ALMA + VLT observations of a damped Lyman-α absorbing galaxy: massive, wide CO emission, gas-rich but with very low SFR

Abstract

We are undertaking an Atacama Large Millimeter Array survey of molecular gas in galaxies selected for their strong H i absorption, so-called damped Lyα absorber (DLA)/sub-DLA galaxies. Here, we report CO(2–1) detection from a DLA galaxy at z = 0.716. We also present optical and near-infrared (NIR) spectra of the galaxy revealing [O ii], Hα, and [N ii] emission lines shifted by ∼170 km s⁻¹ relative to the DLA, and providing an oxygen abundance 3.2 times solar, similar to the absorption metallicity. We report low unobscured SFR∼1 M_⊙ yr⁻¹ given the large reservoir of molecular gas, and also modest obscured SFR |$=4.5_{-2.6}^{+4.4}$| M_⊙ yr⁻¹ based on far-IR and sub-millimetre data. We determine mass components of the galaxy: log[M_*/M|$_{\odot } ]= 10.80^{+0.07}_{-0.14}$|⁠, log[M_mol-gas/M_⊙] = 10.37 ± 0.04, and log[M_dust/M|$_{\odot } ]= 8.45^{+0.10}_{-0.30}$|⁠. Surprisingly, this H i absorption-selected galaxy has no equivalent objects in CO surveys of flux-selected samples. The galaxy falls off current scaling relations for the star formation rate (SFR) to molecular gas mass and CO Tully–Fisher relation. Detailed comparison of kinematical components of the absorbing, ionized, and molecular gas, combined with their spatial distribution, suggests that part of the CO gas is both kinematically and spatially decoupled from the main galaxy. It is thus possible that a major starburst in the past could explain the wide CO profile as well as the low SFR. Support for this also comes from the spectral energy distribution favouring an instantaneous burst of age ≈0.5 Gyr. Our survey will establish whether flux-selected surveys of molecular gas are missing a key stage in the evolution of galaxies and their conversion of gas to stars.

1 INTRODUCTION

Since the discovery of the first damped Lyα absorber (DLA) at high redshift (z = 2.31; Lowrance et al. 1972) almost half a century ago, the nature of the objects causing this absorption has been an important open problem in galaxy evolution. A wide range of suggestions, including large galaxy discs, gas-rich dwarf galaxies, circumgalactic gas, tidally stripped gas in merger events, and starburst-driven outflows, have been put forward to explain various properties of the absorbers (e.g. Prochaska & Wolfe 1997; Haehnelt, Steinmetz & Rauch 1998; Nulsen, Barcons & Fabian 1998; Fynbo, Møller & Warren 1999; Schaye 2001). However, no single galaxy class has so far been able to explain all the properties of the known DLA sample. Of course, the absorber sample is likely to be drawn from the full galaxy population at any redshift and will thus certainly contain galaxies of different types.

The main limitation in addressing the question of the nature of the DLA host galaxies has been simply the fact that the bulk of our information on DLAs stems from absorption studies, which trace only a pencil beam through the host galaxy. While such studies have provided detailed information on the H i column density distribution function, the elemental abundances, the gas-phase metallicities and dust depletions, the absorption kinematics, the spin temperatures, and the molecular gas fractions in DLAs at all redshifts (e.g. Prochaska & Wolfe 1997; Pettini et al. 1997; Storrie-Lombardi & Wolfe 2000; Kanekar & Chengalur 2003; Prochaska et al. 2003; Kulkarni et al. 2005; Prochaska et al. 2005; Ledoux et al. 2006; Noterdaeme et al. 2008; Rafelski et al. 2012; Noterdaeme et al. 2012a; Kanekar et al. 2014), they have offered limited insight into the link to emission selected samples. Unfortunately, direct optical imaging of the DLA hosts is difficult due to their proximity to the bright background quasars, especially in case of faint DLA host galaxies. To improve the chance of detection, one can use the damped profile itself as a natural coronagraph (Smith et al. 1989; Møller & Warren 1993), or image below the Lyman limit of a line of sight neighbour absorber (Steidel & Hamilton 1992; Christensen et al. 2009; Fumagalli et al. 2015).

Over the last decade, some progress has been made in understanding the nature of DLA galaxies via the use of correlations, especially in absorption properties. For example, DLAs appear to follow luminosity–metallicity (Møller, Fynbo & Fall 2004), mass–metallicity (Ledoux et al. 2006; Prochaska et al. 2008), and spin temperature–metallicity (Kanekar et al. 2009, 2014) relations. The use of these correlations has helped optimize target selection for imaging studies: specifically, focused observations of high-metallicity DLAs have resulted in a significant increase in the number of identifications of z ≥ 2 DLA hosts in recent years (e.g. Fynbo et al. 2010, 2011, 2013; Péroux et al. 2011, 2012; Noterdaeme et al. 2012b; Bouché et al. 2012; Krogager et al. 2013; Jorgenson & Wolfe 2014; Hartoog et al. 2015; Zafar et al. 2017). This has allowed a global description of metallicity scaling relations in DLA host galaxies, and their evolution with redshift (Møller et al. 2013; Neeleman et al. 2013), especially the mass–metallicity relation in DLAs, whose predictions have recently been verified by independent methods (Christensen et al. 2014). However, it should be recalled that a pre-requisite for an imaging study is that the DLA host be bright enough to be imaged; this effectively limits continuum studies of DLA hosts, similar to the case of a flux-limited sample. Specifically, while the H i selection allows to select a sample including low-luminosity galaxies, the majority of DLA hosts are still undetectable in optical or near-infrared (NIR) imaging studies due to the steep dependence of luminosity on metallicity (Møller et al. 2004; Fynbo et al. 2008).

The pre-selection of the highest metallicity systems for emission searches in many current observing programmes is thus in reality a flux-limiting sample pre-selection. While this has significantly increased the efficiency of those programmes (to a success rate of ≈50 per cent, Krogager et al. 2017), one is restricted to studies of only a small fraction of the population. Imaging studies that do not have such pre-selection have a far lower success rate in imaging the DLA hosts (Charlot & Fall 1991; Lowenthal et al. 1995; Le Brun et al. 1997; Lanzetta et al. 1997; Guillemin & Bergeron 1997; Turnshek et al. 2001; Warren et al. 2001; Møller et al. 2002; Colbert & Malkan 2002; Kanekar et al. 2002; Lacy et al. 2003; Chen & Lanzetta 2003; Rao et al. 2003; Kulkarni et al. 2006; Chun et al. 2006; Straka et al. 2010; Battisti et al. 2012; Fumagalli et al. 2014, 2015; Srianand et al. 2016).

Because high metallicity traces more massive, and therefore larger, galaxies a side effect of metallicity pre-selection is that the median impact parameter of the detected hosts is larger for those samples. This effect was predicted by Møller et al. (2004) and Fynbo et al. (2008) and has been verified observationally by Krogager et al. (2012). Therefore, both long-slit and integral-field-unit observations of high-metallicity DLA hosts may miss the target completely if the field of view is small. For example, Wang, Kanekar & Prochaska (2015) reported Hα non-detections of three hosts out to impact parameters of only ≈12.5 kpc and used this to place limits on their star formation rates (SFRs). However, Krogager et al. (2017) investigated this question for several complete samples containing a large variety of DLA observations, and found that the fraction of non-detections precisely matches that predicted based on metallicity and limited field coverage. Latest, Neeleman et al. (2017) reported [C ii] 158 μm emission at impact parameters 45 and 18 kpc from two DLAs at z ≈ 4, confirming that even at this redshift the impact parameter will affect detection probability.

Thus, while much work remains to be done in characterizing the stellar properties of DLA hosts, it is clear that significant progress has been made in this field over the last decade. The situation is much worse in studies of the neutral gas that fuels this star formation, despite the fact that the neutral gas mass of DLA host galaxies and the gas distribution therein are critical pieces in understanding galaxy evolution. Unfortunately, the weakness of the H i 21-cm line has meant that only a handful of DLAs and sub-DLAs have so far been searched for H i 21-cm emission, and all at low redshifts, z ≲ 0.1 (Bowen et al. 2001; Kanekar et al. 2001; Kanekar & Chengalur 2005; Briggs & Barnes 2006; Mazumdar, Kanekar & Prochaska 2014; Chengalur et al. 2015). So far, only one DLA and one sub-DLA have been detected and mapped in H i 21-cm emission (Bowen et al. 2001; Chengalur & Kanekar 2002; Briggs & Barnes 2006; Chengalur et al. 2015), at z ≈ 0.006–0.009, with fairly tight limits (≈few × 10⁹ M_⊙) available on the H i mass of three other systems (Mazumdar et al. 2014). Despite this difficulty in measuring the atomic gas content of individual DLA systems, the connection between DLAs and H i emission in low-redshift galaxies has been made statistically by Zwaan et al. (2005). They show that the characteristics of low-z DLAs are consistent with their originating in gas discs of galaxies like those in the z = 0 population.

