DESI Bright Galaxy Sample (BGS)

• Bright Galaxy Sample (BGS) is a low-redshift, flux-limited survey from DESI targeting over 10 million galaxies across 14,000 deg² for precision cosmology.
• It utilizes straightforward magnitude cuts, Gaia-based star–galaxy separation, and dynamic exposure times to achieve high redshift success and fiber-assignment efficiency.
• BGS underpins advanced clustering analyses, refined mock catalogues, and extended studies in galaxy evolution and AGN, validated across multiple DESI data releases.

Searching arXiv for the core DESI BGS paper and closely related work to ground the article in current literature. The Bright Galaxy Sample (BGS) most commonly denotes the bright-time, low-redshift, flux-limited galaxy survey of the Dark Energy Spectroscopic Instrument (DESI), selected from Legacy Surveys imaging and designed to obtain redshifts for more than $10^7$ galaxies over $14{,}000\,\mathrm{deg}^2$ at $z<0.6$, with the core low-redshift clustering program described as a nearly magnitude-limited sample over $0.05 \le z \le 0.4$ and median $z \approx 0.2$. Its primary scientific purpose is precision measurement of baryon acoustic oscillations (BAO) and redshift-space distortions (RSD) through dense galaxy clustering in the nearby Universe (Ruiz-Macias et al., 2020, Hahn et al., 2022). The acronym is not unique, however: it has also been used for distinct 500 $\mu$m-selected bright-galaxy samples in the Herschel Virgo and Fornax cluster surveys (Davies et al., 2011, Davies et al., 2012).

1. Survey definition and scope

In the DESI program, BGS is the bright-time galaxy survey. It exploits observing conditions that are less suitable for DESI’s dark-time tracers and is explicitly designed as a large, low-redshift, flux-limited sample for cosmology and galaxy evolution. In the DR8-era characterization, it was described as a deepened version of the SDSS Main Galaxy Sample, targeting galaxies to $z \approx 0.6$ with median $z \approx 0.2$, and expected to be roughly 10 times larger than the SDSS-I/II Main Galaxy Sample (Ruiz-Macias et al., 2020). The final survey design retained that broad role while emphasizing that BGS will map the dark-energy-dominated epoch with redshifts of $>10$ million galaxies over $14{,}000\,\mathrm{deg}^2$ (Hahn et al., 2022).

The target sample was organized from the outset around a simple magnitude split. In the preliminary target-selection paper, DESI BGS comprised two target classes: BRIGHT with $14{,}000\,\mathrm{deg}^2$0 and FAINT with $14{,}000\,\mathrm{deg}^2$1, with preliminary densities of $14{,}000\,\mathrm{deg}^2$2 and $14{,}000\,\mathrm{deg}^2$3 objects/deg$14{,}000\,\mathrm{deg}^2$4, respectively. GAMA cross-matching in that stage gave mean redshifts $14{,}000\,\mathrm{deg}^2$5 for BRIGHT, $14{,}000\,\mathrm{deg}^2$6 for FAINT, and $14{,}000\,\mathrm{deg}^2$7 for the combined sample (Ruiz-Macias et al., 2020). The final design preserved BGS Bright as a simple magnitude-limited sample with $14{,}000\,\mathrm{deg}^2$8, but extended BGS Faint to $14{,}000\,\mathrm{deg}^2$9 and added a color-dependent fiber-flux requirement to maintain redshift efficiency; it also formalized a small low-$z<0.6$0 AGN recovery channel (Hahn et al., 2022).

The resulting DESI nomenclature distinguishes $z<0.6$1, $z<0.6$2, a promoted faint subset $z<0.6$3, and the AGN-oriented $z<0.6$4 class. This suggests that “BGS” is best understood not as a single immutable cut, but as a survey family whose core bright sample remained stable while the faint and AGN components were optimized during validation.

2. Target classes and selection logic

The central selection logic is photometric and morphology-aware. Targets are selected from Legacy Surveys optical $z<0.6$5 imaging using extinction-corrected AB magnitudes, with broad color cuts

$z<0.6$6

and a fiber-magnitude criterion

$z<0.6$7

The fiber cut was introduced to suppress large, low-surface-brightness, or problematic galaxies whose total $z<0.6$8-band flux is bright enough but whose predicted fiber flux is too faint or poorly modeled for successful spectroscopy (Ruiz-Macias et al., 2020).

