DES 5-Year Sample Insights

• The dataset provides deep imaging and precise photometry across grizY bands, enabling multi-probe analyses of cosmic structure and dark energy.
• Advanced algorithms and machine learning techniques (e.g., SuperNNova, DNF) are deployed for robust calibration of galaxies, clusters, and type Ia supernovae.
• The survey underpins high-impact cosmological constraints through BAO, cosmic shear, and cluster analyses, achieving precision improvements over earlier measurements.

The Dark Energy Survey (DES) 5-Year Sample is a comprehensive, multi-component dataset originating from a five-year optical imaging campaign with the explicit aim of probing cosmic acceleration through high-precision measurements of large-scale structure, weak lensing, clustering, galaxy clusters, and type Ia supernovae. The dataset spans nearly 5,000 deg² of the southern sky, provides uniform photometry and exquisite astrometric precision across five bands (grizY), and underpins a broad array of high-impact cosmological analyses.

1. Survey Design, Data Acquisition, and Processing

The DES 5-Year Sample results from a uniform deep-wide extragalactic survey and a time-domain transient survey using the Dark Energy Camera (DECam) on the Blanco 4 m telescope at Cerro Tololo Inter-American Observatory. The main wide survey covers ≈5,000 deg² in grizY, with a depth of $\text{g} > 23.5$ mag and typical single-epoch PSF FWHMs of 1.11″ (g), 0.95″ (r), 0.88″ (i), 0.83″ (z), 0.90″ (Y). Relative photometric calibration uncertainties are $<3$ mmag internally and $~10$ mmag (absolute), and the median internal astrometric precision is ~27 mas (Collaboration et al., 2021). All exposures undergo an image-processing pipeline—preprocessing (crosstalk, bias, flat-field, nonlinearity correction), “First Cut” real-time processing (astrometric, photometric calibration, PSF modeling, cosmic ray/bleed-trail masking), and “Final Cut” biannual reprocessing with refined “supercal” calibrations and global astrometry (Morganson et al., 2018). Coaddition strategies produce deep images, maintain photometric homogeneity, and facilitate the construction of extensive star/galaxy catalogs.

A dedicated supernova program covers ~27 deg² to greater depth and cadence for precise transient light curves. The processed data are made available through public data releases (notably DES DR2), accessible via web query portals, APIs, and JupyterLab notebooks (Collaboration et al., 2021).

2. Galaxy, Cluster, and Supernova Samples: Construction and Calibration

Galaxy and BAO Samples

For large-scale structure and BAO measurements, a photometrically selected galaxy sample is constructed using redshift- and color-dependent i-band magnitude cuts of the form $i < 19.64 + 2.894 z_{\text{ph}}$ (depth-limited to $i < 22.5$), optimized for $0.6<z_{\text{ph}}<1.2$ and covering 4,273 deg² with 15,937,556 galaxies (Mena-Fernández et al., 2024). Redshift calibration is performed via a combination of the Directional Neighborhood Fitting (DNF) algorithm, direct calibration with VIPERS spectroscopy, and clustering redshift techniques using SDSS samples. Residual systematics are controlled by iterative decontamination weighting and masking, yielding forecasted BAO measurement precision improvements of ~25% over the previous Y3 analysis.

Galaxy Cluster Cosmology

Clusters are identified via the redMaPPer photometric algorithm. Cosmological analyses jointly model cluster abundances (counts in bins of richness $\lambda$ and redshift) and weak lensing mass calibrations (stacked ΔΣ profiles), accounting for shear calibration and photometric redshift uncertainties (Collaboration et al., 2020, To et al., 2020). A key empirical richness–mass relation is fit,

$$\langle M_{200\text{m}}|\lambda \rangle = 10^{14.351 \pm 0.020} \left(\frac{\lambda}{40}\right)^{1.058 \pm 0.074}\, h^{-1} M_\odot,$$

with systematic controls for selection bias, projection effects, and triaxiality. Imaging data are cross-matched with X-ray and SZ samples to validate cluster identification and purity.

