The ALMA Spectroscopic Survey in the HUDF: Nature and Physical Properties of Gas-mass Selected Galaxies Using MUSE Spectroscopy

We focus on the ASPECS-LP Band 3 observations that have been completed in ALMA Cycle 4. The acquisition and reduction of the Band 3 data are described in detail in Decarli et al. (2019). The final mosaic covers a 4.6 arcmin² area in the HUDF (where the primary beam response is >50% of the peak sensitivity). The data are combined into a single spectral cube with a spatial resolution of ≈175 × 149 (synthesized beam with natural weighting at 99.5 GHz) and a spectral resolution of 7.813 MHz, corresponding to Δv ≈ 23.5 km s⁻¹ at 99.5 GHz. The average root-mean-square (rms) sensitivity is ≈0.2 mJy beam⁻¹ but varies across the frequency range, being deepest (≈0.13 mJy beam⁻¹) around 100 GHz and higher above 110 GHz, due to the spectral setup of the observations (see Gonzalez-Lopez et al. 2019 for details). Throughout this paper, we consider the area that lies within >40% of the primary beam peak sensitivity, which is the shallowest part of the survey over which we still detect CO candidates without preselection (Section 3.1). When comparing to the HST reference frame, we take into account an astrometric offset of Δα = +0076, Δδ = −0279 (Rujopakarn et al. 2016; Dunlop et al. 2017).

We perform an extensive search of the cube for molecular emission lines, as is detailed in Gonzalez-Lopez et al. (2019) and Section 3. With the Band 3 data alone, the ASPECS-LP is sensitive to different CO and $[{\rm{C}}\,{\rm{I}}]$ transitions at specific redshift ranges, which are indicated in the top panel of Figure 1.

Zoom In Zoom Out Reset image size

The HUDF was observed with the MUSE as part of the MUSE Hubble Ultra Deep Field survey (Bacon et al. 2017). The location on the sky of the ASPECS-LP with respect to the MUSE HUDF is shown in Decarli et al. (2019), Figure 1. The MUSE integral-field spectrograph has a 1' × 1' field of view, covering the optical regime (4750–9300 Å) at an average spectral resolution of λ/Δλ ≈ 3000. The HUDF was observed in a two-tier strategy, with the mosaic-region reaching a median depth of 10 hr in a 3' × 3' region and the udf10-pointing reaching 31 hr depth in a 1' × 1' region (3σ emission line depth for a point source of 3.1 and 1.5 × 10⁻¹⁹ erg s⁻¹ cm⁻² at 7000 Å, respectively). The data acquisition and reduction as well as the automated source detection are described in detail in Bacon et al. (2017). The measured seeing in the reduced data cube is 065 full width at half maximum (FWHM) at 7000 Å.

Redshifts were identified semiautomatically and the full spectroscopic catalog is presented in Inami et al. (2017). The spectra were extracted using a weighted extraction, where the weighting was based on the MUSE white light image, to obtain the maximal signal-to-noise. The spectra are modeled with a modified version of platefit (Brinchmann et al. 2004, 2008; Tremonti et al. 2004) to obtain line-flux measurements and equivalent widths for all sources. The typical uncertainty on the redshift measurement is σ_v = 0.00012(1 + z) or ≈40 km s⁻¹ (Inami et al. 2017), which we use to compute the uncertainties in the relative velocities.

In order to compare in detail the relative velocities measured between the UV/optical features in MUSE and CO in ALMA, we need to place both on the same reference frame. The MUSE redshifts are provided in the barycentric reference frame, while the ALMA cube is set to the kinematic local standard of rest (LSRK). When determining detailed velocity offsets we place both on the same reference frame by removing the velocity difference; BARY−LSRK = −16.7 km s⁻¹ (accounting for the angle between the LSRK vector and the observation direction toward the HUDF).

The redshift distribution of the MUSE galaxies that fall within >40% of the primary beam peak sensitivity of the ASPECS-LP footprint in the HUDF is shown in Figure 1, where galaxies are color coded by the primary spectral feature(s) used to identify the redshift (see Inami et al. 2017 for details). The redshifts that correspond to the ASPECS band 3 coverage of the different molecular lines are indicated in the top panel. CO(1–0)[115.27 GHz] is observable at the lowest redshifts (z < 0.3694), where MUSE still covers a major part of the rest-frame optical spectrum that contains a wealth of spectral features, including absorption and (strong) emission lines (e.g., ${\rm{H}}\alpha \ \lambda 6563$, $[{\rm{O}}\,{\rm{III}}]\lambda 4959,5007$, and $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$). The strong lines are the main spectral features used to identify star-forming galaxies all the way up to z < 1.50, where $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$ moves out of the spectral range of MUSE. CO(2–1) [230.54 GHz] is covered by ASPECS at 1.0059 < z < 1.7387, mostly overlapping with $[{\rm{O}}\,{\rm{II}}]$ in MUSE. At z > 1.5, the main features used to identify these galaxies are absorption lines such as $\mathrm{Mg}\,{\rm{II}}\ \lambda 2796,2803$ and $\mathrm{Fe}\,{\rm{II}}\ \lambda 2586,2600$. Over the redshift range of CO(3–2) [345.80 GHz], 2.0088 < z < 3.1080, MUSE only has coverage of weaker UV emission lines (mainly ${\rm{C}}\,{\rm{III}}]\ \lambda 1907,1909$), making redshift identifications more challenging (the "redshift desert"). Here, UV absorption lines are commonly used to identify redshifts, for galaxies where the continuum is strong enough (m_F775W ≲ 26 mag). Above z = 2.9, MUSE flourishes again, with the coverage of Lyα λ1216 all the way out to z ≈ 6.7. Here, ASPECS covers CO(4–3) [461.04 GHz] and transitions with J_up ≥ 4, and atomic carbon lines (${[{\rm{C}}{\rm{I}}]}_{1-0}$ 610 μm and ${[{\rm{C}}{\rm{I}}]}_{2-1}$ 370 μm).

In order to construct spectral energy distributions (SEDs) for the ASPECS-LP sources, we utilize the wealth of available photometric data over the HUDF, summarized below.

We use the photometric compilation by Skelton et al. (2014, see references therein), which includes UV, optical, and near-IR photometry from the Hubble Space Telescope (HST) and ground-based facilities, as well as (deblended) Spitzer/IRAC 3.6, 4.5, 5.8, and 8.0 μm. We also include the corresponding deblended Spitzer/MIPS 24 μm photometry from Whitaker et al. (2014). We take deblended far-infrared (FIR) data from Herschel/PACS 100 and 160 μm from Elbaz et al. (2011), which have a native resolution of 67 and 110, respectively. The PACS 100 and 160 μm have a 3σ depth of 0.8 and 2.4 mJy and are limited by confusion. For the flux uncertainties we use the maximum of the local and simulated noise levels for each source, as recommended by the documentation.³⁰ We further include the 1.2 mm continuum data from the combination of the available ASPECS-LP data with the ALMA observations by Dunlop et al. (2017), taken over the same region, as detailed in Aravena et al. (2019). We also include the ASPECS-LP 3.0 mm continuum data, as presented in Gonzalez-Lopez et al. (2019). For the ASPECS survey we have created a master photometry catalog for the galaxies in the HUDF, adopting the spectroscopic redshifts from MUSE (Section 2.2) and literature sources, as detailed in Decarli et al. (2019).

We use the high-z extension of the SED-fitting code magphys to infer physical parameters from the photometric information of the galaxies in our field (Da Cunha et al. 2008, 2015). The high-z extension of magphys includes a larger library of spectral emission models that extend to higher dust optical depths, higher SFRs and younger ages compared to what is typically found in the local universe. From the spectral emission models, the code can constrain the stellar mass, sSFR, and dust attenuation (A_V) along the line of sight. An energy-balance argument ensures that the amount of absorption at rest-frame UV/optical wavelengths is consistent with the light reradiated in the infrared. The code performs a Bayesian inference of the posterior likelihood distribution of the fitted parameter, to account for uncertainties such as degeneracies in the models, missing data, and nondetections.

