snappy 3.2__cp311-cp311-macosx_10_12_x86_64.whl

snappy/ptolemy/coordinates.py

@@ -0,0 +1,2290 @@
+from .component import ZeroDimensionalComponent
+from .rur import RUR
+from . import matrix
+from . import findLoops
+from . import utilities
+from ..sage_helper import _within_sage
+from ..pari import Gen, pari
+import re
+
+
+class PtolemyCannotBeCheckedError(Exception):
+    def __init__(self):
+        msg = (
+            "Use .cross_ratios().check_against_manifold(...) since checking "
+            "Ptolemy coordinates for non-trivial generalized obstruction "
+            "class is not supported.")
+        Exception.__init__(self, msg)
+
+
+class LogToCloseToBranchCutError(Exception):
+    """
+    An exception raised when taking log(-x) for some real number x
+    Due to numerical inaccuracies, we cannot know in this case whether to take
+    -Pi or Pi as imaginary part.
+    """
+    pass
+
+
+class RelationViolationError(Exception):
+    """
+    An exception raised when some supposed relation doesn't hold exactly
+    or numerical.
+    """
+
+    def __init__(self, value, epsilon, comment):
+        self.value = value
+        self.epsilon = epsilon
+        self.comment = comment
+
+    def __str__(self):
+        r = self.comment + " is violated, "
+        r += "difference is %s" % self.value
+        if self.epsilon is None:
+            return r + " (exact values)"
+        return r + " (epsilon = %s)" % self.epsilon
+
+
+class NotPU21Representation:
+    """
+    Returned by is_pu_2_1_representation if cross ratios do not fulfill
+    conditions to be a PU(2,1)-representation.
+    Contains the reason why cross ratios fail to do so.
+    Cast to bool evaluates to False.
+    """
+
+    def __init__(self, reason):
+        self.reason = reason
+
+    def __repr__(self):
+        return "NotPU21Representation(reason = %r)" % self.reason
+
+    def __bool__(self):
+        return False
+
+    __nonzero__ = __bool__ # backwards compatibility python 2x
+
+
+class NumericalMethodError(Exception):
+    def __init__(self, method):
+        self.method = method
+
+    def __str__(self):
+        return "Method %s only supported for numerical values" % self.method
+
+
+class ExactMethodError(Exception):
+    def __init__(self, method):
+        self.method = method
+
+    def __str__(self):
+        return "Method %s only supported for exact values" % self.method
+
+
+def _check_relation(value, epsilon, comment):
+    if epsilon is None:
+        if not value == 0:
+            raise RelationViolationError(value, epsilon, comment)
+    else:
+        if not abs(value) < epsilon:
+            raise RelationViolationError(value, epsilon, comment)
+
+
+class PtolemyCoordinates(dict):
+    """
+    Represents a solution of a Ptolemy variety as python dictionary.
+
+    === Examples ===
+
+    Construct solution from magma output:
+
+    >>> from snappy.ptolemy.processMagmaFile import _magma_output_for_4_1__sl3, solutions_from_magma
+    >>> from snappy import Manifold
+    >>> solutions = solutions_from_magma(_magma_output_for_4_1__sl3)
+    >>> solution = solutions[2]
+
+    Access a Ptolemy coordinate:
+
+    >>> solution['c_2100_0']
+    1
+
+    >>> solution.number_field()
+    x^2 + x + 1
+
+    Solution is always 0 dimensional:
+
+    >>> solution.dimension
+    0
+
+    Check that it is really a solution, exactly:
+
+    >>> solution.check_against_manifold()
+
+    If the solution was not created through the ptolemy module
+    and thus not associated to a manifold, we need to explicitly
+    specify one:
+
+    >>> myDict = {}
+    >>> for key, value in solution.items():
+    ...     myDict[key] = value
+    >>> mysolution = PtolemyCoordinates(myDict)
+    >>> M = Manifold("4_1")
+    >>> solution.check_against_manifold(M)
+
+    Turn into (Galois conjugate) numerical solutions:
+
+    >>> old_precision = pari.set_real_precision(100) # with high precision
+    >>> numerical_solutions = solution.numerical()
+
+    Check that it is a solution, numerically:
+
+    >>> numerical_solutions[0].check_against_manifold(M, 1e-80)
+    >>> pari.set_real_precision(old_precision) # reset pari engine
+    100
+
+    Compute cross ratios from the Ptolemy coordinates (cross ratios
+    according to SnapPy convention, see help(solution.cross_ratios):
+
+    >>> cross = solution.cross_ratios()
+    >>> cross['z_0001_0']
+    Mod(-x, x^2 + x + 1)
+
+    Compute volumes:
+
+    >>> volumes = cross.volume_numerical()
+
+    Check that volume is 4 times the geometric one:
+
+    >>> volume = volumes[0].abs()
+    >>> diff = abs(4 * M.volume() - volume)
+    >>> diff < 1e-9
+    True
+
+    Compute flattenings:
+
+    >>> flattenings = solution.flattenings_numerical()
+
+    Compute complex volumes:
+
+    >>> cvols = [flattening.complex_volume() for flattening in flattenings]
+    >>> volume = cvols[0].real().abs()
+    >>> chernSimons = cvols[0].imag()
+    >>> diff = abs(4 * M.volume() - volume)
+    >>> diff < 1e-9
+    True
+
+    >>> from snappy import pari
+    >>> normalized = chernSimons * 6 / (pari('Pi')**2)
+
+    Check that Chern Simons is zero up to 6 torsion:
+
+    >>> normalized - normalized.round() < 1e-9
+    True
+    """
+
+    def __init__(self, d, is_numerical=True, py_eval_section=None,
+                 manifold_thunk=lambda : None,
+                 non_trivial_generalized_obstruction_class=False):
+
+        self._manifold_thunk = manifold_thunk
+
+        self._is_numerical = is_numerical
+        self.dimension = 0
+
+        self._non_trivial_generalized_obstruction_class = (
+            non_trivial_generalized_obstruction_class)
+        processed_dict = d
+
+        if py_eval_section is not None:
+            # process the extra information that is given by
+            # ptolemyVariety's py_eval_section
+
+            processed_dict = py_eval_section['variable_dict'](d)
+            if py_eval_section.get(
+                    'non_trivial_generalized_obstruction_class'):
+                self._non_trivial_generalized_obstruction_class = True
+
+        # Caches the matrices that label the short and long edges
+        # of the truncated simplices building the manifold
+        self._edge_cache = {}
+
+        # Caches the images of a fundamental group generator
+        self._matrix_cache = []
+        self._inverse_matrix_cache = []
+
+        super().__init__(processed_dict)
+
+    def __repr__(self):
+        dict_repr = dict.__repr__(self)
+        return "PtolemyCoordinates(%s, is_numerical = %r, ...)" % (
+                                   dict_repr, self._is_numerical)
+
+    def _repr_pretty_(self, p, cycle):
+        if cycle:
+            p.text('PtolemyCoordinates(...)')
+        else:
+            with p.group(4, 'PtolemyCoordinates(',')'):
+                p.breakable()
+                p.pretty(dict(self))
+                p.text(',')
+                p.breakable()
+                p.text('is_numerical = %r, ...' % self._is_numerical)
+
+    def get_manifold(self):
+        """
+        Get the manifold for which this structure represents a solution
+        to the Ptolemy variety.
+        """
+
+        return self._manifold_thunk()
+
+    def num_tetrahedra(self):
+        """
+        The number of tetrahedra for which we have Ptolemy coordinates.
+        """
+        return _num_tetrahedra(self)
+
+    def N(self):
+        """
+        Get the *N* where these coordinates are for SL/PSL(*N*, **C**)-representations.
+        """
+
+        N, has_obstruction = _N_and_has_obstruction_for_ptolemys(self)
+        return N
+
+    def has_obstruction(self):
+        """
+        Whether the Ptolemy variety has legacy obstruction class that
+        modifies the Ptolemy relation to
+        """
+        N, has_obstruction = _N_and_has_obstruction_for_ptolemys(self)
+        return has_obstruction
+
+    def number_field(self):
+        """
+        For an exact solution, return the number_field spanned by the
+        Ptolemy coordinates. If number_field is Q, return None.
+        """
+
+        if self._is_numerical:
+            raise ExactMethodError("number_field")
+
+        return _get_number_field(self)
+
+    def numerical(self):
+        """
+        Turn exact solutions into numerical solutions using pari.
+
+        Take an exact solution:
+
+        >>> from snappy.ptolemy.processMagmaFile import _magma_output_for_4_1__sl3, solutions_from_magma
+        >>> solutions = solutions_from_magma(_magma_output_for_4_1__sl3)
+        >>> solution = solutions[2]
+
+        Turn into a numerical solution:
+
+        >>> old_precision = pari.set_real_precision(100) # with high precision
+        >>> numerical_solutions = solution.numerical()
+        >>> pari.set_real_precision(old_precision) # reset pari engine
+        100
+
+        Pick one of the Galois conjugates:
+
+        >>> numerical_solution = numerical_solutions[0]
+        >>> value = numerical_solution['c_1110_0']
+        """
+
+        if self._is_numerical:
+            return self
+        return ZeroDimensionalComponent(
+            [ PtolemyCoordinates(
+                    d, is_numerical=True,
+                    manifold_thunk=self._manifold_thunk,
+                    non_trivial_generalized_obstruction_class=(
+                        self._non_trivial_generalized_obstruction_class))
+              for d in _to_numerical(self) ])
+
+    def to_PUR(self):
+        """
+        If any Ptolemy coordinates are given as Rational Univariate
+        Representation, convert them to Polynomial Univariate Representation and
+        return the result.
+
+        See to_PUR of RUR.
