1 - https://academic.oup.com/mnrasl/article/537/1/L55/7926647?login=false
ABSTRACT
We present a new, cosmologically model-independent, statistical analysis of the Pantheon
1 INTRODUCTION
The
We perform an empirical cosmologically independent analysis within which both the
A key ingredient of the timescape model is a particular integrability relation for the Buchert equations: the uniform quasi-local Hubble expansion condition. Physically, it is motivated by an extension of Einstein’s Strong Equivalence Principle to cosmological averages at small scales (
In standard cosmology, differences from average FLRW expansion are assumed to be mostly attributed to local Lorentz boosts – i.e. peculiar velocities – of source and observer, with gravitational potentials contributing fractional variations of
Both the standard cosmology and the timescape model agree empirically on a Statistical Homogeneity Scale (SHS), typically given as
To conduct our analysis, we use the largest spectroscopically confirmed SNe Ia data set, Pantheon
where
The observational distance modulus from equation (1) is then compared with the theoretical distance modulus, given by
which is determined using the bolometric flux. The luminosity distance,
As noted in Lane et al. (2024), we omit peculiar velocity corrections. These are typically made using FLRW geometry assumptions, making it impossible to include them while preserving model-independence, or to perform a fair comparison. However, as distinctions between peculiar motion and expansion are central to the further development of timescape, the inclusion of such corrections will be addressed in future work. We would expect such corrections to have a small impact for low-redshift data cuts and negligible impact for
Lane et al. (2024) already presented moderate preference in favour of the timescape model over
2 STATISTICAL ANALYSIS
We determine Bayes factors, B, using the standard Jeffrey’s scale (Kass & Raftery 1995) for model comparison, whereby
Bayesian statistics have already been implemented on SNe Ia data for cosmological analysis, originally in the SDSS one-year sample (Kessler et al. 2009; March et al. 2011) but later extended to the Joint Light curve Analysis (Betoule et al. 2014) sample (Nielsen, Guffanti & Sarkar 2016; Dam et al. 2017) and more recently in the Pantheon
The previous studies implemented a Bayesian hierarchical likelihood construction in the form
where the quantities which are denoted with a hat are considered to be observed values, the true values are the quantities not denoted by a hat, and N is the number of supernovae observations. The true data represents the intrinsic parameters utilised explicitly in the Tripp (Tripp 1998) relation.
Nielsen et al. (2016), Dam et al. (2017), and Lane et al. (2024) follow the analysis of March et al. (2011) and adopt global, independent Gaussian distributions for
To overcome the faulty assumptions of the previous analyses, a full non-Gaussian modelling of the joint distribution for
where
Equivalently, given a single-shot inference for any physical quantity, the best guess for its true value is precisely the one inferred through the observational procedure. The assumption of being the most probable value introduces a caveat that it may, however, potentially overlook astrophysical systematics inherent in the SALT2 light-curve procedure.
Our approach here has essential differences from previous methodology (Nielsen et al. 2016; Dam et al. 2017; Lane et al. 2024), and is not merely a change of priors. Earlier work assumed that all supernovae are drawn from ideal independent Gaussian distributions in stretch (
Therefore, by now following the same procedure as in Lane et al. (2024), we find the likelihood to be
where the distributional error matrix (D) is the block-diagonal matrix with each block defined as
Similarly to Dam et al. (2017) and Lane et al. (2024) we utilize a nested Bayesian sampler PyMultiNest (Buchner et al. 2014), which interacts with the MultiNest (Feroz & Hobson 2008; Feroz, Hobson & Bridges 2009; Feroz et al. 2019) code to compare the spatially flat
Bayesian and frequentist priors on parameters used in the analysis. All priors are uniform on the respective intervals and, importantly, relatively broad for both models to ensure fair comparison.
| Parameter | Priors |
|---|---|
| [0.500,0.799] ( | |
| [0.143,0.487] ( | |
| [0,1] | |
| [0,7] | |
| [ | |
| c | [ |
| [ | |
Finally, in our analysis we reconstruct the
3 RESULTS
Results for the Bayes factor, cosmological and light-curve parameters are shown in Fig. 1.
The Bayes factors and Bayesian Maximum Likelihood Estimate (MLE) parameters for the fitting parameters across different redshift cuts, with Bayes factor uncertainties too small to display in the plot. The top plot shows the Bayes factors, where the upper section (
The Bayesian comparisons are best understood by splitting the minimum redshift cutoff used into three regimes: (i) for
Since timescape’s uniform quasi-local Hubble expansion condition holds down to scales
In comparing two models with different assumptions in the non-linear regime, the redshift distribution of the data becomes particularly important. For example, Lane et al. (2024) found consistent weak preference in favour of timescape using the P+580 subsample in which data from the full sample is truncated at high and low redshifts. While the evidence for the P+1690 sample changes significantly of order
The difference in the Bayes factors for the full P+1690 sample and the P+580 subsample between Lane et al. (2024) and our results. For the subsample, the results from the new analysis presented here align very well with the results by Lane et al. (2024), while for the full sample the new analysis greatly increases the preference in favour of timescape.
The Bayes factors and Bayesian Maximum Likelihood Estimate (MLE) parameters for different redshift cuts are shown. The top panel shows Bayes factors with blue indicating preference for timescape, red for
Lane et al. (2024) introduced an additional empirical data-driven notion of statistical homogeneity, defining SHS
The convergence of the
The Bayesian analysis can be used to find the MLE of the parameters, including the single free cosmological parameter. For
In the case of timescape, we find a void fraction of,
The change in
In the beams with Bias Correction method (Kessler & Scolnic 2017) a galaxy host correction is introduced with an additional parameter,
We examined including this term but found that it does not affect the Bayes factor conclusions, with an average offset of
4 DISCUSSION AND CONCLUSIONS
We performed a new Bayesian statistical analysis on the Pantheon
Our new analysis makes fewer assumptions about any particular statistical distribution of the data. Specifically, the likelihood function is constructed directly from the
Any astrophysical or environmental biases would likely impact both cosmological models. Thus the strong preference for timescape would require an extremely subtle combination of such biases for this to be its prime cause. The largest systematic error in the Pantheon
Since timescape has the same number of free parameters as spatially flat
The results presented in this Letter indicate that the timescape cosmology is not only a viable contender to the
Regardless of what model cosmology is to be the standard in future, exploring more than one model is important. Indeed, the timescape framework is consistent with new analysis of void statistics in numerical relativity simulations using the full Einstein equations (Williams et al. 2024). These are consistent with an emerging kinetic spatial curvature of voids on small scales. Much remains to be done in calibrating the dark matter fraction, primordial sound speed and the BAO scale. However, new results are likely to provide a robust framework for this (Galoppo & Wiltshire 2024; Galoppo, Re & Wiltshire 2024).
