The Magellan Evolution of Galaxies Spectroscopic and Ultraviolet Reference Atlas (MegaSaura). I. The Sample and the Spectra

The MegaSaura sample is comprised of N = 15 bright strongly lensed galaxies, spanning the redshift range 1.68 < z < 3.6, including many of the brightest lensed galaxies known. The criteria for target selection were: highly magnified lensed galaxies, as bright as possible in the observed g-band, with an image plane configuration such that a significant portion of the total emission would fit into a 10'' MageE slit with room for sky on both ends of the slit, with a redshift such that both Lyα and C iv are observable by MagE, and at a sufficiently southern decl. to be observable by Magellan.

Most of the MegaSaura targets are behind cluster–scale lenses that were discovered through two related surveys: SDSS Giant Arcs Survey 1 (SGAS1; M. G. Gladders et al. 2018, in preparation), and SDSS Giant Arcs Survey 2 (SGAS2; H. Dahle et al. 2018, in preparation). SGAS1 has yielded 217 lenses, with a median giant arc redshift of z = 2 (Bayliss et al. 2011; Bayliss 2012). SGAS2 expanded the search space within the Sloan Digital Sky Survey (SDSS), resulting in the discovery of 62 additional lenses.

We also included three targets from other samples: the bright lensed galaxy RCS-GA 032727−132609 (Wuyts et al. 2010), which was discovered through the Red-sequence Cluster Survey-2 (RCS-2; Gilbank et al. 2011), and two bright galaxy-scale lenses: the Cosmic Eye (Smail et al. 2007) and the Cosmic Horseshoe (Belokurov et al. 2007).

Table 1 lists the discovery paper for each arc; several targets were independently discovered by multiple groups. Table 2 lists oxygen abundances as known for the sample; seven galaxies have measurements or constraints. The measurements range from <25% of solar to 81% of solar, with a median of 37% of solar (where solar is from Asplund et al. 2009.) Several of the MegaSaura galaxies have redshifts such that the rest-frame optical oxygen abundance diagnostics are not accessible from the ground; access to these diagnostic lines was not a selection criterion. Oxygen abundances for other SGAS galaxies, not in the MegaSaura sample, were published in Wuyts et al. (2012b); the median and quartile ranges are ${66}_{-36}^{+27} \%$ of solar.

Table 2.  Oxygen Abundances for the MegaSaura Sample

Object	log([N ii]/Hα)	12 + log(O/H)	Method	Notes	Reference
SGAS J000451.7−010321	<−1.4	<8.09	Pettini & Pagel (2004) linear		see Note below
SGAS J003341.5+024217
SGAS J010842.2+062444
RCSGA 032727−132609	−1.19 ± 0.07	8.22 ± 0.04	Pettini & Pagel (2004) linear		Rigby et al. (2011)
SGAS J090003.3+223408		8.18 ± 0.14	Pettini & Pagel (2004) linear	weighted average of 2 measurements	Bian et al. (2010)
SGAS J095738.7+050929
SGAS J105039.6+001730		8.3 ± 0.1	multiple R23 methods		Bayliss et al. (2014)
Cosmic Horseshoe SE	−0.79 ± 0.04	8.45 ± 0.03	Pettini & Pagel (2004) linear	their "Aperture 2"	Hainline et al. (2009)
SGAS J122651.3+215220
SGAS J142954.9+120239
SGAS J145836.1−002358
SGAS J152745.1+065219	<−0.65	<8.5	Pettini & Pagel (2004)		Wuyts et al. (2012a)
SGAS J211118.9−011431
Cosmic Eye		8.6	R23, Pettini & Pagel (2004), Pilyugin & Thuan (2005)	assumed upper branch	Stark et al. (2008)
		${7.97}_{-0.23}^{+0.32}$	Maiolino et al. (2008)		Troncoso et al. (2014)
SGAS J224324.2−093508

Note. Compilation of measured oxygen abundances for the MegaSaura galaxies. We list the measured log([N ii] 6583/Hα) ratio and the resulting inferred oxygen abundance (commonly called the "metallicity") in the format 12 + log(O/H). The measurement for SGAS J0004 is from unpublished FIRE/Magellan spectra, obtained UT 2010 October 14 and UT 2013 September 13, with a total integration time of 4.18 hr.

