The Large Early Galaxy Astrophysics Census (LEGA-C) Data Release 3: 3000 High-quality Spectra of K_s-selected Galaxies at z > 0.6

The successful tenure of ESO's Very Large Telescope/Visible Multi-Object Spectrograph (VIMOS; Le Fèvre et al. 2003) was completed by the implementation of the two most recent Public Spectroscopic Surveys: Large Early Galaxy Astrophysics Census (LEGA)-C (van der Wel et al. 2016; Straatman et al. 2018) and VANDELS (McLure et al. 2018; Pentericci et al. 2018). Both surveys, carried out from 2015 to 2018, are characterized by a focus on depth and data quality, representing a clear departure from VIMOS' previous use as the "redshift machine" for which it was designed. The detector upgrade and instrument refurbishment in 2010 made this possible.

The practical goal of the LEGA-C survey is to collect high signal-to-noise ratio (S/N), high-resolution spectra for thousands of galaxies in the redshift range 0.6 ≲ z ≲ 1. This allows us, for the first time, for the general galaxy population, to probe stellar populations (ages and metallicities) and stellar kinematics (velocity dispersions) of thousands of galaxies at a look-back time of ∼7 Gyr.

The primary science goals span all key open questions in the field of galaxy formation and evolution. First, what are the star formation histories of galaxies? The spectra constrain the bulk formation age of the stellar population, the metal enrichment history, and how bursty the star formation history is. Crucially, at large look-back times, it is easier to resolve the main formation phase; in the present-day universe, most stars formed many gigayears ago, and old stellar populations are difficult to dissect due to their slow spectral evolution. Second, how does star formation quench in massive galaxies? Does this occur rapidly due to strong outflows, possibly associated with a preceding increase in star formation, or is it relatively slowly due to a lack of cooling to supplement fuel? The role of environment, active galactic nuclei (AGNs), and mergers can be examined thanks to the large sample size. Third, how do galaxies evolve after the cessation of star formation? They can evolve merely passively, or continue growth and structural evolution through merging. The kinematic properties and chemical composition constrain this process.

These topics are interconnected and also connect to quantifying the fundamental properties of the galaxy population as a whole. Given the dynamical mass measurements and spectroscopy-based estimates with unprecedented quality of the stellar mass, we will be able to assess the contribution of dark matter (and/or a bottom-heavy stellar initial mass function) across galaxy types, and how this changes with cosmic time. The abundances of heavy elements can be constrained and compared with those of the (ionized) interstellar medium to indirectly constrain the importance of in- and outflows of galaxies.

Initial results based on the first and second data releases (Straatman et al. 2018, hereafter, DR2) illustrate the broad range of applications of LEGA-C data. Bezanson et al. (2018a) showed that z ∼ 1 quiescent galaxies show more rotation in their stellar body than equally massive counterparts at z ∼ 0, providing strong evidence for a reduction in net angular momentum through merging after galaxies become quiescent, with modest environmental dependence (Cole et al. 2020). At the same time, the question of whether galaxies change their structure upon "quenching"—the transition from an actively star-forming state to a quiescent state—remains open. Wu et al. (2020) and D'Eugenio et al. (2020) showed that post-starburst galaxies rapidly form a large number of stars in the central regions of galaxies, producing compact galaxies. But, at the same time, larger galaxies are generally younger (Wu et al. 2018a), reflecting a complexity in galaxy evolution that can only be revealed by high-quality spectroscopic data. The specific implication here is that there are multiple evolutionary pathways to reach a quiescent state. This is just one example of how constraints on stellar populations (Wu et al. 2018b) increase our physical insight into the evolutionary process, and our first attempts to quantify this information demonstrated that the "downsizing" trend—that galaxies with higher stellar masses are older—was already evident at z ∼ 1 (Chauke et al. 2018). The quality of the LEGA-C spectroscopy is also such that, for the first time, we could detect and quantify bursts of star formation at z ∼ 1 that occurred after an initial period of quiescence at z ∼ 2 (Chauke et al. 2019), as well as a novel approach to quantifying dust attenuation in star-forming galaxies. Barišić et al. (2020) showed steep attenuation curves, especially for face-on galaxies, and frequently detected the 2175 Å bump in individual galaxies for the first time.

The ultimate goal to jointly analyze the evolution of stellar populations and the kinematic state of the stellar body is now within reach, which starts with examining the evolution and detailed properties of scaling relations such as the Fundamental Plane (de Graaff et al. 2020, 2021). Detailed comparisons with the latest cosmological simulations of galaxy formation will then improve the fidelity of the physical models used in those simulations. Finally, the large sample size also allows us to probe rare classes of objects such as radio-loud AGNs and explore their kinematic and stellar population properties for the first time (Barišić et al. 2017).

With the full LEGA-C sample, where we increase the sample of spectra with measured redshifts and stellar velocity dispersions from 1447 to 3528 compared with DR2, we enable the community to pursue this line of research. In this paper we will first present the updated sample and the improvements in the data reduction pipeline compared to DR2, as well as an analysis of the Hubble Space Telescope/Advanced Camera for Surveys (HST/ACS) data that are available for nearly all objects in our sample (Section 2). We then describe the basic measurements that we are publishing with this paper: stellar and ionized gas velocity dispersions, spectral indices, and emission-line fluxes, as well as inferred quantities such as dynamical mass estimates (Section 3). Our published measurements are independent of cosmological parameters, with the exception of dynamical mass, which has one element (galaxy size) that requires a conversion from observed to physical units. There we adopt a flat ΛCDM cosmology with Ω_m = 0.3 and H₀ = 70 km s⁻¹.

2.1. The LEGA-C Survey: Observations

The survey goals and design were extensively discussed by van der Wel et al. (2016) and Straatman et al. (2018). The key characteristic of the survey is that it is K_s -band selected. As we will see, the long integration times of ∼20 hr lead to a very high success rate in measurement redshifts, stellar kinematics, and stellar population properties, even for the faintest, reddest targets. From the UltraVISTA catalog (Muzzin et al. 2013b), we first create a parent sample of galaxies with (spectroscopic or photometric) redshifts in the range 0.6 < z < 1 and with K_s magnitudes brighter than a redshift-dependent limit ${K}_{s,\mathrm{LIM}}\,=20.7\mbox{--}7.5\,\mathrm{log}((1+z)/1.8)$ (Figure 1). When designing the slit masks, we included targets ordered by K_s magnitude (brightest first). As the mask fills up, slit collisions prevent the inclusion of fainter galaxies, leading to higher (and precisely quantifiable) completeness for brighter targets. Any remaining mask space is used to include fillers: K_s -bright objects at z > 1, fainter targets in the LEGA-C redshift range, and mixed bag objects at lower redshifts and faint, higher-redshift objects. We refer to Table 1 and, for comparison, to the equivalent Table 1 in the survey paper (van der Wel et al. 2016) for an overview of the LEGA-C sample. The success rate of the redshift and stellar velocity dispersions (discussed in Section 3) is also given.