Similarly, until recently, the low sensitivity of millimetre(mm)-wave telescopes has meant that only a handful of searches have been carried out for molecular gas in emission in DLAs, yielding weak limits on the CO line luminosity, and thence on the molecular gas mass (e.g. Wiklind & Combes 1994). The lack of H i 21-cm and CO emission detections in DLAs has also meant that we have little information on the transverse size of the absorbers. The neutral H i column density can typically only be measured along the sightline to the QSO, except in the rare cases when a single gravitationally lensed or double background QSO can be used to constrain the spatial extent of the neutral gas (e.g. Ellison et al. 2007; Monier, Turnshek & Rao 2009; Cooke et al. 2010; Zafar et al. 2011). Mapping of the H i 21-cm absorption against the extended radio continuum (and thus measuring the spatial extent of the neutral gas), has also so far only been possible in a single DLA, at z = 0.437 towards 3C196 (Briggs, de Bruyn & Vermeulen 2001; Kanekar & Briggs 2004).

With the advent of the Atacama Large Millimeter Array (ALMA), it is now finally possible to directly address the question of the molecular gas content of DLA host galaxies. We have hence initiated an ALMA survey for CO emission in a sample of DLA hosts, to measure their molecular gas masses, kinematical properties of the gas, determine the mass fractions in gas and stars, and their distributions, the star formation efficiencies, and to test whether DLA galaxies have similar properties as emission-selected galaxies at similar redshifts.

For our initial pilot ALMA survey, we chose to observe a sample of high-metallicity DLAs and sub-DLAs, at z ≈ 0.1–0.7. This survey has already yielded the first detection of CO emission in a sub-DLA at z = 0.101 towards PKS 0439−433 (Neeleman et al. 2016b), originally selected as being part of a QSO–galaxy pair (Petitjean et al. 1996). In this paper, we describe our ALMA observations of a second DLA, at z_abs = 0.7163 towards J1323−0021 (SDSS J132323.78−002155.2). The absorber was originally detected by Rao, Turnshek & Nestor (2006) via Hubble Space Telescope (HST) spectroscopy. A companion galaxy was subsequently reported as a candidate DLA host by Hewett & Wild (2007), based on NIR imaging data, but attempts to confirm the identification via detection of optical/NIR emission lines were so far unsuccessful.

Our ALMA data allow us to definitively identify the DLA host galaxy via the detection of CO(2–1) line emission, and to characterize its molecular gas content and distribution. We also present a new Very Large Telescope (VLT) FOcal Reducer and low dispersion Spectrograph (FORS2) spectrum which confirms the identification, as well as new optimized reductions of archival VLT Spectrograph for INtegral Field Observations in the Near Infrared (SINFONI) and Ultraviolet Echelle Spectrograph (UVES) data. In addition, recalibrated Herschel and HST/Space Telescope Imaging Spectrograph (STIS) data are reduced and analysed. The paper is organized as follows: Section 2 describes our new observations and data analysis, as well as our new reductions of archival data, Section 3 presents our results and the extracted physical parameters of the z = 0.7163 DLA host galaxy, while Sections 4 and 5 discuss the new results and summarize our conclusions. Throughout this paper, we assume a flat Λ cold dark matter (ΛCDM) cosmology with H₀ = 70.4 km s⁻¹Mpc⁻¹ and |$\Omega _\Lambda =0.727$| (Komatsu & et al. 2011).

2 OBSERVATIONS AND ARCHIVAL DATA

2.1 ALMA observations

The field of the z = 0.7163 DLA towards J1323−0021 was observed with ALMA using the band 4 receivers as part of project 2013.1.01178.S, on 2014 December 27 (log provided in Table 1). Observing conditions were below average for high-frequency studies, with an average precipitable water vapour measurement of 5.5 mm. The ALMA correlator was configured to use four spectral windows in dual polarization mode. A single high-resolution 1.875 GHz band, sub-divided into 480 channels, was centred at the redshifted CO(2–1) line frequency (134.3 GHz); this yielded a velocity resolution of ≈16.4 km s⁻¹ (after online Hanning smoothing), and a total velocity coverage of 4185 km s⁻¹ around the DLA redshift. To obtain continuum data, the three other bands were used in low-resolution mode, giving 2 GHz bandwidth and 128 channels each. Forty 12-m ALMA antennas were used for the observations, with a maximum baseline of 348.5 m. A standard observing strategy was used, with a phase calibrator (J1319−0049) observed every ≈8 min, and a bandpass calibrator (J1256−0547) observed for 15 min. Flux calibration was obtained through observing the quasar 3C279, whose absolute flux scale is frequently verified by the ALMA observatory. The total on-source time was approximately 40 min.

The data editing, calibration, and imaging were carried out in the casa (Common Astronomy Software Applications) package, using standard procedures. Stable phase solutions were found for most antennas, except for a period of ≈10  min in the middle of the observing run. Data for approximately half the antennas were disregarded for that period. A continuum image of the field was made by combining data from all four observing bands, excluding the channels containing line emission. This image was made using ‘robust’ weighting, with a Briggs parameter of 0.5, and has a root-mean-square (rms) noise level of 16.4 μJy beam⁻¹. We clearly detect continuum emission from the quasar, with a flux density of 0.14 mJy. This is, however, too faint for the purpose of self-calibration. The continuum map shows a tantalizing extension to the north (see Fig. 1), exactly in the direction of the candidate galaxy reported 1.1 arcsec to the NE of the QSO (Hewett & Wild 2007), and at 1.25 arcsec to the NE in a higher resolution adaptive-optics (AO) image (Chun et al. 2010). To investigate this further, we re-imaged the field after subtracting out a point source model at the quasar position from the U-V visibilities. Both the unsubtracted (red contours) and the point-source-subtracted (blue contours) data are shown in Fig.��1. The 140 GHz continuum flux density at the position of the DLA host galaxy is 53 μJy, which corresponds to a 3.2σ detection. The position of the centroid of the continuum emission is in excellent agreement with the galaxy position in the optical image.

A data cube covering the CO(2–1) line redshifted to the redshift of the DLA was made by imaging the high-resolution visibility data set in 50 km s⁻¹  wide spectral channels. Since the data were obtained with slightly higher spatial resolution than requested, we tapered the visibilities beyond 75 kλ during the imaging, in order to improve the surface brightness sensitivity. The resulting spectral cube has a synthesized beam of extent 2.7 arcsec × 2.2 arcsec and an rms noise of 0.25 mJy beam⁻¹ per 50 km s⁻¹ channel. The above angular resolution translates to a spatial resolution of ≈19.8 kpc × 16.1 kpc at the DLA redshift, z = 0.7163. This relatively coarse spatial resolution excludes the possibility that we may be resolving out some of the CO emission.

CO(2–1) emission is clearly detected in the cube, with a peak line flux density of (1.70 ± 0.25) mJy (i.e. at ≈7σ significance). The CO(2–1) line profile is shown in Fig. 2; this was obtained by summing all emission over an aperture of 1.5 arcsec centred on the optical galaxy. The profile is seen to display the steep sides and relatively flat central part referred to as ‘boxy’ profiles or double-horned (Davis et al. 2011), although the signal-to-noise ratio (S/N) is relatively low. We obtained a total CO intensity image by adding the flux from all channels over the central range of ≈500 km s⁻¹, without applying any spatial masking to the channel maps. Fig. 3(a) shows an overlay of this total CO intensity image on the NIR image; it is clear that the CO emission is centred on the NIR emission from the galaxy and not on the QSO. With the clear detection of the CO(2–1) emission line, we identify the neighbouring galaxy as the DLA host: this constitutes the first confirmation of the identification of a DLA host galaxy via molecular line emission. The velocity-integrated CO(2–1) line flux density is S_int = (0.50 ± 0.05) Jy km s⁻¹; the measured flux densities are listed in Table 4.

In addition to the integrated CO image, we also produced two maps where we separately summed the fluxes from the low-velocity CO peak (over the velocity range −400 km s⁻¹ to −200 km s⁻¹) and the high-velocity peak (+200 km s⁻¹ to +400 km s⁻¹). Those are both plotted in Fig. 3(b), as blue and red contours respectively. It is seen that the red set of contours fit the position of the galaxy well, while the centroid of the blue contours is offset by 0.9 arcsec (6.6 kpc) to the west indicating that the molecular gas has a spatially resolved kinematic structure. We shall return to this in more detail in Section 3.2.

2.2 Herschel data

The QSO was observed with Herschel/SPIRE (Griffin et al. 2010) on 2011 July 31 in the 250, 350, and 500 μm bands (OBSID: 1342224987, PI: V. Kulkarni). We obtained the recalibrated maps from the Herschel science archive.

