Skip to main content
Sandbox Physics
Baryon acoustic oscillations

How to read the DESI BAO results

Sandbox Physics ·

A DESI BAO paper reports distances divided by the sound horizon, DM/rdD_M/r_d, DH/rdD_H/r_d or DV/rdD_V/r_d, at a few effective redshifts, with a correlation between each pair. Those ratios are the measurement. Everything after them, a matter density, a Hubble constant, a preference for evolving dark energy, comes from combining them with a cosmological model and often with other data. This guide walks through DESI 2024 III, the galaxy and quasar BAO paper from the first year of data, and DESI DR2 Results II, the three-year cosmology paper, so you can tell which number is which and how firm it is.

The versions read here are arXiv 2404.03000v1 and arXiv 2503.14738v3. This guide does not check later journal versions. It assumes the chapter on the two dilation parameters.

Start with what is measured

Turn first to the tables of results. DESI 2024 III gives two kinds.

Table 15 lists, for each tracer and redshift range, the fitted dilations, together with χ2\chi^2 and the degrees of freedom. These are relative to the fiducial cosmology. For the 300,017 bright galaxies (BGS) at 0.1<z<0.40.1 < z < 0.4, after reconstruction, it gives αiso=0.9829±0.0185\alpha_{\mathrm{iso}} = 0.9829 \pm 0.0185 with χ2=17.3\chi^2 = 17.3 for 19 degrees of freedom. Before reconstruction it was 0.9646±0.02580.9646 \pm 0.0258.

Table 18 converts those into cosmology-independent ratios. For BGS at zeff=0.30z_{\mathrm{eff}} = 0.30, DV/rd=7.93±0.15D_V/r_d = 7.93 \pm 0.15. Only DVD_V is given for BGS. In Table 15 BGS, ELG1 and QSO have an isotropic fit only, while the other samples also have αAP\alpha_{\mathrm{AP}} and its correlation with αiso\alpha_{\mathrm{iso}}.

The abstract summarises the whole set: more than 5.7 million redshifts, BAO detected in all six redshift bins, a combined precision of about 0.52 percent, and a strongest single detection of 9.1σ9.1\sigma at zeff=0.93z_{\mathrm{eff}} = 0.93. BGS is the weakest, at 4.0σ4.0\sigma.

DR2 Results II repeats the exercise with more than 14 million galaxies and quasars. Its BGS sample has 1,188,526 galaxies with a fainter magnitude cut, and gives αiso=0.9857±0.0093\alpha_{\mathrm{iso}} = 0.9857 \pm 0.0093 and DV/rd=7.942±0.075D_V/r_d = 7.942 \pm 0.075 at zeff=0.295z_{\mathrm{eff}} = 0.295. The error has roughly halved.

How the pipeline gets from galaxies to a dilation

Section 4 of DESI 2024 III describes the pipeline. Four choices matter most for reading the result.

Blinding. The catalogues were blinded before the analysis, so that the fitting choices could not be tuned toward an expected answer. All fiducial choices were fixed on mocks and blinded data.

Estimator. Correlation functions use the Landy-Szalay estimator, computed with the pycorr code, compressed into the monopole and quadrupole. The correlation-function fit runs over 4848 to 152 h−1 Mpc152\,h^{-1}\,\mathrm{Mpc} in bins of 4 h−1 Mpc4\,h^{-1}\,\mathrm{Mpc}.

Reconstruction. Gravity moves galaxies by roughly 10 Mpc10\,\mathrm{Mpc} over cosmic history and smears the peak. Reconstruction estimates those displacements from the observed density field and moves galaxies back. DESI 2024 III adopts the RecSym convention. For BGS it smooths the density field on 15 h−1 Mpc15\,h^{-1}\,\mathrm{Mpc} and assumes a bias of 1.51.5 and a growth rate of 0.680.68. DR2 Results II uses the IterativeFFT algorithm in the pyrecon code. Compare the pre- and post-reconstruction BGS errors above to see what it buys.

The model. Equation 4.4 splits the power spectrum into a smooth no-wiggle part and the oscillating wiggle part. Only the wiggle part is dilated by α\alpha. The smooth part is evaluated at the observed coordinates, so that the fit cannot draw distance information from the broadband shape. The wiggles are damped by Gaussian factors with scales Σ⊥\Sigma_\perp and Σ∥\Sigma_\parallel. For BGS these are 6.56.5 and 10.0 h−1 Mpc10.0\,h^{-1}\,\mathrm{Mpc} before reconstruction and 3.03.0 and 8.08.0 after. The remaining broadband is a sum of cubic spline terms spaced by Δ=2π/rd≃0.06 h Mpc−1\Delta = 2\pi/r_d \simeq 0.06\,h\,\mathrm{Mpc^{-1}}, wide enough not to mimic the wiggles.

The Lab and the method guide in this series use a much simpler model, a dilated template plus three power-law terms. They teach what α\alpha means and leave out most of what makes the DESI fit robust.

