DESI: Dark Energy Spectroscopic Instrument Survey

• The survey is a Stage-IV, massively multiplexed spectroscopic campaign that measures cosmic expansion via BAO and RSD across 14,000 deg² with approximately 50 million targets.
• Advanced instrumentation and a robust targeting pipeline enable precise fiber positioning, spectral extraction, and redshift determination using cutting-edge metrology and algorithms.
• DESI forecasts sub-percent BAO and percent-level RSD growth measurements, delivering transformative constraints on dark energy, neutrino mass, and primordial density fluctuations.

The Dark Energy Spectroscopic Instrument Survey (DESI) is a Stage-IV, massively multiplexed spectroscopic campaign conducted on the 4-meter Mayall Telescope at Kitt Peak. The survey is designed to map the large-scale structure of the Universe via precise measurements of baryon acoustic oscillations (BAO) and redshift-space distortions (RSD) across 14,000 deg² and a redshift range 0 < z < 3.5, targeting ∼50 million extragalactic and Galactic sources over five years. DESI aims to provide percent-level constraints on dark energy, cosmic expansion history, neutrino mass, and primordial density fluctuations, leveraging a robust instrumentation platform and a suite of ultra-efficient pipelines for survey operations, targeting, astrometry, spectral extraction, redshift estimation, and exposure control (Collaboration et al., 2016, Collaboration et al., 2016, Schlafly et al., 2023, Collaboration et al., 2023, Myers et al., 2022, Guy et al., 2022, Anand et al., 29 May 2024, Tie et al., 2021, Schlafly et al., 10 Jul 2024, Kent et al., 2023).

1. Survey Design and Targeting Principles

The DESI Survey’s core design divides observing time into “dark” and “bright” programs dependent on moon phase and seeing. Primary dark-time targets are emission-line galaxies (ELGs), luminous red galaxies (LRGs), and quasi-stellar objects (QSOs), optimized for BAO and RSD science. Bright time is dedicated to the Bright Galaxy Survey (BGS) and Milky Way Survey (MWS). Secondary programs (encoded in SCND_TARGET bitmasks) utilize fibers for bespoke targets such as supernova hosts, clusters, and high-z QSOs (Myers et al., 2022).

Survey targeting utilizes the “desitarget” pipeline, which processes all imaging-based targets and assigns unique TARGETIDs via a hierarchical 64-bit signed integer scheme: $$\mathrm{TARGETID} = \mathrm{OBJID} + \mathrm{BRICKID}\cdot2^{28} + \mathrm{RELEASE}\cdot2^{46} + \mathrm{MOCK}\cdot2^{58}+\mathrm{SKY}\cdot2^{59}+\mathrm{GAIADR}\cdot2^{60}$$ Positive TARGETIDs cover all real and simulated objects, while negative TARGETIDs are used for on-the-fly sky assignment to broken fibers (Myers et al., 2022).

Multiple bitmasks within targeting files encode selection flags:

• DESI_TARGET, BGS_TARGET, MWS_TARGET, SCND_TARGET: each 64-bit mask flags target type/class.
• OBSCONDITIONS: encodes DARK/BRIGHT/BACKUP status per target (Myers et al., 2022).

Survey targeting is phase-dependent, with separate masking for CMX (commissioning), SV1/SV2/SV3 (survey validation), and main survey (Myers et al., 2022, Collaboration et al., 2023, Collaboration et al., 2023).

2. Instrumentation: Focal Plane, Spectrograph, and Metrology

DESI is built around a six-element fused-silica corrector, providing a 3.2° diameter focal plane (0.8 m diameter), yielding a plate scale of ~14–18″/mm. The focal plane hosts 5,000 robotic “theta–phi” positioners, each capable of placing a 107 μm fiber with RMS lateral accuracy ~8–11 μm (2-D), verified through laboratory metrology and on-sky dither astrometry (Collaboration et al., 2016, Kent et al., 2023, Fagrelius et al., 2017).

Fiber positioning leverages a closed-loop system with back-illumination and the Fiber View Camera (FVC), which images fiber tips and fiducials through the corrector. Astrometric transformations from sky to focal plane are characterized with distortion models using spin-weighted Zernike polynomials and adaptive compensation for thermal expansion, flexure, and atmospheric dispersion (Kent et al., 2023, Schlafly et al., 10 Jul 2024).