Our results imply profound consequences for cosmology and astrophysics. Indeed, a net preference for the timescape cosmology over the standard FLRW cosmologies may point to a need for revision of the foundations of theoretical cosmology, both ontologically and epistemologically, to better understand inhomogeneities and their backreaction on the average evolution of the Universe.
ACKNOWLEDGEMENTS
DLW, RRH, and ZGL are supported by the Marsden Fund administered by the Royal Society of New Zealand, Te Apārangi under grants M1271 and M1255. RRH is also supported by Rutherford Foundation Postdoctoral Fellowship RFT-UOC2203-PD. We are indebted to the anonymous referee of Lane et al. (2024) for suggesting the analysis framework. We also thank the anonymous referee of this Letter for their constructive and insightful comments. We thank all members of the University of Canterbury Gravity and Cosmology and Astrophysics groups for stimulating discussions, particularly: John Forbes, Christopher Harvey-Hawes, Morag Hills, Emma Johnson, Shreyas Tiruvaskar, Michael Williams, and Manon van Zyl. Finally, we wish to thank Elena Moltchanova for her precious insights into the statistical methods employed.
DATA AVAILABILITY
A complete set of the codes and details used for our analysis and how to use them can be found at Seifert & Lane (2023), and the covariance and input files are made available at Lane & Seifert (2024).
Footnotes
For Pantheon
See Dam et al. (2017, appendix A) for detailed comparisons of luminosity distance calculations in the timescape and FLRW models.
The principal small difference occurs in the geometric homogeneous Eddington bias (McKay 2016), leading to the potential for future tests.
The Lane et al. (2024) value
The
The relative contributions of Hubble constant uncertainty and absolute magnitude uncertainty, respectively
A further reason for not including galaxy-host corrections is the observation that the
REFERENCES
Journal Article
Supernovae evidence for foundational change to cosmological models Open Access
Antonia Seifert , Zachary G Lane , Marco Galoppo , Ryan Ridden-Harper , David L Wiltshire
Monthly Notices of the Royal Astronomical Society: Letters, Volume 537, Issue 1, February 2025, Pages L55–L60, https://doi.org/10.1093/mnrasl/slae112
Published: 19 December 2024 Article history
pdfPDF
Split View
Cite
Permissions Icon Permissions
Share Icon Share
ABSTRACT
We present a new, cosmologically model-independent, statistical analysis of the Pantheon
Type Ia Supernovae spectroscopic data set, improving a standard methodology adopted by Lane et al. We use the Tripp equation for supernova standardization alone, thereby avoiding any potential correlation in the stretch and colour distributions. We compare the standard homogeneous cosmological model, i.e. spatially flat
cold dark matter (
CDM), and the timescape cosmology which invokes backreaction of inhomogeneities. Timescape, while statistically homogeneous and isotropic, departs from average Friedmann–Lemaître–Robertson–Walker evolution, and replaces dark energy by kinetic gravitational energy and its gradients, in explaining independent cosmological observations. When considering the entire Pantheon
sample, we find very strong evidence (
) in favour of timescape over
CDM. Furthermore, even restricting the sample to redshifts beyond any conventional scale of statistical homogeneity,
, timescape is preferred over
CDM with
. These results provide evidence for a need to revisit the foundations of theoretical and observational cosmology.
gravitation, supernovae: general, cosmological parameters, dark energy, cosmology: theory
Issue Section: Letter
1 INTRODUCTION
The
cold dark matter (
CDM) model, which has served as the standard cosmological model for quarter of a century, is facing serious challenges in light of recent results (Abbott et al. 2024; Adame et al. 2024) and may need to be reconsidered at a fundamental level (Di Valentino et al. 2021; Peebles 2022; Aluri et al. 2023). In this Letter, we present definite statistical evidence that the timescape cosmological model (Wiltshire 2007a, b, 2009) outperforms
CDM in matching Type Ia Supernovae (SNe Ia) observations. It may provide not only a viable alternative to the standard cosmological model, but ultimately a preferred one. This result potentially has far-reaching consequences not only for cosmology, but also for other key aspects of astrophysical modelling from late epochs to the early universe.
We perform an empirical cosmologically independent analysis within which both the
CDM and timescape cosmologies may be embedded, and thus compared via Bayesian statistics. The timescape model is a particular implementation of Buchert’s scalar averaging scheme which incorporates backreaction of inhomogeneities (Buchert 2000, 2001; Wiltshire 2014; Buchert, Mourier & Roy 2020). Instead of a matter density parameter relative to average Friedmann–Lemaître–Robertson–Walker model (as in
CDM), timescape is characterized by the void fraction,
, which represents the fractional volume of the expanding regions of the universe made up by voids.
A key ingredient of the timescape model is a particular integrability relation for the Buchert equations: the uniform quasi-local Hubble expansion condition. Physically, it is motivated by an extension of Einstein’s Strong Equivalence Principle to cosmological averages at small scales (
–
Mpc) where perturbations to average isotropic expansion and average isotropic motion cannot be observationally distinguished (Wiltshire 2008).
In standard cosmology, differences from average FLRW expansion are assumed to be mostly attributed to local Lorentz boosts – i.e. peculiar velocities – of source and observer, with gravitational potentials contributing fractional variations of
of average expansion at galaxy and galaxy cluster scales. In timescape, the same fractional variation can be up to
and the equivalence of different choices of background, via the Cosmological Equivalence Principle, means that notions of average isotropic expansion persist well into the non-linear regime of structure formation. The signature of the emergent kinetic spatial curvature of voids has now been identified in cosmological simulations using full numerical general relativity without
(Williams et al. 2024).
Both the standard cosmology and the timescape model agree empirically on a Statistical Homogeneity Scale (SHS), typically given as
by the two-point galaxy correlation function (Hogg et al. 2005; Scrimgeour et al. 2012; Dam, Heinesen & Wiltshire 2017). Timescape offers its most important tests and predictions below the SHS, at scales where the filaments, sheets and voids of the cosmic web are still expanding but in the non-linear regime.
