lalsuite 7.26.2.dev20251207__cp313-cp313-manylinux_2_28_x86_64.whl

lalsimulation/tilts_at_infinity/calc_tilts_prec_avg_regularized.py

@@ -0,0 +1,1424 @@
+"""
+Implement the computation of the tilt angles at infinity or bounds and average values of tilts at a finite separation
+from tilt angles at some frequency, taken to be sufficiently small so that precession-averaged evolution is accurate.
+This implementation is described in the paper, <https://dcc.ligo.org/P2100029>, arXiv:2107.11902.
+
+N. K. Johnson-McDaniel, 2021
+"""
+
+import numpy as np
+
+from numpy.linalg import norm
+
+from numpy.polynomial import Polynomial as P
+
+from scipy.special import ellipe, ellipk
+
+from scipy.integrate import ode
+
+from .tilts_at_infinity_utils import *
+
+from warnings import warn
+
+try:
+    from scipy.integrate import solve_ivp
+except ImportError:
+    solve_ivp_avail = False
+else:
+    solve_ivp_avail = True
+
+try:
+    import mpmath as mp
+except ImportError:
+    mpmath_avail = False
+else:
+    mpmath_avail = True
+
+# Set available PN orders for the orbital angular momentum expression in terms of frequency
+
+LPN_orders = [0, 1, 1.5, 2, 2.5]
+
+LPN_order_string = ', '.join(map(str, LPN_orders))
+
+# Define functions
+
+
+def L_vec_from_masses_spins_and_f(m1, m2, f, PNorder=2, S1_vec=np.zeros(3), S2_vec=np.zeros(3),
+                                  no_mass_spin_input_check=False):
+    """
+    Compute the PN orbital angular momentum from the binary's masses and spins and m = 2 GW frequency
+    using the expression in Eq. (4) of <https://dcc.ligo.org/public/0122/T1500554/023/dLdS.pdf>.
+
+    We orbit-average the contributions from the spins as in Eq. (17) of that reference, but the spins are
+    set to zero by default.
+
+    Inputs:
+
+    - Required:
+
+    m1, m2: Detector frame masses in kg
+    f: m = 2 GW frequency in Hz
+
+    - Optional:
+
+    PNorder: Order of PN corrections to include--choices are 0, 1, 1.5, 2, or 2.5, where the half-integer orders are
+             the spin-orbit contributions, default: 2
+    S1_vec, S2_vec: 3d numpy arrays giving the components of the dimensionful spins (with total mass = 1) in a coordinate
+                    system where the Newtonian orbital angular momentum is in the z-direction, default: [0., 0., 0.]
+    no_mass_spin_input_check: Turn off mass and spin input checking for when this is called by other functions that have
+                              already performed this check, default: False
+
+    Output: 3d array giving the PN orbital angular momentum in total mass = 1 units in a coordinate system where the
+            Newtonian orbital angular momentum is in the z-direction
+    """
+
+    # Check inputs
+
+    if not no_mass_spin_input_check:
+        check_masses(m1, m2)
+
+    if f <= 0.:
+        raise ValueError(
+            "The frequency must be positive, while it is f = %f" % f)
+
+    if PNorder not in LPN_orders:
+        raise ValueError("The PN order to use to compute the orbital angular momentum must be one of %s, while it is "
+                         "PNorder = %f" % (LPN_order_string, PNorder))
+
+    M = m1 + m2
+    Msq = M*M
+
+    S1 = norm(S1_vec)
+    S2 = norm(S2_vec)
+
+    if not no_mass_spin_input_check and (S1 > m1*m1/Msq or S2 > m2*m2/Msq):
+        raise ValueError(
+            "The magnitudes of the spins must be at most (m1/M)^2 = %f and (m2/M)^2 = %f, respectively, "
+            "where M is the total mass, while they are |S1|, |S2| = %f, %f" %
+            (m1*m1/Msq, m2*m2/Msq, S1, S2))
+
+    # Compute useful quantities
+
+    X1 = m1/M
+    X2 = m2/M
+
+    X1sq = X1*X1
+    X2sq = X2*X2
+
+    eta = X1*X2
+
+    zhat = np.array([0., 0., 1.])
+
+    S1z_vec = S1_vec[2]*zhat
+    S2z_vec = S2_vec[2]*zhat
+
+    v = (np.pi*M*kg_to_s*f)**(1./3.)
+
+    v2 = v*v
+    v3 = v2*v
+
+    # Put together L to the desired PN order, starting by giving the magnitude and direction of the Newtonian expression
+
+    LN = eta/v
+
+    L = 1.*zhat
+
+    if PNorder >= 1:
+        L += (1.5+eta/6.)*v2*zhat
+    if PNorder >= 1.5:
+        L += (-0.25*((1. + 3./X1)*S1_vec + (1. + 3./X2)*S2_vec)
+              - ((1. + 27./X1)*S1z_vec + (1. + 27./X2)*S2z_vec)/12.
+              )*v3
+    if PNorder >= 2:
+        L += (3.375 - 2.375*eta + eta*eta/24.)*v2*v2*zhat
+    if PNorder >= 2.5:
+        L += ((-71. + 21./X1 - 13.*X1 - 9.*X1sq)*S1_vec + (-71. + 21./X2 - 13.*X2 - 9.*X2sq)*S2_vec
+              + (1. - 147./X1 - 367.*X1/3. + 13.*X1sq/3.)*S1z_vec
+              + (1. - 147./X2 - 367.*X2/3. + 13.*X2sq/3.)*S2z_vec
+              )*v3*v2/48.
+
+    return LN*L
+
+
+def L_from_masses_spins_and_f(m1, m2, f, PNorder=2, S1_vec=np.zeros(3), S2_vec=np.zeros(3), no_mass_spin_input_check=False):
+    """
+    Compute the magnitude of the PN orbital angular momentum from the binary's masses and spins and m = 2 GW frequency
+    using the expression in Eq. (4) of <https://dcc.ligo.org/public/0122/T1500554/023/dLdS.pdf>.
+
+    We orbit-average the contributions from the spins as in Eq. (17) of that reference, but the spins are
+    set to zero by default.
+
+    This wraps L_vec_from_masses_spins_and_f()
+
+    Inputs:
+
+    - Required:
+
+    m1, m2: Detector frame masses in kg
+    f: m = 2 GW frequency in Hz
+
+    - Optional:
+
+    PNorder: Order of PN corrections to include--choices are 0, 1, 1.5, 2, or 2.5, where the half-integer orders are
+             the spin-orbit contributions, default: 2
+    S1_vec, S2_vec: 3d numpy arrays giving the components of the dimensionful spins (with total mass = 1) in a coordinate
+                    system where the Newtonian orbital angular momentum is in the z-direction, default: [0., 0., 0.]
+    no_mass_spin_input_check: Turn off mass and spin input checking for when this is called by other functions that have
+                              already performed this check, default: False
+
+    Output: Magnitude of the PN orbital angular momentum in total mass = 1 units
+    """
+
+    return norm(L_vec_from_masses_spins_and_f(m1, m2, f, PNorder=PNorder, S1_vec=S1_vec, S2_vec=S2_vec,
+                                              no_mass_spin_input_check=no_mass_spin_input_check))
+
+
+def L_from_masses_a_and_ecc(m1, m2, a, e):
+    """
+    Compute the magnitude of the Newtonian orbital angular momentum from the binary's masses, semimajor axis, and
+    eccentricity
+
+    Inputs:
+
+    m1, m2: Source frame masses in kg
+    a: Semimajor axis in m
+    e: eccentricity
+
+    Output: Magnitude of orbital angular momentum in total mass = 1 units
+    """
+
+    # Check inputs
+
+    check_masses(m1, m2)
+
+    if a <= 0.:
+        raise ValueError(
+            "The semimajor axis must be positive, while it is a = %e" % a)
+
+    if e < 0. or e > 1.:
+        raise ValueError(
+            "The eccentricity must be between 0 and 1, while it is e = %f" % e)
+
+    # Compute
+
+    M = m1 + m2
+
+    eta = m1*m2/M/M
+
+    return eta*(a*(1. - e*e)/(M*kg_to_m))**0.5
+
+
+def E_ratio_term(m):
+    """
+    Compute m*E_ratio(m) [cf. Eq. (B13) in the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902, which gives m^2*E_ratio(m)]
+    """
+
+    omm = 1. - m
+    fpm = 4. + m
+
+    return (3.*(1. + 3.*(4. - m)/(omm*fpm))/(16.*omm**1.5) - 0.5)*m/fpm
+
+
+def eq_coeffs(u, kappaxiq, q, S1, S2, xi, S0sq):
+    """
+    Compute coefficients for the cubic equation in bar{S}^2 to be solved [Eq. (20) in the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902]
+
+    Inputs:
+    u: 1/(2L), where L is the orbital angular momentum (i.e., the integration variable)
+    kappaxiq: The rescaled variable defined in Eq. (14) of the paper
+    q: Mass ratio of the binary, with the < 1 convention
+    S1, S2: Magnitudes of the dimensionful spins of the binary (with total mass = 1)
+    xi: Conserved effective spin, defined in Eq. (2) of the paper
+    S0sq: Square of magnitude of the initial total spin (at u = u0, and with total mass = 1)
+
+    Output: The coefficients of the equation (vareps, Btransf, Ctransf, Dtransf)
+    """
+
+    # Define useful quantities
+
+    q2 = q*q
+
+    eps = 1. - q
+
+    opq = 1. + q
+
+    u2 = u*u
+
+    S1sq = S1*S1
+    S2sq = S2*S2
+
+    kappaxiq2 = kappaxiq*kappaxiq
+
+    # Compute the coefficients of the transformed equation q (1 - q^2) u^2 bar{S}^6 + bar{B} bar{S}^4 +
+    # bar{C} bar{S}^2 + bar{D} = 0 [Eqs. (20-22) of the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902].
