Stable cosmology under flexible covariances: Rescaling and cross-survey correlations in DESI BAO and supernova data
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.
Vera: We're looking at the paper: "Stable cosmology under flexible covariances".
Title and authors: Jocelyn: The authors are Kolonia, Li, Zhang, and Afshordi from the Perimeter Institute for Theoretical Physics at Waterloo. They put forward this framework to test the sensitivity of our results when we allow extra covariance freedom in BAO measurements and survey-dependent systematics in supernova data.
Vera: What’s interesting is that they’re using a simple rescaling parameter, sX, which multiplies the covariance matrix C by sX to define a new matrix C X equals sX times C. This lets them test for additional covariance while keeping the assumption that this scaling factor sX has to be one or greater than one.
Subrahmanyan: That’s a clever way to test if we're just underestimating our error bars, rather than saying the published uncertainties are completely wrong or too small for the data.
Jocelyn: And they also introduce these cross-survey correlations, modeling them through pairwise coefficients like rPP U3 and rPP DES in the full model. This lets them see if how one survey's error correlates with another survey's error changes the final result significantly.
The paper's summary: Vera: So, the paper summarizes that they apply this framework to combine different data sets, specifically DESI DR2 BAO and supernova data from Planck two thousand eighteen temperature and polarization data <ref:2611.13829#pg1,Planck 2018 temperature and polarization data>. They show that allowing sBAO to vary in the DESI DR2 BAO likelihood only produced mild changes in the cosmological constraints.
Jocelyn: And for the supernova datasets, allowing sU3 to vary results in a mild broadening of the constraints, and their best-found effective chi squared difference goes from fifteen point zero zero with fixed scales down to fourteen point two seven when they allow free scales.
Subrahmanyan: That's a bit reassuring because it suggests that even with those extra degrees of freedom we introduce, the shift in the tension between w0waCDM and Lambda CDM is very small, changing only from a chi squared difference of twelve point eight to twelve point three >
Vera: Right. The main point they make is that while these parameters modify the posteriors for those nuisance parameters, they don't produce a clear shift in the marginalized cosmological constraints within the models they looked at.
The paper's improvements: Jocelyn: One of their key methodological improvements is how they handle combining heterogeneous datasets. They use a common redshift basis to combine supernova compilations, specifically using the native twenty-two-point Union3 grid as this common index-aligned basis for all the data <ref:2611.13829#pg2>.
Vera: They also make the joint likelihood construction more robust by rebinning different supernova compilations onto this unified grid before combining them into one likelihood. This allows them to create a single joint covariance matrix that incorporates both survey specific systematics and those phenomenological intersurvey correlations.
Subrahmanyan: From a theoretical side, they treat each sX parameter as a nuisance parameter and vary it simultaneously with the cosmological parameters. They adopt uniform priors for these parameters, like sPP, sU3, and sDES being uniformly distributed between one and five or six.
Jocelyn: They also explicitly state that the baseline value for this rescaling factor is one, which means it recovers the published covariance matrix. Because their priors are bounded below at unity, this analysis tests only for additional covariance freedom and doesn't allow them to reduce the reported uncertainties.
Conclusion: Vera: So, to wrap up, they’re showing that these phenomenological extensions modify the nuisance parameter posteriors but don't produce a clear displacement of the marginalized cosmological constraints in their current models. They find changes of less than one in the best-found effective chi squared when covariance rescaling is allowed, which suggests the data are consistent with their published covariances.
Jocelyn: They provide a simple diagnostic tool for anyone doing combined cosmological analyses to assess how sensitive those constraints are to imperfectly modeled uncertainties without needing a detailed physical model for the error sources.
Subrahmanyan: For me, it means we can be more confident that the tension we see between models is real and not just an artifact of our current way of handling those complex error structures in our data processing pipeline.
Vera: Exactly. The paper, "Stable cosmology under flexible covariances: Rescaling and cross-survey correlations in DESI BAO and supernova data", gives us a way to test the stability of these measurements against residual systematics.