To conduct our analysis, we use the largest spectroscopically confirmed SNe Ia data set, Pantheon
(Scolnic et al. 2022). SNe Ia have been a pillar for informing the distance ladder used for cosmological model comparison, and have a rich history in revolutionising the field (Riess et al. 1998; Perlmutter et al. 1999). More modern methods for standardizing SNe Ia light curves use the SALT2 fitting algorithm (Guy et al. 2007; Taylor et al. 2021), as used by Pantheon
, and more recently SALT3 (Kenworthy et al. 2021) used by the Dark Energy Survey 5-year release (DES5yr; Abbott et al. 2024). The SALT fitting algorithms fit the distance moduli,
, using a modified version of the Tripp formula:
(1)
where
and
are considered constant across all redshifts1,
is the time stretch/decay parameter, c is the colour, and
and
are the apparent and absolute magnitude in the rest frame of the B band filter. Rest-frame measurements are identical for theories obeying the Strong Equivalence Principle of general relativity – in particular, in both the FLRW and timescape models. In our analysis,
, c, and
are taken directly from the Pantheon
data.
The observational distance modulus from equation (1) is then compared with the theoretical distance modulus, given by
(2)
which is determined using the bolometric flux. The luminosity distance,
, can be calculated using the redshift of the supernovae and suitable cosmological model parameters. Typically, these are
for the spatially flat
CDM model and
for the timescape cosmology.2 Thus, the distance modulus constitutes the pillar of cosmological model comparison via supernovae analysis.
As noted in Lane et al. (2024), we omit peculiar velocity corrections. These are typically made using FLRW geometry assumptions, making it impossible to include them while preserving model-independence, or to perform a fair comparison. However, as distinctions between peculiar motion and expansion are central to the further development of timescape, the inclusion of such corrections will be addressed in future work. We would expect such corrections to have a small impact for low-redshift data cuts and negligible impact for
taken within a statistically homogeneous regime (Carr et al. 2022). Furthermore, for the same reasons we do not include other cosmological model and metric-dependent bias corrections, such as Malmquist biases. Such corrections are small and cannot drive any substantial changes to the Bayes factors since the trend with redshift is expected to be very similar3 in both
CDM and timescape.
Lane et al. (2024) already presented moderate preference in favour of the timescape model over
CDM. A similar result was also obtained by the DES team, with
, using the Akaike Information Criterion on the DES5yr supernovae sample (Camilleri et al. 2024). They further noted a change from
(in favour of timescape) to
(in favour of spatially flat
CDM), when SNe Ia data were combined with Baryonic Acoustic Oscillation (BAO) measurements. However, the BAO analysis of Camilleri et al. (2024) assumes purely geometric adjustments to the standard FLRW pipeline, using a
CDM calibration of the BAO drag epoch, which is not the case in timescape. Incorporating detailed BAO analysis into the timescape cosmology requires extraction of the BAO from galaxy clustering statistics, which has already been implemented (Heinesen et al. 2019). However, since the ratio of baryonic matter to non-baryonic dark matter is different from
CDM, matter model calibrations in the early universe must also be revisited.
2 STATISTICAL ANALYSIS
We determine Bayes factors, B, using the standard Jeffrey’s scale (Kass & Raftery 1995) for model comparison, whereby
indicates no statistical preference,
moderate preference, while
and
represent strong and very strong preference respectively. In this Letter, positive (negative)
values indicate a preference for the timescape (spatially flat
CDM) model.
Bayesian statistics have already been implemented on SNe Ia data for cosmological analysis, originally in the SDSS one-year sample (Kessler et al. 2009; March et al. 2011) but later extended to the Joint Light curve Analysis (Betoule et al. 2014) sample (Nielsen, Guffanti & Sarkar 2016; Dam et al. 2017) and more recently in the Pantheon
(Brout et al. 2022a, b; Scolnic et al. 2022) data set (Lane et al. 2024).
The previous studies implemented a Bayesian hierarchical likelihood construction in the form
(3)
where the quantities which are denoted with a hat are considered to be observed values, the true values are the quantities not denoted by a hat, and N is the number of supernovae observations. The true data represents the intrinsic parameters utilised explicitly in the Tripp (Tripp 1998) relation.
Nielsen et al. (2016), Dam et al. (2017), and Lane et al. (2024) follow the analysis of March et al. (2011) and adopt global, independent Gaussian distributions for
,
, and c to determine the probability density of the true parameters. However, both of these simplifying assumptions are ultimately flawed. Indeed, (i) the true values of
and c are expected to be highly correlated as these are effective parameters obtained by coarse-graining the highly complex processes behind supernovae explosions; (ii) both the distributions of
and c present strong non-Gaussian features that cannot be explained away by systematics or biases in the data. Whilst the former always represented an overly simplifying assumption, the latter was a reasonable assumption when it was first implemented, however, the vast increases in observed SNe Ia have shown the second assumption to be flawed (Hinton et al. 2019).
To overcome the faulty assumptions of the previous analyses, a full non-Gaussian modelling of the joint distribution for
and c would be required. This represents non-trivial changes in the likelihood construction and integration, which will be addressed in future work (in prep.). Therefore, in this Letter, we propose an alternative approach to sidestep the issue. Our new approach builds upon the Bayesian hierarchical likelihood construction method by directly seeding the priors of
and c with the inferred values from the SALT2 fitting algorithm (Guy et al. 2005, 2007; Taylor et al. 2021). Specifically, we define the priors over the true values for each supernovae as
(4)
where
is a normal distribution with mean value
and variance
, and
is the Dirac delta distribution. Thus, the prior distribution in
is common to all the supernovae data, while the priors in
and c are supernovae specific. Therefore, our new approach sidesteps the problem of modelling the joint distribution, only requiring five parameters (a cosmological parameter,
,
,
, and
), by assuming that the SALT2 parameters represent the ‘true’ parameters, i.e. the most probable values for both
and c for this version of the SALT model.
Equivalently, given a single-shot inference for any physical quantity, the best guess for its true value is precisely the one inferred through the observational procedure. The assumption of being the most probable value introduces a caveat that it may, however, potentially overlook astrophysical systematics inherent in the SALT2 light-curve procedure.