+
+    # Here we refer to the barred coefficients with the suffix "transf" and the coefficient of bar{S}^6 as
+    # vareps, as in Appendix C of the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902
+
+    Sigma = 0.5*(S0sq - S1sq - S2sq)
+    Upsilon = opq*(2.*q*Sigma + q2*S1sq + S2sq)
+    zeta = q*(Sigma + q*S1sq)*xi
+
+    vareps = q*opq*eps*u2
+
+    Btransf = 0.25*opq*eps*eps + q*eps * \
+        (xi - 2.*opq*kappaxiq)*u + Upsilon*u2
+
+    Ctransf = eps*(opq*(Sigma + q*kappaxiq2) - q*kappaxiq *
+                   xi) + 2.*(zeta - Upsilon*kappaxiq)*u
+
+    Dtransf = opq*(Sigma*Sigma - S1sq*S2sq) + q2*S1sq*xi * \
+        xi/opq - 2.*zeta*kappaxiq + Upsilon*kappaxiq2
+
+    return vareps, Btransf, Ctransf, Dtransf
+
+
+def solve_cubic(coeffs, u, kappaxiq, q, S1, S2, imag_tol, use_mpmath=False, root_tol=1.e-6, polyroots_maxsteps=50, polyroots_extraprec=50):
+    """
+    Solve the cubic equation, check for complex roots, and order the roots.
+    The inputs besides the coefficients are just needed in order to be printed
+    if one of the checks is triggered
+
+    Inputs:
+
+    - Required:
+
+    coeffs: Array of coefficients of cubic [vareps, Btransf, Ctransf, Dtransf]
+    u: 1/(2L), where L is the orbital angular momentum (i.e., the integration variable)
+    kappaxiq: The rescaled variable defined in Eq. (14) of the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902
+    q: Mass ratio of the binary, with the < 1 convention
+    S1, S2: Magnitudes of the dimensionful spins of the binary (with total mass = 1)
+    imag_tol: Tolerance for the imaginary part of roots of the cubic
+
+    - Optional:
+
+    use_mpmath: If True, use mpmath.polyroots to solve the cubic instead of numpy's Polynomial roots method
+                (slower but more accurate), default: False
+    root_tol: Tolerance for the error estimate from mpmath.polyroots, if use_mpmath == True, default: 1.e-6
+    polyroots_maxsteps: The maximum number of steps allowed in mpmath.polyroots if use_mpmath == True, default: 50
+    polyroots_extraprec: The number of extra digits to use in mpmath.polyroots if use_mpmath == True, default: 50
+
+    Output: The roots of the equation ordered from smallest to largest (Ssq3, Ssqm, Ssqp)
+    """
+
+    # Extract coefficients and solve
+
+    vareps, Btransf, Ctransf, Dtransf = coeffs
+
+    if use_mpmath:
+        Btransf_out, Ctransf_out, Dtransf_out = Btransf, Ctransf, Dtransf
+
+        roots, err = mp.polyroots([vareps, Btransf, Ctransf, Dtransf], error=True, maxsteps=polyroots_maxsteps, extraprec=polyroots_extraprec)
+
+        SsqI, SsqII, SsqIII = roots
+
+        # Check that the error on the roots isn't too large
+        if err > root_tol:
+            warn("polyroots error of %e greater than root_tol = %e"%(err, root_tol), RuntimeWarning)
+    else:
+        # The [0] indices here and elsewhere are necessary to avoid a warning
+        Btransf_out, Ctransf_out, Dtransf_out = Btransf[0], Ctransf[0], Dtransf[0]
+
+        cubic = P([Dtransf_out, Ctransf_out, Btransf_out, vareps])
+
+        SsqI, SsqII, SsqIII = cubic.roots()
+
+    # Check that all roots are real
+    if max(abs(SsqI.imag), abs(SsqII.imag), abs(SsqIII.imag)) > imag_tol:
+        raise RuntimeError("At least one of the roots of the cubic in S^2 has an imaginary part larger than the "
+                           "tolerance of %.2g set in the code.\nThe roots are (%.10f, %.10f), (%.10f, %.10f), "
+                           "(%.10f, %.10f), and the coefficients are %.10e, %.10e, %.10e, %.10e at u = %.10f, "
+                           "kappaxiq = %.10f. q = %f, S1, S2 = %f, %f" %
+                           (imag_tol, SsqI.real, SsqI.imag, SsqII.real, SsqII.imag, SsqIII.real, SsqIII.imag,
+                            vareps, Btransf_out, Ctransf_out, Dtransf_out, u, kappaxiq, q, S1, S2))
+
+    # Order the roots
+
+    Ssq3, Ssqm, Ssqp = np.sort([SsqI.real, SsqII.real, SsqIII.real])
+
+    # Check for complex conjugates that passed the previous check, but allow for roots that are real and equal,
+    # which occurs to numerical accuracy for close-to-aligned-spin cases
+    if Ssq3.real == Ssqm.real:
+        if Ssq3.imag != 0. or Ssqm.imag != 0.:
+            warn(
+                r"\bar{S}_3^2 and \bar{S}_-^2 have the same real parts, with values of: %e, %e"%(Ssq3, Ssqm), RuntimeWarning)
+        elif Ssq3 == 0. and Ssqm == 0.:
+            raise NonprecessingError(r"The system is nonprecessing: \bar{S}^2_3 = \bar(S}^2_- = 0") # This system is nonprecessing to numerical accuracy
+
+    if Ssqm.real == Ssqp.real:
+        if Ssqm.imag != 0. or Ssqp.imag != 0.:
+            warn(
+                r"\bar{S}_-^2 and \bar{S}_+^2 have the same real parts, with values of: %e, %e"%(Ssqm, Ssqp), RuntimeWarning)
+        elif Ssqm == 0. and Ssqp == 0.:
+            raise NonprecessingError(r"The system is nonprecessing: \bar{S}^2_- = \bar(S}^2_+ = 0") # This system is nonprecessing to numerical accuracy
+
+    return Ssq3, Ssqm, Ssqp
+
+
+def dkappaxiq_du(u, kappaxiq, q, S1, S2, xi, S0sq, imag_tol, root_tol, lin_tol, root_tol_raise_err, use_mpmath=False,
+                 polyroots_maxsteps=50, polyroots_extraprec=50):
+    """
+    Compute (<S^2>_prec - S0^2)/(1 - q) [= dkappa_{xi q}/du] using Eq. (42) in Chatziioannou, Klein, Yunes, and Cornish PRD 95, 104004 (2017)
+    [rewritten as bar{S}_+^2 + [E(m)/K(m) - 1](bar{S}_+^2 - bar{S}_3^2), where bar{S}_*^2 := (S_*^2 - S_0^2)/(1 - q)].
+    Here kappa_{xi q} := (kappa - S_0^2/(2 L) - q xi/(1 + q))/(1 - q) - xi/4, and S_0 is the magnitude of the
+    initial total spin (i.e., at u = u0).
+
+    Inputs:
+
+    - Required:
+
+    u: 1/(2L), where L is the orbital angular momentum (i.e., the integration variable)
+    kappaxiq: The rescaled variable defined above and in Eq. (14) of the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902
+    q: Mass ratio of the binary, with the < 1 convention
+    S1, S2: Magnitudes of the dimensionful spins of the binary (with total mass = 1)
+    xi: Conserved effective spin, defined in Eq. (2) of the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902
+    S0sq: Square of magnitude of the initial total spin (at u = u0, and with total mass = 1)
+    imag_tol: Tolerance for the imaginary part of roots of the cubic
+    root_tol: Tolerance for the error in the roots of the cubic (is an absolute tolerance for small values of the roots, as
+              generally occurs during the beginning of the evolution, and is a fractional tolerance when the roots are large,
+              which generally occurs towards the end of the evolution; also is reduced automatically when the roots are close
+              together)
+    lin_tol: Tolerance for errors on the r.h.s. of dkappa/du due to linearization in m (is a fractional tolerance when m is
+             large, as generally occurs during the beginning of the evolution, and is an absolute tolerance when m is small,
+             as generally occurs towards the end of the evolution)
+    root_tol_raise_err: Whether to raise an error (True) or just print a warning (False) when the error in the
+                        roots of the cubic is larger than root_tol
+
+    - Optional:
+
+    use_mpmath: If True, use mpmath functions instead of numpy functions to solve the cubic and evaluate the elliptic integrals
+                (slower but more accurate), default: False
+    polyroots_maxsteps: The maximum number of steps allowed in mpmath.polyroots if use_mpmath == True, default: 50
+    polyroots_extraprec: The number of extra digits to use in mpmath.polyroots if use_mpmath == True, default: 50
+
+    Output: (<S^2>_prec - S0^2)/(1 - q) [= dkappa_{xi q}/du]
+
+    """
+
+    # Compute the desired quantity for u > 0, otherwise set to 0
+
+    if u > 0.:
+        # Compute coefficients of equation
+
+        vareps, Btransf, Ctransf, Dtransf = eq_coeffs(
+            u, kappaxiq, q, S1, S2, xi, S0sq)
+
+        # Compute the roots of the equation
+
+        # First, check whether vareps is small enough that we can linearize <S^2>_pr in mm and compute
+        # S_+^2 + S_-^2 to good accuracy by solving the quadratic one obtains by setting eps = 0 in the cubic
+        # (so S_+^2 + S_-^2 = -Ctransf/Btransf).
