Constraints on neutrino physics from DESI DR2 BAO and DR1 full shape

This paper asks how well current large-scale structure (LSS) measurements can constrain neutrino properties—specifically the sum of neutrino masses $\sum m_{\nu }$ and the effective number of relativistic species $N_{\mathrm{eff}}$—and whether the resulting cosmological limits are consistent with neutrino oscillation experiments. The question matters because neutrinos are among the few particle-physics inputs that leave measurable imprints on cosmic expansion and structure formation. In the standard cosmological picture, massive neutrinos suppress the growth of structure on small scales through free streaming, while also affecting the late-time expansion history that sets the distances measured by baryon acoustic oscillations (BAO). As cosmological neutrino-mass limits approach the lower bounds implied by oscillations, any mismatch becomes a potential sign of unmodeled systematics or new physics.

The authors combine two complementary DESI analyses: (i) BAO measurements from DESI Data Release 2 (DR2), using over 14 million galaxy/quasar redshifts and 820,000 Lyman-alpha forest spectra (and cross-correlations), covering roughly $0.1<z<4.2$; and (ii) a full-shape (FS) analysis of the DESI Data Release 1 (DR1) galaxy/quasar power spectrum multipoles, modeled with the Effective Field Theory of Large Scale Structure (EFT) over $0.02<k/(h{\mathrm{Mpc}}^{-1})<0.20$ in six redshift bins. These LSS data are combined with CMB information from Planck and the Atacama Cosmology Telescope (ACT), including CMB lensing. The inference is performed with Cobaya using Metropolis-Hastings MCMC, with Boltzmann solvers CAMB or CLASS and likelihoods for Planck/ACT. External priors include a BBN prior on the physical baryon density ${\omega }_b={\Omega }_b{h}^2=0.02218\pm 0.00055$, and Planck-informed priors on the primordial spectral index $n_s$ (either a tight $\pm 1\sigma$ prior or a looser $10\sigma$ prior). Neutrino oscillation constraints are incorporated via NuFIT 6.0 mass-squared splittings, yielding lower bounds on $\sum m_{\nu }$ of $\sum m_{\nu }>0.05878\pm 0.00023\mathrm{eV}$ for normal ordering (NO) and $\sum m_{\nu }>0.09892\pm 0.00041\mathrm{eV}$ for inverted ordering (IO).

The baseline cosmological model is $\Lambda$CDM with three degenerate massive neutrinos and the physical prior $\sum m_{\nu }>0$. Under this setup, the main result is a tight upper limit from DESI DR2 BAO + CMB: $\sum m_{\nu }<0.0642\mathrm{eV}$ at 95% confidence, with marginalized uncertainty $\sigma (\sum m_{\nu })=0.020\mathrm{eV}$. The paper also constrains early-universe radiation content, finding $N_{\mathrm{eff}}=3.23_{-0.34}^{+0.35}$ (95%), consistent with the Standard Model expectation $N_{\mathrm{eff}}=3.044$. When allowing neutrino mass ordering using an oscillation-informed parametrization in terms of the lightest mass $m_l$, the data prefer the normal ordering with posterior probability $P(\mathrm{NO})=0.91$ (Bayes factor $K=10$), and yield $m_l<0.023\mathrm{eV}$ (95%) for NO (and $m_l<0.024\mathrm{eV}$ for IO).

