)
+
+        **Non-periodic functions**
+
+        If `f(x) = g(x) h(x)` for some function `h(x)` that is not
+        strictly periodic, *omega* or *period* might not work, and it might
+        be necessary to use *zeros*.
+
+        A notable exception can be made for Bessel functions which, though not
+        periodic, are "asymptotically periodic" in a sufficiently strong sense
+        that the sum extrapolation will work out::
+
+            >>> quadosc(j0, [0, inf], period=2*pi)
+            1.0
+            >>> quadosc(j1, [0, inf], period=2*pi)
+            1.0
+
+        More properly, one should provide the exact Bessel function zeros::
+
+            >>> j0zero = lambda n: findroot(j0, pi*(n-0.25))
+            >>> quadosc(j0, [0, inf], zeros=j0zero)
+            1.0
+
+        For an example where *zeros* becomes necessary, consider the
+        complete Fresnel integrals
+
+        .. math ::
+
+            \int_0^{\infty} \cos x^2\,dx = \int_0^{\infty} \sin x^2\,dx
+            = \sqrt{\frac{\pi}{8}}.
+
+        Although the integrands do not decrease in magnitude as
+        `x \to \infty`, the integrals are convergent since the oscillation
+        rate increases (causing consecutive periods to asymptotically
+        cancel out). These integrals are virtually impossible to calculate
+        to any kind of accuracy using standard quadrature rules. However,
+        if one provides the correct asymptotic distribution of zeros
+        (`x_n \sim \sqrt{n}`), :func:`~mpmath.quadosc` works::
+
+            >>> mp.dps = 30
+            >>> f = lambda x: cos(x**2)
+            >>> quadosc(f, [0,inf], zeros=lambda n:sqrt(pi*n))
+            0.626657068657750125603941321203
+            >>> f = lambda x: sin(x**2)
+            >>> quadosc(f, [0,inf], zeros=lambda n:sqrt(pi*n))
+            0.626657068657750125603941321203
+            >>> sqrt(pi/8)
+            0.626657068657750125603941321203
+
+        (Interestingly, these integrals can still be evaluated if one
+        places some other constant than `\pi` in the square root sign.)
+
+        In general, if `f(x) \sim g(x) \cos(h(x))`, the zeros follow
+        the inverse-function distribution `h^{-1}(x)`::
+
+            >>> mp.dps = 15
+            >>> f = lambda x: sin(exp(x))
+            >>> quadosc(f, [1,inf], zeros=lambda n: log(n))
+            -0.25024394235267
+            >>> pi/2-si(e)
+            -0.250243942352671
+
+        **Non-alternating functions**
+
+        If the integrand oscillates around a positive value, without
+        alternating signs, the extrapolation might fail. A simple trick
+        that sometimes works is to multiply or divide the frequency by 2::
+
+            >>> f = lambda x: 1/x**2+sin(x)/x**4
+            >>> quadosc(f, [1,inf], omega=1)  # Bad
+            1.28642190869861
+            >>> quadosc(f, [1,inf], omega=0.5)  # Perfect
+            1.28652953559617
+            >>> 1+(cos(1)+ci(1)+sin(1))/6
+            1.28652953559617
+
+        **Fast decay**
+
+        :func:`~mpmath.quadosc` is primarily useful for slowly decaying
+        integrands. If the integrand decreases exponentially or faster,
+        :func:`~mpmath.quad` will likely handle it without trouble (and generally be
+        much faster than :func:`~mpmath.quadosc`)::
+
+            >>> quadosc(lambda x: cos(x)/exp(x), [0, inf], omega=1)
+            0.5
+            >>> quad(lambda x: cos(x)/exp(x), [0, inf])
+            0.5
+
+        """
+        a, b = ctx._as_points(interval)
+        a = ctx.convert(a)
+        b = ctx.convert(b)
+        if [omega, period, zeros].count(None) != 2:
+            raise ValueError( \
+                "must specify exactly one of omega, period, zeros")
+        if a == ctx.ninf and b == ctx.inf:
+            s1 = ctx.quadosc(f, [a, 0], omega=omega, zeros=zeros, period=period)
+            s2 = ctx.quadosc(f, [0, b], omega=omega, zeros=zeros, period=period)
+            return s1 + s2
+        if a == ctx.ninf:
+            if zeros:
+                return ctx.quadosc(lambda x:f(-x), [-b,-a], lambda n: zeros(-n))
+            else:
+                return ctx.quadosc(lambda x:f(-x), [-b,-a], omega=omega, period=period)
+        if b != ctx.inf:
+            raise ValueError("quadosc requires an infinite integration interval")
+        if not zeros:
+            if omega:
+                period = 2*ctx.pi/omega
+            zeros = lambda n: n*period/2
+        #for n in range(1,10):
+        #    p = zeros(n)
+        #    if p > a:
+        #        break
+        #if n >= 9:
+        #    raise ValueError("zeros do not appear to be correctly indexed")
+        n = 1
+        s = ctx.quadgl(f, [a, zeros(n)])
+        def term(k):
+            return ctx.quadgl(f, [zeros(k), zeros(k+1)])
+        s += ctx.nsum(term, [n, ctx.inf])
+        return s
+
+    def quadsubdiv(ctx, f, interval, tol=None, maxintervals=None, **kwargs):
+        """
+        Computes the integral of *f* over the interval or path specified
+        by *interval*, using :func:`~mpmath.quad` together with adaptive
+        subdivision of the interval.
+
+        This function gives an accurate answer for some integrals where
+        :func:`~mpmath.quad` fails::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = True
+            >>> quad(lambda x: abs(sin(x)), [0, 2*pi])
+            3.99900894176779
+            >>> quadsubdiv(lambda x: abs(sin(x)), [0, 2*pi])
+            4.0
+            >>> quadsubdiv(sin, [0, 1000])
+            0.437620923709297
+            >>> quadsubdiv(lambda x: 1/(1+x**2), [-100, 100])
+            3.12159332021646
+            >>> quadsubdiv(lambda x: ceil(x), [0, 100])
+            5050.0
+            >>> quadsubdiv(lambda x: sin(x+exp(x)), [0,8])
+            0.347400172657248
+
+        The argument *maxintervals* can be set to limit the permissible
+        subdivision::
+
+            >>> quadsubdiv(lambda x: sin(x**2), [0,100], maxintervals=5, error=True)
+            (-5.40487904307774, 5.011)
+            >>> quadsubdiv(lambda x: sin(x**2), [0,100], maxintervals=100, error=True)
+            (0.631417921866934, 1.10101120134116e-17)
+
+        Subdivision does not guarantee a correct answer since, the error
+        estimate on subintervals may be inaccurate::
+
+            >>> quadsubdiv(lambda x: sech(10*x-2)**2 + sech(100*x-40)**4 + sech(1000*x-600)**6, [0,1], error=True)
+            (0.210802735500549, 1.0001111101e-17)
+            >>> mp.dps = 20
+            >>> quadsubdiv(lambda x: sech(10*x-2)**2 + sech(100*x-40)**4 + sech(1000*x-600)**6, [0,1], error=True)
+            (0.21080273550054927738, 2.200000001e-24)
+
+        The second answer is correct. We can get an accurate result at lower
+        precision by forcing a finer initial subdivision::
+
+            >>> mp.dps = 15
+            >>> quadsubdiv(lambda x: sech(10*x-2)**2 + sech(100*x-40)**4 + sech(1000*x-600)**6, linspace(0,1,5))
+            0.210802735500549
+
+        The following integral is too oscillatory for convergence, but we can get a
+        reasonable estimate::
+
+            >>> v, err = fp.quadsubdiv(lambda x: fp.sin(1/x), [0,1], error=True)
+            >>> round(v, 6), round(err, 6)
+            (0.504067, 1e-06)
+            >>> sin(1) - ci(1)
+            0.504067061906928
+
+        """
+        queue = []
+        for i in range(len(interval)-1):
+            queue.append((interval[i], interval[i+1]))
+        total = ctx.zero
+        total_error = ctx.zero
+        if maxintervals is None:
+            maxintervals = 10 * ctx.prec
+        count = 0
+        quad_args = kwargs.copy()
+        quad_args["verbose"] = False
+        quad_args["error"] = True
+        if tol is None:
+            tol = +ctx.eps
+        orig = ctx.prec
+        try:
+            ctx.prec += 5
+            while queue:
+                a, b = queue.pop()
+                s, err = ctx.quad(f, [a, b], **quad_args)
+                if kwargs.get("verbose"):
+                    print("subinterval", count, a, b, err)
+                if err < tol or count > maxintervals:
+                    total += s
+                    total_error += err
+                else:
+                    count += 1
+                    if count == maxintervals and kwargs.get("verbose"):
+                        print("warning: number of intervals exceeded maxintervals")
+                    if a == -ctx.inf and b == ctx.inf:
+                        m = 0
+                    elif a == -ctx.inf:
+                        m = min(b-1, 2*b)
+                    elif b == ctx.inf:
+                        m = max(a+1, 2*a)
+                    else:
+                        m = a + (b - a) / 2
+                    queue.append((a, m))
+                    queue.append((m, b))
+        finally:
+            ctx.prec = orig
+        if kwargs.get("error"):
+            return +total, +total_error
+        else:
+            return +total
+
+if __name__ == '__main__':
+    import doctest
+    doctest.testmod()

lib/python3.11/site-packages/mpmath/ctx_base.py

@@ -0,0 +1,494 @@
+from operator import gt, lt
+
+from .libmp.backend import xrange
+
+from .functions.functions import SpecialFunctions
+from .functions.rszeta import RSCache
+from .calculus.quadrature import QuadratureMethods
+from .calculus.inverselaplace import LaplaceTransformInversionMethods
+from .calculus.calculus import CalculusMethods
+from .calculus.optimization import OptimizationMethods
+from .calculus.odes import ODEMethods
+from .matrices.matrices import MatrixMethods
+from .matrices.calculus import MatrixCalculusMethods
+from .matrices.linalg import LinearAlgebraMethods
+from .matrices.eigen import Eigen
+from .identification import IdentificationMethods
+from .visualization import VisualizationMethods
+
+from . import libmp
+
+class Context(object):
+    pass
+
+class StandardBaseContext(Context,
+    SpecialFunctions,
+    RSCache,
+    QuadratureMethods,
+    LaplaceTransformInversionMethods,
+    CalculusMethods,
+    MatrixMethods,
+    MatrixCalculusMethods,
+    LinearAlgebraMethods,
+    Eigen,
+    IdentificationMethods,
+    OptimizationMethods,
+    ODEMethods,
+    VisualizationMethods):
+
+    NoConvergence = libmp.NoConvergence
+    ComplexResult = libmp.ComplexResult
+
+    def __init__(ctx):
+        ctx._aliases = {}
+        # Call those that need preinitialization (e.g. for wrappers)
+        SpecialFunctions.__init__(ctx)
+        RSCache.__init__(ctx)
+        QuadratureMethods.__init__(ctx)
+        LaplaceTransformInversionMethods.__init__(ctx)
+        CalculusMethods.__init__(ctx)
+        MatrixMethods.__init__(ctx)
+
+    def _init_aliases(ctx):
+        for alias, value in ctx._aliases.items():
+            try:
+                setattr(ctx, alias, getattr(ctx, value))
+            except AttributeError:
+                pass
+
+    _fixed_precision = False
+
+    # XXX
+    verbose = False
+
+    def warn(ctx, msg):
+        print("Warning:", msg)
+
+    def bad_domain(ctx, msg):
+        raise ValueError(msg)
+
+    def _re(ctx, x):
+        if hasattr(x, "real"):
+            return x.real
+        return x
+
+    def _im(ctx, x):
+        if hasattr(x, "imag"):
+            return x.imag
+        return ctx.zero
+
+    def _as_points(ctx, x):
+        return x
+
+    def fneg(ctx, x, **kwargs):
+        return -ctx.convert(x)
+
+    def fadd(ctx, x, y, **kwargs):
+        return ctx.convert(x)+ctx.convert(y)
+
+    def fsub(ctx, x, y, **kwargs):
+        return ctx.convert(x)-ctx.convert(y)
+
+    def fmul(ctx, x, y, **kwargs):
+        return ctx.convert(x)*ctx.convert(y)
+
+    def fdiv(ctx, x, y, **kwargs):
+        return ctx.convert(x)/ctx.convert(y)
+
+    def fsum(ctx, args, absolute=False, squared=False):
+        if absolute:
+            if squared:
+                return sum((abs(x)**2 for x in args), ctx.zero)
+            return sum((abs(x) for x in args), ctx.zero)
+        if squared:
+            return sum((x**2 for x in args), ctx.zero)
+        return sum(args, ctx.zero)
+
+    def fdot(ctx, xs, ys=None, conjugate=False):
+        if ys is not None:
+            xs = zip(xs, ys)
+        if conjugate:
+            cf = ctx.conj
+            return sum((x*cf(y) for (x,y) in xs), ctx.zero)
+        else:
+            return sum((x*y for (x,y) in xs), ctx.zero)
+
+    def fprod(ctx, args):
+        prod = ctx.one
+        for arg in args:
+            prod *= arg
+        return prod
+
+    def nprint(ctx, x, n=6, **kwargs):
+        """
+        Equivalent to ``print(nstr(x, n))``.
+        """
+        print(ctx.nstr(x, n, **kwargs))
+
+    def chop(ctx, x, tol=None):
+        """
+        Chops off small real or imaginary parts, or converts
+        numbers close to zero to exact zeros. The input can be a
+        single number or an iterable::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> chop(5+1e-10j, tol=1e-9)
+            mpf('5.0')
+            >>> nprint(chop([1.0, 1e-20, 3+1e-18j, -4, 2]))
+            [1.0, 0.0, 3.0, -4.0, 2.0]
+
+        The tolerance defaults to ``100*eps``.
+        """
+        if tol is None:
+            tol = 100*ctx.eps
+        try:
+            x = ctx.convert(x)
+            absx = abs(x)
+            if abs(x) < tol:
+                return ctx.zero
+            if ctx._is_complex_type(x):
+                #part_tol = min(tol, absx*tol)
+                part_tol = max(tol, absx*tol)
+                if abs(x.imag) < part_tol:
+                    return x.real
+                if abs(x.real) < part_tol:
+                    return ctx.mpc(0, x.imag)
+        except TypeError:
+            if isinstance(x, ctx.matrix):
+                return x.apply(lambda a: ctx.chop(a, tol))
+            if hasattr(x, "__iter__"):
+                return [ctx.chop(a, tol) for a in x]
+        return x
+
+    def almosteq(ctx, s, t, rel_eps=None, abs_eps=None):
+        r"""
+        Determine whether the difference between `s` and `t` is smaller
+        than a given epsilon, either relatively or absolutely.
+
+        Both a maximum relative difference and a maximum difference
+        ('epsilons') may be specified. The absolute difference is
+        defined as `|s-t|` and the relative difference is defined
+        as `|s-t|/\max(|s|, |t|)`.
+
+        If only one epsilon is given, both are set to the same value.
+        If none is given, both epsilons are set to `2^{-p+m}` where
+        `p` is the current working precision and `m` is a small
+        integer. The default setting typically allows :func:`~mpmath.almosteq`
+        to be used to check for mathematical equality
+        in the presence of small rounding errors.
+
+        **Examples**
+
+            >>> from mpmath import *
+            >>> mp.dps = 15
+            >>> almosteq(3.141592653589793, 3.141592653589790)
+            True
+            >>> almosteq(3.141592653589793, 3.141592653589700)
+            False
+            >>> almosteq(3.141592653589793, 3.141592653589700, 1e-10)
+            True
+            >>> almosteq(1e-20, 2e-20)
+            True
+            >>> almosteq(1e-20, 2e-20, rel_eps=0, abs_eps=0)
+            False
+
+        """
+        t = ctx.convert(t)
+        if abs_eps is None and rel_eps is None:
+            rel_eps = abs_eps = ctx.ldexp(1, -ctx.prec+4)
+        if abs_eps is None:
+            abs_eps = rel_eps
+        elif rel_eps is None:
+            rel_eps = abs_eps
+        diff = abs(s-t)
+        if diff <= abs_eps:
+            return True
+        abss = abs(s)
+        abst = abs(t)
+        if abss < abst:
+            err = diff/abst
+        else:
+            err = diff/abss
+        return err <= rel_eps
+
+    def arange(ctx, *args):
+        r"""
+        This is a generalized version of Python's :func:`~mpmath.range` function
+        that accepts fractional endpoints and step sizes and
+        returns a list of ``mpf`` instances. Like :func:`~mpmath.range`,
+        :func:`~mpmath.arange` can be called with 1, 2 or 3 arguments:
+
+        ``arange(b)``
+            `[0, 1, 2, \ldots, x]`
+        ``arange(a, b)``
+            `[a, a+1, a+2, \ldots, x]`
+        ``arange(a, b, h)``
+            `[a, a+h, a+h, \ldots, x]`
+
+        where `b-1 \le x < b` (in the third case, `b-h \le x < b`).
+
+        Like Python's :func:`~mpmath.range`, the endpoint is not included. To
+        produce ranges where the endpoint is included, :func:`~mpmath.linspace`
+        is more convenient.
+
+        **Examples**
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> arange(4)
+            [mpf('0.0'), mpf('1.0'), mpf('2.0'), mpf('3.0')]
+            >>> arange(1, 2, 0.25)
+            [mpf('1.0'), mpf('1.25'), mpf('1.5'), mpf('1.75')]
+            >>> arange(1, -1, -0.75)
+            [mpf('1.0'), mpf('0.25'), mpf('-0.5')]
+
+        """
+        if not len(args) <= 3:
+            raise TypeError('arange expected at most 3 arguments, got %i'
+                            % len(args))
+        if not len(args) >= 1:
+            raise TypeError('arange expected at least 1 argument, got %i'
+                            % len(args))
+        # set default
+        a = 0
+        dt = 1
+        # interpret arguments
+        if len(args) == 1:
+            b = args[0]
+        elif len(args) >= 2:
+            a = args[0]
+            b = args[1]
+        if len(args) == 3:
+            dt = args[2]
+        a, b, dt = ctx.mpf(a), ctx.mpf(b), ctx.mpf(dt)
+        assert a + dt != a, 'dt is too small and would cause an infinite loop'
+        # adapt code for sign of dt
+        if a > b:
+            if dt > 0:
+                return []
+            op = gt
+        else:
+            if dt < 0:
+                return []
+            op = lt
+        # create list
+        result = []
+        i = 0
+        t = a
+        while 1:
+            t = a + dt*i
+            i += 1
+            if op(t, b):
+                result.append(t)
+            else:
+                break
+        return result
+
+    def linspace(ctx, *args, **kwargs):
+        """
+        ``linspace(a, b, n)`` returns a list of `n` evenly spaced
+        samples from `a` to `b`. The syntax ``linspace(mpi(a,b), n)``
+        is also valid.
+
+        This function is often more convenient than :func:`~mpmath.arange`
+        for partitioning an interval into subintervals, since
+        the endpoint is included::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> linspace(1, 4, 4)
+            [mpf('1.0'), mpf('2.0'), mpf('3.0'), mpf('4.0')]
+
+        You may also provide the keyword argument ``endpoint=False``::
+
+            >>> linspace(1, 4, 4, endpoint=False)
+            [mpf('1.0'), mpf('1.75'), mpf('2.5'), mpf('3.25')]
+
+        """
+        if len(args) == 3:
+            a = ctx.mpf(args[0])
+            b = ctx.mpf(args[1])
+            n = int(args[2])
+        elif len(args) == 2:
+            assert hasattr(args[0], '_mpi_')
+            a = args[0].a
+            b = args[0].b
+            n = int(args[1])
+        else:
+            raise TypeError('linspace expected 2 or 3 arguments, got %i' \
+                            % len(args))
+        if n < 1:
+            raise ValueError('n must be greater than 0')
+        if not 'endpoint' in kwargs or kwargs['endpoint']:
+            if n == 1:
+                return [ctx.mpf(a)]
+            step = (b - a) / ctx.mpf(n - 1)
+            y = [i*step + a for i in xrange(n)]
+            y[-1] = b
+        else:
+            step = (b - a) / ctx.mpf(n)
+            y = [i*step + a for i in xrange(n)]
+        return y
+
+    def cos_sin(ctx, z, **kwargs):
+        return ctx.cos(z, **kwargs), ctx.sin(z, **kwargs)
+
+    def cospi_sinpi(ctx, z, **kwargs):
+        return ctx.cospi(z, **kwargs), ctx.sinpi(z, **kwargs)
+
+    def _default_hyper_maxprec(ctx, p):
+        return int(1000 * p**0.25 + 4*p)
+
+    _gcd = staticmethod(libmp.gcd)
+    list_primes = staticmethod(libmp.list_primes)
+    isprime = staticmethod(libmp.isprime)
+    bernfrac = staticmethod(libmp.bernfrac)
+    moebius = staticmethod(libmp.moebius)
+    _ifac = staticmethod(libmp.ifac)
+    _eulernum = staticmethod(libmp.eulernum)
+    _stirling1 = staticmethod(libmp.stirling1)
+    _stirling2 = staticmethod(libmp.stirling2)
+
+    def sum_accurately(ctx, terms, check_step=1):
+        prec = ctx.prec
+        try:
+            extraprec = 10
+            while 1:
+                ctx.prec = prec + extraprec + 5
+                max_mag = ctx.ninf
+                s = ctx.zero
+                k = 0
+                for term in terms():
+                    s += term
+                    if (not k % check_step) and term:
+                        term_mag = ctx.mag(term)
+                        max_mag = max(max_mag, term_mag)
+                        sum_mag = ctx.mag(s)
+                        if sum_mag - term_mag > ctx.prec:
+                            break
+                    k += 1
+                cancellation = max_mag - sum_mag
+                if cancellation != cancellation:
+                    break
+                if cancellation < extraprec or ctx._fixed_precision:
+                    break
+                extraprec += min(ctx.prec, cancellation)
+            return s
+        finally:
+            ctx.prec = prec
+
+    def mul_accurately(ctx, factors, check_step=1):
+        prec = ctx.prec
+        try:
+            extraprec = 10
+            while 1:
+                ctx.prec = prec + extraprec + 5
+                max_mag = ctx.ninf
+                one = ctx.one
+                s = one
+                k = 0
+                for factor in factors():
+                    s *= factor
+                    term = factor - one
+                    if (not k % check_step):
+                        term_mag = ctx.mag(term)
+                        max_mag = max(max_mag, term_mag)
+                        sum_mag = ctx.mag(s-one)
+                        #if sum_mag - term_mag > ctx.prec:
+                        #    break
+                        if -term_mag > ctx.prec:
+                            break
+                    k += 1
+                cancellation = max_mag - sum_mag
+                if cancellation != cancellation:
+                    break
+                if cancellation < extraprec or ctx._fixed_precision:
+                    break
+                extraprec += min(ctx.prec, cancellation)
+            return s
+        finally:
+            ctx.prec = prec
+
+    def power(ctx, x, y):
+        r"""Converts `x` and `y` to mpmath numbers and evaluates
+        `x^y = \exp(y \log(x))`::
+
+            >>> from mpmath import *
+            >>> mp.dps = 30; mp.pretty = True
+            >>> power(2, 0.5)
+            1.41421356237309504880168872421
+
+        This shows the leading few digits of a large Mersenne prime
+        (performing the exact calculation ``2**43112609-1`` and
+        displaying the result in Python would be very slow)::
+
+            >>> power(2, 43112609)-1
+            3.16470269330255923143453723949e+12978188
+        """
+        return ctx.convert(x) ** ctx.convert(y)
+
+    def _zeta_int(ctx, n):
+        return ctx.zeta(n)
+
+    def maxcalls(ctx, f, N):
+        """
+        Return a wrapped copy of *f* that raises ``NoConvergence`` when *f*
+        has been called more than *N* times::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15
+            >>> f = maxcalls(sin, 10)
+            >>> print(sum(f(n) for n in range(10)))
+            1.95520948210738
+            >>> f(10) # doctest: +IGNORE_EXCEPTION_DETAIL
+            Traceback (most recent call last):
+              ...
+            NoConvergence: maxcalls: function evaluated 10 times
+
+        """
+        counter = [0]
+        def f_maxcalls_wrapped(*args, **kwargs):
+            counter[0] += 1
+            if counter[0] > N:
+                raise ctx.NoConvergence("maxcalls: function evaluated %i times" % N)
+            return f(*args, **kwargs)
+        return f_maxcalls_wrapped
+
+    def memoize(ctx, f):
+        """
+        Return a wrapped copy of *f* that caches computed values, i.e.
+        a memoized copy of *f*. Values are only reused if the cached precision
+        is equal to or higher than the working precision::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = True
+            >>> f = memoize(maxcalls(sin, 1))
+            >>> f(2)
+            0.909297426825682
+            >>> f(2)
+            0.909297426825682
+            >>> mp.dps = 25
+            >>> f(2) # doctest: +IGNORE_EXCEPTION_DETAIL
+            Traceback (most recent call last):
+              ...
+            NoConvergence: maxcalls: function evaluated 1 times
+
+        """
+        f_cache = {}
+        def f_cached(*args, **kwargs):
+            if kwargs:
+                key = args, tuple(kwargs.items())
+            else:
+                key = args
+            prec = ctx.prec
+            if key in f_cache:
+                cprec, cvalue = f_cache[key]
+                if cprec >= prec:
+                    return +cvalue
+            value = f(*args, **kwargs)
+            f_cache[key] = (prec, value)
+            return value
+        f_cached.__name__ = f.__name__
+        f_cached.__doc__ = f.__doc__
+        return f_cached

lib/python3.11/site-packages/mpmath/ctx_fp.py

@@ -0,0 +1,253 @@
+from .ctx_base import StandardBaseContext
+
+import math
+import cmath
+from . import math2
+
+from . import function_docs
+
+from .libmp import mpf_bernoulli, to_float, int_types
+from . import libmp
+
+class FPContext(StandardBaseContext):
+    """
+    Context for fast low-precision arithmetic (53-bit precision, giving at most
+    about 15-digit accuracy), using Python's builtin float and complex.
+    """
+
+    def __init__(ctx):
+        StandardBaseContext.__init__(ctx)
+
+        # Override SpecialFunctions implementation
+        ctx.loggamma = math2.loggamma
+        ctx._bernoulli_cache = {}
+        ctx.pretty = False
+
+        ctx._init_aliases()
+
+    _mpq = lambda cls, x: float(x[0])/x[1]
+
+    NoConvergence = libmp.NoConvergence
+
+    def _get_prec(ctx): return 53
+    def _set_prec(ctx, p): return
+    def _get_dps(ctx): return 15
+    def _set_dps(ctx, p): return
+
+    _fixed_precision = True
+
+    prec = property(_get_prec, _set_prec)
+    dps = property(_get_dps, _set_dps)
+
+    zero = 0.0
+    one = 1.0
+    eps = math2.EPS
+    inf = math2.INF
+    ninf = math2.NINF
+    nan = math2.NAN
+    j = 1j
+
+    # Called by SpecialFunctions.__init__()
+    @classmethod
+    def _wrap_specfun(cls, name, f, wrap):
+        if wrap:
+            def f_wrapped(ctx, *args, **kwargs):
+                convert = ctx.convert
+                args = [convert(a) for a in args]
+                return f(ctx, *args, **kwargs)
+        else:
+            f_wrapped = f
+        f_wrapped.__doc__ = function_docs.__dict__.get(name, f.__doc__)
+        setattr(cls, name, f_wrapped)
+
+    def bernoulli(ctx, n):
+        cache = ctx._bernoulli_cache
+        if n in cache:
+            return cache[n]
+        cache[n] = to_float(mpf_bernoulli(n, 53, 'n'), strict=True)
+        return cache[n]
+
+    pi = math2.pi
+    e = math2.e
+    euler = math2.euler
+    sqrt2 = 1.4142135623730950488
+    sqrt5 = 2.2360679774997896964
+    phi = 1.6180339887498948482
+    ln2 = 0.69314718055994530942
+    ln10 = 2.302585092994045684
+    euler = 0.57721566490153286061
+    catalan = 0.91596559417721901505
+    khinchin = 2.6854520010653064453
+    apery = 1.2020569031595942854
+    glaisher = 1.2824271291006226369
+
+    absmin = absmax = abs
+
+    def is_special(ctx, x):
+        return x - x != 0.0
+
+    def isnan(ctx, x):
+        return x != x
+
+    def isinf(ctx, x):
+        return abs(x) == math2.INF
+
+    def isnormal(ctx, x):
+        if x:
+            return x - x == 0.0
+        return False
+
+    def isnpint(ctx, x):
+        if type(x) is complex:
+            if x.imag:
+                return False
+            x = x.real
+        return x <= 0.0 and round(x) == x
+
+    mpf = float
+    mpc = complex
+
+    def convert(ctx, x):
+        try:
+            return float(x)
+        except:
+            return complex(x)
+
+    power = staticmethod(math2.pow)
+    sqrt = staticmethod(math2.sqrt)
+    exp = staticmethod(math2.exp)
+    ln = log = staticmethod(math2.log)
+    cos = staticmethod(math2.cos)
+    sin = staticmethod(math2.sin)
+    tan = staticmethod(math2.tan)
+    cos_sin = staticmethod(math2.cos_sin)
+    acos = staticmethod(math2.acos)
+    asin = staticmethod(math2.asin)
+    atan = staticmethod(math2.atan)
+    cosh = staticmethod(math2.cosh)
+    sinh = staticmethod(math2.sinh)
+    tanh = staticmethod(math2.tanh)
+    gamma = staticmethod(math2.gamma)
+    rgamma = staticmethod(math2.rgamma)
+    fac = factorial = staticmethod(math2.factorial)
+    floor = staticmethod(math2.floor)
+    ceil = staticmethod(math2.ceil)
+    cospi = staticmethod(math2.cospi)
+    sinpi = staticmethod(math2.sinpi)
+    cbrt = staticmethod(math2.cbrt)
+    _nthroot = staticmethod(math2.nthroot)
+    _ei = staticmethod(math2.ei)
+    _e1 = staticmethod(math2.e1)
+    _zeta = _zeta_int = staticmethod(math2.zeta)
+
+    # XXX: math2
+    def arg(ctx, z):
+        z = complex(z)
+        return math.atan2(z.imag, z.real)
+
+    def expj(ctx, x):
+        return ctx.exp(ctx.j*x)
+
+    def expjpi(ctx, x):
+        return ctx.exp(ctx.j*ctx.pi*x)
+
+    ldexp = math.ldexp
+    frexp = math.frexp
+
+    def mag(ctx, z):
+        if z:
+            return ctx.frexp(abs(z))[1]
+        return ctx.ninf
+
+    def isint(ctx, z):
+        if hasattr(z, "imag"):   # float/int don't have .real/.imag in py2.5
+            if z.imag:
+                return False
+            z = z.real
+        try:
+            return z == int(z)
+        except:
+            return False
+
+    def nint_distance(ctx, z):
+        if hasattr(z, "imag"):   # float/int don't have .real/.imag in py2.5
+            n = round(z.real)
+        else:
+            n = round(z)
+        if n == z:
+            return n, ctx.ninf
+        return n, ctx.mag(abs(z-n))
+
+    def _convert_param(ctx, z):
+        if type(z) is tuple:
+            p, q = z
+            return ctx.mpf(p) / q, 'R'
+        if hasattr(z, "imag"):    # float/int don't have .real/.imag in py2.5
+            intz = int(z.real)
+        else:
+            intz = int(z)
+        if z == intz:
+            return intz, 'Z'
+        return z, 'R'
+
+    def _is_real_type(ctx, z):
+        return isinstance(z, float) or isinstance(z, int_types)
+
+    def _is_complex_type(ctx, z):
+        return isinstance(z, complex)
+
+    def hypsum(ctx, p, q, types, coeffs, z, maxterms=6000, **kwargs):
+        coeffs = list(coeffs)
+        num = range(p)
+        den = range(p,p+q)
+        tol = ctx.eps
+        s = t = 1.0
+        k = 0
+        while 1:
+            for i in num: t *= (coeffs[i]+k)
+            for i in den: t /= (coeffs[i]+k)
+            k += 1; t /= k; t *= z; s += t
+            if abs(t) < tol:
+                return s
+            if k > maxterms:
+                raise ctx.NoConvergence
+
+    def atan2(ctx, x, y):
+        return math.atan2(x, y)
+
+    def psi(ctx, m, z):
+        m = int(m)
+        if m == 0:
+            return ctx.digamma(z)
+        return (-1)**(m+1) * ctx.fac(m) * ctx.zeta(m+1, z)
+
+    digamma = staticmethod(math2.digamma)
+
+    def harmonic(ctx, x):
+        x = ctx.convert(x)
+        if x == 0 or x == 1:
+            return x
+        return ctx.digamma(x+1) + ctx.euler
+
+    nstr = str
+
+    def to_fixed(ctx, x, prec):
+        return int(math.ldexp(x, prec))
+
+    def rand(ctx):
+        import random
+        return random.random()
+
+    _erf = staticmethod(math2.erf)
+    _erfc = staticmethod(math2.erfc)
+
+    def sum_accurately(ctx, terms, check_step=1):
+        s = ctx.zero
+        k = 0
+        for term in terms():
+            s += term
+            if (not k % check_step) and term:
+                if abs(term) <= 1e-18*abs(s):
+                    break
+            k += 1
+        return s

