Abstract
We report molecular gas mass estimates obtained from a stacking analysis of CO line emission in the ALMA Spectroscopic Survey (ASPECS) using the spectroscopic redshifts from the optical integral field spectroscopic survey by the Multi Unit Spectroscopic Explorer (MUSE) of the Hubble Ultra Deep Field (HUDF). Stacking was performed on subsets of the sample of galaxies classified by their stellar mass and position relative to the main-sequence relation (on, above, below). Among all the CO emission lines, from CO(2–1) to CO(6–5), with redshifts accessible via the ASPECS Band 3 and the MUSE data, CO(2–1) provides the strongest constraints on the molecular gas content. We detect CO(2–1) emission in galaxies down to stellar masses of . Below this stellar mass, we present a new constraint on the molecular gas content of main-sequence galaxies by stacking based on the MUSE detections. We find that the molecular gas mass of main-sequence galaxies continuously decreases with stellar mass down to . Assuming a metallicity-based CO–to–H2 conversion factor, the molecular gas-to-stellar mass ratio from to ∼10.0 does not seem to decrease as fast as for , which is in line with simulations and studies at lower redshift. The inferred molecular gas density of MUSE-selected galaxies at is comparable with the one derived in the HUDF with a different CO selection. Using the MUSE data we recover most of the CO emission in our deep ALMA observations through stacking, demonstrating the synergy between volumetric surveys obtained at different wave bands.
Export citation and abstract BibTeX RIS
1. Introduction
Stars form inside dense molecular gas clouds. It is thus critical to reveal how much molecular gas exists in galaxies to characterize galaxy formation and evolution (e.g., Kennicutt & Evans 2012; Carilli & Walter 2013; Hodge & da Cunha 2020; Tacconi et al. 2020). It is well established that the cosmic star formation rate (SFR) density increased from the early stages of the universe toward its peak around , after which it progressively decreased until the current epoch (Madau & Dickinson 2014). A broad consensus is emerging on the cause of this growth, peak, and decline of the star formation history over cosmic time via measurements of the gas that fuels star formation (e.g., Walter et al. 2014; Decarli et al. 2016a; Scoville et al. 2017; Decarli et al. 2019; Liu et al. 2019; Riechers et al. 2019; Lenkić et al. 2020; Magnelli et al. 2020; Tacconi et al. 2020). This evolution could be due to either the available supply of molecular gas for forming stars, a mechanism that causes high efficiency in star formation such as galaxy mergers, or a combination of these processes.
Most of the star formation in the universe occurs in galaxies residing on the so-called "main sequence (MS)," a tight correlation between the SFR and stellar mass (M*) of star-forming galaxies (e.g., Noeske et al. 2007; Salmi et al. 2012; Whitaker et al. 2012, 2014; Schreiber et al. 2015; Popesso et al. 2019a, 2019b). This correlation is observed at redshifts up to at least (e.g., Speagle et al. 2014; Salmon et al. 2015). In other words, most star formation in the universe is long-lasting and evolves steadily, supporting the idea that it is predominantly regulated by the gas accretion of the available fuel supply and feedback processes, rather than stochastic events like galaxy mergers (Dekel & Krumholz 2013; Lilly et al. 2013; Tacchella et al. 2016). Such stochastic events cause enhanced star formation activity, which elevates SFRs significantly above the MS relation. However, these galaxies, referred to as starbursts, are in the minority (e.g., Rodighiero et al. 2011; Sargent et al. 2012; Lutz 2014). To understand these two star formation modes and the efficiency of the star formation process, it is essential to determine the H2 gas supply and deficiency. The most common tracer of the molecular gas, the fuel of star formation, is line emission from carbon monoxide () rotational transitions at low excitation (Bolatto et al. 2013; Carilli & Walter 2013).
The connection between molecular gas content, stellar mass, and SFR for galaxies on and above the MS relation has been investigated in various targeted studies (e.g., Magdis et al. 2012; Tacconi et al. 2013, 2018; Santini et al. 2014; Scoville et al. 2016). However, the targets have so far been limited to massive () and highly star-forming () galaxies. Little is known about the molecular gas reservoirs in galaxies either below the MS or at modest stellar masses, mostly because of sensitivity limits.
The ALMA Spectroscopic Survey in the Hubble Ultra-Deep Field (ASPECS; Walter et al. 2016) has been conducted as a spectroscopic survey over the entire frequency range of ALMA Bands 3 and 6 in the Hubble Ultra Deep Field (HUDF; Beckwith et al. 2006) to perform an unbiased search for multiple rotational transitions of CO emission (Walter et al. 2016; González-López et al. 2019). Spectral line scans have an advantage in assessing the molecular gas content based on a complete line flux-limited sample without any target preselections. González-López et al. (2019) conducted a blind search of line and continuum sources directly in the ASPECS data cube (Band 3) and evaluated its completeness. Using these reliable CO detections, Decarli et al. (2019) constructed CO luminosity functions and presented the evolution of the cosmic gas mass density. A census of the molecular gas content of galaxies that have direct CO detections is shown and discussed in Aravena et al. (2019). These gas measurements were compared with model predictions from cosmological simulations and semi-analytical models in Popping et al. (2019). Uzgil et al. (2019) performed a power spectrum analysis and probed CO emission at below the sensitivity limit of individual detections and gave a constraint on missing CO emission from individually undetected galaxies.
Here, we maximize the sensitivity of the ASPECS data to detect CO emission by stacking ALMA spectra using the Band 3 data. Our stacking analysis is based on optical spectroscopic redshifts from another large, unbiased, blind spectroscopic survey in the HUDF carried out by the integral field unit (IFU) instrument MUSE (Multi Unit Spectroscopic Explorer) on the Very Large Telescope (VLT; Bacon et al. 2017). The combination of the three-dimensional (3D) data obtained by both the ASPECS and MUSE surveys not only made the stacking analysis possible, but also enabled a direct comparison between the molecular gas properties and rest-frame optical/ultraviolet properties (Boogaard et al. 2019).
