The ALPINE–ALMA [C ii] Survey: Multiwavelength Ancillary Data and Basic Physical Measurements

Galaxy evolution undergoes several important phases such as the ionization of neutral Hydrogen at redshifts z > 6 (also known as the Epoch of Reionizaton) as well as a time of highest cosmic star formation rate (SFR) density at z ∼ 2–3. The transition phase at z = 4–6 (a time roughly 0.9 to 1.5 billion years after the Big Bang), often referred to as the early growth phase, is currently in focus of many studies. This time is of great interest for understanding galaxy evolution as it connects primordial galaxy formation during the epoch of reionization with mature galaxy growth at and after the peak of cosmic SFR density. During a time of only 600 Myr, the cosmic stellar mass density in the universe increased by one order of magnitude (Caputi et al. 2011; Davidzon et al. 2017), galaxies underwent a critical morphological transformation to build up their disk and bulge structures (Gnedin et al. 1999; Bournaud et al. 2007; Agertz et al. 2009), and their interstellar medium (ISM) became enriched with metal from sub-solar to solar amounts (Ando et al. 2007; Faisst et al. 2016b), while at the same time the dust attenuation of the UV light significantly increased (Finkelstein et al. 2012; Bouwens et al. 2015; Fudamoto et al. 2017; Popping et al. 2017; Cullen et al. 2018; Ma et al. 2019; Yamanaka & Yamada 2019). Furthermore, the most massive of these galaxies may become the first quiescent galaxies already at z > 4 (Glazebrook et al. 2017; Tanaka et al. 2019; Faisst et al. 2019; Stockmann et al. 2020; Valentino et al. 2020). All of this put together, makes the early growth phase an important puzzle piece to be studied in order to decipher how galaxies formed and evolved to become the galaxies (either star-forming or quiescent) that we observe in the local universe.

It is evident from studies at lower redshift that multiwavelength observations are crucial for us to be able to form a coherent picture of galaxy evolution. To capture several important properties of galaxies, a panchromatic survey must comprise several spectroscopic and imaging data sets that cover a large fraction of the wavelength range of a galaxy’s light emission, including (i) the rest-frame ultra-violet (UV) containing Lyα emission, as well as several absorption lines to study stellar winds and metallicity (Heckman et al. 1997; Maraston et al. 2009; Steidel et al. 2010; Faisst et al. 2016b), (ii) the rest-frame optical containing tracers of age (Balmer break) as well as important emission lines (e.g., Hα) to quantify the star formation and gas metal properties (Kennicutt 1998; Kewley & Ellison 2008), and (iii) the far-infrared (FIR) continuum and several FIR emission lines (e.g,. [] or [] ) that provide insights into the gas and dust properties of galaxies (De Looze et al. 2014; Pavesi et al. 2019).

Fortunately, the early growth phase at redshifts z = 4–6 is at the same time the highest-redshift epoch at which, using current technologies, such a panchromatic study can be carried out. The rest-frame UV part of the energy distribution at these redshifts has been probed in the past thanks to several large spectroscopic (Le Fèvre et al. 2015; Hasinger et al. 2018) and imaging (Capak et al. 2007; McCracken et al. 2012; Aihara et al. 2019) surveys from the ground as well as imaging surveys with the Hubble Space Telescope (HST; Scoville et al. 2007a; Grogin et al. 2011; Koekemoer et al. 2011). In addition, Hα has been accessed successfully through observations with the Spitzer Space Telescope (Stark et al. 2013; de Barros et al. 2014; Smit et al. 2014; Faisst et al. 2016b, 2019; Rasappu et al. 2016; Smit et al. 2016; Lam et al. 2019). However, the FIR of z > 4 galaxies has only been probed sparsely in the past in less than a dozen galaxies using the Atacama Large (Sub-) Millimeter Array (ALMA; Riechers et al. 2014; Capak et al. 2015; Watson et al. 2015; Willott et al. 2015; Strandet et al. 2017; Carniani et al. 2018; Zavala et al. 2018b, 2018a; Casey et al. 2019; Jin et al. 2019) as well as some as part of Herschel surveys in lensed and unlensed fields (e.g., Egami et al. 2010; Casey et al. 2012, 2014; Combes et al. 2012). Commonly targeted by observations with ALMA is singly ionized Carbon (C⁺) at , which is an important coolant for the gas in galaxies and is therefore broadly related to star formation activity and gas masses (Stacey et al. 1991; Carilli & Walter 2013; De Looze et al. 2014). The [] emission line is one of the strongest in the FIR and is in addition conveniently located in the ALMA Band 7 at redshifts z = 4–6 at one of the highest atmospheric transmissions compared to other FIR lines (see, e.g., Faisst et al. 2017). The origin of [] emission is still debated. In addition to photodissociation regions (PDRs) and the cold neutral medium, a significant fraction can also origin from ionized gas regions or CO-dark molecular clouds (Pineda et al. 2013; Vallini et al. 2015; Pavesi et al. 2016). Also, the increasing temperature of the Cosmic Microwave Background has an effect on the relation between [] and star formation (Ferrara et al. 2019). Both potentially complicates the interpretation of [] as SFR indicator at high redshifts. Similar to Hα, [] traces the gas kinematics in a galaxy and is therefore an important component to quantify rotation- and dispersion-dominated systems as well as outflows (Jones et al. 2017; Pavesi et al. 2018; Kohandel et al. 2019; Ginolfi et al. 2020).

The FIR landscape has dramatically changed with the completion of the ALMA Large Program to Investigate C⁺ at Early Times (ALPINE; #2017.1.00428.L). ALPINE is laying the ground work for the exploration of gas and dust properties in 118 main-sequence star-forming galaxies in the early growth phase at 4.4 < z < 5.9 and herewith started the first panchromatic survey of its kind at these redshifts (Figure 1).

In the following, we summarize the scope of the ALPINE survey, we refer to Le Fèvre et al. (2019) for a broader overview of the program. ALPINE is a 69 hr large ALMA program started in Cycle 5 in 2018 May and completed during Cycle 6 in 2019 February. In total, 118 galaxies have been observed in Band 7 (covering [] emission at and its nearby continuum) at a spatial resolution of <10 and with integration times ∼30 minutes on-source depending on their predicted [] flux. The galaxies originate from two fields, namely the Cosmic Evolution Survey field (COSMOS; 105 galaxies, Scoville et al. 2007b) and the Extended Chandra Deep Field South (ECDFS; 13 galaxies, Giacconi et al. 2002). Due to gaps in the transition through the atmosphere, the galaxies are split into two different redshift ranges spanning 4.40 < z < 4.65 and 5.05 < z < 5.90 with medians of and 5.5 and galaxy numbers of 67 and 51, respectively. All galaxies are spectroscopically confirmed by either Lyα emission or rest-UV absorption lines and are selected to be brighter than an absolute UV magnitude of M₁₅₀₀ = −20.2. This limit is roughly equivalent to an SFR cut at and corresponds roughly to a limiting luminosity in [] emission of (assuming the relation derived by De Looze et al. 2014). Assuming a 3.5σ detection limit, the [] detection rate is 64%, and continuum emission is detected in 19% of the galaxies (see Figure 2).

The main science goals enabled by ALPINE are diverse and cover many crucial research topics at high redshifts:

• 1.  
  Connecting [] to star formation at high redshifts,
• 2.  
  Coherent study of the total SFR density at z > 4 including the contribution of dust-obscured star formation,
• 3.  
  Study of gas dynamics and merger statistics from [] kinematics and quantification of UV-faint companion galaxies,
• 4.  
  Study of gas fractions and dust properties at z > 4,
• 5.  
  The first characterization of ISM properties using and []/FIR continuum diagnostics for a large sample at z > 4,
• 6.  
  Quantifying outflows and feedback processes in z > 4 galaxies from [] line profiles.

Note that ALPINE provides at the same time the equivalent of a blind-survey of approximately 25 square-arcminutes. This enables us to estimate the obscured fraction of star formation (mostly below z = 4) by finding UV-faint galaxies with FIR continuum or [] emission. The serendipitous continuum sources and [] detections are discussed in detail in Bethermin et al. (2020) and F. Loiacono et al. (2020, in preparation). A more detailed description of these science goals can be found in our survey overview paper (Le Fèvre et al. 2019).

