The DESI Experiment Part I: Science,Targeting, and Survey Design

1 Overview

• The instrument can take up to 5,000 simultaneous spectra across 360–980 nm using ten three-arm spectrographs on the Mayall telescope.Its resolving power ranges from R = 2000 to 5500 depending on wavelength.Source: §1 · p. 5
• DESI plans a five-year, 14,000 deg2 survey targeting LRGs to z = 1.0, ELGs to z = 1.7, quasars, and about 10 million BGS galaxies.The BGS is observed when moonlight prevents efficient observations of the faint baseline targets; the minimum survey spans 9,000 deg2 and still meets Stage-IV requirements.Source: §1 · p. 5
• More than 30 million redshifts will support BAO and matter-power-spectrum measurements, including redshift-space distortions.The BGS has median z ≈ 0.2, while high-redshift quasars at 2.1 < z < 3.5 provide Ly-α forest absorption tracing neutral hydrogen.Source: §1 · p. 5
• The DETF FoM is 133 for DESI galaxy BAO, more than three times the combined Stage-III galaxy BAO FoM.It reaches 169 with Ly-α forest BAO, 332 with galaxy broadband power to k = 0.1 h Mpc−1, and 704 when that power extends to k < 0.2 h Mpc−1.Source: §1 · p. 5
• DESI provides at least an order-of-magnitude improvement over BOSS in both comoving volume and the number of mapped galaxies, while also enabling constraints on modified gravity and inflation.The survey’s legacy data are expected to support a wider range of projects, and DESI will be a pathfinder for spectroscopic follow-up of imaging surveys such as LSST.Source: §1 · pp. 5–6
• 0.020 eV is the forecast uncertainty on the sum of neutrino masses for kmax < 0.2 h Mpc−1, sufficient for a 3-σ detection under the stated normal-hierarchy, minimal-mass assumption.Under that assumption, DESI can rule out the inverted hierarchy at 99% CL.Source: §1 · pp. 5–6

2.1 Introduction

• DESI’s projected lookback-time measurements illustrate the precision needed to distinguish five constant-w models spanning −0.97 to −1.03.The figure cautions that it varies only one cosmological parameter; full forecasts must marginalize over parameters such as Ωm and H0.Source: Figure 2.1 · p. 8
• Cosmic acceleration may reflect a cosmological constant with w = −1, dynamical dark energy with w(a) ≠ −1, or a failure of General Relativity.The section frames DESI’s goal as distinguishing the cosmological-constant case from dark energy with w near −1.Source: §2.1 · p. 8
• DESI’s more than thirty million spectroscopic redshifts will map matter over 0.4 < z < 3.5, with a denser BGS sample at z < 0.4.These data are intended to help distinguish a mysterious component of the Universe from a cosmic-scale modification of General Relativity and constrain primordial inflation models.Source: §2.1 · pp. 8–9
• BAO localizes a known 150 Mpc feature in galaxy pair separations, yielding distances and the expansion history; RSD anisotropies trace gravity through galaxy motions.BAO depends on galaxy locations rather than shapes or brightnesses, while RSD arise from peculiar velocities generated by density perturbations.Source: §2.1 · p. 9
• The report presents galaxy and Ly-α flux power spectra as probes of neutrino mass, primordial scale dependence, and possible modified gravity.Realizing these probes requires many redshifts across a deep, wide volume and a new instrument with capabilities beyond existing facilities.Source: §2.1 · p. 9
• The report does not discuss DESI’s broader galaxy, quasar, and stellar science because those topics do not drive its requirements.The Milky Way Survey is a brief exception because it uses targets that piggyback on the BGS.Source: §2.1 · p. 9

2.2.1 Theory

• The BAO ruler is about 150 Mpc, and its observed angular and radial scales provide measurements of D_A(z) and H(z).The ruler’s size is calculable to 0.3% from CMB-constrained cosmological parameters; measuring both distances helps distinguish cosmological models.Source: §2.2.1 · p. 10

2.2.2 BAO in Galaxies

• The BOSS z > 0.45 sample measured D_V(z) to z = 0.57 at 1.0% precision, while reconstruction improves statistical precision and suppresses distance-scale shifts.Reconstruction tests show shifts suppressed below 0.1%; SDSS-II demonstrated a 1.7-fold improvement in statistical precision.Source: §2.2.2 · pp. 11–12, Figure 2.2 · p. 11

Observational Systematics

• Fiber-positioning incompleteness in dense regions and sky-line interference can alter measured clustering, though both effects are expected to be correctable to good accuracy.The fiber effect can skew anisotropic correlations, while sky brightness can reduce [O II] completeness at specific redshifts.Source: Observational Systematics · p. 12