A key discriminator is Gaia-based star–galaxy separation. In the preliminary and final DESI formulations, an object is retained as galaxy-like if it is not in Gaia or, for Gaia matches, if

$z<0.6$9

The DR8 characterization showed why this comparison was preferred to purely Tractor morphology: Gaia is complete to the relevant magnitudes and provides robust stellar rejection, while Tractor’s PSF/extended classification can be compromised for compact galaxies and Gaia-AEN-forced PSF fits (Ruiz-Macias et al., 2020). The final design kept the same essential criterion and supplemented it with masking around bright stars and globular clusters, a requirement of coverage in all three optical bands,

$0.05 \le z \le 0.4$0

and a bright-end veto

$0.05 \le z \le 0.4$1

to remove targets likely to contaminate neighboring fibers (Hahn et al., 2022).

The evolution from DR8 to DR9 is important. DR8-era BGS applied hard masks around bright stars, large galaxies, and globular clusters, together with Tractor quality cuts

$0.05 \le z \le 0.4$2

DR9 retained the same general quality logic but adopted less conservative “new FRACS,” requiring these thresholds in two of three bands rather than all three, reduced the bright-star masking radius by a factor of 2 relative to DR8, and removed the hard large-galaxy mask because improved SGA-2020-based source fitting recovered genuine galaxies that DR8 would have excluded (Zarrouk et al., 2021).

The main sample definitions can therefore be summarized as follows.

Configuration	Bright sample	Faint sample
Preliminary	$0.05 \le z \le 0.4$3; $0.05 \le z \le 0.4$4 deg$0.05 \le z \le 0.4$5	$0.05 \le z \le 0.4$6; $0.05 \le z \le 0.4$7 deg$0.05 \le z \le 0.4$8
Final DESI design	$0.05 \le z \le 0.4$9; about $z \approx 0.2$0 deg$z \approx 0.2$1	$z \approx 0.2$2; about $z \approx 0.2$3 deg$z \approx 0.2$4

For the final BGS Faint class, DESI imposed a color-dependent fiber cut,

$z \approx 0.2$5

which was interpreted as a proxy for emission-line strength, especially H$z \approx 0.2$6 and H$z \approx 0.2$7, and was introduced to preserve high redshift efficiency in the faint extension (Hahn et al., 2022).

3. Survey execution, incompleteness, and validation

BGS is operationally tied to DESI’s bright-time strategy. The final design used a survey-speed criterion in which bright-time observations occur when the speed lies in

$z \approx 0.2$8

with exposure times dynamically scaled by the Exposure Time Calculator to maintain homogeneous redshift completeness. The nominal anchor is

$z \approx 0.2$9

defined as the exposure needed to achieve $\mu$0 redshift efficiency for BGS Bright under nominal dark conditions (Hahn et al., 2022).

The geometry of DESI’s focal plane makes incompleteness nontrivial. An early incompleteness study assumed a BGS footprint of $\mu$1 square degrees observed in 3 passes, each pass comprising roughly 2000 DESI tiles of area $\mu$2 square degrees. Because fibers are arranged in 10 wedge-shaped petals and each fiber can move only within a patrol region of radius $\mu$3 arcmin, fiber collisions are not a simple minimum-separation rule; completeness depends strongly on local target surface density and overlapping tile coverage. In low-density regions completeness can exceed $\mu$4 after 3 passes, while in the centers of the most massive haloes it can fall below $\mu$5 (Smith et al., 2018).

That study adopted pair inverse probability (PIP) weighting combined with angular upweighting to correct clustering measurements. Two mitigation steps were central to making the estimator unbiased: dithering the tile pattern by a small random rotation of order $\mu$6, and randomly promoting a small fraction of the lower-priority faint sample to the bright-priority class, with a fiducial promotion of 10%. In mocks, the method recovered the angular correlation function and the redshift-space monopole, quadrupole, and hexadecapole without detectable bias for the full 3-pass survey, and remained unbiased on average even after 1 pass, albeit with much larger scatter (Smith et al., 2018).