lib/python3.11/site-packages/mpmath/ctx_iv.py

@@ -0,0 +1,551 @@
+import operator
+
+from . import libmp
+
+from .libmp.backend import basestring
+
+from .libmp import (
+    int_types, MPZ_ONE,
+    prec_to_dps, dps_to_prec, repr_dps,
+    round_floor, round_ceiling,
+    fzero, finf, fninf, fnan,
+    mpf_le, mpf_neg,
+    from_int, from_float, from_str, from_rational,
+    mpi_mid, mpi_delta, mpi_str,
+    mpi_abs, mpi_pos, mpi_neg, mpi_add, mpi_sub,
+    mpi_mul, mpi_div, mpi_pow_int, mpi_pow,
+    mpi_from_str,
+    mpci_pos, mpci_neg, mpci_add, mpci_sub, mpci_mul, mpci_div, mpci_pow,
+    mpci_abs, mpci_pow, mpci_exp, mpci_log,
+    ComplexResult,
+    mpf_hash, mpc_hash)
+from .matrices.matrices import _matrix
+
+mpi_zero = (fzero, fzero)
+
+from .ctx_base import StandardBaseContext
+
+new = object.__new__
+
+def convert_mpf_(x, prec, rounding):
+    if hasattr(x, "_mpf_"): return x._mpf_
+    if isinstance(x, int_types): return from_int(x, prec, rounding)
+    if isinstance(x, float): return from_float(x, prec, rounding)
+    if isinstance(x, basestring): return from_str(x, prec, rounding)
+    raise NotImplementedError
+
+
+class ivmpf(object):
+    """
+    Interval arithmetic class. Precision is controlled by iv.prec.
+    """
+
+    def __new__(cls, x=0):
+        return cls.ctx.convert(x)
+
+    def cast(self, cls, f_convert):
+        a, b = self._mpi_
+        if a == b:
+            return cls(f_convert(a))
+        raise ValueError
+
+    def __int__(self):
+        return self.cast(int, libmp.to_int)
+
+    def __float__(self):
+        return self.cast(float, libmp.to_float)
+
+    def __complex__(self):
+        return self.cast(complex, libmp.to_float)
+
+    def __hash__(self):
+        a, b = self._mpi_
+        if a == b:
+            return mpf_hash(a)
+        else:
+            return hash(self._mpi_)
+
+    @property
+    def real(self): return self
+
+    @property
+    def imag(self): return self.ctx.zero
+
+    def conjugate(self): return self
+
+    @property
+    def a(self):
+        a, b = self._mpi_
+        return self.ctx.make_mpf((a, a))
+
+    @property
+    def b(self):
+        a, b = self._mpi_
+        return self.ctx.make_mpf((b, b))
+
+    @property
+    def mid(self):
+        ctx = self.ctx
+        v = mpi_mid(self._mpi_, ctx.prec)
+        return ctx.make_mpf((v, v))
+
+    @property
+    def delta(self):
+        ctx = self.ctx
+        v = mpi_delta(self._mpi_, ctx.prec)
+        return ctx.make_mpf((v,v))
+
+    @property
+    def _mpci_(self):
+        return self._mpi_, mpi_zero
+
+    def _compare(*args):
+        raise TypeError("no ordering relation is defined for intervals")
+
+    __gt__ = _compare
+    __le__ = _compare
+    __gt__ = _compare
+    __ge__ = _compare
+
+    def __contains__(self, t):
+        t = self.ctx.mpf(t)
+        return (self.a <= t.a) and (t.b <= self.b)
+
+    def __str__(self):
+        return mpi_str(self._mpi_, self.ctx.prec)
+
+    def __repr__(self):
+        if self.ctx.pretty:
+            return str(self)
+        a, b = self._mpi_
+        n = repr_dps(self.ctx.prec)
+        a = libmp.to_str(a, n)
+        b = libmp.to_str(b, n)
+        return "mpi(%r, %r)" % (a, b)
+
+    def _compare(s, t, cmpfun):
+        if not hasattr(t, "_mpi_"):
+            try:
+                t = s.ctx.convert(t)
+            except:
+                return NotImplemented
+        return cmpfun(s._mpi_, t._mpi_)
+
+    def __eq__(s, t): return s._compare(t, libmp.mpi_eq)
+    def __ne__(s, t): return s._compare(t, libmp.mpi_ne)
+    def __lt__(s, t): return s._compare(t, libmp.mpi_lt)
+    def __le__(s, t): return s._compare(t, libmp.mpi_le)
+    def __gt__(s, t): return s._compare(t, libmp.mpi_gt)
+    def __ge__(s, t): return s._compare(t, libmp.mpi_ge)
+
+    def __abs__(self):
+        return self.ctx.make_mpf(mpi_abs(self._mpi_, self.ctx.prec))
+    def __pos__(self):
+        return self.ctx.make_mpf(mpi_pos(self._mpi_, self.ctx.prec))
+    def __neg__(self):
+        return self.ctx.make_mpf(mpi_neg(self._mpi_, self.ctx.prec))
+
+    def ae(s, t, rel_eps=None, abs_eps=None):
+        return s.ctx.almosteq(s, t, rel_eps, abs_eps)
+
+class ivmpc(object):
+
+    def __new__(cls, re=0, im=0):
+        re = cls.ctx.convert(re)
+        im = cls.ctx.convert(im)
+        y = new(cls)
+        y._mpci_ = re._mpi_, im._mpi_
+        return y
+
+    def __hash__(self):
+        (a, b), (c,d) = self._mpci_
+        if a == b and c == d:
+            return mpc_hash((a, c))
+        else:
+            return hash(self._mpci_)
+
+    def __repr__(s):
+        if s.ctx.pretty:
+            return str(s)
+        return "iv.mpc(%s, %s)" % (repr(s.real), repr(s.imag))
+
+    def __str__(s):
+        return "(%s + %s*j)" % (str(s.real), str(s.imag))
+
+    @property
+    def a(self):
+        (a, b), (c,d) = self._mpci_
+        return self.ctx.make_mpf((a, a))
+
+    @property
+    def b(self):
+        (a, b), (c,d) = self._mpci_
+        return self.ctx.make_mpf((b, b))
+
+    @property
+    def c(self):
+        (a, b), (c,d) = self._mpci_
+        return self.ctx.make_mpf((c, c))
+
+    @property
+    def d(self):
+        (a, b), (c,d) = self._mpci_
+        return self.ctx.make_mpf((d, d))
+
+    @property
+    def real(s):
+        return s.ctx.make_mpf(s._mpci_[0])
+
+    @property
+    def imag(s):
+        return s.ctx.make_mpf(s._mpci_[1])
+
+    def conjugate(s):
+        a, b = s._mpci_
+        return s.ctx.make_mpc((a, mpf_neg(b)))
+
+    def overlap(s, t):
+        t = s.ctx.convert(t)
+        real_overlap = (s.a <= t.a <= s.b) or (s.a <= t.b <= s.b) or (t.a <= s.a <= t.b) or (t.a <= s.b <= t.b)
+        imag_overlap = (s.c <= t.c <= s.d) or (s.c <= t.d <= s.d) or (t.c <= s.c <= t.d) or (t.c <= s.d <= t.d)
+        return real_overlap and imag_overlap
+
+    def __contains__(s, t):
+        t = s.ctx.convert(t)
+        return t.real in s.real and t.imag in s.imag
+
+    def _compare(s, t, ne=False):
+        if not isinstance(t, s.ctx._types):
+            try:
+                t = s.ctx.convert(t)
+            except:
+                return NotImplemented
+        if hasattr(t, '_mpi_'):
+            tval = t._mpi_, mpi_zero
+        elif hasattr(t, '_mpci_'):
+            tval = t._mpci_
+        if ne:
+            return s._mpci_ != tval
+        return s._mpci_ == tval
+
+    def __eq__(s, t): return s._compare(t)
+    def __ne__(s, t): return s._compare(t, True)
+
+    def __lt__(s, t): raise TypeError("complex intervals cannot be ordered")
+    __le__ = __gt__ = __ge__ = __lt__
+
+    def __neg__(s): return s.ctx.make_mpc(mpci_neg(s._mpci_, s.ctx.prec))
+    def __pos__(s): return s.ctx.make_mpc(mpci_pos(s._mpci_, s.ctx.prec))
+    def __abs__(s): return s.ctx.make_mpf(mpci_abs(s._mpci_, s.ctx.prec))
+
+    def ae(s, t, rel_eps=None, abs_eps=None):
+        return s.ctx.almosteq(s, t, rel_eps, abs_eps)
+
+def _binary_op(f_real, f_complex):
+    def g_complex(ctx, sval, tval):
+        return ctx.make_mpc(f_complex(sval, tval, ctx.prec))
+    def g_real(ctx, sval, tval):
+        try:
+            return ctx.make_mpf(f_real(sval, tval, ctx.prec))
+        except ComplexResult:
+            sval = (sval, mpi_zero)
+            tval = (tval, mpi_zero)
+            return g_complex(ctx, sval, tval)
+    def lop_real(s, t):
+        if isinstance(t, _matrix): return NotImplemented
+        ctx = s.ctx
+        if not isinstance(t, ctx._types): t = ctx.convert(t)
+        if hasattr(t, "_mpi_"): return g_real(ctx, s._mpi_, t._mpi_)
+        if hasattr(t, "_mpci_"): return g_complex(ctx, (s._mpi_, mpi_zero), t._mpci_)
+        return NotImplemented
+    def rop_real(s, t):
+        ctx = s.ctx
+        if not isinstance(t, ctx._types): t = ctx.convert(t)
+        if hasattr(t, "_mpi_"): return g_real(ctx, t._mpi_, s._mpi_)
+        if hasattr(t, "_mpci_"): return g_complex(ctx, t._mpci_, (s._mpi_, mpi_zero))
+        return NotImplemented
+    def lop_complex(s, t):
+        if isinstance(t, _matrix): return NotImplemented
+        ctx = s.ctx
+        if not isinstance(t, s.ctx._types):
+            try:
+                t = s.ctx.convert(t)
+            except (ValueError, TypeError):
+                return NotImplemented
+        return g_complex(ctx, s._mpci_, t._mpci_)
+    def rop_complex(s, t):
+        ctx = s.ctx
+        if not isinstance(t, s.ctx._types):
+            t = s.ctx.convert(t)
+        return g_complex(ctx, t._mpci_, s._mpci_)
+    return lop_real, rop_real, lop_complex, rop_complex
+
+ivmpf.__add__, ivmpf.__radd__, ivmpc.__add__, ivmpc.__radd__ = _binary_op(mpi_add, mpci_add)
+ivmpf.__sub__, ivmpf.__rsub__, ivmpc.__sub__, ivmpc.__rsub__ = _binary_op(mpi_sub, mpci_sub)
+ivmpf.__mul__, ivmpf.__rmul__, ivmpc.__mul__, ivmpc.__rmul__ = _binary_op(mpi_mul, mpci_mul)
+ivmpf.__div__, ivmpf.__rdiv__, ivmpc.__div__, ivmpc.__rdiv__ = _binary_op(mpi_div, mpci_div)
+ivmpf.__pow__, ivmpf.__rpow__, ivmpc.__pow__, ivmpc.__rpow__ = _binary_op(mpi_pow, mpci_pow)
+
+ivmpf.__truediv__ = ivmpf.__div__; ivmpf.__rtruediv__ = ivmpf.__rdiv__
+ivmpc.__truediv__ = ivmpc.__div__; ivmpc.__rtruediv__ = ivmpc.__rdiv__
+
+class ivmpf_constant(ivmpf):
+    def __new__(cls, f):
+        self = new(cls)
+        self._f = f
+        return self
+    def _get_mpi_(self):
+        prec = self.ctx._prec[0]
+        a = self._f(prec, round_floor)
+        b = self._f(prec, round_ceiling)
+        return a, b
+    _mpi_ = property(_get_mpi_)
+
+class MPIntervalContext(StandardBaseContext):
+
+    def __init__(ctx):
+        ctx.mpf = type('ivmpf', (ivmpf,), {})
+        ctx.mpc = type('ivmpc', (ivmpc,), {})
+        ctx._types = (ctx.mpf, ctx.mpc)
+        ctx._constant = type('ivmpf_constant', (ivmpf_constant,), {})
+        ctx._prec = [53]
+        ctx._set_prec(53)
+        ctx._constant._ctxdata = ctx.mpf._ctxdata = ctx.mpc._ctxdata = [ctx.mpf, new, ctx._prec]
+        ctx._constant.ctx = ctx.mpf.ctx = ctx.mpc.ctx = ctx
+        ctx.pretty = False
+        StandardBaseContext.__init__(ctx)
+        ctx._init_builtins()
+
+    def _mpi(ctx, a, b=None):
+        if b is None:
+            return ctx.mpf(a)
+        return ctx.mpf((a,b))
+
+    def _init_builtins(ctx):
+        ctx.one = ctx.mpf(1)
+        ctx.zero = ctx.mpf(0)
+        ctx.inf = ctx.mpf('inf')
+        ctx.ninf = -ctx.inf
+        ctx.nan = ctx.mpf('nan')
+        ctx.j = ctx.mpc(0,1)
+        ctx.exp = ctx._wrap_mpi_function(libmp.mpi_exp, libmp.mpci_exp)
+        ctx.sqrt = ctx._wrap_mpi_function(libmp.mpi_sqrt)
+        ctx.ln = ctx._wrap_mpi_function(libmp.mpi_log, libmp.mpci_log)
+        ctx.cos = ctx._wrap_mpi_function(libmp.mpi_cos, libmp.mpci_cos)
+        ctx.sin = ctx._wrap_mpi_function(libmp.mpi_sin, libmp.mpci_sin)
+        ctx.tan = ctx._wrap_mpi_function(libmp.mpi_tan)
+        ctx.gamma = ctx._wrap_mpi_function(libmp.mpi_gamma, libmp.mpci_gamma)
+        ctx.loggamma = ctx._wrap_mpi_function(libmp.mpi_loggamma, libmp.mpci_loggamma)
+        ctx.rgamma = ctx._wrap_mpi_function(libmp.mpi_rgamma, libmp.mpci_rgamma)
+        ctx.factorial = ctx._wrap_mpi_function(libmp.mpi_factorial, libmp.mpci_factorial)
+        ctx.fac = ctx.factorial
+
+        ctx.eps = ctx._constant(lambda prec, rnd: (0, MPZ_ONE, 1-prec, 1))
+        ctx.pi = ctx._constant(libmp.mpf_pi)
+        ctx.e = ctx._constant(libmp.mpf_e)
+        ctx.ln2 = ctx._constant(libmp.mpf_ln2)
+        ctx.ln10 = ctx._constant(libmp.mpf_ln10)
+        ctx.phi = ctx._constant(libmp.mpf_phi)
+        ctx.euler = ctx._constant(libmp.mpf_euler)
+        ctx.catalan = ctx._constant(libmp.mpf_catalan)
+        ctx.glaisher = ctx._constant(libmp.mpf_glaisher)
+        ctx.khinchin = ctx._constant(libmp.mpf_khinchin)
+        ctx.twinprime = ctx._constant(libmp.mpf_twinprime)
+
+    def _wrap_mpi_function(ctx, f_real, f_complex=None):
+        def g(x, **kwargs):
+            if kwargs:
+                prec = kwargs.get('prec', ctx._prec[0])
+            else:
+                prec = ctx._prec[0]
+            x = ctx.convert(x)
+            if hasattr(x, "_mpi_"):
+                return ctx.make_mpf(f_real(x._mpi_, prec))
+            if hasattr(x, "_mpci_"):
+                return ctx.make_mpc(f_complex(x._mpci_, prec))
+            raise ValueError
+        return g
+
+    @classmethod
+    def _wrap_specfun(cls, name, f, wrap):
+        if wrap:
+            def f_wrapped(ctx, *args, **kwargs):
+                convert = ctx.convert
+                args = [convert(a) for a in args]
+                prec = ctx.prec
+                try:
+                    ctx.prec += 10
+                    retval = f(ctx, *args, **kwargs)
+                finally:
+                    ctx.prec = prec
+                return +retval
+        else:
+            f_wrapped = f
+        setattr(cls, name, f_wrapped)
+
+    def _set_prec(ctx, n):
+        ctx._prec[0] = max(1, int(n))
+        ctx._dps = prec_to_dps(n)
+
+    def _set_dps(ctx, n):
+        ctx._prec[0] = dps_to_prec(n)
+        ctx._dps = max(1, int(n))
+
+    prec = property(lambda ctx: ctx._prec[0], _set_prec)
+    dps = property(lambda ctx: ctx._dps, _set_dps)
+
+    def make_mpf(ctx, v):
+        a = new(ctx.mpf)
+        a._mpi_ = v
+        return a
+
+    def make_mpc(ctx, v):
+        a = new(ctx.mpc)
+        a._mpci_ = v
+        return a
+
+    def _mpq(ctx, pq):
+        p, q = pq
+        a = libmp.from_rational(p, q, ctx.prec, round_floor)
+        b = libmp.from_rational(p, q, ctx.prec, round_ceiling)
+        return ctx.make_mpf((a, b))
+
+    def convert(ctx, x):
+        if isinstance(x, (ctx.mpf, ctx.mpc)):
+            return x
+        if isinstance(x, ctx._constant):
+            return +x
+        if isinstance(x, complex) or hasattr(x, "_mpc_"):
+            re = ctx.convert(x.real)
+            im = ctx.convert(x.imag)
+            return ctx.mpc(re,im)
+        if isinstance(x, basestring):
+            v = mpi_from_str(x, ctx.prec)
+            return ctx.make_mpf(v)
+        if hasattr(x, "_mpi_"):
+            a, b = x._mpi_
+        else:
+            try:
+                a, b = x
+            except (TypeError, ValueError):
+                a = b = x
+            if hasattr(a, "_mpi_"):
+                a = a._mpi_[0]
+            else:
+                a = convert_mpf_(a, ctx.prec, round_floor)
+            if hasattr(b, "_mpi_"):
+                b = b._mpi_[1]
+            else:
+                b = convert_mpf_(b, ctx.prec, round_ceiling)
+        if a == fnan or b == fnan:
+            a = fninf
+            b = finf
+        assert mpf_le(a, b), "endpoints must be properly ordered"
+        return ctx.make_mpf((a, b))
+
+    def nstr(ctx, x, n=5, **kwargs):
+        x = ctx.convert(x)
+        if hasattr(x, "_mpi_"):
+            return libmp.mpi_to_str(x._mpi_, n, **kwargs)
+        if hasattr(x, "_mpci_"):
+            re = libmp.mpi_to_str(x._mpci_[0], n, **kwargs)
+            im = libmp.mpi_to_str(x._mpci_[1], n, **kwargs)
+            return "(%s + %s*j)" % (re, im)
+
+    def mag(ctx, x):
+        x = ctx.convert(x)
+        if isinstance(x, ctx.mpc):
+            return max(ctx.mag(x.real), ctx.mag(x.imag)) + 1
+        a, b = libmp.mpi_abs(x._mpi_)
+        sign, man, exp, bc = b
+        if man:
+            return exp+bc
+        if b == fzero:
+            return ctx.ninf
+        if b == fnan:
+            return ctx.nan
+        return ctx.inf
+
+    def isnan(ctx, x):
+        return False
+
+    def isinf(ctx, x):
+        return x == ctx.inf
+
+    def isint(ctx, x):
+        x = ctx.convert(x)
+        a, b = x._mpi_
+        if a == b:
+            sign, man, exp, bc = a
+            if man:
+                return exp >= 0
+            return a == fzero
+        return None
+
+    def ldexp(ctx, x, n):
+        a, b = ctx.convert(x)._mpi_
+        a = libmp.mpf_shift(a, n)
+        b = libmp.mpf_shift(b, n)
+        return ctx.make_mpf((a,b))
+
+    def absmin(ctx, x):
+        return abs(ctx.convert(x)).a
+
+    def absmax(ctx, x):
+        return abs(ctx.convert(x)).b
+
+    def atan2(ctx, y, x):
+        y = ctx.convert(y)._mpi_
+        x = ctx.convert(x)._mpi_
+        return ctx.make_mpf(libmp.mpi_atan2(y,x,ctx.prec))
+
+    def _convert_param(ctx, x):
+        if isinstance(x, libmp.int_types):
+            return x, 'Z'
+        if isinstance(x, tuple):
+            p, q = x
+            return (ctx.mpf(p) / ctx.mpf(q), 'R')
+        x = ctx.convert(x)
+        if isinstance(x, ctx.mpf):
+            return x, 'R'
+        if isinstance(x, ctx.mpc):
+            return x, 'C'
+        raise ValueError
+
+    def _is_real_type(ctx, z):
+        return isinstance(z, ctx.mpf) or isinstance(z, int_types)
+
+    def _is_complex_type(ctx, z):
+        return isinstance(z, ctx.mpc)
+
+    def hypsum(ctx, p, q, types, coeffs, z, maxterms=6000, **kwargs):
+        coeffs = list(coeffs)
+        num = range(p)
+        den = range(p,p+q)
+        #tol = ctx.eps
+        s = t = ctx.one
+        k = 0
+        while 1:
+            for i in num: t *= (coeffs[i]+k)
+            for i in den: t /= (coeffs[i]+k)
+            k += 1; t /= k; t *= z; s += t
+            if t == 0:
+                return s
+            #if abs(t) < tol:
+            #    return s
+            if k > maxterms:
+                raise ctx.NoConvergence
+
+
+# Register with "numbers" ABC
+#     We do not subclass, hence we do not use the @abstractmethod checks. While
+#     this is less invasive it may turn out that we do not actually support
+#     parts of the expected interfaces.  See
+#     http://docs.python.org/2/library/numbers.html for list of abstract
+#     methods.
+try:
+    import numbers
+    numbers.Complex.register(ivmpc)
+    numbers.Real.register(ivmpf)
+except ImportError:
+    pass