Zoom In Zoom Out Reset image size

Table 1. Sample Selection and Success Rate

Sample	Redshift	K-band limit	# in UltraVISTA	# in LEGA-C	with z	with ${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$
Parent	Any	<23.4 (90% compl.)	157,667	4081	3937	3528
Primary	0.6 < z < 1.0	<20.7–7.5 log((1 + z)/1.8)	8655	3029	3018	2932
Filler I	z > 1	<20.36	3511	260	184	163
Filler II	0.6 < z < 1.0	>20.7–7.5 log((1 + z)/1.8)	32,261	590	553	380
Filler III	Other	Other	113,240	202	182	53

Note. Sample selection criteria for the primary sample and the three categories of fillers. The columns with # refer to the numbers of objects within the LEGA-C footprint in the Ultra-VISTA catalog from Muzzin et al. (2013a) and the LEGA-C survey targets, and the last two columns are the targets with successful redshift measurements and the targets with successful stellar velocity dispersion measurements.

Download table as:  ASCIITypeset image

The final survey footprint is created by 30 slightly overlapping pointings, together covering 1.4255 square degrees (Figure 2), giving a total survey comoving volume of 3.664 × 10⁶ Mpc³ between z = 0.6 and z = 1.0, equivalent to the local volume out to ∼100 Mpc (z ∼ 0.023).

Zoom In Zoom Out Reset image size

Two additional pointings, contained within the main footprint, were chosen for specific purposes. One of these has the same selection approach as the 30 main pointings, but the slits are oriented in the east–west direction, whereas all other masks and slits are oriented in the north–south direction. The goal here is to obtain kinematic information along two perpendicular axes. The other extra mask has a different selection function and is optimized to include red, low-mass galaxies that are otherwise rarely included in the survey. This increases the dynamic range in scaling relations.

The 32 masks produced 4081 galaxy spectra, with 3741 unique objects (340 are observed more than once). We have 3029 spectra of primary targets (compare their K magnitude distribution with the K magnitude distribution of the parent sample in Figure 1), and 1052 spectra of fillers. The success rate (in terms of making key measurements) is very high; the fraction of primary spectra with successful redshift and stellar velocity dispersion measurements is close to 100% as described in detail below (Section 3). The completeness of the LEGA-C sample in relation to both the parent sample and the full galaxy population is discussed in detail in Appendix A.

Observations were carried out between 2014 December and 2018 March. In total, 211 (mostly partial) nights were spent observing in Visitor Mode, for a total on-source integration time of 695 hr out of the allocated 1107 hr. The resulting efficiency of 63% includes not only the usual overheads, but also weather and technical losses. The anticipated efficiency in the survey management plan was also precisely 63% (including weather losses); from a scheduling and execution perspective, LEGA-C can then be considered a success. The motivation for executing the survey in Visitor Mode is obvious: even without weather and technical losses, Service Mode observations would have at most 50% efficiency given standard calibration requirements and scheduling constraints. The increased efficiency is mainly due to an efficient calibration plan and flexibility in designing the observations, circumventing the need for hourly re-acquisition and alignment. On a general note, given that VIMOS was not originally designed for long exposures and tracking objects for many hours, the performance of instrument and telescope has been very good.

2.2. Custom LEGA-C Pipeline

As described in the DR2 paper, we designed a custom pipeline built on top of ESO-provided data handling infrastructure. Here, for DR3, we make three improvements in the data analysis related to (1) sky subtraction/object extraction, (2) telluric absorption correction, and (3) flux calibration.

2.2.1. Sky Subtraction/Object Extraction

For DR2 we created sky+object models in the spatial direction where the object was represented by a Gaussian. We knew that a Gaussian imperfectly describes seeing-convolved galaxy light profiles, and we had already addressed this by refitting the object in the co-added, two-dimensional spectra with a Moffat profile. For most galaxies this worked fine, but for the many bright galaxies that illuminate more than half of the slit, this approach was insufficient. For DR3 we make use of the fact that most galaxies have been imaged by HST, for which we derived Sérsic light profiles (see Section 2.3). Based on the observed light profile in the LEGA-C spectra, collapsed in the wavelength direction, we construct a galaxy+seeing profile by convolving the Sérsic profile with a Moffat profile, and propagating the resulting profile through a rectangular slit with width 1''. The two Moffat parameters are optimized in this fit, producing a spectral extraction kernel that encompasses both the galaxy and the seeing. As before (see DR2), this kernel is used at each wavelength bin to fit for the sky and the object flux.

After coaddition of the sky-subtracted two-dimensional spectra, we check for systematic effects in the sky subtraction. For some galaxies located near the edge of their slit, or where multiple objects lie in the same slit, the sky subtraction was imperfect. In those cases, as for DR2, a new sky subtraction was performed by fitting a new Moffat profile, producing a new sky and a new object extraction.

2.2.2. Telluric Absorption

For DR2 we created a telluric absorption spectrum for each of the 32 masks based on the spectrum of a blue star included in those masks. This left two unaddressed problems. First, for some blue stars, the wavelength coverage was insufficient to probe the strong telluric features around 9000 Å. Second, a fixed telluric spectrum for each of the >100 objects in a mask leaves very significant high-frequency residuals (see Figure 3) due to two effects: (1) errors in the wavelength calibration at the level of ∼0.1–0.2 Å due to small, but unavoidable variations in object-slit alignments (if the blue star has a different alignment than a galaxy, then there will be a wavelength shift in the spectrum and, consequently, a difference in the telluric absorption spectrum); and (2) galaxies are not point sources, such that the line-spread function is broader for objects with larger angular sizes, leading to a smoother telluric absorption spectra for galaxies than for stars (this effect varies from galaxy to galaxy).