We run magphys on all the galaxies in our catalog, using the available photometric information in all the bands (listed in Appendix B). We do not include the Spitzer/MIPS and Herschel/PACS photometry in the fits of the general sample because the angular resolution of these observations is relatively modest (>5''), thus a delicate deblending analysis would be required (the average sky density of galaxies in the HUDF is ≳1 galaxy per 3 arcsec²). For the CO-detected galaxies we repeat the magphys fits including these bands (Section 4.1). In order to take into account systematic errors in the zero-point fitting for these sources, we add the zero-point errors (Skelton et al. 2014) in quadrature to the flux errors in all filters except HST, and include a 5% error-floor to further account for systematic errors in the physical models (following Leja et al. 2018). The filter selection of the general sample provides excellent photometric coverage of the stellar population. Paired with the wealth of spectroscopic redshifts (see Decarli et al. 2019 for a detailed description), this enables robust constraints on properties such as M_*, SFR, and A_V. We do note that while the formal uncertainties on the inferred properties are generally small, systematic uncertainties can be of order ∼0.3 dex (e.g., Conroy 2013).

To identify AGN in the field, we use the Chandra X-ray data available over the GOODS-S region from Luo et al. (2017), which reaches the full depth of 7 Ms over the HUDF area. In total, there are 36 X-ray sources within the ASPECS-LP region of the HUDF (i.e., within 40% of the primary beam). We spatially cross-match the X-ray catalog to the closest source within 1'' in our MUSE and multiwavelength catalog over the ASPECS-LP area, visually inspecting all matches used in this paper to ensure they are accurately identified.

At the depth of the X-ray data, there are multiple physical mechanisms (e.g., AGN and star formation) that may produce the X-ray emission detected at 0.5–7 keV. Luo et al. (2017) adopt the following six criteria to distinguish X-ray AGN from other sources of X-ray emission, of which at least one needs to be satisfied to be classified as AGN (we refer the reader to Xue et al. 2011; Luo et al. 2017 and references therein for details): (1) L_X ≥ 3 × 10⁴² erg s⁻¹, identifying luminous X-ray sources; (2) an effective photon index Γ_eff ≤ 1.0 indicating hard X-ray sources, identifying obscured AGN; (3) X-ray-to-R-band flux ratio of $\mathrm{log}\left({f}_{X}/{f}_{R}\right)\gt -1;$ (4) spectroscopically classified as AGN via, e.g., broad emission lines and/or high excitation lines; (5) X-ray-to-radio flux ratio of ${L}_{X}/{L}_{1.4\mathrm{GHz}}\,\geqslant 2.4\times {10}^{18}$, indicating an excess of X-ray emission over the level expected from pure star formation; and (6) X-ray-to-K-band flux ratio of $\mathrm{log}\left({f}_{X}/{f}_{{K}_{s}}\right)\gt -1.2$. Note that even with these criteria it is possible that some X-ray sources host low-luminosity or heavily obscured AGN and are currently misclassified.

Overall, there are six X-ray AGN in the ASPECS-LP volume at 1.0 < z < 1.7, all of which have a MUSE redshift (one being a broad-line AGN). In the ASPECS-LP volume at 2.0 < z < 3.1, there are seven X-ray AGN, three of which have spectroscopic redshifts from MUSE (including one broad-line AGN), and four with a photometric redshift (we discard one source in the catalog with a photometric redshift in this regime for which we cannot securely identify a counterpart in HST). There is one X-ray AGN at a higher redshift, which is also identified by MUSE as a broad-line AGN at z = 3.188.

An extensive description of the line search is provided in Gonzalez-Lopez et al. (2019). In summary, three independent methods were combined to search for CO lines in the ASPECS-LP band 3 data without any preselection: LineSeeker (González-López et al. 2017), FindClump (Decarli et al. 2014; Walter et al. 2016), and MF3D (Pavesi et al. 2018). The fidelity³¹ of these line candidates was estimated from the ratio of the number of lines with a negative and positive flux detected at a given S/N. Lastly, the completeness of the sample was estimated by ingesting simulated emission lines into the real data cube.

In total, there are 16 emission line candidates for which the fidelity is ≥0.9. Statistical analysis shows that this sample is free from false positives (the sum of their fidelities, based on the ALMA data alone, is 15.9; Gonzalez-Lopez et al. 2019). These 16 sources form the primary, line search sample and are shown in Figure 2. All these candidates have an S/N ≥ 6.4.

Zoom In Zoom Out Reset image size

For all sources in the primary sample, one or multiple potential counterpart galaxies are visible in the deep HST imaging shown in Figure 2. In order to confidently identify a single CO emission line, an independent redshift measurement of the potential counterpart measurement is needed. Given the wealth of multiwavelength photometry in the HUDF, photometric redshifts can often already provide sufficient constraints to discern between different rotation transitions of CO in the case of isolated galaxies at redshifts z ≲ 3. However, complex systems of several galaxies, or projected superpositions of independent galaxies at distinct redshifts, can make redshift assignments more complicated. Fortunately, the integral-field spectroscopy from MUSE is ideally suited to disentangle spectral features belonging to different galaxies, allowing us to confidently assign redshifts to the CO emission lines. The frequency of a CO line can correspond to different rotational transitions, each with a unique associated redshift. With the potential redshift solutions in hand, we systematically identify the CO line candidates from the line search. We provide a summary of the redshift identifications here. A detailed description of the individual sources and their redshift identifications can be found in Appendix A, where we also show the MUSE spectra for all sources (Figures 13–16).

First, we correlate the spatial position and potential redshifts of the CO lines with known spectroscopic redshifts from MUSE (Inami et al. 2017). From the MUSE redshifts alone, we immediately identify most (11/16) of the CO lines with the highest fidelity. The brightest (ASPECS-LP.3mm.01) is a CO(3–2) emitter at z = 2.54, showing a wealth of UV absorption features. The other 10 galaxies are a diverse sample of CO(2–1) emitters spanning the redshift range over which we are sensitive: 1.01 < z < 1.74. They show a variety of spectra at different levels of S/N, covering a range of UV and optical absorption and emission features. Notably, $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$ is detected in all galaxies where it is covered by MUSE, while $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$ is detected in some of the higher S/N spectra.

Next, we extract MUSE spectra for the remaining five (5/16) sources without a cataloged redshift and investigate their spectra for a redshift solution matching the observed CO line. We discover two new spectroscopic redshifts at z = 2.54 (ASPECS-LP.3mm.12) and z = 2.69 (associated with ASPECS-LP.3mm.09) confirming detections of CO(3–2), which were both not included in the catalog of Inami et al. (2017) as their spectra are blended with foreground sources. The former in particular demonstrates the key use of MUSE in disentangling a spatially overlapping system comprised of a foreground $[{\rm{O}}\,{\rm{II}}]$ emitter and a faint background galaxy, which is detected at S/N > 4 both via cross-correlation with a z ≈ 2.5 spectral template and by stacking absorption features (see Figure 18). For ASPECS-LP.3mm.03 and ASPECS-LP.3mm.07 we leverage the absence of spectral features (e.g., $[{\rm{O}}\,{\rm{II}}]$, $\mathrm{Ly}\alpha$), consistent with their faint magnitudes (m_F775W > 27 mag) and a redshift in the MUSE redshift desert, in combination with photometric redshifts in the z = 2–3 regime from the deep multiwavelength data, to confirm detections of CO(3–2). Lastly, we find ASPECS-LP.3mm.13 being CO(4–3) at z = 3.601, based on the photometric redshifts suggesting z ≈ 3.5 and the absence of a lower redshift solution from the spectrum. Lyα λ1216 is not detected for this source, but we caution that at this redshift $\mathrm{Ly}\alpha$ falls very close to the $[{\rm{O}}\,{\rm{I}}]\ \lambda 5577$ skyline. Furthermore, given that the source potentially contains significant amounts of dust, no $\mathrm{Ly}\alpha$ emission may be expected at all.