lib/python3.11/site-packages/mpmath/ctx_mp.py

@@ -0,0 +1,1339 @@
+"""
+This module defines the mpf, mpc classes, and standard functions for
+operating with them.
+"""
+__docformat__ = 'plaintext'
+
+import functools
+
+import re
+
+from .ctx_base import StandardBaseContext
+
+from .libmp.backend import basestring, BACKEND
+
+from . import libmp
+
+from .libmp import (MPZ, MPZ_ZERO, MPZ_ONE, int_types, repr_dps,
+    round_floor, round_ceiling, dps_to_prec, round_nearest, prec_to_dps,
+    ComplexResult, to_pickable, from_pickable, normalize,
+    from_int, from_float, from_str, to_int, to_float, to_str,
+    from_rational, from_man_exp,
+    fone, fzero, finf, fninf, fnan,
+    mpf_abs, mpf_pos, mpf_neg, mpf_add, mpf_sub, mpf_mul, mpf_mul_int,
+    mpf_div, mpf_rdiv_int, mpf_pow_int, mpf_mod,
+    mpf_eq, mpf_cmp, mpf_lt, mpf_gt, mpf_le, mpf_ge,
+    mpf_hash, mpf_rand,
+    mpf_sum,
+    bitcount, to_fixed,
+    mpc_to_str,
+    mpc_to_complex, mpc_hash, mpc_pos, mpc_is_nonzero, mpc_neg, mpc_conjugate,
+    mpc_abs, mpc_add, mpc_add_mpf, mpc_sub, mpc_sub_mpf, mpc_mul, mpc_mul_mpf,
+    mpc_mul_int, mpc_div, mpc_div_mpf, mpc_pow, mpc_pow_mpf, mpc_pow_int,
+    mpc_mpf_div,
+    mpf_pow,
+    mpf_pi, mpf_degree, mpf_e, mpf_phi, mpf_ln2, mpf_ln10,
+    mpf_euler, mpf_catalan, mpf_apery, mpf_khinchin,
+    mpf_glaisher, mpf_twinprime, mpf_mertens,
+    int_types)
+
+from . import function_docs
+from . import rational
+
+new = object.__new__
+
+get_complex = re.compile(r'^\(?(?P<re>[\+\-]?\d*(\.\d*)?(e[\+\-]?\d+)?)??'
+                         r'(?P<im>[\+\-]?\d*(\.\d*)?(e[\+\-]?\d+)?j)?\)?$')
+
+if BACKEND == 'sage':
+    from sage.libs.mpmath.ext_main import Context as BaseMPContext
+    # pickle hack
+    import sage.libs.mpmath.ext_main as _mpf_module
+else:
+    from .ctx_mp_python import PythonMPContext as BaseMPContext
+    from . import ctx_mp_python as _mpf_module
+
+from .ctx_mp_python import _mpf, _mpc, mpnumeric
+
+class MPContext(BaseMPContext, StandardBaseContext):
+    """
+    Context for multiprecision arithmetic with a global precision.
+    """
+
+    def __init__(ctx):
+        BaseMPContext.__init__(ctx)
+        ctx.trap_complex = False
+        ctx.pretty = False
+        ctx.types = [ctx.mpf, ctx.mpc, ctx.constant]
+        ctx._mpq = rational.mpq
+        ctx.default()
+        StandardBaseContext.__init__(ctx)
+
+        ctx.mpq = rational.mpq
+        ctx.init_builtins()
+
+        ctx.hyp_summators = {}
+
+        ctx._init_aliases()
+
+        # XXX: automate
+        try:
+            ctx.bernoulli.im_func.func_doc = function_docs.bernoulli
+            ctx.primepi.im_func.func_doc = function_docs.primepi
+            ctx.psi.im_func.func_doc = function_docs.psi
+            ctx.atan2.im_func.func_doc = function_docs.atan2
+        except AttributeError:
+            # python 3
+            ctx.bernoulli.__func__.func_doc = function_docs.bernoulli
+            ctx.primepi.__func__.func_doc = function_docs.primepi
+            ctx.psi.__func__.func_doc = function_docs.psi
+            ctx.atan2.__func__.func_doc = function_docs.atan2
+
+        ctx.digamma.func_doc = function_docs.digamma
+        ctx.cospi.func_doc = function_docs.cospi
+        ctx.sinpi.func_doc = function_docs.sinpi
+
+    def init_builtins(ctx):
+
+        mpf = ctx.mpf
+        mpc = ctx.mpc
+
+        # Exact constants
+        ctx.one = ctx.make_mpf(fone)
+        ctx.zero = ctx.make_mpf(fzero)
+        ctx.j = ctx.make_mpc((fzero,fone))
+        ctx.inf = ctx.make_mpf(finf)
+        ctx.ninf = ctx.make_mpf(fninf)
+        ctx.nan = ctx.make_mpf(fnan)
+
+        eps = ctx.constant(lambda prec, rnd: (0, MPZ_ONE, 1-prec, 1),
+            "epsilon of working precision", "eps")
+        ctx.eps = eps
+
+        # Approximate constants
+        ctx.pi = ctx.constant(mpf_pi, "pi", "pi")
+        ctx.ln2 = ctx.constant(mpf_ln2, "ln(2)", "ln2")
+        ctx.ln10 = ctx.constant(mpf_ln10, "ln(10)", "ln10")
+        ctx.phi = ctx.constant(mpf_phi, "Golden ratio phi", "phi")
+        ctx.e = ctx.constant(mpf_e, "e = exp(1)", "e")
+        ctx.euler = ctx.constant(mpf_euler, "Euler's constant", "euler")
+        ctx.catalan = ctx.constant(mpf_catalan, "Catalan's constant", "catalan")
+        ctx.khinchin = ctx.constant(mpf_khinchin, "Khinchin's constant", "khinchin")
+        ctx.glaisher = ctx.constant(mpf_glaisher, "Glaisher's constant", "glaisher")
+        ctx.apery = ctx.constant(mpf_apery, "Apery's constant", "apery")
+        ctx.degree = ctx.constant(mpf_degree, "1 deg = pi / 180", "degree")
+        ctx.twinprime = ctx.constant(mpf_twinprime, "Twin prime constant", "twinprime")
+        ctx.mertens = ctx.constant(mpf_mertens, "Mertens' constant", "mertens")
+
+        # Standard functions
+        ctx.sqrt = ctx._wrap_libmp_function(libmp.mpf_sqrt, libmp.mpc_sqrt)
+        ctx.cbrt = ctx._wrap_libmp_function(libmp.mpf_cbrt, libmp.mpc_cbrt)
+        ctx.ln = ctx._wrap_libmp_function(libmp.mpf_log, libmp.mpc_log)
+        ctx.atan = ctx._wrap_libmp_function(libmp.mpf_atan, libmp.mpc_atan)
+        ctx.exp = ctx._wrap_libmp_function(libmp.mpf_exp, libmp.mpc_exp)
+        ctx.expj = ctx._wrap_libmp_function(libmp.mpf_expj, libmp.mpc_expj)
+        ctx.expjpi = ctx._wrap_libmp_function(libmp.mpf_expjpi, libmp.mpc_expjpi)
+        ctx.sin = ctx._wrap_libmp_function(libmp.mpf_sin, libmp.mpc_sin)
+        ctx.cos = ctx._wrap_libmp_function(libmp.mpf_cos, libmp.mpc_cos)
+        ctx.tan = ctx._wrap_libmp_function(libmp.mpf_tan, libmp.mpc_tan)
+        ctx.sinh = ctx._wrap_libmp_function(libmp.mpf_sinh, libmp.mpc_sinh)
+        ctx.cosh = ctx._wrap_libmp_function(libmp.mpf_cosh, libmp.mpc_cosh)
+        ctx.tanh = ctx._wrap_libmp_function(libmp.mpf_tanh, libmp.mpc_tanh)
+        ctx.asin = ctx._wrap_libmp_function(libmp.mpf_asin, libmp.mpc_asin)
+        ctx.acos = ctx._wrap_libmp_function(libmp.mpf_acos, libmp.mpc_acos)
+        ctx.atan = ctx._wrap_libmp_function(libmp.mpf_atan, libmp.mpc_atan)
+        ctx.asinh = ctx._wrap_libmp_function(libmp.mpf_asinh, libmp.mpc_asinh)
+        ctx.acosh = ctx._wrap_libmp_function(libmp.mpf_acosh, libmp.mpc_acosh)
+        ctx.atanh = ctx._wrap_libmp_function(libmp.mpf_atanh, libmp.mpc_atanh)
+        ctx.sinpi = ctx._wrap_libmp_function(libmp.mpf_sin_pi, libmp.mpc_sin_pi)
+        ctx.cospi = ctx._wrap_libmp_function(libmp.mpf_cos_pi, libmp.mpc_cos_pi)
+        ctx.floor = ctx._wrap_libmp_function(libmp.mpf_floor, libmp.mpc_floor)
+        ctx.ceil = ctx._wrap_libmp_function(libmp.mpf_ceil, libmp.mpc_ceil)
+        ctx.nint = ctx._wrap_libmp_function(libmp.mpf_nint, libmp.mpc_nint)
+        ctx.frac = ctx._wrap_libmp_function(libmp.mpf_frac, libmp.mpc_frac)
+        ctx.fib = ctx.fibonacci = ctx._wrap_libmp_function(libmp.mpf_fibonacci, libmp.mpc_fibonacci)
+
+        ctx.gamma = ctx._wrap_libmp_function(libmp.mpf_gamma, libmp.mpc_gamma)
+        ctx.rgamma = ctx._wrap_libmp_function(libmp.mpf_rgamma, libmp.mpc_rgamma)
+        ctx.loggamma = ctx._wrap_libmp_function(libmp.mpf_loggamma, libmp.mpc_loggamma)
+        ctx.fac = ctx.factorial = ctx._wrap_libmp_function(libmp.mpf_factorial, libmp.mpc_factorial)
+
+        ctx.digamma = ctx._wrap_libmp_function(libmp.mpf_psi0, libmp.mpc_psi0)
+        ctx.harmonic = ctx._wrap_libmp_function(libmp.mpf_harmonic, libmp.mpc_harmonic)
+        ctx.ei = ctx._wrap_libmp_function(libmp.mpf_ei, libmp.mpc_ei)
+        ctx.e1 = ctx._wrap_libmp_function(libmp.mpf_e1, libmp.mpc_e1)
+        ctx._ci = ctx._wrap_libmp_function(libmp.mpf_ci, libmp.mpc_ci)
+        ctx._si = ctx._wrap_libmp_function(libmp.mpf_si, libmp.mpc_si)
+        ctx.ellipk = ctx._wrap_libmp_function(libmp.mpf_ellipk, libmp.mpc_ellipk)
+        ctx._ellipe = ctx._wrap_libmp_function(libmp.mpf_ellipe, libmp.mpc_ellipe)
+        ctx.agm1 = ctx._wrap_libmp_function(libmp.mpf_agm1, libmp.mpc_agm1)
+        ctx._erf = ctx._wrap_libmp_function(libmp.mpf_erf, None)
+        ctx._erfc = ctx._wrap_libmp_function(libmp.mpf_erfc, None)
+        ctx._zeta = ctx._wrap_libmp_function(libmp.mpf_zeta, libmp.mpc_zeta)
+        ctx._altzeta = ctx._wrap_libmp_function(libmp.mpf_altzeta, libmp.mpc_altzeta)
+
+        # Faster versions
+        ctx.sqrt = getattr(ctx, "_sage_sqrt", ctx.sqrt)
+        ctx.exp = getattr(ctx, "_sage_exp", ctx.exp)
+        ctx.ln = getattr(ctx, "_sage_ln", ctx.ln)
+        ctx.cos = getattr(ctx, "_sage_cos", ctx.cos)
+        ctx.sin = getattr(ctx, "_sage_sin", ctx.sin)
+
+    def to_fixed(ctx, x, prec):
+        return x.to_fixed(prec)
+
+    def hypot(ctx, x, y):
+        r"""
+        Computes the Euclidean norm of the vector `(x, y)`, equal
+        to `\sqrt{x^2 + y^2}`. Both `x` and `y` must be real."""
+        x = ctx.convert(x)
+        y = ctx.convert(y)
+        return ctx.make_mpf(libmp.mpf_hypot(x._mpf_, y._mpf_, *ctx._prec_rounding))
+
+    def _gamma_upper_int(ctx, n, z):
+        n = int(ctx._re(n))
+        if n == 0:
+            return ctx.e1(z)
+        if not hasattr(z, '_mpf_'):
+            raise NotImplementedError
+        prec, rounding = ctx._prec_rounding
+        real, imag = libmp.mpf_expint(n, z._mpf_, prec, rounding, gamma=True)
+        if imag is None:
+            return ctx.make_mpf(real)
+        else:
+            return ctx.make_mpc((real, imag))
+
+    def _expint_int(ctx, n, z):
+        n = int(n)
+        if n == 1:
+            return ctx.e1(z)
+        if not hasattr(z, '_mpf_'):
+            raise NotImplementedError
+        prec, rounding = ctx._prec_rounding
+        real, imag = libmp.mpf_expint(n, z._mpf_, prec, rounding)
+        if imag is None:
+            return ctx.make_mpf(real)
+        else:
+            return ctx.make_mpc((real, imag))
+
+    def _nthroot(ctx, x, n):
+        if hasattr(x, '_mpf_'):
+            try:
+                return ctx.make_mpf(libmp.mpf_nthroot(x._mpf_, n, *ctx._prec_rounding))
+            except ComplexResult:
+                if ctx.trap_complex:
+                    raise
+                x = (x._mpf_, libmp.fzero)
+        else:
+            x = x._mpc_
+        return ctx.make_mpc(libmp.mpc_nthroot(x, n, *ctx._prec_rounding))
+
+    def _besselj(ctx, n, z):
+        prec, rounding = ctx._prec_rounding
+        if hasattr(z, '_mpf_'):
+            return ctx.make_mpf(libmp.mpf_besseljn(n, z._mpf_, prec, rounding))
+        elif hasattr(z, '_mpc_'):
+            return ctx.make_mpc(libmp.mpc_besseljn(n, z._mpc_, prec, rounding))
+
+    def _agm(ctx, a, b=1):
+        prec, rounding = ctx._prec_rounding
+        if hasattr(a, '_mpf_') and hasattr(b, '_mpf_'):
+            try:
+                v = libmp.mpf_agm(a._mpf_, b._mpf_, prec, rounding)
+                return ctx.make_mpf(v)
+            except ComplexResult:
+                pass
+        if hasattr(a, '_mpf_'): a = (a._mpf_, libmp.fzero)
+        else: a = a._mpc_
+        if hasattr(b, '_mpf_'): b = (b._mpf_, libmp.fzero)
+        else: b = b._mpc_
+        return ctx.make_mpc(libmp.mpc_agm(a, b, prec, rounding))
+
+    def bernoulli(ctx, n):
+        return ctx.make_mpf(libmp.mpf_bernoulli(int(n), *ctx._prec_rounding))
+
+    def _zeta_int(ctx, n):
+        return ctx.make_mpf(libmp.mpf_zeta_int(int(n), *ctx._prec_rounding))
+
+    def atan2(ctx, y, x):
+        x = ctx.convert(x)
+        y = ctx.convert(y)
+        return ctx.make_mpf(libmp.mpf_atan2(y._mpf_, x._mpf_, *ctx._prec_rounding))
+
+    def psi(ctx, m, z):
+        z = ctx.convert(z)
+        m = int(m)
+        if ctx._is_real_type(z):
+            return ctx.make_mpf(libmp.mpf_psi(m, z._mpf_, *ctx._prec_rounding))
+        else:
+            return ctx.make_mpc(libmp.mpc_psi(m, z._mpc_, *ctx._prec_rounding))
+
+    def cos_sin(ctx, x, **kwargs):
+        if type(x) not in ctx.types:
+            x = ctx.convert(x)
+        prec, rounding = ctx._parse_prec(kwargs)
+        if hasattr(x, '_mpf_'):
+            c, s = libmp.mpf_cos_sin(x._mpf_, prec, rounding)
+            return ctx.make_mpf(c), ctx.make_mpf(s)
+        elif hasattr(x, '_mpc_'):
+            c, s = libmp.mpc_cos_sin(x._mpc_, prec, rounding)
+            return ctx.make_mpc(c), ctx.make_mpc(s)
+        else:
+            return ctx.cos(x, **kwargs), ctx.sin(x, **kwargs)
+
+    def cospi_sinpi(ctx, x, **kwargs):
+        if type(x) not in ctx.types:
+            x = ctx.convert(x)
+        prec, rounding = ctx._parse_prec(kwargs)
+        if hasattr(x, '_mpf_'):
+            c, s = libmp.mpf_cos_sin_pi(x._mpf_, prec, rounding)
+            return ctx.make_mpf(c), ctx.make_mpf(s)
+        elif hasattr(x, '_mpc_'):
+            c, s = libmp.mpc_cos_sin_pi(x._mpc_, prec, rounding)
+            return ctx.make_mpc(c), ctx.make_mpc(s)
+        else:
+            return ctx.cos(x, **kwargs), ctx.sin(x, **kwargs)
+
+    def clone(ctx):
+        """
+        Create a copy of the context, with the same working precision.
+        """
+        a = ctx.__class__()
+        a.prec = ctx.prec
+        return a
+
+    # Several helper methods
+    # TODO: add more of these, make consistent, write docstrings, ...
+
+    def _is_real_type(ctx, x):
+        if hasattr(x, '_mpc_') or type(x) is complex:
+            return False
+        return True
+
+    def _is_complex_type(ctx, x):
+        if hasattr(x, '_mpc_') or type(x) is complex:
+            return True
+        return False
+
+    def isnan(ctx, x):
+        """
+        Return *True* if *x* is a NaN (not-a-number), or for a complex
+        number, whether either the real or complex part is NaN;
+        otherwise return *False*::
+
+            >>> from mpmath import *
+            >>> isnan(3.14)
+            False
+            >>> isnan(nan)
+            True
+            >>> isnan(mpc(3.14,2.72))
+            False
+            >>> isnan(mpc(3.14,nan))
+            True
+
+        """
+        if hasattr(x, "_mpf_"):
+            return x._mpf_ == fnan
+        if hasattr(x, "_mpc_"):
+            return fnan in x._mpc_
+        if isinstance(x, int_types) or isinstance(x, rational.mpq):
+            return False
+        x = ctx.convert(x)
+        if hasattr(x, '_mpf_') or hasattr(x, '_mpc_'):
+            return ctx.isnan(x)
+        raise TypeError("isnan() needs a number as input")
+
+    def isfinite(ctx, x):
+        """
+        Return *True* if *x* is a finite number, i.e. neither
+        an infinity or a NaN.
+
+            >>> from mpmath import *
+            >>> isfinite(inf)
+            False
+            >>> isfinite(-inf)
+            False
+            >>> isfinite(3)
+            True
+            >>> isfinite(nan)
+            False
+            >>> isfinite(3+4j)
+            True
+            >>> isfinite(mpc(3,inf))
+            False
+            >>> isfinite(mpc(nan,3))
+            False
+
+        """
+        if ctx.isinf(x) or ctx.isnan(x):
+            return False
+        return True
+
+    def isnpint(ctx, x):
+        """
+        Determine if *x* is a nonpositive integer.
+        """
+        if not x:
+            return True
+        if hasattr(x, '_mpf_'):
+            sign, man, exp, bc = x._mpf_
+            return sign and exp >= 0
+        if hasattr(x, '_mpc_'):
+            return not x.imag and ctx.isnpint(x.real)
+        if type(x) in int_types:
+            return x <= 0
+        if isinstance(x, ctx.mpq):
+            p, q = x._mpq_
+            if not p:
+                return True
+            return q == 1 and p <= 0
+        return ctx.isnpint(ctx.convert(x))
+
+    def __str__(ctx):
+        lines = ["Mpmath settings:",
+            ("  mp.prec = %s" % ctx.prec).ljust(30) + "[default: 53]",
+            ("  mp.dps = %s" % ctx.dps).ljust(30) + "[default: 15]",
+            ("  mp.trap_complex = %s" % ctx.trap_complex).ljust(30) + "[default: False]",
+        ]
+        return "\n".join(lines)
+
+    @property
+    def _repr_digits(ctx):
+        return repr_dps(ctx._prec)
+
+    @property
+    def _str_digits(ctx):
+        return ctx._dps
+
+    def extraprec(ctx, n, normalize_output=False):
+        """
+        The block
+
+            with extraprec(n):
+                <code>
+
+        increases the precision n bits, executes <code>, and then
+        restores the precision.
+
+        extraprec(n)(f) returns a decorated version of the function f
+        that increases the working precision by n bits before execution,
+        and restores the parent precision afterwards. With
+        normalize_output=True, it rounds the return value to the parent
+        precision.
+        """
+        return PrecisionManager(ctx, lambda p: p + n, None, normalize_output)
+
+    def extradps(ctx, n, normalize_output=False):
+        """
+        This function is analogous to extraprec (see documentation)
+        but changes the decimal precision instead of the number of bits.
+        """
+        return PrecisionManager(ctx, None, lambda d: d + n, normalize_output)
+
+    def workprec(ctx, n, normalize_output=False):
+        """
+        The block
+
+            with workprec(n):
+                <code>
+
+        sets the precision to n bits, executes <code>, and then restores
+        the precision.
+
+        workprec(n)(f) returns a decorated version of the function f
+        that sets the precision to n bits before execution,
+        and restores the precision afterwards. With normalize_output=True,
+        it rounds the return value to the parent precision.
+        """
+        return PrecisionManager(ctx, lambda p: n, None, normalize_output)
+
+    def workdps(ctx, n, normalize_output=False):
+        """
+        This function is analogous to workprec (see documentation)
+        but changes the decimal precision instead of the number of bits.
+        """
+        return PrecisionManager(ctx, None, lambda d: n, normalize_output)
+
+    def autoprec(ctx, f, maxprec=None, catch=(), verbose=False):
+        r"""
+        Return a wrapped copy of *f* that repeatedly evaluates *f*
+        with increasing precision until the result converges to the
+        full precision used at the point of the call.
+
+        This heuristically protects against rounding errors, at the cost of
+        roughly a 2x slowdown compared to manually setting the optimal
+        precision. This method can, however, easily be fooled if the results
+        from *f* depend "discontinuously" on the precision, for instance
+        if catastrophic cancellation can occur. Therefore, :func:`~mpmath.autoprec`
+        should be used judiciously.
+
+        **Examples**
+
+        Many functions are sensitive to perturbations of the input arguments.
+        If the arguments are decimal numbers, they may have to be converted
+        to binary at a much higher precision. If the amount of required
+        extra precision is unknown, :func:`~mpmath.autoprec` is convenient::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15
+            >>> mp.pretty = True
+            >>> besselj(5, 125 * 10**28)    # Exact input
+            -8.03284785591801e-17
+            >>> besselj(5, '1.25e30')   # Bad
+            7.12954868316652e-16
+            >>> autoprec(besselj)(5, '1.25e30')   # Good
+            -8.03284785591801e-17
+
+        The following fails to converge because `\sin(\pi) = 0` whereas all
+        finite-precision approximations of `\pi` give nonzero values::
+
+            >>> autoprec(sin)(pi) # doctest: +IGNORE_EXCEPTION_DETAIL
+            Traceback (most recent call last):
+              ...
+            NoConvergence: autoprec: prec increased to 2910 without convergence
+
+        As the following example shows, :func:`~mpmath.autoprec` can protect against
+        cancellation, but is fooled by too severe cancellation::
+
+            >>> x = 1e-10
+            >>> exp(x)-1; expm1(x); autoprec(lambda t: exp(t)-1)(x)
+            1.00000008274037e-10
+            1.00000000005e-10
+            1.00000000005e-10
+            >>> x = 1e-50
+            >>> exp(x)-1; expm1(x); autoprec(lambda t: exp(t)-1)(x)
+            0.0
+            1.0e-50
+            0.0
+
+        With *catch*, an exception or list of exceptions to intercept
+        may be specified. The raised exception is interpreted
+        as signaling insufficient precision. This permits, for example,
+        evaluating a function where a too low precision results in a
+        division by zero::
+
+            >>> f = lambda x: 1/(exp(x)-1)
+            >>> f(1e-30)
+            Traceback (most recent call last):
+              ...
+            ZeroDivisionError
+            >>> autoprec(f, catch=ZeroDivisionError)(1e-30)
+            1.0e+30
+
+
+        """
+        def f_autoprec_wrapped(*args, **kwargs):
+            prec = ctx.prec
+            if maxprec is None:
+                maxprec2 = ctx._default_hyper_maxprec(prec)
+            else:
+                maxprec2 = maxprec
+            try:
+                ctx.prec = prec + 10
+                try:
+                    v1 = f(*args, **kwargs)
+                except catch:
+                    v1 = ctx.nan
+                prec2 = prec + 20
+                while 1:
+                    ctx.prec = prec2
+                    try:
+                        v2 = f(*args, **kwargs)
+                    except catch:
+                        v2 = ctx.nan
+                    if v1 == v2:
+                        break
+                    err = ctx.mag(v2-v1) - ctx.mag(v2)
+                    if err < (-prec):
+                        break
+                    if verbose:
+                        print("autoprec: target=%s, prec=%s, accuracy=%s" \
+                            % (prec, prec2, -err))
+                    v1 = v2
+                    if prec2 >= maxprec2:
+                        raise ctx.NoConvergence(\
+                        "autoprec: prec increased to %i without convergence"\
+                        % prec2)
+                    prec2 += int(prec2*2)
+                    prec2 = min(prec2, maxprec2)
+            finally:
+                ctx.prec = prec
+            return +v2
+        return f_autoprec_wrapped
+
+    def nstr(ctx, x, n=6, **kwargs):
+        """
+        Convert an ``mpf`` or ``mpc`` to a decimal string literal with *n*
+        significant digits. The small default value for *n* is chosen to
+        make this function useful for printing collections of numbers
+        (lists, matrices, etc).
+
+        If *x* is a list or tuple, :func:`~mpmath.nstr` is applied recursively
+        to each element. For unrecognized classes, :func:`~mpmath.nstr`
+        simply returns ``str(x)``.
+
+        The companion function :func:`~mpmath.nprint` prints the result
+        instead of returning it.
+
+        The keyword arguments *strip_zeros*, *min_fixed*, *max_fixed*
+        and *show_zero_exponent* are forwarded to :func:`~mpmath.libmp.to_str`.
+
+        The number will be printed in fixed-point format if the position
+        of the leading digit is strictly between min_fixed
+        (default = min(-dps/3,-5)) and max_fixed (default = dps).
+
+        To force fixed-point format always, set min_fixed = -inf,
+        max_fixed = +inf. To force floating-point format, set
+        min_fixed >= max_fixed.
+
+            >>> from mpmath import *
+            >>> nstr([+pi, ldexp(1,-500)])
+            '[3.14159, 3.05494e-151]'
+            >>> nprint([+pi, ldexp(1,-500)])
+            [3.14159, 3.05494e-151]
+            >>> nstr(mpf("5e-10"), 5)
+            '5.0e-10'
+            >>> nstr(mpf("5e-10"), 5, strip_zeros=False)
+            '5.0000e-10'
+            >>> nstr(mpf("5e-10"), 5, strip_zeros=False, min_fixed=-11)
+            '0.00000000050000'
+            >>> nstr(mpf(0), 5, show_zero_exponent=True)
+            '0.0e+0'
+
+        """
+        if isinstance(x, list):
+            return "[%s]" % (", ".join(ctx.nstr(c, n, **kwargs) for c in x))
+        if isinstance(x, tuple):
+            return "(%s)" % (", ".join(ctx.nstr(c, n, **kwargs) for c in x))
+        if hasattr(x, '_mpf_'):
+            return to_str(x._mpf_, n, **kwargs)
+        if hasattr(x, '_mpc_'):
+            return "(" + mpc_to_str(x._mpc_, n, **kwargs)  + ")"
+        if isinstance(x, basestring):
+            return repr(x)
+        if isinstance(x, ctx.matrix):
+            return x.__nstr__(n, **kwargs)
+        return str(x)
+
+    def _convert_fallback(ctx, x, strings):
+        if strings and isinstance(x, basestring):
+            if 'j' in x.lower():
+                x = x.lower().replace(' ', '')
+                match = get_complex.match(x)
+                re = match.group('re')
+                if not re:
+                    re = 0
+                im = match.group('im').rstrip('j')
+                return ctx.mpc(ctx.convert(re), ctx.convert(im))
+        if hasattr(x, "_mpi_"):
+            a, b = x._mpi_
+            if a == b:
+                return ctx.make_mpf(a)
+            else:
+                raise ValueError("can only create mpf from zero-width interval")
+        raise TypeError("cannot create mpf from " + repr(x))
+
+    def mpmathify(ctx, *args, **kwargs):
+        return ctx.convert(*args, **kwargs)
+
+    def _parse_prec(ctx, kwargs):