ALPINE is based on a rich set of ancillary data, which makes it the first panchromatic survey at these high redshifts including imaging and spectroscopic observations at FIR wavelengths (see Figure 1). The backbone for a successful selection of galaxies are rest-frame UV spectroscopic observations from the Keck telescope in Hawaii as well as the European Very Large Telescope (VLT) in Chile. These are complemented by ground-based imaging observations from rest-frame UV to optical, HST observations in the rest-frame UV, and Spitzer coverage above the Balmer break at rest-frame . The latter is crucial for the robust measurement of stellar masses at these redshifts (e.g., Faisst et al. 2016a).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For a survey overview of ALPINE see Le Fèvre et al. (2019) and for details on the data analysis see Bethermin et al. (2020). In this paper, we present these valuable ancillary data products and detail several basic measurements for the ALPINE galaxies. The outline of the paper is sketched in Figure 1. Specifically, in Section 2, we present the spectroscopic data and detail the spectroscopic selection of the ALPINE galaxies. In the same section, we also present stacked spectra and touch on velocity offsets between Lyα, [], and absorption line redshifts. Section 3 is devoted to the photometric data products, which include ground- and space-based photometry. In Section 4.1, we detail the derivation of several galaxy properties from the observed photometry. These include stellar masses, SFRs, UV luminosities, UV continuum slopes, as well as Hα emission derived from Spitzer colors. We conclude and summarize in Section 5. All presented data products are available in the online printed version of this paper.²⁸ The different catalogs and their columns are described in detail in the Appendix A. HST cutouts and rest-frame UV spectra for each of the ALPINE galaxies are shown in Appendix B.

Throughout the paper, we assume the ΛCDM cosmology with , Ω_Λ = 0.70, and Ω_m = 0.30. All magnitudes are given in the AB system (Oke 1974) and stellar masses and SFRs are normalized to a Chabrier (2003) initial mass function (IMF).

The ALPINE survey is only possible due to a spectroscopic pre-selection of galaxies from large spectroscopic surveys on COSMOS and ECDFS. This is because the ALMA frequency bands are narrow (), and in order to observe [] emission, the redshift has to be known within a precision of . The galaxy selection is refined to optimize the efficiency of the ALMA observations by creating groups of galaxies in spectral dimensions. Our sample also includes seven galaxies that were previously observed with ALMA by Riechers et al. (2014) and Capak et al. (2015). These are HZ1, HZ2, HZ3, HZ4, HZ5, HZ6/LBG-1, and HZ8, which correspond to the ALPINE galaxies DC_536534, DC_417567, DC_683613, DC_494057, DC_845652, DC_848185, and DC_873321, respectively. Furthermore, four galaxies from the VUDS survey (vc_5101288969, vc_5100822662, and vc_510786441 in COSMOS and ve_530029038 in ECDFS) are observed twice (resulting in a total number of 122 observations). The duplicate observations are used for quality assessment. Bethermin et al. (2020) describes the combination of these observations.

The rest-frame UV spectroscopic data from which the ALPINE sample is selected combine various large surveys on the COSMOS and ECDFS fields. Out of the 105 ALPINE galaxies on the COSMOS field, 84 are obtained by the large DEIMOS spectroscopic survey (Capak et al. 2004; Mallery et al. 2012; Hasinger et al. 2018) at the Keck telescope in Hawaii. The remaining spectra on the COSMOS field are obtained from the VIMOS Ultra Deep Survey (VUDS; Le Fèvre et al. 2015; Tasca et al. 2017) at the VLT in Chile. In total, six of the VUDS spectra are independently also observed as part of the Keck/DEIMOS survey (vc_5100559223, vc_5100822662, vc_5101218326, vc_5101244930, vc_5101288969, vc_510786441). The redshifts are consistent within , and we do not find any systematic offsets between the two observations (see also Section 2.4.1). Out of the 13 galaxies in the ECDFS field, 11 are obtained from spectroscopic observations with VIMOS (9) and FORS2 (2²⁹ ) at the VLT (Vanzella et al. 2007, 2008; Balestra et al. 2010), and two come from the HST grism survey GRAPES (Malhotra et al. 2005; Rhoads et al. 2009). The spectral resolution of the different data set varies between R  ∼  100 (ECDFS/GRAPES grism), R ∼ 180 (ECDFS/VIMOS), R ∼ 230 (COSMOS/VUDS), R ∼ 660 (ECDFS/FORS2), and R ∼ 2500 (COSMOS/DEIMOS).

Biases toward dust-poor star-forming galaxies with strong rest-frame UV emission lines (such as Lyα) can be common in purely spectroscopically selected samples. To minimized such biases as much as possible, the spectroscopically observed galaxies have been pre-selected through a variety of different selection methods. The largest fraction of galaxies in ALPINE is drawn from the Keck/DEIMOS and VUDS surveys on the COSMOS field. Both surveys include galaxies pre-selected in various ways, resulting in the most representative and inclusive spectroscopic high-redshift galaxy sample. Specifically, the VUDS survey combines predominantly a photometric redshift selection with a color-selected Lyman Break Galaxy (LBG) selection (Le Fèvre et al. 2015), known as the Lyman-break drop-out technique (see, e.g., Steidel et al. 1996; Dickinson 1998). The Keck/DEIMOS survey (providing 71% of the total ALPINE sample) consists of galaxies that are selected by narrowband surveys at z ∼ 4.5 (7%) and z ∼ 5.7 (27%), the drop-out technique (color selection) over the whole redshift range (49%), as well as purely by photometric redshifts (11%). In addition, four galaxies are selected by a 4.5 μm excess and one galaxy was pre-selected through X-ray emission using the Chandra Observatory. On the ECDFS field, the galaxies are mostly color-selected. Table 1 summarizes the different selections and corresponding numbers of galaxies and provides a complete list of references. We also list the numbers of galaxies with Lyα emission (76%) and weak Lyα emission or Lyα absorption (∼24%). Note that the Keck/DEIMOS and VUDS samples have similar Lyα emission properties. However, note that above z = 5, the ALPINE sample is strongly dominated by narrowband-selected galaxies.

Table 1.  Spectroscopy and Selection of ALPINE Galaxies

Survey	Selection	Number	Ref.
COSMOS field (105 galaxies)
Keck/DEIMOSa		84	1
	narrowband (z ∼ 4.5)c	6
	narrowband (z ∼ 5.7)d	23
	LBG (color)e	41
	pure photo-zf	9
	excess	4
	X-ray (Chandra)	1
	with Lyα emission	66
	weak Lyα emission or absorption	18
VUDS		21	2
	photo-z + LBG	21
	[narrowband (z ∼ 4.5)	3]b
	[narrowband (z ∼ 5.7)	1]b
	[LBG (color)	1]b
	excess	1]b
	with Lyα emission	16
	weak Lyα emission or absorption	5
ECDFS field (13 galaxies)
VLT GOODS-S		11	3
	primarily LBG (color)	11
	total with Lyα emission	6
	total without Lyα emission	5
HST/GRAPES		2	4
	Grism (no a priori selection)	2
	with Lyα emission	2
	weak Lyα emission or absorption	0

Notes.

^aFor a detailed description of the selection criteria, we refer the reader to Mallery et al. (2012) and Hasinger et al. (2018). ^bSix of these galaxies are also observed as part of the Keck/DEIMOS survey (ref. 1). The corresponding number per selection from the Keck/DEIMOS program is given in square-brackets for those six galaxies. ^cLyα emitters selected with NB711. ^dLyα emitters selected with NB814. ^eColor-selected galaxies in B, g⁺, V, r⁺, and z⁺⁺ using the criteria from Ouchi et al. (2004), Capak et al. (2004, 2011), Iwata et al. (2003), Hildebrandt et al. (2009). ^fGalaxies with a photometric redshift z > 4 with a probability of > 50% based on the Ilbert et al. (2010) photo-z catalog.

References: (1) Capak et al. (2004), Mallery et al. (2012), Hasinger et al. (2018), (2) Le Fèvre et al. (2015), (3) Vanzella et al. (2007, 2008), Balestra et al. (2010), (4) Malhotra et al. (2005), Rhoads et al. (2009).