Zoom In Zoom Out Reset image size

To remedy these effects, we revised the telluric absorption correction for DR3 compared to DR1 and DR2:

• 1.  
  All telluric absorption features in the 6000–9300 Å wavelength range arise due to only two species: H₂O and O₂. We choose a well-calibrated white dwarf spectrum taken with the ESO's X-shooter spectrograph (WD0839-327, Programme ID 098.D-0392(A)) and create H₂O and O₂ telluric absorption template spectra using the wavelength ranges 6860 < λ/Å < 6925 and 7580 < λ/Å < 7710 (the A and B bands) for O₂ and the wavelength ranges 6925 < λ/Å < 7580 and λ/Å > 7710 for H₂O. Note that the combination covers the entire wavelength range λ > 6860 Å, that is, most of the LEGA-C spectra.
• 2.  
  We fit the H₂O and O₂ templates to the blue star spectrum, varying the strength of the H₂O and O₂ features. The key here is all features arising due to O₂ vary together (primarily as a function of airmass) as do the H₂O features (mostly as a function of humidity). This produces a master telluric absorption spectrum for each mask. This approach remedies incomplete wavelength coverage for the blue stars in some of our masks.
• 3.  
  For each galaxy, the coadded one-dimensional spectrum is divided by the master absorption spectrum for its mask, but with two parameters allowed to vary: a shift in wavelength (to account for slit misalignments) and a smoothing factor (to account for the fact that galaxies have broader line-spread functions than do stars). Note that the strengths of the telluric absorption features are kept fixed for this step.

The resulting telluric absorption correction spectra are divided into the galaxy spectra and propagated into the variance spectra.

2.2.3. Flux Calibration

The standard procedure for flux calibration, based on spectroscopic observations of standard stars with the same instrumental setup, is not feasible for LEGA-C. Slit/galaxy alignment varies from exposure to exposure, and throughout the field of view, so that it is impossible to construct a calibration spectrum that can be applied to all galaxy spectra. Instead, we calibrate our galaxy spectra with the aid of well-calibrated photometric spectral energy distributions (SEDs).

The approach used for DR3 is conceptually the same as described in the DR2 paper, but several improvements were made. As demonstrated in Appendix B, the photometric zero-points are improved compared to the original photometry from Muzzin et al. (2013b), and we only use a subset of photometric bands (BVrizYJ) for the SED fit. For consistency with full-spectral fitting results that will be described in a forthcoming paper, we perform the photometric SED fits with the bagpipes code (Carnall et al. 2018) instead of FAST (Kriek et al. 2009), which was used for DR2. We use bagpipes to fit the new photometric SED with double power-law star formation histories (instead of exponentially declining star formation histories, which were used for DR2). Recent work (e.g., Carnall et al. 2019) shows that, especially for star-forming galaxies, rising-then-declining star formation rates are a necessary choice to avoid biases in, e.g., stellar mass estimates. But for our purpose of flux calibration, this change is not particularly relevant, and only made for consistency. The best-fitting templates from the SED fits are compared with the uncalibrated LEGA-C spectra by fitting a fifth-order Legendre polynomial to the ratio of the two. The calibrated spectra (as well as the variance spectra) are then obtained by multiplying the polynomials with the uncalibrated spectra. The resulting, calibrated DR3 spectra differ slightly in overall shape and normalization compared to DR2, as illustrated in Figure 3 for a typical example.

2.3. Hubble Space Telescope Imaging

3.1. Redshifts

3.2. Stellar and Ionized Gas Kinematics

We use the recent Python implementation of pPXF (Cappellari 2017) to estimate stellar and ionized gas velocity dispersions. The procedure is explained in detail by Bezanson et al. (2018b) and in the DR2 paper. Briefly, we use a set of single stellar population models based on high-resolution (R = 10,000) theoretical templates from C. Conroy (private communication). This high resolution allows us to take maximum advantage of the high effective resolution of the LEGA-C spectra (see DR2 paper: R ∼ 3500; FWHM = 86 km s⁻¹ or σ = 36 km s⁻¹). In addition to the stellar continuum model, we add a set of emission lines (see Table 2). The pPXF fit consists of optimizing the combination of the templates, emission lines, a third-order multiplicative polynomial, and an additive polynomial to reproduce the integrated, one-dimensional spectra. The integrated stellar velocity dispersions ${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$ and ${\sigma }_{g,\mathrm{int}}^{{\prime} }$ ¹⁷ are estimated by broadening the templates with Gaussians (no higher-order moments are fit); the stellar and emission-line dispersions are allowed to be different, but emission lines all have the same velocity dispersion in a given spectrum. This produces a stellar velocity dispersion for 3528 of the 3937 spectra with redshift measurements (96.8% for primary targets; 56.7% for fillers—see Table 1). The precision of the resulting uncertainties is assessed in Appendix C. The pPXF fit also refines the initial redshift measurement described in Section 3.1. If no successful pPXF fit was performed, the initial redshift is adopted as the final redshift estimate.