Our approach here has essential differences from previous methodology (Nielsen et al. 2016; Dam et al. 2017; Lane et al. 2024), and is not merely a change of priors. Earlier work assumed that all supernovae are drawn from ideal independent Gaussian distributions in stretch (
) and in colour (c), with mean values and standard deviations derived from the cosmological fit. In contrast, this study does not assume any particular statistical distribution for
and c, nor do we assume these parameters follow the same ideal distribution across the supernova sample. Instead,
and c are treated as fixed, with values provided by the SALT2 fit. Taylor et al. (2021) show through simulations that SALT2 reliably recovers input supernova parameters. To compare this method with the previous one, we use the same data set as Lane et al. (2024).
Therefore, by now following the same procedure as in Lane et al. (2024), we find the likelihood to be
(5)
where the distributional error matrix (D) is the block-diagonal matrix with each block defined as
,
is the
statistical and systematic covariance matrix given by Lane et al. (2024, section 2), and the residual vector X is defined by
(6)
Similarly to Dam et al. (2017) and Lane et al. (2024) we utilize a nested Bayesian sampler PyMultiNest (Buchner et al. 2014), which interacts with the MultiNest (Feroz & Hobson 2008; Feroz, Hobson & Bridges 2009; Feroz et al. 2019) code to compare the spatially flat
CDM and timescape models with a tolerance of
and
for nine parameters. We choose the same priors as Lane et al. (2024, table B2 & section 3) summarized in Table 1.
Table 1.Open in new tabBayesian and frequentist priors on parameters used in the analysis. All priors are uniform on the respective intervals and, importantly, relatively broad for both models to ensure fair comparison.
Parameter Priors
[0.500,0.799] (
bound)
[0.143,0.487] (
bound)
[0,1]
[0,7]
[
20,20]
c [
20,20]
[
20.3,18.3]
[
10,4]
Finally, in our analysis we reconstruct the
by applying a boost (Fixsen et al. 1996) to the Pantheon + heliocentric redshifts, excluding peculiar velocity corrections. We then remove all supernovae with
for varying redshift cuts
and fit the cosmological model to the remaining supernova events. This allows us to examine how the Bayes factor, cosmological parameters, and Tripp parameters vary across different redshift regimes.
3 RESULTS
Results for the Bayes factor, cosmological and light-curve parameters are shown in Fig. 1.
The Bayes factors and Bayesian Maximum Likelihood Estimate (MLE) parameters for the fitting parameters across different redshift cuts, with Bayes factor uncertainties too small to display in the plot. The top plot shows the Bayes factors, where the upper section ($\ln B > 1$) favours timescape, the unshaded section favours neither hypothesis and the lower section ($\ln B < -1$) favours $\Lambda$CDM. The following plots show the various MLE parameter estimates, with values beyond SHS$_\alpha = 0.054^{+0.007}_{-0.012}$ indicated by the dashed vertical line.
Figure 1.The Bayes factors and Bayesian Maximum Likelihood Estimate (MLE) parameters for the fitting parameters across different redshift cuts, with Bayes factor uncertainties too small to display in the plot. The top plot shows the Bayes factors, where the upper section (
) favours timescape, the unshaded section favours neither hypothesis and the lower section (
) favours
CDM. The following plots show the various MLE parameter estimates, with values beyond SHS
indicated by the dashed vertical line.
Open in new tabDownload slide
The Bayesian comparisons are best understood by splitting the minimum redshift cutoff used into three regimes: (i) for
we find very strong to strong evidence on the Jeffrey’s scale (Kass & Raftery 1995) in favour of timescape over
CDM; (ii) for
we enter the calibration regime,4 finding moderate to no significant preference for timescape; (iii) for
, beyond any measure of a SHS5, we find exclusively moderate preference for the timescape cosmology. Notably, the log-evidence,
, values found here for both models are
greater compared to the previous analysis by Lane et al. (2024).
Since timescape’s uniform quasi-local Hubble expansion condition holds down to scales
–
Mpc, as we decrease
an increase in the Bayesian evidence favouring timescape is expected if the model accurately captures the average cosmic expansion deep in the non-linear regime of structure formation. Beyond the SHS,
CDM of course provides an excellent description of our Universe. However, the evidence in favour of timescape remains small but modest (
) at the highest redshift cuts,
, pointing to the ability of the model to describe the Universe’s expansion history on scales greater than the SHS. This moderate evidence (
) can be interpreted as resulting from the integrated effects across the redshift range
, reflecting the 1–3 per cent variations in the expansion history between timescape and
CDM.
In comparing two models with different assumptions in the non-linear regime, the redshift distribution of the data becomes particularly important. For example, Lane et al. (2024) found consistent weak preference in favour of timescape using the P+580 subsample in which data from the full sample is truncated at high and low redshifts. While the evidence for the P+1690 sample changes significantly of order
–
in our revised analysis, the P+580 subsample result remains consistent (Fig. 2). The discrepancy between the results of the full data set and the subsample suggest the need for further analysis on how the redshift distribution of supernovae, and the probed redshift range impact evidence for cosmological models. The uncertainty in the Bayes factor,
, is so small that it does not influence the Jeffrey’s scale classifications or the conclusions drawn.
The difference in the Bayes factors for the full P+1690 sample and the P+580 subsample between Lane et al. (2024) and our results. For the subsample, the results from the new analysis presented here align very well with the results by Lane et al. (2024), while for the full sample the new analysis greatly increases the preference in favour of timescape.
Figure 2.The difference in the Bayes factors for the full P+1690 sample and the P+580 subsample between Lane et al. (2024) and our results. For the subsample, the results from the new analysis presented here align very well with the results by Lane et al. (2024), while for the full sample the new analysis greatly increases the preference in favour of timescape.
Open in new tabDownload slide
The Bayes factors and Bayesian Maximum Likelihood Estimate (MLE) parameters for different redshift cuts are shown. The top panel shows Bayes factors with blue indicating preference for timescape, red for
CDM, and white for neither. The subsequent plots show MLE parameter estimates, with values beyond the scale of the statistical homogeneity (SHS) marked by the dashed vertical line.
Lane et al. (2024) introduced an additional empirical data-driven notion of statistical homogeneity, defining SHS
from a power law fitted to the
degenerate parameter. The analogous SHS
defined from
does not yield a true convergence for the analysis by Lane et al. (2024), nor for this analysis, due to Malmquist bias not being accounted for. While the SHS
appears to converge below
, this is not the case for higher redshift cuts. For the reanalysis presented in Fig. 3 we find SHS
, which is 1.2
greater than the maximum value of the SHS gathered from the two-point galaxy correlation function (Hogg et al. 2005; Scrimgeour et al. 2012) and somewhat lower but within 2.3
of the result Lane et al. (2024) determined. The differences with respect to the analysis in Lane et al. (2024) derive from the lifting of the Gaussian assumption of the underlying distributions of
and c.
