In a 12-parameter extended cosmology, the combination CMB+BAO+DESY5+WL yields a + preference for nonzero neutrino mass: (95%), stated as 2.1.
Briefing
This paper asks whether recent large-scale-structure measurements from DESI Data Release 2 (DR2) BAO, when combined with CMB, supernovae, and weak-lensing data, provide evidence for physics beyond the standard CDM model—specifically (i) nonzero neutrino masses, (ii) dynamical dark energy, (iii) a possible “lensing anomaly” captured phenomenologically by a rescaling of the CMB lensing amplitude, and (iv) whether these extensions can ease the Hubble tension. The question matters because neutrino masses and dark energy dynamics are fundamental inputs to both cosmology and particle physics, while the lensing anomaly and Hubble tension are long-standing inconsistencies that may indicate either unmodeled systematics or new physics.
The author performs a Bayesian parameter inference in a 12-parameter cosmological model. The baseline six parameters are the standard CDM set: (cold dark matter density), (baryon density), (sound-horizon to angular-diameter-distance ratio at decoupling), (reionization optical depth), (scalar spectral index), and (primordial amplitude). The six extensions are: Chevallier–Polarski–Linder (CPL) dynamical dark energy parameters , the sum of neutrino masses , the effective number of non-photon radiation species , the running of the scalar spectral index , and a phenomenological lensing-amplitude scaling . The analysis assumes a degenerate neutrino mass hierarchy (three equal masses) and imposes the prior . The primordial power spectrum includes running via a standard Taylor expansion in .
Methodologically, the study uses Markov Chain Monte Carlo sampling with Cobaya, computing cosmological predictions with CAMB. Convergence is checked using Gelman–Rubin statistics with the criterion . Flat priors are applied (broad ranges listed in the paper), and the author uses Planck PR4 likelihoods for CMB temperature/polarization (HiLLiPoP and LoLLiPoP, plus Commander for low- TT) and Planck PR4 + ACT DR6 for CMB lensing. For BAO, the likelihood is DESI DR2 (“DESI2”), including multiple tracers across redshift ranges (BGS, LRG, ELG, LRG+ELG, QSO, and Ly). For supernovae, the paper uses uncalibrated Type Ia SN likelihoods from Pantheon+ (1550 SNe) and DES Year 5 (DESY5; 1635 SNe), but never uses them simultaneously to avoid double counting. Weak lensing comes from DES Year 1 (1321 deg two) combining galaxy clustering and shear.
The key results are presented as marginalized constraints for several dataset combinations. The most prominent finding is a neutrino-mass preference that becomes statistically significant only when weak-lensing data are included. With the combination CMB (Planck PR4 + lensing) + DESI2 BAO + DESY5 SNe + WL, the paper reports a first + detection of nonzero neutrino mass with at 95% confidence. The corresponding significance is stated as 2.1 for this dataset set. Replacing DESY5 with Pantheon+ yields a slightly weaker but still near-significant signal: with CMB+BAO+Pantheon++WL. Without weak lensing, the neutrino-mass posteriors peak at positive values but do not reach a 2 detection; for example, with CMB+BAO+lensing+DESI2+DESY5 (no WL) the paper gives (1) with an upper limit at 95% C.L. The author attributes the WL-driven enhancement to a strong negative correlation between and : WL data prefer lower , and in this extended model that can be accommodated by increasing neutrino mass, which suppresses structure growth.
The paper also finds dataset-dependent evidence for dynamical dark energy. When using CMB+BAO+Pantheon+, the cosmological constant case lies at the edge of the 95% contour. However, when DESY5 is used instead of Pantheon+, the cosmological constant is excluded at more than 2. The author emphasizes that adding WL has negligible impact on the dark-energy constraints, so the conclusion about dynamical dark energy remains sensitive to which SN dataset is chosen.
For the lensing anomaly, the study introduces and finds that including weak lensing shifts the inferred lensing amplitude away from unity. Specifically, with CMB+BAO+SNe+WL, the paper states that is excluded at more than 2. In the parameter table, the corresponding marginalized values show with uncertainties such that the 2 interval excludes 1 when WL is included (while it remains consistent with 1 at the 2 level without WL). The author interprets this as evidence that the apparent Planck PR4 lensing anomaly may depend on non-CMB datasets.
Regarding the effective number of relativistic species and the Hubble tension, the paper reports that remains consistent with the standard value (the table shows values around – with uncertainties ). The Hubble tension persists: using the derived values from the table, the discrepancy with SH0ES is quoted as 3.6–4.2 depending on the SN dataset. Weak lensing has minimal impact on these numbers. Thus, the simple extensions explored here do not reduce the Hubble tension below 2.