Table 2. Catalog Contents

Catalog Entry	Unit	No. Objects	Value Range	Uncertainty Range	Comment
			(min,max) or (16, 50, 84%-ile)	(16, 50, 84%-ile)
ID_LEGAC	INT	4081	(1, 4081)
ID	INT	4081	(3945, 262197)		Muzzin et al. (2013b)
MASK	INT	4081	(1, 102)		LEGA-C mask id
RAJ2000A	degrees	4081	(149.379760, 150.774770)		J2000
DECJ2000	degrees	4081	(1.747992, 2.804280)		J2000
PRIMARY	INT	(1053, 3028)	(0, 1)		primary (1)/filler (0)
Z		3937	(0.672, 0.808, 0.964)		LEGA-C spec.-z
SIGMA_STARS_PRIME	km s−1	3528	(106, 161, 222)	(7, 13, 24)	${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$
SIGMA_GAS_PRIME	km s−1	2674	(68, 128, 201)	(2, 7, 21) a	${\sigma }_{g,\mathrm{int}}^{{\prime} }$
LICK_HD_A	Å	3147	(0.3, 3.8, 6.3)	(0.3, 0.6, 1.4)	Absorption line
LICK_HD_F		3175	(1.5, 3.3, 4.9)	(0.2, 0.4, 0.9)	""
LICK_HG_A	Å	2899	(−3.2, 1.4, 4.9)	(0.4, 0.7, 1.5)	""
LICK_HG_F	Å	2913	(−0.1, 2.5, 4.5)	(0.2, 0.4, 0.9)	""
LICK_HB	Å	1579	(2.0, 3.3, 4.9)	(0.4, 0.7, 1.6)	""
LICK_CN1	mag	3029	(−0.12, −0.06, 0.03)	(0.01, 0.02, 0.04)	""
LICK_CN2	mag	3033	(−0.07, −0.01, 0.08)	(0.01, 0.02, 0.05)	""
LICK_CA4227	Å	3115	(0.1, 0.7, 1.1)	(0.2, 0.4, 0.8)	""
LICK_G4300	Å	3032	(0.2, 2.5, 4.8)	(0.4, 0.7, 1.5)	""
LICK_FE4383	Å	2800	(0.8, 2.7, 4.3)	(0.5, 1.0, 2.2)	""
LICK_CA4455	Å	2701	(0.1, 0.9, 1.4)	(0.3, 0.5, 1.1)	""
LICK_FE4531	Å	2470	(1.2, 2.5, 3.5)	(0.4, 0.8, 1.9)	""
LICK_C4668	Å	2026	(0.5, 3.6, 6.1)	(0.7, 1.3, 2.9)	""
LICK_FE5015	Å	1120	(1.6, 3.8, 5.2)	(0.6, 1.2, 2.8)	""
LICK_MG1	mag	438	(0.00, 0.03, 0.08)	(0.01, 0.01, 0.03)	""
LICK_MG2	mag	432	(0.046, 0.11, 0.20)	(0.01, 0.02, 0.04)	""
LICK_MGB	Å	761	(1.0, 2.3, 3.7)	(0.3, 0.6, 1.5)	""
LICK_FE5270	Å	525	(0.8, 2.0, 2.8)	(0.4, 0.7, 1.7)	""
LICK_FE5335	Å	435	(0.9, 1.9, 2.7)	(0.4, 0.7, 2.0)	""
LICK_FE5406	Å	320	(0.4, 1.1, 1.8)	(0.3, 0.6, 1.4)	""
D4000_N		2838	(1.23, 1.40, 1.76)	(0.02, 0.03, 0.06)
OII_3727_FLUX	10−19 erg s−1 cm−2 Å−1	2777	(53, 316, 863)	(12, 21, 41) a	Emission line
OII_3727_EW	Å	2765	(1.51, 8.74, 25.9)	(0.32, 0.6, 1.2) a	""
NEIII_3869_FLUX	10−19 erg s−1 cm−2 Å−1	3139	(−5, 14, 47)	(8, 13, 24) a	""
NEIII_3869_EW	Å	3130	((−0.11, 0.28, 0.98)	(0.14, 0.28, 0.55) a	""
HEI_3890_FLUX	10−19 erg s−1 cm−2 Å−1	3190	(3, 27, 71)	(8, 14, 25) a	""
HEI_3890_EW	Å	3181	(0.09, 0.62, 1.55)	(0.18, 0.33, 0.66) a	""
Hd_FLUX	10−19 erg s−1 cm−2 Å−1	3392	(−3, 29, 103)	(8, 13, 25) a	""
Hd_EW	Å	3368	(−0.04, 0.59, 2.13)	(0.13, 0.25, 0.52) a	""
Hg_FLUX	10−19 erg s−1 cm−2 Å−1	3195	(1, 61, 212)	(8, 16, 28) a	""
Hg_EW	Å	3163	(0.04, 1.25, 4.29)	(0.13, 0.26, 0.56) a	""
OIII_4363_FLUX	10−19 erg s−1 cm−2 Å−1	3167	(−15, 2, 19)	(8, 14, 26) a	""
OIII_4363_EW	Å	3205	(−0.21, 0.03, 0.29)	(0.11, 0.20, 0.42) a	""
Hb_FLUX	10−19 erg s−1 cm−2 Å−1	1858	(7, 164, 580)	(11, 19, 34) a	""
Hb_EW	Å	1830	(0.09, 3.26, 10.8)	(0.13, 0.28, 0.58) a	""
OIII_4959_FLUX	10−19 erg s−1 cm−2 Å−1	1501	(4.2, 44, 149)	99, 17, 30) a	""
OIII_4959_EW	Å	1529	(0.03, 0.43, 2.0)	(0.06, 0.11, 0.24) a	""
OIII_5007_FLUX	10−19 erg s−1 cm−2 Å−1	1372	(28, 127, 429)	(12, 22, 42) a	""
OIII_5007_EW	Å	1348	(0.36, 1.5, 6.7)	(0.13, 0.26, 0.61) a	""
SERSIC_RE	arcsec	3800	(0.29, 0.61, 1.13)	(0.03, 0.07, 0.13)	Eff. radius (maj.axis)
SERSIC_N		3800	(0.82, 2.53, 5.32)	(0.14, 0.44, 0.92)	Sérsic index
SERSIC_Q		3800	(0.38, 0.64, 0.85)	(0.03, 0.03, 0.03)	Axis ratio
SERSIC_PA	degrees	3800	(−62, 2, 61)	(2.8, 2.8, 2.8)	From North to East
SERSIC_MAG	F814W AB mag	3800	(20.7, 21.5, 22.7)	(0.18, 0.18, 0.18)	Total mag
SERSIC_LIMIT	INT	(3341, 459)	(0, 1)		Galfit constraint flag
SIGMA_STARS_VIR	km s−1	3100	(106, 160, 222)	(7, 12, 24)	Virial vel. disp.
LOG_MVIR	M⊙	3100	(10.80, 11.22, 11.59)	(0.06, 0.08, 0.14)	Virial mass SN
Å−1	3933	(5.3, 12, 24.8) b		Avg. S/N
SN_RF_4000	Å−1	3764	(7, 14.8, 26.4) b		S/N at rest 4000 Å
SN_OBS_8030	Å−1	3904	(6.8, 15.8, 30.9) b		S/N at obs 8030 Å
TCOR		2830	(1.5, 2.4, 6.3)		Completeness corr.
SCOR		2830	(1.7, 2.6, 4.4)		Selection corr.
VCOR		2830	(1.0, 1.0, 1.5)		Volume corr.
FLAG_MORPH	INT	(3824, 243, 14)	(0, 1, 2)		Morphology flag
FLAG_SPEC	INT	(3949, 107, 25)	(0, 1, 2)		Spectroscopy flag

Note. Data release catalog. Percentile value ranges are given for those objects for which the quantity was measured; that is, nondetections and failed measurements are not included.

^a These are the formal measurement uncertainties. As described in Section 3.3, there is a systematic uncertainty of 15% due to unknown variations in slit losses. ^b The S/N values are given for the full sample, not just the primary targets that have higher S/Ns than nonprimary targets.(This table is available in its entirety in FITS format.)