+        if kwargs:
+            if kwargs.get('exact'):
+                return 0, 'f'
+            prec, rounding = ctx._prec_rounding
+            if 'rounding' in kwargs:
+                rounding = kwargs['rounding']
+            if 'prec' in kwargs:
+                prec = kwargs['prec']
+                if prec == ctx.inf:
+                    return 0, 'f'
+                else:
+                    prec = int(prec)
+            elif 'dps' in kwargs:
+                dps = kwargs['dps']
+                if dps == ctx.inf:
+                    return 0, 'f'
+                prec = dps_to_prec(dps)
+            return prec, rounding
+        return ctx._prec_rounding
+
+    _exact_overflow_msg = "the exact result does not fit in memory"
+
+    _hypsum_msg = """hypsum() failed to converge to the requested %i bits of accuracy
+using a working precision of %i bits. Try with a higher maxprec,
+maxterms, or set zeroprec."""
+
+    def hypsum(ctx, p, q, flags, coeffs, z, accurate_small=True, **kwargs):
+        if hasattr(z, "_mpf_"):
+            key = p, q, flags, 'R'
+            v = z._mpf_
+        elif hasattr(z, "_mpc_"):
+            key = p, q, flags, 'C'
+            v = z._mpc_
+        if key not in ctx.hyp_summators:
+            ctx.hyp_summators[key] = libmp.make_hyp_summator(key)[1]
+        summator = ctx.hyp_summators[key]
+        prec = ctx.prec
+        maxprec = kwargs.get('maxprec', ctx._default_hyper_maxprec(prec))
+        extraprec = 50
+        epsshift = 25
+        # Jumps in magnitude occur when parameters are close to negative
+        # integers. We must ensure that these terms are included in
+        # the sum and added accurately
+        magnitude_check = {}
+        max_total_jump = 0
+        for i, c in enumerate(coeffs):
+            if flags[i] == 'Z':
+                if i >= p and c <= 0:
+                    ok = False
+                    for ii, cc in enumerate(coeffs[:p]):
+                        # Note: c <= cc or c < cc, depending on convention
+                        if flags[ii] == 'Z' and cc <= 0 and c <= cc:
+                            ok = True
+                    if not ok:
+                        raise ZeroDivisionError("pole in hypergeometric series")
+                continue
+            n, d = ctx.nint_distance(c)
+            n = -int(n)
+            d = -d
+            if i >= p and n >= 0 and d > 4:
+                if n in magnitude_check:
+                    magnitude_check[n] += d
+                else:
+                    magnitude_check[n] = d
+                extraprec = max(extraprec, d - prec + 60)
+            max_total_jump += abs(d)
+        while 1:
+            if extraprec > maxprec:
+                raise ValueError(ctx._hypsum_msg % (prec, prec+extraprec))
+            wp = prec + extraprec
+            if magnitude_check:
+                mag_dict = dict((n,None) for n in magnitude_check)
+            else:
+                mag_dict = {}
+            zv, have_complex, magnitude = summator(coeffs, v, prec, wp, \
+                epsshift, mag_dict, **kwargs)
+            cancel = -magnitude
+            jumps_resolved = True
+            if extraprec < max_total_jump:
+                for n in mag_dict.values():
+                    if (n is None) or (n < prec):
+                        jumps_resolved = False
+                        break
+            accurate = (cancel < extraprec-25-5 or not accurate_small)
+            if jumps_resolved:
+                if accurate:
+                    break
+                # zero?
+                zeroprec = kwargs.get('zeroprec')
+                if zeroprec is not None:
+                    if cancel > zeroprec:
+                        if have_complex:
+                            return ctx.mpc(0)
+                        else:
+                            return ctx.zero
+
+            # Some near-singularities were not included, so increase
+            # precision and repeat until they are
+            extraprec *= 2
+            # Possible workaround for bad roundoff in fixed-point arithmetic
+            epsshift += 5
+            extraprec += 5
+
+        if type(zv) is tuple:
+            if have_complex:
+                return ctx.make_mpc(zv)
+            else:
+                return ctx.make_mpf(zv)
+        else:
+            return zv
+
+    def ldexp(ctx, x, n):
+        r"""
+        Computes `x 2^n` efficiently. No rounding is performed.
+        The argument `x` must be a real floating-point number (or
+        possible to convert into one) and `n` must be a Python ``int``.
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> ldexp(1, 10)
+            mpf('1024.0')
+            >>> ldexp(1, -3)
+            mpf('0.125')
+
+        """
+        x = ctx.convert(x)
+        return ctx.make_mpf(libmp.mpf_shift(x._mpf_, n))
+
+    def frexp(ctx, x):
+        r"""
+        Given a real number `x`, returns `(y, n)` with `y \in [0.5, 1)`,
+        `n` a Python integer, and such that `x = y 2^n`. No rounding is
+        performed.
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> frexp(7.5)
+            (mpf('0.9375'), 3)
+
+        """
+        x = ctx.convert(x)
+        y, n = libmp.mpf_frexp(x._mpf_)
+        return ctx.make_mpf(y), n
+
+    def fneg(ctx, x, **kwargs):
+        """
+        Negates the number *x*, giving a floating-point result, optionally
+        using a custom precision and rounding mode.
+
+        See the documentation of :func:`~mpmath.fadd` for a detailed description
+        of how to specify precision and rounding.
+
+        **Examples**
+
+        An mpmath number is returned::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fneg(2.5)
+            mpf('-2.5')
+            >>> fneg(-5+2j)
+            mpc(real='5.0', imag='-2.0')
+
+        Precise control over rounding is possible::
+
+            >>> x = fadd(2, 1e-100, exact=True)
+            >>> fneg(x)
+            mpf('-2.0')
+            >>> fneg(x, rounding='f')
+            mpf('-2.0000000000000004')
+
+        Negating with and without roundoff::
+
+            >>> n = 200000000000000000000001
+            >>> print(int(-mpf(n)))
+            -200000000000000016777216
+            >>> print(int(fneg(n)))
+            -200000000000000016777216
+            >>> print(int(fneg(n, prec=log(n,2)+1)))
+            -200000000000000000000001
+            >>> print(int(fneg(n, dps=log(n,10)+1)))
+            -200000000000000000000001
+            >>> print(int(fneg(n, prec=inf)))
+            -200000000000000000000001
+            >>> print(int(fneg(n, dps=inf)))
+            -200000000000000000000001
+            >>> print(int(fneg(n, exact=True)))
+            -200000000000000000000001
+
+        """
+        prec, rounding = ctx._parse_prec(kwargs)
+        x = ctx.convert(x)
+        if hasattr(x, '_mpf_'):
+            return ctx.make_mpf(mpf_neg(x._mpf_, prec, rounding))
+        if hasattr(x, '_mpc_'):
+            return ctx.make_mpc(mpc_neg(x._mpc_, prec, rounding))
+        raise ValueError("Arguments need to be mpf or mpc compatible numbers")
+
+    def fadd(ctx, x, y, **kwargs):
+        """
+        Adds the numbers *x* and *y*, giving a floating-point result,
+        optionally using a custom precision and rounding mode.
+
+        The default precision is the working precision of the context.
+        You can specify a custom precision in bits by passing the *prec* keyword
+        argument, or by providing an equivalent decimal precision with the *dps*
+        keyword argument. If the precision is set to ``+inf``, or if the flag
+        *exact=True* is passed, an exact addition with no rounding is performed.
+
+        When the precision is finite, the optional *rounding* keyword argument
+        specifies the direction of rounding. Valid options are ``'n'`` for
+        nearest (default), ``'f'`` for floor, ``'c'`` for ceiling, ``'d'``
+        for down, ``'u'`` for up.
+
+        **Examples**
+
+        Using :func:`~mpmath.fadd` with precision and rounding control::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fadd(2, 1e-20)
+            mpf('2.0')
+            >>> fadd(2, 1e-20, rounding='u')
+            mpf('2.0000000000000004')
+            >>> nprint(fadd(2, 1e-20, prec=100), 25)
+            2.00000000000000000001
+            >>> nprint(fadd(2, 1e-20, dps=15), 25)
+            2.0
+            >>> nprint(fadd(2, 1e-20, dps=25), 25)
+            2.00000000000000000001
+            >>> nprint(fadd(2, 1e-20, exact=True), 25)
+            2.00000000000000000001
+
+        Exact addition avoids cancellation errors, enforcing familiar laws
+        of numbers such as `x+y-x = y`, which don't hold in floating-point
+        arithmetic with finite precision::
+
+            >>> x, y = mpf(2), mpf('1e-1000')
+            >>> print(x + y - x)
+            0.0
+            >>> print(fadd(x, y, prec=inf) - x)
+            1.0e-1000
+            >>> print(fadd(x, y, exact=True) - x)
+            1.0e-1000
+
+        Exact addition can be inefficient and may be impossible to perform
+        with large magnitude differences::
+
+            >>> fadd(1, '1e-100000000000000000000', prec=inf)
+            Traceback (most recent call last):
+              ...
+            OverflowError: the exact result does not fit in memory
+
+        """
+        prec, rounding = ctx._parse_prec(kwargs)
+        x = ctx.convert(x)
+        y = ctx.convert(y)
+        try:
+            if hasattr(x, '_mpf_'):
+                if hasattr(y, '_mpf_'):
+                    return ctx.make_mpf(mpf_add(x._mpf_, y._mpf_, prec, rounding))
+                if hasattr(y, '_mpc_'):
+                    return ctx.make_mpc(mpc_add_mpf(y._mpc_, x._mpf_, prec, rounding))
+            if hasattr(x, '_mpc_'):
+                if hasattr(y, '_mpf_'):
+                    return ctx.make_mpc(mpc_add_mpf(x._mpc_, y._mpf_, prec, rounding))
+                if hasattr(y, '_mpc_'):
+                    return ctx.make_mpc(mpc_add(x._mpc_, y._mpc_, prec, rounding))
+        except (ValueError, OverflowError):
+            raise OverflowError(ctx._exact_overflow_msg)
+        raise ValueError("Arguments need to be mpf or mpc compatible numbers")
+
+    def fsub(ctx, x, y, **kwargs):
+        """
+        Subtracts the numbers *x* and *y*, giving a floating-point result,
+        optionally using a custom precision and rounding mode.
+
+        See the documentation of :func:`~mpmath.fadd` for a detailed description
+        of how to specify precision and rounding.
+
+        **Examples**
+
+        Using :func:`~mpmath.fsub` with precision and rounding control::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fsub(2, 1e-20)
+            mpf('2.0')
+            >>> fsub(2, 1e-20, rounding='d')
+            mpf('1.9999999999999998')
+            >>> nprint(fsub(2, 1e-20, prec=100), 25)
+            1.99999999999999999999
+            >>> nprint(fsub(2, 1e-20, dps=15), 25)
+            2.0
+            >>> nprint(fsub(2, 1e-20, dps=25), 25)
+            1.99999999999999999999
+            >>> nprint(fsub(2, 1e-20, exact=True), 25)
+            1.99999999999999999999
+
+        Exact subtraction avoids cancellation errors, enforcing familiar laws
+        of numbers such as `x-y+y = x`, which don't hold in floating-point
+        arithmetic with finite precision::
+
+            >>> x, y = mpf(2), mpf('1e1000')
+            >>> print(x - y + y)
+            0.0
+            >>> print(fsub(x, y, prec=inf) + y)
+            2.0
+            >>> print(fsub(x, y, exact=True) + y)
+            2.0
+
+        Exact addition can be inefficient and may be impossible to perform
+        with large magnitude differences::
+
+            >>> fsub(1, '1e-100000000000000000000', prec=inf)
+            Traceback (most recent call last):
+              ...
+            OverflowError: the exact result does not fit in memory
+
+        """
+        prec, rounding = ctx._parse_prec(kwargs)
+        x = ctx.convert(x)
+        y = ctx.convert(y)
+        try:
+            if hasattr(x, '_mpf_'):
+                if hasattr(y, '_mpf_'):
+                    return ctx.make_mpf(mpf_sub(x._mpf_, y._mpf_, prec, rounding))
+                if hasattr(y, '_mpc_'):
+                    return ctx.make_mpc(mpc_sub((x._mpf_, fzero), y._mpc_, prec, rounding))
+            if hasattr(x, '_mpc_'):
+                if hasattr(y, '_mpf_'):
+                    return ctx.make_mpc(mpc_sub_mpf(x._mpc_, y._mpf_, prec, rounding))
+                if hasattr(y, '_mpc_'):
+                    return ctx.make_mpc(mpc_sub(x._mpc_, y._mpc_, prec, rounding))
+        except (ValueError, OverflowError):
+            raise OverflowError(ctx._exact_overflow_msg)
+        raise ValueError("Arguments need to be mpf or mpc compatible numbers")
+
+    def fmul(ctx, x, y, **kwargs):
+        """
+        Multiplies the numbers *x* and *y*, giving a floating-point result,
+        optionally using a custom precision and rounding mode.
+
+        See the documentation of :func:`~mpmath.fadd` for a detailed description
+        of how to specify precision and rounding.
+
+        **Examples**
+
+        The result is an mpmath number::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fmul(2, 5.0)
+            mpf('10.0')
+            >>> fmul(0.5j, 0.5)
+            mpc(real='0.0', imag='0.25')
+
+        Avoiding roundoff::
+
+            >>> x, y = 10**10+1, 10**15+1
+            >>> print(x*y)
+            10000000001000010000000001
+            >>> print(mpf(x) * mpf(y))
+            1.0000000001e+25
+            >>> print(int(mpf(x) * mpf(y)))
+            10000000001000011026399232
+            >>> print(int(fmul(x, y)))
+            10000000001000011026399232
+            >>> print(int(fmul(x, y, dps=25)))
+            10000000001000010000000001
+            >>> print(int(fmul(x, y, exact=True)))
+            10000000001000010000000001
+
+        Exact multiplication with complex numbers can be inefficient and may
+        be impossible to perform with large magnitude differences between
+        real and imaginary parts::
+
+            >>> x = 1+2j
+            >>> y = mpc(2, '1e-100000000000000000000')
+            >>> fmul(x, y)
+            mpc(real='2.0', imag='4.0')
+            >>> fmul(x, y, rounding='u')
+            mpc(real='2.0', imag='4.0000000000000009')
+            >>> fmul(x, y, exact=True)
+            Traceback (most recent call last):
+              ...
+            OverflowError: the exact result does not fit in memory
+
+        """
+        prec, rounding = ctx._parse_prec(kwargs)
+        x = ctx.convert(x)
+        y = ctx.convert(y)
+        try:
+            if hasattr(x, '_mpf_'):
+                if hasattr(y, '_mpf_'):
+                    return ctx.make_mpf(mpf_mul(x._mpf_, y._mpf_, prec, rounding))
+                if hasattr(y, '_mpc_'):
+                    return ctx.make_mpc(mpc_mul_mpf(y._mpc_, x._mpf_, prec, rounding))
+            if hasattr(x, '_mpc_'):
+                if hasattr(y, '_mpf_'):
+                    return ctx.make_mpc(mpc_mul_mpf(x._mpc_, y._mpf_, prec, rounding))
+                if hasattr(y, '_mpc_'):
+                    return ctx.make_mpc(mpc_mul(x._mpc_, y._mpc_, prec, rounding))
+        except (ValueError, OverflowError):
+            raise OverflowError(ctx._exact_overflow_msg)
+        raise ValueError("Arguments need to be mpf or mpc compatible numbers")
+
+    def fdiv(ctx, x, y, **kwargs):
+        """
+        Divides the numbers *x* and *y*, giving a floating-point result,
+        optionally using a custom precision and rounding mode.
+
+        See the documentation of :func:`~mpmath.fadd` for a detailed description
+        of how to specify precision and rounding.
+
+        **Examples**
+
+        The result is an mpmath number::
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fdiv(3, 2)
+            mpf('1.5')
+            >>> fdiv(2, 3)
+            mpf('0.66666666666666663')
+            >>> fdiv(2+4j, 0.5)
+            mpc(real='4.0', imag='8.0')
+
+        The rounding direction and precision can be controlled::
+
+            >>> fdiv(2, 3, dps=3)    # Should be accurate to at least 3 digits
+            mpf('0.6666259765625')
+            >>> fdiv(2, 3, rounding='d')
+            mpf('0.66666666666666663')
+            >>> fdiv(2, 3, prec=60)
+            mpf('0.66666666666666667')
+            >>> fdiv(2, 3, rounding='u')
+            mpf('0.66666666666666674')
+
+        Checking the error of a division by performing it at higher precision::
+
+            >>> fdiv(2, 3) - fdiv(2, 3, prec=100)
+            mpf('-3.7007434154172148e-17')
+
+        Unlike :func:`~mpmath.fadd`, :func:`~mpmath.fmul`, etc., exact division is not
+        allowed since the quotient of two floating-point numbers generally
+        does not have an exact floating-point representation. (In the
+        future this might be changed to allow the case where the division
+        is actually exact.)
+
+            >>> fdiv(2, 3, exact=True)
+            Traceback (most recent call last):
+              ...
+            ValueError: division is not an exact operation
+
+        """
+        prec, rounding = ctx._parse_prec(kwargs)
+        if not prec:
+            raise ValueError("division is not an exact operation")
+        x = ctx.convert(x)
+        y = ctx.convert(y)
+        if hasattr(x, '_mpf_'):
+            if hasattr(y, '_mpf_'):
+                return ctx.make_mpf(mpf_div(x._mpf_, y._mpf_, prec, rounding))
+            if hasattr(y, '_mpc_'):
+                return ctx.make_mpc(mpc_div((x._mpf_, fzero), y._mpc_, prec, rounding))
+        if hasattr(x, '_mpc_'):
+            if hasattr(y, '_mpf_'):
+                return ctx.make_mpc(mpc_div_mpf(x._mpc_, y._mpf_, prec, rounding))
+            if hasattr(y, '_mpc_'):
+                return ctx.make_mpc(mpc_div(x._mpc_, y._mpc_, prec, rounding))
+        raise ValueError("Arguments need to be mpf or mpc compatible numbers")
+
+    def nint_distance(ctx, x):
+        r"""
+        Return `(n,d)` where `n` is the nearest integer to `x` and `d` is
+        an estimate of `\log_2(|x-n|)`. If `d < 0`, `-d` gives the precision
+        (measured in bits) lost to cancellation when computing `x-n`.
+
+            >>> from mpmath import *
+            >>> n, d = nint_distance(5)
+            >>> print(n); print(d)
+            5
+            -inf
+            >>> n, d = nint_distance(mpf(5))
+            >>> print(n); print(d)
+            5
+            -inf
+            >>> n, d = nint_distance(mpf(5.00000001))
+            >>> print(n); print(d)
+            5
+            -26
+            >>> n, d = nint_distance(mpf(4.99999999))
+            >>> print(n); print(d)
+            5
+            -26
+            >>> n, d = nint_distance(mpc(5,10))
+            >>> print(n); print(d)
+            5
+            4
+            >>> n, d = nint_distance(mpc(5,0.000001))
+            >>> print(n); print(d)
+            5
+            -19
+
+        """
+        typx = type(x)
+        if typx in int_types:
+            return int(x), ctx.ninf
+        elif typx is rational.mpq:
+            p, q = x._mpq_
+            n, r = divmod(p, q)
+            if 2*r >= q:
+                n += 1
+            elif not r:
+                return n, ctx.ninf
+            # log(p/q-n) = log((p-nq)/q) = log(p-nq) - log(q)
+            d = bitcount(abs(p-n*q)) - bitcount(q)
+            return n, d
+        if hasattr(x, "_mpf_"):
+            re = x._mpf_
+            im_dist = ctx.ninf
+        elif hasattr(x, "_mpc_"):
+            re, im = x._mpc_
+            isign, iman, iexp, ibc = im
+            if iman:
+                im_dist = iexp + ibc
+            elif im == fzero:
+                im_dist = ctx.ninf
+            else:
+                raise ValueError("requires a finite number")
+        else:
+            x = ctx.convert(x)
+            if hasattr(x, "_mpf_") or hasattr(x, "_mpc_"):
+                return ctx.nint_distance(x)
+            else:
+                raise TypeError("requires an mpf/mpc")
+        sign, man, exp, bc = re
+        mag = exp+bc
+        # |x| < 0.5
+        if mag < 0:
+            n = 0
+            re_dist = mag
+        elif man:
+            # exact integer
+            if exp >= 0:
+                n = man << exp
+                re_dist = ctx.ninf
+            # exact half-integer
+            elif exp == -1:
+                n = (man>>1)+1
+                re_dist = 0
+            else:
+                d = (-exp-1)
+                t = man >> d
+                if t & 1:
+                    t += 1
+                    man = (t<<d) - man
+                else:
+                    man -= (t<<d)
+                n = t>>1   # int(t)>>1
+                re_dist = exp+bitcount(man)
+            if sign:
+                n = -n
+        elif re == fzero:
+            re_dist = ctx.ninf
+            n = 0
+        else:
+            raise ValueError("requires a finite number")
+        return n, max(re_dist, im_dist)
+
+    def fprod(ctx, factors):
+        r"""
+        Calculates a product containing a finite number of factors (for
+        infinite products, see :func:`~mpmath.nprod`). The factors will be
+        converted to mpmath numbers.
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fprod([1, 2, 0.5, 7])
+            mpf('7.0')
+
+        """
+        orig = ctx.prec
+        try:
+            v = ctx.one
+            for p in factors:
+                v *= p
+        finally:
+            ctx.prec = orig
+        return +v
+
+    def rand(ctx):
+        """
+        Returns an ``mpf`` with value chosen randomly from `[0, 1)`.
+        The number of randomly generated bits in the mantissa is equal
+        to the working precision.
+        """
+        return ctx.make_mpf(mpf_rand(ctx._prec))
+
+    def fraction(ctx, p, q):
+        """
+        Given Python integers `(p, q)`, returns a lazy ``mpf`` representing
+        the fraction `p/q`. The value is updated with the precision.
+
+            >>> from mpmath import *
+            >>> mp.dps = 15
+            >>> a = fraction(1,100)
+            >>> b = mpf(1)/100
+            >>> print(a); print(b)
+            0.01
+            0.01
+            >>> mp.dps = 30
+            >>> print(a); print(b)      # a will be accurate
+            0.01
+            0.0100000000000000002081668171172
+            >>> mp.dps = 15
+        """
+        return ctx.constant(lambda prec, rnd: from_rational(p, q, prec, rnd),
+            '%s/%s' % (p, q))
+
+    def absmin(ctx, x):
+        return abs(ctx.convert(x))
+
+    def absmax(ctx, x):
+        return abs(ctx.convert(x))
+
+    def _as_points(ctx, x):
+        # XXX: remove this?
+        if hasattr(x, '_mpi_'):
+            a, b = x._mpi_
+            return [ctx.make_mpf(a), ctx.make_mpf(b)]
+        return x
+
+    '''
+    def _zetasum(ctx, s, a, b):
+        """
+        Computes sum of k^(-s) for k = a, a+1, ..., b with a, b both small
+        integers.
+        """
+        a = int(a)
+        b = int(b)
+        s = ctx.convert(s)
+        prec, rounding = ctx._prec_rounding
+        if hasattr(s, '_mpf_'):
+            v = ctx.make_mpf(libmp.mpf_zetasum(s._mpf_, a, b, prec))
+        elif hasattr(s, '_mpc_'):
+            v = ctx.make_mpc(libmp.mpc_zetasum(s._mpc_, a, b, prec))
+        return v
+    '''
+
+    def _zetasum_fast(ctx, s, a, n, derivatives=[0], reflect=False):
+        if not (ctx.isint(a) and hasattr(s, "_mpc_")):
+            raise NotImplementedError
+        a = int(a)
+        prec = ctx._prec
+        xs, ys = libmp.mpc_zetasum(s._mpc_, a, n, derivatives, reflect, prec)
+        xs = [ctx.make_mpc(x) for x in xs]
+        ys = [ctx.make_mpc(y) for y in ys]
+        return xs, ys
+
+class PrecisionManager:
+    def __init__(self, ctx, precfun, dpsfun, normalize_output=False):
+        self.ctx = ctx
+        self.precfun = precfun
+        self.dpsfun = dpsfun
+        self.normalize_output = normalize_output
+    def __call__(self, f):
+        @functools.wraps(f)
+        def g(*args, **kwargs):
+            orig = self.ctx.prec
+            try:
+                if self.precfun:
+                    self.ctx.prec = self.precfun(self.ctx.prec)
+                else:
+                    self.ctx.dps = self.dpsfun(self.ctx.dps)
+                if self.normalize_output:
+                    v = f(*args, **kwargs)
+                    if type(v) is tuple:
+                        return tuple([+a for a in v])
+                    return +v
+                else:
+                    return f(*args, **kwargs)
+            finally:
+                self.ctx.prec = orig
+        return g
+    def __enter__(self):
+        self.origp = self.ctx.prec
+        if self.precfun:
+            self.ctx.prec = self.precfun(self.ctx.prec)
+        else:
+            self.ctx.dps = self.dpsfun(self.ctx.dps)
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.ctx.prec = self.origp
+        return False
+
+
+if __name__ == '__main__':
+    import doctest
+    doctest.testmod()