+
+        # Check if vareps is small enough that we can just solve the quadratic for S_+^2 + S_-^2
+        # This is currently just implemented for the case where Btransf, Ctransf, Dtransf > 0
+        # This is the case where there are issues in practice and also is the one that is easy to analyze
+
+        only_quadratic = False
+
+        # The second check is that vareps is small enough that the square root in z_eps_abs makes sense
+        if min(Btransf, Ctransf, Dtransf) > 0 and 4.*Ctransf*vareps < Btransf*Btransf:
+            z_eps_abs = abs(
+                (-Btransf + (Btransf*Btransf - 4.*Ctransf*vareps)**0.5)/(2.*Ctransf))
+
+            # Error on -Ctransf/Btransf expression for S_+^2 + S_-^2, E_\varepsilon^\text{sum} [Eq. (C5) in the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902]
+            sum_err_bound = z_eps_abs * \
+                (abs(Ctransf*Ctransf - Btransf*Dtransf) + Ctransf*Dtransf*z_eps_abs) / \
+                (Btransf*(Btransf - Ctransf*z_eps_abs))
+
+            # Check that the additional inequality is satisfied
+            if z_eps_abs*(Ctransf/Btransf + sum_err_bound) < 1:
+                m_bound = z_eps_abs*(Ctransf/Btransf + sum_err_bound) / \
+                    (1. - z_eps_abs*(Ctransf/Btransf + sum_err_bound))
+
+                # This bounds S_+^2 - S_-^2 by abs(S_+^2 + S_-^2) in the first part, which is not sharp
+                # We check that m_bound < 1 since E_ratio_term is only real for m_bound < 1
+                # It should be possible to restrict computing E_ratio_term only for smaller m_bound values,
+                # but it's not clear if this is necessary
+                if m_bound < 1.:
+                    if E_ratio_term(m_bound) <= 0.5*lin_tol*max(1., 1./(Ctransf/Btransf + sum_err_bound)) \
+                       and sum_err_bound < lin_tol:
+                        only_quadratic = True
+
+        if only_quadratic:
+            Ssq_pr = -0.5*Ctransf/Btransf
+
+            # Check that Ssq_pr [= (bar{S}_+^2 + bar{S}_-^2)/2] gives an unbarred value that is nonnegative
+
+            Ssq_pr_unbarred = (1. - q)*Ssq_pr + S0sq
+
+            if Ssq_pr_unbarred < 0.:
+                raise RuntimeError("The unbarred S_+^2 + S_-^2 value (which was all that was necessary to compute to this accuracy) is negative, "
+                                   "indicating a problem with the evolution, with a value of %e"%(2.*Ssq_pr_unbarred))
+        else:
+            Ssq3, Ssqm, Ssqp = solve_cubic(
+                [vareps, Btransf, Ctransf, Dtransf], u, kappaxiq, q, S1, S2, imag_tol, use_mpmath=use_mpmath, root_tol=root_tol,
+                polyroots_maxsteps=polyroots_maxsteps, polyroots_extraprec=polyroots_extraprec)
+
+            # Check that Ssqp and Ssqm give unbarred values that are nonnegative
+
+            Ssqp_unbarred = (1. - q)*Ssqp + S0sq
+            Ssqm_unbarred = (1. - q)*Ssqm + S0sq
+
+            if Ssqp_unbarred < 0. or Ssqm_unbarred < 0.:
+                raise RuntimeError("At least one of the unbarred S_+^2 and S_-^2 values are negative, indicating a problem with the evolution, "
+                                   "with values of %e, %e"%(Ssqp_unbarred, Ssqm_unbarred))
+
+            # Compute <S^2>_pr
+
+            if Ssqp == Ssq3:
+                raise RuntimeError(
+                    "Degenerate nonprecessing case with all roots equal. Cannot continue with computation.")
+
+            # When abs(Ssq3) is large, mm is small, and one can write Ssq_pr to good accuracy in a form that does not
+            # involve Ssq3 (which has a large residual when put back into the equation when it is large, likely due to
+            # catastrophic cancellation), but the residual is still smaller than Ssq3, so we take the value of Ssq3
+            # output by the solver to give us an accurate enough value to use in checking the size of mm
+
+            mm = (Ssqp - Ssqm)/(Ssqp - Ssq3)  # Eq. (25) in Chatziioannou, Klein, Yunes, and Cornish PRD 95, 104004 (2017)
+
+            # Check that the equation is satisfied to good accuracy (by checking that the residual changes sign over an
+            # interval given by root_tol) and compute Ssq_pr
+
+            def cubic(Ssq):
+                if use_mpmath:
+                    return mp.polyval([vareps, Btransf, Ctransf, Dtransf], Ssq)
+                else:
+                    return ((vareps*Ssq + Btransf)*Ssq + Ctransf)*Ssq + Dtransf
+
+            def cubic_deriv(Ssq):
+                if use_mpmath:
+                    return mp.polyval([3*vareps, 2*Btransf, Ctransf], Ssq)
+                else:
+                    return (3.*vareps*Ssq + 2.*Btransf)*Ssq + Ctransf
+
+            # If the difference between Ssqp and Ssqm or Ssqm and Ssq3 is smaller than the difference set by root_tol,
+            # shrink root_tol appropriately
+            if Ssqp - Ssqm < 1.1*max(0.5*abs(Ssqp + Ssqm), 1.)*root_tol:
+                root_tol_pm = 0.5*(Ssqp - Ssqm)/max(0.5*abs(Ssqp + Ssqm), 1.)
+            else:
+                root_tol_pm = root_tol
+
+            if Ssqm - Ssq3 < 1.1*max(0.5*abs(Ssqm + Ssq3), 1.)*root_tol:
+                root_tol_m3 = 0.5*(Ssqm - Ssq3)/max(0.5*abs(Ssqm + Ssq3), 1.)
+            else:
+                root_tol_m3 = root_tol
+
+            root_tol_message_extra_bit_flag = False
+
+            root_tol_orig = root_tol
+
+            if root_tol_pm < root_tol and root_tol_pm <= root_tol_m3:
+                root_tol = root_tol_pm
+
+                root_tol_message_extra_bit_flag = True
+
+                root_tol_message_extra_bit_insert = "S_+^2 - S_-^2"
+            elif root_tol_m3 < root_tol:
+                root_tol = root_tol_m3
+
+                root_tol_message_extra_bit_flag = True
+
+                root_tol_message_extra_bit_insert = "S_-^2 - S_3^2"
+
+            if root_tol_message_extra_bit_flag:
+                root_tol_message_extra_bit = " the internally modified (since %s is small) version "\
+                                             "of"%root_tol_message_extra_bit_insert
+            else:
+                root_tol_message_extra_bit = ""
+
+            Ssqp_resp, Ssqp_resm = cubic(
+                Ssqp + max(abs(Ssqp), 1.)*root_tol), cubic(Ssqp - max(abs(Ssqp), 1.)*root_tol)
+            Ssqm_resp, Ssqm_resm = cubic(
+                Ssqm + max(abs(Ssqm), 1.)*root_tol), cubic(Ssqm - max(abs(Ssqm), 1.)*root_tol)
+
+            # Check for special cases and if we can linearize
+
+            if use_mpmath:
+                Btransf_out, Ctransf_out, Dtransf_out = Btransf, Ctransf, Dtransf
+            else:
+                Btransf_out, Ctransf_out, Dtransf_out = Btransf[0], Ctransf[0], Dtransf[0]
+
+            if Ssqp == Ssqm:
+                # Since here we have a double root, we will check that the derivative of the cubic changes sign
+                # If the difference between Ssqm and Ssq3 is smaller than the difference set by root_tol,
+                # shrink root_tol appropriately
+                if Ssqm - Ssq3 < 1.1*max(0.5*abs(Ssqm + Ssq3), 1.)*root_tol_orig:
+                    root_tol_m0 = 0.5*(Ssqm - Ssq3)/max(0.5*abs(Ssqm + Ssq3), 1.)
+
+                    root_tol_message_extra_bit = " the internally modified (since S_-^2 - S_3^2 is small) "\
+                                                 "version of"
+                else:
+                    root_tol_m0 = root_tol_orig
+
+                    root_tol_message_extra_bit = ""
+                Ssqp_resp, Ssqp_resm = cubic_deriv(
+                    Ssqp + max(abs(Ssqp), 1.)*root_tol_m0), cubic_deriv(Ssqp - max(abs(Ssqp), 1.)*root_tol_m0)
+
+                if np.sign(Ssqp_resp) == np.sign(Ssqp_resm):
+                    root_tol_message = "Error in S_+^2 is greater than%s the tolerance root_tol = %e--the "\
+                                       "residual does not change sign over the interval determined by root_tol "\
+                                       "(S_+^2 = S_-^2, so m = 0 and we do not need to consider S_3^2)\n"\
+                                       "Printing the residuals for S^2 = S_+^2 + max(abs(S_+^2),1)*root_tol, S_+^2 "\
+                                       "- max(abs(S_+^2),1)*root_tol, as well as the values of u, S_+^2, "\
+                                       "q*(1 - q^2)*u^2, bar{B}, bar{C}, bar{D}\n%e %e %e %f %e %e %e %e" % \
+                                       (root_tol_message_extra_bit, root_tol_m0, Ssqp_resp, Ssqp_resm, u, Ssqp, vareps,
+                                        Btransf_out, Ctransf_out, Dtransf_out)
+
+                    if root_tol_raise_err:
+                        raise RuntimeError(root_tol_message)
+                    else:
+                        warn(root_tol_message, RuntimeWarning)
+
+                Ssq_pr = Ssqp # Here m = 0
+            elif Ssqm == Ssq3:
+                # Since here we have a double root, we will check that the derivative of the cubic changes sign
+                # If the difference between Ssqp and Ssqm is smaller than the difference set by root_tol,
+                # shrink root_tol appropriately
+                if Ssqp - Ssqm < 1.1*max(0.5*abs(Ssqp + Ssqm), 1.)*root_tol_orig:
+                    root_tol_m1 = 0.5*(Ssqp - Ssqm)/max(0.5*abs(Ssqp + Ssqm), 1.)