In summary, we determine a redshift solution for all (16/16) candidates from the line search. Twelve are directly confirmed by MUSE spectroscopy, while the remaining four are supported by their photometric redshifts and indirect spectroscopic evidence. We highlight that some of these counterparts are very faint, even in the reddest HST bands, and their identifications would not have been possible without the exquisite depth of both the HST and MUSE data over the HUDF. Similar objects would typically not have robust photometric counterparts in areas of the sky with inferior coverage (let alone have independent spectroscopic confirmation).

The identifications of the CO transitions, along with their MUSE counterparts, are presented in Table 1. We show the spatial extent of the CO emission on top of the HST images in Figure 2. The MUSE spectra for the individual sources are shown in Figures 13–16 and discussed in Appendix A.

Table 1.  ASPECS-LP CO-detected Sources from the Line Search, with MUSE Spectroscopic Counterparts

ID	R.A.	Decl.	νCO	CO trans.	zCO	MUSE ID	zMUSE	Δv
	(J2000)	(J2000)	(GHz)	(${J}_{\mathrm{up}}\to {J}_{\mathrm{low}}$)				(km s−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
3mm.01	03:32:38.54	−27:46:34.6	97.584 ± 0.003	3 → 2	2.5436	35	2.5432	−15.5 ± 41.0
3mm.02	03:32:42.38	−27:47:07.9	99.510 ± 0.005	2 → 1	1.3167	996	1.3172a	73.5 ± 42.7
3mm.03	03:32:41.02	−27:46:31.5	100.131 ± 0.005	3 → 2	2.4534	⋯	⋯	⋯
3mm.04	03:32:34.44	−27:46:59.8	95.501 ± 0.006	2 → 1	1.4140	1117	1.4147	102.9 ± 44.2
3mm.05	03:32:39.76	−27:46:11.5	90.393 ± 0.006	2 → 1	1.5504	1001	1.5509	71.7 ± 44.7
3mm.06	03:32:39.90	−27:47:15.1	110.038 ± 0.005	2 → 1	1.0951	8	1.0955	79.2 ± 42.3
3mm.07	03:32:43.53	−27:46:39.4	93.558 ± 0.008	3 → 2	2.6961	⋯	⋯	⋯
3mm.08	03:32:35.58	−27:46:26.1	96.778 ± 0.002	2 → 1	1.3821	6415	1.3820	−0.1 ± 40.5
3mm.09	03:32:44.03	−27:46:36.0	93.517 ± 0.003	3 → 2	2.6977b	⋯	⋯	⋯
3mm.10	03:32:42.98	−27:46:50.4	113.192 ± 0.009	2 → 1	1.0367	1011	1.0362a	−53.7 ± 46.6
3mm.11	03:32:39.80	−27:46:53.7	109.966 ± 0.003	2 → 1	1.0964	16	1.0965	19.8 ± 40.8
3mm.12	03:32:36.21	−27:46:27.7	96.757 ± 0.004	3 → 2	2.5739	1124c	2.5739a	16.8 ± 41.9
3mm.13	03:32:35.56	−27:47:04.3	100.209 ± 0.006	4 → 3	3.6008	⋯	⋯	⋯
3mm.14	03:32:34.84	−27:46:40.7	109.877 ± 0.009	2 → 1	1.0981	924	1.0981	15.0 ± 46.9
3mm.15	03:32:36.48	−27:46:31.9	109.971 ± 0.005	2 → 1	1.0964	6870	1.0979	240.4 ± 42.3
3mm.16	03:32:39.92	−27:46:07.4	100.503 ± 0.004	2 → 1	1.2938	925	1.2942	66.3 ± 41.7

Notes. The CO frequencies are taken from Gonzalez-Lopez et al. (2019). (1) ASPECS-LP 3mm ID. (2)–(3) Coordinates. (4) CO line frequency. (5) Identified CO transition (Section 3.1). (6) CO redshift. (7) MUSE ID. (8) MUSE redshift. (9) Velocity offset between MUSE and ALMA (${\rm{\Delta }}v=({z}_{{\text{}}\mathrm{MUSE}}-{z}_{\mathrm{CO}})/(1+{z}_{\mathrm{CO}});$ after converting both to the same reference frame).

^aUpdated from Inami et al. (2017), see Appendix A. ^bAdditionally supported by matching absorption found in MUSE#6941, at z = 2.695, 07 to the north. ^cAdditional redshift for MUSE#1124, which is cataloged as the foreground $[{\rm{O}}\,{\rm{II}}]$-emitter at z = 1.098 (see Figure 18).

Download table as:  ASCIITypeset image

Table 2.  ASPECS-LP CO(2–1) Detected Sources Based On a Spectroscopic Redshift Prior from MUSE

ID	R.A.	Decl.	νCO	CO trans.	zCO	MUSE ID	zMUSE	Δv
	(J2000)	(J2000)	(GHz)	(${J}_{\mathrm{up}}\to {J}_{\mathrm{low}}$)				(km s−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
MP.3mm.01	03:32:37.30	−27:45:57.8	109.978 ± 0.011	2 → 1	1.0962	985	1.0959	−28.2 ± 50.6
MP.3mm.02	03:32:35.48	−27:46:26.5	110.456 ± 0.007	2 → 1	1.0872	879	1.0874	55.8 ± 44.3

Note. (1) ASPECS-LP MUSE prior (MP) ID. (2)–(3) Coordinates. (4) CO line frequency. (5) CO transition. (6) CO redshift. (7) MUSE ID. (8) MUSE redshift. (9) Velocity offset between MUSE and ALMA (${\rm{\Delta }}v=({z}_{{MUSE}}-{z}_{\mathrm{CO}})/(1+{z}_{\mathrm{CO}});$ after converting both to the same reference frame).

Download table as:  ASCIITypeset image

The CO-line detections from Gonzalez-Lopez et al. (2019) are selected to have the highest fidelity and are therefore the highest S/N (≥6.4) candidates over the ASPECS-LP area. In Figure 3, we plot the stellar mass–SFR relation for all MUSE sources at 1.01 < z < 1.74, where we indicate all the galaxies that have been detected in CO(2–1) in the line search.³² There are several galaxies in the field with properties similar to the ASPECS-LP galaxies that are not detected in the line search. This raises the following question: why are these galaxies not detected? Given their physical properties, we may expect some of these galaxies to harbor molecular gas and therefore to have CO signal in the ASPECS-LP cube. The reason that we did not detect these sources in the line search may, therefore, simply be due to the fact that they are present at lower S/N, which puts them in the regime where the decreasing fidelity makes it challenging to identify them among the spurious sources.

Zoom In Zoom Out Reset image size

However, the physical properties of the galaxies themselves provide an extra piece of information that can guide us in detecting CO for these sources. In particular, we can use the spectroscopic redshifts from MUSE to obtain a measurement of the CO flux for each source, either identifying them at lower S/N, or putting an upper limit on their molecular gas mass. We aim at the CO transitions covered at z < 2.9, where the features in the MUSE spectrum typically provide a systemic redshift. At higher redshift the main spectral feature used to identify redshifts is often $\mathrm{Ly}\alpha$, which can be offset from the systemic redshift by a few hundred km s⁻¹ (e.g., Shapley et al. 2003; Rakic et al. 2011; Verhamme et al. 2018).

We extract a single-pixel spectrum from the 3'' tapered cube at the position of each MUSE source in the redshift range, after correcting for the astrometric offset (Section 2.1). We then fit the lines with a Gaussian curve, using a custom-made Bayesian Markov chain Monte Carlo routine with the following priors:

• 1.  
  line peak velocity: a Gaussian distribution centered at Δv = 0 (based on the MUSE redshift) and σ = 100 km s⁻¹ (the MUSE spectral resolution).
• 2.  
  line width: a Maxwellian distribution with a width of 100 km s⁻¹.
• 3.  
  line flux: a Gaussian distribution centered at zero, with σ = 0.5 Jy km s⁻¹, allowing both positive and negative line fluxes to be fitted.

We choose a strong prior on the velocity difference, as we only search for lines at the exact MUSE redshift. The Gaussian prior on the line flux is important to estimate the fidelity of our measurements, allowing an unbiased comparison of positive versus negative line fluxes (see Gonzalez-Lopez et al. 2019 for details). The Maxwellian prior is chosen because it is bound to produce positive values of the line width, depends on a single scale parameter, and has a non-null tail at very large line widths. The uncertainties are computed from the 16th and 84th percentiles of the posterior distributions of each parameter.