This paper is structured as follows: we first introduce the observations and data taken with the ASPECS and MUSE–HUDF surveys and the ancillary data in Section 2. We describe the method of the stacking analysis along with the sample used for stacking in Section 3. In Section 4, the stacked CO spectra are presented. We then convert the measured CO emission to molecular gas mass and discuss the gas mass content in Section 5. We summarize and conclude our findings in Section 6. A flat
2. The ASPECS and MUSE Datasets
The MUSE survey covered the entire area of the HUDF (Bacon et al. 2017), whereas the ASPECS Large Program (LP) observed almost the entire region of the Hubble eXtremely Deep Field (XDF; Illingworth et al. 2013). MUSE is an optical IFU (Bacon et al. 2015) on the VLT Yepun (UT4) of the European Southern Observatory (ESO) with a wide field-of-view (FoV; ), high sensitivity, wide wavelength coverage (), and high spectral resolution (). In particular, the large size of its FoV facilitates spectroscopic redshift surveys without requiring any target preselection, achieving a spatially homogeneous spectroscopic completeness. It offers redshift determination at based on rest-frame ultraviolet and optical emission and absorption lines.
Complementing the MUSE spectroscopic survey, the ASPECS line scan survey in ALMA Band 3 (3mm) can detect multiple CO rotational transition lines from to in the redshift range of (with gaps 28 at and ; Walter et al. 2016). The MUSE redshift coverage overlaps well with the redshift coverage of the ASPECS Band 3 line scan survey at (Figure 1), except for a small gap at .
In this paper, we concentrate on the ASPECS Band 3 data where the full MUSE redshift coverage can be exploited (see Section 3.2). The significant redshift overlap of these two unbiased spectroscopic surveys in the same field is beneficial for performing stacking analyses on ASPECS CO spectral lines based on the optically determined MUSE redshifts. In this section, we will briefly present the survey designs of the HUDF conducted by ASPECS and MUSE.
2.1. ASPECS Observations and Data
The detailed survey strategy and data reduction of the ASPECS Pilot and Large Program surveys are presented in Walter et al. (2016) and González-López et al. (2019), respectively. The observational setups of the LP survey were the same as the Pilot survey, except its coverage is extended to . In this work, we used the combined data of the ASPECS Pilot and LP surveys taken with ALMA Band 3. ASPECS LP carried out a full frequency scan in Band 3 (84−115 GHz) over the XDF, which resides in the HUDF. With a total of 17 pointings centered at (R.A., decl.) = (03:32:38.5, −27:47:00), which completely cover the Pilot region, the total area with a primary beam response in the LP survey is at (the central frequency of Band 3).
The Common Astronomy Software Applications (CASA) software was used to calibrate and image the data. With the C40-3 array configuration, we obtained a synthesized beam size of with a position angle at by using natural weighting in CASA when imaging the data. The frequency channel was rebinned to 7.813 MHz ( at 99.5 GHz). The sensitivity for this channel bin size was throughout the entire scanned frequency range. The 5
Using the same data cube, González-López et al. (2019) reported CO emission line detections from an unbiased blind search without prior knowledge of source positions and observed CO frequencies. They found 16 high significance CO emitting sources, among which 11 were identified as CO(2–1) emission. We refer the readers to González-López et al. (2019) for comprehensive discussions on the detection methods. Aravena et al. (2019) assessed the molecular gas properties of these sources, which will be used for comparison in this paper.
2.2. MUSE Observations and Data
The MUSE–HUDF deep survey was conducted as a two layer survey of different depths (Bacon et al. 2017). The deep survey observed the entire HUDF region, whereas the ultra-deep survey was carried out near the center of the deep survey area. The and exposure times, respectively, reached 3
The MUSE spectra were extracted with two methods: prior extractions based on the Ultraviolet Hubble Ultra Deep Field catalog (UVUDF; Rafelski et al. 2015) and a blind search for emission lines in the data cubes. In the former case, the coordinates for the source extraction were from UVUDF, 29 whereas for the latter, the coordinates were determined from the MUSE data. For a more detailed description of the survey and data treatments, see Bacon et al. (2017). Following Dunlop et al. (2017), we corrected the known systematic offset of the Hubble Space Telescope (HST) positions to match the radio astrometric reference frame by applying in decl. and in R.A.
The MUSE spectroscopic redshifts were measured from the spectral features in the extracted spectra. Each redshift has an associated confidence level of 3 (secure redshift, determined by multiple spectral lines), 2 (secure redshift, determined by a single spectral line), or 1 (possible redshift, determined by a single spectral line whose spectral identification remains uncertain). The typical MUSE redshift uncertainty is or , which is smaller than the typical line width of CO emission. We refer to Inami et al. (2017) for details about the redshift determination and redshift catalogs of the MUSE–HUDF field. In this work, we only use the reliable MUSE redshifts of confidence levels 2 and 3.
The MUSE–HUDF survey obtained 1338 reliable redshifts in total, a factor of eight increase over the previously available spectroscopic redshifts in this field. The simultaneous wavelength coverage of and the spectral resolution of of MUSE allowed detections and unambiguous identifications of major rest-frame ultraviolet and optical emission lines, including H
Among the spectroscopic redshifts assessed in the MUSE–HUDF field, 503 sources with a confidence level of 2 or higher lie within the ASPECS survey region (LP primary beam response ). Out of these sources, 107 sources are [O ii] emitters covering and 363 sources are Ly
2.3. Ancillary Data and Physical Parameters Derived from SED Fitting
Owing to the same HUDF coverage of the MUSE and ASPECS observations, there are abundant ancillary data available. We assembled optical and near-infrared photometric data from Skelton et al. (2014). These photometry catalogs and data include ultraviolet to infrared from the HST, various ground-based telescopes, and all of the Spitzer IRAC channels.
Based on this photometric data set, we inferred physical parameters such as SFR and stellar mass (M*) via modeling with the high-redshift extension of the spectral energy distribution (SED) fitting code MAGPHYS (da Cunha et al. 2008, 2015). In addition to Spitzer/MIPS and Herschel/PACS, the ALMA 1.2 and 3 mm photometry from the ASPECS data (González-López et al. 2019, 2020) were also used for the SED fitting for a subset of the sources whose CO or continuum emission was detected. The same procedure was carried out in the other ASPECS work (Aravena et al. 2020; Boogaard et al. 2020). The SED fits were computed based on the MUSE spectroscopic redshift. See Boogaard et al. (2019) for the detailed process of the SED fitting.