The convergence of the $\alpha x_{\lower2pt\hbox{$\scriptstyle 1$}}$ light-curve parameter for the spatially flat $\Lambda$CDM model across various redshift cuts, where $x _{\lower2pt\hbox{$\scriptstyle 1$}}$ is the median value from the distribution. A power-law model has been fit to the data, and the shaded band represents within 5 per cent of the median value within the range $0.1 \le z_{\rm min}\lt 0.14$ indicating when the model converges. The vertical dotted line represents the SHS$_\alpha$ found at $z_{\rm min}= 0.054^{+0.007}_{-0.012}$. The power-law uncertainty is smaller than the plotted line.
Figure 3.The convergence of the
light-curve parameter for the spatially flat
CDM model across various redshift cuts, where
is the median value from the distribution. A power-law model has been fit to the data, and the shaded band represents within 5 per cent of the median value within the range
indicating when the model converges. The vertical dotted line represents the SHS
found at
. The power-law uncertainty is smaller than the plotted line.
Open in new tabDownload slide
The Bayesian analysis can be used to find the MLE of the parameters, including the single free cosmological parameter. For
cuts beyond the SHS
, (
), for
CDM we find
, within 1.2
of the value found from the DES5yr release (Abbott et al. 2024), and just outside of 2
of Pantheon
(Brout et al. 2022a).6
In the case of timescape, we find a void fraction of,
, within 2
of the Camilleri et al. (2024) DES5yr value. Significantly, our
value is also within 2
of independent values predicted from the Planck CMB power spectrum,
(Duley, Nazer & Wiltshire 2013); and well within 1
of strong gravitational lensing distance ratios,
, (Harvey-Hawes & Wiltshire 2024). We also find the evolution of the Tripp constants,
,
, and
with varying
cuts following Dam et al. (2017), Lane et al. (2024). To avoid the underlying degeneracy between
and
, we fix
for both models as a nuisance parameter.7 Moreover, although the values for the individual parameters differ between the two statistical methods, the Tripp distance modulus,
, changes on average by only
for redshift cuts beyond
. This variation is observed when comparing the median values of
and c in the Tripp methodology, to the general Gaussian distribution fit values.
The change in
between this work and Lane et al. (2024) is thus not statistically significant in this regime. However, it is expected that differences in the prior distribution cause differences in the fitted parameters. This behaviour will be investigated further for supernovae statistics built on skewed, non-Gaussian distributions in future work (in prep.).
In the beams with Bias Correction method (Kessler & Scolnic 2017) a galaxy host correction is introduced with an additional parameter,
, defined by the mass-step
(7)
We examined including this term but found that it does not affect the Bayes factor conclusions, with an average offset of
compared to the uncorrected value. Furthermore, the statistical cost of introducing additional free parameters can be assessed by the relative Bayesian Information Criterion (BIC) statistic (Schwarz 1978; Kass & Raftery 1995)
for k free parameters, a sample size, N, and likelihood Z. We find that independent of cosmology the model with a mass step is strongly disfavoured relative to the uncorrected Tripp model, with
at
. Furthermore, there is no significant change in the value of the cosmological parameter, with
, which is well within the 1
range of our statistical and systematic uncertainties. Thus our final results are stated without galaxy host corrections.8
4 DISCUSSION AND CONCLUSIONS
We performed a new Bayesian statistical analysis on the Pantheon
supernovae data set, accounting for the non-Gaussian
and c features of the supernovae parameter distributions. The Bayesian evidence yields very strong to strong evidence for the timescape model in the low-redshift regime. This late-universe result could be expected, as the timescape models accounts for non-kinematic differential expansion on scales
where the local inhomogeneous structure of our nearby cosmic web most impacts measurements. On the other hand, for samples strongly weighted by SNe Ia in the calibration regime of the
CDM model (
) there is no significant preference either way, the two models being statistically equivalent. With a restriction to higher redshifts, well beyond any scale of statistical homogeneity generally accepted (Lane et al. 2024), Bayesian evidence is driven once again in favour of timescape.
Our new analysis makes fewer assumptions about any particular statistical distribution of the data. Specifically, the likelihood function is constructed directly from the
and c values obtained using the SALT2 algorithm – values employed in most SNe Ia analyses. The empirical SNe Ia data obtained via the cosmology independent SALT2 fit strongly favours the timescape model over
CDM.
Any astrophysical or environmental biases would likely impact both cosmological models. Thus the strong preference for timescape would require an extremely subtle combination of such biases for this to be its prime cause. The largest systematic error in the Pantheon
analysis is the standardization of the heterogeneous mix of low-z sample light curves (Abbott et al. 2024; Lane et al. 2024). Future improvements with the new DES5yr sample (Abbott et al. 2024) will allow for a more homogeneous and careful selection of the low-z sample. However, in this Letter we concentrate on the impact of the new statistical method on cosmological model selection, and therefore we use the same data as Lane et al. (2024).
Since timescape has the same number of free parameters as spatially flat
CDM, Bayesian evidence offers the best comparison. To expand our results to include other popular FLRW-type alternative cosmological models, which contain more parameters, e.g. wCDM, we determine the BIC statistic (Schwarz 1978; Kass & Raftery 1995) for fair model comparison. For the full sample, we find that relative to timescape
CDM models with FLRW curvature are very strongly disfavoured with
, while wCDM is also very strongly disfavoured with
.
The results presented in this Letter indicate that the timescape cosmology is not only a viable contender to the
CDM framework, but may also provide new insights to the astrophysics of modelling SNe Ia. Timescape’s non-FLRW average evolution reveals degeneracies between cosmological parameters and empirical SNe Ia model parameters that were already partly uncovered in earlier work (Dam et al. 2017) but which are striking with Pantheon
, as shown by Lane et al. (2024) and the present Letter.
Regardless of what model cosmology is to be the standard in future, exploring more than one model is important. Indeed, the timescape framework is consistent with new analysis of void statistics in numerical relativity simulations using the full Einstein equations (Williams et al. 2024). These are consistent with an emerging kinetic spatial curvature of voids on small scales. Much remains to be done in calibrating the dark matter fraction, primordial sound speed and the BAO scale. However, new results are likely to provide a robust framework for this (Galoppo & Wiltshire 2024; Galoppo, Re & Wiltshire 2024).