lib/python3.11/site-packages/mpmath/ctx_mp_python.py

@@ -0,0 +1,1149 @@
+#from ctx_base import StandardBaseContext
+
+from .libmp.backend import basestring, exec_
+
+from .libmp import (MPZ, MPZ_ZERO, MPZ_ONE, int_types, repr_dps,
+    round_floor, round_ceiling, dps_to_prec, round_nearest, prec_to_dps,
+    ComplexResult, to_pickable, from_pickable, normalize,
+    from_int, from_float, from_npfloat, from_Decimal, from_str, to_int, to_float, to_str,
+    from_rational, from_man_exp,
+    fone, fzero, finf, fninf, fnan,
+    mpf_abs, mpf_pos, mpf_neg, mpf_add, mpf_sub, mpf_mul, mpf_mul_int,
+    mpf_div, mpf_rdiv_int, mpf_pow_int, mpf_mod,
+    mpf_eq, mpf_cmp, mpf_lt, mpf_gt, mpf_le, mpf_ge,
+    mpf_hash, mpf_rand,
+    mpf_sum,
+    bitcount, to_fixed,
+    mpc_to_str,
+    mpc_to_complex, mpc_hash, mpc_pos, mpc_is_nonzero, mpc_neg, mpc_conjugate,
+    mpc_abs, mpc_add, mpc_add_mpf, mpc_sub, mpc_sub_mpf, mpc_mul, mpc_mul_mpf,
+    mpc_mul_int, mpc_div, mpc_div_mpf, mpc_pow, mpc_pow_mpf, mpc_pow_int,
+    mpc_mpf_div,
+    mpf_pow,
+    mpf_pi, mpf_degree, mpf_e, mpf_phi, mpf_ln2, mpf_ln10,
+    mpf_euler, mpf_catalan, mpf_apery, mpf_khinchin,
+    mpf_glaisher, mpf_twinprime, mpf_mertens,
+    int_types)
+
+from . import rational
+from . import function_docs
+
+new = object.__new__
+
+class mpnumeric(object):
+    """Base class for mpf and mpc."""
+    __slots__ = []
+    def __new__(cls, val):
+        raise NotImplementedError
+
+class _mpf(mpnumeric):
+    """
+    An mpf instance holds a real-valued floating-point number. mpf:s
+    work analogously to Python floats, but support arbitrary-precision
+    arithmetic.
+    """
+    __slots__ = ['_mpf_']
+
+    def __new__(cls, val=fzero, **kwargs):
+        """A new mpf can be created from a Python float, an int, a
+        or a decimal string representing a number in floating-point
+        format."""
+        prec, rounding = cls.context._prec_rounding
+        if kwargs:
+            prec = kwargs.get('prec', prec)
+            if 'dps' in kwargs:
+                prec = dps_to_prec(kwargs['dps'])
+            rounding = kwargs.get('rounding', rounding)
+        if type(val) is cls:
+            sign, man, exp, bc = val._mpf_
+            if (not man) and exp:
+                return val
+            v = new(cls)
+            v._mpf_ = normalize(sign, man, exp, bc, prec, rounding)
+            return v
+        elif type(val) is tuple:
+            if len(val) == 2:
+                v = new(cls)
+                v._mpf_ = from_man_exp(val[0], val[1], prec, rounding)
+                return v
+            if len(val) == 4:
+                if val not in (finf, fninf, fnan):
+                    sign, man, exp, bc = val
+                    val = normalize(sign, MPZ(man), exp, bc, prec, rounding)
+                v = new(cls)
+                v._mpf_ = val
+                return v
+            raise ValueError
+        else:
+            v = new(cls)
+            v._mpf_ = mpf_pos(cls.mpf_convert_arg(val, prec, rounding), prec, rounding)
+            return v
+
+    @classmethod
+    def mpf_convert_arg(cls, x, prec, rounding):
+        if isinstance(x, int_types): return from_int(x)
+        if isinstance(x, float): return from_float(x)
+        if isinstance(x, basestring): return from_str(x, prec, rounding)
+        if isinstance(x, cls.context.constant): return x.func(prec, rounding)
+        if hasattr(x, '_mpf_'): return x._mpf_
+        if hasattr(x, '_mpmath_'):
+            t = cls.context.convert(x._mpmath_(prec, rounding))
+            if hasattr(t, '_mpf_'):
+                return t._mpf_
+        if hasattr(x, '_mpi_'):
+            a, b = x._mpi_
+            if a == b:
+                return a
+            raise ValueError("can only create mpf from zero-width interval")
+        raise TypeError("cannot create mpf from " + repr(x))
+
+    @classmethod
+    def mpf_convert_rhs(cls, x):
+        if isinstance(x, int_types): return from_int(x)
+        if isinstance(x, float): return from_float(x)
+        if isinstance(x, complex_types): return cls.context.mpc(x)
+        if isinstance(x, rational.mpq):
+            p, q = x._mpq_
+            return from_rational(p, q, cls.context.prec)
+        if hasattr(x, '_mpf_'): return x._mpf_
+        if hasattr(x, '_mpmath_'):
+            t = cls.context.convert(x._mpmath_(*cls.context._prec_rounding))
+            if hasattr(t, '_mpf_'):
+                return t._mpf_
+            return t
+        return NotImplemented
+
+    @classmethod
+    def mpf_convert_lhs(cls, x):
+        x = cls.mpf_convert_rhs(x)
+        if type(x) is tuple:
+            return cls.context.make_mpf(x)
+        return x
+
+    man_exp = property(lambda self: self._mpf_[1:3])
+    man = property(lambda self: self._mpf_[1])
+    exp = property(lambda self: self._mpf_[2])
+    bc = property(lambda self: self._mpf_[3])
+
+    real = property(lambda self: self)
+    imag = property(lambda self: self.context.zero)
+
+    conjugate = lambda self: self
+
+    def __getstate__(self): return to_pickable(self._mpf_)
+    def __setstate__(self, val): self._mpf_ = from_pickable(val)
+
+    def __repr__(s):
+        if s.context.pretty:
+            return str(s)
+        return "mpf('%s')" % to_str(s._mpf_, s.context._repr_digits)
+
+    def __str__(s): return to_str(s._mpf_, s.context._str_digits)
+    def __hash__(s): return mpf_hash(s._mpf_)
+    def __int__(s): return int(to_int(s._mpf_))
+    def __long__(s): return long(to_int(s._mpf_))
+    def __float__(s): return to_float(s._mpf_, rnd=s.context._prec_rounding[1])
+    def __complex__(s): return complex(float(s))
+    def __nonzero__(s): return s._mpf_ != fzero
+
+    __bool__ = __nonzero__
+
+    def __abs__(s):
+        cls, new, (prec, rounding) = s._ctxdata
+        v = new(cls)
+        v._mpf_ = mpf_abs(s._mpf_, prec, rounding)
+        return v
+
+    def __pos__(s):
+        cls, new, (prec, rounding) = s._ctxdata
+        v = new(cls)
+        v._mpf_ = mpf_pos(s._mpf_, prec, rounding)
+        return v
+
+    def __neg__(s):
+        cls, new, (prec, rounding) = s._ctxdata
+        v = new(cls)
+        v._mpf_ = mpf_neg(s._mpf_, prec, rounding)
+        return v
+
+    def _cmp(s, t, func):
+        if hasattr(t, '_mpf_'):
+            t = t._mpf_
+        else:
+            t = s.mpf_convert_rhs(t)
+            if t is NotImplemented:
+                return t
+        return func(s._mpf_, t)
+
+    def __cmp__(s, t): return s._cmp(t, mpf_cmp)
+    def __lt__(s, t): return s._cmp(t, mpf_lt)
+    def __gt__(s, t): return s._cmp(t, mpf_gt)
+    def __le__(s, t): return s._cmp(t, mpf_le)
+    def __ge__(s, t): return s._cmp(t, mpf_ge)
+
+    def __ne__(s, t):
+        v = s.__eq__(t)
+        if v is NotImplemented:
+            return v
+        return not v
+
+    def __rsub__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if type(t) in int_types:
+            v = new(cls)
+            v._mpf_ = mpf_sub(from_int(t), s._mpf_, prec, rounding)
+            return v
+        t = s.mpf_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t - s
+
+    def __rdiv__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if isinstance(t, int_types):
+            v = new(cls)
+            v._mpf_ = mpf_rdiv_int(t, s._mpf_, prec, rounding)
+            return v
+        t = s.mpf_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t / s
+
+    def __rpow__(s, t):
+        t = s.mpf_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t ** s
+
+    def __rmod__(s, t):
+        t = s.mpf_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t % s
+
+    def sqrt(s):
+        return s.context.sqrt(s)
+
+    def ae(s, t, rel_eps=None, abs_eps=None):
+        return s.context.almosteq(s, t, rel_eps, abs_eps)
+
+    def to_fixed(self, prec):
+        return to_fixed(self._mpf_, prec)
+
+    def __round__(self, *args):
+        return round(float(self), *args)
+
+mpf_binary_op = """
+def %NAME%(self, other):
+    mpf, new, (prec, rounding) = self._ctxdata
+    sval = self._mpf_
+    if hasattr(other, '_mpf_'):
+        tval = other._mpf_
+        %WITH_MPF%
+    ttype = type(other)
+    if ttype in int_types:
+        %WITH_INT%
+    elif ttype is float:
+        tval = from_float(other)
+        %WITH_MPF%
+    elif hasattr(other, '_mpc_'):
+        tval = other._mpc_
+        mpc = type(other)
+        %WITH_MPC%
+    elif ttype is complex:
+        tval = from_float(other.real), from_float(other.imag)
+        mpc = self.context.mpc
+        %WITH_MPC%
+    if isinstance(other, mpnumeric):
+        return NotImplemented
+    try:
+        other = mpf.context.convert(other, strings=False)
+    except TypeError:
+        return NotImplemented
+    return self.%NAME%(other)
+"""
+
+return_mpf = "; obj = new(mpf); obj._mpf_ = val; return obj"
+return_mpc = "; obj = new(mpc); obj._mpc_ = val; return obj"
+
+mpf_pow_same = """
+        try:
+            val = mpf_pow(sval, tval, prec, rounding) %s
+        except ComplexResult:
+            if mpf.context.trap_complex:
+                raise
+            mpc = mpf.context.mpc
+            val = mpc_pow((sval, fzero), (tval, fzero), prec, rounding) %s
+""" % (return_mpf, return_mpc)
+
+def binary_op(name, with_mpf='', with_int='', with_mpc=''):
+    code = mpf_binary_op
+    code = code.replace("%WITH_INT%", with_int)
+    code = code.replace("%WITH_MPC%", with_mpc)
+    code = code.replace("%WITH_MPF%", with_mpf)
+    code = code.replace("%NAME%", name)
+    np = {}
+    exec_(code, globals(), np)
+    return np[name]
+
+_mpf.__eq__ = binary_op('__eq__',
+    'return mpf_eq(sval, tval)',
+    'return mpf_eq(sval, from_int(other))',
+    'return (tval[1] == fzero) and mpf_eq(tval[0], sval)')
+
+_mpf.__add__ = binary_op('__add__',
+    'val = mpf_add(sval, tval, prec, rounding)' + return_mpf,
+    'val = mpf_add(sval, from_int(other), prec, rounding)' + return_mpf,
+    'val = mpc_add_mpf(tval, sval, prec, rounding)' + return_mpc)
+
+_mpf.__sub__ = binary_op('__sub__',
+    'val = mpf_sub(sval, tval, prec, rounding)' + return_mpf,
+    'val = mpf_sub(sval, from_int(other), prec, rounding)' + return_mpf,
+    'val = mpc_sub((sval, fzero), tval, prec, rounding)' + return_mpc)
+
+_mpf.__mul__ = binary_op('__mul__',
+    'val = mpf_mul(sval, tval, prec, rounding)' + return_mpf,
+    'val = mpf_mul_int(sval, other, prec, rounding)' + return_mpf,
+    'val = mpc_mul_mpf(tval, sval, prec, rounding)' + return_mpc)
+
+_mpf.__div__ = binary_op('__div__',
+    'val = mpf_div(sval, tval, prec, rounding)' + return_mpf,
+    'val = mpf_div(sval, from_int(other), prec, rounding)' + return_mpf,
+    'val = mpc_mpf_div(sval, tval, prec, rounding)' + return_mpc)
+
+_mpf.__mod__ = binary_op('__mod__',
+    'val = mpf_mod(sval, tval, prec, rounding)' + return_mpf,
+    'val = mpf_mod(sval, from_int(other), prec, rounding)' + return_mpf,
+    'raise NotImplementedError("complex modulo")')
+
+_mpf.__pow__ = binary_op('__pow__',
+    mpf_pow_same,
+    'val = mpf_pow_int(sval, other, prec, rounding)' + return_mpf,
+    'val = mpc_pow((sval, fzero), tval, prec, rounding)' + return_mpc)
+
+_mpf.__radd__ = _mpf.__add__
+_mpf.__rmul__ = _mpf.__mul__
+_mpf.__truediv__ = _mpf.__div__
+_mpf.__rtruediv__ = _mpf.__rdiv__
+
+
+class _constant(_mpf):
+    """Represents a mathematical constant with dynamic precision.
+    When printed or used in an arithmetic operation, a constant
+    is converted to a regular mpf at the working precision. A
+    regular mpf can also be obtained using the operation +x."""
+
+    def __new__(cls, func, name, docname=''):
+        a = object.__new__(cls)
+        a.name = name
+        a.func = func
+        a.__doc__ = getattr(function_docs, docname, '')
+        return a
+
+    def __call__(self, prec=None, dps=None, rounding=None):
+        prec2, rounding2 = self.context._prec_rounding
+        if not prec: prec = prec2
+        if not rounding: rounding = rounding2
+        if dps: prec = dps_to_prec(dps)
+        return self.context.make_mpf(self.func(prec, rounding))
+
+    @property
+    def _mpf_(self):
+        prec, rounding = self.context._prec_rounding
+        return self.func(prec, rounding)
+
+    def __repr__(self):
+        return "<%s: %s~>" % (self.name, self.context.nstr(self(dps=15)))
+
+
+class _mpc(mpnumeric):
+    """
+    An mpc represents a complex number using a pair of mpf:s (one
+    for the real part and another for the imaginary part.) The mpc
+    class behaves fairly similarly to Python's complex type.
+    """
+
+    __slots__ = ['_mpc_']
+
+    def __new__(cls, real=0, imag=0):
+        s = object.__new__(cls)
+        if isinstance(real, complex_types):
+            real, imag = real.real, real.imag
+        elif hasattr(real, '_mpc_'):
+            s._mpc_ = real._mpc_
+            return s
+        real = cls.context.mpf(real)
+        imag = cls.context.mpf(imag)
+        s._mpc_ = (real._mpf_, imag._mpf_)
+        return s
+
+    real = property(lambda self: self.context.make_mpf(self._mpc_[0]))
+    imag = property(lambda self: self.context.make_mpf(self._mpc_[1]))
+
+    def __getstate__(self):
+        return to_pickable(self._mpc_[0]), to_pickable(self._mpc_[1])
+
+    def __setstate__(self, val):
+        self._mpc_ = from_pickable(val[0]), from_pickable(val[1])
+
+    def __repr__(s):
+        if s.context.pretty:
+            return str(s)
+        r = repr(s.real)[4:-1]
+        i = repr(s.imag)[4:-1]
+        return "%s(real=%s, imag=%s)" % (type(s).__name__, r, i)
+
+    def __str__(s):
+        return "(%s)" % mpc_to_str(s._mpc_, s.context._str_digits)
+
+    def __complex__(s):
+        return mpc_to_complex(s._mpc_, rnd=s.context._prec_rounding[1])
+
+    def __pos__(s):
+        cls, new, (prec, rounding) = s._ctxdata
+        v = new(cls)
+        v._mpc_ = mpc_pos(s._mpc_, prec, rounding)
+        return v
+
+    def __abs__(s):
+        prec, rounding = s.context._prec_rounding
+        v = new(s.context.mpf)
+        v._mpf_ = mpc_abs(s._mpc_, prec, rounding)
+        return v
+
+    def __neg__(s):
+        cls, new, (prec, rounding) = s._ctxdata
+        v = new(cls)
+        v._mpc_ = mpc_neg(s._mpc_, prec, rounding)
+        return v
+
+    def conjugate(s):
+        cls, new, (prec, rounding) = s._ctxdata
+        v = new(cls)
+        v._mpc_ = mpc_conjugate(s._mpc_, prec, rounding)
+        return v
+
+    def __nonzero__(s):
+        return mpc_is_nonzero(s._mpc_)
+
+    __bool__ = __nonzero__
+
+    def __hash__(s):
+        return mpc_hash(s._mpc_)
+
+    @classmethod
+    def mpc_convert_lhs(cls, x):
+        try:
+            y = cls.context.convert(x)
+            return y
+        except TypeError:
+            return NotImplemented
+
+    def __eq__(s, t):
+        if not hasattr(t, '_mpc_'):
+            if isinstance(t, str):
+                return False
+            t = s.mpc_convert_lhs(t)
+            if t is NotImplemented:
+                return t
+        return s.real == t.real and s.imag == t.imag
+
+    def __ne__(s, t):
+        b = s.__eq__(t)
+        if b is NotImplemented:
+            return b
+        return not b
+
+    def _compare(*args):
+        raise TypeError("no ordering relation is defined for complex numbers")
+
+    __gt__ = _compare
+    __le__ = _compare
+    __gt__ = _compare
+    __ge__ = _compare
+
+    def __add__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if not hasattr(t, '_mpc_'):
+            t = s.mpc_convert_lhs(t)
+            if t is NotImplemented:
+                return t
+            if hasattr(t, '_mpf_'):
+                v = new(cls)
+                v._mpc_ = mpc_add_mpf(s._mpc_, t._mpf_, prec, rounding)
+                return v
+        v = new(cls)
+        v._mpc_ = mpc_add(s._mpc_, t._mpc_, prec, rounding)
+        return v
+
+    def __sub__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if not hasattr(t, '_mpc_'):
+            t = s.mpc_convert_lhs(t)
+            if t is NotImplemented:
+                return t
+            if hasattr(t, '_mpf_'):
+                v = new(cls)
+                v._mpc_ = mpc_sub_mpf(s._mpc_, t._mpf_, prec, rounding)
+                return v
+        v = new(cls)
+        v._mpc_ = mpc_sub(s._mpc_, t._mpc_, prec, rounding)
+        return v
+
+    def __mul__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if not hasattr(t, '_mpc_'):
+            if isinstance(t, int_types):
+                v = new(cls)
+                v._mpc_ = mpc_mul_int(s._mpc_, t, prec, rounding)
+                return v
+            t = s.mpc_convert_lhs(t)
+            if t is NotImplemented:
+                return t
+            if hasattr(t, '_mpf_'):
+                v = new(cls)
+                v._mpc_ = mpc_mul_mpf(s._mpc_, t._mpf_, prec, rounding)
+                return v
+            t = s.mpc_convert_lhs(t)
+        v = new(cls)
+        v._mpc_ = mpc_mul(s._mpc_, t._mpc_, prec, rounding)
+        return v
+
+    def __div__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if not hasattr(t, '_mpc_'):
+            t = s.mpc_convert_lhs(t)
+            if t is NotImplemented:
+                return t
+            if hasattr(t, '_mpf_'):
+                v = new(cls)
+                v._mpc_ = mpc_div_mpf(s._mpc_, t._mpf_, prec, rounding)
+                return v
+        v = new(cls)
+        v._mpc_ = mpc_div(s._mpc_, t._mpc_, prec, rounding)
+        return v
+
+    def __pow__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if isinstance(t, int_types):
+            v = new(cls)
+            v._mpc_ = mpc_pow_int(s._mpc_, t, prec, rounding)
+            return v
+        t = s.mpc_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        v = new(cls)
+        if hasattr(t, '_mpf_'):
+            v._mpc_ = mpc_pow_mpf(s._mpc_, t._mpf_, prec, rounding)
+        else:
+            v._mpc_ = mpc_pow(s._mpc_, t._mpc_, prec, rounding)
+        return v
+
+    __radd__ = __add__
+
+    def __rsub__(s, t):
+        t = s.mpc_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t - s
+
+    def __rmul__(s, t):
+        cls, new, (prec, rounding) = s._ctxdata
+        if isinstance(t, int_types):
+            v = new(cls)
+            v._mpc_ = mpc_mul_int(s._mpc_, t, prec, rounding)
+            return v
+        t = s.mpc_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t * s
+
+    def __rdiv__(s, t):
+        t = s.mpc_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t / s
+
+    def __rpow__(s, t):
+        t = s.mpc_convert_lhs(t)
+        if t is NotImplemented:
+            return t
+        return t ** s
+
+    __truediv__ = __div__
+    __rtruediv__ = __rdiv__
+
+    def ae(s, t, rel_eps=None, abs_eps=None):
+        return s.context.almosteq(s, t, rel_eps, abs_eps)
+
+
+complex_types = (complex, _mpc)
+
+
+class PythonMPContext(object):
+
+    def __init__(ctx):
+        ctx._prec_rounding = [53, round_nearest]
+        ctx.mpf = type('mpf', (_mpf,), {})
+        ctx.mpc = type('mpc', (_mpc,), {})
+        ctx.mpf._ctxdata = [ctx.mpf, new, ctx._prec_rounding]
+        ctx.mpc._ctxdata = [ctx.mpc, new, ctx._prec_rounding]
+        ctx.mpf.context = ctx
+        ctx.mpc.context = ctx
+        ctx.constant = type('constant', (_constant,), {})
+        ctx.constant._ctxdata = [ctx.mpf, new, ctx._prec_rounding]
+        ctx.constant.context = ctx
+
+    def make_mpf(ctx, v):
+        a = new(ctx.mpf)
+        a._mpf_ = v
+        return a
+
+    def make_mpc(ctx, v):
+        a = new(ctx.mpc)
+        a._mpc_ = v
+        return a
+
+    def default(ctx):
+        ctx._prec = ctx._prec_rounding[0] = 53
+        ctx._dps = 15
+        ctx.trap_complex = False
+
+    def _set_prec(ctx, n):
+        ctx._prec = ctx._prec_rounding[0] = max(1, int(n))
+        ctx._dps = prec_to_dps(n)
+
+    def _set_dps(ctx, n):
+        ctx._prec = ctx._prec_rounding[0] = dps_to_prec(n)
+        ctx._dps = max(1, int(n))
+
+    prec = property(lambda ctx: ctx._prec, _set_prec)
+    dps = property(lambda ctx: ctx._dps, _set_dps)
+
+    def convert(ctx, x, strings=True):
+        """
+        Converts *x* to an ``mpf`` or ``mpc``. If *x* is of type ``mpf``,
+        ``mpc``, ``int``, ``float``, ``complex``, the conversion
+        will be performed losslessly.
+
+        If *x* is a string, the result will be rounded to the present
+        working precision. Strings representing fractions or complex
+        numbers are permitted.
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> mpmathify(3.5)
+            mpf('3.5')
+            >>> mpmathify('2.1')
+            mpf('2.1000000000000001')
+            >>> mpmathify('3/4')
+            mpf('0.75')
+            >>> mpmathify('2+3j')
+            mpc(real='2.0', imag='3.0')
+
+        """
+        if type(x) in ctx.types: return x
+        if isinstance(x, int_types): return ctx.make_mpf(from_int(x))
+        if isinstance(x, float): return ctx.make_mpf(from_float(x))
+        if isinstance(x, complex):
+            return ctx.make_mpc((from_float(x.real), from_float(x.imag)))
+        if type(x).__module__ == 'numpy': return ctx.npconvert(x)
+        if isinstance(x, numbers.Rational): # e.g. Fraction
+            try: x = rational.mpq(int(x.numerator), int(x.denominator))
+            except: pass
+        prec, rounding = ctx._prec_rounding
+        if isinstance(x, rational.mpq):
+            p, q = x._mpq_
+            return ctx.make_mpf(from_rational(p, q, prec))
+        if strings and isinstance(x, basestring):
+            try:
+                _mpf_ = from_str(x, prec, rounding)
+                return ctx.make_mpf(_mpf_)
+            except ValueError:
+                pass
+        if hasattr(x, '_mpf_'): return ctx.make_mpf(x._mpf_)
+        if hasattr(x, '_mpc_'): return ctx.make_mpc(x._mpc_)
+        if hasattr(x, '_mpmath_'):
+            return ctx.convert(x._mpmath_(prec, rounding))
+        if type(x).__module__ == 'decimal':
+            try: return ctx.make_mpf(from_Decimal(x, prec, rounding))
+            except: pass
+        return ctx._convert_fallback(x, strings)
+
+    def npconvert(ctx, x):
+        """
+        Converts *x* to an ``mpf`` or ``mpc``. *x* should be a numpy
+        scalar.
+        """
+        import numpy as np
+        if isinstance(x, np.integer): return ctx.make_mpf(from_int(int(x)))
+        if isinstance(x, np.floating): return ctx.make_mpf(from_npfloat(x))
+        if isinstance(x, np.complexfloating):
+            return ctx.make_mpc((from_npfloat(x.real), from_npfloat(x.imag)))
+        raise TypeError("cannot create mpf from " + repr(x))
+
+    def isnan(ctx, x):
+        """
+        Return *True* if *x* is a NaN (not-a-number), or for a complex
+        number, whether either the real or complex part is NaN;
+        otherwise return *False*::
+
+            >>> from mpmath import *
+            >>> isnan(3.14)
+            False
+            >>> isnan(nan)
+            True
+            >>> isnan(mpc(3.14,2.72))
+            False
+            >>> isnan(mpc(3.14,nan))
+            True
+
+        """
+        if hasattr(x, "_mpf_"):
+            return x._mpf_ == fnan
+        if hasattr(x, "_mpc_"):
+            return fnan in x._mpc_
+        if isinstance(x, int_types) or isinstance(x, rational.mpq):
+            return False
+        x = ctx.convert(x)
+        if hasattr(x, '_mpf_') or hasattr(x, '_mpc_'):
+            return ctx.isnan(x)
+        raise TypeError("isnan() needs a number as input")
+
+    def isinf(ctx, x):
+        """
+        Return *True* if the absolute value of *x* is infinite;
+        otherwise return *False*::
+
+            >>> from mpmath import *
+            >>> isinf(inf)
+            True
+            >>> isinf(-inf)
+            True
+            >>> isinf(3)
+            False
+            >>> isinf(3+4j)
+            False
+            >>> isinf(mpc(3,inf))
+            True
+            >>> isinf(mpc(inf,3))
+            True
+
+        """
+        if hasattr(x, "_mpf_"):
+            return x._mpf_ in (finf, fninf)
+        if hasattr(x, "_mpc_"):
+            re, im = x._mpc_
+            return re in (finf, fninf) or im in (finf, fninf)
+        if isinstance(x, int_types) or isinstance(x, rational.mpq):
+            return False
+        x = ctx.convert(x)
+        if hasattr(x, '_mpf_') or hasattr(x, '_mpc_'):
+            return ctx.isinf(x)
+        raise TypeError("isinf() needs a number as input")
+
+    def isnormal(ctx, x):
+        """
+        Determine whether *x* is "normal" in the sense of floating-point
+        representation; that is, return *False* if *x* is zero, an
+        infinity or NaN; otherwise return *True*. By extension, a
+        complex number *x* is considered "normal" if its magnitude is
+        normal::
+
+            >>> from mpmath import *
+            >>> isnormal(3)
+            True
+            >>> isnormal(0)
+            False
+            >>> isnormal(inf); isnormal(-inf); isnormal(nan)
+            False
+            False
+            False
+            >>> isnormal(0+0j)
+            False
+            >>> isnormal(0+3j)
+            True
+            >>> isnormal(mpc(2,nan))
+            False
+        """
+        if hasattr(x, "_mpf_"):
+            return bool(x._mpf_[1])
+        if hasattr(x, "_mpc_"):
+            re, im = x._mpc_
+            re_normal = bool(re[1])
+            im_normal = bool(im[1])
+            if re == fzero: return im_normal
+            if im == fzero: return re_normal
+            return re_normal and im_normal
+        if isinstance(x, int_types) or isinstance(x, rational.mpq):
+            return bool(x)
+        x = ctx.convert(x)
+        if hasattr(x, '_mpf_') or hasattr(x, '_mpc_'):
+            return ctx.isnormal(x)
+        raise TypeError("isnormal() needs a number as input")
+
+    def isint(ctx, x, gaussian=False):
+        """
+        Return *True* if *x* is integer-valued; otherwise return
+        *False*::
+
+            >>> from mpmath import *
+            >>> isint(3)
+            True
+            >>> isint(mpf(3))
+            True
+            >>> isint(3.2)
+            False
+            >>> isint(inf)
+            False
+
+        Optionally, Gaussian integers can be checked for::
+
+            >>> isint(3+0j)
+            True
+            >>> isint(3+2j)
+            False
+            >>> isint(3+2j, gaussian=True)
+            True
+
+        """
+        if isinstance(x, int_types):
+            return True
+        if hasattr(x, "_mpf_"):
+            sign, man, exp, bc = xval = x._mpf_
+            return bool((man and exp >= 0) or xval == fzero)
+        if hasattr(x, "_mpc_"):
+            re, im = x._mpc_
+            rsign, rman, rexp, rbc = re
+            isign, iman, iexp, ibc = im
+            re_isint = (rman and rexp >= 0) or re == fzero
+            if gaussian:
+                im_isint = (iman and iexp >= 0) or im == fzero
+                return re_isint and im_isint
+            return re_isint and im == fzero
+        if isinstance(x, rational.mpq):
+            p, q = x._mpq_
+            return p % q == 0
+        x = ctx.convert(x)
+        if hasattr(x, '_mpf_') or hasattr(x, '_mpc_'):
+            return ctx.isint(x, gaussian)
+        raise TypeError("isint() needs a number as input")
+
+    def fsum(ctx, terms, absolute=False, squared=False):
+        """
+        Calculates a sum containing a finite number of terms (for infinite
+        series, see :func:`~mpmath.nsum`). The terms will be converted to
+        mpmath numbers. For len(terms) > 2, this function is generally
+        faster and produces more accurate results than the builtin
+        Python function :func:`sum`.
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> fsum([1, 2, 0.5, 7])
+            mpf('10.5')
+
+        With squared=True each term is squared, and with absolute=True
+        the absolute value of each term is used.
+        """
+        prec, rnd = ctx._prec_rounding
+        real = []
+        imag = []
+        for term in terms:
+            reval = imval = 0
+            if hasattr(term, "_mpf_"):
+                reval = term._mpf_
+            elif hasattr(term, "_mpc_"):
+                reval, imval = term._mpc_
+            else:
+                term = ctx.convert(term)
+                if hasattr(term, "_mpf_"):
+                    reval = term._mpf_
+                elif hasattr(term, "_mpc_"):
+                    reval, imval = term._mpc_
+                else:
+                    raise NotImplementedError
+            if imval:
+                if squared:
+                    if absolute:
+                        real.append(mpf_mul(reval,reval))
+                        real.append(mpf_mul(imval,imval))
+                    else:
+                        reval, imval = mpc_pow_int((reval,imval),2,prec+10)
+                        real.append(reval)
+                        imag.append(imval)
+                elif absolute:
+                    real.append(mpc_abs((reval,imval), prec))
+                else:
+                    real.append(reval)
+                    imag.append(imval)
+            else:
+                if squared:
+                    reval = mpf_mul(reval, reval)
+                elif absolute:
+                    reval = mpf_abs(reval)
+                real.append(reval)
+        s = mpf_sum(real, prec, rnd, absolute)
+        if imag:
+            s = ctx.make_mpc((s, mpf_sum(imag, prec, rnd)))
+        else:
+            s = ctx.make_mpf(s)
+        return s
+
+    def fdot(ctx, A, B=None, conjugate=False):
+        r"""
+        Computes the dot product of the iterables `A` and `B`,
+
+        .. math ::
+
+            \sum_{k=0} A_k B_k.
+
+        Alternatively, :func:`~mpmath.fdot` accepts a single iterable of pairs.
+        In other words, ``fdot(A,B)`` and ``fdot(zip(A,B))`` are equivalent.
+        The elements are automatically converted to mpmath numbers.
+
+        With ``conjugate=True``, the elements in the second vector
+        will be conjugated:
+
+        .. math ::
+
+            \sum_{k=0} A_k \overline{B_k}
+
+        **Examples**
+
+            >>> from mpmath import *
+            >>> mp.dps = 15; mp.pretty = False
+            >>> A = [2, 1.5, 3]
+            >>> B = [1, -1, 2]
+            >>> fdot(A, B)
+            mpf('6.5')
+            >>> list(zip(A, B))
+            [(2, 1), (1.5, -1), (3, 2)]
+            >>> fdot(_)
+            mpf('6.5')
+            >>> A = [2, 1.5, 3j]
+            >>> B = [1+j, 3, -1-j]
+            >>> fdot(A, B)
+            mpc(real='9.5', imag='-1.0')
+            >>> fdot(A, B, conjugate=True)
+            mpc(real='3.5', imag='-5.0')
+
+        """
+        if B is not None:
+            A = zip(A, B)
+        prec, rnd = ctx._prec_rounding
+        real = []
+        imag = []
+        hasattr_ = hasattr
+        types = (ctx.mpf, ctx.mpc)
+        for a, b in A:
+            if type(a) not in types: a = ctx.convert(a)
+            if type(b) not in types: b = ctx.convert(b)
+            a_real = hasattr_(a, "_mpf_")
+            b_real = hasattr_(b, "_mpf_")
+            if a_real and b_real:
+                real.append(mpf_mul(a._mpf_, b._mpf_))
+                continue
+            a_complex = hasattr_(a, "_mpc_")
+            b_complex = hasattr_(b, "_mpc_")
+            if a_real and b_complex:
+                aval = a._mpf_
+                bre, bim = b._mpc_
+                if conjugate:
+                    bim = mpf_neg(bim)
+                real.append(mpf_mul(aval, bre))
+                imag.append(mpf_mul(aval, bim))
+            elif b_real and a_complex:
+                are, aim = a._mpc_
+                bval = b._mpf_
+                real.append(mpf_mul(are, bval))
+                imag.append(mpf_mul(aim, bval))
+            elif a_complex and b_complex:
+                #re, im = mpc_mul(a._mpc_, b._mpc_, prec+20)
+                are, aim = a._mpc_
+                bre, bim = b._mpc_
+                if conjugate:
+                    bim = mpf_neg(bim)
+                real.append(mpf_mul(are, bre))
+                real.append(mpf_neg(mpf_mul(aim, bim)))
+                imag.append(mpf_mul(are, bim))
+                imag.append(mpf_mul(aim, bre))
+            else:
+                raise NotImplementedError
+        s = mpf_sum(real, prec, rnd)
+        if imag:
+            s = ctx.make_mpc((s, mpf_sum(imag, prec, rnd)))
+        else:
+            s = ctx.make_mpf(s)
+        return s
+
+    def _wrap_libmp_function(ctx, mpf_f, mpc_f=None, mpi_f=None, doc="<no doc>"):
+        """
+        Given a low-level mpf_ function, and optionally similar functions
+        for mpc_ and mpi_, defines the function as a context method.
+
+        It is assumed that the return type is the same as that of
+        the input; the exception is that propagation from mpf to mpc is possible
+        by raising ComplexResult.
+
+        """
+        def f(x, **kwargs):
+            if type(x) not in ctx.types:
+                x = ctx.convert(x)
+            prec, rounding = ctx._prec_rounding
+            if kwargs:
+                prec = kwargs.get('prec', prec)
+                if 'dps' in kwargs:
+                    prec = dps_to_prec(kwargs['dps'])
+                rounding = kwargs.get('rounding', rounding)
+            if hasattr(x, '_mpf_'):
+                try:
+                    return ctx.make_mpf(mpf_f(x._mpf_, prec, rounding))
+                except ComplexResult:
+                    # Handle propagation to complex
+                    if ctx.trap_complex:
+                        raise
+                    return ctx.make_mpc(mpc_f((x._mpf_, fzero), prec, rounding))
+            elif hasattr(x, '_mpc_'):
+                return ctx.make_mpc(mpc_f(x._mpc_, prec, rounding))
+            raise NotImplementedError("%s of a %s" % (name, type(x)))
+        name = mpf_f.__name__[4:]
+        f.__doc__ = function_docs.__dict__.get(name, "Computes the %s of x" % doc)
+        return f
+
+    # Called by SpecialFunctions.__init__()
+    @classmethod
+    def _wrap_specfun(cls, name, f, wrap):
+        if wrap:
+            def f_wrapped(ctx, *args, **kwargs):
+                convert = ctx.convert
+                args = [convert(a) for a in args]
+                prec = ctx.prec
+                try:
+                    ctx.prec += 10
+                    retval = f(ctx, *args, **kwargs)
+                finally:
+                    ctx.prec = prec
+                return +retval
+        else:
+            f_wrapped = f
+        f_wrapped.__doc__ = function_docs.__dict__.get(name, f.__doc__)
+        setattr(cls, name, f_wrapped)
+
+    def _convert_param(ctx, x):
+        if hasattr(x, "_mpc_"):
+            v, im = x._mpc_
+            if im != fzero:
+                return x, 'C'
+        elif hasattr(x, "_mpf_"):
+            v = x._mpf_
+        else:
+            if type(x) in int_types:
+                return int(x), 'Z'
+            p = None
+            if isinstance(x, tuple):
+                p, q = x
+            elif hasattr(x, '_mpq_'):
+                p, q = x._mpq_
+            elif isinstance(x, basestring) and '/' in x:
+                p, q = x.split('/')
+                p = int(p)
+                q = int(q)
+            if p is not None:
+                if not p % q:
+                    return p // q, 'Z'
+                return ctx.mpq(p,q), 'Q'
+            x = ctx.convert(x)
+            if hasattr(x, "_mpc_"):
+                v, im = x._mpc_
+                if im != fzero:
+                    return x, 'C'
+            elif hasattr(x, "_mpf_"):
+                v = x._mpf_
+            else:
+                return x, 'U'
+        sign, man, exp, bc = v
+        if man:
+            if exp >= -4:
+                if sign:
+                    man = -man
+                if exp >= 0:
+                    return int(man) << exp, 'Z'
+                if exp >= -4:
+                    p, q = int(man), (1<<(-exp))
+                    return ctx.mpq(p,q), 'Q'
+            x = ctx.make_mpf(v)
+            return x, 'R'
+        elif not exp:
+            return 0, 'Z'
+        else:
+            return x, 'U'
+
+    def _mpf_mag(ctx, x):
+        sign, man, exp, bc = x
+        if man:
+            return exp+bc
+        if x == fzero:
+            return ctx.ninf
+        if x == finf or x == fninf:
+            return ctx.inf
+        return ctx.nan
+
+    def mag(ctx, x):
+        """
+        Quick logarithmic magnitude estimate of a number. Returns an
+        integer or infinity `m` such that `|x| <= 2^m`. It is not
+        guaranteed that `m` is an optimal bound, but it will never
+        be too large by more than 2 (and probably not more than 1).
+
+        **Examples**
+
+            >>> from mpmath import *
+            >>> mp.pretty = True
+            >>> mag(10), mag(10.0), mag(mpf(10)), int(ceil(log(10,2)))
+            (4, 4, 4, 4)
+            >>> mag(10j), mag(10+10j)
+            (4, 5)
+            >>> mag(0.01), int(ceil(log(0.01,2)))
+            (-6, -6)
+            >>> mag(0), mag(inf), mag(-inf), mag(nan)
+            (-inf, +inf, +inf, nan)
+
+        """
+        if hasattr(x, "_mpf_"):
+            return ctx._mpf_mag(x._mpf_)
+        elif hasattr(x, "_mpc_"):
+            r, i = x._mpc_
+            if r == fzero:
+                return ctx._mpf_mag(i)
+            if i == fzero:
+                return ctx._mpf_mag(r)
+            return 1+max(ctx._mpf_mag(r), ctx._mpf_mag(i))
+        elif isinstance(x, int_types):
+            if x:
+                return bitcount(abs(x))
+            return ctx.ninf
+        elif isinstance(x, rational.mpq):
+            p, q = x._mpq_
+            if p:
+                return 1 + bitcount(abs(p)) - bitcount(q)
+            return ctx.ninf
+        else:
+            x = ctx.convert(x)
+            if hasattr(x, "_mpf_") or hasattr(x, "_mpc_"):
+                return ctx.mag(x)
+            else:
+                raise TypeError("requires an mpf/mpc")
+
+
+# Register with "numbers" ABC
+#     We do not subclass, hence we do not use the @abstractmethod checks. While
+#     this is less invasive it may turn out that we do not actually support
+#     parts of the expected interfaces.  See
+#     http://docs.python.org/2/library/numbers.html for list of abstract
+#     methods.
+try:
+    import numbers
+    numbers.Complex.register(_mpc)
+    numbers.Real.register(_mpf)
+except ImportError:
+    pass