Download table as:  ASCIITypeset images: 1 2

3.3. Emission-line Fluxes and Equivalent Widths

3.4. Absorption Line Indices

4.1. Spectra

The primary data product consists of the spectra and corresponding weights (inverse variance spectra). Figure 4 compiles 2707 emission-line subtracted spectra with measured stellar velocity dispersions and Hγ_A absorption indices, visualizing the rich content of the data set. First, we see how the Balmer break seen for young galaxies gradually transitions into the 4000Å break for old galaxies. Second, although the old galaxies show stronger metal features, the spectra of individual young galaxies show the same features, enabling the measurement of their star formation histories and metallicities.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For a more detailed look into the characteristics of the data, we examine a small subset of galaxies in Figure 6, which shows four very different types of galaxies, from blue/late-type to red/early-type. All steps in the analysis described above are illustrated here. The Doppler broadening of the absorption and emission-line features—decomposed into two components: continuum and emission lines—produces velocity dispersion measurements (for the two examples in the top two panels, the gas and stellar dispersions agree well). The bottom panel demonstrates the difficulty of finding a single-size-fits-all analysis method: the high-spectral resolution theoretical stellar population models used for stellar velocity dispersion measurements do not fit the spectra of old galaxies as well as models based on empirical stellar libraries. This mismatch, which may be the result of missing species in the theoretical templates and/or nonsolar abundance ratios, will be explored in future work.

Zoom In Zoom Out Reset image size

Figure 7 shows four examples of nonstandard objects. We see an early stage merger, a projected pair (that would have been classified as a merger without a spectrum), and a strong gravitational lens—these three objects are flagged in the catalog as described in Section 4.2. Finally, we show an outflow seen in Mg absorption at z = 1.3. The LEGA-C sample size guarantees the frequent occurrence of such cases, and makes for a fascinating treasure trove.

Zoom In Zoom Out Reset image size

4.2. Catalog

In Table 2 we list the contents of the electronic public data release. For the 4081 spectra, we provide 61 quantities and, where relevant, their uncertainties. For each quantity, we indicate for how many of the spectra it has been measured and with what range in values and uncertainties. For example, we have a median redshift (Z_SPEC) of z = 0.808 and 3528 measured stellar velocity dispersions ${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$ (SIGMA_STARS_PRIME) with a median value of 161 km s⁻¹ (the sixteenth to eighty-fourth percentile range is 106 to 222 km s⁻¹) and median uncertainty of 13 km s⁻¹ (the sixteenth to eighty-fourth percentile range is 7 to 24 km s⁻¹).

The user should be aware of various meta parameters and peculiarities of the data set. ID_LEGAC identifies unique spectra, but because of the duplicate observations, there are cases with multiple spectra of the same objects, and some ID entries from the Muzzin et al. (2013a) catalog appear more than once. The PRIMARY flag indicates whether the galaxy was selected from the K_s -band selected parent sample (1) or as a filler (0).

We have created two flags that should be kept in mind when assigning meaning to the measured quantities. FLAG_SPEC: a subset of spectra show clear evidence of AGNs affecting the continuum shape, compromising the interpretation of the index measurements. Upon visual inspection, we decided to flag all 107 galaxies with mid-infrared and/or X-ray AGNs even if the majority do not show an obvious issue; the catalog entry FLAG_SPEC = 1 indicates such AGNs. Narrow-line and radio AGNs do not present a problem for our spectral analysis and are not flagged. However, in 25 cases, we found that the photometry-based flux calibration showed significant imperfections, potentially compromising the measurement of absorption and emission indices. These are indicated by FLAG_SPEC = 2.

FLAG_MORPH: in most cases, the light coming through the slit is from a single galaxy with a regular morphology, but in a significant minority of cases, this is not the case. In order to address this, we devise the catalog parameter FLAG_MORPH with value 0, 1, or 2. Single galaxies with regular morphologies have a value 0. In spectra for which the light coming through the slit does not come from single, regular galaxies get a flag value 1; these are often irregular galaxies such as merger remnants, but also multiple galaxies that are separated in the HST image, but not in the spectrum. The guiding principle when assigning flag values of 1 is that the combination of the stellar velocity dispersion (from the spectrum) and structural parameters (from the HST image) cannot be included in an analysis of the fundamental plane or other scaling relations. Spectra that have, on top of this, the problem that the light comes from galaxies at different redshifts receive a flag value 2. Again, the guiding principle is that the presence of a secondary object at the different redshift prevents us from using the stellar velocity dispersion and the structural parameters for a scaling relation analysis. In total, 257 are flagged, 14 of which have flag value 2. Many more galaxies have low-level "contamination" from secondary sources in the slit, but not to the extent that the measurements presented here are ambiguous in their interpretation.

Note that the measurements as presented in the catalog are not wrong for spectra with FLAG_MORPH value 1 or 2—the flag simply alerts the user to a possible issue with the interpretation of the measurement, not an issue with the measurements themselves. For example, we have a spectrum with a correctly measured velocity dispersion of 623 km s⁻¹, and close inspection of the HST image reveals a double nucleus. This combination indicates that we are seeing an early stage merger where the two components happen to be aligned almost perfectly or closely separated in real space. Obviously, the high stellar velocity dispersion does not trace the internal dynamical structure of either component, but rather the orbital motion of the pair.

The structural parameters (Section 2.3) for 3800 objects are also included in the catalog. Most missing values are the result of imperfect overlap between the LEGA-C and COSMOS-ACS footprints. We give an SERSIC_LIMIT flag to indicate for which objects the parameters reached the galfit limiting value (e.g., n = 6).

Finally, we publish the virial mass (LOG_MVIR) and its uncertainty, derived as explained in an accompanying paper (J. van Houdt et al. 2021, in preparation). In short, it is given by

$$\begin{eqnarray}&&{M}_{\mathrm{vir}}=K(n)\displaystyle \frac{{\sigma }_{\star ,\mathrm{vir}}^{2}{R}_{\mathrm{sma}}}{G},\end{eqnarray} \tag{ 1 }$$

where n is SERSIC_N, K(n) = 8.87–0.831n + 0.0241n², R_sma is the semimajor axis length SERSIC_RE in kpc, G the gravitational constant, and σ_⋆,vir (SIGMA_STARS_VIR) is the inclination- and aperture-corrected velocity dispersion:

$$\begin{eqnarray}&&{\sigma }_{\star ,\mathrm{vir}}={\sigma }_{\star ,\mathrm{int}}^{{\prime} 2}\left(0.87+0.39\cdot {e}^{-3.78(1-q)}\right),\end{eqnarray} \tag{ 2 }$$

and q is the projected axis ratio (SERSIC_Q). In principle, at least for flat galaxies, one might expect a difference in σ_⋆,vir (and ${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$), depending on the angle between the major axis of the galaxy and the slit, but we find no significant evidence. This might be due to the small sample size of just ten duplicate observations for flat galaxies (b/a < 0.5) and for which the position angle differs by more than 45 degrees, and for those, we can only constrain the effect on σ_⋆,vir (and ${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$) to be less than 10%. Larger samples are needed to provide a more precise constraint.