The effective full width at half-maximum (FWHM) in the three bands are 17.6, 23.9, and 35.2 arcsec, respectively, and because of the small separation between the QSO and DLA galaxy, it was not possible to deconvolve their emission into two distinct components. Visual inspection of the maps showed that the QSO plus DLA galaxy is clearly detected in the 250 and 350 μm maps, but not in the 500 μm map, which was excluded from further analysis.

To measure the integrated flux, the Herschel level 2 maps were divided by the effective beam area and multiplied by the pixel sizes to get maps in units of Jy pixel⁻¹  . We performed aperture photometry following the description of SPIRE point source photometry in Pearson et al. (2014) choosing a small aperture size, and including an aperture correction to derive the total flux. Errors were obtained using the same apertures on the Herschel error maps and adding errors in quadrature. The integrated aperture flux was verified using the data analysis software ‘Herschel Interactive Processing Environment (hipe)’ (Ott 2010).

The fractional flux detected in the ALMA continuum band from the DLA galaxy relative to the QSO is 0.27 ± 0.07. To estimate the Herschel flux from the DLA galaxy alone, we assumed that the fractional flux detected in the ALMA continuum band from the DLA galaxy is the same also at shorter wavelengths. The final results are reported in Table 2. We discuss, and validate, this assumption in Section 3.4.3.

2.3 VLT–FORS2 data

The galaxy identified in the ALMA image as the absorber, was also part of a VLT–FORS2 programme to obtain spectra of DLA galaxy candidates (programme ID 095.A-0890(A); PI: L. Christensen). The galaxy was observed on 2015 April 17 and 18, with FORS2 long-slit spectroscopy using the grism 600RI, which covers the wavelength range 5120–8250 Å at a spectral resolution of R ∼ 900 with the chosen slit width of 1.3 arcsec. The total integration time was 4× 1100 s on target with the slit aligned at PA = 42° east of north, to cover both the QSO and the candidate galaxy based on the relative coordinates in Rao et al. (2011). The observing conditions were non-photometric, and the seeing measured in the spectra varied in time and wavelength over the range 0.9–1.2 arcsec, with better values measured on April 18 (Table 1).

The VLT–FORS2 data were analysed in the standard manner with bias subtraction, flat fielding, wavelength calibration, and sky background subtraction using a median value from pixels adjacent to the QSO and the galaxy. To flux calibrate the data, we used an observation of the spectrophotometric standard star EG274 observed on April 17. However, as the night was not photometric systematic flux calibration errors are expected, along with slit losses. Comparing the flux-calibrated QSO spectrum with a spectrum obtained in the Sloan Digital Sky Survey (SDSS) on 2001 February 19, we find that the FORS2 spectrum has a 40 per cent lower flux level than the SDSS spectrum over the entire wavelength range. A further comparison of the FORS2 R-band acquisition image with an SDSS r-band image reveals that 10 bright stars in the field have the same magnitudes (within 0.01 mag.) in the FORS2 and SDSS images, while the QSO in the FORS2 image is 0.1 mag fainter than in the SDSS image. We cannot determine how much the variation of the flux level in the FORS2 spectra is due to intrinsic time variations of the QSO flux or slit losses and hence simply report the measured flux from the DLA host galaxy, with the caveat that the flux is likely to be low by a factor of ≈2.

A high-resolution Keck K-band image, obtained with the use of AO, shows a spatial offset of 1.25 arcsec between the galaxy and the QSO (Chun et al. 2010). This implies that the two objects are blended in the moderate seeing of our FORS2 observations. A careful subtraction of the QSO emission is therefore necessary to recover any spectral signatures from the galaxy. To search for the [O ii] λλ3727,3729 doublet, we assume that the QSO continuum emission is a pure power law in two wavelength ranges immediately bluewards and redwards of the expected [O ii] emission feature, and that the spatial point spread function (PSF) modelled in the adjacent wavelengths does not depend on wavelength. The measured QSO flux is then scaled to the different spectral wavelengths and subtracted out, a process known as spectral PSF (or SPSF) subtraction. The same approach has been used to find metal emission lines from high-redshift DLA galaxies (Weatherley et al. 2005; Fynbo et al. 2013; Krogager et al. 2013). Fig. 4 shows the result of the subtraction, which reveals a clear detection of excess emission at exactly the spatial position of the galaxy.

The [O ii] line doublet is not well resolved at the relatively low FORS2 resolution and we therefore perform a simultaneous fit to both lines, where we lock the FWHM of the two lines to be identical, but allow the flux ratio of the lines to vary freely. We find that the FWHM is <300 km s⁻¹ with a heliocentric wavelength corresponding to [O ii] at z = 0.7165 ± 0.0004, where the redshift uncertainty reflects the uncertainty of wavelength position in the emission-line fit from propagated statistical errors. There are additional systematic errors from strong subtraction residuals close to the quasar trace which we take into account as follows. The SPSF subtraction at the expected galaxy position (from imaging) is excellent with insignificant residuals, and the extracted [O ii] line at this position provides a redshift z = 0.7176 ± 0.0004. We take the difference here to be the result of the sum of systematic and statistical errors, plus possibly a component from rotation (as suggested from Fig. 4). Conservatively, we assign it all to the combined error and adopt the average (z = 0.7170 ± 0.0006) as the best redshift estimator where the error spans both extremes.

The integrated flux from the two components is (2.8 ± 0.3) × 10⁻¹⁷ erg cm⁻² s⁻¹, after correcting for Galactic extinction of E_B−V = 0.024 (Schlafly & Finkbeiner 2011), but not for intrinsic extinction in the DLA galaxy itself. The full results of the FORS2 analysis are listed in Table 4. We note, in passing, that the low [O ii] emission-line flux is consistent with the non-detection of [O ii] emission by Straka et al. (2015).

2.4 VLT SINFONI data

Péroux et al. (2011) presented VLT SINFONI J-band integral field unit spectroscopy of the field of the z = 0.7163 DLA. They reported a non-detection of Hα emission from the candidate host galaxy. To complete our multiwavelength study, we retrieved the SINFONI data set (programme ID: 079.A-0673) from the European Southern Observatory (ESO) archive and rereduced it. The SINFONI J-band spectra cover the spectral range 11000–14000 Å at a resolution of R ∼ 2000, and the field of view 8 arcsec×8 arcsec. A total of 13 × 900 s of integration was split over three different observing nights in 2007, using a dither pattern for sky subtraction. A large offset dither pattern avoids nodding into the galaxy, thus avoiding over subtraction.

Our data processing was done with the recent version (2.6) of the SINFONI pipeline; the individual steps involved the same basic routines used by Péroux et al. (2011), apart from the flux calibration and correction of telluric absorption lines. Specifically, telluric absorption-line corrections may cause differences in our results compared to the previous reported ones. We do not initially include a telluric correction of the data cubes using the calibration observation of a hot star just before or after the science exposures as implemented in the pipelines, but instead we derive the correction directly from the QSO spectrum. We find that this is particularly important in the spectral region between the Y and J bands, where the Hα emission line from the DLA galaxy falls. Observations on each of the three observing nights of hot B-type stars with known magnitudes from the Hipparcos catalogue and known spectral types were used for flux calibration.

To extract information on the galaxy, we first construct an intermediate-band filter image by summing over an ≈50 nm wide region in the data cube centred on 13  500 Å. The upper left-hand panel in Fig. 5 shows the resulting image, which is clearly dominated by the QSO emission, but also shows an extension of fainter emission to the north-east at a distance of 1.2 arcsec (equal to the impact parameter of the candidate DLA host). Fitting a 2D Moffat function to the QSO location gives a spatial FWHM of 0.9 arcsec × 0.7 arcsec in the east-west and north-south directions, respectively. We then model the QSO emission by fitting a 2D Moffat profile with a spatial location and an FWHM that vary smoothly with wavelength through the data cube and subtract this model from the initial data cube. After this, the same spectral region of the data cube is summed as illustrated in the upper right panel in Fig. 5. Some residuals within the FWHM of the QSO position are seen, but the continuum emission of the DLA galaxy is now clearly visible in the image.

We extract a 1D spectrum of the galaxy within an elliptical aperture of 2 arcsec × 1.5 arcsec by co-adding spatial pixels (‘spaxels’) associated with the galaxy. After application of corrections for telluric absorption and flux calibration, the continuum flux level in the chosen region around λ = 13 500 Å is measured to be (6.7 ± 1.3) × 10⁻¹⁸ erg cm⁻² s⁻¹ Å⁻¹. This flux density corresponds to an AB magnitude of 19.9 ± 0.25 at this wavelength. The large uncertainty reflects the added uncertainty from the QSO subtraction residuals. At the bluest end of the spectral range, the 1D spectrum corresponds to a magnitude m₁₁₂₀₀ = 20.20 ± 0.20. We include those continuum magnitudes in our analysis of the spectral energy distribution (SED) of the galaxy in Section 3.4.1.

