Source code for oxygen

#! /usr/bin/env python
# -*- coding: utf-8 -*-
"""
This module computes the neutral and ionized populations of O in the upper
atmosphere.
"""

from __future__ import (division, print_function, absolute_import,
                        unicode_literals)
import numpy as np
import astropy.units as u
import astropy.constants as c
from scipy.integrate import simps, solve_ivp, odeint
from p_winds import tools, microphysics
import warnings


__all__ = ["radiative_processes", "electron_impact_ionization", "recombination",
           "charge_transfer", "ion_fraction"]


# Some hard coding based on the astrophysical literature
_SOLAR_OXYGEN_ABUNDANCE_ = 8.69  # Asplund et al. 2009
_SOLAR_OXYGEN_FRACTION_ = 10 ** (_SOLAR_OXYGEN_ABUNDANCE_ - 12.00)


# Photoionization of O I (neutral)
[docs] def radiative_processes(spectrum_at_planet): """ Calculate the photoionization rate of oxygen at null optical depth based on the EUV spectrum arriving at the planet. Parameters ---------- spectrum_at_planet : ``dict`` Spectrum of the host star arriving at the planet covering fluxes at least up to the wavelength corresponding to the energy to ionize oxygen (13.62 eV, or 910 Angstrom). Returns ------- phi_oi : ``float`` Ionization rate of O I at null optical depth in unit of 1 / s. a_oi : ``float`` Flux-averaged photoionization cross-section of O I in unit of cm ** 2. a_h_oi : ``float`` Flux-averaged photoionization cross-section of H I in the range absorbed by O I in unit of cm ** 2. a_he : ``float`` Flux-averaged photoionization cross-section of He I in unit of cm ** 2. """ wavelength = (spectrum_at_planet['wavelength'] * spectrum_at_planet['wavelength_unit']).to(u.angstrom).value flux_lambda = (spectrum_at_planet['flux_lambda'] * spectrum_at_planet[ 'flux_unit']).to(u.erg / u.s / u.cm ** 2 / u.angstrom).value energy = ((c.h * (c.c / wavelength / u.angstrom).to(u.Hz)).to(u.eV)).value energy_erg = (energy * u.eV).to(u.erg).value # Auxiliary definitions parameters_dict = microphysics.sigma_properties_v1996() energy_threshold_oi = parameters_dict['O I'][0] # Ionization threshold in # eV wl_break_oi = 12398.42 / energy_threshold_oi # O I ionization threshold in # angstrom wl_break_he = 504 # He ionization threshold in angstrom i0 = tools.nearest_index(wavelength, wl_break_he) i1 = tools.nearest_index(wavelength, wl_break_oi) wavelength_cut_0 = wavelength[:i0] flux_lambda_cut_0 = flux_lambda[:i0] wavelength_cut_1 = wavelength[:i1] flux_lambda_cut_1 = flux_lambda[:i1] energy_cut_1 = energy_erg[:i1] # Calculate the photoionization cross-section a_lambda_oi = microphysics.general_cross_section(wavelength_cut_1, species='O I') # The flux-averaged photoionization cross-section of O I a_oi = abs(simps(flux_lambda_cut_1 * a_lambda_oi, wavelength_cut_1) / simps(flux_lambda_cut_1, wavelength_cut_1)) # The flux-averaged photoionization cross-section of H is also going to be # needed because it adds to the optical depth that O I see. a_lambda_h_oi = microphysics.hydrogen_cross_section( wavelength=wavelength_cut_1) a_h_oi = abs(simps(flux_lambda_cut_1 * a_lambda_h_oi, wavelength_cut_1) / simps(flux_lambda_cut_1, wavelength_cut_1)) # Same for the He atoms, but only up to the He ionization threshold a_lambda_he = microphysics.helium_total_cross_section(wavelength_cut_0) a_he = abs(simps(flux_lambda_cut_0 * a_lambda_he, wavelength_cut_0) / simps(flux_lambda_cut_0, wavelength_cut_0)) # Calculate the photoionization rates phi_oi = abs(simps(flux_lambda_cut_1 * a_lambda_oi / energy_cut_1, wavelength_cut_1)) return phi_oi, a_oi, a_h_oi, a_he