3. Methods
3.1. ALMA Spectral Extraction and Stacking
From the ALMA data, we first extracted a sub-cube centered at the position of each MUSE source with a secure spectroscopic redshift. This sub-cube has a size of . The primary beam correction using the combined Pilot and LP primary beam response was applied after the sub-cube extraction.
The uncertainty of the extracted spectra was calculated based on the region in the data cube where the combined beam response is of the peak sensitivity at the phase center. The uncertainty for each spectrum is then scaled by the combined primary beam response of the location of the objects under consideration.
To perform stacking, the channel frequencies of the extracted spectra were converted to the rest frame, then the spectra were resampled onto a common frequency grid. In the rest frame, a CO line detected at the lower frequency end of the spectrum (i.e., galaxies at higher redshift) has a coarser spectral sampling due to the correction of the redshift. Thus, the common frequency grid was given at the coarsest sampling, corresponding to the CO line detected at the lowest frequency end of Band 3. In the case of CO(2–1), we set the final spectral range and the channel width of the extracted spectra to be , centered on the CO(2–1) rest-frame frequency ( GHz), and (), respectively. This resulted in 108 channels in the extracted spectra. The final extracted rest-frame frequency range is GHz (). The same procedure is adopted for stacking high-J CO lines.
This conversion was implemented by first shifting the channel frequencies of the spectra to the rest frame, according to the MUSE redshift. We then applied a Gaussian decimation filter to the rest-frame spectra via the Fourier plane. This filter removes fine scale (high sampling frequency) noise in the spectrum which would otherwise be aliased into spurious noise when sub-sampled to the lower target resolution (Lyons 2010). A better signal-to-noise ratio (S/N) was thus obtained when we binned all spectra onto the coarser common frequency grid in the later step. In the Fourier plane, the width of the Gaussian filter was set to give an attenuation of 40
Next, we interpolated the filtered spectrum onto the frequency grid of the stacked spectra. The mean of the resulting spectra was then calculated to obtain the final stacked spectrum. 30 We did not apply any weighting when performing the stack.
We also carried out two different random extractions following the same procedure described above, to produce random stacked spectra for comparisons. One random extraction involved assigning random redshifts to each spectrum before combining them at the location of the known MUSE position. The other involved using the correct MUSE redshifts, but extracting the spectra at random positions within the region where the LP primary beam response is , instead of at the known source position. The number of randomized extractions is the same as the number of CO emission samples. Neither of these methods should produce stacked spectra with real features. This random spectral extraction highlights which features in the real stacked spectra should be discounted as noise.
In Figure 2, we show the standard deviation of randomly stacked spectra against the number of the stacks. For this plot, we randomize both redshift and position for the spectral extractions. The standard deviation of stacked spectra roughly decreases with .
Download figure:
Standard image High-resolution image3.2. Sample Selection
We took advantage of the significant redshift overlap between the ASPECS and MUSE surveys to carry out a stacking analysis on CO line emission: we focused on the ASPECS Band 3 data where the full MUSE redshift coverage can be exploited. In particular, over the range , the ASPECS survey has the highest sensitivity for detecting CO(2–1) among all of the observable CO emission features (Figure 9 of Boogaard et al. 2019). In addition, this redshift range is where MUSE is efficient at detecting spectral features (up to for [O ii]) as shown in Figure 1.
We first selected a subset of the MUSE sources for stacking CO(2–1) emission in the ASPECS Band 3 data cube. The criteria were the following: (1) , (2) MUSE spec confidence levels , and (3) location within ALMA Band 3 LP primary beam response . For high-J CO emission, we used the same selection criteria, except for the redshift range. The ranges were for CO(3–2), for CO(4–3), for CO(5–4), and for CO(6–5).
The resulting number of MUSE sources for stacking CO(2–1) was 111 in total. 31 For 104 sources, the MUSE spectroscopic redshift identifications were based on [O ii], five sources used absorption features, one source used C iii], and one quasar had strong Mg ii emission. Among these 111 sources, 10 sources 32 were identified with CO(2–1) emission by the ASPECS blind search: MUSE IDs 8 (ASPECS-LP-3mm.06), 16 (3mm.11), 924 (3mm.14), 925 (3mm.16), 996 (3mm.02), 1001 (3mm.05), 1011 (3mm.10), 1117 (3mm.04), 6415 (3mm.08), and 6870 (3mm.15) (Boogaard et al. 2019; Aravena et al. 2019).
We performed the stacking in each group of the sample galaxies classified by the stellar mass and specific SFR () derived from the MAGPHYS SED fits (see Section 2.3). The relation of our sample is shown in Figure 3. The specific star formation rate (SSFR) classification was based on the MS relation from Equation 11 of Boogaard et al. (2018), using the mean redshift of the CO(2–1) line detectable in Band 3 ( Walter et al. 2016). Galaxies lying within, above, and below the intrinsic scatter (0.44 dex 33 ) of this relation are referred to as the "MS," "above" the MS, and "below" the MS, respectively, throughout this paper. We here use the MS relation from Boogaard et al. (2018) because their correlation is assessed with the objects whose spectral features and redshifts were measured based on the MUSE data. Compared with earlier studies such as Whitaker et al. (2014) and Schreiber et al. (2015), whose samples are mostly massive galaxies, Boogaard et al. (2018) better constrained the low-mass end of the relation. The numbers of galaxies in each group for stacking are presented in Table 1.
Download figure:
Standard image High-resolution imageTable 1. Numbers of Galaxies in the Stellar Mass and SSFR Bins for the CO(2–1) Stacking Sample
Above the MS | 11 (11) | 4 (3) | 1 (1) | 0 (0) |
On the MS | 38 (38) | 32 (32) | 12 (7) | 3 (0) |
Below the MS | 0 (0) | 5 (5) | 2 (2) | 3 (2) |
Note. The numbers in parentheses are after excluding the galaxies which have the CO(2–1) line identified by the blind search (see Section 3.2).