Download table as:  ASCIITypeset image

Figure 3 shows the distribution of redshifts of the ALPINE galaxies in the COSMOS and ECDFS fields. The colored histogram bars show stacked numbers of galaxies that are pre-selected by the different methods discussed above. The bins with galaxies in the ECDFS field are hatched. The narrowband-selected galaxies are prominent at z ∼ 5.7 and represent the largest fraction of galaxies at z > 5 in ALPINE. On the other hand, the z < 5 sample consists mostly of color-selected galaxies. The VUDS galaxies are most represented at z < 5, while the DEIMOS spectra and the galaxies in ECDFS cover the whole redshift range.

Zoom In Zoom Out Reset image size

Figure 4 shows the distribution of observed magnitudes as well as rest-frame 1500 Å and luminosity of galaxies selected by the different methods. The photometry that is used is explained in detail in Section 3. The 1500 Å rest-frame luminosity is derived from spectral energy distribution (SED) fitting (see Section 4.2 for details). The 4000 Å rest-frame luminosity is derived directly from the UltraVISTA K_s and VLT K_s^v magnitude for galaxies in the COSMOS and GOODS-S field, respectively. The magnitudes and luminosities are not corrected for dust attenuation. Note that the K-band is rest-frame 3000 Å at the highest redshifts (z = 5.9); hence at these redshifts, older and dustier galaxies would be biased to lower luminosities. As expected for spectroscopically selected galaxies, the ALPINE sample covers the brighter part of the galaxy magnitude and luminosity distribution. The different selection methods on their own are distributed differently in this parameter space. Most noticeably, the z ∼ 5.7 narrowband-selected galaxies reside at the faintest luminosities, while the 4.5 μm continuum excess selected galaxies are among the brightest. The Chandra X-Ray Observatory–detected galaxy DC_845652 (green star) at z = 5.3 outshines all of the galaxies in UV luminosity.

Zoom In Zoom Out Reset image size

All in all, although naturally biased to the brightest galaxies, this diverse selection function makes ALPINE an exemplary panchromatic survey that enables the study of a representative high-z galaxy sample at UV, optical, and FIR wavelengths.

All of the rest-frame UV spectra discussed in Section 2.1 are relative-flux corrected to remove sensitivity variations across the spectrograph as well as to correct atmospheric absorption features. However, not all of the spectra have been absolute-flux calibrated, which is important to measure absolute quantities such as their Lyα emission. Hence, we recalibrate the spectra using the Galactic extinction-corrected total broad-, intermediate-, and narrowband photometry of the ALPINE galaxies (see Section 3 for details on the photometry). It turns out that the absolute-flux calibrated spectra are in excellent agreement (within better than 5% in flux) with our measured photometry, and the recalibration is not necessary in these cases. As the spectra come from different surveys, we convert them to a common format during the recalibration procedure.

To perform the absolute-flux calibration, we convolve each of the spectra with the transmission functions of the various optical broad-, intermediate-, and narrowband filters that exist on the COSMOS and ECDFS fields, respectively. On average, we use 4–9 filters for galaxies at z < 5 and 2–4 at z > 5. If the filter extends further than the spectrum, we extrapolate the spectrum by its medium continuum value. If the filter extends significantly beyond the spectrum (>50%), we do not consider the filter. We then compare the photometry obtained from the spectra to the total and Galactic extinction-corrected photometry discussed in Section 3, which allows us to obtain an average correction factor for each spectrum. We found that a single number for this correction per galaxy is enough for the calibration as the spectra already have been relative-flux calibrated. Since the uncalibrated spectra are mostly in units of counts, this correction is on the order of 10⁻²¹ for most galaxies. Our recalibration corrects for slit-losses and seeing variations. We also scale the variance in order to conserve the signal-to-noise ratio (S/N) of the spectrum. The final precision of our calibration is around 5%–10% in flux, which corresponds to the 1σ uncertainty in the photometry. Note that we do not consider undetected spectroscopic fluxes in this procedure; however, we use the constraints gained from the upper limits in the photometry for the calibration. Figure 5 shows three absolute-calibrated spectra at z ∼ 4.53, z ∼ 4.56 (with weak Lyα), and z ∼ 5.68 to visualize our method. The filters that were used for the calibration are indicated in colors.

Zoom In Zoom Out Reset image size

Figures 6 and 7 show stacks of different spectra. In order to create median-stacks of the spectra, we resample the spectra to a common wavelength grid and stack them in rest frame using their respective redshifts derived from [] or, if not available, from rest-UV absorption lines or Lyα emission. All stacks are subsequently binned to a resolution of 2 Å for visual purposes to emphasize the UV absorption features. To obtain a per-pixel uncertainty from the sky background for each stack (visualize by the gray line), we simply combine the inverse variances in the individual spectra in quadrature. The latter are the original inverse variance that we adjusted to the new normalization described in Section 2.2.

Zoom In Zoom Out Reset image size

Panels 1a and 1b of Figure 6 compare the full stacked spectra of galaxies at z < 5 in COSMOS from observations with DEIMOS and as part of VUDS, as well as at z > 5 from observations in COSMOS from DEIMOS and in ECDFS from VIMOS and FORS2. In the former case, we adjust the resolution of the DEIMOS spectra (R ∼ 2500) to that of the VUDS observations (R ∼ 230) by applying a 1D Gaussian smoothing. The spectra are normalized to the median flux in the rest-frame wavelength range between 1300 and 1400 Å before stacking. For stacks of galaxies at z > 5, the rest-frame wavelength reaches up to 1500 Å, while for lower-redshift stacks, we show wavelengths up to rest frame 1700 Å. Several prominent spectral features are visible in the stacks in both redshift bins (indicated by gray bars). These include the Lyα emission line and  at 1241 Å and in addition UV absorption lines such as  at 1260 Å, the -[]- complex at 1301 Å, the two  lines at 1398 Å, as well as , , and  at 1527 Å, 1548 Å, and 1640 Å, respectively. Furthermore, we see indication of  absorption at 1608 Å in the COSMOS/DEIMOS spectra stack at z < 5. The depth of the UV absorption features are comparable for the different observations with the different instruments, verifying similar quality and little biases. However, note that the features in the ECDFS spectra are less pronounced due to the factor ∼6 smaller number of spectra contributing to the stacks compared to the DEIMOS stacks. Panels 2a through 2e show the stacks for variously selected data sets below and above z = 5. The spectra are not normalized before stacking in these cases to provide a comparison of the absolute-flux values for the different redshifts and samples to the reader. The number of spectra per wavelength are shown on the top right for each panel. Note again that the number of high-redshift spectra drops toward redder wavelengths. This has to be kept in mind when analyzing the spectral features in the stacks. Emission and absorption lines are indicated as in the other panels. As expected, the stacked spectra at higher redshifts are fainter, but still significant UV absorption features are present (see also Faisst et al. 2016b; Khusanova et al. 2019; Pahl et al. 2020).

Figure 7 shows stacked spectra in COSMOS observed with DEIMOS for the different selection categories (see also Table 1 and Figure 3). We split the LBG category in galaxies below and above z = 5. All of the spectra are smoothed with a Savitzky–Golay filter with size of 2 Å for visualization purposes. The total number of spectra per stack is indicated in the upper left corner. All panels are scaled the same way to emphasize differences in brightness. The X-ray detected galaxy at z = 5.3 is UV bright compared to the other stacks and shows strong  emission with overlaid  absorption as well as broad  emission. LBGs (i.e., color-selected galaxies) are preferentially fainter but of similar continuum brightness as narrowband-selected galaxies at z ∼ 4.5. The latter show significant , , and  absorption. As expected, narrowband-selected galaxies at z ∼ 5.7 show strong Lyα emission and a faint continuum such that the S/N is too low to detect UV absorption features at great significance. The stack of galaxies selected by photometric redshifts shows to first-order similar properties as the LBGs. The 4.5 μm-excess continuum selected galaxies are, on average, the continuum brightest galaxies and show significant Lyα emission as well as absorption features.

Zoom In Zoom Out Reset image size

2.4.1. Measurements

2.4.2. Velocity Offsets with Respect to [C ii] FIR Redshifts

The detection of [] by ALMA provides the systemic redshift of a galaxy. This enables us to study velocity offsets of rest-frame UV absorption lines and Lyα emission that will inform further about the properties of the ISM in these galaxies similarly to studies at lower redshifts using Hα and ] (e.g., Steidel et al. 2010; Marchi et al. 2019). Here, we give an overview of the velocity properties and compare them for galaxies with high and low specific SFRs.