Theoretical Systematics

• The BAO scale is altered by 0.3% by nonlinear density evolution at the present epoch, with smaller effects expected at DESI’s higher redshifts.Galaxy-formation shifts can reach about 0.5% in extreme cases, while reconstruction reduces these shifts to 0.1%; a relative-velocity effect is expected to be negligible for DESI galaxies but cannot be ruled out theoretically.Source: Theoretical Systematics · p. 13
• BAO-scale theoretical systematics are argued to be intrinsically small or correctable to below DESI’s required 0.1% level.The discussion includes nonlinear evolution, galaxy-formation shifts, and a possible baryon–dark-matter relative-velocity effect.Source: Theoretical Systematics · p. 13

2.2.3 BAO in the Ly-α Forest

• Ly-α forest flux fluctuations trace neutral hydrogen and, on BAO-relevant scales, provide three-dimensional correlations for measuring BAO beyond redshift 2.The forest consists of absorption features blueward of quasar Ly-α emission; simulations and analytical work support linear biasing on the relevant scales.Source: §2.2.3 · p. 13
• 5σ BAO detection in the Ly-α forest was reported using the nearly complete BOSS sample.Ly-α measurements at z ≳ 2 provide access to the high-redshift expansion history; their radial modes have higher signal-to-noise than transverse modes, making H(z) more precise than D_A(z).Source: §2.2.3 · pp. 13–14

Systematics

• Ly-α systematics may leave isolated BAO features measurable while contaminating broadband power, neutrino-mass, and warm-dark-matter analyses if not carefully modeled.Potential contaminants include non-gravitational fluctuations, targeting selection, and metal absorption; the text describes modeling and simulation approaches for addressing them.Source: Systematics · pp. 14–15

2.3.1 Theory

• Peculiar velocities make redshift-inferred galaxy clustering anisotropic, and the resulting signal constrains growth; because f is degenerate with power-spectrum amplitude, the observable is fσ8.The Kaiser relation describes the linear-regime anisotropy, while fσ8 combines the growth rate with the power-spectrum normalization.Source: §2.3.1 · pp. 16–17
• The Bright Galaxy Survey maps z < 0.4 at higher number density and with more diverse galaxies, using multiple tracers to suppress sample variance and test modified gravity.The survey also supports tests using cluster and group abundances, weak-lensing maps, velocity fields, and screening on smaller scales.Source: §2.3.1 · p. 17

2.3.2 Systematics

• Nonlinear velocities, infall motions, and scale-dependent galaxy bias complicate RSD modeling; these effects can reach 10% at k ∼ 0.1 h/Mpc, while models reproduce them at percent level for k < 0.05–0.1 h/Mpc.Improving bias models and simulation realism is needed to extend modeling to smaller scales.Source: §2.3.2 · pp. 17–18
• Estimating the radial selection function from observed redshifts removes some cosmological radial modes and biases monopole–quadrupole amplitudes by < 0.2σ.The systematic-to-statistical uncertainty ratio is expected to remain relatively constant with survey area for a given redshift distribution.Source: §2.3.2 · p. 18

2.3.3 Current Status of RSD Measurements

• RSD surveys measure growth to about 3% at low redshift, and adding BOSS DR11 quadrupole information tightens the flat-wCDM constraint on w from −1.01 ± 0.08 to −0.993 ± 0.056.Most measurements use anisotropy in two-point galaxy-density correlations, often compressed into multipoles or wedges.Source: §2.3.3 · pp. 18–19
• 16% constraint on the growth index from BOSS DR11; DESI is forecast to improve growth-constraint precision by ∼4–10× and reach significantly higher redshifts.The right panel of Figure 2.4 shows the BOSS DR11 constraints in the Ωm–γ plane.Source: Figure 2.4 · p. 19, §2.3.3 · p. 20

2.4.1 Forecasting Overview

• The Fisher forecasts use flat ΛCDM; isolated BAO gives robust but pessimistic constraints, while broadband power recovers additional two-point clustering information.The broadband treatment includes power-spectrum shape, redshift-space distortions, Alcock-Paczynski information, and BAO.Source: §2.4.1 · pp. 21–22
• Broadband forecasts bracket recoverable nonlinear-scale information with kmax = 0.1 and 0.2 h Mpc−1, the latter described as a stretch goal.How well nonlinear corrections and bias marginalization can recover information will determine parameter-measurement precision.Source: §2.4.1 · p. 23
• Aggregate DESI fσ precision is about 0.74% for kmax = 0.1 h Mpc−1 and 0.38% for kmax = 0.2 h Mpc−1.The forecasts use two kmax assumptions to represent different levels of recovered broadband information.Source: §2.4.1 · p. 23