+
+        This conversion might lead to very large coefficients.
+        """
+
+        return PtolemyCoordinates(
+            _apply_to_RURs(self, RUR.to_PUR),
+            is_numerical=self._is_numerical,
+            manifold_thunk=self._manifold_thunk,
+            non_trivial_generalized_obstruction_class=(
+                self._non_trivial_generalized_obstruction_class))
+
+    def multiply_terms_in_RUR(self):
+        """
+        If a Ptolemy coordinate is given as Rational Univariate Representation
+        with numerator and denominator being a product, multiply the terms and
+        return the result.
+
+        See multiply_terms of RUR.
+
+        This loses information about how the numerator and denominator are
+        factorised.
+        """
+
+        return PtolemyCoordinates(
+            _apply_to_RURs(self, RUR.multiply_terms),
+            is_numerical=self._is_numerical,
+            manifold_thunk=self._manifold_thunk,
+            non_trivial_generalized_obstruction_class=(
+                self._non_trivial_generalized_obstruction_class))
+
+    def multiply_and_simplify_terms_in_RUR(self):
+        """
+        If a Ptolemy coordinate is given as Rational Univariate Representation
+        with numerator and denominator being a product, multiply the terms,
+        reduce the fraction and return the result.
+
+        See multiply_and_simplify_terms of RUR.
+
+        This loses information about how the numerator and denominator are
+        factorised.
+
+        """
+
+        return PtolemyCoordinates(
+            _apply_to_RURs(self, RUR.multiply_and_simplify_terms),
+            is_numerical=self._is_numerical,
+            manifold_thunk=self._manifold_thunk,
+            non_trivial_generalized_obstruction_class=(
+                self._non_trivial_generalized_obstruction_class))
+
+    def cross_ratios(self):
+        """
+        Compute cross ratios from Ptolemy coordinates. The cross ratios are
+        according to the SnapPy convention, so we have::
+
+             z = 1 - 1/zp, zp = 1 - 1/zpp, zpp = 1 - 1/z
+
+        where::
+
+             z   is at the edge 01 and equal to   s0 * s1 * (c_1010 * c_0101) / (c_1001 * c_0110)
+             zp  is at the edge 02 and equal to - s0 * s2 * (c_1001 * c_0110) / (c_1100 * c_0011)
+             zpp is at the edge 03 and equal to   s0 * s3 * (c_1100 * c_0011) / (c_0101 * c_1010).
+
+        Note that this is different from the convention used in
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        Take an exact solution:
+
+        >>> from snappy.ptolemy.processMagmaFile import _magma_output_for_4_1__sl3, solutions_from_magma
+        >>> solutions = solutions_from_magma(_magma_output_for_4_1__sl3)
+        >>> solution = solutions[2]
+
+        Turn into cross Ratios:
+
+        >>> crossRatios = solution.cross_ratios()
+
+        Get a cross ratio:
+
+        >>> crossRatios['zp_0010_0']
+        Mod(-x, x^2 + x + 1)
+
+        Check the relationship between cross ratios:
+
+        >>> crossRatios['z_0010_0'] == 1 - 1 / crossRatios['zp_0010_0']
+        True
+
+        >>> crossRatios['zp_0010_0'] == 1 - 1 / crossRatios['zpp_0010_0']
+        True
+
+        >>> crossRatios['zpp_0010_0'] == 1 - 1 / crossRatios['z_0010_0']
+        True
+
+        Get information about what one can do with cross ratios
+        """
+        return CrossRatios(_ptolemy_to_cross_ratio(self)[0],
+                           is_numerical=self._is_numerical,
+                           manifold_thunk=self._manifold_thunk)
+
+    def cross_ratios_numerical(self):
+        """
+        Turn exact solutions into numerical and then compute cross ratios.
+        See numerical() and cross_ratios().
+        """
+
+        if self._is_numerical:
+            return self.cross_ratios()
+        else:
+            return ZeroDimensionalComponent(
+                [num.cross_ratios() for num in self.numerical()])
+
+    def flattenings_numerical(self):
+        """
+        Turn into numerical solutions and compute flattenings, see
+        help(snappy.ptolemy.coordinates.Flattenings)
+        Also see numerical()
+
+        Get Ptolemy coordinates.
+
+        >>> from snappy.ptolemy.processMagmaFile import _magma_output_for_4_1__sl3, solutions_from_magma
+        >>> solutions = solutions_from_magma(_magma_output_for_4_1__sl3)
+        >>> solution = solutions[2]
+
+        Compute a numerical solution
+
+        >>> flattenings = solution.flattenings_numerical()
+
+        Get more information with help(flattenings[0])
+        """
+
+        if self._is_numerical:
+            # Used as a factor when taking log's to shift the branch slightly
+            # from the standard branch cut at the negative real line
+            branch_factor = 1
+
+            # Try different branch cuts 1000 times
+            for i in range(1000):
+                try:
+                    # get the dictionary containing flattenings
+                    # and the evenN that describes in what
+                    # flavor of the Extended Bloch group the result lives in
+                    d, evenN = _ptolemy_to_cross_ratio(
+                        self,
+                        branch_factor,
+                        self._non_trivial_generalized_obstruction_class,
+                        as_flattenings=True)
+
+                    return Flattenings(d,
+                                       manifold_thunk=self._manifold_thunk,
+                                       evenN=evenN)
+                except LogToCloseToBranchCutError:
+                    # Values to close to the branch cut, just multiply
+                    # by a small offset
+                    branch_factor *= pari('exp(0.0001 * I)')
+
+            raise Exception("Could not find non-ambiguous branch cut for log")
+        else:
+            return ZeroDimensionalComponent(
+                [num.flattenings_numerical() for num in self.numerical()])
+
+    def volume_numerical(self, drop_negative_vols=False):
+        """
+        Turn into (Galois conjugate) numerical solutions and compute volumes.
+        If already numerical, only return the one volume.
+        See numerical().
+
+        If drop_negative_vols = True is given as optional argument,
+        only return non-negative volumes.
+        """
+        if self._is_numerical:
+            return self.cross_ratios().volume_numerical()
+        else:
+            vols = ZeroDimensionalComponent(
+                [num.volume_numerical() for num in self.numerical()])
+            if drop_negative_vols:
+                return [vol for vol in vols if vol > -1e-12]
+            return vols
+
+    def complex_volume_numerical(self,
+                                 drop_negative_vols=False,
+                                 with_modulo=False):
+        """
+        Turn into (Galois conjugate) numerical solutions and compute complex
+        volumes. If already numerical, return the volume.
+
+        Complex volume is defined up to i*pi**2/6.
+
+        See numerical(). If drop_negative_vols = True is given as optional
+        argument, only return complex volumes with non-negative real part.
+        """
+
+        if self._is_numerical:
+            return self.flattenings_numerical().complex_volume(
+                with_modulo=with_modulo)
+        else:
+            cvols = ZeroDimensionalComponent(
+                [ num.flattenings_numerical().complex_volume(
+                        with_modulo=with_modulo)
+                  for num in self.numerical()])
+            if drop_negative_vols:
+                return [cvol for cvol in cvols if cvol.real() > -1e-12]
+            return cvols
+
+    def _coordinate_at_tet_and_point(self, tet, pt):
+        """
+        Given the index of a tetrahedron and a quadruple (any iterable) of
+        integer to mark an integral point on that tetrahedron, returns the
+        associated Ptolemy coordinate.
+        If this is a vertex Ptolemy coordinate, always return 1 without
+        checking for it in the dictionary.
+        """
+
+        # Handle the vertex Ptolemy coordinate case
+        if sum(pt) in pt:
+            return 1
+
+        # Normal case
+        return self['c_%d%d%d%d' % tuple(pt) + '_%d' % tet]
+
+    def _get_obstruction_variable(self, face, tet):
+        """
+        Get the obstruction variable sigma_face for the given face and
+        tetrahedron. Return 1 if there is no such obstruction class.
+        """
+
+        key = "s_%d_%d" % (face, tet)
+        return self.get(key, +1) # Default to 1 if no obstruction class given
+
+    @staticmethod
+    def _three_perm_sign(v0, v1, v2):
+        """
+        Returns the sign of the permutation necessary to bring the three
+        elements v0, v1, v2 in order.
+        """
+        if v0 < v2 and v2 < v1:
+            return -1
+        if v1 < v0 and v0 < v2:
+            return -1
+        if v2 < v1 and v1 < v0:
+            return -1
+        return +1
+
+    def diamond_coordinate(self, tet, v0, v1, v2, pt):
+        """
+        Returns the diamond coordinate for tetrahedron with index tet
+        for the face with vertices v0, v1, v2 (integers between 0 and 3) and
+        integral point pt (quadruple adding up to N-2).
