def make_sum_of_masses_plot(log_mass: bool = False):
""" Plot the posterior distribution for the sum of neutrino masses for two hierarchies and
a range of upper bounds."""
data = load_neutrino_constraints()
for i, hierarchy in enumerate(Hierarchy):
for j, upper_bound in enumerate(SUM_OF_MASSES_ONE_SIGMA):
data.sum_of_masses_one_sigma = upper_bound
data.sum_of_masses_offset = SUM_OF_MASSES_OFFSET[j]
linestyle = LINESTYLES[j]
likeligrid = get_posterior(hierarchy, data, n_samples=N_SAMPLES)
y = likeligrid.mass_posterior[:, 3]
if log_mass:
plt.plot(likeligrid.mass_log_bins, y, linestyle=linestyle, color=COLOURS[j])
else:
mass = likeligrid.mass_bins
y /= np.sum(y)
plt.plot(mass, y, linestyle=linestyle, color=COLOURS[j])
plt.xlim([0, 0.2])
plt.xlabel('$\Sigma_\\nu$ [eV]')
plt.ylabel('p($\Sigma_\\nu$ )')
print_figure('sum_of_masses_posteriors')
plt.show()
def make_mass_plot(log_mass: bool = False):
data = load_neutrino_constraints()
fig, axes = plt.subplots(nrows=1, ncols=2, sharex=True, sharey=False, figsize=(12, 6))
for i, hierarchy in enumerate(Hierarchy):
for j, upper_bound in enumerate(SUM_OF_MASSES_ONE_SIGMA):
data.sum_of_masses_one_sigma = upper_bound
data.sum_of_masses_offset = SUM_OF_MASSES_OFFSET[j]
linestyle = LINESTYLES[j]
likeligrid = get_posterior(hierarchy, data, n_samples=N_SAMPLES)
for m in range(3):
y = likeligrid.mass_posterior[:, m]
if log_mass:
axes[i].plot(likeligrid.mass_log_bins, y, linestyle=linestyle, color=COLOURS[m])
else:
mass = likeligrid.mass_bins
y /= np.sum(y)
axes[i].plot(mass, y, linestyle=linestyle, color=COLOURS[m])
axes[i].set_xlabel('m/eV')
axes[i].set_ylabel('p(m)')
axes[i].set_xlim([0, 0.09])
print_figure('mass_posteriors')
plt.show()
def make_mu_sigma_figures(sum_of_masses_one_sigma=0.0445):
""" Make plots for the two hierarchies with matching colour scheme. """
data = load_neutrino_constraints(sum_of_masses_one_sigma)
posterior_list = []
for i, hierarchy in enumerate(Hierarchy):
posterior = get_posterior(hierarchy, data, n_samples=N_SAMPLES)
posterior_list.append(posterior)
print_evidence(posterior_list)
make_plots(posterior_list, data.sum_of_masses_one_sigma)
def calculate_evidence_value(sum_of_masses_one_sigma: float, offset: float,
new_splittings: bool):
""" Get evidences for the two hierarchies. """
data = load_neutrino_constraints(sum_of_masses_one_sigma,
sum_of_masses_offset=offset,
new_splittings=new_splittings)
posterior_list = []
for i, hierarchy in enumerate(Hierarchy):
posterior = get_posterior(hierarchy, data)
posterior_list.append(posterior)
delta_chi2_boost = DELTA_CHI2_BOOSTS[
1] if new_splittings else DELTA_CHI2_BOOSTS[0]
print_evidence(posterior_list, delta_chi2_boost)
""" Estimate the model evidence using nested sampling. """
import dynesty as dy
import numpy as np
from inference.sampling import log_pdf
from neutrinos.constraints import load_neutrino_constraints
from neutrinos.hierarchies import Hierarchy
MAX_MASS = 1.0
DATA = load_neutrino_constraints()
HIERARCHY = Hierarchy.Normal
def prior_map(cube):
""" Our three masses have uniform priors so mapping of the prior quantile is trivial. """
return cube * MAX_MASS
def evaluate_log_likelihood_of_parameters(param_vector):
# Order the neutrino masses for ease of likelihood eval
if not np.all(np.diff(param_vector) >= 0):
return -1e100 # Faster to exclude unsorted region then compensate Z for missing volume. (factor of 6)
if HIERARCHY == Hierarchy.Normal:
msqr1 = param_vector[1]**2 - param_vector[
0]**2 # M ^ 2 - S ^ 2 ie m2 - m1
msqr2 = param_vector[2]**2 - param_vector[
1]**2 # L ^ 2 - M ^ 2 ie m3 - m2
import numpy as np
from inference.ultranest_log import run_ultranest
from neutrinos.constraints import load_neutrino_constraints
from neutrinos.hierarchies import Hierarchy
SUM_OF_MASSES_ONE_SIGMA = np.array([0.089, 0.12, 0.102]) * 0.5
SUM_OF_MASSES_OFFSET = [0, 0, 0] # Corresponding offsets
MAX_CALLS = 300_000
data = load_neutrino_constraints()
for hierarchy in [Hierarchy.Inverted]:
for j, upper_bound in enumerate(SUM_OF_MASSES_ONE_SIGMA):
data.sum_of_masses_one_sigma = upper_bound
data.sum_of_masses_offset = SUM_OF_MASSES_OFFSET[j]
run_ultranest(hierarchy, data, MAX_CALLS)