As narrow lines are more easily caused by noise in the cube (Gonzalez-Lopez et al. 2019), we rerun the fit with a broader prior on the line width of 200 km s⁻¹. We also independently fit the spectrum with a uniform prior over ±1 GHz around the MUSE redshift. We select only the sources in which the same feature was recovered with S/N > 3 in all three fits. In order to select a sample that is as pure as possible, we select only the objects that have a velocity offset of <80 km s⁻¹ from the MUSE systemic redshift (≈×2 the typical uncertainty on the MUSE redshift). In addition, we only keep objects with a line width of >100 km s⁻¹, to avoid including spurious narrow lines. We note that, while these cuts potentially remove other sources that are detected at lower S/N, we do not attempt to be complete. Rather, we aim to have the prior-based sample as clean as possible.

The prior-based search reveals two additional sources detected in CO(2–1) with an S/N > 3 (see Table 2). Both sources lie within the area in which the sensitivity is >40% of the primary beam peak sensitivity. We show the HST cutouts with the CO spectra of these sources in Figure 4, ordered by S/N. ASPECS-LP-MP.3mm.02 is the foreground spiral galaxy of ASPECS-LP.3mm.08. This source was already found in the ASPECS-Pilot (Decarli et al. 2016, see Appendix A).

Zoom In Zoom Out Reset image size

Because the molecular gas mass is to first order correlated with the SFR, we expect to detect CO in the galaxies with the highest SFRs at a given redshift. Sorting all the galaxies by their SFR indeed reveals a clear correlation between the SFR and the S/N in CO, suggesting there are additional sources in the ASPECS-LP data cube at lower S/N. This can also be clearly seen from Figure 3, where our stringent sample of prior-based sources all lie at log SFR[${M}_{\odot }$ yr⁻¹] > 0.5. Qualitatively, it becomes clear that the ASPECS-LP is sensitive enough to detect molecular gas in most massive MS galaxies at 1.01 < z < 1.74 (a quantitative discussion of the detection fraction for the full sample is provided in Section 6). For many galaxies, the reason these are not unveiled in the line search may simply be because their lower CO luminosity and/or smaller line width puts them below the conservative S/N threshold we adopt in the line search. Using the MUSE redshifts as prior information, it is possible to unveil their molecular gas reservoirs at lower S/N.

The full ASPECS-LP CO line sample consists of 18 galaxies with a CO detection in the HUDF; 16 detections without preselection and 2 MUSE redshift prior-based detections. These galaxies span a range of redshifts between 1 < z < 4. The lowest redshift galaxy is detected in CO(2–1) at z = 1.04, while the highest redshift galaxy is detected (without prior) in CO(4–3) at z = 3.60. We show a histogram of the redshifts of the line-search and prior-based detections in Figure 5.

Zoom In Zoom Out Reset image size

Twelve sources are detected in CO(2–1) at 1.01 < z < 1.74, where the combination of molecular line sensitivity and survey volume are optimal. Most prominently, we detect five galaxies at the same redshift of z ≈ 1.1. These galaxies are all part of an overdensity of galaxies in the HUDF at z = 1.096, visible in Figure 1.

Five sources are detected in CO(3–2) at 2.01 < z < 3.11, including the brightest CO emitter in the field at z = 2.54 (ASPECS-LP.3mm.01; see also Decarli et al. 2016) and a pair of galaxies (ASPECS-LP.3mm.07 and #9) at z ≈ 2.697 (see Section 3.1). All five CO(3–2) sources are detected in 1 mm dust continuum (Aravena et al. 2016; Dunlop et al. 2017) with flux densities below 1 mJy. However, only one of these sources (ASPECS-LP.3mm.01) previously had a spectroscopic redshift (Walter et al. 2016; Inami et al. 2017).

For all the CO-detected sources, we derive the SFR (and M_* and A_V) from the UV-FIR data (including 24–160 μm and ASPECS-LP 1.2 and 3.0 mm) using Magphys (see Section 2.3), which are provided in Table 3. The full SED fits are shown in Figure 19.

Table 3.  Physical Properties of the ASPECS-LP Sources from the Line Search and the MUSE Prior-based Search, with Formal Uncertainties

ID	z	$\mathrm{log}{M}_{* ,\mathrm{SED}}$	SFRSED	${A}_{V,\mathrm{SED}}$	X-ray	XID
		(${M}_{\odot }$)	(${M}_{\odot }$ yr−1)	(mag)
(1)	(2)	(3)	(4)	(5)	(6)	(7)
ASPECS-LP.3mm.01	2.5436	${10.4}_{-0.0}^{+0.0}$	${233}_{-0}^{+0}$	${2.7}_{-0.0}^{+0.0}$	AGN	718
ASPECS-LP.3mm.02	1.3167	${11.2}_{-0.0}^{+0.0}$	${11}_{-0}^{+2}$	${1.7}_{-0.0}^{+0.1}$
ASPECS-LP.3mm.03	2.4534	${10.7}_{-0.1}^{+0.1}$	${68}_{-20}^{+19}$	${3.1}_{-0.3}^{+0.1}$
ASPECS-LP.3mm.04	1.4140	${11.3}_{-0.0}^{+0.0}$	${61}_{-12}^{+3}$	${2.9}_{-0.0}^{+0.1}$
ASPECS-LP.3mm.05	1.5504	${11.5}_{-0.0}^{+0.0}$	${62}_{-19}^{+5}$	${2.3}_{-0.3}^{+0.1}$	AGN	748
ASPECS-LP.3mm.06	1.0951	${10.6}_{-0.0}^{+0.0}$	${34}_{-0}^{+0}$	${0.8}_{-0.0}^{+0.0}$	X	749
ASPECS-LP.3mm.07	2.6961	${11.1}_{-0.1}^{+0.1}$	${187}_{-16}^{+35}$	${3.2}_{-0.1}^{+0.1}$
ASPECS-LP.3mm.08	1.3821	${10.7}_{-0.0}^{+0.0}$	${35}_{-5}^{+8}$	${0.9}_{-0.1}^{+0.1}$
ASPECS-LP.3mm.09	2.6977	${11.1}_{-0.0}^{+0.1}$	${318}_{-35}^{+35}$	${3.6}_{-0.1}^{+0.1}$	AGN	805
ASPECS-LP.3mm.10	1.0367	${11.1}_{-0.1}^{+0.0}$	${18}_{-1}^{+1}$	${3.0}_{-0.1}^{+0.0}$
ASPECS-LP.3mm.11	1.0964	${10.2}_{-0.0}^{+0.0}$	${10}_{-1}^{+0}$	${0.8}_{-0.1}^{+0.0}$
ASPECS-LP.3mm.12	2.5739	${10.6}_{-0.1}^{+0.0}$	${31}_{-3}^{+18}$	${0.8}_{-0.1}^{+0.2}$	AGN	680
ASPECS-LP.3mm.13	3.6008	${9.8}_{-0.1}^{+0.1}$	${41}_{-9}^{+15}$	${1.4}_{-0.2}^{+0.3}$
ASPECS-LP.3mm.14	1.0981	${10.6}_{-0.1}^{+0.1}$	${27}_{-4}^{+1}$	${1.6}_{-0.2}^{+0.0}$
ASPECS-LP.3mm.15	1.0964	${9.7}_{-0.0}^{+0.3}$	${62}_{-4}^{+0}$	${2.9}_{-0.0}^{+0.0}$	AGN	689
ASPECS-LP.3mm.16	1.2938	${10.3}_{-0.0}^{+0.1}$	${11}_{-3}^{+1}$	${0.5}_{-0.2}^{+0.1}$
ASPECS-LP-MP.3mm.01	1.0959	${10.1}_{-0.0}^{+0.1}$	${8}_{-2}^{+3}$	${1.3}_{-0.2}^{+0.2}$
ASPECS-LP-MP.3mm.02	1.0874	${10.4}_{-0.0}^{+0.0}$	${25}_{-0}^{+0}$	${1.0}_{-0.0}^{+0.0}$	X	661

Note. (1) ASPECS-LP ID number. (2) Source redshift. (3) Stellar mass (M_*). (4) Star formation rate (SFR). (5) Visual attenuation (A_V). (6)–(7) X-ray classification as active galactic nucleus (AGN) or other X-ray source (X) from Luo et al. (2017) and corresponding X-ray ID (XID).