Validation of the imaging-side target definition proceeded in stages. The DR8 characterization used GAMA DR4 as an external truth table and found that, after masking and selection, the dominant recoverable incompleteness came from galaxies whose photometry was degraded by Gaia-AEN-based PSF-only fitting; if that issue were fixed, residual incompleteness would drop to about 0.62%. The same work identified the largest systematic correlation as a 7 per cent suppression of the target density in regions of high stellar density, motivating a linear stellar-density weight (Ruiz-Macias et al., 2020).

DR9 reduced these concerns. Cross-matching to GAMA DR4 in the DR9 bright-target study showed $\mu$7 completeness for the nominal bright selection, with only about 4.5 objects/deg$\mu$8 from GAMA missing, mostly due to star–galaxy separation and a smaller contribution from QC cuts (Zarrouk et al., 2021). Final survey validation with SV and realistic simulations then showed that BGS targets have stellar contamination $\mu$9 fiber-assignment efficiency, and BGS Bright and Faint both reach $z \approx 0.6$0 redshift success rates with no significant dependence on observing conditions (Hahn et al., 2022).

4. Clustering measurements and statistical characterization

The BGS selection was optimized for precision clustering, and its angular and three-dimensional statistics became a major validation axis. Using DR9 imaging, the bright sample’s angular correlation function $z \approx 0.6$1 was measured in BASS/MzLS NGC, DECaLS NGC, and DECaLS SGC and fit with a power-law model derived from

$z \approx 0.6$2

For the full BGS Bright sample, the fitted values were $z \approx 0.6$3, $z \approx 0.6$4, and $z \approx 0.6$5 in those three regions, with corresponding $z \approx 0.6$6, $z \approx 0.6$7, and $z \approx 0.6$8. The work also showed the expected luminosity and color dependence: brighter galaxies cluster more strongly, and red galaxies are more strongly clustered than blue galaxies (Zarrouk et al., 2021).

The One-Percent Survey and subsequent population analyses extended this statistical program beyond two-point clustering. PROVABGS used Bayesian SED modeling and hierarchical inference to derive a probabilistic stellar mass function (pSMF) for BGS galaxies, explicitly propagating stellar-mass posterior uncertainties and incorporating fiber-assignment, redshift-failure, and $z \approx 0.6$9-type selection weights. Over $z \approx 0.2$0, the pSMFs showed good agreement with previous measurements and no significant redshift evolution within the quoted uncertainties, while the split at average specific SFR $z \approx 0.2$1 recovered distinct star-forming and quiescent populations (Hahn et al., 2023).

A later and methodologically distinct line of work used DESI DR1 BGS as a testbed for large-scale statistical homogeneity. Rather than analyzing the raw flux-limited catalog, that study constructed four volume-limited BGS subsamples—VL2, VL3, VL4, and VL5—with $z \approx 0.2$2 equal to $z \approx 0.2$3, $z \approx 0.2$4, $z \approx 0.2$5, and $z \approx 0.2$6 Mpc/$z \approx 0.2$7, and measured the conditional average density

$z \approx 0.2$8

over a broad intermediate range. It reported no clear transition to homogeneity up to $z \approx 0.2$9 Mpc/$>10$0, interpreted the flattening at large $>10$1 as a finite-size effect, found variance scaling roughly as $>10$2 on small scales and $>10$3 on larger scales, and argued that the counts-in-spheres fluctuation PDF is better described by a Gumbel distribution than by a Gaussian (Labini et al., 26 Nov 2025). This result is best read as a specific statistical interpretation based on conditional-density estimators and conservative boundary handling, rather than as part of the target-selection or survey-design literature.

5. Mock catalogues and forward models

Because BGS is flux-limited, dense, and sensitive to both luminosity-dependent clustering and survey incompleteness, it motivated a substantial mock-catalogue program. An early high-fidelity reference mock, Rosella, populated the P-Millennium simulation at the snapshot $>10$4 using subhalo abundance matching (SHAM) on $>10$5 with luminosity-dependent scatter. It assigned rest-frame $>10$6-band absolute magnitudes and $>10$7 colors, the latter through a formation-redshift-based age-matching scheme. Rosella was designed as a single-snapshot BGS mock at the median survey redshift and intended for approximate-mock calibration, systematics testing, and interpretation of non-cosmology-focused BGS analyses (Safonova et al., 2020).