+
+                    root_tol_message_extra_bit = " the internally modified (since S_+^2 - S_-^2 is small) "\
+                                                 "version of"
+                else:
+                    root_tol_m1 = root_tol_orig
+
+                    root_tol_message_extra_bit = ""
+                Ssqm_resp, Ssqm_resm = cubic_deriv(
+                    Ssqm + max(abs(Ssqm), 1.)*root_tol_m1), cubic_deriv(Ssqm - max(abs(Ssqm), 1.)*root_tol_m1)
+
+                if np.sign(Ssqm_resp) == np.sign(Ssqm_resm):
+                    root_tol_message = "Error in S_-^2 is greater than%s the tolerance root_tol = %e--the "\
+                                       "residual does not change sign over the interval determined by root_tol "\
+                                       "(S_-^2 = S_3^2, so m = 1 and we do not need to consider S_+^2)\n"\
+                                       "Printing the residuals for S^2 = S_-^2 + max(abs(S_-^2),1)*root_tol, S_-^2 "\
+                                       "- max(abs(S_-^2),1)*root_tol, as well as the values of u, S_-^2, "\
+                                       "q*(1 - q^2)*u^2, bar{B}, bar{C}, bar{D}\n%e %e %e %f %e %e %e %e" % \
+                                       (root_tol_message_extra_bit, root_tol_m1, Ssqm_resp, Ssqm_resm, u, Ssqm, vareps,
+                                        Btransf_out, Ctransf_out, Dtransf_out)
+
+                    if root_tol_raise_err:
+                        raise RuntimeError(root_tol_message)
+                    else:
+                        warn(root_tol_message, RuntimeWarning)
+
+                Ssq_pr = Ssq3 # Here m = 1
+            elif E_ratio_term(mm) <= lin_tol*max(1., 1./(Ssqp - Ssqm)): # Linearization check
+                if np.sign(Ssqp_resp) == np.sign(Ssqp_resm) or np.sign(Ssqm_resp) == np.sign(Ssqm_resm):
+                    root_tol_message = "Error in S_+^2 and/or S_-^2 is greater than%s the tolerance root_tol = %e--the "\
+                                       "residual does not change sign over the interval determined by root_tol "\
+                                       "(m is small enough to be linearized in, so we do not consider S_3^2)\n"\
+                                       "Printing the residuals for S^2 = S_+^2 + max(abs(S_+^2),1)*root_tol, S_+^2 "\
+                                       "- max(abs(S_+^2),1)*root_tol, S_-^2 + max(abs(S_-^2),1)*root_tol, "\
+                                       "S_-^2 - max(abs(S_-^2),1)*root_tol, as well as the values of u, S_+^2, S_-^2, "\
+                                       "q*(1 - q^2)*u^2, bar{B}, bar{C}, bar{D}\n%e %e %e %e %e %f %f %e %e %e %e" % \
+                                       (root_tol_message_extra_bit, root_tol, Ssqp_resp, Ssqp_resm, Ssqm_resp, Ssqm_resm,
+                                        u, Ssqp, Ssqm, vareps, Btransf_out, Ctransf_out, Dtransf_out)
+
+                    if root_tol_raise_err:
+                        raise RuntimeError(root_tol_message)
+                    else:
+                        warn(root_tol_message, RuntimeWarning)
+
+                # Eq. (42) in Chatziioannou, Klein, Yunes, and Cornish PRD 95, 104004 (2017), with the elliptic functions linearized in m
+                Ssq_pr = 0.5*(Ssqp + Ssqm)
+            else:
+                Ssq3_resp, Ssq3_resm = cubic(
+                    Ssq3 + max(abs(Ssq3), 1.)*root_tol), cubic(Ssq3 - max(abs(Ssq3), 1.)*root_tol)
+
+                if np.sign(Ssqp_resp) == np.sign(Ssqp_resm) or np.sign(Ssqm_resp) == np.sign(Ssqm_resm) \
+                   or np.sign(Ssq3_resp) == np.sign(Ssq3_resm):
+                    root_tol_message = "Error in S_+^2, S_-^2, and/or S_3^2 is greater than%s the tolerance root_tol = "\
+                                       "%e--the residual does not change sign over the interval determined by "\
+                                       "root_tol\nPrinting the residuals for S^2 = S_+^2 + max(abs(S_+^2),1)*root_tol, "\
+                                       "S_+^2 - max(abs(S_+^2),1)*root_tol, S_-^2 + max(abs(S_-^2),1)*root_tol, "\
+                                       "S_-^2 - max(abs(S_-^2),1)*root_tol, S_3^2 + max(abs(S_3^2),1)*root_tol, "\
+                                       "S_3^2 - max(abs(S_-^2),1)*root_tol, as well as the values of u, S_+^2, S_-^2, "\
+                                       "S_3^2, q*(1 - q^2)*u^2, bar{B}, bar{C}, bar{D}\n"\
+                                       "%e %e %e %e %e %e %e %f %f %f %e %e %e %e" % \
+                                       (root_tol_message_extra_bit, root_tol, Ssqp_resp, Ssqp_resm, Ssqm_resp, Ssqm_resm,
+                                        Ssq3_resp, Ssq3_resm, u, Ssqp, Ssqm, Ssq3, vareps, Btransf_out, Ctransf_out,
+                                        Dtransf_out)
+
+                    if root_tol_raise_err:
+                        raise RuntimeError(root_tol_message)
+                    else:
+                        warn(root_tol_message, RuntimeWarning)
+
+                if use_mpmath:
+                    ellip_ratio = mp.ellipe(mm)/mp.ellipk(mm)
+                else:
+                    ellip_ratio = ellipe(mm)/ellipk(mm)
+
+                # Eq. (17) in the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902
+                Ssq_pr = Ssqp + (ellip_ratio - 1.)*(Ssqp - Ssq3)
+    else:
+        Ssq_pr = 0.
+
+    return Ssq_pr
+
+
+def kappaxiq_evol(q, S1, S2, S1L, xi, S0sq, u0, uf, du, method, imag_tol, root_tol, lin_tol, ode_atol, ode_rtol,
+                  root_tol_raise_err, use_solve_ivp=False, solve_ivp_lsoda_min_step=0., use_mpmath=False,
+                  polyroots_maxsteps=50, polyroots_extraprec=50, odefun_degree=None):
+    """
+    Compute kappaxiq when the binary has u = uf starting from u = u0
+
+    Inputs:
+
+    - Required:
+
+    q: Mass ratio of the binary, with the < 1 convention
+    S1, S2: Magnitudes of the dimensionful spins of the binary (with total mass = 1)
+    S1L: Component of the dimensionful spin of hole 1 along the orbital angular momentum (with total mass = 1)
+    xi: Conserved effective spin, defined in Eq. (2) of the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902
+    S0sq: Square of initial dimensionful total spin (in total mass = 1 units)
+    u0: Initial value of u
+    uf: Final value of u
+    du: Step in u for the evolution
+    method: method to use for scipy integration ("vode", "lsoda", "dopri5", or "dop853" for the ode interface and
+            "RK45", "RK23", "DOP853", or "LSODA" for the solve_ivp interface); not used if use_mpmath == True
+    imag_tol: Tolerance for the imaginary part of roots of the cubic
+    root_tol: Tolerance for the error in the roots of the cubic (is an absolute tolerance for small values of the roots, as
+              generally occurs during the beginning of the evolution, and is a fractional tolerance when the roots are large,
+              which generally occurs towards the end of the evolution; also is reduced automatically when the roots are close
+              together)
+    lin_tol: Tolerance for errors on the r.h.s. of dkappa/du due to linearization in m (is a fractional tolerance when m is
+             large, as generally occurs during the beginning of the evolution, and is an absolute tolerance when m is small,
+             as generally occurs towards the end of the evolution)
+    ode_atol: Absolute tolerance for the default scipy ode solver, also the single tolerance setting for the mpmath ode solver
+              if it is used
+    ode_rtol: Relative tolerance for the default scipy ode solver
+    root_tol_raise_err: Whether to raise an error (True) or just print a warning (False) when the error in the roots of
+                        the cubic is larger than root_tol
+
+    - Optional:
+
+    use_solve_ivp: If True, use scipy's solve_ivp function for the ODE integration, which gives access to a different set of
+                   integrators and allows tighter tolerances for LSODA than the default ode function, but can hang for certain
+                   parameters, default: False
+    solve_ivp_lsoda_min_step: Minimum step to use in the solve_ivp LSODA integrator, default: 0.