Eleni-Anna Kolonia, Haoye Li, Hanyu Zhang, Niayesh Afshordi
Perimeter Institute for Theoretical Physics · University of Waterloo · Swansea University
astro-ph.CO
Submitted: 2026-11-10
Updated: 2026-11-10
Comments: 21 pages, 12 figures, 7 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 74/100
The gist: Combinations of cosmic microwave background (CMB), baryon acoustic oscillation (BAO), and Type Ia supernova (SN) data currently favor evolving dark energy over ΛCDM at the ∼ 3–4σ level…
Key concepts
- Covariance Rescaling Parameter (sX)
- This parameter multiplies a covariance matrix by sX to test for additional systematic uncertainty. If sX > 1, it indicates that the data suggest more covariance than initially modeled. This allows researchers to quantify the extra uncertainty allowed by the data without assuming the published matrices are too small.
- Joint Likelihood Framework
- The authors create a unified analysis framework for supernovae by combining data from multiple surveys (like Pantheon+, Union3, and DES–Dovekie) using a common redshift basis. This allows for the inclusion of survey-specific systematics and phenomenological inter-survey correlations in one likelihood function.
- Cross-Survey Correlations
- These coefficients (e.g., rPP,U3) model how uncertainties or systematic errors between different surveys are related. Including these correlations tests the robustness of constraints against imperfect modeling of how different datasets might influence each other's error structure.
- Robustness Diagnostic Tool
- The framework serves as a diagnostic tool to assess the reliability of combined cosmological analyses. By seeing if marginalized cosmological constraints shift when covariance parameters are allowed to vary, researchers can determine if residual systematics significantly alter the final scientific conclusions.
Terminology
Summary
Combinations of cosmic microwave background (CMB), baryon acoustic oscillation (BAO), and Type Ia supernova (SN) data currently favor evolving dark energy over ΛCDM at the ∼ 3–4σ level, making it important to test whether this preference could be driven by imperfectly modeled uncertainties<ref:2611. p. 013829#pg2>. This work introduces a phenomenological framework where published covariance is rescaled and cross-survey correlations are allowed in a joint likelihood to assess the robustness of cosmological constraints against imperfectly modeled uncertainties<ref:2611. p. 013829#pg3>.
Methodological Framework
The authors develop a phenomenological framework to incorporate residual systematics and inter-survey correlations into a unified analysis, focusing on rescaling covariance matrices and modeling cross-correlations directly at the likelihood level<ref:2611.13829#pg3>. For systematic uncertainty, they introduce a covariancerescaling parameter, denoted by sX, which multiplies the covariance matrix C by sX to define C'X = sXCX<ref:2611.13829#pg4>. This rescaling tests for additional covariance while maintaining that the adopted priors satisfy sX ≥ 1<ref:2611. p. 113829#pg4>.
Data Treatment and Basis Alignment
The analysis utilizes a common redshift basis to combine heterogeneous supernova compilations, using the native 22-point Union3 grid as this common index-aligned basis<ref:2611.13829#pg9>. The joint likelihood for supernovae contains 63 observed entries by combining the 22 Pantheon+ entries, the 22 Union3 entries, and only the 19 populated bins from DES–Dovekie<ref:2611. p. 713829#pg18>.
Results on BAO and Supernovae
The analysis shows that allowing sBAO to vary in the DESI DR2 BAO likelihood produces only mild changes in the marginalized cosmological constraints, with the improvement of w0waCDM over ΛCDM changing only from ∆χ2 = 12.8 to 12.3<ref:2611. p. 013829#pg6>. Similarly, for supernova datasets, allowing sU3 to vary results in a mild broadening of cosmological constraints, and the best-found effective χ2 difference decreases from 15. p. 500 with fixed scales to 14. p. 527 with free scales<ref:2611. p. 513829#pg8>.
Impact of Correlations and Systematics
The inclusion of inter-survey correlations, modeled through pairwise coefficients rPP,U3, rPP,DES, and rU3,DES in the full model, has a relatively small impact on the inferred cosmological constraints<ref:2611.13829#pg11>. The resulting cosmological contours substantially overlap across different covariance prescriptions<ref:2611. p. 913829#pg8>. Overall, these results indicate that while residual systematics and inter-survey correlations may influence the detailed nuisance parameter structure of the combined likelihood, they do not significantly alter the resulting cosmological constraints within the current observational precision<ref:2611. p. 813829#pg16>.