A complementary approach used AbacusSummit simulations and a fast HOD-fitting framework tailored to flux-limited samples. Rather than fitting independent HODs for separate thresholds, the method simultaneously fit a family of absolute-magnitude threshold samples, roughly from $>10$8 to $>10$9, using 17 meta-parameters to enforce physically nested samples and avoid unphysical HOD crossing. The resulting cubic-box and cut-sky mocks were compared to DESI one-percent BGS measurements for the $14{,}000\,\mathrm{deg}^2$0 sample and found to reproduce number densities well, with projected clustering described as reasonable but improvable by fitting directly to BGS clustering measurements (Smith et al., 2023).

By DR2, the DESI collaboration had also developed Uchuu-BGS reference mocks. These focused on BGS-BRIGHT ($14{,}000\,\mathrm{deg}^2$1) and used SHAM on the large Uchuu simulation, with $14{,}000\,\mathrm{deg}^2$2 as the halo proxy and a luminosity-dependent scatter calibrated to reproduce BGS clustering. For the full flux-limited BGS-BRIGHT sample, the Uchuu mock reproduced the observed redshift evolution of clustering with better than 5\% agreement for $14{,}000\,\mathrm{deg}^2$3 and below 10\% for $14{,}000\,\mathrm{deg}^2$4. It also matched luminosity-dependent BGS clustering in volume-limited subsamples and yielded a bias–luminosity relation consistent with the Y3 data, with fitted parameters $14{,}000\,\mathrm{deg}^2$5, $14{,}000\,\mathrm{deg}^2$6, and $14{,}000\,\mathrm{deg}^2$7 for the mock, versus $14{,}000\,\mathrm{deg}^2$8, $14{,}000\,\mathrm{deg}^2$9, and $14{,}000\,\mathrm{deg}^2$00 in the data (Fernández-García et al., 2 Jul 2025).

Taken together, these efforts indicate that BGS has functioned as a forcing case for mock construction: its flux limit, low-redshift leverage, and luminosity dependence require forward models that simultaneously control $14{,}000\,\mathrm{deg}^2$01, absolute magnitudes, color distributions, and real- and redshift-space clustering.

6. Extensions to galaxy evolution, AGN, and velocity-field studies

The scientific reach of BGS extends well beyond BAO and RSD. In the PAC framework, the DESI Y1 BGS Bright sample served as the spectroscopic tracer population around which the excess surface density of DECaLS photometric galaxies was measured. With an effective overlap area of 5349 deg$14{,}000\,\mathrm{deg}^2$02 and average completeness about 0.656, this analysis combined $14{,}000\,\mathrm{deg}^2$03 from photometric–spectroscopic cross-correlations with $14{,}000\,\mathrm{deg}^2$04 measured from BGS to infer the galaxy stellar mass function down to $14{,}000\,\mathrm{deg}^2$05 for blue galaxies and $14{,}000\,\mathrm{deg}^2$06 for red galaxies. The fitted low-mass slopes were $14{,}000\,\mathrm{deg}^2$07 and $14{,}000\,\mathrm{deg}^2$08, with red galaxies becoming dominant below $14{,}000\,\mathrm{deg}^2$09 (Xu et al., 3 Mar 2025).

BGS has also become a platform for AGN-recovery work. The standard Gaia/Tractor star–galaxy separation rejects many quasar host galaxies and point-source-dominated AGN, so a dedicated BGS-AGN selection was introduced beginning with SV3. Using optical–IR cuts such as

$14{,}000\,\mathrm{deg}^2$10

plus WISE-quality and magnitude constraints, the resulting sample was found to be uniformly distributed over the DESI footprint at roughly 3–4 targets per square degree and spectroscopically dominated by quasars, with about 93–94\% QSOs, $14{,}000\,\mathrm{deg}^2$11 narrow-line AGN/blazar-like objects, and $14{,}000\,\mathrm{deg}^2$12 each of galaxy and stellar contamination. Its redshift distribution peaks around $14{,}000\,\mathrm{deg}^2$13, intermediate between the quasars surviving ordinary BGS cuts and the higher-redshift DESI QSO sample (Juneau et al., 2024).