A machine-readable version of the table is available.

Download table as:  DataTypeset image

The MegaSaura sources are exceptionally bright: the brightest has g_AB = 19.15 (Wuyts et al. 2010), and most have g_AB ∼ 20–21 (Koester et al. 2010; Figure 5 of Wuyts et al. 2012a). For comparison, individual field galaxies that went into the composite spectrum of Shapley et al. (2003) have g_AB ∼ 24.7 (Erb et al. 2006). The MegaSaura targets are many of the brightest lensed sources selected from the SDSS.

Table 1 lists the targets, coordinates, and total integration times. Figures 1 and 2 contain findercharts for the MegaSaura sample.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Portions of the MegaSaura spectra have been published previously. Koester et al. (2010) published smoothed spectra for SGAS J122651.3+215220 and SGAS J152745.1+065219. Rigby et al. (2014) pointed out the lack of correlation between the Mg ii emission and Lyα emission in five of these galaxies. Bayliss et al. (2014) published the MagE spectrum of the highest redshift galaxy in the sample, SGAS J105039.6+001730 at z = 3.6252. Rigby et al. (2015) analyzed [C iii] 1907, C iii] 1909 Å emission for the sample. Bordoloi et al. (2016) analyzed variation in the Mg ii and Fe ii emission and absorption among four star-forming knots within lensed galaxy RCS-GA 032727−132609.

Observations were obtained with The Magellan Echellette (MagE) spectrograph (Marshall et al. 2008) on the twin 6.5 m Magellan telescopes at the Las Campanas Observatory in Chile. The majority of observations were conducted with MagE mounted on the Clay telescope; some observations were made after 2015 October, when MagE was moved to the Baade telescope. Targets were acquired by blind offsets from bright nearby stars; target acquisition was verified via the slit-viewing guider camera. MagE has eight available slits, all 10'' in length. We used slit widths of 07, 085, 10, 15, and 2'', with the slit width chosen to match the atmospheric seeing and the compactness of the emission.

For two galaxies, RCS-GA 032727−132609 and SGAS J152745.1+065219, we obtained spectra of multiple physically distinct regions, as noted in Table 1 and on the findercharts in Figures 1 and 2.

The broad wavelength coverage of MagE makes it important to keep the slit position angle close to the parallactic angle, especially when the airmass is significant (Filippenko 1982). This was done by advancing the slit angle to where the parallactic angle was expected 30 minutes hence, acquiring the object, exposing (typically for one hour), and repeating. A few exceptions to this rule were made to keep contaminating objects out of the slit. The slit position angle and parallactic angle are noted in Table 3.

Because these lensed galaxies are highly spatially extended, and because we maintained the parallactic angle, we generally obtained spectra for only a portion of each lensed galaxy. This spatial coverage is illustrated in Figures 1 and 2, which show the slit position angles obtained for each galaxy in the sample.

We obtained calibrations nightly, following the recommended sequences in the MagE manual. Internal Th–Ar lamps were obtained for wavelength calibration. Xe-flash lamp exposures, using the same slit as the science observations, were obtained to define where the spectral orders fall on the detector.

The wide wavelength coverage of MagE makes it a challenging instrument to flat field. Our strategy was to take three kinds of flat-field calibrations: calibrations for the bluest two orders, for the other blue orders, and for the red orders. To obtain a high-quality flat field for the bluest two orders, we obtained a sequence of calibration frames ("very blue flats") using the internal Xe-flash lamp, long (100 s) integration times, and with the 5'' slit and the instrument defocused to blur out the broad Xe emission lines. These frames have good counts in the two bluest orders and saturate the others. To calibrate the remaining blue orders, we obtained shorter (10–15 s) Xe-flash exposures ("blue flats"), again with the 5'' slit and the instrument defocused. To calibrate the redder orders, we obtained calibration frames ("red flats") using a quartz lamp mounted in the secondary cage and the same slit as the science observations.

Nightly, we observed spectrophotometric standard stars, chosen from the Nearby Supernova Factory Spectroscopy Standards list.¹⁰

For each night, calibrations were generated and the data were processed using the MagE pipeline, which is part of the Carnegie Python Distribution.¹¹ Here, we summarize what the pipeline does.