The extracted spectrum of the galaxy also shows a clear Hα emission line at λ = 11272.2 ± 0.8 Å. We create a pure narrow-band Hα image by co-adding the relevant wavelength slices around the emission line and subtracting the adjacent continuum emission from the galaxy; the result is shown in the lower left panel of Fig. 5. To extract the final, telluric absorption corrected, emission-line spectrum, we first define a 2 arcsec × 1.5 arcsec rectangular aperture centred on the DLA galaxy. We sum all spaxels of the original data cube within this aperture, apply the telluric correction determined from spaxels only containing QSO continuum, and finally subtract the continuum from the resulting 1D spectrum. The spectrum is shown in the lower right-hand panel of Fig. 5 where we also see a clear detection of [N ii] λ6584 emission. The other line of the doublet [N ii] λ6548 is not detected, but this is consistent with the noise level of the data. The emission lines indicate a galaxy redshift of z = 0.7171 ± 0.0001.

Fitting Gaussian functions to the emission lines, we measure a line flux of f(Hα) = (12.0 ± 3.9) × 10⁻¹⁷ erg cm⁻² s⁻¹ and f([N ii]) = (5.6 ± 1.8) × 10⁻¹⁷ erg cm⁻² s⁻¹. The Hα line is resolved and has an FWHM of 238 ± 34 km s⁻¹, after correcting for the instrumental resolution. On both the extracted J-band (m₁₃₅₀₀) and narrow-band images, we measure the position of the DLA galaxy relative to the QSO. The resulting values (PA and impact parameter) are provided in Table 4.

In passing we note that we also detect an emission-line system at z = 1.38, at an impact parameter to the QSO sightline of 2.3 arcsec to the north-east. This system was serendipitously discovered by Péroux et al. (2011), and does not have a corresponding detection in continuum emission (Hewett & Wild 2007; Rao et al. 2011; Chun et al. 2010). Additionally, given the high line flux ratio log f([O iii])/f(Hβ) ∼ 1 which indicates a hard ionizing source, this system is most likely an extended emission-line region at the redshift of the quasar, and ionized by it. This emission-line system does not in any way contaminate the results obtained for the DLA host galaxy.

2.5 VLT UVES and HST STIS data

We obtained the advanced data product VLT-UVES spectrum of the QSO, presented in Péroux et al. (2006), from the ESO data archive. We corrected wavelengths to the heliocentric frame and converted from air to vacuum wavelengths before combining individual exposures to produce the final spectrum. The saturated absorption lines in the spectrum, such as Fe ii λ2344, λ2382, and λ2600, show multiple spectral components at redshifts in the range z = 0.71531–0.71668, equivalent to a total velocity spread of ≈240 km s⁻¹. Using the definition of Δv₉₀ (Prochaska & Wolfe 1997), which is based on the optical depth profile of an unsaturated low-ionization line, we measure a velocity spread of Δv₉₀ = 141 km s⁻¹ from the Mn iiλ2576 absorption line (see Table 4).

The large velocity spread and number of spectral components makes it difficult to assign a unique redshift to the absorption system. Also here we then use the definition of Δv₉₀ and assign the median optical depth (z_τ:50, the redshift at which 50 per cent of the optical depth is on either side) as the effective redshift of the system. Using this definition, we find z_τ:50 = 0.71612 which implies that the absorber is kinematically offset from the Hα emission by 171 ± 18 km s⁻¹. In addition, we determine the redshift at which the system has its largest optical depth, z_τ:max = 0.71625.

Péroux et al. (2006) report that 16 individual components make up the complex line profile of the DLA and derive a metallicity [Zn/H] = 0.61 ± 0.20, for log[N(H i)/cm⁻²] |$=20.21^{+0.21}_{-0.18}$| (obtained by Khare et al. 2004). We note that the identification of the absorber as a DLA or a sub-DLA and the value of the metallicity is critically dependent on the H i column density estimate. Specifically, Rao et al. (2006) derive log N(H i) |$= 20.54^{+0.16}_{-0.15}$| from the same HST STIS spectrum. We therefore downloaded the STIS data from the HST archive and performed an independent determination of the H i column density. Our result agrees with the later analysis of Rao et al. (2006), but does not rule out the original result of Khare et al. (2004). In the following analysis, we will use our estimate, log[N(H i)/cm⁻²] |$= 20.4^{+0.3}_{-0.4}$|⁠, where the adopted errors span the full 1σ range of the two other estimates. This then implies a gas-phase metallicity of [Zn/H] = +0.4 ± 0.3. We note that Neeleman et al. (2016a) did not include this spectrum in their sample because its S/N is lower than their threshold of 4.

3 RESULTS

3.1 The molecular gas mass

The velocity-integrated CO J = 2–1 line flux density is S_int = (0.50 ± 0.05) Jy km s⁻¹. This can be used to compute the molecular gas mass of the z = 0.7163 galaxy if we know the CO-to-H₂ conversion factor α_CO and the nature of the CO excitation. Both of these depend on the nature of the galaxy: for example, spiral disc galaxies like the Milky Way have α_CO ≈ 4 M_⊙ (K km s⁻¹pc²)⁻¹, while QSOs and starburst galaxies (e.g. ultraluminous IR galaxies (ULIRGs) and sub-mm galaxies) have far lower values, α_CO ≈ 1 M_⊙ (K km s⁻¹pc²)⁻¹, both after including a factor of 1.36 to correct for the contribution of helium (e.g. Bolatto, Wolfire & Leroy 2013; Carilli & Walter 2013). The conversion factor also has a strong dependence on metallicity, with far higher α_CO values ( ≳ 30 M_⊙ (K km s⁻¹pc²)⁻¹) obtained in low-metallicity systems (e.g. Leroy et al. 2011) than in high-metallicity galaxies (α_CO ≲ 4 M_⊙ (K km s⁻¹pc²)⁻¹). In the present case, the high metallicity of the DLA host galaxy indicates a low α_CO value.

Next, CO in QSOs and starburst galaxies has typically been found to be thermalized up to the J = 3 level, with sub-thermal excitation of the higher J levels (e.g. Weiß et al. 2007; Danielson et al. 2011). Conversely, even the J = 2 level is typically sub-thermally excited in spiral galaxies (e.g. Fixsen, Bennett & Mather 1999; Dannerbauer et al. 2009; Carilli & Walter 2013). In thermal equilibrium, the brightness temperature of the CO J = 2–1 line would be equal to that of the J = 1–0 line, while in the Milky Way, the J = 2–1 brightness temperature is ≈0.6 times that of the J = 1–0 line. We will assume sub-thermal excitation of the J = 2 level and a brightness temperature line ratio of r₂₁ ≈ 0.6 (⁠|$r_{21}=L^{\prime }_{\rm CO(2-1)} / L^{\prime }_{\rm CO(1-0)} = S_{\rm CO(2-1)} / (4\times S_{\rm CO(1-0)})$|⁠) for the J = 2–1 and 1–0 lines, consistent with the assumption of a Milky Way value for α_CO.

For the assumed ΛCDM cosmology, the CO J = 2–1 flux density of S_int = (0.50 ± 0.05) Jy km s⁻¹ implies a CO J = 1–0 flux density of (0.20 ± 0.02) Jy km s⁻¹, and thence, a CO J = 1–0 line luminosity of L΄(1 − 0) = (5.55 ± 0.55) × 10⁹ K km s⁻¹pc². Using α_CO = 4.2 M_⊙ (K km s⁻¹pc²)⁻¹ (recommended by Bolatto et al. 2013) then yields a total molecular gas mass of M|$_{\rm mol{\rm -}gas} = (2.33 \pm 0.23) \times [\alpha _{\rm CO}/4.2] \times (0.6/r_{21}) \times 10^{10} \, {\rm M}_{\odot }$| for the DLA host galaxy. Conversely, if we were to assume that the DLA host contains an obscured starburst, with α_CO ≈ 1 and r₂₁ ≈ 1 (i.e. that the J = 2 level is thermalized), we obtain a molecular gas mass of M|$_{\rm mol{\rm -}gas} = (3.5 \pm 0.35) \times [\alpha _{\rm CO}/1.0] \times (1.0/r_{21}) \times 10^{9} \, {\rm M}_{\odot }$|⁠. Since the SFRs estimated from both the Hα line and the dust continuum are low ( ≲ 5 M_⊙ yr⁻¹ see Sections 3.4.2 and 3.4.3), we consider the starburst scenario to be unlikely. This is further supported by the galaxy's gas-to-dust ratio (see Section 4.4). We therefore conclude that the most likely scenario is the former, and that the molecular gas mass of the DLA host galaxy is likely to be very high. The values of α_CO applied here already include the correction (1.36) for He.