In the following, we define the velocity difference for two redshifts (z₁ and z₂) as where . The measurement of the [] redshifts are detailed in Bethermin et al. (2020). They are defined as the peak of a Gaussian fit to the [] line with spectral resolution of . The uncertainty of the redshift measurements was estimated by a Monte Carlo simulations with perturbed fluxes according to the error per spectral bin. The average uncertainty is roughly . For the absorption lines, we require that is measured from at least three absorption lines, and we only show galaxies that have not been flagged by our visual inspection with SpecPro as unreliable (flag −99). The average intrinsic measurement error per galaxy is . In relation to that, a systematic uncertainty of in the rest-frame wavelength of the absorption lines (e.g,. due to calibration issues) turns into a velocity shift of .

Figure 8 shows stacked histograms of velocity differences. The number of galaxies used as well as the median of the distribution with scatter (not error on the median) are indicated as well. The upper panel compares the velocities measured from Lyα and the IS+wind absorption lines. We find a median offset on the order of , which is consistent with other measurements at the same redshifts (see, e.g., Faisst et al. 2016b; Pahl et al. 2020) as well as at z ∼ 2−3 (Steidel et al. 2010). The center and bottom panels compare the IS+wind and Lyα redshifts to the systemic redshift (here defined as the [] redshift, Bethermin et al. 2020). For the former, we find an offset of , and for the latter, we find . These negative and positive velocity offsets can be related in a simple physical model involving the resonant scattering of Lyα photons and outflowing gas in the outskirts of galaxies (see detailed discussion in Steidel et al. 2010). The redshifted Lyα emission line (with respect to systemic) can be explained by resonant scattering of the Lyα photons. Preferentially, redshifted Lyα photons scattered from the back of the galaxy can make it unscattered through the intervening gas inside the galaxy along the line of sight. The blueshift of IS absorption may depend on the outflow velocity of the absorbing gas as well as its covering fraction (or optical depth) inside the galaxy along the line of sight toward the observer. For a more in-depth discussion, we refer the reader to a companion paper by Cassata et al. (2020). Overall, we do not see a significant dependence of the velocity differences on the various selection techniques (color-coded in the figure).

Zoom In Zoom Out Reset image size

Figure 9 compares the velocity offsets between IS (, [], , ) and wind (, ) lines. We require that at least three IS lines and one wind line is measured. In addition, only galaxies that pass our visual classification (i.e., have flags other than −99, see above) are used. Overall, we do not see any statistical difference between IS and wind lines, although there is a tail toward higher blueshifts in the case of wind lines. However, wind and outflows may be increased in galaxies with high and spatially dense star formation and young stellar populations. Therefore, we would expect different velocity shifts for the absorption lines with respect to the systemic redshift for highly star-forming galaxies. In Figure 10, we investigate this picture by stacking galaxies at the extreme ends of the sSFR distribution (we refer to Section 4 for details on the measurement of the physical properties of our galaxies), namely low () and high () sSFR, in their corresponding rest-frames defined by the systemic redshift (i.e., [] redshift). The sSFR is a good proxy of the star formation density in a galaxy as well as the age of the current stellar population (see, e.g., Cowie et al. 2011). We show the stacked spectra in five wavelength regions covering prominent absorption lines for each sSFR bin. The vertical dashed lines show the different absorption lines in the [] rest frame. First, we verify that the shifts between IS and wind lines are very similar for each sSFR bin (in concordance with Figure 8). However, it is intriguing that in the low sSFR stack, all absorption lines agree well with the [] redshift, while in the high sSFR stack, the lines are significantly blueshifted by . We also note that in the high sSFR stack, the  line shows a noticeable P-Cygni profile indicative of strong winds and outflows (Castor & Lamers 1979). These findings fit well into a picture of strong winds and outflows produced by the high star formation in these galaxies, which is also in line with recent results obtained through the stacking of ALPINE [] spectra (Ginolfi et al. 2020).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.4.3. How Well Does [C ii] Trace Systemic Redshift? Comparison to Optical [O ii] Emission

The extended nature of [] may be indicative of its origin in the diffuse interstellar medium in addition to PDRs (Stacey et al. 1991; Gullberg et al. 2015; Vallini et al. 2015; Faisst et al. 2017). Moreover, recent work by Ginolfi et al. (2020) shows that [] emission is significantly affected by large-scale outflows caused by a high rate of star formation in these galaxies. However, as shown by the same study, the outflows seem to be symmetric, and therefore, we do not expect them to significantly change the centroid of the [] emission line.

During 2019 January 13–15, we were able to obtain a near-IR spectrum of one of our ALPINE galaxies (DC_881725 at ) using the Multi-Object Spectrometer For Infra-Red Exploration (MOSFIRE; McLean et al. 2010, 2012) at the 10 m Keck I telescope on Maunakea in Hawaii. The observations of a total on-source integration time of in K band () were carried out under clear weather conditions with an excellent average seeing FWHM of . We performed a standard data reduction using the MOSFIRE data reduction pipeline³⁴ (Version 2018). From the produced 2D spectrum and variance map, we extract the 1D spectrum at the spatial location of the galaxy using a weighted mean across ±3.5 spatial pixels ().

We are able to detect the optical [] doublet (3727.09 Å and 3729.88 Å) at the spatial position of the galaxy at a level of >5σ. Note that this is the first detection of optical [] in a galaxy with [] measurement from ALMA, which allows us, for the first time, to compare these two lines at these redshifts. In the bottom panel of Figure 11, we show the final spectrum in the rest frame of the [] emission. The width of ∼4 Å includes both [] lines. The theoretical rest-frame wavelength of the doublet is indicated by the black arrows. The position of the line agrees perfectly with the [] redshift derived from ALMA, indicating that FIR [] and optical [] trace the same systemic redshift. In addition, the top panel of the Figure shows the [] FIR line for comparison of the central wavelength of the lines. Note that the actual width of the [] line is more than 100 times larger than the the one of the optical [] line.

Zoom In Zoom Out Reset image size

The photometry for the galaxies on the ECDFS field is taken directly from the 3D-HST catalog, which provides ground-based observations as well as a wealth of data from HST imaging. The photometry (total fluxes and magnitudes) is corrected for Galactic extinction, PSF size as well as other biases; therefore, no further corrections are applied. The ALPINE galaxies are matched visually to the spatially closest 3D-HST counterpart using the HST WFC3/IR F160W image as reference. The spectroscopic redshifts match the photometric redshifts within their uncertainty (∼0.1–0.2), ensuring that we identified the correct counterpart.

The ground-based photometry available in ECDFS (including references) is listed in Table 2. Summarizing, this includes the U38, b, v, R_c, and I broadband filters from the Wide Field Imager on the 2.2 meter MPG/ESO telescope, the U and R bands from VIMOS on the VLT, the near-IR filters J^v, H^v, and from ISAAC on the VLT, J_w and data taken by WIRCam on the CFHT, as well as 14 intermediate-band filter from the Suprime-Cam on the Subaru telescope. For galaxies at z = 4.5 and 5.5, the Lyman-break falls roughly in the v and the R_c-band, and therefore, the galaxies are expected to be only faintly (or not at all) visible in these and blueward filters. On the other hand, the galaxies are bright at observed near-IR wavelengths, i.e., filters redward of the z band (corresponding to roughly the F850LP filter). Figure 12 shows the stacked F850LP (z), J^v (J), and (K) magnitude distributions of the ECDFS ALPINE galaxies split in z < 5 (hatched blue) and z > 5 (hatched red). As expected, the latter sample occupies slightly fainter magnitudes.