Our results imply profound consequences for cosmology and astrophysics. Indeed, a net preference for the timescape cosmology over the standard FLRW cosmologies may point to a need for revision of the foundations of theoretical cosmology, both ontologically and epistemologically, to better understand inhomogeneities and their backreaction on the average evolution of the Universe.
ACKNOWLEDGEMENTS
DLW, RRH, and ZGL are supported by the Marsden Fund administered by the Royal Society of New Zealand, Te Apārangi under grants M1271 and M1255. RRH is also supported by Rutherford Foundation Postdoctoral Fellowship RFT-UOC2203-PD. We are indebted to the anonymous referee of Lane et al. (2024) for suggesting the analysis framework. We also thank the anonymous referee of this Letter for their constructive and insightful comments. We thank all members of the University of Canterbury Gravity and Cosmology and Astrophysics groups for stimulating discussions, particularly: John Forbes, Christopher Harvey-Hawes, Morag Hills, Emma Johnson, Shreyas Tiruvaskar, Michael Williams, and Manon van Zyl. Finally, we wish to thank Elena Moltchanova for her precious insights into the statistical methods employed.
DATA AVAILABILITY
A complete set of the codes and details used for our analysis and how to use them can be found at Seifert & Lane (2023), and the covariance and input files are made available at Lane & Seifert (2024).
Footnotes
1
For Pantheon
, Scolnic et al. (2022) adopt values of
and
, respectively, for their nominal fit.
2
See Dam et al. (2017, appendix A) for detailed comparisons of luminosity distance calculations in the timescape and FLRW models.
3
The principal small difference occurs in the geometric homogeneous Eddington bias (McKay 2016), leading to the potential for future tests.
4
This is the regime beyond which average homogeneity and isotropy are assumed to apply to all observations. Hogg et al. (2005); Scrimgeour et al. (2012) take this range as
–
(corresponding to a redshift range of approximately 0.023–0.04).
5
The Lane et al. (2024) value
is larger than other estimates and thus gives a robust upper bound for the SHS.
6
The
and
values reported by Scolnic et al. (2022) and Lane et al. (2024) are derived at various stages of the cosmological fitting pipeline, and are influenced by the specific subsample used (Lane et al. 2024). Any slight differences in cosmological parameters can be attributed to these methodological variations and to the omission of cosmology dependent bias corrections.
7
The relative contributions of Hubble constant uncertainty and absolute magnitude uncertainty, respectively
and
, propagate according to
This makes the two contributions impossible to unravel and explains the larger uncertainty,
, relative to uncertainties in other parameters from the fitting (see Fig. 1).
8
A further reason for not including galaxy-host corrections is the observation that the
parameter exhibits inconsistent behaviour across different redshift cuts for a simple mass-step function. This inconsistency most likely arises from the heterogeneous subsamples of low-z data, as
is well constrained in a more statistically homogeneous sample. It is possible, but less likely, that these fluctuations arise from other astrophysical factors explored by the DES5yr team and recent studies (Dixon et al. 2024), but these are beyond the scope of this Letter.
REFERENCES
Abbott T. M. C. et al. , 2024, ApJ, 973, L14 10.3847/2041-8213/ad6f9f
Crossref
Adame A. G. et al. , 2024, preprint (arXiv)
Aluri P. K. et al. , 2023, Class. Quantum Gravity, 40, 094001 10.1088/1361-6382/acbefc
Crossref
Betoule M. et al. , 2014, A&A, 568, A22 10.1051/0004-6361/201423413
Crossref
Brout D. et al. , 2022a, ApJ, 938, 110 10.3847/1538-4357/ac8e04
Crossref
Brout D. et al. , 2022b, ApJ, 938, 111 10.3847/1538-4357/ac8bcc
Crossref
Buchert T., 2000, Gen. Relativ. Gravit., 32, 105 10.1023/A:1001800617177
Crossref
Buchert T., 2001, Gen. Relativ. Gravit., 33, 1381 10.1023/A:1012061725841
Crossref
Buchert T., Mourier P., Roy X., 2020, Gen. Relativ. Gravit., 52, 27 10.1007/s10714-020-02670-6
Crossref
Buchner J. et al. , 2014, A&A, 564, A125 10.1051/0004-6361/201322971
Crossref
Camilleri R. et al. , 2024, MNRAS, 533, 2615 10.1093/mnras/stae1988
Crossref
Carr A., Davis T. M., Scolnic D., Said K., Brout D., Peterson E. R., Kessler R., 2022, Publ. Astron. Soc. Aust., 39, e046 10.1017/pasa.2022.41
Crossref
Dam L. H., Heinesen A., Wiltshire D. L., 2017, MNRAS, 472, 835 10.1093/mnras/stx1858
Crossref
Di Valentino E. et al. , 2021, Class. Quantum Gravity, 38, 153001 10.1088/1361-6382/ac086d
Crossref
Dixon M. et al. , 2024, preprint (arXiv)
Duley J. A. G., Nazer M. A., Wiltshire D. L., 2013, Class. Quantum Gravity, 30, 175006 10.1088/0264-9381/30/17/175006
Crossref
Feroz F., Hobson M. P., 2008, MNRAS, 384, 449 10.1111/j.1365-2966.2007.12353.x
Crossref
Feroz F., Hobson M. P., Bridges M., 2009, MNRAS, 398, 1601 10.1111/j.1365-2966.2009.14548.x
Crossref
Feroz F., Hobson M. P., Cameron E., Pettitt A. N., 2019, Open J. Astrophys., 2, 1 10.21105/astro.1306.2144
Crossref
Fixsen D. J., Cheng E. S., Gales J. M., Mather J. C., Shafer R. A., Wright E. L., 1996, ApJ, 473, 576 10.1086/178173
Crossref
Galoppo M., Wiltshire D. L., 2024, preprint (arXiv)
Galoppo M., Re F., Wiltshire D. L., 2024, preprint (arXiv)
Guy J., Astier P., Nobili S., Regnault N., Pain R., 2005, A&A, 443, 781 10.1051/0004-6361:20053025
Crossref