Data frames first were subtracted of their bias levels, using the overscan regions. Second, Th Ar lamp frames were used to define the regions of the chip covered by the spectral orders. The inter-order regions were then interpolated to generate a non-parametric model of the scattered light.

Next, two flat-fields were generated: the blue flat frames and very blue flat frames were masked of their saturated regions and combined using a weighted average, to produce a blue flat with high signal-to-noise in the blue but saturated in the red. The red flat frames were combined in the same way to produce a red flat, which is unsaturated in the red, but has low signal-to-noise in the blue. The blue flat and the red flat were then combined by masking out saturated regions, and combining with a weighted average to create a flat field. This flat field was then used to flat field the individual spectral exposures.

Individual spectra were then sky subtracted. The sky emission was modeled using routines described in Kelson (2003), but with explicit masking of pixels that are contaminated by the target. The requisite rectification transformations to account for line curvature were derived using routines described in Kelson et al. (2000) that analyze the tracings of cross correlations of image sections in the comparison Th Ar lamp frames.

After sky subtraction, one-dimensional (1D) spectra for the objects were extracted using a multi-exposure optimal extraction algorithm¹² in which the spatial profile of the object in each exposure is characterized by a Gauss–Hermite decomposition with moments that are polynomial functions of order number and position across the detector. With the spatial profile of the object characterized in this manner for each exposure, a single univariate B-spline (Dierckx 1983) representation of the object spectrum can be derived that, in a least-squares sense, best represents all of the available data simultaneously.

The output of the pipeline is an extracted, one-dimensional, wavelength-calibrated spectrum for each echelle order.

In order to flux the spectra, we computed the sensitivity function for each night using the IRAF¹³ tools standard and sensfunc in the noao.onedspec package. The sensitivity functions for each standard star observation typically show a spread in normalization of up to 20%–30% peak-to-peak in the red and up to 40% in the bluest orders; this is consistent with expected seeing variations, as gauged by the Magellan seeing monitor. We scaled the sensitivity functions to the star with the highest throughput to create a composite sensitivity function. To flux calibrate, these sensitivity functions were applied to each spectrum using the IRAF tool onedspec.calibrate.

After fluxing, we combined the overlapping echelle orders, using a weighted average, weighted by the inverse variance, to make a continuous spectrum for each target for each night. Each spectrum was then corrected to vacuum barycentric wavelength. For each galaxy, the spectra from multiple nights were combined with a weighted average, weighted by the inverse variance. For the two galaxies where we obtained spectra of multiple distinct physical regions, as noted in Table 1, we separately combine the spectra of each distinct region.

The output spectra have vacuum barycentric wavelength in Å and flux density f_ν in erg s⁻¹ cm⁻² Hz⁻¹.

There are significant telluric absorption features redward of 6400 Å, primarily the A-band of molecular oxygen, which covers 7590–7660 Å (Wark & Mercer 1965) and the B-band (6860–6890 Å). For observing runs where the featureless standard star EG 131 was observed, we used that star and the IRAF tool noao.onedspec.telluric to correct those telluric features in the two spectral orders that are centered at 6730 and 7570 Å.

The brightest night sky lines, [O i] 5577 Å and [O i] 6300 Å, frequently saturate individual exposures, which results in artifacts at those wavelengths.

The MagE spectrograph is extremely sensitive in the blue, with sensitivity down to 3000 Å. However, the Earth's atmospheric transmission at 3000 Å is only 0.6%, as compared to 17.7% at 3200 Å and 28.7% at 3400 Å (Cox 2000 Table 11.25.) Based on examination of MagE spectra of spectrophotometric standard stars, we chose not to use spectra from the bluest order of MagE and not to use spectra < 3200 Å in the next order. Due to detector fringing, the pipeline does not extract the two reddest orders of MagE. Therefore, we extract spectra over the wavelength range 3200–8280 Å.

The spectra of six MegaSaura galaxies have coverage blueward of the Lyman limit at rest-frame 912 Å. We have summed the flat-fielded, sky-subtracted, two-dimensional spectra for each of these galaxies and note the bluest wavelength where we see the continuum trace; in each case, we see flux down to the Lyman limit, but not blueward. Therefore, we trim the final 1D spectrum of each galaxy at an observed wavelength that is the greater of 3200 Å or (1 + z)912 Å.