Zoom In Zoom Out Reset image size

The space-based photometry includes the four Spitzer bands at 3.6, , , and . In addition, the public 3D-HST catalog includes a wealth of HST photometry. Specifically, it contains measurements in the ACS bands F435W, F606W, F775W, F814W, and F850LP as well as in the WFC3/IR bands F125W, and F160W bands for all 13 ALPINE galaxies. Only 10 galaxies have measurements in the WFC3/IR band F140W. The HST photometry is measured on PSF-matched images. As described in Skelton et al. (2014), the Spitzer and ground-based photometry are measured using the MOPHONGO (Labbé et al. 2006; Wuyts et al. 2007; Whitaker et al. 2011), which uses a high-resolution image (here the HST imaging) as spatial prior to estimate the contributions from neighboring blended sources in the lower resolution image. The different depths of these observations as well as references are listed in Table 2. A query of the Barbara A. Mikulski Archive for Space Telescopes (MAST³⁷ ) using the mastquery Python package³⁸ shows that in addition to the HST measurements contained in the 3D-HST catalog, four, ten, and two galaxies have coverage in the WFC3/IR bands F098M, F105W, and F110W, respectively. None of the galaxies has ACS F475W coverage. These additional data that are not published in the 3D-HST catalog come from various other observation programs in and around the ECDFS field. We subsequently measure this additional photometry for all ALPINE galaxies in ECDFS using SExtractor (version 2.19.5, Bertin & Arnouts 1996) in different aperture sizes ( and 3″) as well as auto magnitudes. For this, we first create a mosaic of all of the HST pointings that overlap with the ALPINE galaxies using the AWS-drizzler³⁹ tool that is part of the grizli⁴⁰ Python package (G. Brammer 2020, in preparation.). We use a 006 pixel scale and all HST images are registered to Gaia (see Section 3.3). SExtractor is run with relative THRESH_TYPE, and we set DETECT_MINAREA, DETECT_THRESH, DEBLEND_MINCONT, and DEBLEND_NTHRESH to 3, 1.5, 1.5, 0.001, and 64, respectively. If no object is detected above the threshold (1.5σ) within 07 (roughly the ground-based seeing) of the original ALPINE coordinates, we consider the galaxy as undetected in a given band and replace its flux by a 1σ limit that is computed from the rms noise at the position of the galaxy. The photometry measured by SExtractor is subsequently corrected for galactic foreground extinction, which we assume to be constant for all galaxies in ECDSF at . Figure 13 summarizes the HST data available for the galaxies in the ECDFS field. The blue squares show the layout of all of the HST pointings as of 2019 October, with darker shades of blue indicating more observations. The ALPINE galaxies at z < 5 and z > 5 are indicated with orange circles and squares, respectively.

Zoom In Zoom Out Reset image size

Most of the ALPINE galaxies (105 out of 118) reside in the COSMOS field, and we match them to the latest photometric measurements presented in the COSMOS2015 catalog. The matching is again done on a galaxy-by-galaxy basis using the HST/ACS F814W as well as UltraVISTA K_s images as references. We also match against the photometric redshifts given in the catalog in order to identify the correct counterpart.⁴¹

In Table 3, we list all of the photometric measurements available for the ALPINE galaxies on the COSMOS field. Summarizing, these include u*-band observations from MegaCam on CFHT, the B, V, r⁺, i⁺, z⁺⁺ as well as 12 intermediate-band and two narrowband filters from the Suprime-Cam on Subaru, the Y_HSC-band from the Hyper Suprime-Cam on Subaru as well as near-IR bands H_w and from WIRCam on CFHT and Y, J, H, and K_s from VIRCAM on the VISTA telescope. In addition, the galaxies are covered by the four Spitzer channels from 3.6 μm to from the SPLASH survey⁴² (Capak et al. 2012; Steinhardt et al. 2014; Laigle et al. 2016). As described in (Laigle et al. 2016), the Spitzer photometry is measured using IRACLEAN (Hsieh et al. 2012), which uses positional priors from higher-resolution imaging (in this case the zYJKHK_s detection χ²-image) to deblend the photometry.

Table 3.  Photometry Available on the COSMOS Field

Observatory/Instrument	Filter	Central λ	3σ depth	Ref.
		(Å)	(mag)a
Ground-based
CFHT/MegaCam	u*	3823.3	26.6	1
Subaru/Suprime-Cam	B	4458.3	27.0	1, 2
	V	5477.8	26.2	1,2
	r+	6288.7	26.5	1, 2
	i+	7683.9	26.2	1, 2
	z++	9105.7	25.9	1, 2
	IA427	4263.4	25.9	1, 2
	IA464	4635.1	25.9	1, 2
	IA484	4849.2	25.9	1, 2
	IA505	5062.5	25.7	1, 2
	IA527	5261.1	26.1	1, 2
	IA574	5764.8	25.5	1, 2
	IA624	6233.1	25.9	1, 2
	IA679	6781.1	25.4	1, 2
	IA709	7073.6	25.7	1, 2
	IA738	7361.6	25.6	1, 2
	IA767	7684.9	25.3	1, 2
	IA827	8244.5	25.2	1, 2
	NB711	7119.9	25.1	1, 2
	NB816	8149.4	25.2	1, 2
Subaru/HSC	YHSC	9791.4	24.4	1
CFHT/WIRCam	Hw	16311.4	23.5	1
		21590.4	23.4	1
VISTA/VIRCAM	Y	10214.2	24.8 (25.3)b	1
	J	12534.6	24.7 (24.9)b	1
	H	16453.4	24.3 (24.6)b	1
	Ks	21539.9	24.0 (24.7)b	1
Space-based
HST/ACS	F435W	4328.7	−c	−
	F475W	4792.3	−c	−
	F606W	5924.8	−c	−
	F814W	8058.2	29.2	3
	F850LP	9181.2	−c	−
HST/WFC3	F098M	9877.4	−c	−
	F105W	10584.9	−c	−
	F110W	11623.8	−b	−
	F125Wd	12516.3	27.6c,d	4
	F140W	13969.4	−c	−
	F160Wd	15391.1	27.5(25.0)c,d,e	4, 5
Spitzer/IRAC	ch1	35634.3	25.5	1, 6, 7
	ch2	45110.1	25.5	1, 6, 7
	ch3	57593.4	23.0	1, 7
	ch4	79594.9	22.9	1, 7

Notes.

^aDepth is measured in aperture for ground-based photometry. ^bDepths of the ultra-deep area are given in parenthesis. ^cAncillary pointings; therefore, depth varies depending on the specific observations. ^dMost of these data come from the CANDELS survey and their depth is indicated. ^eSome of the F160W data comes from the DASH survey (depth given in parenthesis).

References: (1) see Laigle et al. (2016), (2) Taniguchi et al. (2007, 2015) (3) Scoville et al. (2007b), Koekemoer et al. (2007), (4) Grogin et al. (2011), Koekemoer et al. (2011), (5) Momcheva et al. (2017), (6) Capak et al. (2012), Steinhardt et al. (2014), (7) Sanders et al. (2007).

Download table as:  ASCIITypeset image

Contrary to the 3D-HST photometry catalog, the fluxes and magnitudes in the COSMOS2015 catalog are not total and not corrected for systematic biases and Galactic extinction. To perform these corrections, we follow the steps outlined in the appendix of Laigle et al. (2016). Specifically, we use the 3″ diameter aperture magnitudes (), which we correct for photometric (o_i, see their Equation (4)) and systematic offsets (s_f, see their Table 3) by applying

where i is the object identifier and f denotes the different filters. The total magnitudes are subsequently corrected for Galactic extinction by applying

where is the Galactic extinction from the Schlegel et al. (1998) maps on COSMOS for each object as given in the catalog and F_f are the extinction factors per filter given in Table 3 of Laigle et al. (2016).

Figure 12 shows the stacked magnitude distribution in (z), J, and K_s (K) bands for the z < 5 (blue) and z > 5 (red) sub-samples. As expected, the high-redshift galaxies are fainter in all bands. In addition, we find the magnitude distributions between the galaxies in ECDFS and COSMOS to be similar. This indicates no major discrepancies in photometric (and, hence, physical) properties between the two samples.

In terms of HST imaging, all galaxies except one are observed in ACS F814W (Koekemoer et al. 2007; Scoville et al. 2007a). In addition to this, a MAST search shows that several galaxies are covered by other observing programs in the ACS bands F435W (3), F475W (5), F606W (21), and F850LP (5) as well as in the WFC3/IR bands, F105W (11), F110W (5), F125W (16), F140W (13), and F160W (53).⁴³ Note that the observations in F160W primarily come from the CANDELS survey (covering the central part of COSMOS; Grogin et al. 2011; Koekemoer et al. 2011) as well as the “drift and shift” (DASH; Momcheva et al. 2017) survey. While the CANDELS imaging is deep (>27.5 magnitudes at 3σ), the data from the DASH survey is much shallower (25.0 magnitudes at 3σ), and therefore, only half of the galaxies are detected. Furthermore, the spatial sampling in the latter does not allow for a detailed study of the structure of the galaxies. Figure 14 summarizes the HST pointings on the COSMOS field (blue). The CANDELS area in the center of the COSMOS field as well as the three DASH stripes are evident. The locations of the ALPINE galaxies are indicated with circles (z < 5) and squares (z > 5). The photometry available is summarized in Table 3 including depths (where applicable) and references.