+
+        See Definition 10.1:
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+        """
+
+        # Integral points that are indices of Ptolemy coordinates
+        pt_v0_v0 = [ a + 2 * _kronecker_delta(v0, i)
+                          for i, a in enumerate(pt) ]
+        pt_v0_v1 = [ a + _kronecker_delta(v0, i) + _kronecker_delta(v1, i)
+                          for i, a in enumerate(pt) ]
+        pt_v0_v2 = [ a + _kronecker_delta(v0, i) + _kronecker_delta(v2, i)
+                          for i, a in enumerate(pt) ]
+        pt_v1_v2 = [ a + _kronecker_delta(v1, i) + _kronecker_delta(v2, i)
+                          for i, a in enumerate(pt) ]
+
+        # Ptolemy coordinates involved
+        c_pt_v0_v0 = self._coordinate_at_tet_and_point(tet, pt_v0_v0)
+        c_pt_v0_v1 = self._coordinate_at_tet_and_point(tet, pt_v0_v1)
+        c_pt_v0_v2 = self._coordinate_at_tet_and_point(tet, pt_v0_v2)
+        c_pt_v1_v2 = self._coordinate_at_tet_and_point(tet, pt_v1_v2)
+
+        # Obstruction variable
+        # See Definition 9.23 of
+        # Garoufalidis, Thurston, Zickert
+        # The Complex Volume of SL(n,C)-Representations of 3-Manifolds
+        # httpss://arxiv.org/abs/1111.2828
+        face = next(iter(set(range(4)) - {v0, v1, v2}))
+        obstruction = self._get_obstruction_variable(face, tet)
+
+        # The epsilon permutation sign from Definition 10.1 of
+        # Garoufalidis, Goerner, Zickert:
+        # Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        # https://arxiv.org/abs/1207.6711
+        s = PtolemyCoordinates._three_perm_sign(v0, v1, v2)
+
+        # The equation from the same Definition
+        return - (obstruction * s *
+                  (c_pt_v0_v0 * c_pt_v1_v2) /
+                  (c_pt_v0_v1 * c_pt_v0_v2))
+
+    def ratio_coordinate(self, tet, v0, v1, pt):
+        """
+        Returns the ratio coordinate for tetrahedron with index tet
+        for the edge from v0 to v1 (integers between 0 and 3) and integral
+        point pt (quadruple adding up N-1) on the edge.
+
+        See Definition 10.2:
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        Note that this definition turned out to have the wrong sign. Multiply
+        the result by -1 if v1 < v0 and N is even.
+        """
+
+        # Integral points on the edge
+        pt_v0 = [ a + _kronecker_delta(v0, i) for i, a in enumerate(pt) ]
+        pt_v1 = [ a + _kronecker_delta(v1, i) for i, a in enumerate(pt) ]
+
+        # Ptolemy coordinates at those integral points
+        c_pt_v0 = self._coordinate_at_tet_and_point(tet, pt_v0)
+        c_pt_v1 = self._coordinate_at_tet_and_point(tet, pt_v1)
+
+        # Sign
+        s = (-1) ** pt[v1]
+
+        if v1 < v0 and (self.N() % 2 == 0):
+            s *= -1
+
+        # Equation from Definition 10.2
+        return s * c_pt_v1 / c_pt_v0
+
+    def _get_identity_matrix(self):
+
+        # Get N
+        N = self.N()
+
+        return [[_kronecker_delta(i, j) for i in range(N)] for j in range(N)]
+
+    def long_edge(self, tet, v0, v1, v2):
+        """
+        The matrix that labels a long edge from v0 to v1 (integers between 0
+        and 3) of a (doubly) truncated simplex corresponding to an ideal
+        tetrahedron with index tet.
+
+        This matrix was labeled alpha^{v0v1v2} (v2 does not matter for non
+        double-truncated simplex) in Figure 18 of
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        It is computed using equation 10.4. Note that the ratio coordinate
+        is different from the definition in the paper (see ratio_coordinate).
+
+        The resulting matrix is given as a python list of lists.
+        """
+
+        # Key for the cache
+        key = 'long_%d_%d%d' % (tet, v0, v1)
+
+        # Fill cache if necessary
+        if key not in self._edge_cache:
+
+            # Get N
+            N = self.N()
+
+            # Start with the 0 matrix
+            m = [[0 for i in range(N)] for j in range(N)]
+
+            # Traverse the edge to fill the counter diagonal elements
+            # with the ratio coordinates
+            for c in range(N):
+                r = N - 1 - c
+                pt = [ r * _kronecker_delta(v0, i) + c * _kronecker_delta(v1, i)
+                       for i in range(4) ]
+                m[r][c] = self.ratio_coordinate(tet, v0, v1, pt)
+
+            # Set in cache
+            self._edge_cache[key] = m
+
+        # Return
+        return self._edge_cache[key]
+
+    def middle_edge(self, tet, v0, v1, v2):
+        """
+        The matrix that labels a middle edge on the face v0, v1, v2 (integers
+        between 0 and 3) of a doubly truncated simplex (or a short edge of the
+        truncated simplex) corresponding to an ideal tetrahedron with index
+        tet.
+
+        This matrix was labeled beta^{v0v1v2} in Figure 18 of
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        It is computed using equation 10.4.
+
+        The resulting matrix is given as a python list of lists.
+        """
+
+        # Key for the cache
+        key = 'middle_%d_%d%d%d' % (tet, v0, v1, v2)
+
+        # Fill cache if necessary
+        if key not in self._edge_cache:
+
+            # Get N
+            N = self.N()
+
+            # Start with identity
+            m = self._get_identity_matrix()
+
+            # Compute the product in equation 10.4
+            for a0, a1, a2 in utilities.triples_with_fixed_sum_iterator(N - 2):
+
+                # Get integral point for diamond coordinate
+                pt = [ a1 * _kronecker_delta(v0, i) +
+                       a2 * _kronecker_delta(v1, i) +
+                       a0 * _kronecker_delta(v2, i) for i in range(4) ]
+
+                # Compute diamond coordinate
+                diamond = self.diamond_coordinate(tet, v0, v1, v2, pt)
+
+                # Multiply result with the x matrix
+                m = matrix.matrix_mult(m, _X(N, a1 + 1, diamond))
+
+            # Fill cache
+            self._edge_cache[key] = m
+
+        return self._edge_cache[key]
+
+    def short_edge(self, tet, v0, v1, v2):
+        """
+        Returns the identity. This is because we can think of the matrices
+        given by Ptolemy coordinates of living on truncated simplices which
+        can be though of as doubly truncated simplices where all short edges
+        are collapsed, hence labeled by the identity.
+
+        See equation 10.4 in
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+        """
+
+        # Key for the cache
+        key = 'short'
+
+        # Fill cache if necessary
+        if key not in self._edge_cache:
+
+            # Get N
+            N = self.N()
+
+            # Take the identity matrix
+            m = self._get_identity_matrix()
+
+            # Fill cache
+            self._edge_cache[key] = m
+
+        return self._edge_cache[key]
+
+    def _init_matrix_and_inverse_cache(self):
+        # Fill the caches of matrices corresponding to the
+        # fundamental group generators and their inverses
+
+        if self._matrix_cache and self._inverse_matrix_cache:
+            return
+
+        # Compute all the matrices for the generators and there inverses
+        # The short edges of the doubly truncated simplices are all identities
+        # (they will be collapset in the truncated simplex), thus we give them
+        # no penalty.
+        self._matrix_cache, self._inverse_matrix_cache = (
+            findLoops.images_of_original_generators(self,
+                                                    penalties=(0, 1, 1)))
+
+    def evaluate_word(self, word, G=None):
+        """
+        Given a word in the generators of the fundamental group,
+        compute the corresponding matrix. By default, these are the
+        generators of the unsimplified presentation of the fundamental
+        group. An optional SnapPy fundamental group can be given if the
+        words are in generators of a different presentation, e.g.,
+        c.evaluate_word(word, M.fundamental_group(True)) to
+        evaluate a word in the simplified presentation returned by
+        M.fundamental_group(True).
+
+        For now, the matrix is returned as list of lists.
+        """
+
+        # Init the matrices corresponding to generators
+        self._init_matrix_and_inverse_cache()
+
+        return findLoops.evaluate_word(
+            self._get_identity_matrix(),
+            self._matrix_cache,
+            self._inverse_matrix_cache,
+            word,
+            G)
+
+    def _testing_assert_identity(self, m,
+                                 allow_sign_if_obstruction_class=False):
+
+        N = self.N()
+
+        null = [[0 for i in range(N)] for j in range(N)]
+        identity = self._get_identity_matrix()
+
+        if allow_sign_if_obstruction_class and self.has_obstruction():
+
+            if not (matrix.matrix_add(m, identity) == null or
+                    matrix.matrix_sub(m, identity) == null):
+                raise Exception("Cocycle condition violated: %s" % m)
+
+        else:
+
+            if not matrix.matrix_sub(m, identity) == null:
+                raise Exception("Cocycle condition violated: %s" % m)
+
+    def _testing_check_cocycles(self):
+        for tet in range(self.num_tetrahedra()):
+            # Check middle edges is inverse when direction reversed
+            for v in [(0,1,2),(0,1,3),(0,2,1),(0,2,3),(0,3,1),(0,3,2),
+                      (1,0,2),(1,0,3),(1,2,0),(1,2,3),(1,3,0),(1,3,2),
+                      (2,0,1),(2,0,3),(2,1,0),(2,1,3),(2,3,0),(2,3,1),
+                      (3,0,1),(3,0,2),(3,1,0),(3,1,2),(3,2,0),(3,2,1)]:
+                m1 = self.middle_edge(tet,v[0],v[1],v[2])
+                m2 = self.middle_edge(tet,v[0],v[2],v[1])
+                self._testing_assert_identity(
+                    matrix.matrix_mult(m1, m2))
+
+            # Check long edges is inverse when direction reversed
+            for v in [(0,1,2), (0,2,1), (0,3,1),
+                      (1,0,2), (1,2,0), (1,3,0),
+                      (2,0,1), (2,1,0), (2,3,0),
+                      (3,0,1), (3,1,0), (3,2,0)]:
+                m1 = self.long_edge(tet,v[0],v[1],v[2])
+                m2 = self.long_edge(tet,v[1],v[0],v[2])
+                self._testing_assert_identity(
+                    matrix.matrix_mult(m1, m2))
+
+            # Check triangle for each vertex
+            for v in [(0,1,2,3), (1,2,3,0), (2,3,0,1), (3,0,1,2)]:
+                m1 = self.middle_edge(tet, v[0], v[1], v[2])
+                m2 = self.middle_edge(tet, v[0], v[2], v[3])
+                m3 = self.middle_edge(tet, v[0], v[3], v[1])
+
+                self._testing_assert_identity(
+                    matrix.matrix_mult(
+                        m1, matrix.matrix_mult(m2, m3)))
+
+            # Check hexagon for each face
+            for v in [(0,1,2), (0,1,3), (0,2,3), (1,2,3)]:
+                m1 = self.middle_edge(tet,v[0],v[1],v[2])
+                m2 = self.long_edge(  tet,v[0],v[2],v[1])
+                m3 = self.middle_edge(tet,v[2],v[0],v[1])
+                m4 = self.long_edge(  tet,v[2],v[1],v[0])
+                m5 = self.middle_edge(tet,v[1],v[2],v[0])
+                m6 = self.long_edge(  tet,v[1],v[0],v[2])
+                self._testing_assert_identity(
+                    matrix.matrix_mult(
+                        m1,
+                        matrix.matrix_mult(
+                            m2,
+                            matrix.matrix_mult(
+                                m3,
+                                matrix.matrix_mult(
+                                    m4,
+                                    matrix.matrix_mult(m5,m6))))), True)
+
+    def check_against_manifold(self, M=None, epsilon=None):
+        """
+        Checks that the given solution really is a solution to the Ptolemy
+        variety of a manifold. See help(ptolemy.PtolemyCoordinates) for
+        more example.