+
+    use_mpmath: If True, use mpmath functions and ode integration instead of numpy/scipy (slower but more accurate), default: False
+    polyroots_maxsteps: The maximum number of steps allowed in mpmath.polyroots if use_mpmath == True, default: 50
+    polyroots_extraprec: The number of extra digits to use in mpmath.polyroots if use_mpmath == True, default: 50
+    odefun_degree: Degree of polynomial used by mpmath.odefun if use_mpmath == True; calculated automatically by
+                   mpmath.odefun if None, default: None
+
+    Output: kappaxiq at Lf
+    """
+
+    # Compute initial value of kappaxiq
+
+    kappaxiq0 = S1L
+
+    if use_solve_ivp:
+        # Use scipy.integrate.solve_ivp
+
+        # Set up kwargs for solver (necessary since one gets a warning if one passes min_step with an integrator other than LSODA)
+        solve_ivp_kwargs = {'atol': ode_atol, 'rtol': ode_rtol}
+
+        if method == "LSODA":
+            solve_ivp_kwargs['min_step'] = solve_ivp_lsoda_min_step
+
+        # Solve
+        soln = solve_ivp(dkappaxiq_du, [u0, uf], [kappaxiq0], method=method, t_eval=[uf], args=(q, S1, S2, xi, S0sq, imag_tol,
+                         root_tol, lin_tol, root_tol_raise_err), **solve_ivp_kwargs)
+
+        # Return kappaxiq at u = uf
+
+        return soln.y[0]
+    elif use_mpmath:
+        # Use mpmath.odefun
+
+        # Solve (we have to write it in terms of ub := -u so that the final value of ub is greater than the initial value, as
+        # required by mpmath.odefun)
+        soln = mp.odefun(lambda ub, kappaxiq: -dkappaxiq_du(-ub, kappaxiq, q, S1, S2, xi, S0sq, imag_tol, root_tol,
+                                                             lin_tol, root_tol_raise_err, use_mpmath=True,
+                                                             polyroots_maxsteps=polyroots_maxsteps,
+                                                             polyroots_extraprec=polyroots_extraprec),
+                         -u0, kappaxiq0, ode_atol, odefun_degree)
+
+        # Return kappaxiq at u = uf
+
+        return soln(-uf)
+    else:
+        # Use scipy.integrate.ode
+
+        # initialize the ODE solver
+        solver = ode(dkappaxiq_du)
+        solver.set_integrator(method, atol=ode_atol, rtol=ode_rtol)
+        solver.set_initial_value(kappaxiq0, u0)
+        solver.set_f_params(q, S1, S2, xi, S0sq, imag_tol,
+                            root_tol, lin_tol, root_tol_raise_err)
+
+        # Solve
+
+        end = False
+
+        while solver.successful() and not end:
+            new_t = solver.t - du
+
+            if new_t <= uf:
+                new_t = uf
+                end = True
+
+            solver.integrate(new_t, step=1)
+
+        if abs(solver.t - uf) > 0.75*du:
+            raise RuntimeError(
+                "Final u of %e is more than 0.75*du different from uf of %e" % (solver.t, uf))
+
+        # Return kappaxiq at u = uf
+
+        return solver.y
+
+
+def prec_avg_tilt_comp_vec_inputs(m1, m2, chi1_vec, chi2_vec, fref, L0_vec=None, Lf=None, du=1.e-3, ode_atol_base=1.e-8,
+                                  ode_atol_inplane_spin_scaling=True, ode_atol_floor=1.e-13, ode_rtol=-1, method="lsoda",
+                                  use_solve_ivp=False, solve_ivp_lsoda_min_step=0., use_fallback=True, du_fallback=1.e-5,
+                                  ode_atol_fallback=1.e-13, ode_rtol_fallback=-1, method_fallback="lsoda",
+                                  use_mpmath_fallback=True, mpmath_dps=30, ode_tol_mpmath=1.e-15, odefun_degree=None,
+                                  polyroots_maxsteps=50, polyroots_extraprec=50, LPNorder=2, LPNspins=True, imag_tol=1.e-6,
+                                  root_tol=1.e-6, lin_tol=-1, lin_tol_fallback=-1, root_tol_raise_err=False,
+                                  failure_mode='None', version='v1'):
+    """
+    Compute tilt angles at infinite separation or bounds and average value of tilt angles at finite separation
+    from masses, spins, tilt angles, and in-plane spin angle at some frequency, taken to be small enough that
+    the precession averaged spin evolution from Gerosa et al. PRD 92, 064016 (2015) [implemented following
+    Chatziioannou, Klein, Yunes, and Cornish PRD 95, 104004 (2017)] is accurate, using the regularized expressions
+    from <https://dcc.ligo.org/P2100029>, arXiv:2107.11902. This version takes vectors of spins as input (as opposed
+    to spin angles) and allows one optionally to specify the initial orbital angular momentum instead of computing it
+    from fref (and the binary's parameters).
+
+    Inputs:
+
+    - Required:
+
+    m1, m2: Detector frame masses of the binary, in kg
+    chi1_vec, chi2_vec: 3d numpy arrays with the dimensionless spins of the binary; these must be in coordinates where
+                        the Newtonian orbital angular momentum is in the z-direction if L0_vec is None (the default),
+                        otherwise all that is necessary is that chi1_vec, chi2_vec, and L0_vec are all in the same
+                        coordinate system
+    fref: Reference frequency, in Hz; ignored when L0_vec is not None
+
+    - Optional initial and final L settings:
+
+    L0_vec: 3d numpy array with the initial orbital angular momentum in units where the binary's total mass = 1 and in
+            the same coordinates as the spins; this is computed from fref and the binary parameters if None is passed
+            and in that case given in coordinates where the Newtonian orbital angular momentum is in the z-direction,
+            default: None
+    Lf: Final magnitude of orbital angular momentum in total mass = 1 units for output at finite separation,
+        None gives the output at infinity, default: None
+
+    - Optional settings for the evolution:
+
+    du: Step in u for the evolution (external to the ODE integrator), default: 1.e-3
+    ode_atol_base: Base value of absolute tolerance for ode solver, default: 1.e-8
+    ode_atol_inplane_spin_scaling: Scale ode_atol_base by min(chi1*np.sin(tilt1),chi2*np.sin(tilt2)), with a floor of
+                                   ode_atol_floor, to give good accuracy for small in-plane spin cases, default: True
+    ode_atol_floor: Floor value for absolute tolerance for ode solver used when ode_atol_inplane_spin_scaling == True;
+                    should be several orders of magnitude less than ode_atol_base, default: 1.e-13
+    ode_rtol: Relative tolerance for ode solver, -1 sets this to 0.1*ode_atol, default: -1
+    method: method to use for integration (either from scipy's ode interface, where "vode", "lsoda", "dopri5", and
+            "dop853" have been tested, though "lsoda" is the recommended one; "RK45", "RK23", "DOP853", or "LSODA" for
+            scipy's solve_ivp interface, if use_solve_ivp == True ("lsoda" and "dop853" are equivalent to "LSODA" and
+            "DOP853" here, for compatibility); or "odefun" to use mpmath.odefun with its default Taylor method, which
+            is all that is available as of v1.2.0; this uses the mpmath fallback settings and disables the fallback
+            evolution--it is also quite slow), default: "lsoda"
+    use_solve_ivp: If True, use scipy's solve_ivp interface and its integrators instead of the default ode interface; note
+                   that the solve_ivp integrators apparently hang for certain difficult cases, default: False
+    solve_ivp_lsoda_min_step: Minimum step to use in the solve_ivp LSODA integrator, default: 0.
+    use_fallback: Whether to use the fallback integration if the primary integration fails (True) or not (False),
+                  not available if method == "odefun", default: True
+    du_fallback: Step in u for the evolution if the first evolution fails, default: 1.e-5
+    ode_atol_fallback: Absolute tolerance for ode solver if the first evolution fails, default: 1.e-13
+    ode_rtol_fallback: Relative tolerance for ode solver if the first evolution fails, -1 sets this to
+                       0.1*ode_atol_fallback, default: -1
+    method_fallback: method to use for integration if the first evolution fails (any of the scipy integrators available for method),
+                     default: "lsoda"
+    use_mpmath_fallback: Use the slow but accurate mpmath integration if the first fallback integration fails (True) or not (False),
+                         default: True
+    mpmath_dps: The number of decimal places to use for the mpmath computations, if the mpmath integration is triggered or
+                method == "odefun", default: 30
+    ode_tol_mpmath: Tolerance for the mpmath ode solver, only used if the mpmath fallback evolution is triggered or
+                    method == "odefun", default: 1.e-15
+    odefun_degree: Degree of polynomial used by mpmath.odefun if the mpmath fallback evolution is triggered or method == "odefun";
+                   calculated automatically by mpmath.odefun if None, default: None
+    polyroots_maxsteps: The maximum number of steps allowed in mpmath.polyroots if the mpmath fallback evolution is triggered or
+                        method == "odefun", default: 50
+    polyroots_extraprec: The number of extra digits to use in mpmath.polyroots if the mpmath fallback evolution is triggered or
+                         method == "odefun", default: 50
+    LPNorder: PN order to use in computing the intial orbital angular momentum from fref--choices are
+              0, 1, 1.5, 2, or 2.5, where the half-integer orders are the spin contributions; ignored when L0_vec is
+              not None, default: 2
+    LPNspins: Whether to include the orbit-averaged spin contributions in the computation of the initial orbital
+              angular momentum; ignored when L0_vec is not None, default: True
+    imag_tol: Tolerance for the imaginary part of roots of the cubic, default: 1.e-6
+    root_tol: Tolerance for the error in the roots of the cubic (is an absolute tolerance for small values of the roots, as
+              generally occurs during the beginning of the evolution, and is a fractional tolerance when the roots are large,
+              which generally occurs towards the end of the evolution; also is reduced automatically when the roots are close
+              together), default: 1.e-6
+    lin_tol: Tolerance for errors on the r.h.s. of dkappa/du due to linearization in m (is a fractional tolerance when m is
+             large, as generally occurs during the beginning of the evolution, and is an absolute tolerance when m is small,
+             as generally occurs towards the end of the evolution); -1 sets this to be the same as ode_atol, except for the
+             mpmath evolution, where it is always set to the ode_tol_mpmath value, default: -1
+    lin_tol_fallback: Tolerance for errors on the r.h.s. of dkappa/du due to linearization in m if the first evolution
+                      fails, -1 sets this to be the same as ode_atol_fallback, default: -1