Zoom In Zoom Out Reset image size

In order to measure the photometry in these HST bands, we use SExtractor in the same way as described in Section 3.1. We first create a mosaic (006 pixel sizes and registered to Gaia) using the AWS-drizzler. We subsequently measure the HST photometry of on each of the images for all ALPINE galaxies with coverage using SExtractor in apertures (07 and 3″) as well as auto magnitudes. If no object is detected above the set threshold level within 07 of the original ALPINE coordinates, we consider the galaxy as undetected in a given band. Similar to Equation (2), we correct the HST photometry for galactic foreground extinction using the Schlegel et al. (1998) extinction map. We compare these measurements to the ground-based photometry, by first performing a PSF matching by smoothing the original HST images with a Gaussian kernel with FWHM of 07 and measure the photometry in a 3″ aperture (as used in the COSMOS2015 catalog). We also compare the ground-based (i⁺, Y, J, H) with the HST (F814W, F105W, F125W, F160W) photometry in approximately matching filter bands. We find an overall agreement on average of 0.2 magnitudes (dark gray region) with an expected increase of scatter at fainter magnitudes. For the four brightest sources in H-band (<23.4 AB), the ground-based measured photometry is systematically up to 0.5 magnitudes brighter in three out of four cases. Due to the low number of galaxies, it is difficult to investigate this statistically. A more detailed measurement of the HST photometry including deblending of specific sources such as merging or clumpy galaxies, will be provided in a forthcoming paper.

Astrometric accuracy is crucial for many cases. First, accurate spatial offsets of light emission from rest-frame UV, optical, and sub-mm wavelengths reveal the properties of the interstellar medium such as the location and interplay of stars, dust, and gas. Second, the robustness of the identification of sub-mm counterparts needs a high astrometric accuracy of all involved data sets. As described in Sections 3.1 and 3.2, the HST images produced for the ALPINE galaxies are all aligned to Gaia. Our tests show that the average offsets are less than 15 mas in both R.A. and decl. with a scatter of no more than 30 mas (G. Brammer 2020, private communication). Unfortunately, the positional accuracy for current catalogs (such as 3D-HST or COSMOS2015 that are used here) are lower, and we expect significant offsets between those astrometric solution and ALMA data products. In the following, we characterize these offsets.

According to the technical handbook, ALMA observations are currently registered to the International Celestial Reference Frame (ICRF) to an accuracy better than ∼5 mas. The ICRF is based on hundreds of extragalactic radio sources such as quasars distributed over the whole sky. The positional accuracy Δp of single observations can be estimated by

where ν is the observed frequency in GHz, B is the maximum baseline length in kilometers, and σ is the S/N at the peak of emission. For ALPINE (ν = 330 GHz and B = 0.2 km for C43-1), this leads to . The calibrators are detected well above 50σ, which leads to an absolute positional accuracy of ∼20 mas or better.

To check the astrometric alignment of the photometric catalogs used here, we make use of the Gaia DR2 catalog (Mignard et al. 2018), which provides currently the most accurate absolute astrometry. As shown in Mignard & Klioner (2018), there are no significant offsets between this reference frame and the ICRF frame used by ALMA; hence, this test directly reveals potential differences in astrometry between the 3D-HST and COSMOS2015 catalogs and our ALMA observations. Using the proper motion information of the stars from Gaia, we project their positions back in time to the year of calibration of the data products by using the equations

where α and δ denote the R.A. and decl. (and and their proper motion), t_gaia is the Gaia reference frame in years (here 2015.5), and t_ref is the reference frame of the calibration of the catalogs. To increase accuracy, we only include stars with a proper motion in both coordinates of less than in the following. Note that no parallax motion is included in the above formulae, which would result in less than astrometric shifts.

The COSMOS2015 catalog is calibrated to the MegaCam i-band data that was taken in 2004 (Laigle et al. 2016). For ECDFS, the catalog (3D-HST) is calibrated to the same reference system as the CANDELS HST images, namely ground-based R-band data taken in 2001 (Koekemoer et al. 2011; Skelton et al. 2014). However, the exact year is not important as we have selected stars with a relatively slow proper motion. In order to select stars on the COSMOS field, we use the ACS/F814W SExtractor catalog (Scoville et al. 2007a) and select sources with a magnitude brighter than 23 AB and SExtractor star/galaxy classifier value of >0.8. These stars are then matched to the COSMOS2015 catalog to obtain their position in that catalog. For ECDFS, we extract stars directly from the 3D-HST catalog by selecting sources brighter than 23 AB in F160W with star_flag = 1 and a SExtractor star/galaxy classifier value larger than 0.8. Other selections (e.g., different magnitude cuts) do not affect the following results. The Gaia stars are then matched to the star catalogs in COSMOS and ECDFS to obtain the astrometric offsets. In total, 47 and 2724 Gaia stars are used in ECDFS and COSMOS, respectively.

Figure 15 shows scatter plots and histograms comparing the position of the Gaia stars to the positions in the catalogs. We find significant systematic offsets in the astrometry in ECDFS of 98 mas in R.A. and −256 mas in decl.. These offsets are consistent with what was found in earlier studies (Dunlop et al. 2017; Franco et al. 2018; Whitaker et al. 2019). For COSMOS, we only find a significant offset in R.A. of −64 mas. In addition to that, there is a significant scatter in the astrometry on the order of 100 mas in both coordinates in both fields.

Zoom In Zoom Out Reset image size

To compute the astrometric offset of individual ALPINE galaxies, we make use of the fact that the HST images are already aligned to the Gaia reference frame (see details in Sections 3.1 and 3.2). Specifically, we compute the offsets between the coordinates measured on the ACS/F814W images and the original coordinates from the 3D-HST or COSMOS2015 catalog. If no ACS F814W image is available or if the galaxy is not detected (which happens for redshifts z > 5), we use the deepest image red of F814W. If none are available (in four cases), we report the average offset as shown in Figure 15.

Note that the coordinates given in the final ancillary data catalog are not corrected for these offsets. However, we give the offsets for each galaxy in the columns delta_RA and delta_DEC, which can be added to the original coordinates to obtain Gaia-corrected R.A. and declination (see Appendix A).

4.1.1. Fitting Method

4.1.2. Systematic Uncertainties in Physical Properties from Modeling Assumptions

Depending on their exact methods, different SED-fitting codes may measure different physical properties even with the same photometry provided as input. In addition to more physical reasons (such as different assumptions on the stellar population models and implementation of emission lines), the different treatment of undetected fluxes (or fitting in magnitude space), varying minimization techniques and weightings, and different scaling of the error of the input photometry can contribute to these discrepancies.

To investigate the amplitude of such differences, we compare the measurements from LePhare to a modified version of the HyperZ code (Bolzonella et al. 2000) that includes the effects of nebular emission (Schaerer & de Barros 2009), to estimate such systematic uncertainties in the fitted parameters. To minimize the degeneracy with other assumptions, we run HyperZ on the exact same photometry and with the exact same model SEDs (i.e., same metallicity, age, SFH, and dust attenuation law) as described in Section 4.1.1. Figure 16 compares the stellar mass (left panel), SED-derived SFR (middle panel), and absolute UV magnitude (right panel, see also Section 4.2) derived by the two codes. The symbols are color-coded by difference in stellar population age (positive values indicate younger ages derived by HyperZ). We find that the stellar masses derived by HyperZ are systematically smaller by 0.3–0.4 dex. Similarly, the SFRs are systematically larger by ∼0.3 dex.

Zoom In Zoom Out Reset image size

As expected, the absolute UV magnitudes are largely in agreement, as they are, to the first order, independent of the physical parameters and just represent a translation of the fitted UV flux.