Type Ia Supernova Samples

The SN Ia cosmology sample from the 5-year DES program consists of 1,635 photometrically classified SNe Ia between $0.10Collaboration et al., 2024, Möller et al., 2022). Classification employs SuperNNova (SNN), an LSTM-based RNN trained on simulations, and uses normalization (“cosmo_quantile”) to preserve color/shape without biasing on brightness (Möller et al., 2022, Möller et al., 2024). Ensemble methods—combining multiple networks—achieve >98% balanced accuracy. Redshifts are typically secured from host-galaxy spectroscopy; however, an expanded 2,298-object SN Ia sample is constructed via a photometry-only pipeline (no host-galaxy requirement), fitting simultaneous photometric redshifts and light-curve parameters with a modified SALT2 model, and imposing weak cosmology-based priors to break degeneracies (Möller et al., 2024).

3. Key Scientific Results and Cosmological Constraints

Large-Scale Structure and BAO

The consensus angular BAO measurement at $z_{\rm eff}=0.85$ yields

$<3$0

(2.1% precision), via three independent estimators—Angular Correlation Function (ACF), Angular Power Spectrum (APS), and Projected Correlation Function (PCF)—with $<3$1 being 2.1$<3$2 below Planck ΛCDM predictions (Collaboration et al., 2024). This represents the most precise photometric BAO measurement at $<3$3 to date.

Weak Lensing and Joint Probes

First-year joint “3×2pt” (galaxy clustering, cosmic shear, galaxy-galaxy lensing) and extended “5×2pt” analyses including cross-correlations with CMB lensing from SPT/Planck constrain $<3$4 and $<3$5 (Abbott et al., 2018). The joint-probe methodology is robust, as demonstrated by decisive Bayesian evidence ratios and posterior predictive distribution (PPD) tests indicating no significant tension with Planck lensing auto-spectrum results.

Cluster Cosmology

A focal point of DES cluster cosmology analyses is the strong dependency of $<3$6 and $<3$7 constraints on the inclusion of low-richness clusters ($<3$8), with such bins yielding notably lower values (e.g., $<3$9, $~10$0) in significant tension (up to $~10$1) with other probes (Collaboration et al., 2020). Higher richness thresholds bring the cluster results into concordance with joint-probe and Planck CMB measurements. The multi-probe Y1 analysis ($~10$2, $~10$3) confirms that robust systematics calibration is mandatory for the 5-year sample (To et al., 2020).

Type Ia Supernova Cosmology

The full 5-year SN Ia sample yields

• $~10$4 (SN data only, flat ΛCDM),
• $~10$5 (flat $~10$6CDM),
• $~10$7 (flat $~10$8CDM),

all consistent with a cosmological constant within %%%%29$~10$30%%%%, and systematic errors subdominant to statistical errors (Collaboration et al., 2024). The data alone require cosmic acceleration ($i < 19.64 + 2.894 z_{\text{ph}}$1 in ΛCDM) at over 5$i < 19.64 + 2.894 z_{\text{ph}}$2 confidence. Combined with Planck, SDSS BAO, and DES 3×2pt data, joint constraints sharpen to $i < 19.64 + 2.894 z_{\text{ph}}$3.

Probes of the Matter Power Spectrum and Compact Objects

The weak lensing magnification of SN Ia residuals provides sensitivity to non-linear clustering at $i < 19.64 + 2.894 z_{\text{ph}}$4. DES Year-5 data constrain the amplitude of non-linear power via an empirical parameter $i < 19.64 + 2.894 z_{\text{ph}}$5, yielding $i < 19.64 + 2.894 z_{\text{ph}}$6 (68% credible around the median), hinting at, but not requiring, a mild power suppression relative to standard CDM (Shah et al., 31 Jan 2025). Lensing signatures in SN Ia also place a 95% upper limit, $i < 19.64 + 2.894 z_{\text{ph}}$7, on the fraction of the matter density composed of compact objects (stars, stellar remnants, PBHs, $i < 19.64 + 2.894 z_{\text{ph}}$8), robust against cosmological priors and systematics at the $i < 19.64 + 2.894 z_{\text{ph}}$9 level (Shah et al., 2024).