2.4.2 Baseline Survey

• The baseline assumes a five-year survey over 14,000 deg2 targeting BGS, LRGs, ELGs, and quasars; a 9,000 deg2 minimum survey is also forecast.Tracer number densities follow selection criteria, and the forecasts assume constant b(z)D(z) for each target class.Source: §2.4.2 · p. 23

2.4.3 Summary of Forecasts

• FoM 121 for the 9,000 deg2 survey with galaxy and Ly-α forest BAO exceeds the Stage IV threshold of 110; DESI galaxy BAO alone also exceeds it.This 9,000 deg2 survey is the quantitative basis for the Level 1 Science Requirement, with BAO performance aggregated across three redshift ranges.Source: §2.4.3 · pp. 25, 29, Table 2.9 · p. 29
• Forecasted growth measurements can distinguish ΛCDM from the scale-dependent f(R) predictions at two scales and from DGP.The projected expansion-rate measurements also extend into early epochs when matter, rather than dark energy, controlled the rate.Source: §2.4.3 · pp. 25, 29, Figure 2.12 · p. 30

Ly-α Forest

• Ly-α forecasts are limited to z = 2–2.7 and are described as conservative; BAO forecasts extend the forest wavelength range to 985–1200 Å to include Ly-β.The wavelength expansion applies to BAO only; broadband forecasts vary bias and damping parameters and include one-dimensional power-spectrum information.Source: Ly-α Forest · pp. 30–31

2.5 Cosmology Beyond Dark Energy

• Broad power-spectrum measurements and scale-dependent halo bias extend DESI’s BAO and RSD program to tests of inflation, neutrino masses, and possible modified gravity.Large-scale structure measurements complement CMB tests of inflation, which the text says are not expected to improve greatly after Planck.Source: §2.5 · p. 32, §2.5.1 · p. 32

Spectral Index and Its Running

• Up to a factor of three improvement over Planck alone in n_s constraints is forecast, assuming no significant running; excellent control of Ly-α forest and galaxy systematics is required.For V(φ) ∝ φ^m models, the improved spectral-index measurement would determine inflation’s duration more precisely; a running detection above slow-roll expectations would imply multiple fields, broken slow roll, or non-canonical kinetic terms.Source: Spectral Index and Its Running · p. 33, Table 2.10 · p. 33

Primordial non-Gaussianity

• DESI’s large-scale clustering is forecast to constrain local primordial non-Gaussianity competitively with Planck; the projected DESI-only error is σ(fNL) ≃ 5.The method uses the scale-dependent halo bias, whose local-model signature varies as k^-2.Source: Primordial non-Gaussianity · p. 34
• DESI forecasts σ(fNL) ≃ 5 for the local model alone, with about a factor-of-two better precision when combined with final Planck temperature and polarization data.The authors note that these constraints may also detect late-time nonlinear photon-baryon interactions with gravity.Source: Primordial non-Gaussianity · p. 34
• For running non-Gaussianity, DESI plus Planck is forecast to shrink the f∗NL–nfNL constraint area by about 100×, contingent on sufficiently biased galaxies and further halo-mass and systematics studies.Combining high- and low-bias tracers may further reduce sampling variance, though low number density may require additional structure tracers.Source: Primordial non-Gaussianity · p. 35

2.5.2 Neutrinos

• Neutrino masses alter background expansion and suppress small-scale fluctuations by a few percent below the free-streaming scale, enabling mass limits.The free-streaming scale is roughly the distance a typical neutrino travels while relativistic.Source: Neutrino Mass · p. 36

Neutrino Mass

• 0.020 eV projected resolution would measure the neutrino-mass sum independently of hierarchy; for minimal masses and a normal hierarchy, DESI could exclude the inverted hierarchy at 2σ.The hierarchy-independent measurement would determine the absolute neutrino mass scale.Source: Neutrino Mass · pp. 36–37

Dark Radiation (e.g., sterile neutrinos)

• DESI forecasts Nν,eff precision of about 10% or better, constraining alternatives involving sterile neutrinos, axions, or partly thermalized species.The standard-model value quoted is Nν,eff = 3.043.Source: Dark Radiation (e.g., sterile neutrinos) · p. 37