lib/python3.11/site-packages/mpmath/functions/__init__.py

@@ -0,0 +1,14 @@
+from . import functions
+# Hack to update methods
+from . import factorials
+from . import hypergeometric
+from . import expintegrals
+from . import bessel
+from . import orthogonal
+from . import theta
+from . import elliptic
+from . import signals
+from . import zeta
+from . import rszeta
+from . import zetazeros
+from . import qfunctions

lib/python3.11/site-packages/mpmath/functions/bessel.py

@@ -0,0 +1,1108 @@
+from .functions import defun, defun_wrapped
+
+@defun
+def j0(ctx, x):
+    """Computes the Bessel function `J_0(x)`. See :func:`~mpmath.besselj`."""
+    return ctx.besselj(0, x)
+
+@defun
+def j1(ctx, x):
+    """Computes the Bessel function `J_1(x)`.  See :func:`~mpmath.besselj`."""
+    return ctx.besselj(1, x)
+
+@defun
+def besselj(ctx, n, z, derivative=0, **kwargs):
+    if type(n) is int:
+        n_isint = True
+    else:
+        n = ctx.convert(n)
+        n_isint = ctx.isint(n)
+        if n_isint:
+            n = int(ctx._re(n))
+    if n_isint and n < 0:
+        return (-1)**n * ctx.besselj(-n, z, derivative, **kwargs)
+    z = ctx.convert(z)
+    M = ctx.mag(z)
+    if derivative:
+        d = ctx.convert(derivative)
+        # TODO: the integer special-casing shouldn't be necessary.
+        # However, the hypergeometric series gets inaccurate for large d
+        # because of inaccurate pole cancellation at a pole far from
+        # zero (needs to be fixed in hypercomb or hypsum)
+        if ctx.isint(d) and d >= 0:
+            d = int(d)
+            orig = ctx.prec
+            try:
+                ctx.prec += 15
+                v = ctx.fsum((-1)**k * ctx.binomial(d,k) * ctx.besselj(2*k+n-d,z)
+                    for k in range(d+1))
+            finally:
+                ctx.prec = orig
+            v *= ctx.mpf(2)**(-d)
+        else:
+            def h(n,d):
+                r = ctx.fmul(ctx.fmul(z, z, prec=ctx.prec+M), -0.25, exact=True)
+                B = [0.5*(n-d+1), 0.5*(n-d+2)]
+                T = [([2,ctx.pi,z],[d-2*n,0.5,n-d],[],B,[(n+1)*0.5,(n+2)*0.5],B+[n+1],r)]
+                return T
+            v = ctx.hypercomb(h, [n,d], **kwargs)
+    else:
+        # Fast case: J_n(x), n int, appropriate magnitude for fixed-point calculation
+        if (not derivative) and n_isint and abs(M) < 10 and abs(n) < 20:
+            try:
+                return ctx._besselj(n, z)
+            except NotImplementedError:
+                pass
+        if not z:
+            if not n:
+                v = ctx.one + n+z
+            elif ctx.re(n) > 0:
+                v = n*z
+            else:
+                v = ctx.inf + z + n
+        else:
+            #v = 0
+            orig = ctx.prec
+            try:
+                # XXX: workaround for accuracy in low level hypergeometric series
+                # when alternating, large arguments
+                ctx.prec += min(3*abs(M), ctx.prec)
+                w = ctx.fmul(z, 0.5, exact=True)
+                def h(n):
+                    r = ctx.fneg(ctx.fmul(w, w, prec=max(0,ctx.prec+M)), exact=True)
+                    return [([w], [n], [], [n+1], [], [n+1], r)]
+                v = ctx.hypercomb(h, [n], **kwargs)
+            finally:
+                ctx.prec = orig
+        v = +v
+    return v
+
+@defun
+def besseli(ctx, n, z, derivative=0, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    if not z:
+        if derivative:
+            raise ValueError
+        if not n:
+            # I(0,0) = 1
+            return 1+n+z
+        if ctx.isint(n):
+            return 0*(n+z)
+        r = ctx.re(n)
+        if r == 0:
+            return ctx.nan*(n+z)
+        elif r > 0:
+            return 0*(n+z)
+        else:
+            return ctx.inf+(n+z)
+    M = ctx.mag(z)
+    if derivative:
+        d = ctx.convert(derivative)
+        def h(n,d):
+            r = ctx.fmul(ctx.fmul(z, z, prec=ctx.prec+M), 0.25, exact=True)
+            B = [0.5*(n-d+1), 0.5*(n-d+2), n+1]
+            T = [([2,ctx.pi,z],[d-2*n,0.5,n-d],[n+1],B,[(n+1)*0.5,(n+2)*0.5],B,r)]
+            return T
+        v = ctx.hypercomb(h, [n,d], **kwargs)
+    else:
+        def h(n):
+            w = ctx.fmul(z, 0.5, exact=True)
+            r = ctx.fmul(w, w, prec=max(0,ctx.prec+M))
+            return [([w], [n], [], [n+1], [], [n+1], r)]
+        v = ctx.hypercomb(h, [n], **kwargs)
+    return v
+
+@defun_wrapped
+def bessely(ctx, n, z, derivative=0, **kwargs):
+    if not z:
+        if derivative:
+            # Not implemented
+            raise ValueError
+        if not n:
+            # ~ log(z/2)
+            return -ctx.inf + (n+z)
+        if ctx.im(n):
+            return ctx.nan * (n+z)
+        r = ctx.re(n)
+        q = n+0.5
+        if ctx.isint(q):
+            if n > 0:
+                return -ctx.inf + (n+z)
+            else:
+                return 0 * (n+z)
+        if r < 0 and int(ctx.floor(q)) % 2:
+            return ctx.inf + (n+z)
+        else:
+            return ctx.ninf + (n+z)
+    # XXX: use hypercomb
+    ctx.prec += 10
+    m, d = ctx.nint_distance(n)
+    if d < -ctx.prec:
+        h = +ctx.eps
+        ctx.prec *= 2
+        n += h
+    elif d < 0:
+        ctx.prec -= d
+    # TODO: avoid cancellation for imaginary arguments
+    cos, sin = ctx.cospi_sinpi(n)
+    return (ctx.besselj(n,z,derivative,**kwargs)*cos - \
+        ctx.besselj(-n,z,derivative,**kwargs))/sin
+
+@defun_wrapped
+def besselk(ctx, n, z, **kwargs):
+    if not z:
+        return ctx.inf
+    M = ctx.mag(z)
+    if M < 1:
+        # Represent as limit definition
+        def h(n):
+            r = (z/2)**2
+            T1 = [z, 2], [-n, n-1], [n], [], [], [1-n], r
+            T2 = [z, 2], [n, -n-1], [-n], [], [], [1+n], r
+            return T1, T2
+    # We could use the limit definition always, but it leads
+    # to very bad cancellation (of exponentially large terms)
+    # for large real z
+    # Instead represent in terms of 2F0
+    else:
+        ctx.prec += M
+        def h(n):
+            return [([ctx.pi/2, z, ctx.exp(-z)], [0.5,-0.5,1], [], [], \
+                [n+0.5, 0.5-n], [], -1/(2*z))]
+    return ctx.hypercomb(h, [n], **kwargs)
+
+@defun_wrapped
+def hankel1(ctx,n,x,**kwargs):
+    return ctx.besselj(n,x,**kwargs) + ctx.j*ctx.bessely(n,x,**kwargs)
+
+@defun_wrapped
+def hankel2(ctx,n,x,**kwargs):
+    return ctx.besselj(n,x,**kwargs) - ctx.j*ctx.bessely(n,x,**kwargs)
+
+@defun_wrapped
+def whitm(ctx,k,m,z,**kwargs):
+    if z == 0:
+        # M(k,m,z) = 0^(1/2+m)
+        if ctx.re(m) > -0.5:
+            return z
+        elif ctx.re(m) < -0.5:
+            return ctx.inf + z
+        else:
+            return ctx.nan * z
+    x = ctx.fmul(-0.5, z, exact=True)
+    y = 0.5+m
+    return ctx.exp(x) * z**y * ctx.hyp1f1(y-k, 1+2*m, z, **kwargs)
+
+@defun_wrapped
+def whitw(ctx,k,m,z,**kwargs):
+    if z == 0:
+        g = abs(ctx.re(m))
+        if g < 0.5:
+            return z
+        elif g > 0.5:
+            return ctx.inf + z
+        else:
+            return ctx.nan * z
+    x = ctx.fmul(-0.5, z, exact=True)
+    y = 0.5+m
+    return ctx.exp(x) * z**y * ctx.hyperu(y-k, 1+2*m, z, **kwargs)
+
+@defun
+def hyperu(ctx, a, b, z, **kwargs):
+    a, atype = ctx._convert_param(a)
+    b, btype = ctx._convert_param(b)
+    z = ctx.convert(z)
+    if not z:
+        if ctx.re(b) <= 1:
+            return ctx.gammaprod([1-b],[a-b+1])
+        else:
+            return ctx.inf + z
+    bb = 1+a-b
+    bb, bbtype = ctx._convert_param(bb)
+    try:
+        orig = ctx.prec
+        try:
+            ctx.prec += 10
+            v = ctx.hypsum(2, 0, (atype, bbtype), [a, bb], -1/z, maxterms=ctx.prec)
+            return v / z**a
+        finally:
+            ctx.prec = orig
+    except ctx.NoConvergence:
+        pass
+    def h(a,b):
+        w = ctx.sinpi(b)
+        T1 = ([ctx.pi,w],[1,-1],[],[a-b+1,b],[a],[b],z)
+        T2 = ([-ctx.pi,w,z],[1,-1,1-b],[],[a,2-b],[a-b+1],[2-b],z)
+        return T1, T2
+    return ctx.hypercomb(h, [a,b], **kwargs)
+
+@defun
+def struveh(ctx,n,z, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    # http://functions.wolfram.com/Bessel-TypeFunctions/StruveH/26/01/02/
+    def h(n):
+        return [([z/2, 0.5*ctx.sqrt(ctx.pi)], [n+1, -1], [], [n+1.5], [1], [1.5, n+1.5], -(z/2)**2)]
+    return ctx.hypercomb(h, [n], **kwargs)
+
+@defun
+def struvel(ctx,n,z, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    # http://functions.wolfram.com/Bessel-TypeFunctions/StruveL/26/01/02/
+    def h(n):
+        return [([z/2, 0.5*ctx.sqrt(ctx.pi)], [n+1, -1], [], [n+1.5], [1], [1.5, n+1.5], (z/2)**2)]
+    return ctx.hypercomb(h, [n], **kwargs)
+
+def _anger(ctx,which,v,z,**kwargs):
+    v = ctx._convert_param(v)[0]
+    z = ctx.convert(z)
+    def h(v):
+        b = ctx.mpq_1_2
+        u = v*b
+        m = b*3
+        a1,a2,b1,b2 = m-u, m+u, 1-u, 1+u
+        c, s = ctx.cospi_sinpi(u)
+        if which == 0:
+            A, B = [b*z, s], [c]
+        if which == 1:
+            A, B = [b*z, -c], [s]
+        w = ctx.square_exp_arg(z, mult=-0.25)
+        T1 = A, [1, 1], [], [a1,a2], [1], [a1,a2], w
+        T2 = B, [1], [], [b1,b2], [1], [b1,b2], w
+        return T1, T2
+    return ctx.hypercomb(h, [v], **kwargs)
+
+@defun
+def angerj(ctx, v, z, **kwargs):
+    return _anger(ctx, 0, v, z, **kwargs)
+
+@defun
+def webere(ctx, v, z, **kwargs):
+    return _anger(ctx, 1, v, z, **kwargs)
+
+@defun
+def lommels1(ctx, u, v, z, **kwargs):
+    u = ctx._convert_param(u)[0]
+    v = ctx._convert_param(v)[0]
+    z = ctx.convert(z)
+    def h(u,v):
+        b = ctx.mpq_1_2
+        w = ctx.square_exp_arg(z, mult=-0.25)
+        return ([u-v+1, u+v+1, z], [-1, -1, u+1], [], [], [1], \
+            [b*(u-v+3),b*(u+v+3)], w),
+    return ctx.hypercomb(h, [u,v], **kwargs)
+
+@defun
+def lommels2(ctx, u, v, z, **kwargs):
+    u = ctx._convert_param(u)[0]
+    v = ctx._convert_param(v)[0]
+    z = ctx.convert(z)
+    # Asymptotic expansion (GR p. 947) -- need to be careful
+    # not to use for small arguments
+    # def h(u,v):
+    #    b = ctx.mpq_1_2
+    #    w = -(z/2)**(-2)
+    #    return ([z], [u-1], [], [], [b*(1-u+v)], [b*(1-u-v)], w),
+    def h(u,v):
+        b = ctx.mpq_1_2
+        w = ctx.square_exp_arg(z, mult=-0.25)
+        T1 = [u-v+1, u+v+1, z], [-1, -1, u+1], [], [], [1], [b*(u-v+3),b*(u+v+3)], w
+        T2 = [2, z], [u+v-1, -v], [v, b*(u+v+1)], [b*(v-u+1)], [], [1-v], w
+        T3 = [2, z], [u-v-1, v], [-v, b*(u-v+1)], [b*(1-u-v)], [], [1+v], w
+        #c1 = ctx.cospi((u-v)*b)
+        #c2 = ctx.cospi((u+v)*b)
+        #s = ctx.sinpi(v)
+        #r1 = (u-v+1)*b
+        #r2 = (u+v+1)*b
+        #T2 = [c1, s, z, 2], [1, -1, -v, v], [], [-v+1], [], [-v+1], w
+        #T3 = [-c2, s, z, 2], [1, -1, v, -v], [], [v+1], [], [v+1], w
+        #T2 = [c1, s, z, 2], [1, -1, -v, v+u-1], [r1, r2], [-v+1], [], [-v+1], w
+        #T3 = [-c2, s, z, 2], [1, -1, v, -v+u-1], [r1, r2], [v+1], [], [v+1], w
+        return T1, T2, T3
+    return ctx.hypercomb(h, [u,v], **kwargs)
+
+@defun
+def ber(ctx, n, z, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    # http://functions.wolfram.com/Bessel-TypeFunctions/KelvinBer2/26/01/02/0001/
+    def h(n):
+        r = -(z/4)**4
+        cos, sin = ctx.cospi_sinpi(-0.75*n)
+        T1 = [cos, z/2], [1, n], [], [n+1], [], [0.5, 0.5*(n+1), 0.5*n+1], r
+        T2 = [sin, z/2], [1, n+2], [], [n+2], [], [1.5, 0.5*(n+3), 0.5*n+1], r
+        return T1, T2
+    return ctx.hypercomb(h, [n], **kwargs)
+
+@defun
+def bei(ctx, n, z, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    # http://functions.wolfram.com/Bessel-TypeFunctions/KelvinBei2/26/01/02/0001/
+    def h(n):
+        r = -(z/4)**4
+        cos, sin = ctx.cospi_sinpi(0.75*n)
+        T1 = [cos, z/2], [1, n+2], [], [n+2], [], [1.5, 0.5*(n+3), 0.5*n+1], r
+        T2 = [sin, z/2], [1, n], [], [n+1], [], [0.5, 0.5*(n+1), 0.5*n+1], r
+        return T1, T2
+    return ctx.hypercomb(h, [n], **kwargs)
+
+@defun
+def ker(ctx, n, z, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    # http://functions.wolfram.com/Bessel-TypeFunctions/KelvinKer2/26/01/02/0001/
+    def h(n):
+        r = -(z/4)**4
+        cos1, sin1 = ctx.cospi_sinpi(0.25*n)
+        cos2, sin2 = ctx.cospi_sinpi(0.75*n)
+        T1 = [2, z, 4*cos1], [-n-3, n, 1], [-n], [], [], [0.5, 0.5*(1+n), 0.5*(n+2)], r
+        T2 = [2, z, -sin1], [-n-3, 2+n, 1], [-n-1], [], [], [1.5, 0.5*(3+n), 0.5*(n+2)], r
+        T3 = [2, z, 4*cos2], [n-3, -n, 1], [n], [], [], [0.5, 0.5*(1-n), 1-0.5*n], r
+        T4 = [2, z, -sin2], [n-3, 2-n, 1], [n-1], [], [], [1.5, 0.5*(3-n), 1-0.5*n], r
+        return T1, T2, T3, T4
+    return ctx.hypercomb(h, [n], **kwargs)
+
+@defun
+def kei(ctx, n, z, **kwargs):
+    n = ctx.convert(n)
+    z = ctx.convert(z)
+    # http://functions.wolfram.com/Bessel-TypeFunctions/KelvinKei2/26/01/02/0001/
+    def h(n):
+        r = -(z/4)**4
+        cos1, sin1 = ctx.cospi_sinpi(0.75*n)
+        cos2, sin2 = ctx.cospi_sinpi(0.25*n)
+        T1 = [-cos1, 2, z], [1, n-3, 2-n], [n-1], [], [], [1.5, 0.5*(3-n), 1-0.5*n], r
+        T2 = [-sin1, 2, z], [1, n-1, -n], [n], [], [], [0.5, 0.5*(1-n), 1-0.5*n], r
+        T3 = [-sin2, 2, z], [1, -n-1, n], [-n], [], [], [0.5, 0.5*(n+1), 0.5*(n+2)], r
+        T4 = [-cos2, 2, z], [1, -n-3, n+2], [-n-1], [], [], [1.5, 0.5*(n+3), 0.5*(n+2)], r
+        return T1, T2, T3, T4
+    return ctx.hypercomb(h, [n], **kwargs)
+
+# TODO: do this more generically?
+def c_memo(f):
+    name = f.__name__
+    def f_wrapped(ctx):
+        cache = ctx._misc_const_cache
+        prec = ctx.prec
+        p,v = cache.get(name, (-1,0))
+        if p >= prec:
+            return +v
+        else:
+            cache[name] = (prec, f(ctx))
+            return cache[name][1]
+    return f_wrapped
+
+@c_memo
+def _airyai_C1(ctx):
+    return 1 / (ctx.cbrt(9) * ctx.gamma(ctx.mpf(2)/3))
+
+@c_memo
+def _airyai_C2(ctx):
+    return -1 / (ctx.cbrt(3) * ctx.gamma(ctx.mpf(1)/3))
+
+@c_memo
+def _airybi_C1(ctx):
+    return 1 / (ctx.nthroot(3,6) * ctx.gamma(ctx.mpf(2)/3))
+
+@c_memo
+def _airybi_C2(ctx):
+    return ctx.nthroot(3,6) / ctx.gamma(ctx.mpf(1)/3)
+
+def _airybi_n2_inf(ctx):
+    prec = ctx.prec
+    try:
+        v = ctx.power(3,'2/3')*ctx.gamma('2/3')/(2*ctx.pi)
+    finally:
+        ctx.prec = prec
+    return +v
+
+# Derivatives at z = 0
+# TODO: could be expressed more elegantly using triple factorials
+def _airyderiv_0(ctx, z, n, ntype, which):
+    if ntype == 'Z':
+        if n < 0:
+            return z
+        r = ctx.mpq_1_3
+        prec = ctx.prec
+        try:
+            ctx.prec += 10
+            v = ctx.gamma((n+1)*r) * ctx.power(3,n*r) / ctx.pi
+            if which == 0:
+                v *= ctx.sinpi(2*(n+1)*r)
+                v /= ctx.power(3,'2/3')
+            else:
+                v *= abs(ctx.sinpi(2*(n+1)*r))
+                v /= ctx.power(3,'1/6')
+        finally:
+            ctx.prec = prec
+        return +v + z
+    else:
+        # singular (does the limit exist?)
+        raise NotImplementedError
+
+@defun
+def airyai(ctx, z, derivative=0, **kwargs):
+    z = ctx.convert(z)
+    if derivative:
+        n, ntype = ctx._convert_param(derivative)
+    else:
+        n = 0
+    # Values at infinities
+    if not ctx.isnormal(z) and z:
+        if n and ntype == 'Z':
+            if n == -1:
+                if z == ctx.inf:
+                    return ctx.mpf(1)/3 + 1/z
+                if z == ctx.ninf:
+                    return ctx.mpf(-2)/3 + 1/z
+            if n < -1:
+                if z == ctx.inf:
+                    return z
+                if z == ctx.ninf:
+                    return (-1)**n * (-z)
+        if (not n) and z == ctx.inf or z == ctx.ninf:
+            return 1/z
+        # TODO: limits
+        raise ValueError("essential singularity of Ai(z)")
+    # Account for exponential scaling
+    if z:
+        extraprec = max(0, int(1.5*ctx.mag(z)))
+    else:
+        extraprec = 0
+    if n:
+        if n == 1:
+            def h():
+                # http://functions.wolfram.com/03.07.06.0005.01
+                if ctx._re(z) > 4:
+                    ctx.prec += extraprec
+                    w = z**1.5; r = -0.75/w; u = -2*w/3
+                    ctx.prec -= extraprec
+                    C = -ctx.exp(u)/(2*ctx.sqrt(ctx.pi))*ctx.nthroot(z,4)
+                    return ([C],[1],[],[],[(-1,6),(7,6)],[],r),
+                # http://functions.wolfram.com/03.07.26.0001.01
+                else:
+                    ctx.prec += extraprec
+                    w = z**3 / 9
+                    ctx.prec -= extraprec
+                    C1 = _airyai_C1(ctx) * 0.5
+                    C2 = _airyai_C2(ctx)
+                    T1 = [C1,z],[1,2],[],[],[],[ctx.mpq_5_3],w
+                    T2 = [C2],[1],[],[],[],[ctx.mpq_1_3],w
+                    return T1, T2
+            return ctx.hypercomb(h, [], **kwargs)
+        else:
+            if z == 0:
+                return _airyderiv_0(ctx, z, n, ntype, 0)
+            # http://functions.wolfram.com/03.05.20.0004.01
+            def h(n):
+                ctx.prec += extraprec
+                w = z**3/9
+                ctx.prec -= extraprec
+                q13,q23,q43 = ctx.mpq_1_3, ctx.mpq_2_3, ctx.mpq_4_3
+                a1=q13; a2=1; b1=(1-n)*q13; b2=(2-n)*q13; b3=1-n*q13
+                T1 = [3, z], [n-q23, -n], [a1], [b1,b2,b3], \
+                    [a1,a2], [b1,b2,b3], w
+                a1=q23; b1=(2-n)*q13; b2=1-n*q13; b3=(4-n)*q13
+                T2 = [3, z, -z], [n-q43, -n, 1], [a1], [b1,b2,b3], \
+                    [a1,a2], [b1,b2,b3], w
+                return T1, T2
+            v = ctx.hypercomb(h, [n], **kwargs)
+            if ctx._is_real_type(z) and ctx.isint(n):
+                v = ctx._re(v)
+            return v
+    else:
+        def h():
+            if ctx._re(z) > 4:
+                # We could use 1F1, but it results in huge cancellation;
+                # the following expansion is better.
+                # TODO: asymptotic series for derivatives
+                ctx.prec += extraprec
+                w = z**1.5; r = -0.75/w; u = -2*w/3
+                ctx.prec -= extraprec
+                C = ctx.exp(u)/(2*ctx.sqrt(ctx.pi)*ctx.nthroot(z,4))
+                return ([C],[1],[],[],[(1,6),(5,6)],[],r),
+            else:
+                ctx.prec += extraprec
+                w = z**3 / 9
+                ctx.prec -= extraprec
+                C1 = _airyai_C1(ctx)
+                C2 = _airyai_C2(ctx)
+                T1 = [C1],[1],[],[],[],[ctx.mpq_2_3],w
+                T2 = [z*C2],[1],[],[],[],[ctx.mpq_4_3],w
+                return T1, T2
+        return ctx.hypercomb(h, [], **kwargs)
+
+@defun
+def airybi(ctx, z, derivative=0, **kwargs):
+    z = ctx.convert(z)
+    if derivative:
+        n, ntype = ctx._convert_param(derivative)
+    else:
+        n = 0
+    # Values at infinities
+    if not ctx.isnormal(z) and z:
+        if n and ntype == 'Z':
+            if z == ctx.inf:
+                return z
+            if z == ctx.ninf:
+                if n == -1:
+                    return 1/z
+                if n == -2:
+                    return _airybi_n2_inf(ctx)
+                if n < -2:
+                    return (-1)**n * (-z)
+        if not n:
+            if z == ctx.inf:
+                return z
+            if z == ctx.ninf:
+                return 1/z
+        # TODO: limits
+        raise ValueError("essential singularity of Bi(z)")
+    if z:
+        extraprec = max(0, int(1.5*ctx.mag(z)))
+    else:
+        extraprec = 0
+    if n:
+        if n == 1:
+            # http://functions.wolfram.com/03.08.26.0001.01
+            def h():
+                ctx.prec += extraprec
+                w = z**3 / 9
+                ctx.prec -= extraprec
+                C1 = _airybi_C1(ctx)*0.5
+                C2 = _airybi_C2(ctx)
+                T1 = [C1,z],[1,2],[],[],[],[ctx.mpq_5_3],w
+                T2 = [C2],[1],[],[],[],[ctx.mpq_1_3],w
+                return T1, T2
+            return ctx.hypercomb(h, [], **kwargs)
+        else:
+            if z == 0:
+                return _airyderiv_0(ctx, z, n, ntype, 1)
+            def h(n):
+                ctx.prec += extraprec
+                w = z**3/9
+                ctx.prec -= extraprec
+                q13,q23,q43 = ctx.mpq_1_3, ctx.mpq_2_3, ctx.mpq_4_3
+                q16 = ctx.mpq_1_6
+                q56 = ctx.mpq_5_6
+                a1=q13; a2=1; b1=(1-n)*q13; b2=(2-n)*q13; b3=1-n*q13
+                T1 = [3, z], [n-q16, -n], [a1], [b1,b2,b3], \
+                    [a1,a2], [b1,b2,b3], w
+                a1=q23; b1=(2-n)*q13; b2=1-n*q13; b3=(4-n)*q13
+                T2 = [3, z], [n-q56, 1-n], [a1], [b1,b2,b3], \
+                    [a1,a2], [b1,b2,b3], w
+                return T1, T2
+            v = ctx.hypercomb(h, [n], **kwargs)
+            if ctx._is_real_type(z) and ctx.isint(n):
+                v = ctx._re(v)
+            return v
+    else:
+        def h():
+            ctx.prec += extraprec
+            w = z**3 / 9
+            ctx.prec -= extraprec
+            C1 = _airybi_C1(ctx)
+            C2 = _airybi_C2(ctx)
+            T1 = [C1],[1],[],[],[],[ctx.mpq_2_3],w
+            T2 = [z*C2],[1],[],[],[],[ctx.mpq_4_3],w
+            return T1, T2
+        return ctx.hypercomb(h, [], **kwargs)
+
+def _airy_zero(ctx, which, k, derivative, complex=False):
+    # Asymptotic formulas are given in DLMF section 9.9
+    def U(t): return t**(2/3.)*(1-7/(t**2*48))
+    def T(t): return t**(2/3.)*(1+5/(t**2*48))
+    k = int(k)
+    if k < 1:
+        raise ValueError("k cannot be less than 1")
+    if not derivative in (0,1):
+        raise ValueError("Derivative should lie between 0 and 1")
+    if which == 0:
+        if derivative:
+            return ctx.findroot(lambda z: ctx.airyai(z,1),
+                -U(3*ctx.pi*(4*k-3)/8))
+        return ctx.findroot(ctx.airyai, -T(3*ctx.pi*(4*k-1)/8))
+    if which == 1 and complex == False:
+        if derivative:
+            return ctx.findroot(lambda z: ctx.airybi(z,1),
+                -U(3*ctx.pi*(4*k-1)/8))
+        return ctx.findroot(ctx.airybi, -T(3*ctx.pi*(4*k-3)/8))
+    if which == 1 and complex == True:
+        if derivative:
+            t = 3*ctx.pi*(4*k-3)/8 + 0.75j*ctx.ln2
+            s = ctx.expjpi(ctx.mpf(1)/3) * T(t)
+            return ctx.findroot(lambda z: ctx.airybi(z,1), s)
+        t = 3*ctx.pi*(4*k-1)/8 + 0.75j*ctx.ln2
+        s = ctx.expjpi(ctx.mpf(1)/3) * U(t)
+        return ctx.findroot(ctx.airybi, s)
+
+@defun
+def airyaizero(ctx, k, derivative=0):
+    return _airy_zero(ctx, 0, k, derivative, False)
+
+@defun
+def airybizero(ctx, k, derivative=0, complex=False):
+    return _airy_zero(ctx, 1, k, derivative, complex)
+
+def _scorer(ctx, z, which, kwargs):
+    z = ctx.convert(z)
+    if ctx.isinf(z):
+        if z == ctx.inf:
+            if which == 0: return 1/z
+            if which == 1: return z
+        if z == ctx.ninf:
+            return 1/z
+        raise ValueError("essential singularity")
+    if z:
+        extraprec = max(0, int(1.5*ctx.mag(z)))
+    else:
+        extraprec = 0
+    if kwargs.get('derivative'):
+        raise NotImplementedError
+    # Direct asymptotic expansions, to avoid
+    # exponentially large cancellation
+    try:
+        if ctx.mag(z) > 3:
+            if which == 0 and abs(ctx.arg(z)) < ctx.pi/3 * 0.999:
+                def h():
+                    return (([ctx.pi,z],[-1,-1],[],[],[(1,3),(2,3),1],[],9/z**3),)
+                return ctx.hypercomb(h, [], maxterms=ctx.prec, force_series=True)
+            if which == 1 and abs(ctx.arg(-z)) < 2*ctx.pi/3 * 0.999:
+                def h():
+                    return (([-ctx.pi,z],[-1,-1],[],[],[(1,3),(2,3),1],[],9/z**3),)
+                return ctx.hypercomb(h, [], maxterms=ctx.prec, force_series=True)
+    except ctx.NoConvergence:
+        pass
+    def h():
+        A = ctx.airybi(z, **kwargs)/3
+        B = -2*ctx.pi
+        if which == 1:
+            A *= 2
+            B *= -1
+        ctx.prec += extraprec
+        w = z**3/9
+        ctx.prec -= extraprec
+        T1 = [A], [1], [], [], [], [], 0
+        T2 = [B,z], [-1,2], [], [], [1], [ctx.mpq_4_3,ctx.mpq_5_3], w
+        return T1, T2
+    return ctx.hypercomb(h, [], **kwargs)
+
+@defun
+def scorergi(ctx, z, **kwargs):
+    return _scorer(ctx, z, 0, kwargs)
+
+@defun
+def scorerhi(ctx, z, **kwargs):
+    return _scorer(ctx, z, 1, kwargs)
+
+@defun_wrapped
+def coulombc(ctx, l, eta, _cache={}):
+    if (l, eta) in _cache and _cache[l,eta][0] >= ctx.prec:
+        return +_cache[l,eta][1]
+    G3 = ctx.loggamma(2*l+2)
+    G1 = ctx.loggamma(1+l+ctx.j*eta)
+    G2 = ctx.loggamma(1+l-ctx.j*eta)
+    v = 2**l * ctx.exp((-ctx.pi*eta+G1+G2)/2 - G3)
+    if not (ctx.im(l) or ctx.im(eta)):
+        v = ctx.re(v)
+    _cache[l,eta] = (ctx.prec, v)
+    return v
+
+@defun_wrapped
+def coulombf(ctx, l, eta, z, w=1, chop=True, **kwargs):
+    # Regular Coulomb wave function
+    # Note: w can be either 1 or -1; the other may be better in some cases
+    # TODO: check that chop=True chops when and only when it should
+    #ctx.prec += 10
+    def h(l, eta):
+        try:
+            jw = ctx.j*w
+            jwz = ctx.fmul(jw, z, exact=True)
+            jwz2 = ctx.fmul(jwz, -2, exact=True)
+            C = ctx.coulombc(l, eta)
+            T1 = [C, z, ctx.exp(jwz)], [1, l+1, 1], [], [], [1+l+jw*eta], \
+                [2*l+2], jwz2
+        except ValueError:
+            T1 = [0], [-1], [], [], [], [], 0
+        return (T1,)
+    v = ctx.hypercomb(h, [l,eta], **kwargs)
+    if chop and (not ctx.im(l)) and (not ctx.im(eta)) and (not ctx.im(z)) and \
+        (ctx.re(z) >= 0):
+        v = ctx.re(v)
+    return v
+
+@defun_wrapped
+def _coulomb_chi(ctx, l, eta, _cache={}):
+    if (l, eta) in _cache and _cache[l,eta][0] >= ctx.prec:
+        return _cache[l,eta][1]
+    def terms():
+        l2 = -l-1
+        jeta = ctx.j*eta
+        return [ctx.loggamma(1+l+jeta) * (-0.5j),
+            ctx.loggamma(1+l-jeta) * (0.5j),
+            ctx.loggamma(1+l2+jeta) * (0.5j),
+            ctx.loggamma(1+l2-jeta) * (-0.5j),
+            -(l+0.5)*ctx.pi]
+    v = ctx.sum_accurately(terms, 1)
+    _cache[l,eta] = (ctx.prec, v)
+    return v
+
+@defun_wrapped
+def coulombg(ctx, l, eta, z, w=1, chop=True, **kwargs):
+    # Irregular Coulomb wave function
+    # Note: w can be either 1 or -1; the other may be better in some cases
+    # TODO: check that chop=True chops when and only when it should
+    if not ctx._im(l):
+        l = ctx._re(l)  # XXX: for isint
+    def h(l, eta):
+        # Force perturbation for integers and half-integers
+        if ctx.isint(l*2):
+            T1 = [0], [-1], [], [], [], [], 0
+            return (T1,)
+        l2 = -l-1
+        try:
+            chi = ctx._coulomb_chi(l, eta)
+            jw = ctx.j*w
+            s = ctx.sin(chi); c = ctx.cos(chi)
+            C1 = ctx.coulombc(l,eta)
+            C2 = ctx.coulombc(l2,eta)
+            u = ctx.exp(jw*z)
+            x = -2*jw*z
+            T1 = [s, C1, z, u, c], [-1, 1, l+1, 1, 1], [], [], \
+                [1+l+jw*eta], [2*l+2], x
+            T2 = [-s, C2, z, u],   [-1, 1, l2+1, 1],    [], [], \
+                [1+l2+jw*eta], [2*l2+2], x
+            return T1, T2
+        except ValueError:
+            T1 = [0], [-1], [], [], [], [], 0
+            return (T1,)
+    v = ctx.hypercomb(h, [l,eta], **kwargs)
+    if chop and (not ctx._im(l)) and (not ctx._im(eta)) and (not ctx._im(z)) and \
+        (ctx._re(z) >= 0):
+        v = ctx._re(v)
+    return v
+
+def mcmahon(ctx,kind,prime,v,m):
+    """
+    Computes an estimate for the location of the Bessel function zero
+    j_{v,m}, y_{v,m}, j'_{v,m} or y'_{v,m} using McMahon's asymptotic
+    expansion (Abramowitz & Stegun 9.5.12-13, DLMF 20.21(vi)).
+
+    Returns (r,err) where r is the estimated location of the root
+    and err is a positive number estimating the error of the
+    asymptotic expansion.
+    """
+    u = 4*v**2
+    if kind == 1 and not prime: b = (4*m+2*v-1)*ctx.pi/4
+    if kind == 2 and not prime: b = (4*m+2*v-3)*ctx.pi/4
+    if kind == 1 and prime: b = (4*m+2*v-3)*ctx.pi/4
+    if kind == 2 and prime: b = (4*m+2*v-1)*ctx.pi/4
+    if not prime:
+        s1 = b
+        s2 = -(u-1)/(8*b)
+        s3 = -4*(u-1)*(7*u-31)/(3*(8*b)**3)
+        s4 = -32*(u-1)*(83*u**2-982*u+3779)/(15*(8*b)**5)
+        s5 = -64*(u-1)*(6949*u**3-153855*u**2+1585743*u-6277237)/(105*(8*b)**7)
+    if prime:
+        s1 = b
+        s2 = -(u+3)/(8*b)
+        s3 = -4*(7*u**2+82*u-9)/(3*(8*b)**3)
+        s4 = -32*(83*u**3+2075*u**2-3039*u+3537)/(15*(8*b)**5)
+        s5 = -64*(6949*u**4+296492*u**3-1248002*u**2+7414380*u-5853627)/(105*(8*b)**7)
+    terms = [s1,s2,s3,s4,s5]
+    s = s1
+    err = 0.0
+    for i in range(1,len(terms)):
+        if abs(terms[i]) < abs(terms[i-1]):
+            s += terms[i]
+        else:
+            err = abs(terms[i])
+    if i == len(terms)-1:
+        err = abs(terms[-1])
+    return s, err
+
+def generalized_bisection(ctx,f,a,b,n):
+    """
+    Given f known to have exactly n simple roots within [a,b],
+    return a list of n intervals isolating the roots
+    and having opposite signs at the endpoints.
+
+    TODO: this can be optimized, e.g. by reusing evaluation points.
+    """
+    if n < 1:
+        raise ValueError("n cannot be less than 1")
+    N = n+1
+    points = []
+    signs = []
+    while 1:
+        points = ctx.linspace(a,b,N)
+        signs = [ctx.sign(f(x)) for x in points]
+        ok_intervals = [(points[i],points[i+1]) for i in range(N-1) \
+            if signs[i]*signs[i+1] == -1]
+        if len(ok_intervals) == n:
+            return ok_intervals
+        N = N*2
+
+def find_in_interval(ctx, f, ab):
+    return ctx.findroot(f, ab, solver='illinois', verify=False)
+
+def bessel_zero(ctx, kind, prime, v, m, isoltol=0.01, _interval_cache={}):
+    prec = ctx.prec
+    workprec = max(prec, ctx.mag(v), ctx.mag(m))+10
+    try:
+        ctx.prec = workprec
+        v = ctx.mpf(v)
+        m = int(m)
+        prime = int(prime)
+        if v < 0:
+            raise ValueError("v cannot be negative")
+        if m < 1:
+            raise ValueError("m cannot be less than 1")
+        if not prime in (0,1):
+            raise ValueError("prime should lie between 0 and 1")
+        if kind == 1:
+            if prime: f = lambda x: ctx.besselj(v,x,derivative=1)
+            else:     f = lambda x: ctx.besselj(v,x)
+        if kind == 2:
+            if prime: f = lambda x: ctx.bessely(v,x,derivative=1)
+            else:     f = lambda x: ctx.bessely(v,x)
+        # The first root of J' is very close to 0 for small
+        # orders, and this needs to be special-cased
+        if kind == 1 and prime and m == 1:
+            if v == 0:
+                return ctx.zero
+            if v <= 1:
+                # TODO: use v <= j'_{v,1} < y_{v,1}?
+                r = 2*ctx.sqrt(v*(1+v)/(v+2))
+                return find_in_interval(ctx, f, (r/10, 2*r))
+        if (kind,prime,v,m) in _interval_cache:
+            return find_in_interval(ctx, f, _interval_cache[kind,prime,v,m])
+        r, err = mcmahon(ctx, kind, prime, v, m)
+        if err < isoltol:
+            return find_in_interval(ctx, f, (r-isoltol, r+isoltol))
+        # An x such that 0 < x < r_{v,1}
+        if kind == 1 and not prime: low = 2.4
+        if kind == 1 and prime: low = 1.8
+        if kind == 2 and not prime: low = 0.8
+        if kind == 2 and prime: low = 2.0
+        n = m+1
+        while 1:
+            r1, err = mcmahon(ctx, kind, prime, v, n)
+            if err < isoltol:
+                r2, err2 = mcmahon(ctx, kind, prime, v, n+1)
+                intervals = generalized_bisection(ctx, f, low, 0.5*(r1+r2), n)
+                for k, ab in enumerate(intervals):
+                    _interval_cache[kind,prime,v,k+1] = ab
+                return find_in_interval(ctx, f, intervals[m-1])
+            else:
+                n = n*2
+    finally:
+        ctx.prec = prec
+
+@defun
+def besseljzero(ctx, v, m, derivative=0):
+    r"""
+    For a real order `\nu \ge 0` and a positive integer `m`, returns
+    `j_{\nu,m}`, the `m`-th positive zero of the Bessel function of the
+    first kind `J_{\nu}(z)` (see :func:`~mpmath.besselj`). Alternatively,
+    with *derivative=1*, gives the first nonnegative simple zero
+    `j'_{\nu,m}` of `J'_{\nu}(z)`.
+
+    The indexing convention is that used by Abramowitz & Stegun
+    and the DLMF. Note the special case `j'_{0,1} = 0`, while all other
+    zeros are positive. In effect, only simple zeros are counted
+    (all zeros of Bessel functions are simple except possibly `z = 0`)
+    and `j_{\nu,m}` becomes a monotonic function of both `\nu`
+    and `m`.
+
+    The zeros are interlaced according to the inequalities
+
+    .. math ::
+
+        j'_{\nu,k} < j_{\nu,k} < j'_{\nu,k+1}
+
+        j_{\nu,1} < j_{\nu+1,2} < j_{\nu,2} < j_{\nu+1,2} < j_{\nu,3} < \cdots
+
+    **Examples**
+
+    Initial zeros of the Bessel functions `J_0(z), J_1(z), J_2(z)`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> besseljzero(0,1); besseljzero(0,2); besseljzero(0,3)
+        2.404825557695772768621632
+        5.520078110286310649596604
+        8.653727912911012216954199
+        >>> besseljzero(1,1); besseljzero(1,2); besseljzero(1,3)
+        3.831705970207512315614436
+        7.01558666981561875353705
+        10.17346813506272207718571
+        >>> besseljzero(2,1); besseljzero(2,2); besseljzero(2,3)
+        5.135622301840682556301402
+        8.417244140399864857783614
+        11.61984117214905942709415
+
+    Initial zeros of `J'_0(z), J'_1(z), J'_2(z)`::
+
+        0.0
+        3.831705970207512315614436
+        7.01558666981561875353705
+        >>> besseljzero(1,1,1); besseljzero(1,2,1); besseljzero(1,3,1)
+        1.84118378134065930264363
+        5.331442773525032636884016
+        8.536316366346285834358961
+        >>> besseljzero(2,1,1); besseljzero(2,2,1); besseljzero(2,3,1)
+        3.054236928227140322755932
+        6.706133194158459146634394
+        9.969467823087595793179143
+
+    Zeros with large index::
+
+        >>> besseljzero(0,100); besseljzero(0,1000); besseljzero(0,10000)
+        313.3742660775278447196902
+        3140.807295225078628895545
+        31415.14114171350798533666
+        >>> besseljzero(5,100); besseljzero(5,1000); besseljzero(5,10000)
+        321.1893195676003157339222
+        3148.657306813047523500494
+        31422.9947255486291798943
+        >>> besseljzero(0,100,1); besseljzero(0,1000,1); besseljzero(0,10000,1)
+        311.8018681873704508125112
+        3139.236339643802482833973
+        31413.57032947022399485808
+
+    Zeros of functions with large order::
+
+        >>> besseljzero(50,1)
+        57.11689916011917411936228
+        >>> besseljzero(50,2)
+        62.80769876483536093435393
+        >>> besseljzero(50,100)
+        388.6936600656058834640981
+        >>> besseljzero(50,1,1)
+        52.99764038731665010944037
+        >>> besseljzero(50,2,1)
+        60.02631933279942589882363
+        >>> besseljzero(50,100,1)
+        387.1083151608726181086283
+
+    Zeros of functions with fractional order::
+
+        >>> besseljzero(0.5,1); besseljzero(1.5,1); besseljzero(2.25,4)
+        3.141592653589793238462643
+        4.493409457909064175307881
+        15.15657692957458622921634
+
+    Both `J_{\nu}(z)` and `J'_{\nu}(z)` can be expressed as infinite
+    products over their zeros::
+
+        >>> v,z = 2, mpf(1)
+        >>> (z/2)**v/gamma(v+1) * \
+        ...     nprod(lambda k: 1-(z/besseljzero(v,k))**2, [1,inf])
+        ...
+        0.1149034849319004804696469
+        >>> besselj(v,z)
+        0.1149034849319004804696469
+        >>> (z/2)**(v-1)/2/gamma(v) * \
+        ...     nprod(lambda k: 1-(z/besseljzero(v,k,1))**2, [1,inf])
+        ...
+        0.2102436158811325550203884
+        >>> besselj(v,z,1)
+        0.2102436158811325550203884
+
+    """
+    return +bessel_zero(ctx, 1, derivative, v, m)
+
+@defun
+def besselyzero(ctx, v, m, derivative=0):
+    r"""
+    For a real order `\nu \ge 0` and a positive integer `m`, returns
+    `y_{\nu,m}`, the `m`-th positive zero of the Bessel function of the
+    second kind `Y_{\nu}(z)` (see :func:`~mpmath.bessely`). Alternatively,
+    with *derivative=1*, gives the first positive zero `y'_{\nu,m}` of
+    `Y'_{\nu}(z)`.
+
+    The zeros are interlaced according to the inequalities
+
+    .. math ::
+
+        y_{\nu,k} < y'_{\nu,k} < y_{\nu,k+1}
+
+        y_{\nu,1} < y_{\nu+1,2} < y_{\nu,2} < y_{\nu+1,2} < y_{\nu,3} < \cdots
+
+    **Examples**
+
+    Initial zeros of the Bessel functions `Y_0(z), Y_1(z), Y_2(z)`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> besselyzero(0,1); besselyzero(0,2); besselyzero(0,3)
+        0.8935769662791675215848871
+        3.957678419314857868375677
+        7.086051060301772697623625
+        >>> besselyzero(1,1); besselyzero(1,2); besselyzero(1,3)
+        2.197141326031017035149034
+        5.429681040794135132772005
+        8.596005868331168926429606
+        >>> besselyzero(2,1); besselyzero(2,2); besselyzero(2,3)
+        3.384241767149593472701426
+        6.793807513268267538291167
+        10.02347797936003797850539
+
+    Initial zeros of `Y'_0(z), Y'_1(z), Y'_2(z)`::
+
+        >>> besselyzero(0,1,1); besselyzero(0,2,1); besselyzero(0,3,1)
+        2.197141326031017035149034
+        5.429681040794135132772005
+        8.596005868331168926429606
+        >>> besselyzero(1,1,1); besselyzero(1,2,1); besselyzero(1,3,1)
+        3.683022856585177699898967
+        6.941499953654175655751944
+        10.12340465543661307978775
+        >>> besselyzero(2,1,1); besselyzero(2,2,1); besselyzero(2,3,1)
+        5.002582931446063945200176
+        8.350724701413079526349714
+        11.57419546521764654624265
+
+    Zeros with large index::
+
+        >>> besselyzero(0,100); besselyzero(0,1000); besselyzero(0,10000)
+        311.8034717601871549333419
+        3139.236498918198006794026
+        31413.57034538691205229188
+        >>> besselyzero(5,100); besselyzero(5,1000); besselyzero(5,10000)
+        319.6183338562782156235062
+        3147.086508524556404473186
+        31421.42392920214673402828
+        >>> besselyzero(0,100,1); besselyzero(0,1000,1); besselyzero(0,10000,1)
+        313.3726705426359345050449
+        3140.807136030340213610065
+        31415.14112579761578220175
+
+    Zeros of functions with large order::
+
+        >>> besselyzero(50,1)
+        53.50285882040036394680237
+        >>> besselyzero(50,2)
+        60.11244442774058114686022
+        >>> besselyzero(50,100)
+        387.1096509824943957706835
+        >>> besselyzero(50,1,1)
+        56.96290427516751320063605
+        >>> besselyzero(50,2,1)
+        62.74888166945933944036623
+        >>> besselyzero(50,100,1)
+        388.6923300548309258355475
+
+    Zeros of functions with fractional order::
+
+        >>> besselyzero(0.5,1); besselyzero(1.5,1); besselyzero(2.25,4)
+        1.570796326794896619231322
+        2.798386045783887136720249
+        13.56721208770735123376018
+
+    """
+    return +bessel_zero(ctx, 2, derivative, v, m)