4. Methodological Innovations and Systematic Error Control

Advances in the DES 5-Year Sample include:

• Implementation of deep-learning–based photometric supernova classification (SuperNNova), ensemble approaches, and Bayesian neural networks for both standard and uncertainty-aware predictions (Möller et al., 2022, Möller et al., 2024).
• Use of likelihood formalisms incorporating non-Gaussian scatter (e.g., lensing-induced in SN Ia Hubble residuals) and joint marginalization over cosmological and nuisance parameters (Shah et al., 2024, Shah et al., 31 Jan 2025).
• Progressive refinement of galaxy sample selection, photometric redshift calibration, and systematics weights using algorithms such as DNF, VIPERS calibration, and iterative systematics decontamination (Mena-Fernández et al., 2024).
• Strong control of systematics through redundant cross-checks with ancillary datasets (VIPERS spectroscopy, SDSS clustering redshifts, X-ray, SZ) and public pipelines supporting reproducibility.

5. Astrophysical Discoveries and Dark Matter Constraints

The depth, uniformity, and area of DES imaging have enabled the detection and characterization of new thin stellar streams, globular cluster extra-tidal features, and ultra-faint star clusters—expanding the stream census by eleven candidates such as Indus, Jhelum, and Chenab (Shipp et al., 2018). The matched-filter technique leverages synthetic metal-poor isochrones and precise photometric calibration ($i < 22.5$01%) to detect structures down to $i < 22.5$1 mag arcsec$i < 22.5$2 and out to $i < 22.5$350 kpc. The spatial and kinematic granularity in these tidal features provides an empirical route to constraining both the Milky Way potential and the dark subhalo population (enabling studies of ΛCDM substructure predictions and the “missing satellites” problem).

Constraints from SN Ia lensing—$i < 22.5$4 for the fraction of matter in compact objects—limit contributions from compact object populations, including primordial black holes, as dominant dark matter constituents (Shah et al., 2024).

6. Data Accessibility, Community Use, and Implications for Future Surveys

DES DR2 (encompassing the 5-year sample plus a sixth-year extension) provides public access to $i < 22.5$5700 million objects and detailed value-added products via DESaccess, LIneA, and Astro Data Lab (Collaboration et al., 2021). Coadded images, object catalogs, depth/coverage maps, and toolkits for in situ analysis enable community-led science at scale.

Findings from the 5-year sample validate advanced photometric classification techniques, pipeline-based calibration, and systematic error controls, presaging the analytic paradigms that will be imperative for next-generation facilities with higher data rates and lower per-object spectroscopic follow-up yield (e.g., Rubin Observatory/LSST, Euclid). Empirical methodologies from DES—including joint-probe analysis, machine-learning classification, and scale-specific systematics marginalization—form the template for LSST- and Roman-scale cosmology.

7. Summary Table: Highlights of the DES 5-Year Sample

Science Probe	Area/Coverage	Key Results/Precision
Galaxy clustering/BAO	4,273 deg², 16M galaxies	$i < 22.5$6 (2.1%)
Cosmic shear, galaxy–CMB lensing	Up to 5,000 deg²	$i < 22.5$7, $i < 22.5$8
Cluster abundance + weak lensing	Up to 5,000 deg²	$i < 22.5$9, $0.6<z_{\text{ph}}<1.2$0
SN Ia Hubble diagram cosmology	1,635 SNe (DES), $0.6<z_{\text{ph}}<1.2$1	$0.6<z_{\text{ph}}<1.2$2 (DES-only)
SN Ia lensing (compact object fraction)	1,532 SNe	$0.6<z_{\text{ph}}<1.2$3 (95% CL), $0.6<z_{\text{ph}}<1.2$4
Stellar streams (Milky Way halo)	$0.6<z_{\text{ph}}<1.2$55,000 deg²	15 new streams/overdensities; $0.6<z_{\text{ph}}<1.2$6 mag arcsec$0.6<z_{\text{ph}}<1.2$7

The DES 5-Year Sample thus serves both as a cosmological benchmark and a methodological blueprint for future wide-field surveys, enabling precision tests of fundamental physics and the detailed mapping of cosmic structure.