3.2 Kinematics of the molecular, neutral, and ionized gas

In Fig. 6, we provide an enlarged view of the CO J = 2–1 emission-line profile shown in the right-hand panel of Fig. 2. The flat-topped appearance of a ‘boxy’ profile (Davis et al. 2011) is here evident. The flat profile appears to consist of three individual CO velocity components, but the details of this sub-structure is difficult to verify at the current S/N and must await better data. Overlaid in Fig. 6, we show the spectral components of the ionized gas seen in the VLT–SINFONI and VLT–FORS2 emission spectra (marked in red), and the absorbing DLA gas seen in the VLT–UVES spectrum (blue). For the absorption profiles, we mark the velocity spread containing 90 per cent of the integrated optical depth of low-ionization unsaturated metal lines, Δv₉₀.

Curiously, Fig. 6 indicates that the Hα emission is kinematically related only to the two highest redshift molecular components, while the lowest redshift molecular feature appears to be kinematically distinct from both the optical emission and absorption lines. The lowest velocity molecular peak spans the bins from which we produced the blue contour map in Fig. 3, where it was seen that this peak also appears spatially offset from the optical galaxy. The velocity span of the metal-line absorption lies kinematically between the reddest and bluest CO velocity components, which are plotted as red and blue contours in Fig. 3.

3.2.1 Centroid positions of emission sub-components

In order to address this issue of position of the various emission sub-components quantitatively, we determine the positions of the dust and of the red and blue CO components. For DLA hosts, this information is traditionally provided in the form of the position angle (PA) and the impact parameter (b_DLA), relative to the quasar. Based on the high-resolution K-band data, Chun et al. (2010) reported (PA, b_DLA) = (24.5°, 1.25 arcsec). From the images of the dust continuum, the red CO component, and the blue CO component, we obtain (PA, b_DLA) = (17 ± 8°, 1.46 arcsec ± 0.2 arcsec), (32 ± 14°, 1.03 arcsec ± 0.3 arcsec), and (−18 ± 14°, 1.15 arcsec ± 0.3 arcsec), respectively. It is apparent that the first two sets of values agree well with the optical position of Chun et al. (2010), whereas the last does not. More specifically, measuring the distance between the K-band position of the host and the centroid of the image of the blue CO component, we find 0.9 arcsec ± 0.3 arcsec. This is statistically significant evidence that the blue CO component is spatially offset from the other three (optical, dust, and red CO) components.

From the results summary in Table 4, we see that there is agreement (to within the errors) between the relative position of the DLA galaxy detected in K-band, J-band, dust continuum, Hα and ‘red CO component’, i.e. all of those emission components of the Damped absorber appear to be centred on the optical galaxy.

3.3 The CO line width and the dynamical mass

As noted in Section 2.1, the CO J = 2–1 line profile matches the definition of a boxy profile. To obtain measures for the line width which are directly comparable to those of the COLD GASS galaxy sample (Saintonge et al. 2011), we use the same fitting formula for a double-horn profile (equation A2 in Tiley et al. 2016). A fit of this function to the unsmoothed data (shown in Fig. 7, left-hand panel) finds a CO central redshift of z = 0.71645(9), a W50 of 647 ± 54 km s⁻¹ and W20 of 652 ± 54 km s⁻¹ (uncorrected for inclination). Our CO map does not have the resolution to obtain an inclination angle of the disc, but from the high-resolution AO image presented by Chun et al. (2010, their figs 1 and 2), we estimate that the inclination is in the range 40°–48° (including also the effect of a slightly elongated residual PSF). Using the most conservative value, we get W50/sin(48°) = 871 km s⁻¹.

This is an extremely large velocity width for CO emission, comparable to those obtained in the most massive sub-mm galaxies and QSOs (e.g. Carilli & Walter 2013). As noted above, it is possible that the bluest feature of the profile may be due to a separate object which is kinematically detached from the main galaxy. In particular, the Hα line is offset from the CO central redshift. To test this possibility, we also perform a second fit using the same functional form, but now locking the central frequency of the fit to that of the Hα line. The fit is restricted to the spectral range from −200 km s⁻¹ and above, and gives W50 = 378 ± 52 km s⁻¹and W20 = 384 ± 52 km s⁻¹ which is more in line with typical galaxies of similar stellar mass. In addition, we fit a single gauss profile to the remaining blue peak which covers the spectral range used for the blue contours in Fig. 3. The best-fitting gauss has σ = 61 km s⁻¹ and an offset of −212 km s⁻¹ relative to the fiducial DLA redshift (0.7163), resulting in a total offset between this blue component and the Hα line of 352 km s⁻¹. The combined fit is shown in the right-hand panel of Fig. 7. The flaring parameter (R_2,5 =W20/W50) is found to be 1.01 and 1.02 for the two fits, respectively, confirming the profile shape as boxy (R_{2, 5} < 1.2).

The dynamical mass of the DLA host can be roughly estimated as M_dyn ∼ (ΔV² × R)/G, where ΔV is the characteristic velocity, and R the characteristic size. We will assume that the entire CO emission arises from a single object. A lower limit on its spatial extent can be obtained from the transverse separation between the optical and the blue CO components; the angular separation of 0.9 arcsec corresponds to a spatial separation of ≈6.5 kpc. Assuming ΔV ≈ (W50/sin(48°))/2 = 436 km s⁻¹ and R ≳ 6.5 kpc, we obtain M_dyn ≳ 2.9 × 10¹¹ M_⊙.

3.4 Results from the optical and near-IR data

The excellent spatial alignment between the CO emission map and the optical image of the candidate host galaxy in Fig. 3(a), and the good agreement between the CO and the metal-line absorption and emission velocities in Section 3.2 clearly demonstrate that the CO J = 2–1 emission originates in gas associated with the galaxy detected in the K-band and optical images. This constitutes the first case of spectroscopic identification of a DLA host galaxy via its molecular line emission. This is an important result in its own right, because it shows that ALMA has opened up a new window for the identification of DLA hosts, a task that has been fraught with much difficulty in the past. Of course, once the host has been identified via its molecular emission lines, additional data can be obtained and already existing data can be exploited, as has been done here.

3.4.1 The stellar mass of the z = 0.7163 galaxy

To derive the stellar mass of the z = 0.7163 galaxy, we use the standard approach of fitting to the SED, using release version 12.2 of hyperz (Bolzonella, Miralles & Pelló 2000). We assume a Chabrier initial mass function (IMF), and derive different stellar masses, using templates with various star formation histories and metallicities between 0.2 and 2.5 times solar values (see Christensen et al. 2014, for further details).

The input photometry of the galaxy in ugri filters is taken from Rao et al. (2011). However, strong residuals from the subtraction of the QSO's PSF have a strong effect on the i-band measurement (Rao et al. 2011)¹ and we hence choose to not include this band in the SED fit. The NIR K-band data are from Hewett & Wild (2007), who obtained a magnitude similar to that obtained from the K-band AO data (Chun et al. 2010). In addition, we also use the magnitudes measured from our VLT–SINFONI data, derived in Section 2.4. All photometric data are listed in Table 3.

We obtain a best-fitting stellar mass of log|$[{\rm M}_*/{\rm M}_{\odot }]=10.80^{+0.07}_{-0.14}$|⁠, as illustrated in Fig. 8, with a reddening of |$A_{\rm V}=0.4^{+0.3}_{-0.3}$| mag, 2.5 times solar metallicity, and an age of ≈0.5 Gyr, assuming an instantaneous burst of star formation. To derive the 68 per cent confidence intervals for the results, we assume a Gaussian error distribution, propagate the magnitude errors, and perform SED fits to 10³ random realizations. Including the uncertain i-band data point in the fit (shown as a red open circle in Fig. 8) significantly increases the χ² for the best fit, but yields a very similar derived stellar mass, log|$[{\rm M}_*/{\rm M}_{\odot }]=10.94^{+0.05}_{-0.08}$|⁠.

3.4.2 Star formation rates

The galaxy has been previously suggested to be devoid of emission lines and most likely a ‘red-and-dead’ type. With our detection of multiple emission lines and 140 GHz continuum, we are now able to estimate the SFR using several tracers: the UV continuum emission, the [O ii] and Hα lines, but also the obscured SFR can be determined (see Section 3.4.3). We convert the measured flux densities and line fluxes to luminosities in the ΛCDM model adopted here, and then to SFRs, using the conversion factors in Kennicutt (1998). This calibration assumes a Salpeter IMF; a simple downward correction factor of 1.8 is finally used to shift to the assumption of a Chabrier IMF, for consistency with our SED fitting.

The u-band central transmission corresponds to a rest-frame wavelength of ∼2000Å which implies an unobscured SFR_UV = 0.43 ± 0.23 M_⊙ yr⁻¹. The large uncertainty reflects the large u-band magnitude error.