Guy J. et al. , 2007, A&A, 466, 11 10.1051/0004-6361:20066930
Crossref
Harvey-Hawes C., Wiltshire D. L., 2024, MNRAS, 534, 3364 10.1093/mnras/stae2306
Crossref
Heinesen A., Blake C., Li Y.-Z., Wiltshire D. L., 2019, J. Cosmol. Astropart. Phys., 03, 003 10.1088/1475-7516/2019/03/003
Crossref
Hinton S. R. et al. , 2019, ApJ, 876, 15 10.3847/1538-4357/ab13a3
Crossref
Hogg D. W., Eisenstein D. J., Blanton M. R., Bahcall N. A., Brinkmann J., Gunn J. E., Schneider D. P., 2005, ApJ, 624, 54 10.1086/429084
Crossref
Kass R. E., Raftery A. E., 1995, J. Am. Stat. Assoc., 90, 773 10.1080/01621459.1995.10476572
Crossref
Kenworthy W. D. et al. , 2021, ApJ, 923, 265 10.3847/1538-4357/ac30d8
Crossref
Kessler R., Scolnic D., 2017, ApJ, 836, 56 10.3847/1538-4357/836/1/56
Crossref
Kessler R. et al. , 2009, ApJS, 185, 32 10.1088/0067-0049/185/1/32
Crossref
Lane Z. G., Seifert A., 2024, P+1690 Covariance Matrix, Zenodo. Available at: 10.5281/zenodo.12729746
Lane Z. G., Seifert A., Ridden-Harper R., Wiltshire D. L., 2024, MNRAS, 536, 1752 10.1093/mnras/stae2437
Crossref
McKay J. H., 2016, MSc thesis, Univ. Canterbury 10.26021/6882
March M. C., Trotta R., Berkes P., Starkman G. D., Vaudrevange P. M., 2011, MNRAS, 418, 2308 10.1111/j.1365-2966.2011.19584.x
Crossref
Nielsen J. T., Guffanti A., Sarkar S., 2016, Sci. Rep., 6, 35596 10.1038/srep35596
CrossrefPubMed
Peebles P. J. E., 2022, Ann. Phys., 447, 169159 10.1016/j.aop.2022.169159
Crossref
Perlmutter S. et al. , 1999, ApJ, 517, 565 10.1086/307221
Crossref
Riess A. G. et al. , 1998, AJ, 116, 1009 10.1086/300499
Crossref
Schwarz G., 1978, Ann. Stat., 6, 461 10.1214/aos/1176344136
Crossref
Scolnic D. et al. , 2022, ApJ, 938, 113 10.3847/1538-4357/ac8b7a
Crossref
Scrimgeour M. I. et al. , 2012, MNRAS, 425, 116 10.1111/j.1365-2966.2012.21402.x
Crossref
Seifert A., Lane Z. G., 2023, Code for Supernova Analysis with Pantheon+ by Lane et al. 2023, GitHub. Available at: https://github.com/antosft/SNe-PantheonPlus-Analysis
Google ScholarGoogle PreviewWorldCatCOPAC
Taylor G., Lidman C., Tucker B. E., Brout D., Hinton S. R., Kessler R., 2021, MNRAS, 504, 4111 10.1093/mnras/stab962
Crossref
Tripp R., 1998, A&A, 331, 815
Williams M. J., Macpherson H. J., Wiltshire D. L., Stevens C., 2024, preprint (arXiv)
Wiltshire D. L., 2007a, New J. Phys., 9, 377 10.1088/1367-2630/9/10/377
Crossref
Wiltshire D. L., 2007b, Phys. Rev. Lett., 99, 251101 10.1103/PhysRevLett.99.251101
Crossref
Wiltshire D. L., 2008, Phys. Rev. D, 78, 084032 10.1103/PhysRevD.78.084032
Crossref
Wiltshire D. L., 2009, Phys. Rev. D, 80, 123512 10.1103/PhysRevD.80.123512
Crossref
Wiltshire D. L., 2014, in Perez Bergliaffa S., Novello M., eds, Proceedings of the XVth Brazilian School of Cosmology and Gravitation. Cambridge Scientific Publishers, Cambridge, UK, p. 203
Google ScholarGoogle PreviewWorldCatCOPAC
© 2024 The Author(s). Published by Oxford University Press on behalf of Royal Astronomical Society.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
RESUMO
Apresentamos uma nova análise estatística do Panteão, cosmologicamente independente do modelo.
1 INTRODUÇÃO
O
Um ingrediente-chave do modelo de paisagem temporal é uma relação de integrabilidade particular para as equações de Buchert: a condição de expansão uniforme quase local de Hubble . Fisicamente, é motivado por uma extensão do Princípio da Equivalência Forte de Einstein para médias cosmológicas em pequenas escalas (
Na cosmologia padrão, supõe-se que as diferenças em relação à expansão média de FLRW sejam atribuídas principalmente aos impulsos locais de Lorentz – ou seja, velocidades peculiares – da fonte e do observador, com potenciais gravitacionais contribuindo com variações fracionárias de
Para conduzir nossa análise, usamos o maior conjunto de dados SNe Ia confirmado espectroscopicamente, Pantheon
O módulo de distância observacional da equação ( 1 ) é então comparado com o módulo de distância teórica, dado por
Conforme observado em Lane et al. ( 2024 ), omitimos correções de velocidade peculiar. Estas são tipicamente feitas usando suposições de geometria FLRW, tornando impossível incluí-las preservando a independência do modelo ou realizando uma comparação justa. No entanto, como as distinções entre movimento peculiar e expansão são centrais para o desenvolvimento posterior da paisagem temporal, a inclusão de tais correções será abordada em trabalhos futuros. Esperamos que tais correções tenham um impacto pequeno para cortes de dados de baixo desvio para o vermelho e um impacto insignificante para
2 ANÁLISE ESTATÍSTICA
Determinamos os fatores de Bayes, B , usando a escala padrão de Jeffrey (Kass & Raftery 1995 ) para comparação de modelos, onde
Estatísticas bayesianas já foram implementadas em dados SNe Ia para análise cosmológica, originalmente na amostra de um ano do SDSS (Kessler et al. 2009 ; March et al. 2011 ), mas posteriormente estendidas para a amostra da curva de luz conjunta (Betoule et al. 2014 ) (Nielsen, Guffanti & Sarkar 2016 ; Dam et al. 2017 ) e mais recentemente no Pantheon
Os estudos anteriores implementaram uma construção de verossimilhança hierárquica bayesiana na forma
onde as grandezas denotadas por um chapéu são consideradas valores observados, os valores verdadeiros são as grandezas não denotadas por um chapéu e N é o número de observações de supernovas. Os dados verdadeiros representam os parâmetros intrínsecos utilizados explicitamente na relação de Tripp (Tripp 1998 ).