Our new definition of the virial mass is different from (and up to 50% higher than) what has been the norm in the high-redshift community: instead of a circularized radius, we use the major-axis length, and we use an inclination-corrected velocity dispersion relying on the observational result that most galaxies are oblate rotators (e.g., Weijmans et al. 2014; Foster et al. 2017). (Most intrinsically round ellipticals have q ∼ 0.8, and therefore have an inclination correction near unity.) These masses were calibrated against dynamical models applied to spatially resolved stellar kinematic measurements.

In order to visualize the increase in sample size and parameter space compared to previous work that LEGA-C represents (not counting previous LEGA-C data releases), we show the size-${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$ distribution in Figure 8. Before LEGA-C, the combined efforts of observing programs with the goal to collect spectra with sufficient quality to determine the stellar velocity dispersions for individual galaxies with HST imaging in the 0.6 ≲ z ≲ 1 redshift range yielded 231 such measurements (van der Wel et al. 2005; Treu et al. 2005; Gallazzi et al. 2014; Bezanson et al. 2015), of which the vast majority are quiescent and very massive due to sample selection. A subset of 119 of these have stellar population measurements from Gallazzi et al. (2014) and Shetty & Cappellari (2015). ²¹ We note that the Smithsonian Hectospec Lensing Survey (Geller et al. 2016), which mostly targets galaxies at z ∼ 0.3, has a high-redshift tail of ∼60 galaxies at z ∼ 0.6. Their σ_* and stellar population properties have been used in publications (Zahid et al. 2016) but not published in tabulated form.

Zoom In Zoom Out Reset image size

LEGA-C expands the sample of galaxies with measured and stellar population properties and velocity dispersions to over 3000, while also greatly expanding the parameter space of galaxy mass and type. Note that we only counted "field" galaxies in this comparison—dozens of cluster galaxies in this redshift range also have σ_* measurements (van Dokkum et al. 1998; Wuyts et al. 2004; Holden et al. 2010; Jørgensen & Chiboucas 2013). In addition, field galaxy measurements at z > 1 now also number in the dozens (van Dokkum et al. 2009; Newman et al. 2010; Toft et al. 2012; van de Sande et al. 2013, 2014; Belli et al. 2014; Longhetti et al. 2014; Newman et al. 2015; Hill et al. 2016; Belli et al. 2017; Toft et al. 2017; Saracco et al. 2019).

The near-absence of a correlation between galaxy size and velocity dispersion—this is an almost face-on projection of the fundamental plane—vividly reveals the connection between galaxy structure and stellar population content. High-σ galaxies have weaker Hδ absorption (that is, older ages), a trend that is seen for the population as a whole as well as for UVJ-selected star-forming galaxies. Within the quiescent population, there is a smattering of compact galaxies with strong Hδ absorption, which we can readily identify as post-starburst galaxies (Wu et al. 2018a, 2020). It goes far beyond the scope of this paper, but Figure 8 clearly invites an analysis as presented for present-day galaxies by Graves et al. (2009) with the added advantage of being able to measure evolution in the size-${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$ plane.

Figure 9 makes the explicit connection between galaxy morphology, structure, kinematics, and stellar population age. Morphologically late-type galaxies generally have young ages and low stellar velocity dispersion, whereas early-type galaxies have old ages and high σ_*. At fixed σ_*, we see, from young to old age, a change from late- to early-type morphology, and an increase in Sérsic index. Galaxy morphology at fixed Hδ shows less variety or trend than at fixed σ_*; although for late types, we do see an increased Sérsic index at high σ_*, that is, increased bulge prominence. In short, the visual (morphological) appearance of a galaxy in this redshift range first and foremost reflects its age, rather than its dynamical structure. A more in-depth analysis is beyond the scope of this data release paper, but Figure 9 shows the potential of LEGA-C to produce a significant leap forward in our understanding of galaxy formation and evolution by providing for the first time Sloan Digital Sky Survey–quality spectroscopy for the general galaxy population 7 Gyr ago. We encourage the community to take advantage of this unique, publicly available data set.

Zoom In Zoom Out Reset image size

Based on observations made with ESO Telescopes at the La Silla Paranal Observatory under program IDs 194-A.2005 and 1100.A-0949 (The LEGA-C Public Spectroscopy Survey). This project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement No. 683184). C.M.S.S. acknowledges support from Research Foundation—Flanders (FWO) through Fellowship 12ZC120N. C.P. is supported by the Canadian Space Agency under a contract with NRC Herzberg Astronomy and Astrophysics.

Software: Astropy (Astropy Collaboration et al. 2013), Matplotlib (Hunter 2007), galfit (Peng et al. 2010), XCSAO, (Kurtz et al. 1992).

The measurements from high-redshift galaxy spectroscopic surveys are often difficult to connect to the full galaxy population in terms of their completeness. Three factors play a key role in establishing this connection:

• 1.  
  How does the parent sample from which the spectroscopic targets are chosen connect to the full galaxy population in terms of its physical parameters? (For example, for LEGA-C, how does the K_s magnitude limit connect to a stellar mass limit?)
• 2.  
  Given the parent sample, how are targets prioritized for observation? (If chosen randomly, this is not an issue; for LEGA-C, brighter targets in K_s have higher priority.)
• 3.  
  What is the success rate of measuring the desired quantities? (For LEGA-C, these are stellar velocity dispersions and stellar population properties for which the success rate is very high: 97%.)

The LEGA-C survey is designed to control all three factors. Factor 1: our K_s -band selection of the parent sample from which the targets are chosen is fairly close (but clearly not identical) to a stellar mass selection, with the added advantage that the stellar mass is a model-dependent quantity that may change over time, while the interpretation of a magnitude is fixed. To illustrate the effect of our parent sample selection—${K}_{s}\lt {K}_{{\rm{s}},\mathrm{LIM}}$ as described in Section 2 and Figure 1—on completeness in terms of stellar mass, we show in Figure A1 the stellar mass-redshift distribution of the primary targets. ²² ${K}_{{\rm{s}},\mathrm{LIM}}$ results in a redshift-dependent stellar mass limit above which all galaxies satisfy ${K}_{s}\lt {K}_{{\rm{s}},\mathrm{LIM}}$, and below which this is not the case.