The [O ii] emission line yields SFR|$_{\mathrm{[O{\small II}]}}=0.51\pm 0.05$| M_⊙ yr⁻¹, while the Hα line yields SFR_Hα = 1.6 ± 0.4 M_⊙ yr⁻¹. Apart from the fact that Hα and [O ii] trace SFRs on slightly different time-scales, which could lead to this difference, we also stress that the Hα emission is measured in a larger aperture than that used for the [O ii] emission. In addition, the [O ii] flux might be underestimated due to the non-photometric observing conditions and will be underestimated due to intrinsic extinction.

3.4.3 Dust mass, dust temperature, and the obscured SFR

To determine the dust mass in the host galaxy, we use magphys (da Cunha, Charlot & Elbaz 2008) with the photometry in Table 3, the ALMA 140 GHz measurement of 53 ± 17 μJy, and the Herschel photometry in Table 2, again fixing the redshift to that of the galaxy. magphys assumes a Chabrier IMF to generate galaxy templates, which makes the results directly comparable to the output of hyperz.

The Herschel resolution does not allow us to determine which fraction of the flux is from the DLA galaxy and which is from the QSO. In the higher resolution ALMA data, we find that the relative fraction is 0.27/0.73, and in Section 2.2, we assumed the same ratio to hold for the Herschel data. Here, we test this assumption. We proceed by forming a series of photometric tables where we split the Herschel flux 0.10/0.90, 0.20/0.80,... 0.90/0.10 in steps of 0.10. For each table, we derive the best-fitting model templates, and compute the combined χ² from the Herschel and ALMA data. The best fit is found where 0.30 of the Herschel flux comes from the DLA galaxy, supporting the assumption used in Section 2.2. With this assumption, the maximum likelihood for the dust mass is found to be |$\log [{\rm M}_{\mathrm{dust}}/{\rm M}_{\odot }]= 8.45^{+0.10}_{-0.30}$|⁠, with a luminosity of log [L_dust/L_⊙] = 11.15 ± 0.20. The maximum likelihood peaks at a stellar mass of log [M_*/M_⊙] = 10.85 ± 0.20, in full agreement with our result in Section 3.4.1 (⁠|$10.80^{+0.07}_{-0.14}$|⁠). The fit yields an obscured SFR of |$4.5^{+4.4}_{-2.6}$| M_⊙ yr⁻¹ and a best-fitting dust temperature of 21.5 ± 2 K in the ambient interstellar medium.

In summary, the three optical/UV tracers yield low SFRs, in the range ≈0.4–1.6 M_⊙ yr⁻¹. While the obscured SFR inferred from the fit to the dust emission is a little higher, it is still relatively low for a galaxy with this high stellar mass; its specific SFR is = SFR/M_* ≈ 10⁻¹¹ yr⁻¹, which places it in the red sequence of passively evolving galaxies in the correlation between SFRs and stellar masses (Wuyts et al. 2011).

3.4.4 Emission-line metallicity

Having detected both Hα and [N ii] emission lines, we can use their line flux ratio N2 = log f([N ii] 6584)/f(Hα) to infer the oxygen abundance of the host (e.g. Pettini & Pagel 2004; Kewley & Ellison 2008). We note that the relation between N2 and the metallicity is known to saturate at high metallicities (e.g. Steidel et al. 2014), and that one has to rely on extrapolations above solar metallicity (Pettini & Pagel 2004). Here, we use the calibration relation of Maiolino et al. (2008), which has been calibrated to higher metallicity SDSS galaxies, using photoionization models (Kewley & Dopita 2002). Equally important, this calibration has been used to tie the conventional galaxy mass–metallicity relation (Tremonti et al. 2004) to the mass–metallicity relation in DLAs (Møller et al. 2013; Christensen et al. 2014).

From the N2 index, we infer 12+log(O/H) = 9.20 ± 0.09. In addition to the measurement error, the N2 calibration itself has an intrinsic spread of ≈0.2 dex. The solar oxygen abundance is 12+log(O/H) = 8.69 (Asplund et al. 2009), yielding an oxygen metallicity of [O/H] = +0.51 ± 0.09 (not including the calibration uncertainty), i.e. 3.2 times the solar value. This is in excellent agreement with both the independent result from the best-fitting SED model which had 2.5 times the solar value, and with the DLA absorption metallicity (Section 2.5).

4 DISCUSSION

In the vast array of possible techniques to select high-z galaxies (e.g. Lyman-break, Lyman-α emission, sub-mm emission, QSOs, ULIRGs, and radio galaxies), the selection via damped Lyman-α absorption has always held a unique position. The use of absorption towards a background QSO implies that this selection is not subject to the usual luminosity bias that affects all emission-based methods, rather it is solely a function of the H i absorption cross-section. Consequently, some of the galaxies identified via the DLA approach may well not be identifiable by any other means; however, there does exist some overlap with traditional flux-limited samples, and this overlap allows one to use DLA samples to extend galaxy scaling relations out beyond the boundaries of emission-selected, flux-limited samples. Nevertheless, one of the critical questions in DLA studies has long been: ‘Is selection via damped Lyman-α absorption simply an alternative way to identify the same galaxies that are detected via flux-limited surveys, or does this selection also allow detection of galaxies that are missed in other surveys?’. Several papers have addressed this question over the past decades considering all previously observable properties, but one important property, the molecular gas content, has so far not been possible to study for DLA galaxies. Now, with ALMA in operation, we are finally able to start filling in this missing section of parameter space.

Above we have reported the discovery of a CO line-emitting galaxy (ALMA J132323.81−002154.1) at z_CO = 0.71645 and at the position of a known optical candidate DLA host. In this section, we now compare the properties of this DLA galaxy to the properties of emission-selected galaxy samples. In particular, we ask if it lies within the range of the scaling relations known from flux-selected samples. Where possible, we also include our previous detection of the absorbing galaxy towards PKS 0439−433 = QSO J0441−4313 (Neeleman et al. 2016b). The H i column density of the absorber in PKS 0439−433 is below the classic definition of a DLA, but for simplicity we shall in what follows refer to both of the galaxies simply as ‘DLA galaxies’, but keeping in mind the sub-DLA nature of the absorber towards PKS 0439-433.

4.1 The relation between molecular gas mass and stellar mass

Fig. 9 plots the molecular gas mass of the two absorbing galaxies versus their stellar mass, along with data from three comparison galaxy samples: COLD GASS, at z ≈ 0.02 (Saintonge et al. 2011, in blue), ALLSMOG at z ≈ 0.04 (Bothwell et al. 2014; Cicone et al. 2017, in red), and PHIBSS at z ≈ 1.2 and ≈2.2 (Tacconi et al. 2013, in yellow). For convenience, we use the same CO-to-H₂ conversion factor, α_CO = 4.2 M_⊙ (K km s⁻¹pc²)⁻¹ for all galaxies in the figure. The left-hand panel shows that the COLD GASS and ALLSMOG samples lie on the well-known correlation between molecular gas mass and stellar mass. Conversely, the two absorbers lie at or above the upper boundary of this wide relation. For easy reference in this figure, and in the remainder of the paper, we use the short-hand notation g0439 and g1323 for the two galaxies. In Table 4, we list the known properties of g1323; Table 5 provides a summary of the relevant properties of g0439.

Galaxies from the PHIBSS sample are seen to have systematically larger gas masses than the two absorbers, as well as than the flux-selected samples in the nearby Universe. However, to obtain a fair comparison between the galaxies at different redshifts, we must take the redshift evolution of this scaling relation into account. The simplest view is that the relation between molecular gas-mass and stellar mass scales (empirically) as (1 + z)⁻³ between redshifts 0 and 2.5 (Genzel et al. 2015); this empirical correction is applied in the right-hand panel of Fig. 9. The figure shows that the high-z PHIBSS points have now moved down to perfectly overlap with the two low-z galaxy samples. The two absorbers remain in the upper part of the distribution, but now lie within the spread of the distribution of the emission-selected galaxies. We note that both absorbers appear to be rich in molecular gas; notably, g0439 lies on the upper boundary of the distribution.