+
+        === Arguments ===
+
+        * M --- manifold to check this for
+        * epsilon --- maximal allowed error when checking the relations, use
+          None for exact comparison.
+        """
+
+        if M is None:
+            M = self.get_manifold()
+
+        if M is None:
+            raise Exception("Need to give manifold")
+
+        if self._non_trivial_generalized_obstruction_class:
+            raise PtolemyCannotBeCheckedError()
+
+        num_tets = self.num_tetrahedra()
+        N, has_obstruction_class = _N_and_has_obstruction_for_ptolemys(self)
+
+        if not M.num_tetrahedra() == num_tets:
+            raise Exception("Number tetrahedra not matching")
+
+        if has_obstruction_class:
+            # check cocycle condition
+            for tet in range(num_tets):
+                _check_relation(
+                    self._get_obstruction_variable(0, tet) *
+                    self._get_obstruction_variable(1, tet) *
+                    self._get_obstruction_variable(2, tet) *
+                    self._get_obstruction_variable(3, tet) - 1,
+                    epsilon,
+                    "Obstruction cocycle")
+            # check identified faces
+            for dummy_sign, power, var1, var2 in (
+                    M._ptolemy_equations_identified_face_classes()):
+                _check_relation(
+                    self[var1] - self[var2],
+                    epsilon,
+                    "Identification of face classes")
+
+        # Check identified Ptolemy coordinates
+        for sign, power, var1, var2 in (
+                M._ptolemy_equations_identified_coordinates(N)):
+            _check_relation(
+                self[var1] - sign * self[var2],
+                epsilon,
+                "Identification of Ptolemy coordinates")
+
+        # Check Ptolemy relationship
+        for tet in range(num_tets):
+            for index in utilities.quadruples_with_fixed_sum_iterator(N - 2):
+
+                def get_ptolemy_coordinate(addl_index):
+                    total_index = matrix.vector_add(index, addl_index)
+                    key = "c_%d%d%d%d" % tuple(total_index) + "_%d" % tet
+                    return self[key]
+
+                s0 = self._get_obstruction_variable(0, tet)
+                s1 = self._get_obstruction_variable(1, tet)
+                s2 = self._get_obstruction_variable(2, tet)
+                s3 = self._get_obstruction_variable(3, tet)
+
+                rel = (  s0 * s1 * get_ptolemy_coordinate((1,1,0,0))
+                                 * get_ptolemy_coordinate((0,0,1,1))
+                       - s0 * s2 * get_ptolemy_coordinate((1,0,1,0))
+                                 * get_ptolemy_coordinate((0,1,0,1))
+                       + s0 * s3 * get_ptolemy_coordinate((1,0,0,1))
+                                 * get_ptolemy_coordinate((0,1,1,0)))
+
+                _check_relation(rel,
+                                epsilon,
+                                "Ptolemy relation")
+
+    def is_geometric(self, epsilon=1e-6):
+        """
+        Returns true if all shapes corresponding to this solution have positive
+        imaginary part.
+
+        If the solutions are exact, it returns true if one of the corresponding
+        numerical solutions is geometric.
+
+        An optional epsilon can be given. An imaginary part of a shape is
+        considered positive if it is larger than this epsilon.
+        """
+
+        if self._is_numerical:
+            return self.cross_ratios().is_geometric(epsilon)
+        else:
+            for numerical_sol in self.numerical():
+                if numerical_sol.is_geometric(epsilon):
+                    return True
+            return False
+
+
+class Flattenings(dict):
+    """
+    Represents a flattening assigned to each edge of a simplex as dictionary.
+
+    We assign to each pair of parallel edges of each simplex a triple (w, z, p)
+    such that::
+
+           w = log(z) + p * (2 * pi * i / N)   where N is fixed and even.
+
+    For N = 2, the three triples belonging to a simplex form a combinatorial
+    flattening (w0, w1, w2) as defined in Definition 3.1 in
+    Walter D. Neumann, Extended Bloch group and the Cheeger-Chern-Simons class
+    https://arxiv.org/abs/math.GT/0307092
+
+    For N > 2, the three triples form a generalized combinatorial flattening
+    (w0, w1, w2) that gives an element in the generalized Extended Bloch group
+    which is the Extended Bloch group corresponding to the Riemann surface
+    given by::
+
+                 u1 * e^w0 + u2 * e^w1 = 1
+
+    where u1^N = u2^N = 1.
+
+    A representation in SL(n,C) and SL(n,C)/{+1,-1} with n even gives an element
+    in the generalized Extended Bloch group for N = 2.
+    A representation in PSL(n,C) with n even in the group for N = n.
+    A representation in PSL(n,C) with n odd in the group for N = 2 * n.
+
+    This work has not been published yet.
+
+    If f is a flattening, then in the notation of Neumann, the value of::
+
+        f['z_xxxx_y']    is (w0, z, p)
+        f['zp_xxxx_y']   is (w1, z', q)
+        f['zpp_xxxx_y']  is (w2, z'', r).
+    """
+
+    def __init__(self, d, manifold_thunk=lambda : None, evenN=2):
+        super().__init__(d)
+        self._is_numerical = True
+        self._manifold_thunk = manifold_thunk
+        self.dimension = 0
+
+        # The N for which we get the generalized Extended Bloch group
+        self._evenN = evenN
+
+    def __repr__(self):
+        dict_repr = dict.__repr__(self)
+        return "Flattenings(%s, ...)" % dict_repr
+
+    def _repr_pretty_(self, p, cycle):
+        if cycle:
+            p.text('Flattenings(...)')
+        else:
+            with p.group(4, 'Flattenings(',')'):
+                p.breakable()
+                p.pretty(dict(self))
+                p.text(', ...')
+
+    def get_manifold(self):
+        """
+        Get the manifold for which this structure represents a flattening.
+        """
+
+        return self._manifold_thunk()
+
+    def num_tetrahedra(self):
+        """
+        The number of tetrahedra for which we have cross ratios.
+        """
+        return _num_tetrahedra(self)
+
+    def N(self):
+        """
+        Get the N such that these cross ratios are for
+        SL/PSL(N,C)-representations.
+        """
+
+        return _N_for_shapes(self)
+
+    @classmethod
+    def from_tetrahedra_shapes_of_manifold(cls, M):
+        """
+        Takes as argument a manifold and produces (weak) flattenings using
+        the tetrahedra_shapes of the manifold M.
+
+        >>> from snappy import Manifold
+        >>> M = Manifold("5_2")
+        >>> flattenings = Flattenings.from_tetrahedra_shapes_of_manifold(M)
+        >>> flattenings.check_against_manifold(M)
+        >>> flattenings.check_against_manifold()
+        """
+
+        PiI = pari('Pi * I')
+
+        num_tets = M.num_tetrahedra()
+
+        z_cross_ratios = M.tetrahedra_shapes(
+            part='rect', dec_prec=pari.get_real_precision())
+
+        all_cross_ratios = sum(
+            [ [z, 1 / (1-z), 1 - 1/z] for z in z_cross_ratios], [])
+
+        log_all_cross_ratios = [ z.log() for z in all_cross_ratios ]
+
+        def flattening_condition(r):
+            return (3 * r * [0]
+                    + 3 * [1]
+                    + 3 * (num_tets - r - 1) * [0])
+
+        flattening_conditions = [
+            flattening_condition(r) for r in range(num_tets)]
+
+        try:  # works for snappy.SnapPy.SimpleMatrix
+            equations = M.gluing_equations().data
+        except AttributeError:  # works Sage's matrix class
+            equations = [
+                [ int(c) for c in row] for row in M.gluing_equations().rows()]
+
+        all_equations = equations + flattening_conditions
+
+        u, v, d_mat = matrix.smith_normal_form(all_equations)
+
+        extra_cols = len(all_equations[0]) - len(all_equations)
+
+        d = [d_mat[r][r + extra_cols] for r in range(len(d_mat))]
+
+        # errors to the gluing equations and flattening condition
+        # when using the logarithms without adding p * pi * i as complex
+        # numbers
+        errors = matrix.matrix_mult_vector(all_equations,
+                                           log_all_cross_ratios)
+
+        # divide by pi * i and turn into integers
+        int_errors = [ (x / PiI).real().round() for x in errors ]
+
+        int_errors_in_other_basis = matrix.matrix_mult_vector(u, int_errors)
+
+        def quotient(x, y):
+            if x == 0 and y == 0:
+                return 0
+
+            assert x % y == 0, "%s %s" % (x, y)
+            return x / y
+
+        flattenings_in_other_basis = (
+            extra_cols * [0] +
+            [ - quotient(x, y)
+              for x, y in zip(int_errors_in_other_basis, d) ])
+
+        flattenings = matrix.matrix_mult_vector(v, flattenings_in_other_basis)
+
+        assert (matrix.matrix_mult_vector(all_equations, flattenings) ==
+                [-x for x in int_errors])
+
+        keys = sum([ ['z_0000_%d' % i,
+                      'zp_0000_%d' % i,
+                      'zpp_0000_%d' % i] for i in range(num_tets)],[])
+
+        Mcopy = M.copy()
+
+        return Flattenings(
+            {k: (log + PiI * p, z, p)
+                   for k, log, z, p in zip(keys, log_all_cross_ratios,
+                                           all_cross_ratios, flattenings)},
+            manifold_thunk=lambda : Mcopy)
+
+    def get_order(self):
+        """
+        Returns the number N. This flattening represents an element in the
+        generalized Extended Bloch group for the Riemann surface given by
+        u1 * e^w0 + u2 * e^w1 = 1 where u1^N = u2^N = 1.