Da mesma forma, dada uma inferência única para qualquer grandeza física, a melhor estimativa para seu valor verdadeiro é precisamente aquela inferida por meio do procedimento observacional. A suposição de ser o valor mais provável introduz a ressalva de que pode, no entanto, potencialmente ignorar a sistemática astrofísica inerente ao procedimento de curva de luz SALT2.
Portanto, seguindo agora o mesmo procedimento de Lane et al. ( 2024 ), encontramos a probabilidade de ser
| Parâmetro | Priores |
|---|---|
| [0,500,0,799] ( | |
| [0,143,0,487] ( | |
| [0,1] | |
| [0,7] | |
| [ | |
| c | [ |
| [ | |
3 RESULTADOS
Os resultados para o fator de Bayes, parâmetros cosmológicos e de curva de luz são mostrados na Fig. 1 .
Os fatores de Bayes e os parâmetros da Estimativa de Máxima Verossimilhança Bayesiana (MLE) para os parâmetros de ajuste em diferentes cortes de redshift, com incertezas dos fatores de Bayes muito pequenas para serem exibidas no gráfico. O gráfico superior mostra os fatores de Bayes, enquanto a seção superior (
As comparações bayesianas são melhor compreendidas dividindo o corte mínimo de desvio para o vermelho usado em três regimes: (i) para
Uma vez que a condição de expansão uniforme quase local de Hubble do timescape se mantém em escalas
Ao comparar dois modelos com diferentes suposições no regime não linear, a distribuição do desvio para o vermelho dos dados torna-se particularmente importante. Por exemplo, Lane et al. ( 2024 ) encontraram uma preferência fraca consistente em favor do cenário temporal usando a subamostra P+580, na qual os dados da amostra completa são truncados em desvios para o vermelho altos e baixos. Embora a evidência para a amostra P+1690 mude significativamente de ordem
A diferença nos fatores de Bayes para a amostra completa de P+1690 e a subamostra de P+580 entre Lane et al. ( 2024 ) e nossos resultados. Para a subamostra, os resultados da nova análise apresentada aqui se alinham muito bem com os resultados de Lane et al. ( 2024 ), enquanto para a amostra completa a nova análise aumenta consideravelmente a preferência em favor do escape temporal.
Os fatores de Bayes e os parâmetros da Estimativa de Máxima Verossimilhança Bayesiana (MLE) para diferentes cortes de redshift são mostrados. O painel superior mostra os fatores de Bayes, com azul indicando preferência por paisagem temporal e vermelho para
A convergência do
A mudança em
Nos feixes com método de correção de polarização (Kessler & Scolnic 2017 ), uma correção do hospedeiro da galáxia é introduzida com um parâmetro adicional,
4 DISCUSSÃO E CONCLUSÕES
Quaisquer vieses astrofísicos ou ambientais provavelmente impactariam ambos os modelos cosmológicos. Assim, a forte preferência pela paisagem temporal exigiria uma combinação extremamente sutil de tais vieses para que esta fosse sua causa principal. O maior erro sistemático no Panteão
Como o timescape tem o mesmo número de parâmetros livres que o espacialmente plano
Os resultados apresentados nesta Carta indicam que a cosmologia da paisagem temporal não é apenas uma candidata viável à
Independentemente de qual modelo cosmológico se tornará o padrão no futuro, explorar mais de um modelo é importante. De fato, a estrutura de paisagem temporal é consistente com novas análises de estatísticas de vazios em simulações de relatividade numérica usando as equações completas de Einstein (Williams et al. 2024 ). Estas são consistentes com uma curvatura espacial cinética emergente de vazios em pequenas escalas. Ainda há muito a ser feito na calibração da fração de matéria escura, da velocidade do som primordial e da escala BAO. No entanto, novos resultados provavelmente fornecerão uma estrutura robusta para isso (Galoppo & Wiltshire 2024 ; Galoppo, Re & Wiltshire 2024 ).
Nossos resultados implicam consequências profundas para a cosmologia e a astrofísica. De fato, uma preferência pela cosmologia de paisagem temporal em detrimento das cosmologias FLRW padrão pode apontar para a necessidade de revisão dos fundamentos da cosmologia teórica, tanto ontológica quanto epistemologicamente, para melhor compreender as inomogeneidades e sua reação à evolução média do Universo.
AGRADECIMENTOS
DLW, RRH e ZGL são apoiados pelo Fundo Marsden administrado pela Royal Society of New Zealand, Te Apārangi sob bolsas M1271 e M1255. RRH também é apoiado pela Bolsa de Pós-Doutorado da Fundação Rutherford RFT-UOC2203-PD. Somos gratos ao revisor anônimo de Lane et al. ( 2024 ) por sugerir a estrutura de análise. Também agradecemos ao revisor anônimo desta Carta por seus comentários construtivos e perspicazes. Agradecemos a todos os membros dos grupos de Gravidade, Cosmologia e Astrofísica da Universidade de Canterbury por estimularem as discussões, particularmente: John Forbes, Christopher Harvey-Hawes, Morag Hills, Emma Johnson, Shreyas Tiruvaskar, Michael Williams e Manon van Zyl. Finalmente, gostaríamos de agradecer a Elena Moltchanova por seus preciosos insights sobre os métodos estatísticos empregados.
DISPONIBILIDADE DE DADOS
Um conjunto completo de códigos e detalhes usados para nossa análise e como usá-los pode ser encontrado em Seifert & Lane ( 2023 ), e os arquivos de covariância e entrada são disponibilizados em Lane & Seifert ( 2024 ).
Notas de rodapé
Para o Panteão
Veja Dam et al. ( 2017 , apêndice A) para comparações detalhadas de cálculos de distância de luminosidade nos modelos de paisagem temporal e FLRW.
A principal pequena diferença ocorre no viés geométrico homogêneo de Eddington (McKay 2016 ), o que leva ao potencial para testes futuros.
O valor de Lane et al. ( 2024 )
O
As contribuições relativas da incerteza constante de Hubble e da incerteza de magnitude absoluta, respectivamente
Uma razão adicional para não incluir correções de galáxias hospedeiras é a observação de que
Comentários
Postar um comentário