4.2 The SFR, molecular gas fraction, and stellar mass relations

The other view on the influence of redshift on scaling relations is that of ‘fundamental relation invariance’. In this view, the values of physical parameters describing the galaxies are tied to each other to only allow galaxies to fill small sub-sets of physical parameter space, and galaxy samples then ‘drift’ through this parameter space, yielding an apparent redshift evolution in individual parameters. Santini et al. (2014) formulated such a fundamental relation between the SFR, the gas fraction (f_gas = M_gas/(M_* + M_gas), where M_gas is the total gas mass), and the stellar mass, with the claim that all galaxies, at all redshifts (out to z ≈ 2.5, the upper limit of the redshifts of the galaxies of their sample), must lie on this relation. In order to test if a similar relation exists for molecular gas mass alone (⁠|$f_{\rm mol{\rm -}gas}= M_{\rm mol{\rm -}gas}/(M_* + M_{\rm mol{\rm -}gas})$|⁠), and in particular to test if g1323 then lies on this relation, we first use the same comparison samples as above, but now cut out a thin slice in stellar mass of each sample, centred on the M_* of g1323. In Figs 10 and 11, we plot SFR versus |$f_{\rm mol{\rm -}gas}$| of galaxies within the mass range 10^10.6 ≤ M_*/M_⊙ ≤ 10^11.0. Several different indicators for the SFR are available, applicable at various wavelengths, and in order to provide appropriate and unbiased comparisons we use different symbols for UV- and Hα-based SFR in Fig. 10, while total SFRs corrected for the dust-obscured component are plotted in Fig. 11. For comparison, we also plot the relevant total gas-mass relations within which the galaxies should lie (computed from equation 9 and table 1 of Santini et al. 2014). For completeness, we note here that the calibration of the relation as parametrized by Santini et al. (2014) is based on a Salpeter IMF. For an internally consistent comparison, we have converted their SFR values to a Chabrier IMF, by subtracting 0.24 dex from their reported log[M_*/M_⊙] and 0.15 dex from their log[SFR].

In order to obtain SFRs for the three comparison samples, we proceeded as follows. COLD GASS galaxies were cross-matched with the SDSS data base, and Hα-based SFRs were extracted from aperture corrected SFRs (Brinchmann et al. 2004).² For ALLSMOG galaxies, Cicone et al. (2017) (ALLSMOG website³) list the SFRs (again Hα and based on SDSS data base values), and we also obtained Hα-based SFRs for part of the PHIBSS sample from Erb et al. (2006). For the remaining galaxies in the COSMOS and GOODS-N fields, we obtained estimates of the UV SFRs from the Rainbow data base,⁴ which reports the UV-based SFRs derived from SED fits with fast (Kriek et al. 2009). Tacconi et al. (2013) report a total SFR from either extinction corrected Hα emission (filled green circles in Fig. 11), or from the sum of UV- and IR-based SFRs (filled yellow circles in Fig. 11). The systematic uncertainties are reported to be ±0.15 dex. We compute the total SFR of g1323 by adding our Hα-based SFR and the dust-obscured SFR and find SFR|$_{\rm total}=6.1^{+4.4}_{-2.6}$| M_⊙ yr⁻¹.

From Fig. 10, we first note that the UV-based SFRs (square symbols) all fall far below the relation, and that g1323 is in the very low range of this wide distribution. Considering now only the Hα-based SFRs (filled circles), they are all seen to lie closer to the relation, and are likely much better SFR indicators for galaxies with large molecular gas-mass fractions. It is seen that most points fall well below the corresponding relation for total gas mass, and increasingly so for larger gas-mass fractions. Still, even for |$f_{\rm mol{\rm -}gas}$| as large as 0.4–0.5, the PHIBSS galaxies fall only marginally below the relation. In contrast, the Hα-based SFR of g1323 falls significantly below it (note that the flux-based uncertainty is linear in SFR, but we are plotting log(SFR) in the figures). While g1323 obviously has a large |$f_{\rm mol{\rm -}gas}$| compared to the COLD GASS galaxies, it is still surprising that it falls so far below the relation followed by the low |$f_{\rm mol{\rm -}gas}$| galaxies, and even far below the PHIBSS galaxies that have larger |$f_{\rm mol{\rm -}gas}$|⁠. Either g1323 could have an even larger hidden SFR than the PHIBSS galaxies, or it simply has an extremely low SFR for its stellar mass and gas fraction.

To find out which is true, we therefore plot, in Fig. 11, the ‘total SFR’ corrected for dust. We do not have the information to compute the dust correction for the COLD GASS galaxies, but total SFRs are available for both sub-samples of the PHIBSS galaxies. The corrected PHIBSS sample is seen now to lie closer to the Santini et al. (2014) relation but it still lies somewhat below and with a scatter slightly larger than the mass slice would predict. In order to obtain a better evaluation of how the relation based on |$f_{\rm mol{\rm -}gas}$| compares to that of the total gas-mass relation, we also include CANDELS galaxies from Popping et al. (2015) for which total SFRs are also known. The molecular gas mass for the CANDELS galaxies has not been measured, rather it has been inferred from a suite of optical data and the model presented in Popping et al. (2015). It is seen that for |$f_{\rm mol{\rm -}gas} < 0.2$|⁠, the CANDELS sample follows the Santini et al. (2014) relation well, whereas it for larger |$f_{\rm mol{\rm -}gas}$| connects perfectly with the PHIBSS sample. The CANDELS sample is very large allowing us to select an even narrower mass slice (10^10.75 ≤ M_*/M_⊙ ≤ 10^10.85) than for PHIBSS. While care must be exercised in not over interpreting the partly model-based CANDELS data points, the general trend of the relation from small to large |$f_{\rm mol{\rm -}gas}$| appears now to be well established. The scatter of the CANDELS sample is seen to be smaller than for the PHIBSS sample, which was expected because the CANDELS galaxies are picked from a narrower mass range. g1323 is again seen to fall well below both samples. We conclude this discussion summarizing our two main results: (i) there is indeed a fundamental relation for (⁠|$f_{\rm mol{\rm -}gas}$|⁠, SFR, and M_*) similar to that of total gas-mass fraction but shifted a little towards larger |$f_{\rm mol{\rm -}gas}$| and, (ii) g1323 lies well below this relation. It therefore appears that g1323, seen in this observed parameter space and stellar mass range, does not have counterparts in flux-selected samples. It is noteworthy that Neeleman et al. (2016b) reported a low SFR and large gas consumption time-scale for g0439.

There are ways to apply reasonable assumptions to force g1323 on to the relation of course, but before we turn to those we emphasize that such assumptions have also not been applied to any of the objects of the comparison samples, so at face value, with the zeroth-order assumptions, g1323 is significantly inconsistent, both considering Hα-based and total SFR, with the relation of the flux-selected samples.

For g1323, we found (Section 3.4.3) that there is indeed evidence for some obscured SFR (as included in Fig. 11). One could hypothesize that even more SFR is hidden by dust, thereby pushing the point further upwards towards the relation. This hypothesis is not supported by the data analysed in Section 3.4.3. Alternatively, one could assume a lower value for α_CO for g1323, which would push the point towards the left. Assuming e.g. the extreme example (Section 3.1) of a starburst galaxy, with α_CO ≈ 1 and r₂₁ ≈ 1, would imply |$f_{\rm mol{\rm -}gas} = 0.05$|⁠. The assumption of such a highly obscured starburst is also not supported by our observations, as described in Section 4.4.

4.3 CO-Tully–Fisher relation

In Fig. 12, we compare our absorbing galaxies to the COLD GASS Tully–Fisher relation (TF, fig. 2b in Tiley et al. 2016). For g0439, we fit the same double-horned profile as for g1323 (Section 3.3) and find the values for W50 and W20 listed in Table 5. Both galaxies are seen to fall to the right of the relation (i.e. they have large velocity widths for their stellar masses). g0439 is fully consistent with the scatter of the COLD GASS sample, but g1323 is seen to fall significantly right of the relation.

As discussed in Section 3.2, there is some evidence that part of the CO line profile (the lowest redshift peak) may be due to a gas component which is separate from the rest of the galaxy. If we instead use W50 = 378 km s⁻¹ from the ‘two object fit’ (Section 3.3) then the green square moves left by 0.23 dex in Fig. 12, which brings it in agreement with the COLD GASS distribution. The ‘outlier’ nature of g1323 in the TF relation would therefore find a simple explanation if the existence of this separate component is confirmed.

4.4 Gas-to-dust mass versus metallicity relation

For local galaxies, it is known that the ratio of gas to dust mass is related to the emission-line metallicity of the galaxy. We did not detect any dust emission continuum in our ALMA data of g0439, but for g1323, we obtained a detection and inferred a dust mass of |$\log [{\rm M}_{\mathrm{dust}}/{\rm M}_{\odot }]= 8.45^{+0.10}_{-0.30}$| (Section 3.4.3). In Fig. 13, we now compare g1323 (large green square) to the sample of Local Group galaxies (small black squares) and local ULIRGs (red stars) compiled in Magdis et al. (2011). In order to compare to the Magdis et al. (2011) sample, we need to first convert our measured emission-line metallicity to the calibration by Pettini & Pagel (2004) and we find [12+log(O/H)]_PP04 = 8.71 ± 0.11. As Magdis et al. (2011, their fig. 3, right), we plot |$L^{\prime }_{{\rm CO}}$| rather than M_gas in order to avoid the additional uncertainty arising from α_CO.