Each fiber feeds one of ten identical three-arm spectrographs, with dichroic splits at 553 nm and 752 nm, covering 360–980 nm. The spectral resolution varies:

• Blue: R ≈ 2,000 at 360 nm, rising to ~3,000 at 5800 Å
• Red: R ≈ 3,000–4,000
• NIR: R ≈ 4,000–5,500 to 980 nm This throughput enables reliable [O II] doublet and LRG continuum measurements for targeted redshifts (Collaboration et al., 2016, Collaboration et al., 2023).

3. Survey Operations, Tiling, and Exposure Control

DESI employs a “depth-first,” overlapping-tile strategy: 9,929 “dark” tiles (7 passes) and 5,676 “bright” tiles (4 passes) over the 14,246 deg² footprint, yielding mean coverage of ~5.2 passes per dark tile and ~3.2 for bright tiles (Schlafly et al., 2023).

Nightly observing is orchestrated via automated field selection, based on real-time survey speed S = (S/N)^2, airmass, tile priority, moon avoidance, and weather-driven masks. Exposure time is dynamically set by the Exposure Time Calculator (ETC), absorbing inputs from guiders (PSF, transparency) and the Sky Continuum Monitor, which measures sky background in real time using discrete “sky fibers.” The ETC ensures a pre-defined SNR threshold for all field types, adjusting for airmass: $t_{\rm exp}(X) = t_0 X^{1.25}$ Uniform effective exposure time (EFFTIME_ETC) is maintained across conditions, maximizing survey speed and scientific homogeneity. Observing overheads are ≲2 min per tile, driven by fiber repositioning, plate acquisition, and readout (Schlafly et al., 2023, Tie et al., 2021, Collaboration et al., 2023).

Performance metrics from early survey operations indicate the survey is running ∼7% faster than forecast, with observed mean survey speed S ≈ 1.15× nominal (Schlafly et al., 2023).

4. Data Reduction, Spectral Extraction, and Redshift Determination

Raw CCD data (30 amplifiers, ten spectrographs × three arms) are processed through a hierarchical pipeline:

• CCD calibration: bias, overscan, gain, cosmic-ray masking
• PSF mapping, wavelength calibration (Legendre polynomials per pixel row)
• Spectroperfectionism: forward-modeling fiber-trace extraction, yielding a decorrelated resolution matrix R and uncorrelated fluxes
• Flat-fielding: per-fiber throughput correction using dome flats
• Sky modeling: least-squares fit of a deconvolved sky spectrum on dedicated sky fibers, reconvolved to each science fiber
• Spectrophotometric calibration: F-star standards, color matching, and aperture corrections
• Redrock: redshift and classification via χ²-matching of observed spectra to PCA or archetype-based templates, supplemented for QSOs by QuasarNET and Mg II line search (Guy et al., 2022, Anand et al., 29 May 2024).

Recently, an archetype-based redshift fitter refines Redrock performance for galaxy spectra, reducing catastrophic failures by 10–40%, improving purity by 0.1–0.3% for LRGs/ELGs, and decreasing sky-fiber false positives by 5–40%, via bounded-variable least squares over physical templates and per-camera Legendre polynomials. Computational cost increases by a factor 2–3, yet remains tractable at scale (Anand et al., 29 May 2024).

All coaddition and model-fitting steps maintain strictly uncorrelated noise properties by avoiding wavelength resampling and rigorously tracking the resolution matrix R per spectrum (Guy et al., 2022).

5. Validation Campaigns, Data Model, and Catalog Products

Survey Validation (SV) campaigns—SV1, SV2, and SV3 (One-Percent Survey)—established reproducible target selection, exposure strategies, completeness mapping, and empirical calibration for main operations, covering 140 deg² in the “One-Percent” pilot. SV data informed the final selection cuts and operational workflows, ensuring dark-energy science requirements for sample sizes, completeness, and n(z) are met (Collaboration et al., 2023, Collaboration et al., 2023).

Data products are organized into FITS and HDF5 formats and include:

• Raw and calibrated exposures
• Single-tile and HEALPix coadded spectra
• Redrock redshift catalogs with ZWARN, DELTACHI2, and ZCAT_PRIMARY flags
• Value-added catalogs (VACs) including visual-inspection redshifts, photometry cross-matches, and large-scale structure science weights (“completeness” w_comp; FKP weights) (Collaboration et al., 2023).