Limitations are not deeply quantified in the excerpt, but several methodological constraints are apparent. First, the neutrino-mass inference is performed with a hard prior , which the author notes could bias the posterior toward larger positive values compared to analyses allowing effective negative masses. Second, the lensing anomaly and neutrino-mass detection both depend on dataset combinations, especially the inclusion of WL, raising the possibility that residual systematics or modeling mismatches in WL (or its interplay with other parameters) could drive the signals. Third, the model is highly extended (12 parameters), which can weaken robustness by introducing degeneracies (the paper notes strong correlations, e.g., between and , , and other parameters).
Practically, the results matter for both cosmology and particle physics. If the + neutrino-mass preference survives future WL updates and alternative lensing datasets, it would provide a cosmological handle on neutrino mass sum in the sub-eV range, relevant to distinguishing normal vs inverted hierarchies. However, because the detection is sensitive to the choice of weak-lensing dataset and the SN compilation, the paper should be read as evidence that is suggestive but not yet definitive. Dark-energy evidence is similarly inconclusive and dataset-dependent. The lensing anomaly result suggests that phenomenological deviations from standard lensing may be entangled with how non-CMB probes are modeled. Finally, the persistence of the Hubble tension implies that resolving it likely requires additional physics beyond the late-time and simple pre-recombination extensions considered here (e.g., more radical modifications to early-universe physics or interactions in the dark sector).
Cornell Notes
The paper performs a 12-parameter Bayesian cosmological analysis combining DESI DR2 BAO with Planck PR4 (including lensing), ACT lensing, Pantheon+ or DESY5 supernovae, and DES Year 1 weak lensing. It finds a WL-dependent + preference for nonzero neutrino mass , dataset-dependent hints of dynamical dark energy, and a WL-driven exclusion of at , while the Hubble tension remains unresolved.
What research question does the paper address?
Whether combining DESI DR2 BAO with CMB, supernovae, and weak lensing in a 12-parameter extended cosmology yields evidence for nonzero neutrino masses, dynamical dark energy, a lensing anomaly (via ), and/or alleviation of the Hubble tension.
What is the cosmological model and parameter set?
A 12-parameter model: six CDM parameters plus with CPL dark energy.
What study design and inference method are used?
Bayesian MCMC sampling with Cobaya, using CAMB for theory predictions, and Planck PR4/ACT DR6 likelihoods for CMB and lensing; convergence is checked with Gelman–Rubin .
Which datasets are combined, and how are supernova datasets handled?
CMB: Planck PR4 (HiLLiPoP/LoLLiPoP + Commander TT). Lensing: Planck PR4 + ACT DR6. BAO: DESI DR2 (“DESI2”). SNe: Pantheon+ or DESY5, but never simultaneously to avoid double counting. WL: DES Year 1 (galaxy clustering + shear).
What is the primary neutrino-mass result and under which dataset combination?
With CMB+BAO+DESY5+WL (plus Planck PR4 lensing + ACT DR6 lensing), the paper reports a + preference for nonzero neutrino mass: (95%), stated as 2.1.
How does the neutrino-mass evidence change if Pantheon+ replaces DESY5?
It remains but weakens to about with CMB+BAO+Pantheon++WL.
What happens to neutrino-mass detection when weak lensing is removed?
Without WL, there is no 2 detection; posteriors still peak at , with only a + hint for some combinations.
What is the lensing anomaly result?
Including WL shifts such that is excluded at for CMB+BAO+SNe+WL, while it is only consistent with unity at the level without WL.
Does the paper find evidence that the Hubble tension is resolved?
No. The Hubble tension persists at roughly 3.6–4.2 depending on the SN dataset, and WL has minimal impact.
Review Questions
Which parameter degeneracy does the author identify as driving the WL-dependent neutrino-mass preference, and what is the sign of the correlation?
Why does the choice between Pantheon+ and DESY5 change the dark-energy conclusion in this analysis?
How does adding weak lensing affect and what does that imply about the robustness of the Planck lensing anomaly?
What aspects of the model (e.g., priors on ) could bias the neutrino-mass posterior, and how does the author acknowledge this?
Key Points
- 1
In a 12-parameter extended cosmology, the combination CMB+BAO+DESY5+WL yields a + preference for nonzero neutrino mass: (95%), stated as 2.1.
- 2
Replacing DESY5 with Pantheon+ reduces the neutrino-mass evidence to , and removing WL eliminates any 2 detection (posteriors still peak at positive ).
- 3
Evidence for dynamical dark energy is dataset-dependent: is at the edge of the 95% contour with Pantheon+ but excluded at with DESY5; WL has negligible impact on .
- 4
The lensing anomaly is sensitive to non-CMB probes: with WL included, is excluded at , while without WL it remains consistent with unity at the level.
- 5
The Hubble tension is not resolved: the paper finds 3.6–4.2 discrepancy with SH0ES depending on the SN dataset; WL does not significantly change this.