2.6 The Milky Way Survey: Near-Field Cosmology from Stellar Spectroscopy

• At magnitude 17, 8–10-minute DESI exposures yield stellar spectra with S/N = 25 per pixel, radial velocities to a few km/s, and metallicities.The BGS is expected to generate at least 10 million such spectra; individual stellar spectra also provide effective temperature, surface gravity, chemical abundances, and approximate age.Source: §2.6 · p. 39
• DESI can extend Gaia’s stellar radial-velocity and metallicity coverage to fainter stars, using its higher multiplex, wide field, and rapid reconfiguration.Gaia’s RVS provides radial velocities for millions of brighter stars, while its flux limit was still under investigation in the passage.Source: §2.6 · p. 39
• Stellar measurements of the Milky Way’s outer rotation, tidal-stream motions, and distant-halo kinematics can substantially reduce mass-profile uncertainties that affect direct and indirect dark matter searches.The section identifies these uncertainties as important systematics in interpreting dark matter searches.Source: §2.6 · p. 39
• Joint metallicity and velocity distributions can trace outer-disk assembly and thick-disk formation, while halo turn-off stars out to 30 kpc can probe the Galaxy’s first two billion years.A survey of millions of stars also offers potential to discover kinematically and chemically peculiar stars in unexplored Galactic regions.Source: §2.6 · p. 39

2.7.1 Synergies with Planck and Future CMB Experiments

• CMB cross-correlations with DESI can probe the ISW effect, improve CMB-lensing constraints, test equality of gravitational potentials, and complement neutrino-mass measurements.DESI’s squeezed-limit bispectrum information complements CMB 3-point measurements for studying inflation; galaxy-velocity cross-correlation with the CMB can measure the kSZ effect at the percent level.Source: §2.7.1 · p. 40

2.7.2 Synergies of DESI with DES and LSST

• Combining DESI with DES or LSST imaging can improve photometric-redshift calibration and enable joint lensing and clustering measurements, including tests suited to modified-gravity models.Galaxy-galaxy lensing measures the spectroscopic sample’s galaxy–mass cross-correlation, probing host halos on small scales and matter clustering on larger scales.Source: §2.7.2 · p. 41
• DESI BAO and DES/LSST supernova samples provide complementary distance-redshift information, with BAO calibrating relative supernova distances and redshift overlap offering a systematic cross-check.The text describes supernovae as especially effective at low redshift, where BAO is more limited by cosmic variance.Source: §2.7.2 · p. 41
• DESI will provide spectroscopic redshifts for many tens of thousands of supernova host galaxies, including more than half of SN Ia hosts in its footprint at z < 0.2 before explosions occur.Pre-existing redshifts can also improve follow-up resource allocation for rare transients.Source: §2.7.2 · pp. 41–43
• DESI BAO with LSST weak lensing is forecast to constrain neutrino-mass sum versus dark-energy density and the Ωm–ΩΛ plane; the figures report stronger constraints using DESI’s full power spectrum.The forecasts assume non-overlapping survey areas, which the text says makes practically no difference.Source: §2.7.2 · p. 41, Figure 2.16 · p. 42, Figure 2.17 · p. 42

2.7.3 Synergies of DESI with Euclid/WFIRST

• DESI’s z < 1 galaxy samples and z > 2 Ly-α measurements complement Euclid and WFIRST, while the space surveys’ weak lensing and imaging support joint analyses and DESI target science.Euclid may surpass DESI at 1 < z < 2 if its slitless spectroscopy works as hoped; WFIRST-AFTA’s smaller area implies weaker direct expansion- and growth-rate measurements.Source: §2.7.3 · p. 43

3 Target Selection

• The baseline survey assumes 14,000 deg2, with target-class bands, redshift ranges, densities, and successful-redshift sample sizes summarized in Table 3.1.Sample sizes scale with survey area at essentially the same surface densities in reduced focal-plane simulations.Source: §3 · p. 44, Table 3.1 · p. 44

Summary of Target Samples

• DESI targets BGS galaxies at 0.05 < z < 0.4, LRGs to about z = 1.0, ELGs to z = 1.6, and QSOs for clustering or Ly-α forest measurements above z = 2.1.The target classes are selected for spectral features expected to support reliable redshifts or Ly-α measurements in DESI’s wavelength range.Source: Summary of Target Samples · pp. 44–45, §3 · p. 44

Summary of Required Imaging

• Target selection uses ground-based optical grz and WISE infrared photometry, with planned imaging depths of 24.0, 23.4, and 22.5 AB in g, r, z and 20.0, 19.3 AB in W1, W2.The imaging plan uses DECam, Bok, and Mayall observations; DESI analyzes three footprint regions separately before combining cosmological constraints.Source: Summary of Required Imaging · pp. 45–46, Table 3.2 · p. 46

3.1.1 Overview of the Sample

• BGS mocks match MGS properties at low redshift and use GAMA and DEEP2 data for higher-redshift calibration; the flux-limited sample is set by bright time and exposure needs.The mock catalogs explore BGS target-sample properties.Source: §3.1.1 · p. 47