BGS spectroscopy has also been used as an astrophysical discovery space in its own right. An unsupervised outlier search in the DESI Early Data Release assembled roughly 250,000 BGS spectra with reliable Redrock redshifts, compressed them with a redshift-invariant autoencoder into a six-dimensional latent space, and used a normalizing flow to identify low-probability objects. The highest-ranked outliers included mergers, blends, irregular or double-peaked emission-line systems, rare quasar types, and one previously unknown Broad Absorption Line system; a significant fraction were stars spectroscopically misclassified as galaxies, leading the authors to argue that the issue likely stemmed from the PCA-based stellar model in the DESI pipeline (Liang et al., 2023).

At the interface between large-scale structure and baryon physics, BGS has become a low-redshift tracer for kinematic Sunyaev–Zel’dovich studies. A DR1 proof-of-concept matched 1.6 million BGS galaxies with $14{,}000\,\mathrm{deg}^2$14 to ACT DR6 CMB maps and measured the pairwise kSZ signal, reaching about $14{,}000\,\mathrm{deg}^2$15 in the best $14{,}000\,\mathrm{deg}^2$16, $14{,}000\,\mathrm{deg}^2$17 configuration and inferring $14{,}000\,\mathrm{deg}^2$18 at $14{,}000\,\mathrm{deg}^2$19 after optical-depth calibration (Hadzhiyska et al., 15 Oct 2025). A DR2 analysis then used the BGS_BRIGHT-20.2 sample with $14{,}000\,\mathrm{deg}^2$20, mean redshift $14{,}000\,\mathrm{deg}^2$21, and about 2.26 million galaxies, combining velocity reconstruction with ACT DR6 temperature and lensing maps. It reported BGS kSZ detections with signal-to-noise ratios up to $14{,}000\,\mathrm{deg}^2$22, a fiducial velocity-reconstruction correlation coefficient $14{,}000\,\mathrm{deg}^2$23, and projected gas-fraction proxies near the virial radius of order

$14{,}000\,\mathrm{deg}^2$24

interpreted as evidence for substantial baryon displacement by feedback (Hadzhiyska et al., 21 Apr 2026).

Further applications use BGS as a parent spectroscopic sample for faint-satellite studies. Around isolated central galaxies drawn from DESI Year-1 BGS, photometric satellite luminosity functions were measured down to $14{,}000\,\mathrm{deg}^2$25 and stellar mass functions down to $14{,}000\,\mathrm{deg}^2$26, with the faint-end slopes steepening toward lower-mass hosts and the steepest reported values equal to $14{,}000\,\mathrm{deg}^2$27 for the satellite LF and $14{,}000\,\mathrm{deg}^2$28 for the satellite SMF (Wang et al., 5 Mar 2025).

7. Other uses of the term

Outside DESI, Bright Galaxy Sample has been used for Herschel cluster-survey subsamples that are unrelated to the DESI program. In the Herschel Virgo Cluster Survey, the BGS denotes a 500 $14{,}000\,\mathrm{deg}^2$29m-selected, optically confirmed sample of 78 bright Virgo galaxies, each detected in all five Herschel bands $14{,}000\,\mathrm{deg}^2$30. That work found peaked far-infrared luminosity distributions rather than power laws, a mean optical depth $14{,}000\,\mathrm{deg}^2$31, and dust SEDs well fit by a single modified blackbody with $14{,}000\,\mathrm{deg}^2$32, yielding mean dust mass $14{,}000\,\mathrm{deg}^2$33 and mean dust temperature $14{,}000\,\mathrm{deg}^2$34 (Davies et al., 2011).

The Herschel Fornax Cluster Survey adopted the same terminology for a directly comparable 11-galaxy 500 $14{,}000\,\mathrm{deg}^2$35m-selected sample in Fornax. It reported a mean optical depth again equal to $14{,}000\,\mathrm{deg}^2$36, dust masses in the range $14{,}000\,\mathrm{deg}^2$37, dust temperatures $14{,}000\,\mathrm{deg}^2$38, and a far-infrared luminosity density about a factor of 3 higher than Virgo, largely because of NGC 1365 (Davies et al., 2012).

This terminological overlap can cause confusion in bibliographic searches. In current large-scale-structure and DESI contexts, however, BGS almost always refers to the DESI Bright Galaxy Survey rather than the Herschel cluster samples.