Download table as: ASCIITypeset image
The MS relation of Boogaard et al. (2018) was also adopted for classifying galaxies which expected to have CO lines. Similar to CO(2–1), we used the mean redshifts of z = 2.61, 3.80, 4.99, and 6.18 for CO(3–2), CO(4–3), CO(5–4), and CO(6–5) lines, respectively (Walter et al. 2016).
3.3. CO Line Flux and Upper Limit Measurements
For each stacked spectrum, we performed a best fit on the 2D image (moment-0) with a 2D Gaussian function to estimate the line flux or upper limit. The x- and y-positions are allowed to vary in the vicinity of the MUSE positions within a radius of 1/3 of the ALMA beam size. The widths were fixed to the beam size. We considered a CO emission line to have been detected when the amplitude of the fitted Gaussian function was higher than 3
Uncertainties in MUSE redshifts can cause some flux losses when we stack spectra. We assumed an emission line with a width of (full width at half maximum; FWHM) and performed a bootstrap simulation to assess the flux loss resulting from (Section 2.2). The total flux was measured within (corresponding to 19 slices in frequency; see Section 4.1.1 for this choice of ) centered at the rest frequency of CO(2–1). This velocity range was the same as the one we used to create coadded 2D images for flux measurements (see Section 4). With 10,000 realizations, we found that 92% of them resulted in less than 3% flux loss. We do not scale up our measured fluxes in this work to take account of this flux loss, because its impact on our line flux measurements is not significant.
4. CO Emission from Stacked Spectra
We performed CO emission line stacking from CO(2–1) to CO(6–5). In this section, we will present the resulting stacked spectra.
4.1. CO(2-1)
4.1.1. Stacking the Entire Sample
We first performed stacking without binning in stellar mass or SSFR to obtain a constraint on the average properties of all galaxies. The entire sample (111 spectra) and a subset of the sample (101 spectra) that excluded 10 objects whose CO emission were detected by the blind search (Section 3.2) were used to search for signals in the stacks (below the noise threshold of individual galaxies).
For these two sets of samples, we inspected 2D images (moment-0) that were coadded between () centered at the CO(2–1) rest frequency () in the corresponding 1D spectrum. This width is consistent with 2
The total line fluxes were obtained by fitting a 2D Gaussian function in the 2D images. This measure helps to avoid problems such as a possible slight positional offset between the optical and CO emission. All of the Gaussian parameters, the peak position, and amplitude were set to be free parameters, whereas the widths were fixed to the mean beam size. We obtained total line fluxes (upper limit) of () for the stacked spectra including (excluding) the sources with direct CO(2–1) detections.
4.1.2. Stacking in Stellar Mass and SSFR Bins
We carried out the stacking on both the entire sample and the sample that excluded the direct CO detections. To investigate the dependence of the molecular gas content on fundamental properties of galaxies, we divided the sample into ranges of stellar mass and SSFR to perform the stacking. As shown in Figure 3, we set the bins to have steps of over the range (four bins) with stellar mass, and galaxies "above" the MS, on the "MS", and "below" the MS based on the MS relation at z = 1.43 discussed in Section 3.2. In the redshift range where CO(2–1) can be detected with the ASPECS Band 3 survey (), the MS relation does not evolve significantly.
The stacked 2D and 1D spectra are shown in Figure 4 and the CO line flux measurements are summarized in Table 2. When the directly detected CO(2–1) emission is included, there are solid detections () at , regardless of their SSFRs (except in the bin where there is only a single source). On the other hand, at , we only obtain a signal in the bin above the MS with . When we exclude the known CO(2–1) emission from the stacks, a significant detection is seen in the MS bin with (seven sources).34
Download figure:
Standard image High-resolution imageTable 2. Summary of Measured Stacked CO(2–1) Line Fluxes ()
Above the MS | () | 0.110 ± 0.027 () | () | ⋯ (⋯) |
On the MS | () | () | 0.130 ± 0.021 (0.095 ± 0.024) | 0.453 ± 0.042 (⋯) |
Below the MS | ⋯ (⋯) | () | 0.119 ± 0.039 (0.119 ± 0.039) | 0.152 ± 0.027 () |
Note. The units are in . Values in parentheses are after excluding the galaxies which have the CO(2–1) line identified by the blind search (see Section 3.2).
Download table as: ASCIITypeset image
Although the bins of the MS galaxies in the ranges of and contain the largest numbers of galaxies, we do not detect a CO(2–1) line. To attempt to detect stacked CO emission for galaxies in these low-mass bins, we adopt a stellar mass bin size in case some hidden emission from a small number of sources has been diluted by averaging too many sources without any emission. This smaller bin size results in 14, 24, 23, and 9 MS sources from to in steps of . No detection is identified even with these finer bins. The estimated 3
Among the bins with stacked detections, although the censoring fraction is unity for the bin of below the MS with stellar mass , all of the remaining bins are . In particular, the fraction is for the MS galaxies.
4.1.3. The CO(2-1) Lines Detected in Individual Galaxies
We inspect individual spectra of galaxies in the MS bin of , where the stacked spectrum has the highest S/N (the spectra are depicted in Appendix A). Excluding the emission already found by the blind search, it is possible to identify the CO(2–1) line with lower S/N in MUSE IDs 879, 985, and 1308. We note that MUSE IDs 879 and 985 have already been reported by Boogaard et al. (2019) as the MUSE prior-based sample (see their Table 2 and Figure 4). A further potential stacked CO(2–1) detection is seen in the range of for the mean of two galaxies lying below the MS. A detected CO(2–1) line is dominated by MUSE ID 928.
4.2. Higher-J CO Stacked Spectra
Along with the CO(2–1) stacking analysis, we attempt to detect higher excitation CO emission, up to , with the same stacking method. The census of the stacked sources is shown in Table 3.