3.1.2 Sample Properties

• At r-band limit 19.5, the BGS mock predicts just over 800 targets deg−2, above the 700 targets deg−2 goal.Its redshift distribution peaks near z = 0.2 and has a tail beyond z = 0.4; its space density exceeds MGS and BOSS through z = 0.4.Source: §3.1.2 · p. 47, Redshift Distribution · pp. 47–48, Figure 3.2 · p. 48

Redshift measurement method

• At a 10-day lunar phase, BGS redshift success is 96% at 6 minutes and at least 99% at 9 minutes; star-forming galaxies reach about 98% at 9 minutes.Success decreases with increasing moon illumination, and star-forming targets have lower success than passive galaxies at fixed conditions.Source: Redshift measurement method · p. 49

Large-scale-structure bias

• Abundance matching predicts BGS bias near unity at low redshift and rising with redshift; increasing luminosity scatter can place mean halo mass below the cutoff mass for the brightest galaxies.Mcut is the mass at which halos have a 50% probability of containing a selected galaxy; the scatter at fixed luminosity increases with luminosity.Source: Large-scale-structure bias · pp. 49–50, Figure 3.4 · p. 50

Target selection efficiency

• Fiber-assignment inefficiency is the dominant target loss, while imaging artifacts lose a few percent; green-valley and low-surface-brightness galaxies may also be incomplete under bright-time conditions.The BGS uses three observing layers to achieve fairly high completeness, while green-valley incompleteness depends on exposure time and observing conditions.Source: Target selection efficiency · p. 50, Areas of risk · p. 51

3.2.1 Overview of the Sample

• Low-redshift LRGs are efficiently selected from SDSS-depth photometry, while targets above z = 0.6 require near-infrared data as their optical colors overlap those of red stars.BOSS-like target selection is used where existing BOSS spectroscopy does not cover the DESI footprint.Source: §3.2.1 · p. 51

3.2.2 Selection Technique for z > 0.6 LRGs

• At z near 1, WISE W1 overlaps the LRG 1.6 µm spectral bump, enabling selection with r − W1 color supplemented by r − z to reject non-LRGs.The bump is the peak in flux density for stellar populations older than about 500 Myr.Source: §3.2.2 · p. 51, Figure 3.5 · p. 52, Figure 3.6 · p. 52

3.2.3 Sample Properties

• The baseline selection uses DECam r- and z-band and WISE W1 photometry, and is already sufficient to meet all DESI design requirements.The sample is limited to zAB = 20.46; selected targets are expected to satisfy r < 23 and W1 < 19.5.Source: §3.2.3 · p. 53

Surface Density

• 350 LRG targets/deg2 are reached at zAB = 20.46, with a mean of two spectroscopic visits per object required for the baseline sample.A conservative completeness estimate based on eBOSS experience is 90–95%, with negligible effect on cosmological constraint forecasts.Source: Surface Density · pp. 53–54, Figure 3.7 · p. 53

Redshift Distribution

• The selection meets or exceeds baseline sample requirements despite redshift-distribution estimates being limited by COSMOS sample variance and missing faint high-redshift ancillary redshifts.The sample has a particularly large density below z = 0.8, where apparent magnitude can be used to sculpt its redshift distribution.Source: Redshift Distribution · p. 54, Figure 3.8 · p. 55

Redshift measurement method

• DESI exposure times are set to match the signal-to-noise achieved by the BOSS ancillary program, supporting expected secure redshifts for more than 98% of targets.The 4000 Å break is covered by the DESI spectrograph to z = 1.45.Source: Redshift measurement method · p. 54

Large-scale-structure bias

• The assumed high-redshift LRG bias is b(z) = 1.7/D(z), extrapolating the SDSS-measured trend to z = 1.The selection efficiency estimate combines redshift success, stellar contamination, redshift range, and red-sequence fractions; COSMOS representativeness and redshift-success scaling remain risks.Source: Large-scale-structure bias · pp. 54–55, Target selection efficiency · p. 55, Areas of risk · pp. 55–56

Areas of risk

• The baseline plan yields about 315 LRGs/deg2 from 350 targets/deg2, with conservative projections of 300 z > 0.6 LRGs/deg2 for BAO measurements across 0 < z < 1.The sample also enables direct comparisons with ELG results in overlapping redshift ranges as a test for systematic effects.Source: Areas of risk · p. 56

3.3.1 Overview of the sample

• ELGs are star-forming galaxies with strong nebular emission, and DESI is designed to resolve their [O II] doublet across 0.6 < z < 1.6.Resolving the doublet avoids ambiguity with single emission lines in lower-resolution spectra.Source: §3.3.1 · p. 56, Figure 3.9 · p. 57