Zoom In Zoom Out Reset image size

In Figure A2 we show the color-dependence on the mass completeness limit. This effect is illustrated in more detail in Figure A3 where we show, for specific stellar mass and redshift bins, the color–color distribution of the LEGA-C targets, the galaxies in the parent sample from which the targets are chosen, and galaxies that are fainter than ${K}_{{\rm{s}},\mathrm{LIM}}$ (if they exist). There we see that below the mass limit the parent sample is incomplete in terms of the reddest (dusty) galaxies. This implies that this completeness limit in reality depends on the science question at hand: if a user is interested in face-on blue disks, the stellar mass completeness limit will be much lower. Figures A1 and A3 merely serve to illustrate the mechanism by which incompleteness can be assessed, depending on one's requirements.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Factor 2: galaxies from the K_s -selected parent sample are sorted by K_s flux and included in the mask designs in order of decreasing flux. As a result, a larger fraction of K_s -bright galaxies from the parent sample are included in the survey compared to K_s -faint galaxies. In fact, the fraction of included galaxies (with respect to the parent sample) is a strictly monotonously increasing function of K_s magnitude, which means that we know the probability of survey inclusion based on (and only on) K_s magnitude. This probability is the inverse of the sample correction factor (Scor in the catalog—see Section 4.2). In Figure A3 the filled data points (included in LEGA-C) are representative of the blue data points (the parent sample) when taking into account the completeness via Scor.

A second correction—the volume correction factor Vcor—accounts for the fact that a galaxy at z = 0.7 that is brighter than ${K}_{{\rm{s}},\mathrm{LIM}}$ might not be sufficiently bright if it had been at z = 0.9. The product of Scor and Vcor—which we call Tcor in the catalog—allows for the calculation of volume-limited quantities such as the stellar velocity dispersion function. The extent to which this is useful depends on the amount of evolution in the time between z = 1 and z = 0.6; the correction Tcor implicitly assumes z = 0.6 and z = 1 galaxies are the same.

Factor 3: crucially, the observed primary sample is highly successful in measuring redshifts and stellar velocity dispersion measurements (Figure A4). Only 11 (0.4%) of primary targets do not have a redshift measurement, for a variety of technical and astrophysical reasons. Two galaxies had previously measured spectroscopic measurements that turned out to be incorrect; three have poor spectra for technical reasons (e.g., vignetting), four spectra suffer from contamination by lower-redshift interlopers, preventing the useful extraction of the spectrum, and two galaxies have good spectra without absorption or emission features. Ninety-seven (3%) do not have a stellar velocity dispersion measurement; a measured velocity dispersion also serves as an indication that the stellar population properties (in the form of absorption line index values) can be measured with reasonable precision. A stellar velocity dispersion measurement most commonly fails among the faintest, reddest galaxies. But even among those 148 galaxies with i-band magnitudes fainter than 22.5, 85% have a successful stellar velocity dispersion measurement, and all have redshift measurements. Other causes that lead to a failed σ_* measurement are AGNs and (very rarely) flux calibration errors. These issues are flagged as described below. The implication is that measurement success plays a negligible role in assessing the survey completeness.

Zoom In Zoom Out Reset image size

In summary, connecting the LEGA-C sample to the full galaxy population at z = 0.6–1 is possible thanks to the high success rate of relevant measurements and the well-understood selection function.

One issue has been neglected in the above analysis: we have assumed that the photometric redshift used for target selection is identical to the actual redshift (as measured on the basis of LEGA-C spectra). In reality there is an rms of 0.02 in Δz = z_phot − z_LEGA−C for primary targets. For 97% of the primary sample, this scatter does not matter: (z_phot and z_LEGA−C are both in the redshift range 0.6 < z < 1.0).

The other 3% can be divided into two groups. First, the group that arises due to random scatter near the redshift limits z = 0.6 and z = 1. We assign to this group all galaxies (2%) with 0.56 < z_LEGA−C < 0.60 and 1.00 < z_LEGA−C < 1.04, that is, within 2σ of the redshift limits. These are still treated as primary targets and included in the above completeness and volume corrections. This does not artificially increase the number density, since there is a set of galaxies that is not included in the survey while they should have been: galaxies with 0.6 < z < 1.0 but with 0.56 < z_phot < 0.6 or 1.00 < z_phot < 1.04. Since the scatter in Δz is symmetric, the sets of galaxies that were accidentally included and accidentally excluded cancel out in the above volume correction calculation.

The second group consists of galaxies (1%) with more strongly deviating redshift measurements: z_LEGA−C < 0.56 or z_LEGA−C > 1.04. These "catastrophic" outliers are excluded from the primary sample and not taken into account when calculating the completeness and volume corrections, which has a negligible effect on the analysis given their small number.

We use a subset of the photometry from Muzzin et al. (2013b, hereafter M13) with revised zero-points. In M13, zero-point corrections are applied that remove systematic differences between the initially calibrated photometry and template SEDs. While this greatly aids photometric redshift estimates, systematic errors in the colors of galaxies can be propagated into the catalog if the templates themselves—which are based on a particular flavor of stellar population synthesis models—contain systematic errors. We will show below that small but significant systematic offsets exist in the rest-frame UV and near-IR parts of the spectrum. But first we reassess the relative flux calibrations of the M13 UltraVISTA catalog.