+    root_tol_raise_err: Whether to raise an error (True) or just print a warning (False) when the error in the roots
+                        of the cubic is larger than root_tol, default: False
+    failure_mode: How the code behaves when the evolution fails (after the fallback evolution, if use_fallback is True, the
+                  default). 'Error' means that the code raises a RuntimeError, while 'NAN' and 'None' mean that the code
+                  raises a RuntimeWarning and returns np.nan or None for the output, respectively, default: 'None'
+    version: Version of the calculation to use--currently only the initial version, v1, is available, default: "v1"
+
+    Output: dictionary with entries 'tilt1_inf', 'tilt2_inf' for evolution to infinity and entries 'tilt1_sep_min',
+            'tilt1_sep_max', 'tilt1_sep_avg', 'tilt2_sep_min', 'tilt2_sep_max', 'tilt2_sep_avg' for evolution to a
+            finite separation (i.e., a finite orbital angular momentum)
+    """
+
+    # Set scipy integrators for which this has been tested
+
+    if use_solve_ivp:
+        tested_scipy_integrators = ["RK45", "RK23", "DOP853", "LSODA"]
+    else:
+        tested_scipy_integrators = ["vode", "lsoda", "dopri5", "dop853"]
+
+    tested_scipy_integrator_string = ', '.join(tested_scipy_integrators)
+
+    # Set known failure modes
+
+    known_failure_modes = ['Error', 'NAN', 'None']
+
+    # Check version
+
+    if version != 'v1':
+        raise ValueError("Only version v1 is available at the moment, while version = %s"%version)
+
+    # Make "lsoda" and "dop853" uppercase when passed with use_solve_ivp == True, for compatibility
+
+    methods_to_uppercase = ["lsoda", "dop853"]
+
+    if use_solve_ivp:
+        if method in methods_to_uppercase:
+            method = method.upper()
+
+        if method_fallback in methods_to_uppercase:
+            method_fallback = method_fallback.upper()
+
+    # Check that inputs are physical and otherwise make sense
+
+    check_masses(m1, m2)
+
+    if type(chi1_vec) is not np.ndarray or type(chi2_vec) is not np.ndarray or len(np.atleast_1d(chi1_vec)) != 3 or \
+       len(np.atleast_1d(chi2_vec)) != 3:
+        raise ValueError(
+            "chi1_vec and chi2_vec must both be numpy arrays of length 3, while they are", chi1_vec, chi2_vec)
+
+    if L0_vec is not None:
+        if len(L0_vec) != 3:
+            raise ValueError(
+                "L0_vec must be either None or a numpy array of length 3, while it is", L0_vec)
+
+    chi1 = norm(chi1_vec)
+    chi2 = norm(chi2_vec)
+
+    if chi1 == 0. or chi1 > 1. or chi2 == 0. or chi2 > 1.:
+        raise ValueError(
+            "The magnitudes of the spins must both be positive and at most 1, while they are chi1, chi2 = %f, %f" %
+            (chi1, chi2))
+
+    if fref is not None:
+        check_fref(fref, m1, m2, "precession-averaged")
+    elif L0_vec is None:
+        raise ValueError("One of fref and L0_vec must not be None")
+
+    if du <= 0. or du_fallback <= 0.:
+        raise ValueError(
+            "The step in u and its fallback value must both be positive, while they are du, du_fallback = %e, %e" %
+            (du, du_fallback))
+
+    if use_solve_ivp and not solve_ivp_avail:
+        import scipy
+
+        raise ValueError("solve_ivp is not available in the version of scipy being used (v%s), so use_solve_ivp == True "
+                         "is not allowed."%scipy.__version__)
+
+    if use_solve_ivp:
+        interface_str = "solve_ivp"
+    else:
+        interface_str = "ode"
+
+    if method not in np.append(tested_scipy_integrators, "odefun") or method_fallback not in tested_scipy_integrators:
+        raise ValueError("The integration method must be odefun or one of the tested scipy integrators available with "
+                         "the %s interface, %s, and the fallback integration method must be one of the tested scipy "
+                         "integrators, while they are method, method_fallback = %s, %s" %
+                         (interface_str, tested_scipy_integrator_string, method, method_fallback))
+
+    if method == "odefun" and not mpmath_avail:
+        raise ValueError("mpmath is not available, so odefun is not a possible choice for the integration method, "
+                         "which must in this case be one of the tested scipy integrators: %s"%tested_scipy_integrator_string)
+
+    if mpmath_dps <= 0 or not isinstance(mpmath_dps, int):
+        raise ValueError("The number of decimal places to use for the mpmath computations must be a natural number, "
+                         "while it is mpmath_dps = %f"%mpmath_dps)
+
+    if odefun_degree is not None and (odefun_degree <= 0 or not isinstance(odefun_degree, int)):
+        raise ValueError("The degree for the mpmath.odefun must be a natural number, while it is odefun_degree = %f"%odefun_degree)
+
+    if polyroots_maxsteps <= 0 or polyroots_extraprec <= 0 or not isinstance(polyroots_maxsteps, int) \
+            or not isinstance(polyroots_extraprec, int):
+        raise ValueError("The maximum number of steps and extra precision for mpmath.polyroots must both be natural "
+                         "numbers, while they are polyroots_maxsteps, polyroots_extraprec = %f, %f"%(polyroots_maxsteps,
+                                                                                                     polyroots_extraprec))
+
+    if L0_vec is None and LPNorder not in LPN_orders:
+        raise ValueError("The PN order to use to compute the initial orbital angular momentum must be one of %s, "
+                         "while it is LPNorder = %f" % (LPN_order_string, LPNorder))
+
+    if failure_mode not in known_failure_modes:
+        known_failure_modes_string = ', '.join(known_failure_modes)
+        raise ValueError("The failure mode must be one of %s, while it is failure_mode = %s" %
+                         (known_failure_modes_string, failure_mode))
+
+    # Print a warning if both fref and L0_vec are not None
+
+    if fref is not None and L0_vec is not None:
+        warn("fref and L0_vec are both not None. In this case, the L0_vec value is used to initialize the "
+             "evolution and fref is not used.", ValueWarning)
+
+    # Set the failure output and the string corresponding to it
+    # These default to None, since they are not used if failure_mode == 'Error'
+
+    if failure_mode == 'NAN':
+        failure_output = np.nan
+        failure_output_string = 'np.nan'
+    else:
+        failure_output = None
+        failure_output_string = 'None'
+
+    # Set failure message for mpmath evolution
+
+    def failure_message_mpmath(err):
+        return "Encountered the exception\n\n%s\n\nusing the mpmath evolution and mpmath_dps = %d, " \
+               "ode_tol_mpmath = %e, odefun_degree = %s, polyroots_maxsteps = %d, polyroots_extraprec " \
+               "= %d. However, it may be possible to evolve this case successfully using more digits " \
+               "and/or a tighter tolerance "%(format_error(err), mpmath_dps, ode_tol_mpmath, str(odefun_degree),
+                                              polyroots_maxsteps, polyroots_extraprec)
+
+    # Check that m1 > m2, and swap the objects if m1 < m2; raise an error if m1 == m2
+    # This does not rotate the spins, since we do not need to track the orientation of the binary
+
+    swap = False
+
+    if m1 < m2:
+        swap = True
+        m1, m2 = m2, m1
+        chi1, chi2 = chi2, chi1
+        chi1_vec, chi2_vec = chi2_vec, chi1_vec
+
+    eq_mass_check(m1, m2, Lf)
+
+    # Save the masses in kg for later use
+
+    m1_kg = m1
+    m2_kg = m2
+
+    # Compute derived quantities (scaling the masses so that their sum is 1)
+
+    q = m2/m1
+
+    M = m1 + m2
+
+    m1 /= M
+    m2 /= M
+
+    S1 = m1*m1*chi1
+    S2 = m2*m2*chi2
+
+    S1_vec = m1*m1*chi1_vec
+    S2_vec = m2*m2*chi2_vec
+
+    # Compute initial orbital angular momentum (in total mass = 1 units)
+    # We need to pass the masses in kg here to convert the frequency when computing from fref
+
+    if L0_vec is not None:
+        L0 = norm(L0_vec)
+    else:
+        if LPNspins:
+            S1_vec_for_L = S1_vec
+            S2_vec_for_L = S2_vec
+        else:
+            S1_vec_for_L = np.zeros(3)
+            S2_vec_for_L = np.zeros(3)
+
+        L0_vec = L_vec_from_masses_spins_and_f(
+            m1_kg, m2_kg, fref, PNorder=LPNorder, S1_vec=S1_vec_for_L, S2_vec=S2_vec_for_L, no_mass_spin_input_check=True)
+
+        L0 = norm(L0_vec)
+
+    if L0 == 0.:
+        raise ValueError("L0_vec must not be the zero vector")
+
+    if Lf is not None and Lf <= L0:
+        raise ValueError(
+            "The magnitude of the final orbital angular momentum must be greater than that of the "
+            "initial orbital angular momentum, %f, while it is Lf = %f" % (L0, Lf))
+
+    # Setting ode_atol for the case when ode_atol_inplane_spin_scaling == True
+
+    L0_hat = L0_vec/L0
+
+    chi1_inplane = norm(chi1_vec - np.dot(chi1_vec, L0_hat)*L0_hat)
+    chi2_inplane = norm(chi2_vec - np.dot(chi2_vec, L0_hat)*L0_hat)
+
+    if ode_atol_inplane_spin_scaling:
+        ode_atol = max(min(chi1_inplane, chi2_inplane)
+                       * ode_atol_base, ode_atol_floor)
+    else:
+        ode_atol = ode_atol_base
+
+    # Take care of the -1 cases for ode_rtol and lin_tol and their fallback versions
+
+    if ode_rtol == -1:
+        ode_rtol = 0.1*ode_atol
+
+    if ode_rtol_fallback == -1:
+        ode_rtol_fallback = 0.1*ode_atol_fallback
+
+    if lin_tol == -1:
+        lin_tol = ode_atol
+
+    if lin_tol_fallback == -1:
+        lin_tol_fallback = ode_atol_fallback
+
+    # Check the tolerance settings
+
+    if ode_atol_base <= 0. or ode_atol_floor <= 0. or ode_rtol <= 0. or ode_atol_fallback <= 0. \
+       or ode_rtol_fallback <= 0. or ode_tol_mpmath <= 0.:
+        raise ValueError(
+            "The ode solver tolerance values and their fallback values must all be positive, while they are "
+            "ode_atol_base, ode_atol_floor, ode_rtol, ode_atol_fallback, ode_rtol_fallback, "
+            "ode_tol_mpmath = %e, %e, %e, %e, %e, %e" %
+            (ode_atol_base, ode_atol_floor, ode_rtol, ode_atol_fallback, ode_rtol_fallback, ode_tol_mpmath))
+
+    if ode_atol_inplane_spin_scaling and ode_atol_floor >= ode_atol_base:
+        raise ValueError("ode_atol_floor should be less than ode_atol_base, likely by several orders of magnitude, "
+                         "while they are ode_atol_floor, ode_atol_base = %e, %e" %
+                         (ode_atol_floor, ode_atol_base))
+
+    if imag_tol < 0. or root_tol < 0. or lin_tol < 0. or lin_tol_fallback < 0.:
+        raise ValueError("The tolerance values must all be nonnegative, while they are imag_tol, root_tol, lin_tol, "
+                         "lin_tol_fallback = %e, %e, %e, %e" % (imag_tol, root_tol, lin_tol, lin_tol_fallback))
+
+    # Compute effective spin [Eq. (12) of Gerosa et al.]