3.3.2 Selection Technique for z > 0.6 ELGs

• The grz color box selects strong [O II] emitters over 0.6 < z < 1.6 while balancing z ∼ 1 target yield against contamination from stars and lower-redshift galaxies.The bluest, flat-spectrum galaxies are excluded because their colors are difficult to use for redshift selection.Source: §3.3.2 · p. 57, Figure 3.10 · p. 58

3.3.3 Sample Properties

• Selection criteria based on DEEP2/EGS data, cross-checked with COSMOS and VVDS, meet DESI science requirements at anticipated DECam Legacy imaging depths.The DEEP2 objects were fainter and had higher spectroscopic signal-to-noise than DESI targets.Source: §3.3.3 · p. 58

Surface Density

• At rAB ≈ 23.4, the ELG selection reaches its goal of 2400 targets/deg2, of which at least 65% are conservatively estimated to be bona fide ELGs with secure-redshift-capable emission.Differences between DECaLS and transformed CFHTLS-Deep density curves are most likely due to photometric-transformation scatter.Source: Surface Density · pp. 58–59, Figure 3.11 · p. 59

Redshift Distribution

• The candidate ELG sample exceeds the n̄P = 1 threshold to z ∼ 1.3, beyond which shot noise dominates BAO measurement errors.For z > 1.4, the estimated redshift distribution uses a power-law extrapolation because DEEP2 generally lacks [O II] redshifts there.Source: Redshift Distribution · p. 59, Figure 3.12 · p. 60

Redshift measurement method

• DESI simulations predict > 7σ emission-line detections for the bulk of the targeted sample above the limiting [O II] flux, while lower-redshift galaxies can also show Hβ and [O III].At z > 1.0, [O II] is the only strong spectroscopic feature.Source: Redshift measurement method · p. 60, Figure 3.13 · p. 61

Large-Scale Structure Bias

• DEEP2 clustering measurements support an ELG bias of b(z) = 0.84/D(z), with observed galaxy clustering constant within errors across the measured redshift bins.Source: Large-Scale Structure Bias · p. 60

Areas of risk

• DESI expects to securely measure redshifts for approximately 1220 ELGs/deg2 in 0.6 < z < 1.6 from 2400 targets/deg2.This sample supports cosmological constraints centered near z ≈ 1 and direct comparison with LRG and quasar samples.Source: Target selection efficiency · p. 61, Areas of risk · p. 62

3.4.1 Overview of the sample

• DESI targets 170 QSOs per deg2, including 50 per deg2 at z > 2.1 for Ly-α forest measurements, while using lower-redshift QSOs as matter tracers.This extends the QSO role beyond BOSS, which selected QSOs only for Ly-α forest use.Source: §3.4.1 · p. 62

3.4.2 Selection Technique

• Optical+IR color cuts identify unresolved sources to r = 23, using QSO infrared excess to separate them from stars; morphology cuts avoid nearly tenfold galaxy contamination.The color selection is supplemented by an investigated Random Forest approach, which performs better particularly above z = 2.1 or at faint magnitudes.Source: §3.4.2 · pp. 63, 65, Figure 3.16 · p. 64

3.4.3 Sample Properties

• The baseline color-box selection yields about 140 QSOs per deg2 at z < 2.1 and 40 per deg2 at z > 2.1 from 260 targets per deg2; average selection efficiency after four passes is about 80%.The Random Forest selection estimates 55 z > 2.1 QSOs per deg2 and 67% completeness.Source: §3.4.3 · pp. 65–66
• The expected QSO redshift distribution uses measured color-selection completeness for r < 23, and about 10% of targets are expected to yield no redshift on the first visit.Confirmed z > 2.1 QSOs are re-observed for higher signal-to-noise Ly-α spectra; clustering at those redshifts uses forest transmitted flux rather than object correlations.Source: §3.4.3 · pp. 65–66, Figure 3.17 · p. 66

3.4.4 Recent and near-term developments for QSO target selection

• Pilot programs built QSO truth samples and updated the faint-end luminosity function; DESI also began developing machine-learning selection with DECaLS optical and WISE infrared data.The pilot sample included 18,000 spectroscopically confirmed QSOs over 120 deg2 to gc < 22.5.Source: §3.4.4 · p. 67

3.4.5 Variability Data Improves Selection of High-Redshift QSOs

• Variability selection found 30% more QSOs in Stripe 82 than time-averaged optical photometry alone, while preliminary four-epoch WISE tests indicate about 10% lower target contamination.The gain over baseline targeting with full WISE depth remains to be evaluated; variability is not part of baseline targeting but may help select high-redshift QSOs where coverage exists.Source: §3.4.5 · pp. 67–68, Figure 3.18 · p. 68