Download table as:  ASCIITypeset image

For the 1 < z < 1.5 subsample, we have access to the $[{\rm{O}}\,{\rm{II}}]\lambda 3726,3729$-doublet. We derive SFRs from $[{\rm{O}}\,{\rm{II}}]\lambda 3726,3729$ following Kewley et al. (2004), adopting a Chabrier (2003) IMF. The observed $[{\rm{O}}\,{\rm{II}}]$ luminosity gives a measurement of the unobscured SFR, which can be compared to the total SFR (including the FIR) to derive the fraction of obscured star formation. For that reason, we do not apply a dust correction when calculating the SFR($[{\rm{O}}\,{\rm{II}}]$).

The derived SFR($[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$) is dependent on the oxygen abundance. We have access to the oxygen abundance directly for some of the sources and can also make an estimate through the mass–metallicity relation (e.g., Zahid et al. 2014). However, because of the additional uncertainties in the calibrations for the oxygen abundance, we instead adopt an average $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$/${\rm{H}}\alpha$ ratio of unity, given that all our sources are massive and hence expected to have high oxygen abundance $12+\mathrm{log}\,{\rm{O}}/{\rm{H}}$ ∼ 8.8, where $[{\rm{O}}\,{\rm{II}}]/{\rm{H}}\alpha \,=1.0$ (e.g., Kewley et al. 2004). For all galaxies with S/N($[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$) > 3, excluding the X-ray AGN, the $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726+\lambda 3729$ line flux measurements and SFRs are presented in Table 4.

Table 4.  Emission Line Flux, Unobscured [O II] SFRs, and Metallicities for the ASPECS-LP Line-search and Prior-based Sources at $z\lt 1.5$ with ${\rm{S}}/{\rm{N}}([{\rm{O}}\,{\rm{II}}])\gt 3$

ID	MUSE ID	zMUSE	F[O ii] λ3726+λ3729	F[Ne iii] λ3869	${\mathrm{SFR}}_{[{\rm{O}}\,{\rm{II}}]}^{\mathrm{no}\,\mathrm{dust}}$	${Z}_{[\mathrm{Ne}{\rm{III}}]/[{\rm{O}}{\rm{II}}],{\rm{M}}08}$
			($\times {10}^{-20}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}$)	($\times {10}^{-20}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}$)	(${M}_{\odot }$ yr−1)	(12 + log(O/H))
(1)	(2)	(3)	(4)	(5)	(6)	(7)
3mm.06	8	1.0955	111.4 ± 1.4	1.9 ± 0.4	3.59 ± 0.05	9.05 ± 0.08
3mm.11	16	1.0965	24.4 ± 0.3	0.9 ± 0.1	0.79 ± 0.01	8.78 ± 0.06
3mm.14	924	1.0981	53.6 ± 1.6	2.4 ± 0.4	1.74 ± 0.05	8.70 ± 0.07
3mm.15	6870	1.0979	13.8 ± 0.4	<0.2 ± 0.1	⋯	⋯
3mm.16	925	1.2942	67.0 ± 4.0	<1.9 ± 0.8	3.26 ± 0.20	>8.79 ± 0.17
MP.3mm.01	985	1.0959	17.8 ± 1.5	<0.6 ± 0.5	0.57 ± 0.05	>8.56 ± 0.29
MP.3mm.02	879	1.0874	245.9 ± 1.1	11.5 ± 0.6	7.78 ± 0.03	8.73 ± 0.02

Notes. (1) ASPECS-LP.3mm ID number. (2) MUSE ID (3) MUSE redshift. (4) $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726+\lambda 3729$ flux (${\rm{S}}/{\rm{N}}\gt 3$). (5) $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$ flux (upper limits are reported if S/N < 3). (6) SFR($[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$) without correction for dust attenuation. (7) Metallicity from $[\mathrm{Ne}\,{\rm{III}}]$/$[{\rm{O}}\,{\rm{II}}]$ based on Maiolino et al. (2008). We do not compute an SFR($[{\rm{O}}\,{\rm{II}}]$) or metallicity for the X-ray detected AGN (3mm.15).

Download table as:  ASCIITypeset image

It is well known that the gas-phase metallicity of galaxies is correlated with their stellar mass, with more massive galaxies having higher metallicities on average (e.g., Tremonti et al. 2004; Maiolino et al. 2008; Mannucci et al. 2010; Zahid et al. 2014). For the 1.0 < z < 1.42 subsample, we have access to $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$, which allows us to derive a metallicity from $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$/$[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$. We follow the relation as presented by Maiolino et al. (2008), who calibrated the $[\mathrm{Ne}\,{\rm{III}}]$/$[{\rm{O}}\,{\rm{II}}]$ line ratio against metallicities inferred from the direct T_e method (at low metallicity; $12+\mathrm{log}\,{\rm{O}}/{\rm{H}}\lt 8.35$) and theoretical models from Kewley & Dopita (2002; at high metallicity, mainly relevant for this paper; $12+\mathrm{log}\,{\rm{O}}/{\rm{H}}\,\gt 8.35$). Since the wavelengths of $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$ and $[{\rm{O}}\,{\rm{II}}]\lambda 3726,3729$ are close, this ratio is practically insensitive to dust attenuation. The physical underpinning lies in the fact that the ratio of the low-ionization $[{\rm{O}}\,{\rm{II}}]$ and high-ionization $[\mathrm{Ne}\,{\rm{III}}]$ lines is a solid tracer of the shape of the ionization field, given that neon closely tracks the oxygen abundance (e.g., Ali et al. 1991; Levesque & Richardson 2014; Feltre et al. 2018). As the ionization parameter decreases with increasing stellar metallicity (Dopita et al. 2006a, 2006b) and the metallicity of the young ionizing stars and their birth clouds is correlated, the ratio of $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$/$[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$ is a reasonable gas-phase metallicity diagnostic, albeit indirect, with significant scatter (Nagao et al. 2006; Maiolino et al. 2008) and sensitive to model assumptions (e.g., Levesque & Richardson 2014). If an AGN contributes significantly to the ionizing spectrum, the emission lines may no longer only trace the properties associated with massive star formation. For this reason, we exclude the sources with an X-ray AGN from the analysis of the metallicity.

We report the $[\mathrm{Ne}\,{\rm{III}}]$ flux measurements and $[\mathrm{Ne}\,{\rm{III}}]$/$[{\rm{O}}\,{\rm{II}}]$ metallicities in Table 4. The solar metallicity is $12+\mathrm{log}\,{\rm{O}}/{\rm{H}}\,=8.76\pm 0.07$ (Caffau et al. 2011).

The derivation of the molecular gas properties of our sources is detailed in Aravena et al. (2019). For reference, we provide a brief summary here.

We convert the observed CO($J\to J-1$) flux to a molecular gas mass (${M}_{\mathrm{mol}}$) using the relations from Carilli & Walter (2013). To convert the higher-order CO transitions to CO(1–0), we need to know the excitation dependent intensity ratio between the CO lines, r_J1. We use the excitation ladder as estimated by Daddi et al. (2015) for galaxies on the MS, where r₂₁ = 0.76 ± 0.09, r₃₁ = 0.42 ± 0.07, and r₄₁ = 0.31 ± 0.06 (see also Decarli et al. 2016). To subsequently convert the CO(1–0) luminosity to ${M}_{\mathrm{mol}}$, we use an ${\alpha }_{\mathrm{CO}}=3.6$ ${M}_{\odot }$ (K km s⁻¹ pc²)⁻¹, appropriate for star-forming galaxies (Daddi et al. 2010; see Bolatto et al. 2013 for a review). This choice of ${\alpha }_{\mathrm{CO}}$ is supported by our finding that the ASPECS-LP sources are mostly on the MS and have (near-)solar metallicity (see Section 5.4).