# Ionization rate of O by electron impact
[docs] def electron_impact_ionization(electron_temperature): """ Calculates the electron impact ionization rate that consumes neutral O and produces singly-ionized O. Based on the formula of Voronov 1997 (https://ui.adsabs.harvard.edu/abs/1997ADNDT..65....1V/abstract). Parameters ---------- electron_temperature : ``float`` Temperature of the plasma where the electrons are embedded in unit of Kelvin. Returns ------- ionization_rate_oi : ``float`` Ionization rate of neutral O into singly-ionized O in unit of cm ** 3 / s. """ boltzmann_constant = 8.617333262145179e-05 # eV / K electron_energy = boltzmann_constant * electron_temperature energy_ratio_oi = 11.3 / electron_energy ionization_rate_oi = 3.59E-8 * (0.073 + energy_ratio_oi) ** (-1) * \ energy_ratio_oi ** 0.34 * np.exp(-energy_ratio_oi) return ionization_rate_oi
# Recombination of singly-ionized O into neutral O
[docs] def recombination(electron_temperature): """ Calculates the rate of recombination of singly-ionized O with an electron to produce a neutral O atom. Based on the formulation of Woodall et al. 2007 (https://ui.adsabs.harvard.edu/abs/2007A%26A...466.1197W/abstract). Parameters ---------- electron_temperature : ``float`` Temperature of the plasma where the electrons are embedded in unit of Kelvin. Returns ------- alpha_rec_oi : ``float`` Recombination rate of O II into O I in units of cm ** 3 / s. """ alpha_rec_oi = 3.25E-12 * (300 / electron_temperature) ** 0.66 return alpha_rec_oi
# Charge transfer between O and H
[docs] def charge_transfer(temperature): """ Calculates the charge exchange rates of O with H nuclei. Based on the formulation of Woodall et al. 2007 (https://ui.adsabs.harvard.edu/abs/2007A%26A...466.1197W/abstract). Parameters ---------- temperature : ``float`` Isothermal temperature of the upper atmosphere in unit of Kelvin. Returns ------- ct_rate_oi_hp : ``float`` Charge transfer rate between neutral O and H+ in units of cm ** 3 / s. ct_rate_oii_h : ``float`` Charge transfer rate between O+ and neutral H in units of cm ** 3 / s. """ # Recombination of O II into O I ct_rate_oii_h = 5.66E-10 * (300 / temperature) ** (-0.36) * \ np.exp(8.6 / temperature) # Ionization of O I into O II ct_rate_oi_hp = 7.31E-10 * (300 / temperature) ** (-0.23) * \ np.exp(-226 / temperature) return ct_rate_oi_hp, ct_rate_oii_h
# Calculation the number fractions of O II
[docs] def ion_fraction(radius_profile, velocity, density, hydrogen_ion_fraction, helium_ion_fraction, planet_radius, temperature, h_fraction, speed_sonic_point, radius_sonic_point, density_sonic_point, spectrum_at_planet, o_fraction=_SOLAR_OXYGEN_FRACTION_, initial_f_o_ion=0.0, relax_solution=False, convergence=0.01, max_n_relax=10, method='Radau', return_rates=False, **options_solve_ivp): """ Calculate the fraction of ionized oxygen in the upper atmosphere in function of the radius in unit of planetary radius. Parameters ---------- radius_profile : ``numpy.ndarray`` Radius in unit of planetary radii. velocity : ``numpy.ndarray`` Velocities sampled at the values of ``radius_profile`` in units of sound speed. Similar to the output of ``parker.structure()``. density : ``numpy.ndarray`` Densities sampled at the values of ``radius_profile`` in units of density at the sonic point. Similar to the output of ``parker.structure()``. hydrogen_ion_fraction : ``numpy.ndarray`` Number fraction of H ion over total H in the upper atmosphere in function of radius. Similar to the output of ``hydrogen.ion_fraction()``. helium_ion_fraction : ``numpy.ndarray`` Number fraction of He ion over total He in the upper atmosphere in function of radius. Similar to the output of ``helium.population_fraction()``, but should be ``1 - f_1_r - f_3_r``. planet_radius : ``float`` Planetary radius in unit of Jupiter radius. temperature : ``float`` Isothermal temperature of the upper atmosphere in unit of Kelvin. h_fraction : ``float`` Total (ion + neutral) H number fraction of the atmosphere. speed_sonic_point : ``float`` Speed of sound in the outflow in units of km / s. radius_sonic_point : ``float`` Radius of the sonic point in unit of Jupiter radius. density_sonic_point : ``float`` Density at the sonic point in units of g / cm ** 3. spectrum_at_planet : ``dict`` Spectrum of the host star arriving at the planet covering fluxes at least up to the wavelength corresponding to the energy to ionize oxygen (13.62 eV, or 910 Angstrom). Can be generated using ``tools.make_spectrum_dict``. o_fraction : ``float``, optional Fraction of total oxygen in the upper atmosphere. Default value assumes solar abundance. initial_f_o_ion : ``float``, optional The initial oxygen ion fraction at the layer near the surface of the planet. Default is 0.0, i.e., 100% neutral. relax_solution : ``bool``, optional The first solution is calculating by initially assuming the entire atmosphere is in neutral state. If ``True``, the solution will be re-calculated in a loop until it converges to a delta_f of 1%, or for a maximum of 10 loops (default parameters). Default is ``False``. convergence : ``float``, optional Value of delta_f at which to stop the relaxation of the solution for ``f_r``. Default is 0.01. max_n_relax : ``int``, optional Maximum number of loops to perform the relaxation of the solution for the ion fractions. Default is 10. method : ``str``, optional If method is ``'odeint'``, then ``scipy.integrate.odeint()`` is used instead of ``scipy.integrate.solve_ivp()`` to calculate the steady-state distribution of helium. Any other method will fall back to an option of ``solve_ivp()`` methods. For example, if ``method`` is set to ``'Radau'``, then use ``solve_ivp(method='Radau')``. Default is ``'Radau'``. return_rates : ``bool``, optional If ``True``, then this function also returns a ``dict`` object containing the various reaction rates in function of radius and in units of 1 / s. Default is ``False``. **options_solve_ivp: Options to be passed to the ``scipy.integrate.solve_ivp()`` solver. You may want to change the options ``atol`` (absolute tolerance; default is 1E-6) or ``rtol`` (relative tolerance; default is 1E-3). If you are having numerical issues, you may want to decrease the tolerance by a factor of 10 or 100, or 1000 in extreme cases. Returns ------- f_oii_r : ``numpy.ndarray`` Fraction of singly-ionized oxygen in function of radius. reaction_rates : ``dict`` Dictionary containing the reaction rates in function of radius and in units of 1 / s. Only returned when ``return_rates`` is set to ``True``. Here is a short description of the dict keys: * `photoionization`: Photoionization of O I into O II * `recombination`: Recombination of O II into O I * `e_impact_ionization`: Electron impact ionization of O I into O II * `charge_exchange_HII`: Charge exchange between O I and H II * `charge_exchange_HI`: Charge exchange between O II and H I """ vs = speed_sonic_point # km / s rs = radius_sonic_point # jupiterRad rhos = density_sonic_point # g / cm ** 3 # Recombination rates of C in unit of rs ** 2 * vs alpha_rec_unit = ((rs * 7.1492E+09) ** 2 * vs * 1E5) # cm ** 3 / s alpha_rec_oi = recombination(temperature) alpha_rec_oi = alpha_rec_oi / alpha_rec_unit # Hydrogen mass in unit of rhos * rs ** 3 m_h_unit = (rhos * (rs * 7.1492E+09) ** 3) # Converted to g m_h = 1.67262192E-24 / m_h_unit # Photoionization rates at null optical depth at the distance of the planet # from the host star, in unit of vs / rs, and the flux-averaged # cross-sections in units of rs ** 2 phi_unit = vs * 1E5 / rs / 7.1492E+09 # 1 / s phi_oi, a_oi, a_h_oi, a_he = radiative_processes(spectrum_at_planet) phi_oi = phi_oi / phi_unit a_oi = a_oi / (rs * 7.1492E+09) ** 2 a_h_oi = a_h_oi / (rs * 7.1492E+09) ** 2 a_he = a_he / (rs * 7.1492E+09) ** 2 # Electron-impact ionization rate for C I in the same unit as the # recombination rates ionization_rate_oi = electron_impact_ionization(temperature) ionization_rate_oi = ionization_rate_oi / alpha_rec_unit # Charge transfer rates in the same unit as the recombination rates ct_rate_oi_hp, ct_rate_oii_h = charge_transfer(temperature) ct_rate_oii_h = ct_rate_oii_h / alpha_rec_unit ct_rate_oi_hp = ct_rate_oi_hp / alpha_rec_unit # We solve the steady-state ionization balance in a similar way that we do # for He # The radius in unit of radius at the sonic point r = radius_profile * planet_radius / rs dr = np.diff(r) dr = np.concatenate((dr, np.array([dr[-1], ]))) # With all this setup done, now we need to assume something about the # distribution of neutral O in the atmosphere. Let's assume it based on the # initial guess input. column_density = np.flip(np.cumsum(np.flip(dr * density))) # Total column # density column_density_h_0 = np.flip( # Column density of atomic H only np.cumsum(np.flip(dr * density * (1 - hydrogen_ion_fraction)))) he_fraction = 1 - h_fraction column_density_he_0 = np.flip( # Column density of atomic He only np.cumsum(np.flip(dr * density * he_fraction * (1 - helium_ion_fraction)))) k1 = h_fraction / (h_fraction + 4 * he_fraction + 16 * o_fraction) / m_h k2 = he_fraction / (h_fraction + 4 * he_fraction + 16 * o_fraction) / m_h k3 = o_fraction / (h_fraction + 4 * he_fraction + 16 * o_fraction) / m_h tau_oi_h = k1 * a_h_oi * column_density_h_0 tau_o_he = k2 * a_he * column_density_he_0 tau_oi = (1 - initial_f_o_ion) * k3 * a_oi * column_density + tau_oi_h + \ tau_o_he # The differential equation def _fun(_r, y, rates=False): f_oii = y _v = np.interp(_r, r, velocity) _rho = np.interp(_r, r, density) f_h_ion = np.interp(_r, r, hydrogen_ion_fraction) # Fraction of H+ f_he_ion = np.interp(_r, r, helium_ion_fraction) # Fraction of He+ # Assume the number density of electrons is equal to the number density # of H ions + He ions # Since we may run into loss of numerical precision here (big numbers), # we manipulate the equations to avoid this problem. It looks a bit # messy, but it is necessary log_term_1 = np.log(k1) + np.log(_rho) # H ions log_term_2 = np.log(k2) + np.log(_rho) # He ions ionization_rate_oi_n_e = \ np.exp(log_term_1 + np.log(ionization_rate_oi)) * f_h_ion + \ np.exp(log_term_2 + np.log(ionization_rate_oi)) * f_he_ion ct_rate_oi_hp_n_h_plus = \ np.exp(log_term_1 + np.log(ct_rate_oi_hp)) * f_h_ion alpha_rec_oi_n_e = \ np.exp(log_term_1 + np.log(alpha_rec_oi)) * f_h_ion + \ np.exp(log_term_2 + np.log(alpha_rec_oi)) * f_he_ion ct_rate_oii_h_n_h0 = \ np.exp(log_term_1 + np.log(ct_rate_oii_h)) * (1 - f_h_ion) # n_e = k1 * _rho * f_h_ion + k2 * _rho * f_he_ion # Number density of # # electrons # n_h_plus = k1 * _rho * f_h_ion # Number density of ionized H # n_h0 = k1 * _rho * (1 - f_h_ion) # Number density of atomic H # Terms of dfoii_dr t_oi = np.interp(_r, r, tau_oi) term1 = (1 - f_oii) * phi_oi * np.exp(-t_oi) # Photoionization term2 = (1 - f_oii) * ionization_rate_oi_n_e # Electron-impact # ionization term3 = (1 - f_oii) * ct_rate_oi_hp_n_h_plus # Charge exchange with # H+ term4 = f_oii * alpha_rec_oi_n_e # Recombination of O II into O I term5 = f_oii * ct_rate_oii_h_n_h0 # Charge exchange of O II with # neutral H dfoii_dr = (term1 + term2 + term3 - term4 - term5) / _v if rates is False: return dfoii_dr else: return np.array([term1, term4, term2, term3, term5]) * phi_unit if method == 'odeint': # Since 'odeint' yields only warnings when precision is lost or when # there is a problem, we transform these warnings into an exception with warnings.catch_warnings(): warnings.filterwarnings("error") try: sol = odeint(_fun, y0=initial_f_o_ion, t=r, tfirst=True) except Warning: raise RuntimeError('The solver ``odeint`` failed to obtain a ' 'solution.') f_oii_r = np.copy(sol).T[0] else: # We solve it using `scipy.solve_ivp` sol = solve_ivp(_fun, (r[0], r[-1],), np.array([initial_f_o_ion, ]), t_eval=r, method=method, **options_solve_ivp) f_oii_r = sol['y'][0] # When `solve_ivp` has problems, it may return an array with different # size than `r`. So we raise an exception if this happens if len(f_oii_r) != len(r): raise RuntimeError('The solver ``solve_ivp`` failed to obtain a' ' solution.') # High densities can be numerically unstable and produce unphysical values # of `f_r`, so we replace negative values with zero and values above 1.0 # with 1.0 f_oii_r[f_oii_r < 0] = 1E-15 f_oii_r[f_oii_r > 1.0] = 1.0 # For the sake of self-consistency, there is the option of repeating the # calculation of f_r by updating the optical depth with the new ion # fractions. if relax_solution is True: for i in range(max_n_relax): previous_f_oii_r = np.copy(f_oii_r) # Re-calculate the column densities tau_oi = \ k3 * a_oi * np.flip(np.cumsum( np.flip(dr * density * (1 - f_oii_r)))) + tau_oi_h + \ tau_o_he # Solve it again if method == 'odeint': sol = odeint(_fun, y0=initial_f_o_ion, t=r, tfirst=True) f_oii_r = np.copy(sol).T[0] else: sol = solve_ivp(_fun, (r[0], r[-1],), np.array([initial_f_o_ion, ]), t_eval=r, method=method, **options_solve_ivp) f_oii_r = sol['y'][0] # Replace negative values with zero and values above 1.0 with # 1.0 f_oii_r[f_oii_r < 0] = 1E-15 f_oii_r[f_oii_r > 1.0] = 1.0 # Calculate the relative change of f_ion in the outer shell of # the atmosphere (where we expect the most important change) relative_delta_f_oii = abs(np.sum(f_oii_r - previous_f_oii_r)) \ / np.sum(previous_f_oii_r) # Break the loop if convergence is achieved if relative_delta_f_oii < convergence: break else: pass else: pass if return_rates is False: return f_oii_r else: ionization_rate, recombination_rate, e_impact_ion, ch_exchange_hii,\ ch_exchange_hi = _fun(r, f_oii_r, rates=True) reaction_rates = {'photoionization': ionization_rate, 'recombination': recombination_rate, 'e_impact_ionization': e_impact_ion, 'charge_exchange_HII': ch_exchange_hii, 'charge_exchange_HI': ch_exchange_hi} return f_oii_r, reaction_rates