To measure the instrumental line spread function, which determines the spectral resolution, we extract and combine the spectra of the night sky line emission in the same way as the object spectra, using the same weights applied to the object spectra. The only difference is that when combining the sky spectra, the observed wavelengths receive no barycentric correction. From the final sky spectra, we identify isolated, bright night sky emission lines, fit each with a Gaussian, and measure the median resolution over the spectrum, where the resolution R ≡ λ/δλ, with δλ defined as the full width at half maximum. The spectral resolving powers range from R = 2500 to R = 4700, with an average of R = 3300, and are tabulated in Table 1. For combined spectra taken with the 2'' slit, the median resolution and standard deviation are R = 2545 ± 37; the observations taken with the 1'' slit have R = 4770 ± 8. As expected for an echellete, the spectral resolution is constant with wavelength. These measured resolutions are higher than the design resolution of R ∼ 4100 for a 1'' slit (Marshall et al. 2008).

We correct the flux density of each spectrum for Milky Way reddening, using the E(B − V) value derived from Pan-STARRS 1 and 2MASS photometry by Green et al. (2015),¹⁴ assuming an extinction-to-reddening constant of R_v = 3.1, and the reddening curve of Cardelli et al. (1989). Table 1 lists the E(B − V) values used.

As discussed above, spectra were only obtained for a portion of each lensed galaxy. The fluxing is therefore appropriate for the observed portion of the galaxy. To illustrate this, we use the example of SGAS J122651.3+215220. Koester et al. (2010) improved on the SDSS DR7 photometry through careful subtraction of neighbors, measuring (g, r, i) AB-magnitudes of 21.14, 20.60, and 20.51. Using kcorrect (Blanton & Roweis 2007), we measure the SDSS AB-magnitudes of the flux-calibrated MagE spectrum to be g, r, i = 22.15, 21.70, and 21.62. The offset between these measurements implies that the slit captured 39%, 36%, and 36% of the total light in g, r, i from this lensed galaxy.

Continuum fitting is necessarily a somewhat arbitrary process. At the high signal-to-noise and spectral resolution of MegaSaura, the problem is compounded because the so-called continuum is not in fact featureless. Rather, it is the sum of the rest-frame ultraviolet spectra of many hot stars, and as such contains numerous weak photospheric absorption features. Different continua are required for different purposes; we have therefore computed several different kinds of continuum fits.

• 1.  
  Hand-fit smooth continua: We interactively fit splines to the continuum of each galaxy, using the x_continuum tool in the XIDL package. The continuum was fit from spectral regions that are free from expected emission or absorption lines at the galaxy redshift (including Lyα), intervening absorption lines, or contamination from night sky lines. x_continuum also employs iterative sigma rejection to remove spectral features. We exclude from the fit the spectral region within λ_r ± 30 Å of C iv 1548, 1551 Å. For each galaxy, we fit the continuum three times, looping over the sample to minimize human error; we then adopt as the continuum the mean of the three fits, and as the uncertainty the error in the mean. This process allows us to quantify the uncertainty due to the subjectivity of manual continuum fitting. The resulting continuum is smooth and has no knowledge of the underlying stellar emission; it is therefore most appropriate for analyzing the velocity profiles of transitions tracing the interstellar medium.
• 2.  
  Automatic continua: We automatically generate smooth continua as follows. For each spectrum, after masking out bad pixels as well as regions with velocities within 500 km s⁻¹ of known emission and absorption features, we smooth by convolving with a boxcar of width λ_rest = 100 Å. The resulting continuum is smooth and has no knowledge of the underlying stellar emission. Because this continuum is automatically generated, it can be applied to a large number of theoretical models.
• 3.  
  Stellar continua: We have generated best-fit spectra from linear combinations of Starburst99 models (Leitherer et al. 1999, 2010, 2014), each an instantaneous starburst with a given age (between 1 and 40 Myr) and stellar continuum metallicity (with metallicities of 0.05, 0.2, 0.4, 1.0, and 2.0 Z_⊙), while simultaneously fitting for a stellar extinction (E(B − V)) from the line of sight dust reddening using a Calzetti extinction law (Calzetti et al. 2000). This fit is made assuming a linear combination of the 50 stellar continuum models, and fitting for the linear coefficient (the fraction of the total light attributed to each Starburst99 model) using a least-squares method following the methodology of Chisholm et al. (2015). Regions within 500 km s⁻¹ of known emission and absorption features were masked before fitting. We use fully theoretically Starburst99 models that are computed using the Geneva stellar evolution tracks with high mass-loss (Meynet et al. 1994), computed using the WM-BASIC method (Leitherer et al. 2010), and assuming a Kroupa IMF with a high (low) mass power-law index of 2.3 (1.3) with a high-mass cut-off of 100 M_⊙. This modeling, and derived constraints on the stellar populations, will be described in a subsequent paper (J. R. Rigby et al. 2018, in preparation) that comprehensively analyzes the diagnostics of hot stars in the MegaSaura spectra. What is relevant for the current paper is that these models fit the underlying stellar spectra and can therefore be used to estimate the redshift of the stars, commonly known as the systemic redshift.