lib/python3.11/site-packages/mpmath/functions/elliptic.py

@@ -0,0 +1,1431 @@
+r"""
+Elliptic functions historically comprise the elliptic integrals
+and their inverses, and originate from the problem of computing the
+arc length of an ellipse. From a more modern point of view,
+an elliptic function is defined as a doubly periodic function, i.e.
+a function which satisfies
+
+.. math ::
+
+    f(z + 2 \omega_1) = f(z + 2 \omega_2) = f(z)
+
+for some half-periods `\omega_1, \omega_2` with
+`\mathrm{Im}[\omega_1 / \omega_2] > 0`. The canonical elliptic
+functions are the Jacobi elliptic functions. More broadly, this section
+includes  quasi-doubly periodic functions (such as the Jacobi theta
+functions) and other functions useful in the study of elliptic functions.
+
+Many different conventions for the arguments of
+elliptic functions are in use. It is even standard to use
+different parameterizations for different functions in the same
+text or software (and mpmath is no exception).
+The usual parameters are the elliptic nome `q`, which usually
+must satisfy `|q| < 1`; the elliptic parameter `m` (an arbitrary
+complex number); the elliptic modulus `k` (an arbitrary complex
+number); and the half-period ratio `\tau`, which usually must
+satisfy `\mathrm{Im}[\tau] > 0`.
+These quantities can be expressed in terms of each other
+using the following relations:
+
+.. math ::
+
+    m = k^2
+
+.. math ::
+
+    \tau = i \frac{K(1-m)}{K(m)}
+
+.. math ::
+
+    q = e^{i \pi \tau}
+
+.. math ::
+
+    k = \frac{\vartheta_2^2(q)}{\vartheta_3^2(q)}
+
+In addition, an alternative definition is used for the nome in
+number theory, which we here denote by q-bar:
+
+.. math ::
+
+    \bar{q} = q^2 = e^{2 i \pi \tau}
+
+For convenience, mpmath provides functions to convert
+between the various parameters (:func:`~mpmath.qfrom`, :func:`~mpmath.mfrom`,
+:func:`~mpmath.kfrom`, :func:`~mpmath.taufrom`, :func:`~mpmath.qbarfrom`).
+
+**References**
+
+1. [AbramowitzStegun]_
+
+2. [WhittakerWatson]_
+
+"""
+
+from .functions import defun, defun_wrapped
+
+@defun_wrapped
+def eta(ctx, tau):
+    r"""
+    Returns the Dedekind eta function of tau in the upper half-plane.
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> eta(1j); gamma(0.25) / (2*pi**0.75)
+        (0.7682254223260566590025942 + 0.0j)
+        0.7682254223260566590025942
+        >>> tau = sqrt(2) + sqrt(5)*1j
+        >>> eta(-1/tau); sqrt(-1j*tau) * eta(tau)
+        (0.9022859908439376463573294 + 0.07985093673948098408048575j)
+        (0.9022859908439376463573295 + 0.07985093673948098408048575j)
+        >>> eta(tau+1); exp(pi*1j/12) * eta(tau)
+        (0.4493066139717553786223114 + 0.3290014793877986663915939j)
+        (0.4493066139717553786223114 + 0.3290014793877986663915939j)
+        >>> f = lambda z: diff(eta, z) / eta(z)
+        >>> chop(36*diff(f,tau)**2 - 24*diff(f,tau,2)*f(tau) + diff(f,tau,3))
+        0.0
+
+    """
+    if ctx.im(tau) <= 0.0:
+        raise ValueError("eta is only defined in the upper half-plane")
+    q = ctx.expjpi(tau/12)
+    return q * ctx.qp(q**24)
+
+def nome(ctx, m):
+    m = ctx.convert(m)
+    if not m:
+        return m
+    if m == ctx.one:
+        return m
+    if ctx.isnan(m):
+        return m
+    if ctx.isinf(m):
+        if m == ctx.ninf:
+            return type(m)(-1)
+        else:
+            return ctx.mpc(-1)
+    a = ctx.ellipk(ctx.one-m)
+    b = ctx.ellipk(m)
+    v = ctx.exp(-ctx.pi*a/b)
+    if not ctx._im(m) and ctx._re(m) < 1:
+        if ctx._is_real_type(m):
+            return v.real
+        else:
+            return v.real + 0j
+    elif m == 2:
+        v = ctx.mpc(0, v.imag)
+    return v
+
+@defun_wrapped
+def qfrom(ctx, q=None, m=None, k=None, tau=None, qbar=None):
+    r"""
+    Returns the elliptic nome `q`, given any of `q, m, k, \tau, \bar{q}`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> qfrom(q=0.25)
+        0.25
+        >>> qfrom(m=mfrom(q=0.25))
+        0.25
+        >>> qfrom(k=kfrom(q=0.25))
+        0.25
+        >>> qfrom(tau=taufrom(q=0.25))
+        (0.25 + 0.0j)
+        >>> qfrom(qbar=qbarfrom(q=0.25))
+        0.25
+
+    """
+    if q is not None:
+        return ctx.convert(q)
+    if m is not None:
+        return nome(ctx, m)
+    if k is not None:
+        return nome(ctx, ctx.convert(k)**2)
+    if tau is not None:
+        return ctx.expjpi(tau)
+    if qbar is not None:
+        return ctx.sqrt(qbar)
+
+@defun_wrapped
+def qbarfrom(ctx, q=None, m=None, k=None, tau=None, qbar=None):
+    r"""
+    Returns the number-theoretic nome `\bar q`, given any of
+    `q, m, k, \tau, \bar{q}`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> qbarfrom(qbar=0.25)
+        0.25
+        >>> qbarfrom(q=qfrom(qbar=0.25))
+        0.25
+        >>> qbarfrom(m=extraprec(20)(mfrom)(qbar=0.25))  # ill-conditioned
+        0.25
+        >>> qbarfrom(k=extraprec(20)(kfrom)(qbar=0.25))  # ill-conditioned
+        0.25
+        >>> qbarfrom(tau=taufrom(qbar=0.25))
+        (0.25 + 0.0j)
+
+    """
+    if qbar is not None:
+        return ctx.convert(qbar)
+    if q is not None:
+        return ctx.convert(q) ** 2
+    if m is not None:
+        return nome(ctx, m) ** 2
+    if k is not None:
+        return nome(ctx, ctx.convert(k)**2) ** 2
+    if tau is not None:
+        return ctx.expjpi(2*tau)
+
+@defun_wrapped
+def taufrom(ctx, q=None, m=None, k=None, tau=None, qbar=None):
+    r"""
+    Returns the elliptic half-period ratio `\tau`, given any of
+    `q, m, k, \tau, \bar{q}`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> taufrom(tau=0.5j)
+        (0.0 + 0.5j)
+        >>> taufrom(q=qfrom(tau=0.5j))
+        (0.0 + 0.5j)
+        >>> taufrom(m=mfrom(tau=0.5j))
+        (0.0 + 0.5j)
+        >>> taufrom(k=kfrom(tau=0.5j))
+        (0.0 + 0.5j)
+        >>> taufrom(qbar=qbarfrom(tau=0.5j))
+        (0.0 + 0.5j)
+
+    """
+    if tau is not None:
+        return ctx.convert(tau)
+    if m is not None:
+        m = ctx.convert(m)
+        return ctx.j*ctx.ellipk(1-m)/ctx.ellipk(m)
+    if k is not None:
+        k = ctx.convert(k)
+        return ctx.j*ctx.ellipk(1-k**2)/ctx.ellipk(k**2)
+    if q is not None:
+        return ctx.log(q) / (ctx.pi*ctx.j)
+    if qbar is not None:
+        qbar = ctx.convert(qbar)
+        return ctx.log(qbar) / (2*ctx.pi*ctx.j)
+
+@defun_wrapped
+def kfrom(ctx, q=None, m=None, k=None, tau=None, qbar=None):
+    r"""
+    Returns the elliptic modulus `k`, given any of
+    `q, m, k, \tau, \bar{q}`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> kfrom(k=0.25)
+        0.25
+        >>> kfrom(m=mfrom(k=0.25))
+        0.25
+        >>> kfrom(q=qfrom(k=0.25))
+        0.25
+        >>> kfrom(tau=taufrom(k=0.25))
+        (0.25 + 0.0j)
+        >>> kfrom(qbar=qbarfrom(k=0.25))
+        0.25
+
+    As `q \to 1` and `q \to -1`, `k` rapidly approaches
+    `1` and `i \infty` respectively::
+
+        >>> kfrom(q=0.75)
+        0.9999999999999899166471767
+        >>> kfrom(q=-0.75)
+        (0.0 + 7041781.096692038332790615j)
+        >>> kfrom(q=1)
+        1
+        >>> kfrom(q=-1)
+        (0.0 + +infj)
+    """
+    if k is not None:
+        return ctx.convert(k)
+    if m is not None:
+        return ctx.sqrt(m)
+    if tau is not None:
+        q = ctx.expjpi(tau)
+    if qbar is not None:
+        q = ctx.sqrt(qbar)
+    if q == 1:
+        return q
+    if q == -1:
+        return ctx.mpc(0,'inf')
+    return (ctx.jtheta(2,0,q)/ctx.jtheta(3,0,q))**2
+
+@defun_wrapped
+def mfrom(ctx, q=None, m=None, k=None, tau=None, qbar=None):
+    r"""
+    Returns the elliptic parameter `m`, given any of
+    `q, m, k, \tau, \bar{q}`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> mfrom(m=0.25)
+        0.25
+        >>> mfrom(q=qfrom(m=0.25))
+        0.25
+        >>> mfrom(k=kfrom(m=0.25))
+        0.25
+        >>> mfrom(tau=taufrom(m=0.25))
+        (0.25 + 0.0j)
+        >>> mfrom(qbar=qbarfrom(m=0.25))
+        0.25
+
+    As `q \to 1` and `q \to -1`, `m` rapidly approaches
+    `1` and `-\infty` respectively::
+
+        >>> mfrom(q=0.75)
+        0.9999999999999798332943533
+        >>> mfrom(q=-0.75)
+        -49586681013729.32611558353
+        >>> mfrom(q=1)
+        1.0
+        >>> mfrom(q=-1)
+        -inf
+
+    The inverse nome as a function of `q` has an integer
+    Taylor series expansion::
+
+        >>> taylor(lambda q: mfrom(q), 0, 7)
+        [0.0, 16.0, -128.0, 704.0, -3072.0, 11488.0, -38400.0, 117632.0]
+
+    """
+    if m is not None:
+        return m
+    if k is not None:
+        return k**2
+    if tau is not None:
+        q = ctx.expjpi(tau)
+    if qbar is not None:
+        q = ctx.sqrt(qbar)
+    if q == 1:
+        return ctx.convert(q)
+    if q == -1:
+        return q*ctx.inf
+    v = (ctx.jtheta(2,0,q)/ctx.jtheta(3,0,q))**4
+    if ctx._is_real_type(q) and q < 0:
+        v = v.real
+    return v
+
+jacobi_spec = {
+  'sn' : ([3],[2],[1],[4], 'sin', 'tanh'),
+  'cn' : ([4],[2],[2],[4], 'cos', 'sech'),
+  'dn' : ([4],[3],[3],[4], '1', 'sech'),
+  'ns' : ([2],[3],[4],[1], 'csc', 'coth'),
+  'nc' : ([2],[4],[4],[2], 'sec', 'cosh'),
+  'nd' : ([3],[4],[4],[3], '1', 'cosh'),
+  'sc' : ([3],[4],[1],[2], 'tan', 'sinh'),
+  'sd' : ([3,3],[2,4],[1],[3], 'sin', 'sinh'),
+  'cd' : ([3],[2],[2],[3], 'cos', '1'),
+  'cs' : ([4],[3],[2],[1], 'cot', 'csch'),
+  'dc' : ([2],[3],[3],[2], 'sec', '1'),
+  'ds' : ([2,4],[3,3],[3],[1], 'csc', 'csch'),
+  'cc' : None,
+  'ss' : None,
+  'nn' : None,
+  'dd' : None
+}
+
+@defun
+def ellipfun(ctx, kind, u=None, m=None, q=None, k=None, tau=None):
+    try:
+        S = jacobi_spec[kind]
+    except KeyError:
+        raise ValueError("First argument must be a two-character string "
+            "containing 's', 'c', 'd' or 'n', e.g.: 'sn'")
+    if u is None:
+        def f(*args, **kwargs):
+            return ctx.ellipfun(kind, *args, **kwargs)
+        f.__name__ = kind
+        return f
+    prec = ctx.prec
+    try:
+        ctx.prec += 10
+        u = ctx.convert(u)
+        q = ctx.qfrom(m=m, q=q, k=k, tau=tau)
+        if S is None:
+            v = ctx.one + 0*q*u
+        elif q == ctx.zero:
+            if S[4] == '1': v = ctx.one
+            else:           v = getattr(ctx, S[4])(u)
+            v += 0*q*u
+        elif q == ctx.one:
+            if S[5] == '1': v = ctx.one
+            else:           v = getattr(ctx, S[5])(u)
+            v += 0*q*u
+        else:
+            t = u / ctx.jtheta(3, 0, q)**2
+            v = ctx.one
+            for a in S[0]: v *= ctx.jtheta(a, 0, q)
+            for b in S[1]: v /= ctx.jtheta(b, 0, q)
+            for c in S[2]: v *= ctx.jtheta(c, t, q)
+            for d in S[3]: v /= ctx.jtheta(d, t, q)
+    finally:
+        ctx.prec = prec
+    return +v
+
+@defun_wrapped
+def kleinj(ctx, tau=None, **kwargs):
+    r"""
+    Evaluates the Klein j-invariant, which is a modular function defined for
+    `\tau` in the upper half-plane as
+
+    .. math ::
+
+        J(\tau) = \frac{g_2^3(\tau)}{g_2^3(\tau) - 27 g_3^2(\tau)}
+
+    where `g_2` and `g_3` are the modular invariants of the Weierstrass
+    elliptic function,
+
+    .. math ::
+
+        g_2(\tau) = 60 \sum_{(m,n) \in \mathbb{Z}^2 \setminus (0,0)} (m \tau+n)^{-4}
+
+        g_3(\tau) = 140 \sum_{(m,n) \in \mathbb{Z}^2 \setminus (0,0)} (m \tau+n)^{-6}.
+
+    An alternative, common notation is that of the j-function
+    `j(\tau) = 1728 J(\tau)`.
+
+    **Plots**
+
+    .. literalinclude :: /plots/kleinj.py
+    .. image :: /plots/kleinj.png
+    .. literalinclude :: /plots/kleinj2.py
+    .. image :: /plots/kleinj2.png
+
+    **Examples**
+
+    Verifying the functional equation `J(\tau) = J(\tau+1) = J(-\tau^{-1})`::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> tau = 0.625+0.75*j
+        >>> tau = 0.625+0.75*j
+        >>> kleinj(tau)
+        (-0.1507492166511182267125242 + 0.07595948379084571927228948j)
+        >>> kleinj(tau+1)
+        (-0.1507492166511182267125242 + 0.07595948379084571927228948j)
+        >>> kleinj(-1/tau)
+        (-0.1507492166511182267125242 + 0.07595948379084571927228946j)
+
+    The j-function has a famous Laurent series expansion in terms of the nome
+    `\bar{q}`, `j(\tau) = \bar{q}^{-1} + 744 + 196884\bar{q} + \ldots`::
+
+        >>> mp.dps = 15
+        >>> taylor(lambda q: 1728*q*kleinj(qbar=q), 0, 5, singular=True)
+        [1.0, 744.0, 196884.0, 21493760.0, 864299970.0, 20245856256.0]
+
+    The j-function admits exact evaluation at special algebraic points
+    related to the Heegner numbers 1, 2, 3, 7, 11, 19, 43, 67, 163::
+
+        >>> @extraprec(10)
+        ... def h(n):
+        ...     v = (1+sqrt(n)*j)
+        ...     if n > 2:
+        ...         v *= 0.5
+        ...     return v
+        ...
+        >>> mp.dps = 25
+        >>> for n in [1,2,3,7,11,19,43,67,163]:
+        ...     n, chop(1728*kleinj(h(n)))
+        ...
+        (1, 1728.0)
+        (2, 8000.0)
+        (3, 0.0)
+        (7, -3375.0)
+        (11, -32768.0)
+        (19, -884736.0)
+        (43, -884736000.0)
+        (67, -147197952000.0)
+        (163, -262537412640768000.0)
+
+    Also at other special points, the j-function assumes explicit
+    algebraic values, e.g.::
+
+        >>> chop(1728*kleinj(j*sqrt(5)))
+        1264538.909475140509320227
+        >>> identify(cbrt(_))      # note: not simplified
+        '((100+sqrt(13520))/2)'
+        >>> (50+26*sqrt(5))**3
+        1264538.909475140509320227
+
+    """
+    q = ctx.qfrom(tau=tau, **kwargs)
+    t2 = ctx.jtheta(2,0,q)
+    t3 = ctx.jtheta(3,0,q)
+    t4 = ctx.jtheta(4,0,q)
+    P = (t2**8 + t3**8 + t4**8)**3
+    Q = 54*(t2*t3*t4)**8
+    return P/Q
+
+
+def RF_calc(ctx, x, y, z, r):
+    if y == z: return RC_calc(ctx, x, y, r)
+    if x == z: return RC_calc(ctx, y, x, r)
+    if x == y: return RC_calc(ctx, z, x, r)
+    if not (ctx.isnormal(x) and ctx.isnormal(y) and ctx.isnormal(z)):
+        if ctx.isnan(x) or ctx.isnan(y) or ctx.isnan(z):
+            return x*y*z
+        if ctx.isinf(x) or ctx.isinf(y) or ctx.isinf(z):
+            return ctx.zero
+    xm,ym,zm = x,y,z
+    A0 = Am = (x+y+z)/3
+    Q = ctx.root(3*r, -6) * max(abs(A0-x),abs(A0-y),abs(A0-z))
+    g = ctx.mpf(0.25)
+    pow4 = ctx.one
+    while 1:
+        xs = ctx.sqrt(xm)
+        ys = ctx.sqrt(ym)
+        zs = ctx.sqrt(zm)
+        lm = xs*ys + xs*zs + ys*zs
+        Am1 = (Am+lm)*g
+        xm, ym, zm = (xm+lm)*g, (ym+lm)*g, (zm+lm)*g
+        if pow4 * Q < abs(Am):
+            break
+        Am = Am1
+        pow4 *= g
+    t = pow4/Am
+    X = (A0-x)*t
+    Y = (A0-y)*t
+    Z = -X-Y
+    E2 = X*Y-Z**2
+    E3 = X*Y*Z
+    return ctx.power(Am,-0.5) * (9240-924*E2+385*E2**2+660*E3-630*E2*E3)/9240
+
+def RC_calc(ctx, x, y, r, pv=True):
+    if not (ctx.isnormal(x) and ctx.isnormal(y)):
+        if ctx.isinf(x) or ctx.isinf(y):
+            return 1/(x*y)
+        if y == 0:
+            return ctx.inf
+        if x == 0:
+            return ctx.pi / ctx.sqrt(y) / 2
+        raise ValueError
+    # Cauchy principal value
+    if pv and ctx._im(y) == 0 and ctx._re(y) < 0:
+        return ctx.sqrt(x/(x-y)) * RC_calc(ctx, x-y, -y, r)
+    if x == y:
+        return 1/ctx.sqrt(x)
+    extraprec = 2*max(0,-ctx.mag(x-y)+ctx.mag(x))
+    ctx.prec += extraprec
+    if ctx._is_real_type(x) and ctx._is_real_type(y):
+        x = ctx._re(x)
+        y = ctx._re(y)
+        a = ctx.sqrt(x/y)
+        if x < y:
+            b = ctx.sqrt(y-x)
+            v = ctx.acos(a)/b
+        else:
+            b = ctx.sqrt(x-y)
+            v = ctx.acosh(a)/b
+    else:
+        sx = ctx.sqrt(x)
+        sy = ctx.sqrt(y)
+        v = ctx.acos(sx/sy)/(ctx.sqrt((1-x/y))*sy)
+    ctx.prec -= extraprec
+    return v
+
+def RJ_calc(ctx, x, y, z, p, r, integration):
+    """
+    With integration == 0, computes RJ only using Carlson's algorithm
+    (may be wrong for some values).
+    With integration == 1, uses an initial integration to make sure
+    Carlson's algorithm is correct.
+    With integration == 2, uses only integration.
+    """
+    if not (ctx.isnormal(x) and ctx.isnormal(y) and \
+        ctx.isnormal(z) and ctx.isnormal(p)):
+        if ctx.isnan(x) or ctx.isnan(y) or ctx.isnan(z) or ctx.isnan(p):
+            return x*y*z
+        if ctx.isinf(x) or ctx.isinf(y) or ctx.isinf(z) or ctx.isinf(p):
+            return ctx.zero
+    if not p:
+        return ctx.inf
+    if (not x) + (not y) + (not z) > 1:
+        return ctx.inf
+    # Check conditions and fall back on integration for argument
+    # reduction if needed. The following conditions might be needlessly
+    # restrictive.
+    initial_integral = ctx.zero
+    if integration >= 1:
+        ok = (x.real >= 0 and y.real >= 0 and z.real >= 0 and p.real > 0)
+        if not ok:
+            if x == p or y == p or z == p:
+                ok = True
+        if not ok:
+            if p.imag != 0 or p.real >= 0:
+                if (x.imag == 0 and x.real >= 0 and ctx.conj(y) == z):
+                    ok = True
+                if (y.imag == 0 and y.real >= 0 and ctx.conj(x) == z):
+                    ok = True
+                if (z.imag == 0 and z.real >= 0 and ctx.conj(x) == y):
+                    ok = True
+        if not ok or (integration == 2):
+            N = ctx.ceil(-min(x.real, y.real, z.real, p.real)) + 1
+            # Integrate around any singularities
+            if all((t.imag >= 0 or t.real > 0) for t in [x, y, z, p]):
+                margin = ctx.j
+            elif all((t.imag < 0 or t.real > 0) for t in [x, y, z, p]):
+                margin = -ctx.j
+            else:
+                margin = 1
+                # Go through the upper half-plane, but low enough that any
+                # parameter starting in the lower plane doesn't cross the
+                # branch cut
+                for t in [x, y, z, p]:
+                    if t.imag >= 0 or t.real > 0:
+                        continue
+                    margin = min(margin, abs(t.imag) * 0.5)
+                margin *= ctx.j
+            N += margin
+            F = lambda t: 1/(ctx.sqrt(t+x)*ctx.sqrt(t+y)*ctx.sqrt(t+z)*(t+p))
+            if integration == 2:
+                return 1.5 * ctx.quadsubdiv(F, [0, N, ctx.inf])
+            initial_integral = 1.5 * ctx.quadsubdiv(F, [0, N])
+            x += N; y += N; z += N; p += N
+    xm,ym,zm,pm = x,y,z,p
+    A0 = Am = (x + y + z + 2*p)/5
+    delta = (p-x)*(p-y)*(p-z)
+    Q = ctx.root(0.25*r, -6) * max(abs(A0-x),abs(A0-y),abs(A0-z),abs(A0-p))
+    g = ctx.mpf(0.25)
+    pow4 = ctx.one
+    S = 0
+    while 1:
+        sx = ctx.sqrt(xm)
+        sy = ctx.sqrt(ym)
+        sz = ctx.sqrt(zm)
+        sp = ctx.sqrt(pm)
+        lm = sx*sy + sx*sz + sy*sz
+        Am1 = (Am+lm)*g
+        xm = (xm+lm)*g; ym = (ym+lm)*g; zm = (zm+lm)*g; pm = (pm+lm)*g
+        dm = (sp+sx) * (sp+sy) * (sp+sz)
+        em = delta * pow4**3 / dm**2
+        if pow4 * Q < abs(Am):
+            break
+        T = RC_calc(ctx, ctx.one, ctx.one+em, r) * pow4 / dm
+        S += T
+        pow4 *= g
+        Am = Am1
+    t = pow4 / Am
+    X = (A0-x)*t
+    Y = (A0-y)*t
+    Z = (A0-z)*t
+    P = (-X-Y-Z)/2
+    E2 = X*Y + X*Z + Y*Z - 3*P**2
+    E3 = X*Y*Z + 2*E2*P + 4*P**3
+    E4 = (2*X*Y*Z + E2*P + 3*P**3)*P
+    E5 = X*Y*Z*P**2
+    P = 24024 - 5148*E2 + 2457*E2**2 + 4004*E3 - 4158*E2*E3 - 3276*E4 + 2772*E5
+    Q = 24024
+    v1 = pow4 * ctx.power(Am, -1.5) * P/Q
+    v2 = 6*S
+    return initial_integral + v1 + v2
+
+@defun
+def elliprf(ctx, x, y, z):
+    r"""
+    Evaluates the Carlson symmetric elliptic integral of the first kind
+
+    .. math ::
+
+        R_F(x,y,z) = \frac{1}{2}
+            \int_0^{\infty} \frac{dt}{\sqrt{(t+x)(t+y)(t+z)}}
+
+    which is defined for `x,y,z \notin (-\infty,0)`, and with
+    at most one of `x,y,z` being zero.
+
+    For real `x,y,z \ge 0`, the principal square root is taken in the integrand.
+    For complex `x,y,z`, the principal square root is taken as `t \to \infty`
+    and as `t \to 0` non-principal branches are chosen as necessary so as to
+    make the integrand continuous.
+
+    **Examples**
+
+    Some basic values and limits::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> elliprf(0,1,1); pi/2
+        1.570796326794896619231322
+        1.570796326794896619231322
+        >>> elliprf(0,1,inf)
+        0.0
+        >>> elliprf(1,1,1)
+        1.0
+        >>> elliprf(2,2,2)**2
+        0.5
+        >>> elliprf(1,0,0); elliprf(0,0,1); elliprf(0,1,0); elliprf(0,0,0)
+        +inf
+        +inf
+        +inf
+        +inf
+
+    Representing complete elliptic integrals in terms of `R_F`::
+
+        >>> m = mpf(0.75)
+        >>> ellipk(m); elliprf(0,1-m,1)
+        2.156515647499643235438675
+        2.156515647499643235438675
+        >>> ellipe(m); elliprf(0,1-m,1)-m*elliprd(0,1-m,1)/3
+        1.211056027568459524803563
+        1.211056027568459524803563
+
+    Some symmetries and argument transformations::
+
+        >>> x,y,z = 2,3,4
+        >>> elliprf(x,y,z); elliprf(y,x,z); elliprf(z,y,x)
+        0.5840828416771517066928492
+        0.5840828416771517066928492
+        0.5840828416771517066928492
+        >>> k = mpf(100000)
+        >>> elliprf(k*x,k*y,k*z); k**(-0.5) * elliprf(x,y,z)
+        0.001847032121923321253219284
+        0.001847032121923321253219284
+        >>> l = sqrt(x*y) + sqrt(y*z) + sqrt(z*x)
+        >>> elliprf(x,y,z); 2*elliprf(x+l,y+l,z+l)
+        0.5840828416771517066928492
+        0.5840828416771517066928492
+        >>> elliprf((x+l)/4,(y+l)/4,(z+l)/4)
+        0.5840828416771517066928492
+
+    Comparing with numerical integration::
+
+        >>> x,y,z = 2,3,4
+        >>> elliprf(x,y,z)
+        0.5840828416771517066928492
+        >>> f = lambda t: 0.5*((t+x)*(t+y)*(t+z))**(-0.5)
+        >>> q = extradps(25)(quad)
+        >>> q(f, [0,inf])
+        0.5840828416771517066928492
+
+    With the following arguments, the square root in the integrand becomes
+    discontinuous at `t = 1/2` if the principal branch is used. To obtain
+    the right value, `-\sqrt{r}` must be taken instead of `\sqrt{r}`
+    on `t \in (0, 1/2)`::
+
+        >>> x,y,z = j-1,j,0
+        >>> elliprf(x,y,z)
+        (0.7961258658423391329305694 - 1.213856669836495986430094j)
+        >>> -q(f, [0,0.5]) + q(f, [0.5,inf])
+        (0.7961258658423391329305694 - 1.213856669836495986430094j)
+
+    The so-called *first lemniscate constant*, a transcendental number::
+
+        >>> elliprf(0,1,2)
+        1.31102877714605990523242
+        >>> extradps(25)(quad)(lambda t: 1/sqrt(1-t**4), [0,1])
+        1.31102877714605990523242
+        >>> gamma('1/4')**2/(4*sqrt(2*pi))
+        1.31102877714605990523242
+
+    **References**
+
+    1. [Carlson]_
+    2. [DLMF]_ Chapter 19. Elliptic Integrals
+
+    """
+    x = ctx.convert(x)
+    y = ctx.convert(y)
+    z = ctx.convert(z)
+    prec = ctx.prec
+    try:
+        ctx.prec += 20
+        tol = ctx.eps * 2**10
+        v = RF_calc(ctx, x, y, z, tol)
+    finally:
+        ctx.prec = prec
+    return +v
+
+@defun
+def elliprc(ctx, x, y, pv=True):
+    r"""
+    Evaluates the degenerate Carlson symmetric elliptic integral
+    of the first kind
+
+    .. math ::
+
+        R_C(x,y) = R_F(x,y,y) =
+            \frac{1}{2} \int_0^{\infty} \frac{dt}{(t+y) \sqrt{(t+x)}}.
+
+    If `y \in (-\infty,0)`, either a value defined by continuity,
+    or with *pv=True* the Cauchy principal value, can be computed.
+
+    If `x \ge 0, y > 0`, the value can be expressed in terms of
+    elementary functions as
+
+    .. math ::
+
+        R_C(x,y) =
+        \begin{cases}
+          \dfrac{1}{\sqrt{y-x}}
+            \cos^{-1}\left(\sqrt{\dfrac{x}{y}}\right),   & x < y \\
+          \dfrac{1}{\sqrt{y}},                          & x = y \\
+          \dfrac{1}{\sqrt{x-y}}
+            \cosh^{-1}\left(\sqrt{\dfrac{x}{y}}\right),  & x > y \\
+        \end{cases}.
+
+    **Examples**
+
+    Some special values and limits::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> elliprc(1,2)*4; elliprc(0,1)*2; +pi
+        3.141592653589793238462643
+        3.141592653589793238462643
+        3.141592653589793238462643
+        >>> elliprc(1,0)
+        +inf
+        >>> elliprc(5,5)**2
+        0.2
+        >>> elliprc(1,inf); elliprc(inf,1); elliprc(inf,inf)
+        0.0
+        0.0
+        0.0
+
+    Comparing with the elementary closed-form solution::
+
+        >>> elliprc('1/3', '1/5'); sqrt(7.5)*acosh(sqrt('5/3'))
+        2.041630778983498390751238
+        2.041630778983498390751238
+        >>> elliprc('1/5', '1/3'); sqrt(7.5)*acos(sqrt('3/5'))
+        1.875180765206547065111085
+        1.875180765206547065111085
+
+    Comparing with numerical integration::
+
+        >>> q = extradps(25)(quad)
+        >>> elliprc(2, -3, pv=True)
+        0.3333969101113672670749334
+        >>> elliprc(2, -3, pv=False)
+        (0.3333969101113672670749334 + 0.7024814731040726393156375j)
+        >>> 0.5*q(lambda t: 1/(sqrt(t+2)*(t-3)), [0,3-j,6,inf])
+        (0.3333969101113672670749334 + 0.7024814731040726393156375j)
+
+    """
+    x = ctx.convert(x)
+    y = ctx.convert(y)
+    prec = ctx.prec
+    try:
+        ctx.prec += 20
+        tol = ctx.eps * 2**10
+        v = RC_calc(ctx, x, y, tol, pv)
+    finally:
+        ctx.prec = prec
+    return +v
+
+@defun
+def elliprj(ctx, x, y, z, p, integration=1):
+    r"""
+    Evaluates the Carlson symmetric elliptic integral of the third kind
+
+    .. math ::
+
+        R_J(x,y,z,p) = \frac{3}{2}
+            \int_0^{\infty} \frac{dt}{(t+p)\sqrt{(t+x)(t+y)(t+z)}}.
+
+    Like :func:`~mpmath.elliprf`, the branch of the square root in the integrand
+    is defined so as to be continuous along the path of integration for
+    complex values of the arguments.
+
+    **Examples**
+
+    Some values and limits::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> elliprj(1,1,1,1)
+        1.0
+        >>> elliprj(2,2,2,2); 1/(2*sqrt(2))
+        0.3535533905932737622004222
+        0.3535533905932737622004222
+        >>> elliprj(0,1,2,2)
+        1.067937989667395702268688
+        >>> 3*(2*gamma('5/4')**2-pi**2/gamma('1/4')**2)/(sqrt(2*pi))
+        1.067937989667395702268688
+        >>> elliprj(0,1,1,2); 3*pi*(2-sqrt(2))/4
+        1.380226776765915172432054
+        1.380226776765915172432054
+        >>> elliprj(1,3,2,0); elliprj(0,1,1,0); elliprj(0,0,0,0)
+        +inf
+        +inf
+        +inf
+        >>> elliprj(1,inf,1,0); elliprj(1,1,1,inf)
+        0.0
+        0.0
+        >>> chop(elliprj(1+j, 1-j, 1, 1))
+        0.8505007163686739432927844
+
+    Scale transformation::
+
+        >>> x,y,z,p = 2,3,4,5
+        >>> k = mpf(100000)
+        >>> elliprj(k*x,k*y,k*z,k*p); k**(-1.5)*elliprj(x,y,z,p)
+        4.521291677592745527851168e-9
+        4.521291677592745527851168e-9
+
+    Comparing with numerical integration::
+
+        >>> elliprj(1,2,3,4)
+        0.2398480997495677621758617
+        >>> f = lambda t: 1/((t+4)*sqrt((t+1)*(t+2)*(t+3)))
+        >>> 1.5*quad(f, [0,inf])
+        0.2398480997495677621758617
+        >>> elliprj(1,2+1j,3,4-2j)
+        (0.216888906014633498739952 + 0.04081912627366673332369512j)
+        >>> f = lambda t: 1/((t+4-2j)*sqrt((t+1)*(t+2+1j)*(t+3)))
+        >>> 1.5*quad(f, [0,inf])
+        (0.216888906014633498739952 + 0.04081912627366673332369511j)
+
+    """
+    x = ctx.convert(x)
+    y = ctx.convert(y)
+    z = ctx.convert(z)
+    p = ctx.convert(p)
+    prec = ctx.prec
+    try:
+        ctx.prec += 20
+        tol = ctx.eps * 2**10
+        v = RJ_calc(ctx, x, y, z, p, tol, integration)
+    finally:
+        ctx.prec = prec
+    return +v
+
+@defun
+def elliprd(ctx, x, y, z):
+    r"""
+    Evaluates the degenerate Carlson symmetric elliptic integral
+    of the third kind or Carlson elliptic integral of the
+    second kind `R_D(x,y,z) = R_J(x,y,z,z)`.
+
+    See :func:`~mpmath.elliprj` for additional information.
+
+    **Examples**
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> elliprd(1,2,3)
+        0.2904602810289906442326534
+        >>> elliprj(1,2,3,3)
+        0.2904602810289906442326534
+
+    The so-called *second lemniscate constant*, a transcendental number::
+
+        >>> elliprd(0,2,1)/3
+        0.5990701173677961037199612
+        >>> extradps(25)(quad)(lambda t: t**2/sqrt(1-t**4), [0,1])
+        0.5990701173677961037199612
+        >>> gamma('3/4')**2/sqrt(2*pi)
+        0.5990701173677961037199612
+
+    """
+    return ctx.elliprj(x,y,z,z)
+
+@defun
+def elliprg(ctx, x, y, z):
+    r"""
+    Evaluates the Carlson completely symmetric elliptic integral
+    of the second kind
+
+    .. math ::
+
+        R_G(x,y,z) = \frac{1}{4} \int_0^{\infty}
+            \frac{t}{\sqrt{(t+x)(t+y)(t+z)}}
+            \left( \frac{x}{t+x} + \frac{y}{t+y} + \frac{z}{t+z}\right) dt.
+
+    **Examples**
+
+    Evaluation for real and complex arguments::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> elliprg(0,1,1)*4; +pi
+        3.141592653589793238462643
+        3.141592653589793238462643
+        >>> elliprg(0,0.5,1)
+        0.6753219405238377512600874
+        >>> chop(elliprg(1+j, 1-j, 2))
+        1.172431327676416604532822
+
+    A double integral that can be evaluated in terms of `R_G`::
+
+        >>> x,y,z = 2,3,4
+        >>> def f(t,u):
+        ...     st = fp.sin(t); ct = fp.cos(t)
+        ...     su = fp.sin(u); cu = fp.cos(u)
+        ...     return (x*(st*cu)**2 + y*(st*su)**2 + z*ct**2)**0.5 * st
+        ...
+        >>> nprint(mpf(fp.quad(f, [0,fp.pi], [0,2*fp.pi])/(4*fp.pi)), 13)
+        1.725503028069
+        >>> nprint(elliprg(x,y,z), 13)
+        1.725503028069
+
+    """
+    x = ctx.convert(x)
+    y = ctx.convert(y)
+    z = ctx.convert(z)
+    zeros = (not x) + (not y) + (not z)
+    if zeros == 3:
+        return (x+y+z)*0
+    if zeros == 2:
+        if x: return 0.5*ctx.sqrt(x)
+        if y: return 0.5*ctx.sqrt(y)
+        return 0.5*ctx.sqrt(z)
+    if zeros == 1:
+        if not z:
+            x, z = z, x
+    def terms():
+        T1 = 0.5*z*ctx.elliprf(x,y,z)
+        T2 = -0.5*(x-z)*(y-z)*ctx.elliprd(x,y,z)/3
+        T3 = 0.5*ctx.sqrt(x)*ctx.sqrt(y)/ctx.sqrt(z)
+        return T1,T2,T3
+    return ctx.sum_accurately(terms)
+
+
+@defun_wrapped
+def ellipf(ctx, phi, m):
+    r"""
+    Evaluates the Legendre incomplete elliptic integral of the first kind
+
+     .. math ::
+
+        F(\phi,m) = \int_0^{\phi} \frac{dt}{\sqrt{1-m \sin^2 t}}
+
+    or equivalently
+
+    .. math ::
+
+        F(\phi,m) = \int_0^{\sin \phi}
+        \frac{dt}{\left(\sqrt{1-t^2}\right)\left(\sqrt{1-mt^2}\right)}.
+
+    The function reduces to a complete elliptic integral of the first kind
+    (see :func:`~mpmath.ellipk`) when `\phi = \frac{\pi}{2}`; that is,
+
+    .. math ::
+
+        F\left(\frac{\pi}{2}, m\right) = K(m).
+
+    In the defining integral, it is assumed that the principal branch
+    of the square root is taken and that the path of integration avoids
+    crossing any branch cuts. Outside `-\pi/2 \le \Re(\phi) \le \pi/2`,
+    the function extends quasi-periodically as
+
+    .. math ::
+
+        F(\phi + n \pi, m) = 2 n K(m) + F(\phi,m), n \in \mathbb{Z}.
+
+    **Plots**
+
+    .. literalinclude :: /plots/ellipf.py
+    .. image :: /plots/ellipf.png
+
+    **Examples**
+
+    Basic values and limits::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> ellipf(0,1)
+        0.0
+        >>> ellipf(0,0)
+        0.0
+        >>> ellipf(1,0); ellipf(2+3j,0)
+        1.0
+        (2.0 + 3.0j)
+        >>> ellipf(1,1); log(sec(1)+tan(1))
+        1.226191170883517070813061
+        1.226191170883517070813061
+        >>> ellipf(pi/2, -0.5); ellipk(-0.5)
+        1.415737208425956198892166
+        1.415737208425956198892166
+        >>> ellipf(pi/2+eps, 1); ellipf(-pi/2-eps, 1)
+        +inf
+        +inf
+        >>> ellipf(1.5, 1)
+        3.340677542798311003320813
+
+    Comparing with numerical integration::
+
+        >>> z,m = 0.5, 1.25
+        >>> ellipf(z,m)
+        0.5287219202206327872978255
+        >>> quad(lambda t: (1-m*sin(t)**2)**(-0.5), [0,z])
+        0.5287219202206327872978255
+
+    The arguments may be complex numbers::
+
+        >>> ellipf(3j, 0.5)
+        (0.0 + 1.713602407841590234804143j)
+        >>> ellipf(3+4j, 5-6j)
+        (1.269131241950351323305741 - 0.3561052815014558335412538j)
+        >>> z,m = 2+3j, 1.25
+        >>> k = 1011
+        >>> ellipf(z+pi*k,m); ellipf(z,m) + 2*k*ellipk(m)
+        (4086.184383622179764082821 - 3003.003538923749396546871j)
+        (4086.184383622179764082821 - 3003.003538923749396546871j)
+
+    For `|\Re(z)| < \pi/2`, the function can be expressed as a
+    hypergeometric series of two variables
+    (see :func:`~mpmath.appellf1`)::
+
+        >>> z,m = 0.5, 0.25
+        >>> ellipf(z,m)
+        0.5050887275786480788831083
+        >>> sin(z)*appellf1(0.5,0.5,0.5,1.5,sin(z)**2,m*sin(z)**2)
+        0.5050887275786480788831083
+
+    """
+    z = phi
+    if not (ctx.isnormal(z) and ctx.isnormal(m)):
+        if m == 0:
+            return z + m
+        if z == 0:
+            return z * m
+        if m == ctx.inf or m == ctx.ninf: return z/m
+        raise ValueError
+    x = z.real
+    ctx.prec += max(0, ctx.mag(x))
+    pi = +ctx.pi
+    away = abs(x) > pi/2
+    if m == 1:
+        if away:
+            return ctx.inf
+    if away:
+        d = ctx.nint(x/pi)
+        z = z-pi*d
+        P = 2*d*ctx.ellipk(m)
+    else:
+        P = 0
+    c, s = ctx.cos_sin(z)
+    return s * ctx.elliprf(c**2, 1-m*s**2, 1) + P
+
+@defun_wrapped
+def ellipe(ctx, *args):
+    r"""
+    Called with a single argument `m`, evaluates the Legendre complete
+    elliptic integral of the second kind, `E(m)`, defined by
+
+        .. math :: E(m) = \int_0^{\pi/2} \sqrt{1-m \sin^2 t} \, dt \,=\,
+            \frac{\pi}{2}
+            \,_2F_1\left(\frac{1}{2}, -\frac{1}{2}, 1, m\right).
+
+    Called with two arguments `\phi, m`, evaluates the incomplete elliptic
+    integral of the second kind
+
+     .. math ::
+
+        E(\phi,m) = \int_0^{\phi} \sqrt{1-m \sin^2 t} \, dt =
+                    \int_0^{\sin z}
+                    \frac{\sqrt{1-mt^2}}{\sqrt{1-t^2}} \, dt.
+
+    The incomplete integral reduces to a complete integral when
+    `\phi = \frac{\pi}{2}`; that is,
+
+    .. math ::
+
+        E\left(\frac{\pi}{2}, m\right) = E(m).
+
+    In the defining integral, it is assumed that the principal branch
+    of the square root is taken and that the path of integration avoids
+    crossing any branch cuts. Outside `-\pi/2 \le \Re(z) \le \pi/2`,
+    the function extends quasi-periodically as
+
+    .. math ::
+
+        E(\phi + n \pi, m) = 2 n E(m) + E(\phi,m), n \in \mathbb{Z}.
+
+    **Plots**
+
+    .. literalinclude :: /plots/ellipe.py
+    .. image :: /plots/ellipe.png
+
+    **Examples for the complete integral**
+
+    Basic values and limits::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> ellipe(0)
+        1.570796326794896619231322
+        >>> ellipe(1)
+        1.0
+        >>> ellipe(-1)
+        1.910098894513856008952381
+        >>> ellipe(2)
+        (0.5990701173677961037199612 + 0.5990701173677961037199612j)
+        >>> ellipe(inf)
+        (0.0 + +infj)
+        >>> ellipe(-inf)
+        +inf
+
+    Verifying the defining integral and hypergeometric
+    representation::
+
+        >>> ellipe(0.5)
+        1.350643881047675502520175
+        >>> quad(lambda t: sqrt(1-0.5*sin(t)**2), [0, pi/2])
+        1.350643881047675502520175
+        >>> pi/2*hyp2f1(0.5,-0.5,1,0.5)
+        1.350643881047675502520175
+
+    Evaluation is supported for arbitrary complex `m`::
+
+        >>> ellipe(0.5+0.25j)
+        (1.360868682163129682716687 - 0.1238733442561786843557315j)
+        >>> ellipe(3+4j)
+        (1.499553520933346954333612 - 1.577879007912758274533309j)
+
+    A definite integral::
+
+        >>> quad(ellipe, [0,1])
+        1.333333333333333333333333
+
+    **Examples for the incomplete integral**
+
+    Basic values and limits::
+
+        >>> ellipe(0,1)
+        0.0
+        >>> ellipe(0,0)
+        0.0
+        >>> ellipe(1,0)
+        1.0
+        >>> ellipe(2+3j,0)
+        (2.0 + 3.0j)
+        >>> ellipe(1,1); sin(1)
+        0.8414709848078965066525023
+        0.8414709848078965066525023
+        >>> ellipe(pi/2, -0.5); ellipe(-0.5)
+        1.751771275694817862026502
+        1.751771275694817862026502
+        >>> ellipe(pi/2, 1); ellipe(-pi/2, 1)
+        1.0
+        -1.0
+        >>> ellipe(1.5, 1)
+        0.9974949866040544309417234
+
+    Comparing with numerical integration::
+
+        >>> z,m = 0.5, 1.25
+        >>> ellipe(z,m)
+        0.4740152182652628394264449
+        >>> quad(lambda t: sqrt(1-m*sin(t)**2), [0,z])
+        0.4740152182652628394264449
+
+    The arguments may be complex numbers::
+
+        >>> ellipe(3j, 0.5)
+        (0.0 + 7.551991234890371873502105j)
+        >>> ellipe(3+4j, 5-6j)
+        (24.15299022574220502424466 + 75.2503670480325997418156j)
+        >>> k = 35
+        >>> z,m = 2+3j, 1.25
+        >>> ellipe(z+pi*k,m); ellipe(z,m) + 2*k*ellipe(m)
+        (48.30138799412005235090766 + 17.47255216721987688224357j)
+        (48.30138799412005235090766 + 17.47255216721987688224357j)
+
+    For `|\Re(z)| < \pi/2`, the function can be expressed as a
+    hypergeometric series of two variables
+    (see :func:`~mpmath.appellf1`)::
+
+        >>> z,m = 0.5, 0.25
+        >>> ellipe(z,m)
+        0.4950017030164151928870375
+        >>> sin(z)*appellf1(0.5,0.5,-0.5,1.5,sin(z)**2,m*sin(z)**2)
+        0.4950017030164151928870376
+
+    """
+    if len(args) == 1:
+        return ctx._ellipe(args[0])
+    else:
+        phi, m = args
+    z = phi
+    if not (ctx.isnormal(z) and ctx.isnormal(m)):
+        if m == 0:
+            return z + m
+        if z == 0:
+            return z * m
+        if m == ctx.inf or m == ctx.ninf:
+            return ctx.inf
+        raise ValueError
+    x = z.real
+    ctx.prec += max(0, ctx.mag(x))
+    pi = +ctx.pi
+    away = abs(x) > pi/2
+    if away:
+        d = ctx.nint(x/pi)
+        z = z-pi*d
+        P = 2*d*ctx.ellipe(m)
+    else:
+        P = 0
+    def terms():
+        c, s = ctx.cos_sin(z)
+        x = c**2
+        y = 1-m*s**2
+        RF = ctx.elliprf(x, y, 1)
+        RD = ctx.elliprd(x, y, 1)
+        return s*RF, -m*s**3*RD/3
+    return ctx.sum_accurately(terms) + P
+
+@defun_wrapped
+def ellippi(ctx, *args):
+    r"""
+    Called with three arguments `n, \phi, m`, evaluates the Legendre
+    incomplete elliptic integral of the third kind
+
+    .. math ::
+
+        \Pi(n; \phi, m) = \int_0^{\phi}
+            \frac{dt}{(1-n \sin^2 t) \sqrt{1-m \sin^2 t}} =
+            \int_0^{\sin \phi}
+            \frac{dt}{(1-nt^2) \sqrt{1-t^2} \sqrt{1-mt^2}}.
+
+    Called with two arguments `n, m`, evaluates the complete
+    elliptic integral of the third kind
+    `\Pi(n,m) = \Pi(n; \frac{\pi}{2},m)`.
+
+    In the defining integral, it is assumed that the principal branch
+    of the square root is taken and that the path of integration avoids
+    crossing any branch cuts. Outside `-\pi/2 \le \Re(\phi) \le \pi/2`,
+    the function extends quasi-periodically as
+
+    .. math ::
+
+        \Pi(n,\phi+k\pi,m) = 2k\Pi(n,m) + \Pi(n,\phi,m), k \in \mathbb{Z}.
+
+    **Plots**
+
+    .. literalinclude :: /plots/ellippi.py
+    .. image :: /plots/ellippi.png
+
+    **Examples for the complete integral**
+
+    Some basic values and limits::
+
+        >>> from mpmath import *
+        >>> mp.dps = 25; mp.pretty = True
+        >>> ellippi(0,-5); ellipk(-5)
+        0.9555039270640439337379334
+        0.9555039270640439337379334
+        >>> ellippi(inf,2)
+        0.0
+        >>> ellippi(2,inf)
+        0.0
+        >>> abs(ellippi(1,5))
+        +inf
+        >>> abs(ellippi(0.25,1))
+        +inf
+
+    Evaluation in terms of simpler functions::
+
+        >>> ellippi(0.25,0.25); ellipe(0.25)/(1-0.25)
+        1.956616279119236207279727
+        1.956616279119236207279727
+        >>> ellippi(3,0); pi/(2*sqrt(-2))
+        (0.0 - 1.11072073453959156175397j)
+        (0.0 - 1.11072073453959156175397j)
+        >>> ellippi(-3,0); pi/(2*sqrt(4))
+        0.7853981633974483096156609
+        0.7853981633974483096156609
+
+    **Examples for the incomplete integral**
+
+    Basic values and limits::
+
+        >>> ellippi(0.25,-0.5); ellippi(0.25,pi/2,-0.5)
+        1.622944760954741603710555
+        1.622944760954741603710555
+        >>> ellippi(1,0,1)
+        0.0
+        >>> ellippi(inf,0,1)
+        0.0
+        >>> ellippi(0,0.25,0.5); ellipf(0.25,0.5)
+        0.2513040086544925794134591
+        0.2513040086544925794134591
+        >>> ellippi(1,1,1); (log(sec(1)+tan(1))+sec(1)*tan(1))/2
+        2.054332933256248668692452
+        2.054332933256248668692452
+        >>> ellippi(0.25, 53*pi/2, 0.75); 53*ellippi(0.25,0.75)
+        135.240868757890840755058
+        135.240868757890840755058
+        >>> ellippi(0.5,pi/4,0.5); 2*ellipe(pi/4,0.5)-1/sqrt(3)
+        0.9190227391656969903987269
+        0.9190227391656969903987269
+
+    Complex arguments are supported::
+
+        >>> ellippi(0.5, 5+6j-2*pi, -7-8j)
+        (-0.3612856620076747660410167 + 0.5217735339984807829755815j)
+
+    Some degenerate cases::
+
+        >>> ellippi(1,1)
+        +inf
+        >>> ellippi(1,0)
+        +inf
+        >>> ellippi(1,2,0)
+        +inf
+        >>> ellippi(1,2,1)
+        +inf
+        >>> ellippi(1,0,1)
+        0.0
+
+    """
+    if len(args) == 2:
+        n, m = args
+        complete = True
+        z = phi = ctx.pi/2
+    else:
+        n, phi, m = args
+        complete = False
+        z = phi
+    if not (ctx.isnormal(n) and ctx.isnormal(z) and ctx.isnormal(m)):
+        if ctx.isnan(n) or ctx.isnan(z) or ctx.isnan(m):
+            raise ValueError
+        if complete:
+            if m == 0:
+                if n == 1:
+                    return ctx.inf
+                return ctx.pi/(2*ctx.sqrt(1-n))
+            if n == 0: return ctx.ellipk(m)
+            if ctx.isinf(n) or ctx.isinf(m): return ctx.zero
+        else:
+            if z == 0: return z
+            if ctx.isinf(n): return ctx.zero
+            if ctx.isinf(m): return ctx.zero
+        if ctx.isinf(n) or ctx.isinf(z) or ctx.isinf(m):
+            raise ValueError
+    if complete:
+        if m == 1:
+            if n == 1:
+                return ctx.inf
+            return -ctx.inf/ctx.sign(n-1)
+        away = False
+    else:
+        x = z.real
+        ctx.prec += max(0, ctx.mag(x))
+        pi = +ctx.pi
+        away = abs(x) > pi/2
+    if away:
+        d = ctx.nint(x/pi)
+        z = z-pi*d
+        P = 2*d*ctx.ellippi(n,m)
+        if ctx.isinf(P):
+            return ctx.inf
+    else:
+        P = 0
+    def terms():
+        if complete:
+            c, s = ctx.zero, ctx.one
+        else:
+            c, s = ctx.cos_sin(z)
+        x = c**2
+        y = 1-m*s**2
+        RF = ctx.elliprf(x, y, 1)
+        RJ = ctx.elliprj(x, y, 1, 1-n*s**2)
+        return s*RF, n*s**3*RJ/3
+    return ctx.sum_accurately(terms) + P