How large the systematic errors are

Section 5 builds a systematic error budget from theory, modelling, observational effects and the choice of fiducial cosmology. Equation 5.1 gives the total as a covariance in percent: a standard deviation of 0.2450.245 percent on αiso\alpha_{\mathrm{iso}} and 0.30.3 percent on αAP\alpha_{\mathrm{AP}}, taken as uncorrelated. Equation 5.4 projects it onto α⊥\alpha_\perp and α∥\alpha_\parallel. The paper says this conservative sum raises the error of even its best αiso\alpha_{\mathrm{iso}} by only about 5 percent.

The theory and modelling part comes from a companion paper by Chen and colleagues. Its abstract concludes that this contribution is no more than 0.10.1 percent for the isotropic and 0.20.2 percent for the anisotropic post-reconstruction distances. Statistical errors dominate.

Questions worth asking at this point: Was the error budget fixed before unblinding? Does it include everything you can think of, such as imaging systematics or redshift failures? DESI 2024 III says the observational effects were found negligible. Judging that claim needs the supporting papers it cites. This guide has read only the main paper and the abstract of the modelling paper.

How the cosmology is inferred

BAO alone measures ratios. In flat Λ\LambdaCDM, DR2 Results II finds from DESI alone

Ωm=0.2975±0.0086,h rd=(101.54±0.73) Mpc,\Omega_m = 0.2975 \pm 0.0086, \qquad h\,r_d = (101.54 \pm 0.73)\,\mathrm{Mpc} ,

with a correlation of −0.92-0.92 between them. The Hubble constant and the sound horizon enter only as their product.

To get H0H_0 alone you need rdr_d, which needs the baryon density. With a big-bang nucleosynthesis prior on ωb\omega_b, DR2 Results II gives H0=68.51±0.58 km s−1 Mpc−1H_0 = 68.51 \pm 0.58\,\mathrm{km\,s^{-1}\,Mpc^{-1}}. Adding the angular acoustic scale θ∗\theta_* from Planck gives 68.45±0.4768.45 \pm 0.47. Each extra input brings its own assumptions about early-universe physics.

Where the results are contested

DESI 2024 III already noted that its measurements are systematically larger than the Planck 2018 Λ\LambdaCDM prediction below z=0.8z = 0.8.

DR2 Results II sharpens this. Within Λ\LambdaCDM, the parameters BAO prefers are in 2.3 sigma tension with those from the cosmic microwave background, even though DESI agrees with the acoustic angle θ∗\theta_*. A dark energy equation of state that evolves, parametrised by w0w_0 and waw_a, fits better. The preference is 3.1 sigma for DESI plus the microwave background, and 2.8 to 4.2 sigma when supernovae are added, depending on which supernova compilation is used.

Three points help when reading these claims. The significance depends on the data combination and on the parametrisation, so it is not a single number for "evolving dark energy". The spread from 2.8 to 4.2 sigma comes from different supernova samples, so supernova systematics matter as much as BAO ones. And the paper's own abstract says that, unless there is an unknown systematic error in one or more datasets, Λ\LambdaCDM is being challenged. Treat it as an open question.

A reading checklist

  1. Find the measured ratios and their effective redshifts before reading any cosmology.
  2. Check whether a quoted dilation is relative to a fiducial cosmology, and whether a quoted distance has already been multiplied by an assumed rdr_d.
  3. Compare pre- and post-reconstruction errors.
  4. Find the systematic budget and see how it compares with the statistical error.
  5. For any cosmological claim, list the extra datasets and priors it uses.
  6. For any significance, note the model it is measured against.

Where this series fits

The method guide measures a correlation function from a tiny patch of the same DR1 bright galaxy sample and shows why it cannot reproduce Table 15. The Lab shows what α∥\alpha_\parallel and α⊥\alpha_\perp do to a synthetic catalogue. Neither reanalyses DESI.

Sources: DESI 2024 III, galaxy and quasar BAO from the first year, DESI DR2 Results II, BAO measurements and cosmological constraints, Chen et al. 2024, theory and modelling systematics for DESI BAO, abstract, Eisenstein, Seo and White 2007, nonlinear displacements.

Open the Every assumed distance too long preset and read it as a measurement of distance over the sound horizon.

Sources

  1. DESI 2024 III, galaxy and quasar BAO from the first year
  2. DESI DR2 Results II, BAO measurements and cosmological constraints
  3. Chen et al. 2024, theory and modelling systematics for DESI BAO, abstract
  4. Eisenstein, Seo and White 2007, nonlinear displacements

Keep reading

What this page links to directly

See this series on the content map →

Baryon acoustic oscillations · Reading path

From a sound wave to galaxy pair counts and DESI distances

  1. ArticleHow can pairs of galaxies measure the universe?
  2. ArticleHow far sound travelled before the drag epoch
  3. ArticleCounting galaxy pairs against a random catalogue
  4. ArticleReading the ruler along and across the line of sight
  5. ArticleMeasure a correlation function from a DESI data patch
  6. You are hereHow to read the DESI BAO results
Something broken, confusing, or missing on this page?Tell us
What kind of feedback?

Sent with this page’s address, its Lab ID, the site language and your device type. Please leave out personal details. Feedback is kept for up to 12 months.Privacy policy