Table 3. Total Number of Galaxies Used for Stacking High-J CO Emission
CO | ||||||
---|---|---|---|---|---|---|
Above the MS | 3-2 | 1 (1) | 5 (5) | 5 (4) | 1 (-) | ⋯ |
4-3 | 4 (4) | 4 (4) | 5 (5) | 1 (1) | ⋯ | |
5-4 | ⋯ | ⋯ | ⋯ | ⋯ | ⋯ | |
6-5 | ⋯ | ⋯ | ⋯ | ⋯ | ⋯ | |
On the MS | 3-2 | 4 (4) | 21 (21) | 9 (9) | 4 (3) | ⋯ |
4-3 | 63 (63) | 79 (79) | 29 (29) | 4 (4) | ⋯ | |
5-4 | 40 (40) | 46 (46) | 26 (26) | 11 (11) | 2 (2) | |
6-5 | 2 (2) | 16 (16) | 14 (14) | 7 (7) | 3 (3) | |
Below the MS | 3-2 | ⋯ | ⋯ | ⋯ | ⋯ | ⋯ |
4-3 | ⋯ | ⋯ | 2 (2) | ⋯ | ⋯ | |
5-4 | 1 (1) | ⋯ | 1 (1) | 1 (1) | ⋯ | |
6-5 | 2 (2) | ⋯ | ⋯ | ⋯ | ⋯ |
Note. The numbers in parentheses are after excluding the galaxies which have the CO line identified by the blind search (see Section 3.2).
Download table as: ASCIITypeset image
For CO(3–2), among our sample selection (Section 3.2), MUSE IDs 35 and 1124 35 are the only sources that were already known from the blind search (ASPECS-LP-3mm.01 and 3mm.12). The former is a galaxy above the MS and the latter is on the MS with and 10.64, respectively. In addition, we consider MUSE ID 35's pair galaxy MUSE ID 24 () to have a CO(3–2) detection in our analysis. This is because the CO spatial extent, which peaks at the location of ID 35, covers ID 24 and it is difficult to distinguish the flux contributions from these two sources. When the stacking was performed in the same way as for CO(2–1), these two sources dominated the detections in their bins (Appendix B). If we discard them, no significant emission remains. This result agrees with the finding of Uzgil et al. (2019) who also did not find additional CO(3–2) emission with a masked auto-power spectrum using all MUSE positions with LP primary beam response . There is also no stack detection for CO emission in our sample with the blind search,36 nor the stacking.
We note that the MUSE spectroscopic redshifts of galaxies at , which allow detections of CO emission, were mostly determined using the Ly
We also visually investigate individual CO spectra with the MUSE redshifts as priors to search for possible CO emission which is washed out by stacking. Two potential CO(5–4) emission lines were identified as shown in Appendix C. If confirmed, they may be one of the first cases of high-J CO detection at for Ly
5. Discussion
In this section, we will concentrate on discussing the molecular gas content at , as we have obtained the most reliable measurements at this redshift with our analysis.
5.1. CO Luminosities and Molecular (H2) Gas Masses
We employed the following equation to obtain the CO luminosities from the measured CO(2–1) emission (Solomon et al. 1997; Solomon & Vanden Bout 2005):
where is the integrated line flux density, DL is the luminosity distance in Mpc, and is the observed frequency in GHz. For the redshift (z), we used the mean redshift of the galaxies in each stacking bin.
We then adopted the following equation and the same conditions as Decarli et al. (2016b) to infer molecular gas masses () from the line luminosities ():
where J is the upper level of the CO excitation and is the CO luminosity to gas mass conversion factor. We assume following Daddi et al. (2015). Based on galaxies detected with the ASPECS survey, Boogaard et al. (2020) estimated , which is comparable to the value from Daddi et al. (2015). Here we adopt the value from the former to be consistent with the molecular mass measurements published in Aravena et al. (2019), which are used for comparison in this paper. For , we used for galaxies in the stellar mass bins of which is one of the best estimates for high-redshift MS star-forming galaxies (Daddi et al. 2010). Among the ASPECS directly detected CO sources with metallicity estimates available from the MUSE spectra, all have roughly solar metallicity (Boogaard et al. 2019), which further justifies our choice of (Aravena et al. 2019). For the galaxies with , because they are likely to have lower metallicity, could be higher (e.g., Bolatto et al. 2013). In this work, we assume that the metallicities of galaxies with and ∼8.5 are and (e.g., Savaglio et al. 2005; Erb et al. 2006; Mannucci et al. 2009; Zahid et al. 2011; Sargent et al. 2014; Sanders et al. 2020). Thus, would increase by a factor of ∼2 and ∼3, respectively, compared to galaxies with (e.g., Genzel et al. 2012; Schruba et al. 2012). We adopt and for galaxies with and , respectively. In these lower stellar mass bins, the estimates with a constant are also shown as dotted symbols in figures for reference. The calculated line luminosities and molecular gas masses from CO(2–1) are summarized in Table 4.
Table 4. Summary of CO(2–1) Line Luminosities () and Molecular Gas Masses ()
Above the MS | (1) | () | 2.47 ± 0.61 () | () | ⋯ (⋯) |
(2) | () | 23.42 ± 5.89 () | () | ⋯ (⋯) | |
On the MS | (1) | () | () | 2.51 ± 0.40 (1.85 ± 0.47) | 10.57 ± 0.97 (⋯) |
(2) | () | () | 11.87 ± 1.95 (8.74 ± 2.26) | 50.07 ± 5.00 (⋯) | |
Below the MS | (1) | ⋯ (⋯) | () | 2.12 ± 0.69 (2.12 ± 0.69) | 3.10 ± 0.55 () |
(2) | ⋯ (⋯) | () | 10.03 ± 3.31 (10.03 ± 3.31) | 14.67 ± 2.68 () |
Note. The units are in and for (1) CO(2–1) line luminosities (the first row in each MS bin) and (2) molecular masses ( the second row), respectively. The CO–to–H2 conversion factors are assumed to be , 7.2, and 10.8 for galaxies in the bins of , , and , respectively (see Section 5.1). The values in parentheses are after excluding the galaxies which have the CO(2–1) line identified by the blind search (see Section 3.2).
