Source code for cosmofit.likelihoods.union3
"""
Union3 binned supernova likelihood.
"""
from __future__ import annotations
import numpy as np
from cosmofit.data.loader import load_union3
from .base import BaseLikelihood, AnalyticOffsetMixin
[docs]
class Union3Likelihood(BaseLikelihood, AnalyticOffsetMixin):
"""
Union3 (Rubin et al. 2023): 2087 supernovae compressed into 22
binned distance moduli with a full 22x22 covariance.
The third of the three modern SN Ia compilations, alongside
:class:`~likelihoods.pantheon.PantheonLikelihood` and
:class:`~likelihoods.des_sn5yr.DESSN5YRLikelihood`. It matters
here because the DESI dark-energy results are argued with all
three, and they do not agree with each other on how far the
data sits from a cosmological constant -- DES-SN5YR pulls
hardest, Pantheon+ least, Union3 in between. A library that can
only fit one of them cannot reproduce that comparison, which is
the actual state of the evidence.
What makes Union3 different is not the sample so much as the
analysis: UNITY1.5 is a Bayesian hierarchical model that fits
light-curve standardization, host-galaxy mass dependence,
selection effects and outliers *jointly with* cosmology, and
marginalizes them internally. The released product is therefore
already a binned distance-modulus vector -- there is no
per-supernova catalogue to re-standardize, and no stretch/colour
nuisance parameters left for a downstream fit to vary.
As with both other SN samples, the overall zero point is
degenerate with H0 and is marginalized analytically (see
:class:`~likelihoods.base.AnalyticOffsetMixin`) rather than
fit -- which, for a 22-point dataset, matters more than it does
for a 1600-point one.
Parameters
----------
cosmology
Cosmology model instance (LCDM, CPL, ...).
version : str, optional
Dataset version.
marginalize_offset : bool, optional
If True (default), the constant distance-modulus offset is
marginalized over analytically. If False,
``cosmology.MB`` is added to the model as an explicit
nuisance parameter instead.
Warning
-------
Do not combine ``"union3"`` with ``"pantheon"`` or
``"des_sn5yr"`` in the same fit. Union3 shares a large majority
of its supernovae with Pantheon+ (both compile essentially the
same literature samples), and its high-redshift half overlaps
the DES sample; treating any two of the three as independent
double-counts most of the data. This is the same rule the
library already applies between Pantheon+ and DES-SN5YR --
one SN Ia compilation per fit.
References
----------
Rubin et al. (2023), "Union Through UNITY: Cosmology with
2,000 SNe Using a Unified Bayesian Framework",
arXiv:2311.12098 (ApJ, accepted).
"""
[docs]
def __init__(
self,
cosmology,
version: str = "union3",
marginalize_offset: bool = True,
):
dataset = load_union3(
version=version,
)
self.marginalize_offset = marginalize_offset
super().__init__(
name="Union3",
dataset=dataset,
cosmology=cosmology,
)
if self.marginalize_offset:
self._setup_offset_marginalization()
# ---------------------------------------------------------
@property
def observable(
self,
):
return "mu(z)"
# ---------------------------------------------------------
def model(
self,
) -> np.ndarray:
"""
Predicted binned distance modulus.
mu = 5 log10[(1 + z_hel) D_M(z_cmb)] + 25
The heliocentric/CMB-frame redshift split is the same one
:class:`~likelihoods.pantheon.PantheonLikelihood` and
:class:`~likelihoods.des_sn5yr.DESSN5YRLikelihood` apply:
the comoving distance is a function of the CMB-frame
redshift, while the ``(1 + z)`` source-frame dilation
factor in the luminosity distance is heliocentric.
Union3's release sets ``zhel = zcmb`` for every bin (the
bins average over many lines of sight, so there is no
single heliocentric correction left to apply), so the two
columns are numerically identical here -- the distinction
is kept anyway so the formula reads the same as the other
two SN likelihoods rather than silently relying on that.
"""
dm_model = self.cosmology.distance.DM(
self.data.z_cmb,
)
dl_model = dm_model * (1.0 + self.data.z_hel)
mu_model = 5.0 * np.log10(dl_model) + 25.0
if self.marginalize_offset:
return mu_model
return mu_model + self.cosmology.MB
# ---------------------------------------------------------
def residuals(
self,
) -> np.ndarray:
"""
Data minus model residuals.
"""
return (
self.data.mu
-
self.model()
)
# ---------------------------------------------------------
def chi2(
self,
) -> float:
"""
Chi-square statistic.
"""
if self.marginalize_offset:
return self.marginalized_chi2()
return self.covariance.chi2(
self.residuals(),
)