Quality cuts typically impose ZWARN = 0 (no pipeline failures) and DELTACHI2 thresholds (e.g. >15 for LRG, >40 for BGS, [O II]-Δχ² composite for ELG) (Collaboration et al., 2023).

6. Astrometry, Fiber Metrology, and Positioning Systematics

Astrometric calibration is achieved via integrated software modules: PlateMaker (sky-to-focal plane conversion), spotmatch (FVC centroid extraction), turbulence-correction for dome seeing, and dither analysis for throughput maximization. Transformations from (α, δ) to focal-plane (x, y) employ gnomonic projections and non-linear Zernike-based distortion models with 13 coefficients, including ADC-induced B-modes.

Positioning errors due to atmospheric turbulence are mitigated via Gaussian-process-based correlation modeling using stationary fiducials and disabled positioners as turbulence probes. This correction reduced RMS errors from 7.3 μm (raw) to 3.5 μm (corrected), yielding ~1.6% survey speed improvement by decreasing miscentering loss under typical conditions (Schlafly et al., 10 Jul 2024, Kent et al., 2023).

Final repeatable positioning accuracy is ~8 μm (1-D) and 11 μm (2-D), with negligible impact on flux calibration uniformity and spectroscopic redshift performance (Kent et al., 2023).

7. Cosmological Deliverables, Forecasts, and Legacy

DESI’s forecasted performance includes sub-percent precision on the BAO ruler and percent-level RSD growth-rate measurements in multi-redshift bins:

• σ_R/R < 0.28% for z < 1.1
• σ_R/R < 0.39% for 1.1 < z < 1.9
• σ_H/H < 1.05% for 1.9 < z < 3.7

Aggregate Fisher-matrix constraints anticipate:

• σ(w₀) ≈ 0.04, σ(w_a) ≈ 0.3
• σ(fσ₈(z)) ~1–2% per bin
• σ(∑m_ν) ≈ 0.02 eV
• DETF Figure-of-Merit (FoM) ≈ 150–250 (BAO+RSD) Compared to prior surveys, DESI delivers ~6× gain in FoM, with sample sizes of 13.8 M BGS, 7.46 M LRG, 15.7 M ELG, 2.87 M QSO, and 7.2 M MWS objects (Collaboration et al., 2023, Collaboration et al., 2016, Levi et al., 2013).

The survey’s full data will deliver a 3-D map of ~35 M galaxies, ~3 M quasars, and ~7 M Milky Way stars, supporting transformative studies of cosmic acceleration, structure growth, neutrino masses, and cosmological initial conditions. Public and collaboration data releases follow robust QA procedures, with all primary science catalogs, resolution matrices, and quality flags documented and supported in the Astropy ecosystem (Collaboration et al., 2023, Guy et al., 2022).

• "The Target-selection Pipeline for the Dark Energy Spectroscopic Instrument" (Myers et al., 2022)
• "ProtoDESI: First On-Sky Technology Demonstration for the Dark Energy Spectroscopic Instrument" (Fagrelius et al., 2017)
• "The DESI Experiment, a whitepaper for Snowmass 2013" (Levi et al., 2013)
• "Unraveling the Universe with DESI" (Vargas-Magana et al., 2019)
• "The DESI Experiment Part I: Science, Targeting, and Survey Design" (Collaboration et al., 2016)
• "The Spectroscopic Data Processing Pipeline for the Dark Energy Spectroscopic Instrument" (Guy et al., 2022)
• "The Early Data Release of the Dark Energy Spectroscopic Instrument" (Collaboration et al., 2023)
• "Validation of the Scientific Program for the Dark Energy Spectroscopic Instrument" (Collaboration et al., 2023)
• "The DESI Sky Continuum Monitor System" (Tie et al., 2021)
• "The DESI Experiment Part II: Instrument Design" (Collaboration et al., 2016)
• "Correcting Turbulence-induced Errors in Fiber Positioning for the Dark Energy Spectroscopic Instrument" (Schlafly et al., 10 Jul 2024)
• "Astrometric Calibration and Performance of the Dark Energy Spectroscopic Instrument Focal Plane" (Kent et al., 2023)
• "Archetype-Based Redshift Estimation for the Dark Energy Spectroscopic Instrument Survey" (Anand et al., 29 May 2024)
• "Survey Operations for the Dark Energy Spectroscopic Instrument" (Schlafly et al., 2023)