Download table as: ASCIITypeset image
Table 5. Summary of Molecular Gas to Stellar Mass Ratios () and Depletion Times ()
Above the MS | (1) | () | 9.94 ± 2.50 () | () | ⋯ (⋯) |
(2) | () | 1.82 ± 0.46 () | () | ⋯ (⋯) | |
On the MS | (1) | () | () | 0.49 ± 0.08 (0.46 ± 0.12) | 0.24 ± 0.02 (⋯) |
(2) | () | () | 0.69 ± 0.11 (0.67 ± 0.17) | 1.13 ± 0.11 (⋯) | |
Below the MS | (1) | ⋯ (⋯) | () | 0.18 ± 0.06 (0.18 ± 0.06) | 0.08 ± 0.01 () |
(2) | ⋯ (⋯) | () | 1.75 ± 0.58 (1.75 ± 0.58) | 3.13 ± 0.57 () |
Note. The first and second rows in each MS bin are (1) molecular-to-stellar mass ratios () and (2) depletion times ( in ), respectively. The CO–to–H2 conversion factors are assumed to be , 7.2, and 10.8 for galaxies in the bins of , , and , respectively (see Section 5.1). The values in parentheses are after excluding the galaxies which have the CO(2–1) line identified by the blind search (see Section 3.2).
Download table as: ASCIITypeset image
In the following subsections, we discuss the gas content of galaxies at , together with their stellar masses and SFRs derived from the SED fitting (Section 2.3). The CO(2–1) emission used here stems from the measurements based on the 2D Gaussian fits in the images presented in Section 4.1.2 and Figure 4. Because we are interested in global properties of the molecular gas, we use the measurements that include the direct CO(2–1) detections from the blind search.
5.2. Molecular Gas and Stellar Mass Scaling Relation of MS Galaxies
Previous observational and theoretical studies of molecular gas in high-redshift galaxies have established a set of scaling relations that relate the molecular gas content to galaxy properties such as stellar masses, SFRs, and source sizes (e.g., Magdis et al. 2012; Somerville et al. 2012; Tacconi et al. 2013, 2018; Santini et al. 2014; Genzel et al. 2015; Scoville et al. 2016; Davé et al. 2017; Aravena et al. 2019; Freundlich et al. 2019; Popping et al. 2019). These relations have been key to understanding how the galaxy growth process has taken place through cosmic time. We first compare average molecular gas properties to the stellar mass of distant galaxies.
In conjunction with the MUSE deep survey in the same field, the stacking analysis facilitates the exploration of the gas scaling relations below stellar masses of , a regime that is rather uncharted in CO emission for individual detections at (see also Appendix D). We focus on discussing the MS galaxies, which have the largest numbers of stacked spectra in our CO(2–1) sample.
In Figure 5 (left), we plot the molecular gas mass from the stacking analysis against stellar mass of the MS galaxies, along with individual ASPECS CO detections (González-López et al. 2019; Aravena et al. 2019). The majority of directly detected galaxies have . At , there is only one direct detection. Although stacking also leads to no detection, the stacks of 38 and 32 spectra provide tight constraints on the average gas mass in the range of at . Taking upper limits into account, an increase of gas mass is found with increasing stellar mass from at least to 12.0.
Download figure:
Standard image High-resolution imageFor comparison, we also show the contours of the molecular gas measurements from the PHIBSS2 survey (included both the CO- and dust-based measurements; Tacconi et al. 2013, 2018; Freundlich et al. 2019). For further comparison with the literature, we show the distributions of the PHIBSS2 galaxies at lower redshifts in Appendix D. The gas masses of the PHIBSS2 sources were derived via a metallicity-based prescription for this parameter (Genzel et al. 2012), resulting in at .
We also compare our measurements with cosmological galaxy formation model calculations of molecular gas mass presented in Popping et al. (2019). In the same diagram we show three different model predictions: the IllustrisTNG hydrodynamical simulations with the "3.5arcsec" and "Grav" apertures and the SC SAM. The former corresponds to all the H2 within a radius of of the source center (similar to ASPECS) and the latter corresponds to all the H2 gravitationally bound to the galaxy. The H2 properties of galaxies were derived based on the molecular hydrogen fraction recipe of Gnedin & Kravtsov (2011).37
We use the gas mass model predictions at z = 1.43 (Popping et al. 2019), which is the mean redshift of the CO(2–1) line detectable in Band 3. At the higher stellar mass end, all of the three models predicted lower gas mass than the observed gas mass. At , about half of the sources with ASPECS direct detections lie within the 2
5.3. Molecular Gas-to-stellar Mass Ratio of MS Galaxies as a Function of Stellar Mass
We depict this plot with a different presentation in Figure 5 (right) to show a more common presentation of molecular gas mass normalized by the stellar mass (molecular gas-to-stellar mass ratio, ; Table 5) as a function of stellar mass. In the literature, a decrease of the gas-to-stellar mass ratio with increasing stellar mass at has been reported at (e.g., Tacconi et al. 2013). However, no observational CO constraints exist at lower stellar masses below . Our stacking analysis facilitates exploration of this lower stellar mass regime.
For galaxies with and 11.5, we estimate 0.49 ± 0.08 and 0.24 ± 0.02, respectively. Similarly to Figure 5 (left), these values are in agreement with the measurements of the PHIBSS2 galaxies lying on the MS relation of Boogaard et al. (2018; gray contours; see Appendix D) for a version using the MS relation of Speagle et al. (2014). In the lower stellar mass bin, , we constrain to have a 3
Beyond , a decrease of the gas-to-stellar mass ratio is discerned with increasing stellar mass in our stack results, as well as in previous work reporting both CO-based and dust-based gas estimates (Magdis et al. 2012; Tacconi et al. 2013, 2018; Genzel et al. 2015; Scoville et al. 2017; Aravena et al. 2019; Liu et al. 2019). As an example, we show in Figure 5 the result from PHIBSS2 with a steady decrease of with an increase of stellar mass (black line).
This decline of does not seem to be as steep at . If a constant is also adopted for these lower-mass bins, the gas-to-stellar mass ratio is consistent with being constant from to 10.5, then turning down around . We note that the models from Popping et al. (2019), despite a discrepancy with the observed results at the high mass side, show a similar constant gas-to-stellar mass ratio up to . This is consistent with our finding if we assume a constant conversion factor.