+        """
+
+        return self._evenN
+
+    def get_zpq_triple(self, key_z):
+        """
+        Gives a flattening as triple [z;p,q] representing an element
+        in the generalized Extended Bloch group similar to the way the
+        triple [z;p,q] is used in Lemma 3.2 in
+        Walter D. Neumann, Extended Bloch group and the Cheeger-Chern-Simons class
+        https://arxiv.org/abs/math.GT/0307092
+        """
+        if not key_z[:2] == 'z_':
+            raise Exception("Need to be called with cross ratio variable z_....")
+        key_zp = 'zp_' + key_z[2:]
+
+        w, z, p = self[key_z]
+        wp, zp, q_canonical_branch_cut = self[key_zp]
+
+        # Note that the q in l(z;p,q) and in Definition 3.1 are different if
+        # z is on the real axis and > 1!!!
+        # Thus we need to compute the q again here according to the formula
+        # for l(z;p,q)
+
+        pari_z = _convert_to_pari_float(z)
+
+        f = pari('2 * Pi * I') / self._evenN
+
+        q_dilog_branch_cut = ((wp + (1-pari_z).log()) / f).round()
+
+        return (z, p, q_dilog_branch_cut)
+
+    def complex_volume(self, with_modulo=False):
+        """
+        Compute complex volume. The complex volume is defined only up to
+        some multiple of m where m = i * pi**2/6 for PSL(2,C) and SL(N,C)
+        and m = i * pi**2/18 for PSL(3,C).
+
+        When called with with_modulo = True, gives a pair
+        (volume, m)
+        """
+
+        if self._evenN == 2:
+            m = pari('Pi^2/6')
+        else:
+            m = pari('Pi^2/18')
+
+        sum_L_functions = sum(
+            [
+                _L_function(
+                    self.get_zpq_triple(key), self._evenN)
+                for key in list(self.keys())
+                if key[:2] == 'z_' ])
+
+        cvol = sum_L_functions / pari('I')
+        vol = cvol.real()
+        cs = cvol.imag() % m
+
+        if cs > m/2 + pari('1e-12'):
+            cs = cs - m
+
+        cvol = vol + cs * pari('I')
+
+        if with_modulo:
+            if self._evenN not in [2, 6]:
+                raise Exception("Unknown torsion")
+
+            return cvol, m * pari('I')
+        return cvol
+
+    def check_against_manifold(self, M=None, epsilon=1e-10):
+        """
+        Checks that the flattening really is a solution to the logarithmic
+        PGL(N,C) gluing equations of a manifold. Usage similar to
+        check_against_manifold of Ptolemy Coordinates, see
+        help(ptolemy.Coordinates) for similar examples.
+
+        === Arguments ===
+
+        M --- manifold to check this for
+        epsilon --- maximal allowed error when checking the equations
+        """
+
+        if M is None:
+            M = self.get_manifold()
+
+        if M is None:
+            raise Exception("Need to give manifold")
+
+        f = pari('2 * Pi * I') / self._evenN
+
+        for w, z, p in list(self.values()):
+            _check_relation(
+                w - (z.log() + f * p),
+                epsilon,
+                "Flattening relation w == log(z) + PiI * p")
+
+        for k in list(self.keys()):
+            if k[:2] == 'z_':
+                w, z, p = self[k]
+                wp, zp, q = self['zp_'+k[2:]]
+                wpp, zpp, r = self['zpp_'+k[2:]]
+                _check_relation(
+                    w + wp + wpp,
+                    epsilon,
+                    "Flattening relation w0 + w1 + w2 == 0")
+
+        some_z = list(self.keys())[0]
+        variable_name, index, tet_index = some_z.split('_')
+        if variable_name not in ['z', 'zp', 'zpp']:
+            raise Exception("Variable not z, zp, or, zpp")
+        if len(index) != 4:
+            raise Exception("Not 4 indices")
+        N = sum([int(x) for x in index]) + 2
+
+        matrix_with_explanations = M.gluing_equations_pgl(
+            N, equation_type='all')
+
+        matrix = matrix_with_explanations.matrix
+        rows = matrix_with_explanations.explain_rows
+        cols = matrix_with_explanations.explain_columns
+
+        for row in range(len(rows)):
+            s = 0
+            for col in range(len(cols)):
+                flattening_variable = cols[col]
+                w, z, p = self[flattening_variable]
+                s = s + w
+            _check_relation(
+                s,
+                epsilon,
+                "Gluing equation %s" % rows[row])
+
+
+class CrossRatios(dict):
+    """
+    Represents assigned shape parameters/cross ratios as
+    dictionary. The cross ratios are according to SnapPy convention, so we
+    have::
+
+        z = 1 - 1/zp, zp = 1 - 1/zpp, zpp = 1 - 1/z
+
+    where::
+
+        z   is at the edge 01 and equal to s0 * s1 * (c_1010 * c_0101) / (c_1001 * c_0110)
+        zp  is at the edge 02 and equal to s0 * s2 * (c_1001 * c_0110) / (c_1100 * c_0011)
+        zpp is at the edge 03 and equal to s0 * s3 * (c_1100 * c_0011) / (c_0101 * c_1010).
+
+    Note that this is different from the convention used in
+    Garoufalidis, Goerner, Zickert:
+    Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+    https://arxiv.org/abs/1207.6711
+    """
+
+    def __init__(self, d, is_numerical=True, manifold_thunk=None):
+        super().__init__(d)
+        self._is_numerical = is_numerical
+        self._manifold_thunk = manifold_thunk
+
+        # Caches the matrices that label the short and long edges
+        # of the truncated simplices building the manifold
+        self._edge_cache = {}
+
+        # Caches the images of a fundamental group generator
+        self._matrix_cache = []
+        self._inverse_matrix_cache = []
+
+        self.dimension = 0
+
+    @staticmethod
+    def from_snappy_manifold(M, dec_prec=None, bits_prec=None,
+                             intervals=False):
+        """
+        Constructs an assignment of shape parameters/cross ratios using
+        the tetrahehdra_shapes method of a given SnapPy manifold. The optional
+        parameters are the same as that of tetrahedra_shapes.
+        """
+
+        shapes = M.tetrahedra_shapes('rect', dec_prec=dec_prec,
+                                     bits_prec=bits_prec,
+                                     intervals=intervals)
+        d = {}
+        for i, shape in enumerate(shapes):
+            d['z_0000_%d' % i] = shape
+            d['zp_0000_%d' % i] = 1 / (1 - shape)
+            d['zpp_0000_%d' % i] = 1 - 1 / shape
+
+        return CrossRatios(d, is_numerical=True,
+                           manifold_thunk=lambda M=M: M)
+
+    def __repr__(self):
+        dict_repr = dict.__repr__(self)
+        return "CrossRatios(%s, is_numerical = %r, ...)" % (
+                            dict_repr, self._is_numerical)
+
+    def _repr_pretty_(self, p, cycle):
+        if cycle:
+            p.text('CrossRatios(...)')
+        else:
+            with p.group(4, 'CrossRatios(',')'):
+                p.breakable()
+                p.pretty(dict(self))
+                p.text(',')
+                p.breakable()
+                p.text('is_numerical = %r, ...' % self._is_numerical)
+
+    def get_manifold(self):
+        """
+        Get the manifold for which this structure represents a solution
+        to the gluing equations.
+        """
+
+        return self._manifold_thunk()
+
+    def num_tetrahedra(self):
+        """
+        The number of tetrahedra for which we have cross ratios.
+        """
+        return _num_tetrahedra(self)
+
+    def N(self):
+        """
+        Get the N such that these cross ratios are for
+        SL/PSL(N,C)-representations.
+        """
+
+        return _N_for_shapes(self)
+
+    def numerical(self):
+        """
+        Turn exact solutions into numerical solutions using pari. Similar to
+        numerical() of PtolemyCoordinates. See help(ptolemy.PtolemyCoordinates)
+        for example.
+        """
+        if self._is_numerical:
+            return self
+        return ZeroDimensionalComponent([
+            CrossRatios(d, is_numerical=True,
+                        manifold_thunk=self._manifold_thunk)
+            for d in _to_numerical(self) ])
+
+    def to_PUR(self):
+        """
+        If any Ptolemy coordinates are given as Rational Univariate
+        Representation, convert them to Polynomial Univariate Representation and
+        return the result.
+
+        See to_PUR of RUR.
+
+        This conversion might lead to very large coefficients.
