---
title: 'KODIAQ-SQUAD: High-Res Quasar Archives'
url: https://www.emergentmind.com/topics/kodiaq-squad
type: topic
---

# KODIAQ-SQUAD: High-Res Quasar Archives

KODIAQ-SQUAD denotes the joint scientific use of two major high-resolution quasar-absorption archives—the Keck Observatory Database of Ionized Absorption toward Quasars (KODIAQ) and the UVES Spectral Quasar Absorption Database (SQUAD)—most prominently in one-dimensional Ly$\alpha$ forest power-spectrum analyses. In the canonical implementation, KODIAQ and SQUAD are combined at the spectrum level in pixel space, while the lower-resolution XQ-100 sample is estimated separately and merged later at the bandpower level, yielding a measurement of the 1D Ly$\alpha$ forest power spectrum over $k<0.1\,\mathrm{s\,km}^{-1}$ and $z=2.0$–4.6 from 538 quasars [2108.10870]. The term is not a formal archive name in the KODIAQ data-release papers: KODIAQ DR2 presents a Keck/HIRES release and does not mention SQUAD, and some later repositories use labels such as “KODIAQ-SQUAD” as a practical grouping rather than as a distinct survey product [1707.07905; 2010.09061].

## 1. Nomenclature and archival provenance

KODIAQ and SQUAD originated as separate public data releases. KODIAQ is the Keck Observatory Database of Ionized Absorption toward Quasars, initially released as a HIRES-based survey aimed at galactic and circumgalactic absorption studies at high redshift, with emphasis on highly ionized gas such as O VI [1505.03529]. SQUAD is the UVES Spectral Quasar Absorption Database, a reproducible archive of fully reduced, continuum-fitted VLT/UVES quasar spectra intended to support damped Ly$\alpha$ studies, absorption-line surveys, and time-variable absorption analyses [1810.06136].

The expression “KODIAQ-SQUAD” is therefore contextual. In the 2021 optimal-quadratic-estimator study, it effectively denotes the combined high-resolution Keck/HIRES and VLT/UVES dataset used to estimate the Ly$\alpha$ forest $P_{\mathrm{1D}}$ [2108.10870]. In later cosmological inference work, the same combination is abbreviated as “KS” and analyzed as a distinct power-spectrum dataset with its own selection effects and nuisance structure [2509.18271]. By contrast, the KODIAQ DR2 release itself describes only KODIAQ and states that the paper does not mention SQUAD; it recommends igmspec as the cross-database integration point when KODIAQ is used alongside SQUAD-like resources [1707.07905]. The same practical distinction appears in KODIAQ DR3, where some portals may group KODIAQ with other quasar-absorption compilations under labels such as “KODIAQ-SQUAD,” but the underlying records retain instrument provenance [2010.09061].

## 2. Constituent surveys and data characteristics

The high-resolution combination is enabled by the close spectroscopic characteristics of KODIAQ DR2 and SQUAD DR1, both of which provide continuum-fitted echelle spectra with comparable resolving power and pixel sampling, while XQ-100 contributes additional redshift path length at lower resolution [1707.07905; 1810.06136; 2108.10870].

| Component | Core characteristics | Role in Ly$\alpha$ $P_{\mathrm{1D}}$ work |
|---|---|---|
| KODIAQ DR2 | Keck/HIRES; 300 quasars at $0.07<z_{\mathrm{em}}<5.29$; 831 continuum-normalized co-added spectra from 1577 exposures; $\sim 4.9$ megaseconds; $36{,}000 \le R \le 103{,}000$ | Combined with SQUAD at the spectrum level on a common $3\,\mathrm{km\,s}^{-1}$ grid |
| SQUAD DR1 | VLT/UVES; 467 fully reduced, continuum-fitted high-resolution quasar spectra; $z=0$–5; total exposure time $10.09\times10^6$ seconds; typically $R \approx 40{,}000$–60,000 | Combined with KODIAQ at the spectrum level; seeing-based resolution correction applied |
| XQ-100 | VLT/X-Shooter; 100 quasars at $3.5<z<4.5$; $R \approx 4{,}000$–7,000 | Estimated separately and combined with KS at the bandpower level |

KODIAQ DR2 extends DR1 by adding 130 new quasars and additional observations of some DR1 targets, bringing the full HIRES sample to 300 QSOs spanning $0.07<z_{\mathrm{em}}<5.29$ [1707.07905]. The observations cover resolving powers $36{,}000 \le R \le 103{,}000$, corresponding to $\Delta v \approx 8.33\,\mathrm{km\,s}^{-1}$ at $R=36{,}000$ and $\Delta v \approx 2.91\,\mathrm{km\,s}^{-1}$ at $R=103{,}000$ via $\Delta v = c/R$. HIRES deckers include C5 and D1 ($R \approx 36{,}000$), C1/C2/B5 ($R \approx 48{,}000$), B2 ($R \approx 72{,}000$), and E3 ($R \approx 103{,}000$), with the majority of spectra obtained using C1 or C5. Raw exposures were uniformly reduced with HIRedux, the TK2048 detector “ink spot” was masked during extraction, and more than 15,000 echelle orders were continuum-fit order by order.