lib/python3.11/site-packages/mpmath/functions/expintegrals.py

@@ -0,0 +1,425 @@
+from .functions import defun, defun_wrapped
+
+@defun_wrapped
+def _erf_complex(ctx, z):
+    z2 = ctx.square_exp_arg(z, -1)
+    #z2 = -z**2
+    v = (2/ctx.sqrt(ctx.pi))*z * ctx.hyp1f1((1,2),(3,2), z2)
+    if not ctx._re(z):
+        v = ctx._im(v)*ctx.j
+    return v
+
+@defun_wrapped
+def _erfc_complex(ctx, z):
+    if ctx.re(z) > 2:
+        z2 = ctx.square_exp_arg(z)
+        nz2 = ctx.fneg(z2, exact=True)
+        v = ctx.exp(nz2)/ctx.sqrt(ctx.pi) * ctx.hyperu((1,2),(1,2), z2)
+    else:
+        v = 1 - ctx._erf_complex(z)
+    if not ctx._re(z):
+        v = 1+ctx._im(v)*ctx.j
+    return v
+
+@defun
+def erf(ctx, z):
+    z = ctx.convert(z)
+    if ctx._is_real_type(z):
+        try:
+            return ctx._erf(z)
+        except NotImplementedError:
+            pass
+    if ctx._is_complex_type(z) and not z.imag:
+        try:
+            return type(z)(ctx._erf(z.real))
+        except NotImplementedError:
+            pass
+    return ctx._erf_complex(z)
+
+@defun
+def erfc(ctx, z):
+    z = ctx.convert(z)
+    if ctx._is_real_type(z):
+        try:
+            return ctx._erfc(z)
+        except NotImplementedError:
+            pass
+    if ctx._is_complex_type(z) and not z.imag:
+        try:
+            return type(z)(ctx._erfc(z.real))
+        except NotImplementedError:
+            pass
+    return ctx._erfc_complex(z)
+
+@defun
+def square_exp_arg(ctx, z, mult=1, reciprocal=False):
+    prec = ctx.prec*4+20
+    if reciprocal:
+        z2 = ctx.fmul(z, z, prec=prec)
+        z2 = ctx.fdiv(ctx.one, z2, prec=prec)
+    else:
+        z2 = ctx.fmul(z, z, prec=prec)
+    if mult != 1:
+        z2 = ctx.fmul(z2, mult, exact=True)
+    return z2
+
+@defun_wrapped
+def erfi(ctx, z):
+    if not z:
+        return z
+    z2 = ctx.square_exp_arg(z)
+    v = (2/ctx.sqrt(ctx.pi)*z) * ctx.hyp1f1((1,2), (3,2), z2)
+    if not ctx._re(z):
+        v = ctx._im(v)*ctx.j
+    return v
+
+@defun_wrapped
+def erfinv(ctx, x):
+    xre = ctx._re(x)
+    if (xre != x) or (xre < -1) or (xre > 1):
+        return ctx.bad_domain("erfinv(x) is defined only for -1 <= x <= 1")
+    x = xre
+    #if ctx.isnan(x): return x
+    if not x: return x
+    if x == 1: return ctx.inf
+    if x == -1: return ctx.ninf
+    if abs(x) < 0.9:
+        a = 0.53728*x**3 + 0.813198*x
+    else:
+        # An asymptotic formula
+        u = ctx.ln(2/ctx.pi/(abs(x)-1)**2)
+        a = ctx.sign(x) * ctx.sqrt(u - ctx.ln(u))/ctx.sqrt(2)
+    ctx.prec += 10
+    return ctx.findroot(lambda t: ctx.erf(t)-x, a)
+
+@defun_wrapped
+def npdf(ctx, x, mu=0, sigma=1):
+    sigma = ctx.convert(sigma)
+    return ctx.exp(-(x-mu)**2/(2*sigma**2)) / (sigma*ctx.sqrt(2*ctx.pi))
+
+@defun_wrapped
+def ncdf(ctx, x, mu=0, sigma=1):
+    a = (x-mu)/(sigma*ctx.sqrt(2))
+    if a < 0:
+        return ctx.erfc(-a)/2
+    else:
+        return (1+ctx.erf(a))/2
+
+@defun_wrapped
+def betainc(ctx, a, b, x1=0, x2=1, regularized=False):
+    if x1 == x2:
+        v = 0
+    elif not x1:
+        if x1 == 0 and x2 == 1:
+            v = ctx.beta(a, b)
+        else:
+            v = x2**a * ctx.hyp2f1(a, 1-b, a+1, x2) / a
+    else:
+        m, d = ctx.nint_distance(a)
+        if m <= 0:
+            if d < -ctx.prec:
+                h = +ctx.eps
+                ctx.prec *= 2
+                a += h
+            elif d < -4:
+                ctx.prec -= d
+        s1 = x2**a * ctx.hyp2f1(a,1-b,a+1,x2)
+        s2 = x1**a * ctx.hyp2f1(a,1-b,a+1,x1)
+        v = (s1 - s2) / a
+    if regularized:
+        v /= ctx.beta(a,b)
+    return v
+
+@defun
+def gammainc(ctx, z, a=0, b=None, regularized=False):
+    regularized = bool(regularized)
+    z = ctx.convert(z)
+    if a is None:
+        a = ctx.zero
+        lower_modified = False
+    else:
+        a = ctx.convert(a)
+        lower_modified = a != ctx.zero
+    if b is None:
+        b = ctx.inf
+        upper_modified = False
+    else:
+        b = ctx.convert(b)
+        upper_modified = b != ctx.inf
+    # Complete gamma function
+    if not (upper_modified or lower_modified):
+        if regularized:
+            if ctx.re(z) < 0:
+                return ctx.inf
+            elif ctx.re(z) > 0:
+                return ctx.one
+            else:
+                return ctx.nan
+        return ctx.gamma(z)
+    if a == b:
+        return ctx.zero
+    # Standardize
+    if ctx.re(a) > ctx.re(b):
+        return -ctx.gammainc(z, b, a, regularized)
+    # Generalized gamma
+    if upper_modified and lower_modified:
+        return +ctx._gamma3(z, a, b, regularized)
+    # Upper gamma
+    elif lower_modified:
+        return ctx._upper_gamma(z, a, regularized)
+    # Lower gamma
+    elif upper_modified:
+        return ctx._lower_gamma(z, b, regularized)
+
+@defun
+def _lower_gamma(ctx, z, b, regularized=False):
+    # Pole
+    if ctx.isnpint(z):
+        return type(z)(ctx.inf)
+    G = [z] * regularized
+    negb = ctx.fneg(b, exact=True)
+    def h(z):
+        T1 = [ctx.exp(negb), b, z], [1, z, -1], [], G, [1], [1+z], b
+        return (T1,)
+    return ctx.hypercomb(h, [z])
+
+@defun
+def _upper_gamma(ctx, z, a, regularized=False):
+    # Fast integer case, when available
+    if ctx.isint(z):
+        try:
+            if regularized:
+                # Gamma pole
+                if ctx.isnpint(z):
+                    return type(z)(ctx.zero)
+                orig = ctx.prec
+                try:
+                    ctx.prec += 10
+                    return ctx._gamma_upper_int(z, a) / ctx.gamma(z)
+                finally:
+                    ctx.prec = orig
+            else:
+                return ctx._gamma_upper_int(z, a)
+        except NotImplementedError:
+            pass
+    # hypercomb is unable to detect the exact zeros, so handle them here
+    if z == 2 and a == -1:
+        return (z+a)*0
+    if z == 3 and (a == -1-1j or a == -1+1j):
+        return (z+a)*0
+    nega = ctx.fneg(a, exact=True)
+    G = [z] * regularized
+    # Use 2F0 series when possible; fall back to lower gamma representation
+    try:
+        def h(z):
+            r = z-1
+            return [([ctx.exp(nega), a], [1, r], [], G, [1, -r], [], 1/nega)]
+        return ctx.hypercomb(h, [z], force_series=True)
+    except ctx.NoConvergence:
+        def h(z):
+            T1 = [], [1, z-1], [z], G, [], [], 0
+            T2 = [-ctx.exp(nega), a, z], [1, z, -1], [], G, [1], [1+z], a
+            return T1, T2
+        return ctx.hypercomb(h, [z])
+
+@defun
+def _gamma3(ctx, z, a, b, regularized=False):
+    pole = ctx.isnpint(z)
+    if regularized and pole:
+        return ctx.zero
+    try:
+        ctx.prec += 15
+        # We don't know in advance whether it's better to write as a difference
+        # of lower or upper gamma functions, so try both
+        T1 = ctx.gammainc(z, a, regularized=regularized)
+        T2 = ctx.gammainc(z, b, regularized=regularized)
+        R = T1 - T2
+        if ctx.mag(R) - max(ctx.mag(T1), ctx.mag(T2)) > -10:
+            return R
+        if not pole:
+            T1 = ctx.gammainc(z, 0, b, regularized=regularized)
+            T2 = ctx.gammainc(z, 0, a, regularized=regularized)
+            R = T1 - T2
+            # May be ok, but should probably at least print a warning
+            # about possible cancellation
+            if 1: #ctx.mag(R) - max(ctx.mag(T1), ctx.mag(T2)) > -10:
+                return R
+    finally:
+        ctx.prec -= 15
+    raise NotImplementedError
+
+@defun_wrapped
+def expint(ctx, n, z):
+    if ctx.isint(n) and ctx._is_real_type(z):
+        try:
+            return ctx._expint_int(n, z)
+        except NotImplementedError:
+            pass
+    if ctx.isnan(n) or ctx.isnan(z):
+        return z*n
+    if z == ctx.inf:
+        return 1/z
+    if z == 0:
+        # integral from 1 to infinity of t^n
+        if ctx.re(n) <= 1:
+            # TODO: reasonable sign of infinity
+            return type(z)(ctx.inf)
+        else:
+            return ctx.one/(n-1)
+    if n == 0:
+        return ctx.exp(-z)/z
+    if n == -1:
+        return ctx.exp(-z)*(z+1)/z**2
+    return z**(n-1) * ctx.gammainc(1-n, z)
+
+@defun_wrapped
+def li(ctx, z, offset=False):
+    if offset:
+        if z == 2:
+            return ctx.zero
+        return ctx.ei(ctx.ln(z)) - ctx.ei(ctx.ln2)
+    if not z:
+        return z
+    if z == 1:
+        return ctx.ninf
+    return ctx.ei(ctx.ln(z))
+
+@defun
+def ei(ctx, z):
+    try:
+        return ctx._ei(z)
+    except NotImplementedError:
+        return ctx._ei_generic(z)
+
+@defun_wrapped
+def _ei_generic(ctx, z):
+    # Note: the following is currently untested because mp and fp
+    # both use special-case ei code
+    if z == ctx.inf:
+        return z
+    if z == ctx.ninf:
+        return ctx.zero
+    if ctx.mag(z) > 1:
+        try:
+            r = ctx.one/z
+            v = ctx.exp(z)*ctx.hyper([1,1],[],r,
+                maxterms=ctx.prec, force_series=True)/z
+            im = ctx._im(z)
+            if im > 0:
+                v += ctx.pi*ctx.j
+            if im < 0:
+                v -= ctx.pi*ctx.j
+            return v
+        except ctx.NoConvergence:
+            pass
+    v = z*ctx.hyp2f2(1,1,2,2,z) + ctx.euler
+    if ctx._im(z):
+        v += 0.5*(ctx.log(z) - ctx.log(ctx.one/z))
+    else:
+        v += ctx.log(abs(z))
+    return v
+
+@defun
+def e1(ctx, z):
+    try:
+        return ctx._e1(z)
+    except NotImplementedError:
+        return ctx.expint(1, z)
+
+@defun
+def ci(ctx, z):
+    try:
+        return ctx._ci(z)
+    except NotImplementedError:
+        return ctx._ci_generic(z)
+
+@defun_wrapped
+def _ci_generic(ctx, z):
+    if ctx.isinf(z):
+        if z == ctx.inf: return ctx.zero
+        if z == ctx.ninf: return ctx.pi*1j
+    jz = ctx.fmul(ctx.j,z,exact=True)
+    njz = ctx.fneg(jz,exact=True)
+    v = 0.5*(ctx.ei(jz) + ctx.ei(njz))
+    zreal = ctx._re(z)
+    zimag = ctx._im(z)
+    if zreal == 0:
+        if zimag > 0: v += ctx.pi*0.5j
+        if zimag < 0: v -= ctx.pi*0.5j
+    if zreal < 0:
+        if zimag >= 0: v += ctx.pi*1j
+        if zimag <  0: v -= ctx.pi*1j
+    if ctx._is_real_type(z) and zreal > 0:
+        v = ctx._re(v)
+    return v
+
+@defun
+def si(ctx, z):
+    try:
+        return ctx._si(z)
+    except NotImplementedError:
+        return ctx._si_generic(z)
+
+@defun_wrapped
+def _si_generic(ctx, z):
+    if ctx.isinf(z):
+        if z == ctx.inf: return 0.5*ctx.pi
+        if z == ctx.ninf: return -0.5*ctx.pi
+    # Suffers from cancellation near 0
+    if ctx.mag(z) >= -1:
+        jz = ctx.fmul(ctx.j,z,exact=True)
+        njz = ctx.fneg(jz,exact=True)
+        v = (-0.5j)*(ctx.ei(jz) - ctx.ei(njz))
+        zreal = ctx._re(z)
+        if zreal > 0:
+            v -= 0.5*ctx.pi
+        if zreal < 0:
+            v += 0.5*ctx.pi
+        if ctx._is_real_type(z):
+            v = ctx._re(v)
+        return v
+    else:
+        return z*ctx.hyp1f2((1,2),(3,2),(3,2),-0.25*z*z)
+
+@defun_wrapped
+def chi(ctx, z):
+    nz = ctx.fneg(z, exact=True)
+    v = 0.5*(ctx.ei(z) + ctx.ei(nz))
+    zreal = ctx._re(z)
+    zimag = ctx._im(z)
+    if zimag > 0:
+        v += ctx.pi*0.5j
+    elif zimag < 0:
+        v -= ctx.pi*0.5j
+    elif zreal < 0:
+        v += ctx.pi*1j
+    return v
+
+@defun_wrapped
+def shi(ctx, z):
+    # Suffers from cancellation near 0
+    if ctx.mag(z) >= -1:
+        nz = ctx.fneg(z, exact=True)
+        v = 0.5*(ctx.ei(z) - ctx.ei(nz))
+        zimag = ctx._im(z)
+        if zimag > 0: v -= 0.5j*ctx.pi
+        if zimag < 0: v += 0.5j*ctx.pi
+        return v
+    else:
+        return z * ctx.hyp1f2((1,2),(3,2),(3,2),0.25*z*z)
+
+@defun_wrapped
+def fresnels(ctx, z):
+    if z == ctx.inf:
+        return ctx.mpf(0.5)
+    if z == ctx.ninf:
+        return ctx.mpf(-0.5)
+    return ctx.pi*z**3/6*ctx.hyp1f2((3,4),(3,2),(7,4),-ctx.pi**2*z**4/16)
+
+@defun_wrapped
+def fresnelc(ctx, z):
+    if z == ctx.inf:
+        return ctx.mpf(0.5)
+    if z == ctx.ninf:
+        return ctx.mpf(-0.5)
+    return z*ctx.hyp1f2((1,4),(1,2),(5,4),-ctx.pi**2*z**4/16)

lib/python3.11/site-packages/mpmath/functions/factorials.py

@@ -0,0 +1,187 @@
+from ..libmp.backend import xrange
+from .functions import defun, defun_wrapped
+
+@defun
+def gammaprod(ctx, a, b, _infsign=False):
+    a = [ctx.convert(x) for x in a]
+    b = [ctx.convert(x) for x in b]
+    poles_num = []
+    poles_den = []
+    regular_num = []
+    regular_den = []
+    for x in a: [regular_num, poles_num][ctx.isnpint(x)].append(x)
+    for x in b: [regular_den, poles_den][ctx.isnpint(x)].append(x)
+    # One more pole in numerator or denominator gives 0 or inf
+    if len(poles_num) < len(poles_den): return ctx.zero
+    if len(poles_num) > len(poles_den):
+        # Get correct sign of infinity for x+h, h -> 0 from above
+        # XXX: hack, this should be done properly
+        if _infsign:
+            a = [x and x*(1+ctx.eps) or x+ctx.eps for x in poles_num]
+            b = [x and x*(1+ctx.eps) or x+ctx.eps for x in poles_den]
+            return ctx.sign(ctx.gammaprod(a+regular_num,b+regular_den)) * ctx.inf
+        else:
+            return ctx.inf
+    # All poles cancel
+    # lim G(i)/G(j) = (-1)**(i+j) * gamma(1-j) / gamma(1-i)
+    p = ctx.one
+    orig = ctx.prec
+    try:
+        ctx.prec = orig + 15
+        while poles_num:
+            i = poles_num.pop()
+            j = poles_den.pop()
+            p *= (-1)**(i+j) * ctx.gamma(1-j) / ctx.gamma(1-i)
+        for x in regular_num: p *= ctx.gamma(x)
+        for x in regular_den: p /= ctx.gamma(x)
+    finally:
+        ctx.prec = orig
+    return +p
+
+@defun
+def beta(ctx, x, y):
+    x = ctx.convert(x)
+    y = ctx.convert(y)
+    if ctx.isinf(y):
+        x, y = y, x
+    if ctx.isinf(x):
+        if x == ctx.inf and not ctx._im(y):
+            if y == ctx.ninf:
+                return ctx.nan
+            if y > 0:
+                return ctx.zero
+            if ctx.isint(y):
+                return ctx.nan
+            if y < 0:
+                return ctx.sign(ctx.gamma(y)) * ctx.inf
+        return ctx.nan
+    xy = ctx.fadd(x, y, prec=2*ctx.prec)
+    return ctx.gammaprod([x, y], [xy])
+
+@defun
+def binomial(ctx, n, k):
+    n1 = ctx.fadd(n, 1, prec=2*ctx.prec)
+    k1 = ctx.fadd(k, 1, prec=2*ctx.prec)
+    nk1 = ctx.fsub(n1, k, prec=2*ctx.prec)
+    return ctx.gammaprod([n1], [k1, nk1])
+
+@defun
+def rf(ctx, x, n):
+    xn = ctx.fadd(x, n, prec=2*ctx.prec)
+    return ctx.gammaprod([xn], [x])
+
+@defun
+def ff(ctx, x, n):
+    x1 = ctx.fadd(x, 1, prec=2*ctx.prec)
+    xn1 = ctx.fadd(ctx.fsub(x, n, prec=2*ctx.prec), 1, prec=2*ctx.prec)
+    return ctx.gammaprod([x1], [xn1])
+
+@defun_wrapped
+def fac2(ctx, x):
+    if ctx.isinf(x):
+        if x == ctx.inf:
+            return x
+        return ctx.nan
+    return 2**(x/2)*(ctx.pi/2)**((ctx.cospi(x)-1)/4)*ctx.gamma(x/2+1)
+
+@defun_wrapped
+def barnesg(ctx, z):
+    if ctx.isinf(z):
+        if z == ctx.inf:
+            return z
+        return ctx.nan
+    if ctx.isnan(z):
+        return z
+    if (not ctx._im(z)) and ctx._re(z) <= 0 and ctx.isint(ctx._re(z)):
+        return z*0
+    # Account for size (would not be needed if computing log(G))
+    if abs(z) > 5:
+        ctx.dps += 2*ctx.log(abs(z),2)
+    # Reflection formula
+    if ctx.re(z) < -ctx.dps:
+        w = 1-z
+        pi2 = 2*ctx.pi
+        u = ctx.expjpi(2*w)
+        v = ctx.j*ctx.pi/12 - ctx.j*ctx.pi*w**2/2 + w*ctx.ln(1-u) - \
+            ctx.j*ctx.polylog(2, u)/pi2
+        v = ctx.barnesg(2-z)*ctx.exp(v)/pi2**w
+        if ctx._is_real_type(z):
+            v = ctx._re(v)
+        return v
+    # Estimate terms for asymptotic expansion
+    # TODO: fixme, obviously
+    N = ctx.dps // 2 + 5
+    G = 1
+    while abs(z) < N or ctx.re(z) < 1:
+        G /= ctx.gamma(z)
+        z += 1
+    z -= 1
+    s = ctx.mpf(1)/12
+    s -= ctx.log(ctx.glaisher)
+    s += z*ctx.log(2*ctx.pi)/2
+    s += (z**2/2-ctx.mpf(1)/12)*ctx.log(z)
+    s -= 3*z**2/4
+    z2k = z2 = z**2
+    for k in xrange(1, N+1):
+        t = ctx.bernoulli(2*k+2) / (4*k*(k+1)*z2k)
+        if abs(t) < ctx.eps:
+            #print k, N      # check how many terms were needed
+            break
+        z2k *= z2
+        s += t
+    #if k == N:
+    #    print "warning: series for barnesg failed to converge", ctx.dps
+    return G*ctx.exp(s)
+
+@defun
+def superfac(ctx, z):
+    return ctx.barnesg(z+2)
+
+@defun_wrapped
+def hyperfac(ctx, z):
+    # XXX: estimate needed extra bits accurately
+    if z == ctx.inf:
+        return z
+    if abs(z) > 5:
+        extra = 4*int(ctx.log(abs(z),2))
+    else:
+        extra = 0
+    ctx.prec += extra
+    if not ctx._im(z) and ctx._re(z) < 0 and ctx.isint(ctx._re(z)):
+        n = int(ctx.re(z))
+        h = ctx.hyperfac(-n-1)
+        if ((n+1)//2) & 1:
+            h = -h
+        if ctx._is_complex_type(z):
+            return h + 0j
+        return h
+    zp1 = z+1
+    # Wrong branch cut
+    #v = ctx.gamma(zp1)**z
+    #ctx.prec -= extra
+    #return v / ctx.barnesg(zp1)
+    v = ctx.exp(z*ctx.loggamma(zp1))
+    ctx.prec -= extra
+    return v / ctx.barnesg(zp1)
+
+'''
+@defun
+def psi0(ctx, z):
+    """Shortcut for psi(0,z) (the digamma function)"""
+    return ctx.psi(0, z)
+
+@defun
+def psi1(ctx, z):
+    """Shortcut for psi(1,z) (the trigamma function)"""
+    return ctx.psi(1, z)
+
+@defun
+def psi2(ctx, z):
+    """Shortcut for psi(2,z) (the tetragamma function)"""
+    return ctx.psi(2, z)
+
+@defun
+def psi3(ctx, z):
+    """Shortcut for psi(3,z) (the pentagamma function)"""
+    return ctx.psi(3, z)
+'''