+
+    xi = np.dot((1. + q)*S1_vec + (1. + 1./q)*S2_vec, L0_hat)
+
+    # Compute projection of S1_vec onto L0_vec
+
+    S1L = np.dot(S1_vec, L0_hat)
+    S2L = np.dot(S2_vec, L0_hat)
+
+    # Return the initial tilts when spins are aligned and not in an unstable configuration
+
+    cos_tilt1 = np.clip(S1L/S1, -1., 1.)
+    cos_tilt2 = np.clip(S2L/S2, -1., 1.)
+
+    tilt1 = np.arccos(cos_tilt1)
+    tilt2 = np.arccos(cos_tilt2)
+
+    if (tilt1 == np.pi and tilt2 in [0., np.pi]) or (tilt1 == 0. and tilt2 == 0.):
+        return package_tilts(tilt1, tilt2, Lf, swap=swap)
+
+    # Compute initial total dimensionful spin and angular momentum in total mass = 1 units
+
+    S0sq = norm(S1_vec + S2_vec)**2
+
+    # Compute kappa at infinity or Lf
+
+    u0 = 0.5/L0
+
+    if Lf is not None:
+        uf = 0.5/Lf
+    else:
+        uf = 0.
+
+    mpmath_tilts = False
+
+    if method == "odefun":
+            try:
+                mpmath_tilts = True
+
+                with mp.workdps(mpmath_dps):
+                    k_evol = kappaxiq_evol(q, S1, S2, S1L, xi, S0sq, u0, uf, du, method, imag_tol, root_tol, ode_tol_mpmath,
+                                           ode_tol_mpmath, ode_rtol, root_tol_raise_err, use_mpmath=True,
+                                           polyroots_maxsteps=polyroots_maxsteps, polyroots_extraprec=polyroots_extraprec,
+                                           odefun_degree=odefun_degree)
+
+            except (RuntimeError, mp.ctx_base.StandardBaseContext.NoConvergence) as err:
+                return evolution_error_handling(failure_mode, failure_message_mpmath(err), failure_output, failure_output_string, Lf, swap)
+
+            except NonprecessingError as ne:
+                return package_tilts(tilt1, tilt2, Lf, swap)
+    else:
+        try:
+            k_evol = kappaxiq_evol(q, S1, S2, S1L, xi, S0sq, u0, uf, du, method, imag_tol, root_tol, lin_tol,
+                                   ode_atol, ode_rtol, root_tol_raise_err, use_solve_ivp=use_solve_ivp,
+                                   solve_ivp_lsoda_min_step=solve_ivp_lsoda_min_step)
+
+        except RuntimeError as re:
+            if use_fallback:
+                warn("Encountered the exception\n\n%s\n\nusing method %s and du = %e, ode_atol = %e, ode_rtol = %e, "
+                     "lin_tol = %e, trying with %s with du = %e, ode_atol = %e, ode_rtol = %e, lin_tol = %e" %
+                     (format_error(re), method, du, ode_atol, ode_rtol, lin_tol, method_fallback, du_fallback,
+                      ode_atol_fallback, ode_rtol_fallback, lin_tol_fallback), RuntimeWarning)
+
+                try:
+                    k_evol = kappaxiq_evol(q, S1, S2, S1L, xi, S0sq, u0, uf, du_fallback, method_fallback, imag_tol,
+                                           root_tol, lin_tol_fallback, ode_atol_fallback, ode_rtol_fallback,
+                                           root_tol_raise_err, solve_ivp_lsoda_min_step=solve_ivp_lsoda_min_step)
+
+                except RuntimeError as re2:
+                    if use_mpmath_fallback and mpmath_avail:
+                        warn("Encountered the exception\n\n%s\n\nin the first fallback evolution. Now trying with the mpmath "
+                             "fallback with mpmath_dps = %d, ode_tol_mpmath = %e, odefun_degree = %s, "
+                             "polyroots_maxsteps = %d, polyroots_extraprec = %d"%(format_error(re2), mpmath_dps, ode_tol_mpmath, str(odefun_degree),
+                                                                                  polyroots_maxsteps, polyroots_extraprec), RuntimeWarning)
+
+                        try:
+                            mpmath_tilts = True
+
+                            with mp.workdps(mpmath_dps):
+                                k_evol = kappaxiq_evol(q, S1, S2, S1L, xi, S0sq, u0, uf, du, method, imag_tol, root_tol, ode_tol_mpmath,
+                                                       ode_tol_mpmath, ode_rtol, root_tol_raise_err, use_mpmath=True,
+                                                       polyroots_maxsteps=polyroots_maxsteps, polyroots_extraprec=polyroots_extraprec,
+                                                       odefun_degree=odefun_degree)
+
+                        except (RuntimeError, mp.ctx_base.StandardBaseContext.NoConvergence) as err:
+                            return evolution_error_handling(failure_mode, failure_message_mpmath(err), failure_output, failure_output_string, Lf, swap)
+
+                        except NonprecessingError as ne:
+                            return package_tilts(tilt1, tilt2, Lf, swap)
+                    else:
+                        if not use_mpmath_fallback:
+                            mpmath_bit = "it is disabled"
+                        else:
+                            mpmath_bit = "mpmath is not available"
+
+                        failure_message = "Fallback evolution failed with the exception\n\n%s\n\nThe mpmath fallback evolution might work in this " \
+                                          "case, but %s. Alternatively, you can try adjusting the tolerance settings for the evolution, but this " \
+                                          "case may not be possible to evolve accurately with the scipy integration." % (format_error(re2), mpmath_bit)
+
+                        return evolution_error_handling(failure_mode, failure_message, failure_output, failure_output_string, Lf, swap)
+
+                except NonprecessingError as ne:
+                    return package_tilts(tilt1, tilt2, Lf, swap)
+
+            else:
+                failure_message = "Evolution failed with the exception\n\n%s\n\nYou can try adjusting the tolerance settings " \
+                                  "for the evolution, but this case may not be possible to evolve accurately with the current " \
+                                  "version of the precession-averaged evolution."%format_error(re)
+
+                return evolution_error_handling(failure_mode, failure_message, failure_output, failure_output_string, Lf, swap)
+
+        except NonprecessingError as ne:
+            return package_tilts(tilt1, tilt2, Lf, swap)
+
+    # Compute the desired output. We have already accounted for the case where at least one spin is zero,
+    # so there are no concerns about division by zero here.
+
+    if Lf is None:
+        # Compute cosines of tilt angles at infinity [Eqs. (15) in the paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902]
+
+        cos_tilt1inf = k_evol/S1
+        cos_tilt2inf = q*(xi/(1. + q) - k_evol)/S2
+
+        # Clip to avoid issues with taking the arccos if there are rounding issues
+
+        cos_tilt1inf = np.clip(cos_tilt1inf, -1., 1.)
+        cos_tilt2inf = np.clip(cos_tilt2inf, -1., 1.)
+
+        # Return tilt angles at infinity
+
+        if mpmath_tilts:
+            tilt1inf = float(mp.acos(cos_tilt1inf))
+            tilt2inf = float(mp.acos(cos_tilt2inf))
+        else:
+            tilt1inf = np.arccos(cos_tilt1inf[0])
+            tilt2inf = np.arccos(cos_tilt2inf[0])
+
+        return package_tilts(tilt1inf, tilt2inf, None, swap)
+    else:
+        vareps, Btransf, Ctransf, Dtransf = eq_coeffs(
+            uf, k_evol, q, S1, S2, xi, S0sq)
+
+        _, Ssqm, Ssqp = solve_cubic(
+            [vareps, Btransf, Ctransf, Dtransf], uf, k_evol, q, S1, S2, imag_tol, use_mpmath=mpmath_tilts,
+            root_tol=root_tol, polyroots_maxsteps=polyroots_maxsteps, polyroots_extraprec=polyroots_extraprec)
+
+        if mpmath_tilts:
+            lin_tol_use = ode_tol_mpmath
+        else:
+            lin_tol_use = lin_tol
+
+        Ssqavg = dkappaxiq_du(uf, k_evol, q, S1, S2, xi, S0sq, imag_tol, root_tol, lin_tol_use, root_tol_raise_err,
+                              use_mpmath=mpmath_tilts, polyroots_maxsteps=polyroots_maxsteps,
+                              polyroots_extraprec=polyroots_extraprec)
+
+        # Compute cosines of tilt angles at separation [Eqs. (23) in paper <https://dcc.ligo.org/P2100029>, arXiv:2107.11902]
+
+        cos_tilt1sep_Sm = (k_evol - uf*Ssqm)/S1
+        cos_tilt2sep_Sm = q*(xi/(1. + q) - S1*cos_tilt1sep_Sm)/S2
+
+        cos_tilt1sep_Sp = (k_evol - uf*Ssqp)/S1
+        cos_tilt2sep_Sp = q*(xi/(1. + q) - S1*cos_tilt1sep_Sp)/S2
+
+        cos_tilt1sep_Savg = (k_evol - uf*Ssqavg)/S1
+        cos_tilt2sep_Savg = q*(xi/(1. + q) - S1*cos_tilt1sep_Savg)/S2
+
+        # Clip to avoid issues with taking the arccos if there are rounding issues
+
+        cos_tilt1sep_Sm = np.clip(cos_tilt1sep_Sm, -1., 1.)
+        cos_tilt2sep_Sp = np.clip(cos_tilt2sep_Sp, -1., 1.)
+
+        cos_tilt1sep_Sm = np.clip(cos_tilt1sep_Sm, -1., 1.)
+        cos_tilt2sep_Sp = np.clip(cos_tilt2sep_Sp, -1., 1.)