SQUAD DR1 is built around transparency and full reproducibility. Final spectra are redispersed onto a log-linear, vacuum–heliocentric wavelength grid, typically with $2.5\,\mathrm{km\,s}^{-1}$ per pixel for 2×2 or 2×1-binned data and about $1.3\,\mathrm{km\,s}^{-1}$ per pixel for unbinned data [1810.06136]. UVES\_popler records automatic parameters and manual actions in a human-readable log, and DR1 distributes final spectra, reduction scripts, extracted orders, calibration products, and metadata sufficient to reproduce or modify the combination. Continuum-to-noise ratios at $5500\,\unicode{x212B}$ span 4–342 per $2.5\,\mathrm{km\,s}^{-1}$ pixel, with a median of 20.

These similarities explain why KODIAQ and SQUAD can be combined at the spectrum level. In the power-spectrum analysis, KODIAQ DR2 has original pixel spacings of 1.3 or $2.6\,\mathrm{km\,s}^{-1}$ and SQUAD is sampled at 1.3–$3.0\,\mathrm{km\,s}^{-1}$; both are resampled to a common $3\,\mathrm{km\,s}^{-1}$ velocity grid to reduce computational cost without affecting the $k$-range of interest [2108.10870].

## 3. Construction of the high-resolution 1D Ly$\alpha$ power spectrum

The principal KODIAQ-SQUAD measurement uses the optimal quadratic estimator (OQE) to extract the 1D Ly$\alpha$ forest power spectrum from high-resolution, high-S/N spectra [2108.10870]. KODIAQ and SQUAD are combined at the spectrum level because they have similar resolution and cadence, whereas XQ-100 is estimated separately because its lower resolution requires larger and more uncertain resolution corrections and a more restrictive upper-$k$ limit. The final sample contains 538 unique quasars, partitioned as 186 KODIAQ, 278 SQUAD, and 74 XQ-100 objects. The Ly$\alpha$ forest rest-frame window is 1050–$1180\,\unicode{x212B}$, the analysis region is $1.7<z<4.7$, and the final conservative range excludes $z=1.8$ because of KODIAQ–SQUAD differences in that lowest bin.

The basic flux variables are
$$
F(\lambda)=\frac{f(\lambda)}{C(\lambda)}, \qquad \delta_F(v)=\frac{F(v)}{\langle F\rangle}-1,
$$
with velocity defined by $v=c\ln(\lambda/\lambda_{\mathrm{Ly}\alpha})$, $z=e^{v/c}-1$, and $\lambda_{\mathrm{Ly}\alpha}=1216\,\unicode{x212B}$. The 1D power spectrum is
$$
P_{\mathrm{1D}}(k,z)=\langle |\tilde{\delta}_F(k,z)|^2 \rangle.
$$
With data vector $x\equiv \delta_F$ and covariance $C\equiv S+N$, the OQE estimates bandpowers through
$$
\hat{p}_\alpha=\frac{1}{2}\,x^T C^{-1} C_{,\alpha} C^{-1}x-b_\alpha,
$$
with Fisher matrix
$$
F_{\alpha\beta}=\frac{1}{2}\,\mathrm{Tr}\!\left[C^{-1}C_{,\alpha}C^{-1}C_{,\beta}\right].
$$

The measurement is binned into 15 redshift bins from $z=1.8$ to 4.6 with $\Delta z=0.2$, and 21 $k$ bins: four linear bins at $\Delta k_{\mathrm{lin}}=0.0022\,\mathrm{s\,km}^{-1}$ starting at $k\approx 0.004\,\mathrm{s\,km}^{-1}$, followed by logarithmic bins up to $0.1\,\mathrm{s\,km}^{-1}$ [2108.10870]. High resolution and high S/N enable robust measurement to $k\approx 0.1\,\mathrm{s\,km}^{-1}$, with particular emphasis on $k\gtrsim 0.02\,\mathrm{s\,km}^{-1}$, where thermal broadening and pressure smoothing become visible.