• Main-sequence F stars are the primary spectrophotometric standards because stellar atmosphere models describe them well; individual templates support spectral-response and atmospheric-absorption calibration.Templates account for each star's temperature, surface gravity, and metallicity.Source: §3.5 · p. 68
• DESI broadens the BOSS low-metallicity color selection to include higher-metallicity F stars, then plans to assess the mix and cross-calibrate the groups during commissioning.The restrictive BOSS selection yields 10 stars per deg2.Source: §3.5 · p. 68
• Blank-sky locations are selected to contain no detectable sources within a fiber diameter in any input band, at a density intended to offer each unassigned fiber a sky option when possible.Source: §3.5 · pp. 68–69

3.6 Baseline Imaging Datasets

• Three telescope programs provide DESI baseline imaging, with DECam covering about two-thirds of the footprint at Dec ≤ 34° and Bok and Mayall imaging the northern region.The targeting catalogs combine optical g, r, z imaging and WISE data.Source: §3.6 · p. 69, §3.6.1 · pp. 69–70

3.6.1 Blanco/DECam Surveys (DEC≤34◦)

• DECam can reach DESI target depths in 100, 100, and 200 seconds in g, r, and z, respectively, and image 9000 deg2 in 81 scheduled nights under the stated observing strategy.DECaLS and other public DECam datasets contribute coverage, with data reprocessed uniformly for target selection.Source: §3.6.1 · pp. 70–71

3.6.2 Bok/90Prime Survey (DEC≥34◦)

• Bok/90Prime provides g- and r-band imaging for the northern footprint; its BASS survey targets 5500 deg2, including 500 deg2 overlapping DECam coverage to assess and correct selection biases.Early data had readout-electronics defects, which were replaced in September 2015.Source: §3.6.2 · pp. 71–72

3.6.3 Mayall/MOSAIC Survey (DEC≥34◦)

• The Mayall/MOSAIC-3 survey supplies northern z-band imaging, and its z-band efficiency is 60% higher than that of MOSAIC-1.1.The survey uses three passes, with at least one scheduled in photometric conditions and seeing better than 1.3 arcsec.Source: §3.6.3 · p. 72

3.6.4 W ISE All-Sky Survey

• DESI uses WISE W1 and W2 photometry, and final Legacy Survey catalogs are expected to reach 0.7 mag fainter than Data Release 1 or the WISE All-Sky Data Release.The catalogs use re-stacked WISE and NEOWISE imaging.Source: §3.6.4 · pp. 72–73, Figure 3.21 · p. 73

3.7 Additional Imaging Data

• Nonuniform additional datasets can aid high-redshift QSO selection because Ly-α forest analysis uses absorption clustering along a line of sight rather than requiring a spatially uniform QSO sample.Their use depends on availability and could improve targeting efficiency beyond the baseline strategy.Source: §3.7 · p. 74

3.7.1 SDSS

• SDSS photometry may provide a reference point for measuring high-redshift QSO variability over long time baselines.Its extragalactic footprint covers 10,000 deg2, with substantial depth variation from seeing.Source: §3.7.1 · p. 74

3.7.2 PanSTARRS-1

• PS1 time-domain photometry may enhance selection of brighter Ly-α QSOs, although its co-added imaging and catalogs were not yet available as of March 2016.Its co-added depths may suit BGS and LRG samples but not ELG or QSO samples.Source: §3.7.2 · p. 74

3.7.3 PTF, iPTF, and ZTF

• PTF/iPTF and ZTF may supplement high-redshift QSO selection, but ZTF's planned nonuniform time sampling makes it unsuitable for baseline targeting.PTF data included 5,000 deg2 in R band and 1,000 deg2 in g band at useful QSO-variability depths.Source: §3.7.3 · pp. 74–75

3.7.4 CFHT

• CFHT-Luau u-band data may aid high-redshift QSO selection because u − g color distinguishes low- and high-redshift QSOs, especially when combined with variability data.The planned survey covers 4000 deg2 of high-Galactic-latitude northern sky.Source: §3.7.4 · p. 75

3.7.5 SCUSS

• SCUSS u-band imaging reaches about u = 23 mag at 5σ and may supplement high-redshift Ly-α QSO selection, especially alongside variability data.The survey covers 4000 deg2 in the South Galactic Cap.Source: §3.7.5 · p. 75