A central theme is that the posterior for $\sum m_{\nu }$ in the baseline analysis peaks at the boundary $\sum m_{\nu }=0$, and the authors argue this is driven by “prior weight effects” rather than a true preference for zero mass. To quantify the tension with oscillation lower limits, they perform both frequentist profile-likelihood analyses and a more general Bayesian analysis using an effective neutrino mass parameter $\sum m_{\nu }^{\mathrm{eff}}$ that can take negative values (implemented consistently at the perturbation level). In the frequentist FeldmanCousins construction (correcting for the physical boundary at zero mass), they obtain a 95% upper limit $\sum m_{\nu }<0.053\mathrm{eV}$ in $\Lambda$CDM, which “breaches” the oscillation lower limit for NO ( $\sum m_{\nu }>0.059\mathrm{eV}$). In the Bayesian $\sum m_{\nu }^{\mathrm{eff}}$ framework, they find $\sum m_{\nu }^{\mathrm{eff}}=-0.101_{-0.056}^{+0.047}\mathrm{eV}$ (68%) for DESI DR2 BAO + baseline CMB, corresponding to a marginalized error $\sigma (\sum m_{\nu }^{\mathrm{eff}})=0.053\mathrm{eV}$ and a $\sim 3.0\sigma$ tension with the oscillation lower limit. The authors emphasize that negative $\sum m_{\nu }^{\mathrm{eff}}$ should be interpreted as an effective parameter signaling inconsistency (e.g., systematics or model inadequacy), not as literal negative neutrino energy density.

They also explore how the tension changes with cosmological model flexibility. In a dynamical dark energy model using the CPL parametrization $w(z)=w_0+w_{a}z/(1+z)$ (w0waCDM), the neutrino-mass constraint relaxes substantially: $\sum m_{\nu }<0.163\mathrm{eV}$ at 95% (and a frequentist upper limit $\sum m_{\nu }<0.177\mathrm{eV}$). This relaxation is interpreted as a consequence of degeneracies between neutrino mass and the late-time expansion history. The paper further notes that the preference for negative $\sum m_{\nu }^{\mathrm{eff}}$ is reduced in w0waCDM, and that viable solutions with positive neutrino masses become possible.

Complementing the expansion-history constraints from BAO, the authors use DESI DR1 FS clustering (neutrino free-streaming imprint on the power-spectrum shape) to obtain a complementary limit. Their strongest free-streaming-based constraint in $\Lambda$CDM is $\sum m_{\nu }<0.193\mathrm{eV}$ at 95% (with selected CMB priors). They also diagnose the information source: by modifying the theory so neutrino masses affect only the background expansion (not perturbations), they show the constraint weakens markedly, indicating that the FS analysis is genuinely sensitive to the scale-dependent free-streaming suppression rather than to amplitude or background effects. They further test the role of redshift-space distortions (RSD) and the primordial slope prior $n_s$, finding that neutrino-mass limits depend strongly on the ability to anchor the power-spectrum shape (especially $n_s$), while being less sensitive to amplitude once RSD-related growth information is marginalized.

Limitations include dependence on assumed cosmological models (especially $\Lambda$CDM vs w0waCDM), sensitivity to CMB likelihood choices (Planck plik vs CamSpec vs LoLLiPoPHiLLiPoP), and reliance on external priors (BBN ${\omega }_b$, $n_s$, and sometimes compressed CMB parameter priors). The authors also acknowledge that the negative- $\sum m_{\nu }^{\mathrm{eff}}$ preference could reflect unidentified systematics or new physics; they do not claim it is evidence for literal negative neutrino masses. While they validate the BAO methodology with high-fidelity neutrino-targeted mocks (Peregrinus suite), the paper’s tension analysis is still limited by statistical power and by the fact that the DR1 FS + DR2 BAO cross-covariance is not yet available, preventing a fully optimal combined dataset.

Practically, the results are important for cosmology and particle-physics communities because they provide the tightest current astrophysical constraints on neutrino masses and show a $\sim 3\sigma$ level tension with oscillation-based lower bounds under $\Lambda$CDM when boundary and prior effects are treated carefully. This suggests either (a) unmodeled systematics (in LSS modeling, CMB likelihoods, or calibration of distance scales), or (b) a need for model extensions such as evolving dark energy or other new physics in the neutrino sector. Researchers planning future DESI analyses (especially DR2 full-shape) should care because the paper indicates where the constraining power comes from (free-streaming shape vs geometric BAO) and how degeneracies with dark energy and CMB parameters can mask or mimic neutrino effects. Finally, the results motivate cross-checks using alternative CMB likelihoods, alternative neutrino parametrizations, and robustness tests against known systematics in BAO and full-shape modeling.