With these conversions in mind, the molecular gas mass and derived quantities we report here can easily be rescaled to different assumptions following: ${M}_{\mathrm{mol}}/{M}_{\odot }\,=\left({\alpha }_{\mathrm{CO}}/{r}_{J1}\right){L}_{\mathrm{CO}(J\to J-1)}^{{\prime} }/$(K km s⁻¹ pc²). The CO line and derived molecular gas properties are all presented in Table 5.

Table 5.  Molecular Gas Properties of the ASPECS-LP Line-search and Prior-based Sources, with Formal Uncertainties

ID	${z}_{\mathrm{CO}}$	Jup	FWHM	Fline	${L}_{\mathrm{line}}^{{\prime} }$	${L}_{\mathrm{CO}(1\mbox{--}0)}^{{\prime} }$	${M}_{\mathrm{mol}}$	${M}_{\mathrm{mol}}/{M}_{* }$	${t}_{\mathrm{depl}}$
			(km s−1)	(Jy km s−1)	(×109 K km s−1 pc2)		(×1010 ${M}_{\odot }$)		(Gyr)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
3mm.01	2.5436	3	517 ± 21	1.02 ± 0.04	33.9 ± 1.3	80.8 ± 13.8	29.1 ± 5.0	12.1 ± 2.1	1.2 ± 0.2
3mm.02	1.3167	2	277 ± 26	0.47 ± 0.04	10.7 ± 0.9	14.1 ± 2.1	5.1 ± 0.7	0.3 ± 0.1	4.5 ± 0.8
3mm.03	2.4534	3	368 ± 37	0.41 ± 0.04	12.8 ± 1.3	30.5 ± 5.9	11.0 ± 2.1	2.2 ± 0.6	1.6 ± 0.6
3mm.04	1.4140	2	498 ± 47	0.89 ± 0.07	23.2 ± 1.8	30.5 ± 4.3	11.0 ± 1.6	0.6 ± 0.1	1.8 ± 0.3
3mm.05	1.5504	2	617 ± 58	0.66 ± 0.06	20.4 ± 1.9	26.9 ± 4.0	9.7 ± 1.4	0.3 ± 0.1	1.6 ± 0.4
3mm.06	1.0951	2	307 ± 33	0.48 ± 0.06	7.7 ± 1.0	10.1 ± 1.7	3.6 ± 0.6	1.0 ± 0.2	1.1 ± 0.2
3mm.07	2.6961	3	609 ± 73	0.76 ± 0.09	27.9 ± 3.3	66.5 ± 13.6	23.9 ± 4.9	2.0 ± 0.5	1.3 ± 0.3
3mm.08	1.3821	2	50 ± 8	0.16 ± 0.03	4.0 ± 0.7	5.3 ± 1.2	1.9 ± 0.4	0.4 ± 0.1	0.5 ± 0.2
3mm.09	2.6977	3	174 ± 17	0.40 ± 0.04	14.7 ± 1.5	35.0 ± 6.8	12.6 ± 2.5	1.0 ± 0.2	0.4 ± 0.1
3mm.10	1.0367	2	460 ± 49	0.59 ± 0.07	8.5 ± 1.0	11.1 ± 1.9	4.0 ± 0.7	0.3 ± 0.1	2.2 ± 0.4
3mm.11	1.0964	2	40 ± 12	0.16 ± 0.03	2.6 ± 0.5	3.4 ± 0.7	1.2 ± 0.3	0.8 ± 0.2	1.2 ± 0.3
3mm.12	2.5739	3	251 ± 40	0.14 ± 0.02	4.8 ± 0.7	11.3 ± 2.5	4.1 ± 0.9	0.9 ± 0.2	1.3 ± 0.5
3mm.13	3.6008	4	360 ± 49	0.13 ± 0.02	4.3 ± 0.7	13.9 ± 3.4	5.0 ± 1.2	8.8 ± 2.8	1.2 ± 0.5
3mm.14	1.0981	2	355 ± 52	0.35 ± 0.05	5.6 ± 0.8	7.4 ± 1.4	2.7 ± 0.5	0.7 ± 0.1	1.0 ± 0.2
3mm.15	1.0964	2	260 ± 39	0.21 ± 0.03	3.4 ± 0.5	4.4 ± 0.8	1.6 ± 0.3	3.2 ± 1.1	0.3 ± 0.1
3mm.16	1.2938	2	125 ± 28	0.08 ± 0.01	1.8 ± 0.2	2.3 ± 0.4	0.8 ± 0.1	0.4 ± 0.1	0.7 ± 0.2
MP.3mm.01	1.0962	2	169 ± 21	0.13 ± 0.03	2.1 ± 0.5	2.8 ± 0.7	1.0 ± 0.2	0.7 ± 0.2	1.3 ± 0.5
MP.3mm.02	1.0872	2	107 ± 30	0.10 ± 0.03	1.6 ± 0.4	2.0 ± 0.6	0.7 ± 0.2	0.3 ± 0.1	0.3 ± 0.1

Note. The CO full width at half maximum (FWHM) and line fluxes are taken from Gonzalez-Lopez et al. (2019). (1) ASPECS-LP ID number. (2) CO redshift. (3) Upper level of CO transition. (4) CO line FWHM. (5) Integrated line flux. (6) Line luminosity. (7) CO(1–0) line luminosity assuming Daddi et al. (2015) excitation (Section 4.3). (8) Molecular gas mass assuming ${\alpha }_{\mathrm{CO}}=3.6$ K (km s⁻¹ pc²)⁻¹. (9) Molecular-to-stellar mass ratio, ${M}_{\mathrm{mol}}/{M}_{* }$. (10) Depletion time, ${t}_{\mathrm{depl}}={M}_{\mathrm{mol}}/\mathrm{SFR}$.

Download table as:  ASCIITypeset image

The majority of the detections consist of CO(2–1) and CO(3–2), at 1 < z < 2 and 2 < z < 3, respectively. A key question is in what part of the galaxy population we detect the largest gas-reservoirs at these redshifts.

We show histograms of the stellar masses and SFRs for the sources detected in CO(2–1) and CO(3–2) in Figure 6. We compare these to the distribution of all galaxies in the field that have a spectroscopic redshift from MUSE and our extended (photometric) catalog of all other galaxies. In the top part of each panel we show the percentage of galaxies we detect in ASPECS, compared to the number of galaxies in reference catalogs.

Zoom In Zoom Out Reset image size

We focus first on the SFRs, shown in the right panels of Figure 6. The galaxies in which we detect molecular gas are the galaxies with the highest SFRs and the detection fraction increases with SFR. This is expected as molecular gas is a prerequisite for star formation and the most highly star-forming galaxies are thought to host the most massive gas reservoirs. The detections from the line search at 1.0 < z < 1.7 alone account for ≈40% of the galaxy population at $10\,\lt \mathrm{SFR}[{M}_{\odot }\,{\mathrm{yr}}^{-1}]\,\lt 30$, increasing to >75% at $\mathrm{SFR}\,\gt 30\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$. Including the prior-based detections, we find 60% of the population at SFR ≈ 20 ${M}_{\odot }$ yr⁻¹. Similarly, at 2.0 < z < 3.1, the detection fraction is highest in the most highly star-forming bin. Notably, however, with ASPECS-LP we probe molecular gas in galaxies down to much lower SFRs as well. The sources span over two orders of magnitude in SFR, from ≈5 to >500 ${M}_{\odot }$ yr⁻¹.

The stellar masses of the ASPECS-LP detections in CO(2–1) and CO(3–2) are shown in the left panels of Figure 6. We detect molecular gas in galaxies spanning over two orders of magnitude in stellar mass, down to $\mathrm{log}\,{M}_{* }[{M}_{\odot }]\sim 9.5$. The completeness increases with stellar mass, which is presumably a consequence of the fact that more massive star-forming galaxies also have a larger gas fraction and higher SFR. At M_* > 10¹⁰ ${M}_{\odot }$, we are ≈40% complete at 1.0 < z < 1.7, while we are ≈50% complete at M_* > 10^10.5 ${M}_{\odot }$ at 2 < z <3.1. The full distribution includes both star-forming and passive galaxies, which would explain why we do not pick-up all galaxies at the highest stellar masses.