+        """
+
+        return CrossRatios(
+            _apply_to_RURs(self, RUR.to_PUR),
+            is_numerical=self._is_numerical,
+            manifold_thunk=self._manifold_thunk)
+
+    def multiply_terms_in_RUR(self):
+        """
+        If a cross ratio is given as Rational Univariate Representation
+        with numerator and denominator being a product, multiply the terms and
+        return the result.
+
+        See multiply_terms of RUR.
+
+        This loses information about how the numerator and denominator are
+        factorised.
+        """
+
+        return CrossRatios(
+            _apply_to_RURs(self, RUR.multiply_terms),
+            is_numerical=self._is_numerical,
+            manifold_thunk=self._manifold_thunk)
+
+    def multiply_and_simplify_terms_in_RUR(self):
+        """
+        If a cross ratio is given as Rational Univariate Representation
+        with numerator and denominator being a product, multiply the terms,
+        reduce the fraction and return the result.
+
+        See multiply_and_simplify_terms of RUR.
+
+        This loses information about how the numerator and denominator are
+        factorised.
+
+        """
+
+        return CrossRatios(
+            _apply_to_RURs(self, RUR.multiply_and_simplify_terms),
+            is_numerical=self._is_numerical,
+            manifold_thunk=self._manifold_thunk)
+
+    def volume_numerical(self, drop_negative_vols=False):
+        """
+        Turn into (Galois conjugate) numerical solutions and compute volumes.
+        If already numerical, only compute the one volume.
+        See numerical().
+
+        If drop_negative_vols = True is given as optional argument,
+        only return non-negative volumes.
+        """
+        if self._is_numerical:
+            return sum([_volume(z) for key, z in list(self.items()) if 'z_' in key])
+        else:
+            vols = ZeroDimensionalComponent(
+                [num.volume_numerical() for num in self.numerical()])
+            if drop_negative_vols:
+                return [vol for vol in vols if vol > -1e-12]
+            return vols
+
+    @staticmethod
+    def _cyclic_three_perm_sign(v0, v1, v2):
+        """
+        Returns +1 or -1. It is +1 if and only if (v0, v1, v2) is in the
+        orbit of (0, 1, 2) under the A4-action.
+        """
+
+        for t in [(v0,v1,v2), (v1,v2,v0), (v2,v0,v1)]:
+            if t in [(0,1,2), (1,3,2), (2,3,0), (3,1,0)]:
+                return +1
+        return -1
+
+    def _shape_at_tet_point_and_edge(self, tet, pt, edge):
+        """
+        Given the index of a tetrahedron and two quadruples (any iterabel) of
+        integers, give the cross ratio at that integral point and edge of that
+        tetrahedron.
+        This method translates the SnapPy conventions of labeling simplices
+        and the conventions in Definition 4.2 of
+
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+        """
+
+        postfix = '_%d%d%d%d' % tuple(pt) + '_%d' % tet
+
+        if tuple(edge) in [(1,1,0,0), (0,0,1,1)]:
+            return self['z' + postfix]
+
+        if tuple(edge) in [(1,0,1,0), (0,1,0,1)]:
+            return self['zp' + postfix]
+
+        if tuple(edge) in [(1,0,0,1), (0,1,1,0)]:
+            return self['zpp' + postfix]
+
+        raise Exception("Invalid edge " + str(edge))
+
+    def x_coordinate(self, tet, pt):
+        """
+        Returns the X-coordinate for the tetrahedron with index tet
+        at the point pt (quadruple of integers adding up to N).
+
+        See Definition 10.9:
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+        """
+
+        result = 1
+
+        for v0 in range(4):
+            for v1 in range(v0 + 1, 4):
+                e = [ _kronecker_delta(v0, i) +
+                      _kronecker_delta(v1, i) for i in range(4) ]
+                p = [ x1 - x2 for x1, x2 in zip(pt, e) ]
+                if all(x >= 0 for x in p):
+                    result *= self._shape_at_tet_point_and_edge(tet, p, e)
+
+        return -result
+
+    def _get_identity_matrix(self):
+
+        # Get N
+        N = self.N()
+
+        return [[_kronecker_delta(i, j) for i in range(N)] for j in range(N)]
+
+    def long_edge(self, tet, v0, v1, v2):
+        """
+        The matrix that labels a long edge starting at vertex (v0, v1, v2)
+        of a doubly truncated simplex corresponding to the ideal tetrahedron
+        with index tet.
+
+        This matrix was labeled alpha^{v0v1v2} in Figure 18 of
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        It is computed using equation 10.22.
+
+        The resulting matrix is given as a python list of lists.
+        """
+
+        # Key for cache
+        key = 'long_edge'
+
+        # Fill cache if necessary
+        if key not in self._edge_cache:
+
+            # Get N
+            N = self.N()
+
+            # It is just the counter diagonal matrix
+            m = [ [ _kronecker_delta(i+j, N-1) for i in range(N) ]
+                  for j in range(N)]
+
+            # Set in cache
+            self._edge_cache[key] = m
+
+        return self._edge_cache[key]
+
+    def middle_edge(self, tet, v0, v1, v2):
+        """
+        The matrix that labels a middle edge starting at vertex (v0, v1, v2)
+        of a doubly truncated simplex corresponding to the ideal tetrahedron
+        with index tet.
+
+        This matrix was labeled beta^{v0v1v2} in Figure 18 of
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        It is computed using equation 10.22.
+
+        The resulting matrix is given as a python list of lists.
+        """
+
+        # Key for the cache
+        key = 'middle_%d_%d%d%d' % (tet, v0, v1, v2)
+
+        # Fill cache if necessary
+        if key not in self._edge_cache:
+
+            # Get N
+            N = self.N()
+
+            # The epsilon permutation sign
+            sgn = CrossRatios._cyclic_three_perm_sign(v0, v1, v2)
+
+            # Start with identity
+            m = self._get_identity_matrix()
+
+            for k in range(1, N):
+                # Compute first product
+                prod1 = self._get_identity_matrix()
+                for i in range(1, N - k + 1):
+                    prod1 = matrix.matrix_mult(prod1, _X(N, i, 1))
+
+                # Compute second product
+                prod2 = self._get_identity_matrix()
+                for i in range(1, N - k):
+                    pt = [ k * _kronecker_delta(v2, j) +
+                           i * _kronecker_delta(v0, j) +
+                           (N-k-i) * _kronecker_delta(v1, j)
+                           for j in range(4) ]
+
+                    # Note that the sgn is different from the paper
+                    # because we are using SnapPy conventions for
+                    # cross ratios here
+
+                    prod2 = matrix.matrix_mult(
+                        prod2,
+                        _H(N, i, self.x_coordinate(tet, pt) ** -sgn))
+
+                m = matrix.matrix_mult(m,
+                                       matrix.matrix_mult(prod1, prod2))
+
+            # Matrix from Equation 10.1
+            dpm = [ [ - (-1) ** (N - i) * _kronecker_delta(i, j)
+                        for i in range(N) ]
+                    for j in range(N) ]
+
+            m = matrix.matrix_mult(m, dpm)
+
+            # Set in cache
+            self._edge_cache[key] = m
+
+        return self._edge_cache[key]
+
+    def short_edge(self, tet, v0, v1, v2):
+        """
+        The matrix that labels a long edge starting at vertex (v0, v1, v2)
+        of a doubly truncated simplex corresponding to the ideal tetrahedron
+        with index tet.
+
+        This matrix was labeled gamma^{v0v1v2} in Figure 18 of
+        Garoufalidis, Goerner, Zickert:
+        Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1207.6711
+
+        It is computed using equation 10.22.
+
+        The resulting matrix is given as a python list of lists.
+        """
+
+        # Key for the cache
+        key = 'short_%d_%d%d%d' % (tet, v0, v1, v2)
+
+        # Fill cache if necessary
+        if key not in self._edge_cache:
+
+            edge = [ _kronecker_delta(v0, i) +
+                     _kronecker_delta(v1, i) for i in range(4) ]
+
+            # The epsilon permutation sign
+            sgn = CrossRatios._cyclic_three_perm_sign(v0, v1, v2)
+
+            # Get N
+            N = self.N()
+
+            # Start with identity
+            m = self._get_identity_matrix()
+
+            # Compute the product in equation 10.22
+            for a0 in range(N-1):
+                a1 = N - 2 - a0
+                pt = [ a0 * _kronecker_delta(v0, i) +
+                       a1 * _kronecker_delta(v1, i) for i in range(4) ]
+
+                cross_ratio = self._shape_at_tet_point_and_edge(tet, pt, edge)
+
+                # Multiply result with the H matrix
+
+                # Note that the sgn is different from the paper
+                # because we are using SnapPy conventions for
+                # cross ratios here
+
+                m = matrix.matrix_mult(m, _H(N, a0 + 1, cross_ratio ** sgn))
+
+            # Fill cache
+            self._edge_cache[key] = m
+
+        return self._edge_cache[key]
+
+    def _init_matrix_and_inverse_cache(self):
+        # Fill the caches of matrices corresponding to the
+        # fundamental group generators and their inverses
+
+        if self._matrix_cache and self._inverse_matrix_cache:
+            return
+
+        # Compute all the matrices for the generators and there inverses
+        # The long edges of the doubly truncated simplex are all unit
+        # counter-diagonal so they do not increase the
+        # size of any polynomial coefficients. We thus don't give them penalty.
+        self._matrix_cache, self._inverse_matrix_cache = (
+            findLoops.images_of_original_generators(self,
+                                                    penalties=(0, 1, 1)))
+
+    def evaluate_word(self, word, G=None):
+        """
+        Given a word in the generators of the fundamental group,
+        compute the corresponding matrix. By default, these are the
+        generators of the unsimplified presentation of the fundamental
+        group. An optional SnapPy fundamental group can be given if the
+        words are in generators of a different presentation, e.g.,
+        c.evaluate_word(word, M.fundamental_group(True)) to
+        evaluate a word in the simplified presentation returned by
+        M.fundamental_group(True).