In Figure B1 (left-hand panel) we show that the near-IR photometry of stars in the original M13 catalog does not coincide with the synthesized photometry of the Pickles (1998) library of stellar spectra. ²³ An examination of the photometry in all nine filters listed in Table B1 made clear that the K_s photometry is systematically offset from the other UltraVISTA filters (Y, J, and H) by 0.07 mag. Applying this shift aligns the observed and synthesized stellar photometry well. We elected to keep the total K_s band magnitude unchanged in order to not change the absolute calibration of the source detection filter (K_s ). For the optical filters V, r, i and z, similar—marginally smaller—offsets were required to align the photometry (Table B1). For the B band, a larger shift was required (see Figure B2). All in all, only the B and K_s filters received a relative zero-point correction larger than 0.03 mag. The filters V, r, i, z, Y, J, and H were already consistent with each other to within 3%.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Now that we have accurate zero-points that are independent of stellar population synthesis models, we are in a position to verify those models for systematic errors. In Figure B3 (left-hand panel) we show the J–H and H–K_s colors with revised zero-points for a sample of 76 quiescent galaxies (UVJ selected) with stellar velocity dispersions σ > 200 km s⁻¹ and in the redshift range 0.72 < z < 0.74, chosen to contain a significant over-density of massive galaxies. The colors are in the observed frame at z = 0.73, and the filters correspond to rest-frame wavelengths of ∼0.7, 0.95, and 1.3 μm. In the right-hand panel, we show the i_p –Y and Y–J colors for the same sample (rest frame: ∼0.44, 0.6, and 0.7 μm). The majority of these galaxies have been shown on the basis of their high-quality optical spectra from LEGA-C to have little or no star formation in the past few gigayears and ages typically in the range of 3–5 Gyr (Chauke et al. 2018). Their colors should resemble those of single stellar populations. We compare these colors with synthesized photometry for a variety of stellar population synthesis (SPS) models with ages in the range of 2 to 6 Gyr. The flexible stellar population synthesis (FSPS) models (Conroy et al. 2009) and the BC03 models use different Padova evolutionary tracks: Marigo & Girardi (2007) and Bertelli et al. (1994), respectively. Neither solar nor super-solar models match the observed H–K_s colors, and attenuation is unlikely to explain the offset both due to mismatched "direction" in color–color space and the direct evidence for low attenuation even at shorter wavelengths for this type of galaxy. Considering this set of photometric measurements in isolation, both the Maraston and FSPS models can match the data by adopting high metallicities, younger ages (0.5–1 Gyr), and a significant amount of attenuation, but these parameter ranges are firmly ruled out by independent constraints on age (from the LEGA-C spectra) and attenuation. In the rest-frame optical wavelength range (probed by the colors used in the right-hand panel of the figure), no such wholesale offsets are apparent, and the model colors generally correspond well with the observed colors. The result that no large changes in metallicity, age, and attenuation are required to match the optical SED suggests that the problems lie with the models in the near-IR.

Zoom In Zoom Out Reset image size

In order to analyze the near-infrared offsets further, we use the fitting code Prospector (Leja et al. 2019). We provide the photometry in the wavelength range B to K_s for the filter set given in Table B1, but H and K_s are not used in the fit; instead, residuals are calculated based on the models that are fit to the B through J photometry. ²⁴ In Figure B4 we show the residuals in H–K across redshift. These differences between observed and inferred H–K_s colors are to first order independent of galaxy type (star formation activity) but exhibit strong change with redshift. The implication is that the color offsets seen for old galaxies (Figure B3) are present in all galaxies in the redshift range z = 0.6–1: the models systematically underpredict the H–K_s colors. The redshift dependence suggest that the photometric measurements are not the problem: these would be redshift independent.

Zoom In Zoom Out Reset image size

We conclude that using photometry longward of rest frame ∼0.8 μm leads to systematic errors in fits to the SED of ∼20%, which will propagate into small but systematic errors in the stellar mass, star formation rate, attenuation, and other parameters that are inferred. For the purpose of calibrating our spectra, we choose the BVrizYJ filter set. We note that if H and K_s are included in the fits, then the attenuation parameters are adjusted to accommodate for the systematic effect. We do not fully explore this here, but one illustrative result is that passive galaxies typically require A_V ∼ 0.5, which is ruled out by other means (optical depth estimates; attenuation measurements from spectra, etc.). However, stellar mass estimates differ by no more than ∼0.05 dex on average if H and K_s are included in SED fits.

The formal uncertainties, based on the noise spectra, may not fully capture the full random error budget due to correlations in the noise spectra and systematics in the spectra that propagate to random errors. For specific measurements, there may be specific contributions: template mismatch for σ_*, the quasi-random wavelength coverage of emission lines for σ_g , flux calibration uncertainties for spectral index measurements, and so forth.

By design, LEGA-C targeted hundreds of galaxies multiple times, providing an automatic assessment of the complete budget of spectrum-to-spectrum random uncertainties. We compare the variance in the differences between duplicate measurements to the expected variance inferred from the formal measurement uncertainties, and correct the latter if necessary. This approach was already used in DR2—here we show that due to improvements in the data reduction and analysis techniques, such corrections are generally smaller than before. For the analysis, we choose the sample of 205 galaxies for which we have duplicate spectra with S/N > 5 and successful redshift measurements. (We omit galaxies in mask 101, which were taken with the slits perpendicular to the rest of the survey. This leads to physically different velocity dispersions due to projection effects along the slit.) The scatter between the duplicate measurements is compared with the random uncertainty (added in quadrature), producing a $\sqrt{{\chi }^{2}/N}$ value to parameterize the excess scatter compared to the scatter that can be expected on the basis of the formal measurement uncertainties. This number serves as the correction factor on our measurement uncertainties, propagated into our published catalog.

We first examine the duplicate measurements of ${\sigma }_{\star ,\mathrm{int}}^{{\prime} }$ (Figure C1, left-hand side), and conclude that the true uncertainties are 50% higher than the formally derived uncertainties. There are several possible causes: template mismatch, correlated noise, and wavelength coverage. For ${\sigma }_{g,\mathrm{int}}^{{\prime} }$ (Figure C1, right-hand side), we infer a similar correction of 1.5, bearing in mind the 15% systematic uncertainty discussed in Section 3.3 due to emission-line-specific slit losses.

Zoom In Zoom Out Reset image size

Figure C2 shows the results for duplicate index measurements for three spectral features. In all cases, the formal uncertainties, indicated with the error bars, are somewhat smaller than the scatter among duplicates. Analogous to the correction factors on the velocity dispersion uncertainties, we measure correction factors for all absorption indices of a factor 1.3. Only LICK_Hβ (1.8) and D_N 4000 (2.0) receive larger correction factors. The measurement of the former is complicated by the often-bright emission line that is subtracted before the absorption strength is measured. The precision of the measurement of the latter is limited by imperfections in the wavelength dependence of the flux calibration.

Zoom In Zoom Out Reset image size

Formal uncertainties on the emission-line fluxes and equivalent widths are more difficult to determine due to the aforementioned systematic problems with variable slit losses. The analysis of the LICK_Hβ absorption and emission uncertainties, driven in large part by the decomposition of the spectrum into ionized gas emission and stellar light, led us to adopt a correction factor of 1.8. This correction factor is applied to all emission-line measurements in the catalog, again keeping in mind a 15% systematic uncertainty that is not propagated. These results compare well with similar tests on emission lines in the Sloan Digital Sky Survey (Brinchmann et al. 2008). They find typical corrections to the uncertainty estimates are a factor 1.2–1.5 for forbidden lines and a factor of 2.0 for Balmer lines and OII_3727.