The differences in stellar mass and SFR are due to the effect of nebular emission, which results in younger ages and larger emission line corrections of the intrinsic rest-frame optical continuum (observed by the Spitzer broad bands) in the HyperZ models, and therefore directly affects the stellar mass measurements (see, e.g., de Barros et al. 2014). This is confirmed by the fact that the stellar masses and SFR measured by the two codes agree well (within a factor of two) if the emission lines are turned off. The effect of nebular emission found by HyperZ may be overestimated for the ALPINE galaxies, e.g., if the emission lines are more strongly attenuated than the continuum (see Section 4.4). For consistency and comparability with other studies on the COSMOS field, we choose the LePhare fitting results as the default.

Next to the systematic offset discussed above, we also find four galaxies (DC_472215, DC_503575, DC_722679, DC_790930) whose stellar mass measurements are significantly discrepant, by more than one order of magnitude, between the two codes. Specifically, they are fitted with an old, low-SFR, and massive galaxy template with HyperZ, while a young, high-SFR, and low-mass galaxy template is preferred by LePhare. Three of these outliers have significantly contaminated Spitzer photometry (indicated by the squares). As a consequence, the apparent Spitzer fluxes and stellar masses are overestimated and the optical colors are artificially reddened, which makes their stellar masses largely unreliable. Furthermore, we artificially increased the errors of the Spitzer photometry in our LePhare measurements (see Section 4.1.1), hence they have smaller weights, which might reduce the effect of photometric contamination on the fit.

In addition, we investigate the effect of different dust reddening laws on the LePhare measurements. For this, we compare a Calzetti et al. reddening with a Small Magellanic Cloud (SMC; Prevot et al. 1984) reddening. The latter might be more suitable for metal-poor low-mass galaxies. Running LePhare with the same settings but adopting an SMC reddening curve, we find only small changes in the stellar mass and SFR measurements. Specifically, for the former, we find an average offset toward lower stellar masses in the case of SMC dust of 0.05 dex and a maximal offset of 0.2 dex. For latter, we find similar offsets toward lower SFRs in the case of SMC dust.

Finally, we note that the galaxy properties derived here are consistent within a factor of two with the ones published in the COSMOS2015 catalog (based on photometric redshifts). We conclude this from comparing galaxies with the same spectroscopic and photometric redshift within 0.1.

4.1.3. The ALPINE Galaxies on the z = 5 Main-sequence and [C ii] Fluxes

Figure 17 summarizes the results of this section by showing the relation between stellar mass and SFR (the main sequence) of our ALPINE galaxies together with the [] emission (in ) measured by ALMA (in color). The measurements are compared to all galaxies (with photometric redshifts) at 4 < z < 6 in the COSMOS2015 catalog (blue points) as well as the main-sequence parameterization by Speagle et al. (2014) at z = 5 (gray band with ±0.3 dex width).

Zoom In Zoom Out Reset image size

The comparison to the photometrically selected COSMOS parent sample indicates that the ALPINE sample is a fair representation of the overall population of star-forming z > 4 galaxies. The sample also includes at higher stellar masses galaxies that lie 2–3σ below the main sequence. Note that two of these galaxies at  ∼ 10.3 and 10.7 have contaminated Spitzer photometry, and therefore, their stellar masses are upper limits. Other two galaxies at and 10.9 do not show [] emission, which is expected if they are systems of low SFR below the main sequence. The galaxy with [] detection at is classified as “extended dispersion dominated” by our [] morpho-kinematic classification (see Le Fèvre et al. 2019) and the HST/ACS imaging suggest a clumpy morphology. Although significantly below the main sequence, [] is still detected in that galaxy, perhaps indicative of dust-obscured star formation.

The color-coding of the points suggests a [] emission increase along the star-forming main sequence. Furthermore, the fraction of [] detected galaxies significantly drops below or an SFR of less than . This could be due to the effect of metallicity on the [] line strength by either indirectly a higher ionization state or directly through lower Carbon abundance (e.g., Narayanan et al. 2018). The relationship between star formation and [] emission will be studied in more detail in a forthcoming paper (Schaerer et al. 2020).

Note that there are galaxies with an SFR of less than in our sample, contrary to our initial selection. We emphasize that the initial selection was based on the observed absolute UV magnitude and not on any property derived from SED fitting (such as SFR). This discrepancy is therefore expected within the uncertainty of measuring SFRs from SED fitting.

The UV luminosities and absolute UV magnitudes at rest frame 1500 Å (not dust corrected) are measured during the SED-fitting process with LePhare and are defined by the transmission curve of the Galaxy Evolution Explorer (GALEX) FUV filter (∼1500 Å). As shown in Section 4.1.2, the absolute UV magnitudes measured by LePhare and HyperZ are in very good agreement.

We also compare these measurements to a more direct method by using the observed flux in a single filter that is closest to rest frame 1500 Å (Subaru z⁺⁺ filter at z < 5 and UltraVISTA Y or Y_HSC band at z > 5 for galaxies in COSMOS and the HST filter F850LP for all of the galaxies in ECDFS). The left panel in Figure 18 shows a very good agreement between the two methods. The scatter is mainly due to the low S/N of the single-filter measurements of the second method (indicated by the large error bars). The scatter is enhanced for galaxies at z > 5 also due to the fact that the UltraVISTA Y-band observations are less deep than the z⁺⁺ observations used for z < 5 galaxies. In the following, we will use the more robust absolute UV magnitude from LePhare as they depend less on the S/N of single observations.

Zoom In Zoom Out Reset image size

The middle panel in Figure 18 shows the distribution of absolute UV magnitudes for galaxies in the COSMOS and ECDFS fields in two redshift bins as in the previous figures. The bulk of galaxies are between M_UV of −22.7 and −20.2 (consistent with the faint absolute UV magnitude limit of the survey, see Section 1.2). One of the galaxies in ECDFS (CANDELS_GOODSS_37) is significantly fainter (M_UV = −19.2). This galaxy has been added to the sample to fill in an empty frequency window. The measurement of the absolute UV magnitude from single band and SED fit agree and the galaxy is compact and isolated (i.e., no contamination in the photometry). The fit to its photometry with LePhare suggests a dust-free, low-mass ( = 9.22) galaxy that is forming stars at a rate typical for the main sequence ().

The right panel of Figure 18 compares the distribution of the UV luminosity of the ALPINE galaxies (stacked filled histogram for low and high redshift) to the same parent sample selected from the COSMOS2015 catalog and used in Figure 17 split in two redshift bins (empty histogram). As expected, the absolute UV magnitude cut applied for the selection of the ALPINE sample (see Section 2.1) causes a bias toward the brighter end of the parent distribution.

4.3.1. Method

4.3.2. Systematic Uncertainties and Dependencies on Dust Attenuation

In addition to photometric uncertainties, several model assumptions affect the measurement of β. We found that the two most important ones are the wavelength range over which β is fit and the assumed dust attenuation law.

We choose the wavelength range over which β is measured to be consistent with the definition of several other studies (Calzetti et al. 1994; Meurer et al. 1999; Finkelstein et al. 2012; Bouwens et al. 2014; Barisic et al. 2017; Fudamoto et al. 2017). In the upper panel of Figure 19, we demonstrate how the β measured for our ALPINE galaxies would change if different wavelength ranges are used. First, we do not find any differences in our measurements compared to the definition by Calzetti et al. (1994), who use 10 discrete fitting windows between 1300 Å and 2600 Å to avoid strong absorption and emission lines (green diamonds). The other symbols show the comparison to different wavelength ranges, and we notice significant offsets from our measurements. For example, defining β between 1600 Å and 2200 Å results in up to Δβ = 0.1 bluer slopes (orange squares). Using a significantly redder wavelength range, 1600 Å to 2600 Å, leads to 0.1–0.35 bluer slopes (blue circles) compared to our definition. Note that the offset varies as a function of β itself—specifically, differences are enhanced toward redder slopes.