+
+        For now, the matrix is returned as list of lists.
+        """
+
+        # Init the matrices corresponding to generators
+        self._init_matrix_and_inverse_cache()
+
+        return findLoops.evaluate_word(
+            self._get_identity_matrix(),
+            self._matrix_cache,
+            self._inverse_matrix_cache,
+            word,
+            G)
+
+    def check_against_manifold(self, M=None, epsilon=None):
+        """
+        Checks that the given solution really is a solution to the PGL(N,C) gluing
+        equations of a manifold. Usage similar to check_against_manifold of
+        PtolemyCoordinates. See help(ptolemy.PtolemtyCoordinates) for example.
+
+        === Arguments ===
+
+        M --- manifold to check this for
+        epsilon --- maximal allowed error when checking the relations, use
+        None for exact comparison.
+        """
+        if M is None:
+            M = self.get_manifold()
+
+        if M is None:
+            raise Exception("Need to give manifold")
+
+        some_z = list(self.keys())[0]
+        variable_name, index, tet_index = some_z.split('_')
+        if variable_name not in ['z', 'zp', 'zpp']:
+            raise Exception("Variable not z, zp, or, zpp")
+        if len(index) != 4:
+            raise Exception("Not 4 indices")
+        N = sum([int(x) for x in index]) + 2
+
+        matrix_with_explanations = M.gluing_equations_pgl(
+            N, equation_type='all')
+
+        matrix = matrix_with_explanations.matrix
+        rows = matrix_with_explanations.explain_rows
+        cols = matrix_with_explanations.explain_columns
+
+        for row in range(len(rows)):
+            product = 1
+            for col in range(len(cols)):
+                cross_ratio_variable = cols[col]
+                cross_ratio_value = self[cross_ratio_variable]
+                product = product * (cross_ratio_value ** matrix[row,col])
+            _check_relation(
+                product - 1,
+                epsilon,
+                "Gluing equation %s" % rows[row])
+
+    def induced_representation(self, N):
+        """
+        Given a PSL(2,C) representation constructs the induced representation
+        for the given N.
+        The induced representation is in SL(N,C) if N is odd and
+        SL(N,C) / {+1,-1} if N is even and is described in the Introduction of
+        Garoufalidis, Thurston, Zickert
+        The Complex Volume of SL(n,C)-Representations of 3-Manifolds
+        https://arxiv.org/abs/1111.2828
+
+        There is a canonical group homomorphism SL(2,C)->SL(N,C) coming from
+        the the natural SL(2,C)-action on the vector space Sym^{N-1}(C^2).
+        This homomorphisms decends to a homomorphism from PSL(2,C) if one
+        divides the right side by {+1,-1} when N is even.
+        Composing a representation with this homomorphism gives the induced
+        representation.
+        """
+
+        num_tetrahedra = self.num_tetrahedra()
+
+        if self.N() != 2:
+            raise Exception(
+                "Cross ratios need to come from a PSL(2,C) representation")
+
+        def key_value_pair(v, t, index):
+            new_key = v + '_%d%d%d%d' % tuple(index) + '_%d' % t
+            old_key = v + '_0000' + '_%d' % t
+            return (new_key, self[old_key])
+
+        d = dict(
+            [ key_value_pair(v, t, index)
+              for v in ['z', 'zp', 'zpp']
+              for t in range(num_tetrahedra)
+              for index in utilities.quadruples_with_fixed_sum_iterator(N-2)])
+
+        return CrossRatios(d,
+                           is_numerical=self._is_numerical,
+                           manifold_thunk=self._manifold_thunk)
+
+    def is_real(self, epsilon):
+        """
+        Returns True if all cross ratios are real (have absolute imaginary
+        part < epsilon where epsilon is given as argument).
+        This means that the corresponding representation is in PSL(N,R).
+        """
+
+        if not self._is_numerical:
+            raise NumericalMethodError("is_real")
+
+        for v in self.values():
+            if v.imag().abs() > epsilon:
+                return False
+        return True
+
+    def is_induced_from_psl2(self, epsilon=None):
+        """
+        For each simplex and each edges, checks that all cross ratios of that
+        simplex that are parallel to that each are the same (maximal absolute
+        difference is the epsilon given as argument).
+        This means that the corresponding representation is induced by a
+        PSL(2,C) representation.
+        """
+
+        # Create an auxiliary dictionary containing one z, zp, zpp per tet
+        d = { }
+
+        for key, value in self.items():
+            variable_name, index, tet_index = key.split('_')
+            if variable_name not in ['z', 'zp', 'zpp']:
+                raise Exception("Variable not z, zp, or, zpp")
+            if len(index) != 4:
+                raise Exception("Not 4 indices")
+
+            # The key in the auxiliary dictionary
+            short_key = variable_name + '_' + tet_index
+
+            # Get the old value in the auxiliary dictionary
+            old_value = d.setdefault(short_key, value)
+
+            if epsilon is None:
+                if value != old_value:
+                    return False
+            else:
+                if (value - old_value).abs() > epsilon:
+                    return False
+
+        return True
+
+    def is_pu_2_1_representation(self, epsilon, epsilon2=None):
+        r"""
+        Returns True if the representation is also a
+        PU(2,1)-representation. This uses Proposition 3.5 and the
+        remark following that proposition in [FKR2013]_.
+
+        If a condition given in that Proposition is violated, the method returns
+        an object whose Boolean value is still False and that indicates which condition
+        was violated. Thus, this method can still be used in ``if`` statements.
+
+        The method tests the following complex equalities and inequalities:
+
+        * the three complex equations given in (3.5.1) of [FKR2013]_.
+        * the inequality z\ :sub:`ijl` :math:`\\not=` -1.
+
+        **Remark:** It does not check whether all z\ :sub:`ij` * z\ :sub:`ji` are real or
+        not as these are still valid CR configurations, see the remark following
+        Proposition 3.5.
+
+        The user has to supply an epsilon: an equality/inequality is considered
+        to be true if and only if the absolute value | LHS - RHS | of difference between the
+        left and right hand side is less/greater than epsilon.
+
+        The user can supply another parameter, epsilon2. If any | LHS - RHS | is in
+        the interval [epsilon, epsilon2], this method fails with an exception
+        as the value of | LHS - RHS | is an ambiguous interval where
+        it is unclear whether inequality fails to hold because it truly does
+        hold or just because of numerical noise.
+        """
+
+        def is_zero(val):
+            if val.abs() < epsilon:
+                return True
+            if epsilon2:
+                if not epsilon2 < val.abs():
+                    raise Exception(
+                        "Ambiguous error when determining whether a "
+                        "condition was fulfilled or nor: %s" % val)
+            return False
+
+        def mainCondition(key_zij, key_zji, key_zkl, key_zlk):
+
+            lhs = (self[key_zij] * self[key_zji])
+            rhs = (self[key_zkl] * self[key_zlk]).conj()
+
+            if not is_zero(lhs - rhs):
+                reason = "%s * %s = conjugate(%s * %s) not fulfilled" % (
+                    key_zij, key_zji, key_zkl, key_zlk)
+                return NotPU21Representation(reason)
+
+            return True
+
+        def tripleRatioCondition(key_zji, key_zki, key_zli):
+
+            tripleRatio = self[key_zji] * self[key_zki] * self[key_zli]
+
+            if is_zero(tripleRatio - 1):
+                reason = 'Triple ratio %s * %s * %s = 1' % (
+                    key_zji, key_zki, key_zli)
+                return NotPU21Representation(reason)
+
+            return True
+
+        if self.N() != 3:
+            raise Exception("PU(2,1)-representations only allowed for N = 3")
+
+        if not self._is_numerical:
+            raise NumericalMethodError("is_pu_2_1_representation")
+
+        for t in range(self.num_tetrahedra()):
+
+            m0 = mainCondition("z_1000_%d" % t, "z_0100_%d" % t,
+                               "z_0010_%d" % t, "z_0001_%d" % t)
+            if not m0:
+                return m0
+
+            m1 = mainCondition("zp_1000_%d" % t, "zp_0010_%d" % t,
+                               "zp_0100_%d" % t, "zp_0001_%d" % t)
+            if not m1:
+                return m1
+
+            m2 = mainCondition("zpp_1000_%d" % t, "zpp_0001_%d" % t,
+                               "zpp_0100_%d" % t, "zpp_0010_%d" % t)
+            if not m2:
+                return m2
+
+            t0 = tripleRatioCondition(  "z_0100_%d" % t,
+                                       "zp_0010_%d" % t,
+                                      "zpp_0001_%d" % t)
+            if not t0:
+                return t0
+
+            t1 = tripleRatioCondition(  "z_1000_%d" % t,
+                                       "zp_0001_%d" % t,
+                                      "zpp_0010_%d" % t)
+            if not t1:
+                return t1
+
+            t2 = tripleRatioCondition(  "z_0001_%d" % t,
+                                       "zp_1000_%d" % t,
+                                      "zpp_0100_%d" % t)
+            if not t2:
+                return t2
+
+            t3 = tripleRatioCondition(  "z_0010_%d" % t,
+                                       "zp_0100_%d" % t,
+                                      "zpp_1000_%d" % t)
+            if not t3:
+                return t3
+
+        return True
+
+    def is_geometric(self, epsilon=1e-6):
+        """
+        Returns true if all shapes corresponding to this solution have positive
+        imaginary part.
+
+        If the solutions are exact, it returns true if one of the corresponding
+        numerical solutions is geometric.