We measure the redshifts of the stars, the nebulae, and the interstellar medium for each galaxy, as follows.

The stellar redshift is an output from the Starburst99 fitting (J. R. Rigby et al. 2018, in preparation), in which we cross-correlate the model and the observed data to find the most likely stellar redshift. We shift the Starburst99 models in velocity-space by 10 km s⁻¹ increments over a range of ±600 km s⁻¹ from the measured nebular redshift. We then produce a likelihood function by taking the inverse exponential of the χ² function using the observed data and the shifted Starburst99 models. We calculate the likelihood at each velocity shift, normalize the likelihood function, and take the expectation value and standard deviation of this probability density function as the best-fit redshift and the redshift uncertainty. As such, the quoted redshift uncertainty does not take into account the full range of possible stellar models.

For 12 of the spectra in the MegaSaura sample, the nebular redshift is measured from the [C iii] 1907, C iii] 1909 Å doublet, as these are generally the brightest rest-frame nebular lines. We determine a redshift for the [C iii] 1907, C iii] 1909 feature by fitting two Gaussians with a common linewidth, holding fixed the ratio of the central wavelengths, using the idl levenbergmarquardt least-squares fitting code mpfitfun (Markwardt 2009), following Rigby et al. (2014). The uncertainty quoted is the uncertainty from MPFITFUN.

Table 4 notes that for five spectra, the nebular redshift is sourced differently, from nebular lines in the rest-frame optical. While lines like Hβ and [O iii] 5007 Å generally have higher fluxes than [C iii] 1907, C iii] 1909 Å, using them may introduce systematic wavelength calibration errors when comparing to stellar redshifts measured from the rest-frame UV continuum.

To test for systematic effects, we measured redshifts individually for [C iii], C iii], and weaker rest-frame UV emission lines (O iii] 1666, Si iii] 1882, Si iii] 1892, C ii] 2325, and [O ii] 2470), in four spectra with high S/N (RCS0327-knotE, S0004–0103, S0108+0624, and S0957+0509.) We note a systematic effect: compared to other rest-frame UV emission lines, [C iii] 1907 is blueshifted in all cases by up to 40 km s⁻¹, while in three of four cases, C iii] 1909 is redshifted by up to 20 km s⁻¹. The standard deviation is 10–20 km s⁻¹. This effect appears to be real; we do not understand its origin. Perhaps the forbidden [C iii] 1907 and semi-forbidden C iii] 1909 arise in different physical regions with different velocities. Perhaps the line profiles are intrinsically asymmetric, driven by winds, and the bias comes from fitting them with Gaussians. In any case, we mitigate this bias by using both transitions to determine the redshift, as described above.

The interstellar medium redshift is measured as the average redshift of Gaussian fits to C ii 1334, Si ii 1526, Al ii 1670, Al iii 1854, and Al iii 1862; the uncertainty is the standard deviation. The measured redshifts are tabulated in Table 4.