## 4. Instrument response, preprocessing, and covariance control

The OQE framework models instrumental effects explicitly in the signal covariance [2108.10870]. The measured power is suppressed by the line-spread function and the pixel window. In $k$-space, the response factors are
$$
W_{\mathrm{res}}(k)=|\tilde{L}(k)|^2,\qquad W_{\mathrm{pix}}(k)=\mathrm{sinc}^2(\pi k \Delta v),
$$
and for a Gaussian LSF with $1\sigma$ velocity width $R$ and a top-hat pixel the analysis uses
$$
W(k)=\exp\!\left(-\frac{k^2R^2}{2}\right)\,\mathrm{sinc}\!\left(\frac{k\,\Delta v}{2}\right),
$$
squared in the signal covariance. KODIAQ/HIRES is treated with $R\gtrsim 36{,}000$, corresponding to $c/R\approx 8.3\,\mathrm{km\,s}^{-1}$ FWHM and $\sigma_v\approx 3.5\,\mathrm{km\,s}^{-1}$ under a Gaussian LSF assumption; SQUAD/UVES is treated with $R\gtrsim 40{,}000$, corresponding to $c/R\approx 7.5\,\mathrm{km\,s}^{-1}$ FWHM and $\sigma_v\approx 3.2\,\mathrm{km\,s}^{-1}$.

SQUAD requires two additional corrections. First, the nominal resolution is underestimated when median seeing $\theta$ is smaller than slit width $s$, so the analysis applies
$$
R_{\mathrm{cor}} = R \times (s/\theta)
$$
for $s>\theta$, with a median 25% correction and a maximum of 150%; the net effect on $P_{\mathrm{1D}}$ remains below 3% even at $k=0.1\,\mathrm{s\,km}^{-1}$ [2108.10870]. Second, SQUAD pipeline variances are underestimated in saturated lines, so the per-pixel $\chi^2_\nu$ about the weighted mean is median-filtered over five pixels and errors are scaled by $\sqrt{\mathrm{median}(\chi^2_\nu)}$ where this exceeds unity.

Continuum and contamination control are equally central. Outlier pixels in the Ly$\alpha$ region are cleaned with robust MAD cuts, retaining only pixels satisfying
$$
-\mathrm{MAD}(F)<F(\lambda)<1+\mathrm{MAD}(F)
$$
and
$$
0<\sigma(\lambda)<\mathrm{median}(\sigma)+3.5\times \mathrm{MAD}(\sigma).
$$
To mitigate large-scale continuum errors, each forest is split into three rest-frame chunks and additive continuum modes per chunk—a constant and a slope in $\ln\lambda$—are marginalized in the OQE. Metal lines are not masked individually; instead, statistical metal power is subtracted using sideband regions redward of Ly$\alpha$, specifically SB1: 1268–$1380\,\unicode{x212B}$ and SB2: 1409–$1524\,\unicode{x212B}$ in the quasar rest frame. C IV doublet-induced oscillations are prominent at $k\approx 1.3\times10^{-2}$ and $2.5\times10^{-2}\,\mathrm{s\,km}^{-1}$. Damped systems are masked with a width
$$
W = 7.3\,(1+z_{\mathrm{abs}})\,\sqrt{\frac{N_{\mathrm{HI}}}{10^{20}\,\mathrm{cm}^{-2}}}\,\unicode{x212B},
$$
while incomplete sub-DLA and LLS removal is carried as a systematic term.

Validation uses 100 log-normal mock datasets matched to real resolution, spacing, and noise, which show unbiased recovery of $P_{\mathrm{1D}}$ for $k<0.1\,\mathrm{s\,km}^{-1}$ [2108.10870]. Because Gaussianity assumptions underestimate small-scale errors, the covariance is estimated from 25,000 bootstrap realizations per configuration and regularized through a two-step procedure: off-diagonal element estimation with positive-definiteness enforcement, followed by eigenvalue flooring to Gaussian limits. In the conservative range $0.004<k<0.1\,\mathrm{s\,km}^{-1}$ and $z>1.8$, the total systematic budget averages about 19% of the statistical error.

## 5. Scientific reach and inference targets

KODIAQ-SQUAD provides the largest number of high-resolution, high-S/N Ly$\alpha$ forest observations used in a single 1D power-spectrum analysis, and its main scientific value lies in the small-scale regime [2108.10870]. These modes, especially $k\gtrsim 0.02\,\mathrm{s\,km}^{-1}$, are not available in SDSS/eBOSS or DESI analyses and are sensitive to the thermal state and reionization history of the intergalactic medium as well as to small-scale suppression in the matter power spectrum. The final KS+XQ-100 bandpowers are therefore particularly relevant for forward modeling with hydrodynamical simulations of thermal broadening, pressure smoothing, and warm-dark-matter-like cutoffs.