Given the limitations of our data, and the unknown dependence of the conversion factor, we cannot determine at which stellar mass a possible downturn of the gas-to-stellar mass ratio occurs. We, however, note that such a "plateau" in the low stellar mass regime has also been identified for local galaxies with direct detections of CO emission: here stays constant from at least and starts decreasing around (e.g., Saintonge et al. 2017; Bothwell et al. 2014). A similar result is found in Tacconi et al. (2018) who compared local star-forming galaxies, where CO measurements for low-mass galaxies are available (including Saintonge et al. 2017), to distant galaxies after removing assumed redshift effects on their gas mass content (see also Appendix D).
The declining at the high stellar mass end can be attributed to stellar feedback (e.g., Davé et al. 2011; Tacconi et al. 2013; Genzel et al. 2015). In contrast, the flatter at may imply that the effects of feedback are weaker. The drop in gas-to-stellar mass ratio seems to appear around , which is relevant to the characteristic stellar mass of star-forming galaxies (e.g., Duncan et al. 2014) where mass-quenching is becoming dominant (Peng et al. 2010).
Furthermore, the gas-to-stellar mass ratio below ∼1 at 3
The estimated molecular gas density at , based on our CO(2–1) stacking measurements, is . This value is lower than, but formally consistent with, the value derived in Decarli et al. (2019) with a different CO selection. The CO emission used in Decarli et al. (2019) included sources that did not enter the present analysis because of the lack of a counterpart with high-quality MUSE redshift. Our total CO flux estimate from stacking is also in line with the result from the CO auto-power spectrum analysis using the MUSE positions (Uzgil et al. 2019).
5.4. Dependence of the Molecular Gas Content on SSFR
In the previous section we only considered MS galaxies; in the remaining discussion, we will include galaxies above and below the MS relation of Boogaard et al. (2018) to discuss the dependence of molecular gas content on SSFR. Below, we keep the same assumed conversion factors as above: 3.6, 7.2, and for galaxies with , ∼9.5, and 8.5, respectively (Section 5.1). We include the constraints using a constant throughout the stellar mass bins in the figures for reference.
5.4.1. Gas-to-stellar Mass Ratios
As shown earlier, the gas-to-stellar mass ratio () provides information on the supply and depletion of gas reservoirs in galaxies. We compare against SSFRs in the left panel of Figure 6. We find that increases with increasing SSFRs, which is also seen in earlier studies of high-redshift galaxies, including the dust-based measurements of the molecular gas mass (e.g., Magdis et al. 2012; Tacconi et al. 2013, 2018; Genzel et al. 2015; Scoville et al. 2016; Liu et al. 2019). With the stacking analysis, we show that this trend holds even for galaxies below the MS at .
Download figure:
Standard image High-resolution imageIn the right panel of Figure 6, we show of stacked sources color-coded by different SSFRs to compare with their evolution (Geach et al. 2011; Magdis et al. 2012). Because our data points are derived from the CO(2–1) line, all of them lie between . A wide spread is related to the variations of across the MS relation seen in the left panel (see also e.g., Tacconi et al. 2018). The lower of low SSFR sources is also found by Spilker et al. (2018) who investigated galaxies lying below the MS with at . The depletion times of our sample below the MS are on average comparable with Spilker et al.'s sample. Following the same assumption as Spilker et al. (2018), if these galaxies continuously consume the existing gas with the observed SFR, then the of the galaxies with (11.5) would reduce to 1/10 at (0.2) and 1/100 at (0.0). Hence, by z = 0 their would be comparable to those of passive galaxies at the current epoch.
5.4.2. Gas Depletion Time
The gas depletion time, , is another way of examining the gas content in galaxies. It estimates the time taken for the gas to be fully consumed at the current SFR without accounting for additional fueling. We show the gas depletion time as a function of SSFR in Figure 7. Our results from the stacking analysis are in good agreement with previous studies. Both the stacked and directly detected sources show decreasing with increasing SSFR, except that the bin above the MS shows an elevated value. The data points from these two sets of measurements occupy the same range in the diagram, following the constant gas-to-stellar mass ratio () from around 0.1 to 1.0. The decreasing gas depletion timescale with SSFR demonstrates that galaxies with more extreme star formation consume their gas at a higher rate. This may be the major cause of the scatter as shown in Figure 6.
Download figure:
Standard image High-resolution image6. Summary and Conclusions
Based on the accurate redshifts from the MUSE IFU survey in the HUDF, we perform a CO emission stacking analysis with the ASPECS Band 3 data. When we split the sample into stellar mass and SSFR bins (on, below, and above the MS relation), we detect CO(2–1) emission () down to , even after removing previously reported CO detections. We do not recover any higher-J CO emission at higher redshift () with stacking when excluding the sources with direct CO detections.
The 3
Furthermore, the gas-to-stellar mass ratios of ∼0.5 and at the stellar masses of and , respectively, imply that the faint-end slope of the CO luminosity function is at least consistent or could be flatter than the stellar mass function. We have successfully recovered the majority of the CO emission at with the stacking analysis. The molecular gas density from this stacking analysis is comparable with the one inferred from a CO-driven selection (Decarli et al. 2019).
When we compare the gas-to-stellar mass ratio () against SSFR, we confirmed the known correlations to also hold for galaxies with low SSFRs at . The scatter in the SSFR correlation seems to be related to the decrease of at for star-forming galaxies. We also show that the gas-to-stellar mass ratios of massive galaxies () below the MS at are comparable to those at (Spilker et al. 2018; Tacconi et al. 2018).
Our stacking analysis of the combined volumetric surveys of ALMA and MUSE has let us explore regimes that were uncharted before and that will remain challenging for investigations that rely on direct CO detections.