Figure 3 shows the per-pixel S/Ns for the final combined MegaSaura spectra. The median per-pixel dispersion is 0.35 Å in the observed frame. A total of eight spectra have per-pixel S/N > 10 over the observed wavelength range of 5000–7000 Å; the median spectrum has a per-pixel S/N ∼ 10. For the average resolution, at 5000 Å this corresponds to S/N = 21 per resolution element.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The final combined MegaSaura spectra are plotted in Figure 4 and are provided in an online supplementary tar.gz archive. The MegaSaura spectra comprise a spectral atlas at "cosmic noon" that has superior S/N and wavelength coverage, albeit lower spectral resolution, than the COS/HST archive of starburst galaxies in the local universe.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Two galaxies in our sample show evidence of active galactic nuclei (AGNs). The MagE spectrum of SGAS J224324.2−093508 shows broad emission lines of Lyα, N v, C iv, and [C iii], that together indicate that this galaxy hosts a broad-line AGN. The HST imagery (Figure 1) shows a spatially extended host galaxy with a central point source. The MagE spectrum of SGAS J003341.5+024217 does not show obvious AGN diagnostics. However, a VLT/SINFONI observation shows that the nucleus exhibits high-ionization [O iii] 5007/Hβ and Hα/[N ii] ratios and also broad wings in Hα (E. Wuyts 2017, private communication.) It is thus implicated as having a weak AGN.

Figure 5 compares the velocity offsets among the interstellar medium, stars, and nebulae of the MegaSaura spectra, measured as described in Section 2.6 and tabulated in Table 4. The top left panel of Figure 5 shows that there is no systematic offset between the redshifts of the hot stars and the nebulae; the median offset and median absolute deviation are −1 ± 31 km s⁻¹; from bootstrapping, the 80% confidence interval on the median is $-{1}_{-11}^{+20}$ km s⁻¹.

Zoom In Zoom Out Reset image size

The top right panel of Figure 5 shows that the ISM lines are systematically blueshifted with respect to the stellar continuum, with a median offset and median absolute deviation of −148 ± 35 km s⁻¹, and bootstrapped 80% confidence interval on the median of $-{148}_{-11}^{+42}$ km s⁻¹. This is to be expected given the ubiquity of galaxy-wide winds in z ∼ 2 galaxies (Shapley et al. 2003; Weiner et al. 2009; Rubin et al. 2014).

We now place these measurements in a larger context. Because purely stellar features like the photospheric absorption lines are only detectable with high signal-to-noise and moderate spectral resolution, it is common practice in the literature to use nebular emission lines as a proxy for the systemic redshift. Doing so makes two assumptions—that the ionized gas is well-mixed with the hot stars, and that outflows do not seriously bias the nebular line redshifts. The local galaxy NGC 7552 shows that such biases can occur: Hα is blueshifted from the stellar redshift by −30 km s⁻¹ due to a wind, but this blueshift is much less than the bulk outflow velocity as traced by rest-frame UV absorption lines. These assumptions are testable for the MegaSaura spectra. In Section 2.6, we reported that the brightest emission lines in the rest-frame UV, [C iii] 1907 and C iii] 1909 Å, may be systematically biased by up to 40 km s⁻¹ with respect to each other, and with respect to still weaker rest-frame UV lines. Because one line is systematically blueshifted and the other redshifted, we attempted to mitigate the bias by jointly fitting the [C iii], C iii] doublet. Indeed, the median offset between the nebular redshifts and systemic redshifts, $-{1}_{-11}^{+20}$ km s⁻¹ (80% confidence) is fully consistent with no offset. This is consistent with the measurement of Shapley et al. (2003) for their stack of Lyman Break Galaxies: −10 ± 35 km s⁻¹.

We conclude that while nebular redshifts can be used as proxies for systemic redshifts, one should be aware of systematic biases at the level of tens of km s⁻¹ when using only one line of the [C iii], C iii] doublet. In the era of the James Webb Space Telescope, such biases may limit kinematic studies. The stellar redshifts we measure, by fitting the whole stellar continuum, should be unbiased to outflows but are less precise than the nebular redshifts (median fractional uncertainty of 3 × 10⁻⁴, versus 4 × 10⁻⁵). Greater precision can be obtained by fitting individual photospheric absorption lines, when they are clearly detected. This is the case for about half the MegaSaura spectra and will be explored in a future paper.