Zoom In Zoom Out Reset image size

The second, more physically driven quantity that affects the measurement of β is the assumed dust attenuation law. As described in Section 4.1.2, the choice of the dust attenuation law has a negligible affect on the stellar masses and SFRs. This is not the case for the β slopes as shown in the lower panels of Figure 19. The left panel compares β derived using Calzetti and SMC dust attenuation. We notice a consistent positive offset of up to Δβ = 0.3 for the reddest slopes. We compared the reduced χ² values output by LePhare for fits using a Calzetti and SMC dust attenuation in order to derive a preference for either of the dust attenuations. We find that the χ² values show insignificant differences, which lets us conclude that we are not able to distinguish between the different dust attenuations based on our SED fitting. Hence, we decided that the best way is to be agnostic about the dust attenuation and combine for each galaxy the two probability density functions P_Calzetti(β) and P_SMC(β) derived from our Monte Carlo approach (Section 4.3.1) assuming equal weighting to derive the median β and its 1σ uncertainties. In the lower left panel of Figure 19, the final combined β slopes are compared to the β derived assuming a Calzetti attenuation. The offset toward redder β is significantly reduced due to narrower probability density functions, assuming Calzetti dust (hence, the average β is drawn to the Calzetti solution in most cases).

4.3.3. The β Slopes of the ALPINE Galaxies in Context

Figure 20 shows our β measurement (marginalized over both Calzetti and SMC dust) as a function of stellar mass split in the different methods of selection (Section 2.1). As the UV slope is mostly affected by the dust attenuation, the strong correlation between β and stellar mass is not surprising as more massive galaxies are expected to be more dusty. The z ∼ 5.5 narrowband-selected galaxies, statistically, have the bluest slopes, indicating their dust-poor nature. The other galaxies are spread out over the whole parameter space. We also show the data from our parent sample at 4 < z < 6 in gray and compare their β slope and stellar mass distribution to the ALPINE sample in the kernel density estimate plots. Note that the β slope distribution of ALPINE galaxies peaks at Δβ ∼ 0.2 bluer values than the parent sample at the same redshift. This is a minor bias (likely caused by our spectroscopic selection) that has to be kept in mind for future analyses.

Zoom In Zoom Out Reset image size

Rest-frame optical emission lines in z > 4 galaxies are out of reach of current spectrographs. Specifically, the Hα emission provides a good tool to study the star formation properties of galaxies in more detail. Fortunately, in the redshift range 4 < z < 5, the Hα line falls in the Spitzer filter, while the filter lacks any strong emission lines. Therefore, the []−[] color can be used to constrain the Hα line flux and its equivalent width (a proxy of recent stellar mass build up). This method leads to Hα emission properties that are statistically as accurate as derived from spectroscopic data (Faisst et al. 2016a). Several such measurements have been carried out in the past with success (Shim et al. 2011; Stark et al. 2013; de Barros et al. 2014; Smit et al. 2014; Mármol-Queraltó et al. 2016; Faisst et al. 2016a; Rasappu et al. 2016; Smit et al. 2016; Caputi et al. 2017; Faisst et al. 2019).

About 55% of the ALPINE sample (66 galaxies) lie in this redshift range. To measure the Hα luminosity and equivalent widths, we follow the same technique as outlined in Faisst et al. (2019; we refer to this paper for more technical details). In brief, this method makes an assumption on the rest-frame optical continuum to which emission lines are added in a consistent manner to reproduce the observed []−[] colors of the galaxies. This approach is robust, as it only depends on the slope of the rest-frame optical continuum, which is well defined and nearly independent of assumptions on age, metallicity, and star formation history for galaxies younger than ∼1 Gyr (mostly the case at z > 4). To describe the rest-frame optical continuum, we use several basis stellar population models based on the Bruzual & Charlot (2003) template library (see Table 4). For the dust correction of the Hα emission, we assume the stellar values derived by LePhare, which we convert to nebular extinction factors by assuming an f-factor⁴⁷ of 0.44 as measured in local starburst galaxies (Calzetti et al. 2000). We also assume a Calzetti and SMC reddening law. Furthermore, we assume an [] to Hα ratio of 0.15, as expected for galaxies at (Faisst et al. 2018), to correct the blending of the [] and Hα lines.

Table 4.  List of Basis Stellar Population Models for the Parameterization of the Rest-frame Optical Continuum at 4 < z < 5 to Derive Hα Emission from Spitzer Colors

Model	SFH	Metallicity	Dust attenuation
		(Z⊙ = 0.02)
A	Constant	0.02	Calzetti
B	Constant	0.004	Calzetti
C	Exp. declininga	0.01	Calzetti
D	Constant	0.02	SMC
E	Constant	0.004	SMC
F	Exp. declininga	0.01	SMC

Note.

^aAssuming .

Download table as:  ASCIITypeset image

In Figure 21, we show systematic uncertainties in the measurement of the Hα luminosity due to the assumptions in our model for the rest-frame optical continuum and the reddening law (models A through F, see Table 4), as well as the f-factor for and 0.4. It is evident that different assumptions in metallicity and SFH have a negligible impact on the measured Hα luminosity. The choice of the reddening law matters as the Hα luminosity decreases by ∼0.3 dex for galaxies at high Hα luminosities () assuming an SMC reddening law. The f-factor is the largest uncertainty in this measurement method and will have to be pinned down by future observations with the JWST. For now, the assumed f = 0.44 provides likely an upper limit on the Hα luminosities. As shown by the arrows in Figure 21, assuming an f-factor equal to unity (which is thought to be more likely based on observations at z ∼ 2) would decrease the Hα luminosities by up to 0.4 dex (0.8 dex) for a stellar dust reddening of 0.2 (0.4) magnitudes. The correction in the case of an SMC reddening law are 0.1–0.2 dex less. We note that these factors also apply to Hα-derived SFRs and any other quantity that depends linearly on the Hα luminosity.

Zoom In Zoom Out Reset image size

The top panel of Figure 22 compares the SFRs derived from SED fitting (Section 4.1) to the Hα luminosity and Hα-derived SFRs for galaxies without contaminated Spitzer photometry. The latter is derived using the standard conversion factor given in Kennicutt (1998),

assuming solar metallicity and a Chabrier IMF. Assuming one-fifth solar metallicity, the inferred SFR is expected to be ∼0.2 dex lower (Ly et al. 2016). As shown in Faisst et al. (2019), the uncertainty of this conversion factor is negligible compared to the impact of the uncertain f-factor. The Hα luminosities in the ALPINE sample range from to assuming f = 0.44 and Calzetti et al. (2000) dust attenuation. The Hα-derived SFRs trace well the SED-derived SFRs above . Below that value, we see a large scatter in Hα derived SFRs, which happens when the Hα emission becomes too faint to be measured reliably using the Spitzer broad bands. Specifically, this is the case roughly at , which corresponds to a 4.5 μm detection of less than 5σ (see Figure 3 and the Appendix in Faisst et al. 2019).

Zoom In Zoom Out Reset image size

The lower panel of Figure 22 relates the Hα luminosity (here in units of solar luminosity) to the [] luminosity measured by ALMA (also solar luminosity). In addition, we show the expected relation between Hα and [] luminosity by combining Equation (5) with the linear relation between [] and SFR derived by De Looze et al. (2014) from local and low-redshift galaxy samples,

for different values of intercepts (γ) and slopes (α). Specifically, we are showing the relation for their entire sample (α = 1.01, γ = − 6.99, blue hatched) and the metal-poor dwarf galaxies (α = 0.80, γ = − 5.73, red hatched). For bright [] galaxies (), we find a good agreement with the entire local sample. For lower [] luminosities, the Hα measurements fall below this relation and are instead more consistent, although with a large scatter, with the relation of metal-poor local dwarf galaxies. The majority of [] undetected galaxies align well with either relation, however, the uncertainty in the Hα measurements becomes substantial as galaxies fall below .

Figure 23 shows the rest-frame Hα EW distribution of our ALPINE galaxies in the context of local galaxies (gray cloud) and other z ∼ 4.5 galaxies on the COSMOS field (gray, Faisst et al. 2019). The Hα EW for the z > 4 galaxies is derived consistently assuming a constant SFH, Calzetti reddening law, and f = 0.44. The ALPINE galaxies cover the parameter space of the other z ∼ 4.5 galaxies and, hence, build a representative sample also in terms of Hα properties. For a fixed stellar mass, the high-redshift galaxies have higher Hα EWs compared to local galaxies, which is expected from a galaxy evolution point of view as galaxies at higher redshift are highly star-forming. Note that two galaxies in the ALPINE sample have similar Hα EW values as massive () local galaxies. Consistently, also their sSFR are low (), which is indicative of them being systems with evolved stellar populations at high redshifts.

Zoom In Zoom Out Reset image size