From the deepest X-ray data over the field we identify five AGN in the ASPECS-LP line search sample (see Table 3). Two of these are detected in CO(2–1); namely, ASPECS-LP.3mm.05 and ASPECS-LP.3mm.15. The remaining three X-ray AGN are ASPECS-LP.3mm.01, 3mm.09, and 3mm.12, detected in CO(3–2). The AGN fraction among the ASPECS-LP sources is thus 2/10 = 20% at 1.0 < z < 1.7 and 3/5 =60% at 2.0 < z < 3.1 (note that including the MUSE-prior sources decreases the AGN fraction). If we consider the total number of X-ray AGN over the field, we detect 2/6 = 30% of the X-ray AGN at 1.0 < z < 1.7 and 3/6 = 50% at 2.0 < z < 3.1, without preselection.

The comoving number density of AGN increases out to z ≈ 2–3 (Hopkins et al. 2007). Using a volume limited sample out to z ∼ 0.7 based on the Sloan Digital Sky Survey and Chandra, Haggard et al. (2010) showed that the AGN fraction increases with both stellar mass and redshift, from a few percent at M_* ∼ 10^10.7 ${M}_{\odot }$, up to 20% in their most massive bin (M_* ∼ 10^11.8 ${M}_{\odot }$). Closer in redshift to the ASPECS-LP sample, Wang et al. (2017) investigated the fraction of X-ray AGN in the GOODS fields and found that among massive galaxies, M_* >10^10.6 ${M}_{\odot }$, 5%–15% and 15%–50% host an X-ray AGN at 0.5 < z < 1.5 and 1.5 < z < 2.5, respectively. The AGN fractions found in ASPECS-LP are broadly consistent with these ranges given the limited numbers and considerable Poisson error.

Given the AGN fraction among the ASPECS-LP sources (20% at z ∼ 1.4 and 60% at z ∼ 2.6), the question arises of whether we detect the galaxies in CO because they are AGN (i.e., AGN-powered), or, whether we detect a population of galaxies that hosts a larger fraction of AGN (e.g., because the higher gas content fuels both the AGN and star formation)? The CO ladders in, e.g., quasar host galaxies can be significantly excited, leading to an increased luminosity in the high-J CO transitions compared to star-forming galaxies at lower excitation (see, e.g., Carilli & Walter 2013; Rosenberg et al. 2015). With the band 3 data we are sensitive to the lower-J transitions, decreasing the magnitude of such a bias toward AGN. At the same time, the ASPECS-LP is sensitive to the galaxies with the largest molecular gas reservoirs, which are typically the galaxies with the highest stellar masses and/or SFRs. As AGN are more common in massive galaxies, it is natural to find a moderate fraction of AGN in the sample, increasing with redshift. Once the ASPECS-LP is complete with the observations of the band 6 (1 mm) data, we can investigate the higher J CO transitions for these sources and possibly test whether the CO is powered by AGN activity.

We investigate the fraction of dust-obscured star formation by comparing the SFR derived from the $[{\rm{O}}\,{\rm{II}}]\ \lambda 3726,3729$ emission line, without dust correction, with the (independent) total SFR from modeling the UV-FIR SED with magphys. We show the ratio between the SFR($[{\rm{O}}\,{\rm{II}}]$) and the total SFR(SED) as a function of the total SFR in Figure 7. We use the observed (unobscured) $[{\rm{O}}\,{\rm{II}}]$ luminosity, yielding a measurement of the fraction of unobscured SFR. Immediately evident is the fact that more highly star-forming galaxies (which are on average more massive) are more strongly obscured. The median ratio (bootstrapped errors) of obscured/unobscured SFR is ${10.8}_{-5.1}^{+3.0}$ for the ASPECS-LP sources from the line search, which have a median mass of ${10}^{10.6}$ ${M}_{\odot }$ (see ${2.2}_{-0.1}^{+0.2}$ for the complete sample of MUSE galaxies, with a median mass of ${10}^{9}$ ${M}_{\odot }$). Including the objects from the prior-based search does not significantly affect this fraction (${10.8}_{-5.1}^{+2.3}$, at a median mass of ${M}_{* }\sim {10}^{10.6}\,{M}_{\odot }$).

Zoom In Zoom Out Reset image size

The molecular gas conversion factor is dependent on the metallicity, which is therefore an important quantity to constrain. Specifically, ${\alpha }_{\mathrm{CO}}$ can be higher in galaxies with significantly subsolar metallicities, where a large fraction of the molecular gas may be CO faint, or lower in (luminous) starburst galaxies, where CO emission originates in a more highly excited molecular medium (e.g., Bolatto et al. 2013).

Given that the majority of the ASPECS-LP sources are reasonably massive, M_* ≥ 10¹⁰ ${M}_{\odot }$, their metallicities are likely to be (super-)solar, based on the mass–metallicity relation (e.g., Zahid et al. 2014).

For the ASPECS-LP sources at 1.0 < z < 1.42, the MUSE coverage includes $[\mathrm{Ne}\,{\rm{III}}]\ \lambda 3869$, which can be used as a metallicity indicator (Section 4.2). We infer a metallicity for ASPECS-LP.3mm.06, 3mm.11, and 3mm.14 and ASPECS-LP-MP.3mm.02. In addition, we can provide a lower limit on the metallicity for ASPECS-LP.3mm.16 and ASPECS-LP-MP.3mm.01, based on the upper limit on the flux of $[\mathrm{Ne}\,{\rm{III}}]$.

In Figure 8, we show the ASPECS-LP sources on the stellar mass—gas-phase metallicity plane. For reference, we show the mass–metallicity relation from Maiolino et al. (2008; that matches the $[\mathrm{Ne}\,{\rm{III}}]$/$[{\rm{O}}\,{\rm{II}}]$ calibration) and Zahid et al. (2014), both converted to the same IMF and metallicity scale (Kewley & Ellison 2008). The AGN-free ASPECS-LP sources span about half a dex in metallicity. They are all metal-rich and consistent with a solar or super-solar metallicity, in line with the expectations from the mass–metallicity relation.

Zoom In Zoom Out Reset image size

The (near-)solar metallicity of our targets supports our choice of ${\alpha }_{\mathrm{CO}}$ = 3.6 ${M}_{\odot }$ (K km s⁻¹ pc²)⁻¹, which was derived for z ≈ 1.5 star-forming galaxies (Daddi et al. 2010) and is similar to the Galactic ${\alpha }_{\mathrm{CO}}$ (see Bolatto et al. 2013).

Being a flux-limited survey, the limiting molecular gas mass of the ASPECS-LP, ${M}_{\mathrm{mol}}(z)$, increases with redshift. Based on the measured flux limit of the survey, we can gain insight into what masses of gas we are sensitive to at different redshifts. The sensitivity of the ASPECS-LP Band 3 data itself is presented and discussed in Gonzalez-Lopez et al. (2019): it is relatively constant across the frequency range, being deepest in the center where the different spectral tunings overlap.

Assuming a CO line full width at half maximum (FWHM) and an ${\alpha }_{\mathrm{CO}}$ and excitation ladder as in Section 4.3, we can convert the root-mean-square noise level of ASPECS-LP in each channel to a sensitivity limit on ${M}_{\mathrm{mol}}(z)$. The result of this is shown in Figure 9. With increasing luminosity distance, ASPECS-LP is sensitive to more massive reservoirs. This is partially compensated by the fact that the first few higher-order transitions are generally more luminous at the typical excitation conditions in star-forming galaxies. The ${M}_{\mathrm{mol}}(z)$ function has a strong dependence on the FWHM, as broader lines at the same total flux are harder to detect (see also Gonzalez-Lopez et al. 2019). As the FWHM is related to the dynamical mass of the system, and we are sensitive to more massive systems at higher redshifts, these effects will conspire in further pushing up the gas-mass limit to more massive reservoirs.