Conclusion on Robustness
The phenomenological covariance extensions modify the nuisance-parameter posteriors but do not produce a clear displacement of the marginalized cosmological constraints within the models considered here<ref:2611.13829#pg21>. The available minimizations show changes of less than one in the best-found effective χ2 when covariance rescaling is allowed, indicating that the data are consistent with their published covariances<ref:2611. p. 013829#pg21>. The framework provides a simple diagnostic tool for assessing the robustness of combined cosmological analyses in the presence of imperfectly modeled uncertainties<ref:2611. p. 213829#pg21>.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improve uncertainty quantification by allowing covariance rescaling parameters to vary simultaneously with cosmological parameters, which
broadens the cosmological constraints without shifting them
and provides asimple and agnostic diagnostic of whether residual systematic uncertainty can affect the inferred cosmological constraints.
-
Enhance model comparison robustness by implementing a framework that allows for
phenomenological inter-survey correlations
andsurvey-dependent systematics,
enabling researchers to assess sensitivity to these factors without assuming a specific physical origin. -
Enable more rigorous systematic testing by allowing the analysis to test whether the published covariance is underestimated, as evidenced by the sX posteriors extending above one, which
quantify the additional covariance allowed by the data rather than demonstrating that the published covariance matrices are too small.
-
Improve joint likelihood construction for Type Ia supernovae by introducing a common redshift binning strategy that allows all datasets to be projected onto a unified basis, leading to a
joint covariance matrix incorporating both survey-specific systematics and phenomenological inter-survey correlations.
-
Enhance correlation modeling fidelity by allowing the three pairwise SN correlation coefficients to vary independently in the final joint analysis, as this isolates
the effect of freeing the covariance scales conditional on allowing inter-survey correlations.
Sources
- Planck 2018 results. VI. Cosmological parameters
- DESI DR2 Results II: Measurements of Baryon Acoustic Oscillations and Cosmological Constraints
- Cosmic Complementarity: H_0 and Omega_m from Combining CMB Experiments and Redshift Surveys
- Probing Dark Energy with Baryonic Acoustic Oscillations from Future Large Galaxy Redshift Surveys
- Probing dark energy using baryonic oscillations in the galaxy power spectrum as a cosmological ruler
- Cosmic Microwave Background Anisotropies
- Observational Evidence from Supernovae for an Accelerating Universe and a Cosmological Constant
- Measurements of Omega and Lambda from 42 High-Redshift Supernovae
- The DESI Experiment, a whitepaper for Snowmass 2013
- The DESI Experiment Part II: Instrument Design
- Overview of the Instrumentation for the Dark Energy Spectroscopic Instrument
- Sloan Digital Sky Survey IV: Mapping the Milky Way, Nearby Galaxies, and the Distant Universe
- The Baryon Oscillation Spectroscopic Survey of SDSS-III
- The SDSS-IV extended Baryon Oscillation Spectroscopic Survey: Overview and Early Data
- SDSS-III: Massive Spectroscopic Surveys of the Distant Universe, the Milky Way Galaxy, and Extra-Solar Planetary Systems
- DESI 2024 VI: Cosmological Constraints from the Measurements of Baryon Acoustic Oscillations
- The Completed SDSS-IV extended Baryon Oscillation Spectroscopic Survey: Cosmological Implications from two Decades of Spectroscopic Surveys at the Apache Point observatory
- The Pantheon+ Analysis: The Full Dataset and Light-Curve Release
- Union Through UNITY: Cosmology with 2,000 SNe Using a Unified Bayesian Framework
- The Dark Energy Survey: Cosmology Results With ~1500 New High-redshift Type Ia Supernovae Using The Full 5-year Dataset
Related papers
- 21cmEMUv3: a hybrid diffusion-LSTM emulator of 21cmFAST summary observables
- Connecting field-level and n-point function measurements of galaxy bias
- The Impact of Cosmic Variance and Satellites on JWST Clustering Measurements at Redshift around 6
- Universality of Primordial Anisotropies in Gravitational Wave Background
- KiDS-1000 cosmic shear reanalysis using MetaCalibration
- 2Flow2SN: Dusty Supernova Cosmology with Photometric Redshifts and Contamination using Neural Simulation-Based Inference