The authors would like to thank the referee whose constructive comments helped improve the manuscript. The authors thank Ian Smail for useful suggestions to improve this study. This work was supported by JSPS KAKENHI grant No. JP19K23462 (H.I.). Este trabajo cont ó con el apoyo de CONICYT + PCI + INSTITUTO MAX PLANCK DE ASTRONOMIA MPG190030. F.W. acknowledges support by ERC Advanced Grant 740246 (Cosmic Gas). T.D-S. acknowledges support from the CASSACA and CONICYT fund CAS-CONICYT Call 2018. D.R. acknowledges support from the National Science Foundation under grant Nos. AST-1614213 and AST-1910107 and from the Alexander von Humboldt Foundation through a Humboldt Research Fellowship for Experienced Researchers. M.K. acknowledges support from the International Max Planck Research School for Astronomy and Cosmic Physics at Heidelberg University (IMPRS-HD). ALMA is a partnership of ESO (representing its member states), NSF (USA) and NINS (Japan), together with NRC (Canada), MOST and ASIAA (Taiwan), and KASI (Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by ESO, AUI/NRAO and NAOJ. This paper makes use of the following ALMA data: ADS/JAO.ALMA#2016.1.00324.L. Based on observations collected at the European Southern Observatory under ESO programme(s): 094.A-2089(B), 095.A-0010(A), 096.A-0045(A), and 096.A- 0045(B).
Facilities: ALMA - Atacama Large Millimeter Array, VLT:Yepun - .
Software: astropy (Astropy Collaboration et al. 2013; The Astropy Collaboration et al. 2018), CASA (McMullin et al. 2007).
Appendix A: The Individual Spectra in the Stacking
The individual spectra for the stacking of the galaxies on and below the MS with are displayed in Figure 8 (Section 4.1.3).
Download figure:
Standard image High-resolution imageAppendix B: The Stacked Images, Spectra, and Measured Upper Limits for High-J CO Lines
The stacked images and spectra for CO(3-2), CO(4–3), CO(5–4), and CO(6–5) are shown in Figure 9. The measured line flux upper limits are summarized in Table 6.
Download figure:
Standard image High-resolution imageTable 6. Estimated Line Flux Upper Limits of the Stacked High-J CO Emission
CO | ||||||
---|---|---|---|---|---|---|
Above the MS | 3-2 | ⋯ | ⋯ | |||
4-3 | ⋯ | |||||
5-4 | ⋯ | ⋯ | ⋯ | ⋯ | ⋯ | |
6-5 | ⋯ | ⋯ | ⋯ | ⋯ | ⋯ | |
On the MS | 3-2 | ⋯ | ||||
4-3 | ⋯ | |||||
5-4 | ||||||
6-5 | ||||||
Below the MS | 3-2 | ⋯ | ⋯ | ⋯ | ⋯ | ⋯ |
4-3 | ⋯ | ⋯ | ⋯ | ⋯ | ||
5-4 | ⋯ | ⋯ | ||||
6-5 | ⋯ | ⋯ | ⋯ | ⋯ |
Note. The units are in .
Download table as: ASCIITypeset image
Appendix C: Tentative Detections of CO(5–4)
In Figure 10, we present the two cases of potential CO(5–4) detections. As discussed in Section 4.2, further observations are required to verify these tentative detections.
Download figure:
Standard image High-resolution imageAppendix D: Comparisons with Earlier Studies
Here we demonstrate our results in the context of earlier studies including molecular gas measurements for galaxies at lower redshifts. We adopt the MS relation of Speagle et al. (2014), which was also used in the analysis of PHIBSS2 (Tacconi et al. 2018), to be consistent with the literature. Note that in the main text of this paper, the MS relation of Boogaard et al. (2018) is utilized throughout, because this is calibrated based on the MUSE-detected sources whose average stellar masses and SFRs are lower than the sample used in Speagle et al. (2014). Therefore, the data points and the boundaries of the MS relation shown in this Appendix are slightly shifted compared to the ones shown in the main text. As presented in the rightmost panels of Figures 11 and 12, the results are consistent even when the MS relation of Speagle et al. (2014) is adopted.
Download figure:
Standard image High-resolution imageIn this study, we put constraints on the molecular gas content through CO emission at , which has only been explored at lower redshift, as illustrated in the two left panels of Figure 11. The majority of the PHIBSS2 sample at are from xCOLD GASS (Saintonge et al. 2011a, 2011b, 2016, 2017). We find that at , the molecular gas mass of the MS galaxies is increasing with increasing stellar mass, at least from (top right). At lower redshift, this trend is seen starting from , but at smaller gas masses (top-left panel; see also Saintonge et al. 2017). In the bottom panels, a constant gas-to-stellar mass ratio between and ∼10.5 is shown at both and for galaxies lying on the MS. Then the gas-to-stellar mass ratio declines toward higher stellar masses.
In Figure 12, we show the same plots as in Figures 6 and 7 in the main text, but now the MS relation of Speagle et al. (2014) is adopted. We also include the best-fit lines from Tacconi et al. (2018).
Download figure:
Standard image High-resolution imageFootnotes
- 28
The ASPECS Band 6 (1mm) survey covers most of these gaps by observing higher-J CO lines.
- 29
In cases where MUSE could not spatially resolve the HST-detected sources, these sources were "merged" into a single MUSE object. Its new coordinates are the HST F775W flux-weighted center of all the merged objects.
- 30
The median stacking also produced similar results, but here we adopt the mean stacking for simpler noise estimates and for a better treatment when there are only two samples to stack.
- 31
There are three galaxies (MUSE IDs 6314, 6450, and 6530) which meet the selection criteria, but their stellar masses and SFRs cannot be constrained because they have no optical counterpart on which to perform an SED fit. These galaxies are found by MUSE (no HST counterpart in the UVUDF catalog due to blending). These objects are not included in our sample of 111.
- 32
Four of them detected by the CO blind search, ASPECS-LP-3mm.02, 04, 05, and 08, have a MUSE redshift confidence level of 1, but they are included in this work.
- 33
This intrinsic scatter is found by assuming a Gaussian function in a model of the star formation sequence. As noted in Boogaard et al. 2018, this value is higher than the average value reported in previous work.
- 34
Note that a detection in the bin below the MS with has already been observed when the known CO(2–1) emission was included, because none of the galaxies in this bin include the directly detected CO(2–1) emission (none of the galaxies have been excluded to make any changes).
- 35
This source has a new additional MUSE redshift z = 2.5739 whose foreground galaxy is identified at z = 1.098 (Boogaard et al. 2019).
- 36
- 37