Zoom In Zoom Out Reset image size

At 1.0 < z < 1.7, the lowest gas mass we can detect at 5σ (using the above assumptions and an FWHM for CO(2–1) of 100 km s⁻¹) is ${M}_{\mathrm{mol}}\sim {10}^{9.5}$ ${M}_{\odot }$, with a median limiting gas mass over the entire redshift range of ${M}_{\mathrm{mol}}\geqslant {10}^{9.7}$ ${M}_{\odot }$ (${M}_{\mathrm{mol}}\geqslant {10}^{9.9}$ ${M}_{\odot }$ at FWHM = 300 km s⁻¹). At 2.0 < z <3.1 the median sensitivity increases to ${M}_{\mathrm{mol}}\gtrsim {10}^{10.3}$ ${M}_{\odot }$, assuming an FWHM of 300 km s⁻¹ for CO(3–2). In reality the assumptions made above can vary significantly for individual galaxies, depending on the physical conditions of their ISM.

As cold molecular gas precedes star formation, the ${M}_{\mathrm{mol}}(z)$ selection function of ASPECS-LP can, to first order, be viewed as an SFR(z) selection function. Since more massive star-forming galaxies have higher SFRs (albeit with significant scatter), a weaker correlation with stellar mass may also be expected. These rough, limiting relations will provide useful context to understand what galaxies we detect with ASPECS.

We show the ASPECS-LP sources in the stellar mass–SFR plane at 1.01 < z < 1.74 and 2.01 < z < 3.11 in Figure 10. On average, star-forming galaxies with a higher stellar mass have a higher SFR, with the overall SFR increasing with redshift for a given mass, a relation usually denoted as the galaxy MS. We show the MS relations from Whitaker et al. (2014, W14) and Schreiber et al. (2015, S15) at the average redshift of the sample. The typical intrinsic scatter in the MS at the more massive end is around 0.3 dex or a factor of 2 (Speagle et al. 2014), which we can use to discern whether galaxies are on, above, or below the MS at a given mass.

Zoom In Zoom Out Reset image size

6.2.1. Systematic Offsets in the MS

6.2.2. Normal Galaxies at 1.01 < z < 1.74

The ASPECS-LP sources at $1.01\lt z\lt 1.74$ are shown in the (M_*, SFR)-plane in the left panel of Figure 10. For comparison we show all sources in this redshift range with a secure spectroscopic redshift, as MUSE is mostly complete for massive, star-forming galaxies in the regime of the ASPECS-LP detections at these redshifts (see Figure 6).

At the depth of the ASPECS-LP, we are sensitive enough to probe molecular gas reservoirs in a variety of galaxies that lie on and even below the MS at z ∼ 1.4. Most of the ASPECS-LP galaxies detected in this redshift range lie on the MS, spanning a mass range of ∼2 decades at the massive end. These galaxies belong to the population of normal star-forming galaxies at these redshifts.

As expected, with the primary sample alone we detect essentially all massive galaxies that lie above the MS, for M_* > 10^9.5 ${M}_{\odot }$. The lowest mass galaxy we detect is ASPECS-LP.3mm.15, which is elevated significantly above the MS and is also an X-ray classified AGN. One galaxy, with the highest SFR of all, is a notable outlier for not being detected: MUSE#872 (M_* = 10^10.2 ${M}_{\odot }$, SFR ∼ 150 ${M}_{\odot }$ yr⁻¹). From the prior-based search we find that no molecular gas emission is seen in this source at lower levels either. While the nondetection of this object is very interesting, we caution that this source is also a broad-line AGN in MUSE and it is possible that its SFR is overestimated.

Notably, we also detect a number of galaxies that lie significantly below the MS (e.g., ASPECS-LP.3mm.02), meaning they have SFRs well below the population average. Despite their low SFR, these sources host a significant gas reservoir and have a gas fraction that is in some cases similar to MS galaxies at their stellar mass. The detection of a significant molecular gas reservoir in these sources is interesting, as these sources would typically not be selected in targeted observations for molecular gas.

Overall, we detect the majority of the galaxies on the massive end of the MS at 1.0 < z < 1.7 in CO. We show the detection rate in bins of stellar mass and SFR in the left panel of Figure 11. At an SFR > 10 ${M}_{\odot }$ yr⁻¹, we detect >60% of galaxies at all masses at these redshifts. If we focus on galaxies with ${M}_{* }\gt {10}^{10}\,{M}_{\odot }$, we are >60% complete down to log SFR[${M}_{\odot }$ yr⁻¹] > 0.5, where we encompass all MS galaxies.

Zoom In Zoom Out Reset image size

6.2.3. Massive Galaxies at 2.01 < z < 3.11

We now provide a brief discussion of the evolution of the molecular gas properties (and the individual outliers) in the full ASPECS-LP sample of 18 sources, including the MUSE prior-based sources, in the context of the MUSE-derived properties. A more detailed discussion of these results will be provided in Aravena et al. (2019).

From systematic surveys of the galaxy population at z ≈ 0, we know that the molecular gas properties of galaxies vary across the MS (e.g., Saintonge et al. 2016, 2017). The same trends are unveiled in the ASPECS-LP sample out to z ≈ 3. To reveal these trends more clearly, we show the MS plot colored by the depletion time (${t}_{\mathrm{depl}}={M}_{\mathrm{mol}}/\mathrm{SFR}$) and gas fraction (indicated by ${M}_{\mathrm{mol}}/{M}_{* }$) in Figure 12. The molecular gas mass and depletion time of the ASPECS-LP sources vary systematically across the MS. On average, galaxies above the MS have higher gas fractions and shorter depletion times than galaxies on the MS, while the contrary is true for galaxies below the MS (longer depletion times, smaller gas fractions).

Zoom In Zoom Out Reset image size

At z ∼ 1.4, the sources span about an order of magnitude in depletion time, from 0.3 to 5 Gyr, with a median depletion time of $\approx 1.2$ Gyr. This is comparable to the average depletion times found in z = 1–3 star-forming galaxies (e.g., Daddi et al. 2010; Tacconi et al. 2013). ASPECS-LP.3mm.02, which appears to harbor a substantial gas reservoir while its SFR puts it significantly below the MS, has a correspondingly long depletion time of several gigayears. Although the numbers are more limited at higher redshifts, we see a similar variety in depletion times at z ∼ 2.6, with a median depletion time of $\approx 1.3$ Gyr. For galaxies of similar masses we do not find a strong evolution in the depletion time between the z ∼ 1.4 and z ∼ 2.6 bins.

The evolution of the gas fraction across the MS is clearly seen for the sources at z ∼ 1.4. The lowest gas-mass fractions we find are of the order of 30%, while the galaxies with the highest gas fractions have about equal mass in stars and in molecular gas, with a median of ${M}_{\mathrm{mol}}/{M}_{* }\approx 0.6$. These are comparable to the gas fractions found at similar redshifts (Daddi et al. 2010; Tacconi et al. 2013). The gas fractions at z ∼ 2.6 are substantially higher than they are at lower redshift. ASPECS-LP.3mm.09 and 12 have substantial gas fractions close to unity, while both ASPECS-LP.3mm.03 and 07, have a molecular gas mass of about ×2 their mass in stars (median ${M}_{\mathrm{mol}}/{M}_{* }\approx 2.0$). ASPECS-LP.3mm.1, 3mm.13, and 3mm.15 are outliers in this picture, with a substantially higher gas fraction than the other sources. Both 3mm.01 and 3mm.15 are also starbursts with a high inferred SFR and show an X-ray detected AGN. This high SFR is consistent with the high gas fraction and a picture in which the large gas reservoir fuels a strong starburst, while some gas powers the AGN simultaneously. As may be expected given the flux-limited nature of the observations, the highest redshift source, ASPECS-LP.3mm.13, also has a substantial gas fraction (${M}_{\mathrm{mol}}/{M}_{* }\,=8.8\pm 2.8$). As a whole, Figure 12 reveals the strength of the ASPECS-LP probing the molecular gas across cosmic time without preselection.