+
+        An optional epsilon can be given. An imaginary part of a shape is
+        considered positive if it is larger than this epsilon.
+        """
+
+        if self._is_numerical:
+            for v in self.values():
+                if not v.imag() > 0:
+                    return False
+            return True
+        else:
+            for numerical_sol in self.numerical():
+                if numerical_sol.is_geometric(epsilon):
+                    return True
+            return False
+
+
+def _ptolemy_to_cross_ratio(solution_dict,
+                            branch_factor=1,
+                            non_trivial_generalized_obstruction_class=False,
+                            as_flattenings=False):
+
+    N, has_obstruction = _N_and_has_obstruction_for_ptolemys(solution_dict)
+    num_tets = _num_tetrahedra(solution_dict)
+
+    if N % 2:
+        evenN = 2 * N
+    else:
+        evenN = N
+
+    if not non_trivial_generalized_obstruction_class:
+        evenN = 2
+
+    if as_flattenings:
+        f = pari('2 * Pi * I') / evenN
+
+    def compute_cross_ratios_and_flattenings(tet, index):
+        def get_ptolemy_coordinate(addl_index):
+            total_index = matrix.vector_add(index, addl_index)
+            key = "c_%d%d%d%d" % tuple(total_index) + "_%d" % tet
+            return solution_dict[key]
+
+        def get_obstruction_variable(face):
+            key = "s_%d_%d" % (face, tet)
+            return solution_dict[key]
+
+        c1010 = get_ptolemy_coordinate((1,0,1,0))
+        c1001 = get_ptolemy_coordinate((1,0,0,1))
+        c0110 = get_ptolemy_coordinate((0,1,1,0))
+        c0101 = get_ptolemy_coordinate((0,1,0,1))
+        c1100 = get_ptolemy_coordinate((1,1,0,0))
+        c0011 = get_ptolemy_coordinate((0,0,1,1))
+
+        z = (c1010 * c0101) / (c1001 * c0110)
+        zp = - (c1001 * c0110) / (c1100 * c0011)
+        zpp = (c1100 * c0011) / (c1010 * c0101)
+
+        if has_obstruction:
+            s0 = get_obstruction_variable(0)
+            s1 = get_obstruction_variable(1)
+            s2 = get_obstruction_variable(2)
+            s3 = get_obstruction_variable(3)
+            z = s0 * s1 * z
+            zp = s0 * s2 * zp
+            zpp = s0 * s3 * zpp
+
+        variable_end = '_%d%d%d%d' % tuple(index) + '_%d' % tet
+
+        if as_flattenings:
+            def make_triple(w, z):
+                z = _convert_to_pari_float(z)
+                return (w, z, ((w - z .log()) / f).round())
+
+            w = _compute_flattening(c1010, c0101, c1001, c0110,
+                                    branch_factor, evenN)
+            wp = _compute_flattening(c1001, c0110, c1100, c0011,
+                                    branch_factor, evenN)
+            wpp = _compute_flattening(c1100, c0011, c1010, c0101,
+                                    branch_factor, evenN)
+
+            return [
+                ('z' + variable_end, make_triple(w  ,z  )),
+                ('zp' + variable_end, make_triple(wp ,zp )),
+                ('zpp' + variable_end, make_triple(wpp,zpp)) ]
+
+        else:
+            return [
+                ('z' + variable_end, z),
+                ('zp' + variable_end, zp),
+                ('zpp' + variable_end, zpp) ]
+
+    return dict(
+        sum([compute_cross_ratios_and_flattenings(tet,index)
+             for tet in range(num_tets)
+             for index in utilities.quadruples_with_fixed_sum_iterator(N - 2)],
+            [])), evenN
+
+
+def _num_tetrahedra(solution_dict):
+    return max( [ int(key.split('_')[-1])
+                  for key in solution_dict.keys() ] ) + 1
+
+
+def _N_for_shapes(solution_dict):
+
+    def get_N(key):
+        m = re.match(r'zp{0,2}_(\d{4})_\d+$', key)
+        if not m:
+            raise Exception("Not a valid shape key: '%s'" % key)
+        return sum([int(char) for char in m.group(1)]) + 2
+
+    l = [ get_N(key) for key in solution_dict.keys() ]
+    if not len(set(l)) == 1:
+        raise Exception("Shape keys for different N")
+
+    return l[0]
+
+
+def _N_and_has_obstruction_for_ptolemys(solution_dict):
+
+    def get_N(key):
+        m = re.match(r'c_(\d{4})_\d+$', key)
+        if not m:
+            raise Exception("Not a valid Ptolemy key: '%s'" % key)
+        return sum([int(char) for char in m.group(1)])
+
+    has_obstruction = False
+
+    l = set()
+    for key in solution_dict.keys():
+        if re.match(r's_\d_\d+$', key):
+            has_obstruction = True
+        else:
+            l.add(get_N(key))
+
+    if not len(l) == 1:
+        raise Exception("Ptolemy keys for different N")
+
+    for N in l:
+        return N, has_obstruction
+
+
+def _get_number_field(d):
+    for value in d.values():
+
+        if isinstance(value, RUR):
+            nf = value.number_field()
+            if nf:
+                return nf
+
+        if type(value) == Gen and value.type() == 't_POLMOD':
+            return value.mod()
+
+    return None
+
+
+def _evaluate_at_root(p, root):
+
+    if type(p) == Gen and p.type() == 't_POLMOD':
+        return p.lift().substpol('x', root)
+
+    if isinstance(p, RUR):
+        return p.evaluate_at_root(root)
+
+    return p
+
+
+def _to_numerical(d):
+
+    number_field = _get_number_field(d)
+
+    if number_field is None:
+        roots = [ pari(0) ]
+    else:
+        # Bug in cypari: pari(str(number_field)).polroots()
+        # gives less precision
+        roots = pari('polroots(%s)' % number_field)
+
+    def evaluate_all_for_root(root):
+
+        def evaluate_key_for_root(key, value):
+
+            v = _evaluate_at_root(value, root)
+
+            if key[:2] == 'z_':
+                z = v
+                zp = 1 / (1 - z)
+                zpp = 1 - 1 / z
+
+                return [(key, z),
+                        ('zp_' + key[2:], zp),
+                        ('zpp_' + key[2:], zpp)]
+            elif key[:3] == 'zp_' or key[:4] == 'zpp_':
+                return []
+            else:
+                return [(key, v)]
+
+        return dict(sum(
+                [ evaluate_key_for_root(key, value)
+                  for key, value in d.items() ], []))
+
+    return [ evaluate_all_for_root(root) for root in roots ]
+
+
+def _apply_to_RURs(d, RUR_method):
+
+    def _apply_to_RUR(v):
+        if isinstance(v, RUR):
+            return RUR_method(v)
+        return v
+
+    return {k: _apply_to_RUR(v) for k, v in d.items()}
+
+
+def _convert_to_pari_float(z):
+
+    if type(z) == Gen and z.type() in ['t_INT', 't_FRAC']:
+        return z * pari('1.0')
+
+    return pari(z)
+
+
+def _compute_flattening(a, b, c, d, branch_factor, N=2):
+
+    PiMinusEpsilon = pari(3.141592)
+
+    def safe_log(z):
+
+        l = (branch_factor * z**N).log()
+
+        if l.imag().abs() > PiMinusEpsilon:
+            raise LogToCloseToBranchCutError()
+
+        return l / N
+
+    a = _convert_to_pari_float(a)
+    b = _convert_to_pari_float(b)
+    c = _convert_to_pari_float(c)
+    d = _convert_to_pari_float(d)
+
+    w = safe_log(a) + safe_log(b) - safe_log(c) - safe_log(d)
+
+    return w
+
+# bug in pari
+
+
+def _dilog(z):
+    return pari("dilog(%s)" % z)
+
+
+def _L_function(zpq_triple, evenN=2):
+
+    z, p, q = zpq_triple
+
+    z = _convert_to_pari_float(z)
+    p = _convert_to_pari_float(p)
+    q = _convert_to_pari_float(q)
+
+    f = pari('2 * Pi * I') / evenN
+    Pi2 = pari('Pi * Pi')
+
+    return (  _dilog(z)
+            + (z.log() + p * f) * ((1-z).log() + q * f) / 2
+            - Pi2 / 6)
+
+
+def _volume(z):
+
+    z = _convert_to_pari_float(z)
+
+    return (1-z).arg() * z.abs().log() + _dilog(z).imag()
+
+
+def _kronecker_delta(i, j):
+    """
+    Kronecker Delta, returns 1 if and only if i and j are equal, other 0.
+    """
+
+    if i == j:
+        return 1
+    else:
+        return 0
+
+
+def _X(N, k, v):
+    """
+    Returns the NxN matrix with off-diagonal entry v at position k, that
+    is the entry at row k and column k+1 is v.
+
+    See (10.2) of
+    Garoufalidis, Goerner, Zickert:
+    Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+    https://arxiv.org/abs/1207.6711
+    """
+
+    m = [[_kronecker_delta(i,j) for i in range(N)] for j in range(N)]
+    m[k-1][k] = v
+    return m
+
+
+def _H(N, k, x):
+    """
+    Returns the NxN diagonal matrix where the first k diagonal entries are x
+    and all other entries are 1.
+
+    See (10.1) of
+    Garoufalidis, Goerner, Zickert:
+    Gluing Equations for PGL(n,C)-Representations of 3-Manifolds
+    https://arxiv.org/abs/1207.6711
+    """
+
+    def _entry(i, j):
+        if i != j:
+            return 0
+        if i < k:
+            return x
+        return 1
+
+    return [[_entry(i,j) for i in range(N)] for j in range(N)]