From Fig. 13, we first conclude that g1323 is lying perfectly on this scaling relation for normal flux-selected galaxies. While it is marginally consistent with the position of ULIRGs, the fit to the relation of local galaxies (black line) is much better, a conclusion which is also supported by our result in Section 3.4.3 that this galaxy has a relatively low amount of obscured SFR for its stellar mass, as well as only modest reddening (A_V = 0.4, Section 3.4.1).

4.5 Relation to emission-selected galaxy samples

DLA galaxies make up a sample of galaxies that are almost exclusively first identified in absorption (i.e. absorption-selected). This is an important characteristic to keep in mind, because it means that we do not a priori know if ‘absorption selection’ is a differently weighted process to find the same galaxies as ‘flux-limited selection’, or if it is a way of finding entirely different objects. This question is not easily answered, and yet it is of fundamental importance in order to be able to correctly interpret flux-limited galaxy samples, and in particular to be able to connect them to the vast reservoir of information gathered via absorption-line studies of DLAs. Finding the answer requires that we identify a sample of DLA galaxies in emission, and then that we carry out a detailed study to test if those objects lie within the same parameter space as emission selected galaxies.

Møller et al. (2002) and Weatherley et al. (2005) previously addressed this exact question using a small set of z = 2–3 DLA galaxies observed in optical and NIR emission. The conclusion at that time was that all the measured emission properties were consistent with the conjecture that DLA galaxies are Lyman break galaxies. One of the key purposes of our ALMA survey is now to extend this fundamental test into the ALMA frequency range, where we are able to ask the same question for properties related to the molecular gas. In this paper, we have outlined this procedure, and compared our first two detections to galaxy gas scaling relations determined from flux-limited samples. Here, we provide a short summary of this comparison.

• M_gasversus M_*: both galaxies are rich in molecular gas; notably, g0439 lies on the upper boundary of the known distribution. Both are within the parameter space of flux-limited samples.

• SFR versus molecular gas fraction: Neeleman et al. (2016b) reported a low SFR and large gas consumption time-scale for g0439. Similarly, we have here found that the SFR of g1323 is significantly below the flux-limited relation for galaxies of similar stellar mass and molecular gas-mass fraction, both considering purely Hα-based SFRs, and total (dust corrected) SFRs.

• Tully–Fisher relation: the two galaxies have a large CO line velocity width for their stellar mass. g0439 is here within the standard relation of the COLD GASS sample, but g1323 is also in this respect at best marginally within the flux-limited relation.

• Gas-to-dust versus metallicity: we were only able to obtain a dust mass of g1323. This galaxy lies directly on the known relation of normal (i.e. non-starburst) main-sequence galaxies.

The answer is therefore not as clear as it was in the earlier works where only stellar and ionized gas components were considered. g1323 appears to be the first case of a DLA-selected galaxy that does not have counterparts in flux-limited samples as it fails our consistency test in two out of four relations. This galaxy may simply be a very unique and rare case, but it may in fact represent an ‘in between burst’, or ‘post-starburst’ low SFR phase of galactic evolution. If such a phase follows an event of strong ejection of gas (both molecular and neutral), then the ejected molecular kinematic component could explain the large CO emission-line width, and also explain why galaxies in such a phase are more easily found as DLA galaxies due to their larger cross-section for H i absorption.

Some support for this interpretation may be found in our detailed analysis of the ALMA data presented here which suggests that (i) g1323 consists of two kinematically independent and interacting sub-clouds and (ii) the SED fit to g1323 favours an instantaneous burst of star formation with an age of 0.5 Gyr. The current data do not rule out the possibility that both clouds may have a stellar component, but only one of them is visible on the current (ground-based AO) optical image. An HST image should be able to determine if the additional cloud belongs to a separate and interacting galaxy.

A key conclusion from our previous study was that ‘the strongest molecular absorption component of g0439 cannot arise from the molecular disc’ (Neeleman et al. 2016b) suggesting that also that system has more than one molecular gas component. A larger sample must be collected in order to address how common these type of objects are in DLA-selected samples. Our Cycle 3 data are currently being analysed and they will already significantly enlarge our current sample of detections (Kanekar et al. in preparation) .

5 SUMMARY AND CONCLUSIONS

We have initiated an ALMA survey for CO emission from galaxies known to harbour DLA or sub-DLA absorbers. In this paper, we present an ALMA map of the field around the known DLA absorber towards the quasar J1323−0021. Based on the detection of CO(2–1) line emission at a redshift of z_CO = 0.71645, close to that of the absorber (z_abs = 0.71625), we identify the galaxy responsible for the absorption. This is the second DLA/sub-DLA galaxy in which CO emission has been detected, and it is the highest redshift DLA CO detection to date. This is also the first case of a DLA galaxy identified via molecular gas emission, and as such it highlights a new and promising route to the identification of DLA galaxies. With ALMA, we also detect the dust continuum and therefore measure both the molecular and the dust mass of the host galaxy.

In addition, we present new VLT–FORS2 data and we re-analyse archival VLT–SINFONI data cubes, Herschel maps, and VLT–UVES and HST STIS spectra. This large suite of data allows us to dissect the DLA host galaxy in terms of stellar mass, gas mass, and dust mass; and in terms of kinematical components of the cold (CO) gas, the ionized (Hα and [O ii]) gas and the absorbing gas. Also, we are able to determine the SFR using several different indicators (UV, [O ii], Hα, and IR).

Our results can be summarized as follows:

• The DLA galaxy g1323 is the second most massive DLA-selected galaxy known with log[M_*/M|$_{\odot } ]= 10.80^{+0.07}_{-0.14}$|⁠, log[M|$_{\rm mol{\rm -}gas}$|/M_⊙] = 10.37 ± 0.04, and log[M_dust/M|$_{\odot } ]= 8.45^{+0.10}_{-0.30}$|⁠. This also implies that it is very gas rich, in the upper part of comparable galaxies from flux-limited samples.

• The emission-line metallicity of g1323 is super solar (≈3.2 × solar), fully consistent with previously published absorption metallicities. Its gas-to-dust ratio for this metallicity is perfectly on the relation for normal flux-selected galaxies. It is somewhat below, but within the errors also marginally consistent with local ULIRGs.

• The velocity width of the CO emission is extremely wide for its stellar mass. It falls outside the scatter of the CO TF relation of COLD GASS galaxies. There is some evidence suggesting that the wide CO line could be due to an additional and separate (both dynamically and spatially), interacting CO cloud, e.g. a previous outflow or/and current infall. This cloud could also be related to a separately, yet undiscovered, interacting galaxy.

• We determine the SFR in several different ways and find that for a galaxy with its stellar mass and molecular gas-mass fraction, its SFR is extremely low. With respect to its poor ability to convert molecular gas into stars, this galaxy has no counterparts in flux-limited samples. It is intriguing that a low SFR and large gas consumption time-scale was also found for our first CO-detected absorber.

As in previous (but optically based) studies, we perform the fundamental test if the properties of g1323 are consistent with the conjecture that DLA-selected galaxies and flux-selected galaxies are the same, only selected in different ways. We find that for the first time we have identified a DLA galaxy that does not appear to have counterparts in flux-limited samples. The galaxy g1323 has too low SFR for a galaxy with its stellar mass and gas-mass fraction, and it also has too wide a CO emission line to fit on to the well-known TF relation.

Our survey is ongoing, and our Cycle 3 data have already provided additional detections. If more DLA galaxies of this survey in similar ways do not conform to the relations of emission-selected samples, this could have significant impact on our understanding of the connection between the gaseous and the stellar component of galaxies, and how they relate to star formation. Notably the non-disc molecular gas component of DLA galaxies could form a large previously unexplored reservoir.

ACKNOWLEDGEMENTS

We are grateful to Mark R. Chun for providing the AO K-band image, to Gergely Popping for providing the unpublished CANDELS molecular masses, and to Giorgos Magdis and Celine Peroux for useful discussions. LC and HR are supported by YDUN-DFF grant ID 4090-00079. NK acknowledges support from the Department of Science and Technology via a Swarnajayanti Fellowship (DST/SJF/PSA-01/2012-13). This paper makes use of the following ALMA data: ADS/JAO.ALMA#2013.1.01178.S. ALMA is a partnership of ESO (representing its member states), NSF (USA) and NINS (Japan), together with NRC (Canada), NSC 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 work has made use of the Rainbow Cosmological Surveys Database, which is operated by the Universidad Complutense de Madrid, partnered with the University of California Observatories at Santa Cruz (UCO/Lick, UCSC). This is based on observations at ESO; Programme IDs: 095.A-0890(A)(FORS), 079.A-0673(A)(Sinfoni), 274.A-5030(A)(UVES) and 2013.1.01178.S (ALMA). Herschel is an ESA space observatory with science instruments provided by European-led Principal Investigator consortia and with important participation from NASA.

Footnotes

REFERENCES