KODIAQ itself already expanded high-resolution coverage of the Ly$\alpha$ forest, C IV, and the Lyman limit, with well over 200 quasars covering rest-frame Ly$\alpha$ and C IV and approximately 100 covering the quasar Lyman limit [1707.07905]. The resulting spectra support Ly$\alpha$ forest thermodynamics, small-scale coherence, opacity studies, LLS/DLA surveys, and metal-line analyses. In the combined power-spectrum setting, these archival strengths are transformed into a precision statistical measurement over $z=2.0$–4.6 and $k<0.1\,\mathrm{s\,km}^{-1}$ [2108.10870].

The improvement in small-scale sensitivity is quantified explicitly. Using a single-parameter cutoff forecast, the analysis finds $\sigma_{\mathrm{cut}}=6.4\,\mathrm{s\,km}^{-1}$ for statistical errors alone and $\sigma_{\mathrm{cut}}=7.8\,\mathrm{s\,km}^{-1}$ with statistical and systematic errors, compared with $19.0\,\mathrm{s\,km}^{-1}$ for Walther et al. (2017), $20.1\,\mathrm{s\,km}^{-1}$ for Chabanier et al. (2019), and $32.9\,\mathrm{s\,km}^{-1}$ for Iršič et al. (2017) [2108.10870]. This corresponds to an improvement in sensitivity to a small-scale cutoff by more than a factor of 2. The same work releases $P_{\mathrm{1D}}(k,z)$ bandpowers, sideband power spectra, bootstrap-derived covariance and Fisher matrices, and code for the OQE pipeline and preprocessing.

## 6. Selection effects, later cosmological reinterpretation, and practical usage

Later inference with the PRIYA emulator sharpened the interpretation of KODIAQ-SQUAD by showing that the KS power spectrum is not only thermally informative but also highly sensitive to archival target-selection bias [2509.18271]. In that study, KS combines 767 high-resolution Keck/HIRES and VLT/UVES sightlines covering $2\le z\le 5$, but the authors restrict the likelihood analysis to $z=2.6$–4.2 and $k=0.0055$–$0.065\,\mathrm{s\,km}^{-1}$. The key concern is that KS is assembled from targeted archival observations from more than 300 PIs, including programs that deliberately selected known DLAs and O IV absorbers. Even after masking identified DLAs, residual sub-DLAs and LLSs remain more abundant than in a blind selection.

This bias propagates directly into cosmological fits. For KS alone over $z=2.6$–4.2 and $k=0.0055$–$0.065\,\mathrm{s\,km}^{-1}$, the posterior runs to high primordial-amplitude and tilt values, yielding $A_P>2.26\times10^{-9}$ and $n_P>1.01$, while the HCD template fit favors $\alpha_{\mathrm{LLS}}=1.60^{+0.41}$ [2509.18271]. When eBOSS priors are imposed on $(A_P,n_P)$, the reduced $\chi^2$ remains essentially unchanged, $\chi^2/\mathrm{dof}\approx 1.58 \rightarrow 1.57$, but $\alpha_{\mathrm{LLS}}$ rises to $2.79\pm0.41$ over $z=2.6$–4.2 and to $5.0^{+1.9}_{-1.4}$ in $z=3.8$–4.2. The same analysis therefore concludes that the apparent cosmological preference of KS is driven by selection bias toward high-column density absorbers rather than by primordial power alone. It also states that the $P_{\mathrm{1D}}$ at $k>0.045\,\mathrm{s\,km}^{-1}$ is more sensitive to Lyman limit system contamination and thermal history.

The recommended mitigation is correspondingly specific. Restricting KS to $z=3.4$–4.2 and $k<0.045\,\mathrm{s\,km}^{-1}$ brings the cosmology back into agreement with eBOSS, giving $A_P\approx 1.85^{+0.24}_{-0.63}\times10^{-9}$ and $\alpha_{\mathrm{LLS}}<1.49$ [2509.18271]. This suggests a useful scale separation: cosmological parameters $(A_P,n_P)$ are primarily constrained by larger scales, whereas thermal history and HCD nuisance parameters dominate the smallest scales.

For practical data work, the archival distinction remains important. KODIAQ DR2 is publicly available at the KOA and in igmspec v03, and its own documentation recommends igmspec as the cross-database integration point for scripted workflows that combine KODIAQ with SQUAD-like resources [1707.07905]. KODIAQ DR3 likewise notes that some portals may present joint collections under labels such as “KODIAQ-SQUAD,” but the underlying records retain instrument provenance, so users should filter by dataset and instrument and harmonize wavelength frames and resolutions before stacking or statistical combination [2010.09061].

Source: https://www.emergentmind.com/topics/kodiaq-squad