3.8 The Tractor Photometry for Target Selection

• Joint pixel-level fitting across exposures and bands applies a consistent point-source or extended-source light profile to each source, producing comparable fluxes and colors.The approach accounts for differing point-spread functions and pixel sampling across optical and WISE images.Source: §3.8 · p. 76
• Forced WISE photometry agrees with catalog measurements for bright detections while measuring fainter sources and optically resolved blends without catalog matching.The comparisons are illustrated in Figures 3.22 and 3.23; the latter contrasts catalog matching with forced photometry.Source: §3.8 · p. 76, Figure 3.22 · p. 77, Figure 3.23 · p. 77
• For QSO targeting, the Tractor distinguishes WISE-detected from generally undetected quasars where catalog matching would fail, including cases with low-significance WISE measurements.The cited passage notes broad absorption lines, high-redshift Ly-α band dropouts, and optical limits that correspond to below-5σ WISE measurements.Source: §3.8 · p. 76
• DESI adapts the forced-photometry approach used for eBOSS, replacing SDSS profiles with optical images from DECam, Bok, and Mayall surveys.For eBOSS, SDSS galaxy profiles were convolved with the WISE point-spread function and fit to WISE imaging.Source: §3.8 · p. 76
• DECaLS DR2 Tractor catalogs cover about 260 million sources in 97,554 bricks, with full-coverage median 5σ depths of g=24.65, r=23.61, and z=22.84.The passage reports approximately 1,800 deg2 imaged in all three optical filters and WISE forced-photometry fluxes for all extracted sources.Source: §3.8 · pp. 76, 78
• Catalog validation targets uniform photometry with systematic RMS errors below 1% in g and r, below 2% in z, and below 2% from morphological misclassifications.The requirements also specify footprint depth, z-band image quality, and astrometric error thresholds; validation is ongoing.Source: §3.8 · p. 78

4.1 Introduction; 4.2 Survey Footprint

• The survey is optimized for a large Mayall-accessible footprint clear of the Milky Way; its baseline covers 14,000 deg2 in contiguous northern and southern regions using 2,000 tiles per layer and five layers.The footprint is defined within 1.605 degrees of selected tile centers; imaging coverage and observing conditions further constrain its boundaries.Source: §4.2 · p. 79, §4.1 · p. 79

4.3 Field Centers

• Five-layer tiling uses rotated center patterns to fill focal-plane gaps; geometric coverage averages about 5.1 fibers per coordinate, with 3.5% of the footprint below three fibers.A 9,000-square-degree descoped design uses a different tiling solution; the coverage comparison is shown for full and reduced focal planes.Source: §4.3 · pp. 79–81, Figure 4.3 · p. 81, Figure 4.2 · p. 81

4.4.1 Sequence of Observations

• Grey-time observations are planned for ELGs, while the darkest 80% of scheduled time prioritizes QSOs and LRGs; simulations will select tile order to optimize intermediate and final cosmology results.The plan also targets an early complete sample over 10% of the footprint and annual data-product and analysis milestones; survey validation will examine targeting, exposure depth, data quality, and completeness.Source: §4.4.1 · p. 82

4.4.2 Exposure Times and Margin

• For the 14,000-square-degree survey, the simulated average exposure is 1,800 seconds and the zenith-equivalent fiducial time is 1,226 seconds, leaving a projected 22% margin.The estimate assumes 57% of scheduled time yields usable data, defined as dome open with seeing better than 1.5 arcsec; the 9,000-square-degree design has a 27% projected margin.Source: §4.4.2 · p. 83

4.5.1 Introduction; 4.5.2 Survey Footprint

• Bright-time operations are planned for about 250 open-dome hours per year, supporting a Bright Galaxy Survey and a simultaneous Milky Way Survey on the same 14,000-square-degree footprint.BGS targets come from the dark-time imaging data, while MWS selection uses Gaia photometry and proper motions.Source: §4.5.1 · p. 84, §4.5.2 · p. 84

4.5.3 Field Centers

• The three-layer Bright Time Survey plans 6,000 tiles and about 27 million science-fiber assignments for roughly 20 million BGS targets; simulations project 92% completeness for bright and 77% for faint targets.The projected overall fiber completeness is 86%, or roughly 17 million targets; brighter targets receive higher fiber-assignment priority.Source: §4.5.3 · p. 84

4.5.4 Observation Strategy

• Bright-time observations plan 300-second reference exposures, leaving a 33% margin; simulations project redshift success of 97% for galaxies to r=19.5 and 92% for 19.5 < r < 20.0.Five-minute stellar exposures are projected to provide S/N per Angstrom of 14 above 650 nm for stars of r magnitude 16.5–18; the program expects spectra for roughly 10 million BGS galaxies and 10 million MWS stars.Source: §4.5.4 · pp. 84–85
• Because the Bright Time Survey uses three layers, failed BGS targets may be re-exposed, with fiber-assignment simulations intended to optimize faint-target redshift and radial-velocity completeness.The simulations will combine MWS and BGS samples while accounting for galaxy clustering and variation in stellar density.Source: §4.5.4 · p. 85