+
+        cos_tilt1sep_Savg = np.clip(cos_tilt1sep_Savg, -1., 1.)
+        cos_tilt2sep_Savg = np.clip(cos_tilt2sep_Savg, -1., 1.)
+
+        # Return tilt angles at separation
+
+        if swap:
+            cos_tilt1sep_Sm, cos_tilt2sep_Sm, cos_tilt1sep_Sp, cos_tilt2sep_Sp = cos_tilt2sep_Sp, cos_tilt1sep_Sp, \
+                                                                                 cos_tilt2sep_Sm, cos_tilt1sep_Sm
+            cos_tilt1sep_Savg, cos_tilt2sep_Savg = cos_tilt2sep_Savg, cos_tilt1sep_Savg
+
+        if mpmath_tilts:
+            tilt1sepmin = float(mp.acos(cos_tilt1sep_Sm))
+            tilt1sepmax = float(mp.acos(cos_tilt1sep_Sp))
+            tilt1sepavg = float(mp.acos(cos_tilt1sep_Savg))
+
+            tilt2sepmin = float(mp.acos(cos_tilt2sep_Sp))
+            tilt2sepmax = float(mp.acos(cos_tilt2sep_Sm))
+            tilt2sepavg = float(mp.acos(cos_tilt2sep_Savg))
+        else:
+            tilt1sepmin = np.arccos(cos_tilt1sep_Sm[0])
+            tilt1sepmax = np.arccos(cos_tilt1sep_Sp[0])
+            tilt1sepavg = np.arccos(cos_tilt1sep_Savg[0])
+
+            tilt2sepmin = np.arccos(cos_tilt2sep_Sp[0])
+            tilt2sepmax = np.arccos(cos_tilt2sep_Sm[0])
+            tilt2sepavg = np.arccos(cos_tilt2sep_Savg[0])
+
+        return {'tilt1_sep_min': tilt1sepmin, 'tilt1_sep_max': tilt1sepmax, 'tilt1_sep_avg': tilt1sepavg,
+                'tilt2_sep_min': tilt2sepmin, 'tilt2_sep_max': tilt2sepmax, 'tilt2_sep_avg': tilt2sepavg}
+
+
+def prec_avg_tilt_comp(m1, m2, chi1, chi2, tilt1, tilt2, phi12, fref, Lf=None, **kwargs):
+    """
+    Compute tilt angles at infinite separation or bounds and average value of tilt angles at finite separation from
+    masses, spins, tilt angles, and in-plane spin angle at some frequency, taken to be small enough that the precession
+    averaged spin evolution from Gerosa et al. PRD 92, 064016 (2015) [implemented following Chatziioannou, Klein, Yunes,
+    and Cornish PRD 95, 104004 (2017)] is accurate, using the regularized expressions. This takes in spin angles and a
+    reference frequency to initialize the evolution.
+
+    Inputs:
+
+    - Required:
+
+    m1, m2: Detector frame masses of the binary, in kg
+    chi1, chi2: Dimensionless spin magnitudes of the binary
+    tilt1, tilt2: Tilt angles of the binary's spins (w.r.t. the Newtonian orbital angular momentum) at fref
+    phi12: Angle between the in-plane components of the spins at fref
+    fref: Reference frequency, in Hz
+
+    - Optional settings [passed to prec_avg_tilt_comp_vec_inputs(); all but Lf passed through kwargs]:
+
+    Lf: Final magnitude of orbital angular momentum in total mass = 1 units for output at finite separation
+        None gives the output at infinity, default: None
+
+    du: Step in u for the evolution (external to the ODE integrator), default: 1.e-3
+    ode_atol_base: Base value of absolute tolerance for ode solver, default: 1.e-8
+    ode_atol_inplane_spin_scaling: Scale ode_atol_base by min(chi1*np.sin(tilt1),chi2*np.sin(tilt2)), with a floor of
+                                   ode_atol_floor, to give good accuracy for small in-plane spin cases, default: True
+    ode_atol_floor: Floor value for absolute tolerance for ode solver used when ode_atol_inplane_spin_scaling == True;
+                    should be several orders of magnitude less than ode_atol_base, default: 1.e-13
+    ode_rtol: Relative tolerance for ode solver, -1 sets this to 0.1*ode_atol, default: -1
+    method: method to use for integration (either from scipy's ode interface, where "vode", "lsoda", "dopri5", and
+            "dop853" have been tested, though "lsoda" is the recommended one; "RK45", "RK23", "DOP853", or "LSODA" for
+            scipy's solve_ivp interface, if use_solve_ivp == True ("lsoda" and "dop853" are equivalent to "LSODA" and
+            "DOP853" here, for compatibility); or "odefun" to use mpmath.odefun with its default Taylor method, which
+            is all that is available as of v1.2.0; this uses the mpmath fallback settings and disables the fallback
+            evolution--it is also quite slow), default: "lsoda"
+    use_solve_ivp: If True, use scipy's solve_ivp interface and its integrators instead of the default ode interface; note
+                   that the solve_ivp integrators apparently hang for certain difficult cases, default: False
+    solve_ivp_lsoda_min_step: Minimum step to use in the solve_ivp LSODA integrator, default: 0.
+    use_fallback: Whether to use the fallback integration if the primary integration fails (True) or not (False),
+                  not available if method == "odefun", default: True
+    du_fallback: Step in u for the evolution if the first evolution fails, default: 1.e-5
+    ode_atol_fallback: Absolute tolerance for ode solver if the first evolution fails, default: 1.e-13
+    ode_rtol_fallback: Relative tolerance for ode solver if the first evolution fails, -1 sets this to
+                       0.1*ode_atol_fallback, default: -1
+    method_fallback: method to use for integration if the first evolution fails (any of the scipy integrators available for method),
+                     default: "lsoda"
+    use_mpmath_fallback: Use the slow but accurate mpmath integration if the first fallback integration fails (True) or not (False),
+                         default: True
+    mpmath_dps: The number of decimal places to use for the mpmath computations, if the mpmath integration is triggered or
+                method == "odefun", default: 30
+    ode_tol_mpmath: Tolerance for the mpmath ode solver, only used if the mpmath fallback evolution is triggered or
+                    method == "odefun", default: 1.e-15
+    odefun_degree: Degree of polynomial used by mpmath.odefun if the mpmath fallback evolution is triggered or method == "odefun";
+                   calculated automatically by mpmath.odefun if None, default: None
+    polyroots_maxsteps: The maximum number of steps allowed in mpmath.polyroots if the mpmath fallback evolution is triggered or
+                        method == "odefun", default: 50
+    polyroots_extraprec: The number of extra digits to use in mpmath.polyroots if the mpmath fallback evolution is triggered or
+                         method == "odefun", default: 50
+    LPNorder: PN order to use in computing the intial orbital angular momentum from fref--choices are
+              0, 1, 1.5, 2, or 2.5, where the half-integer orders are the spin contributions, default: 2
+    LPNspins: Whether to include the orbit-averaged spin contributions in the computation of the initial orbital
+              angular momentum, default: True
+    imag_tol: Tolerance for the imaginary part of roots of the cubic, default: 1.e-6
+    root_tol: Tolerance for the error in the roots of the cubic (is an absolute tolerance for small values of the roots, as
+              generally occurs during the beginning of the evolution, and is a fractional tolerance when the roots are large,
+              which generally occurs towards the end of the evolution; also is reduced automatically when the roots are close
+              together), default: 1.e-6
+    lin_tol: Tolerance for errors on the r.h.s. of dkappa/du due to linearization in m (is a fractional tolerance when m is
+             large, as generally occurs during the beginning of the evolution, and is an absolute tolerance when m is small,
+             as generally occurs towards the end of the evolution); -1 sets this to be the same as ode_atol, except for the
+             mpmath evolution, where it is always set to the ode_tol_mpmath value, default: -1
+    lin_tol_fallback: Tolerance for errors on the r.h.s. of dkappa/du due to linearization in m if the first evolution
+                      fails, -1 sets this to be the same as ode_atol_fallback, default: -1
+    root_tol_raise_err: Whether to raise an error (True) or just print a warning (False) when the error in the roots
+                        of the cubic is larger than root_tol, default: False
+    failure_mode: How the code behaves when the evolution fails (after the fallback evolution, if use_fallback is True, the
+                  default). 'Error' means that the code raises a RuntimeError, while 'NAN' and 'None' mean that the code
+                  raises a RuntimeWarning and returns np.nan or None for the output, respectively, default: 'None'
+    version: Version of the calculation to use--currently only the initial version, v1, is available, default: "v1"
+
+    Output: dictionary with entries 'tilt1_inf', 'tilt2_inf' for evolution to infinity and entries 'tilt1_sep_min',
+            'tilt1_sep_max', 'tilt1_sep_avg', 'tilt2_sep_min', 'tilt2_sep_max', 'tilt2_sep_avg' for evolution to a
+            finite separation (i.e., a finite orbital angular momentum)
+    """
+
+    # Return initial tilts for single-spin or nonspinning cases
+
+    if chi1 == 0. or chi2 == 0.:
+        return package_tilts(tilt1, tilt2, Lf, False)
+
+    # Check that spins and tilt angles are physical/make sense [other checks carried out in prec_avg_tilt_comp_vec_inputs()]
+
+    check_spin_mags(chi1, chi2)
+
+    check_tilts(tilt1, tilt2)
+
+    # Convert spin magnitudes and angles to vectors
+
+    chi1_vec = chi1*np.array([np.sin(tilt1), 0., np.cos(tilt1)])
+
+    chi2_vec = chi2*np.array([np.sin(tilt2)*np.cos(phi12),
+                              np.sin(tilt2)*np.sin(phi12), np.cos(tilt2)])
+
+    # Pass to prec_avg_tilt_comp_vec_inputs() to do calculation
+
+    return prec_avg_tilt_comp_vec_inputs(m1, m2, chi1_vec, chi2_vec, fref, Lf=Lf, **kwargs)