Free Access
Issue
A&A
Volume 619, November 2018
Article Number A90
Number of page(s) 21
Section Extragalactic astronomy
DOI https://doi.org/10.1051/0004-6361/201833040
Published online 13 November 2018

© ESO 2018

1. Introduction

The observed scaling relations between the properties of supermassive black holes (SMBHs) and galaxies found in the local Universe suggest a close connection between the galaxy evolution and the growth of the central SMBH (see Kormendy & Ho 2013 for a recent review). Both theoretical and observational arguments (e.g. Alexander et al. 2005; Hopkins et al. 2008; Menci et al. 2008) support a picture, first presented by Sanders et al. (1988), where a dusty, starburst (SB)-dominated system arising from a merger evolves into an optically luminous quasar (QSO). The QSO phase would emerge after the accreting SMBH experienced the “feedback” phase (see e.g. Hopkins et al. 2008), invoked to reproduce the host galaxy-SMBH scaling relations (but see also e.g. Jahnke & Macciò 2011). During this feedback phase, the active galactic nucleus (AGN) releases radiative and kinetic energy in the form of powerful outflowing winds which may be capable of clearing the entire galaxy of dust and gas thereby causing star formation in the host to cease (see Somerville & Davé 2015 for a recent review). We refer to this evolutionary picture as the SB-QSO evolutionary sequence.

In the past decade, the properties of cold molecular gas reservoirs of galaxies have been the subject of intense investigation. This gas phase is regarded as the immediate fuel from which stars form, and can be used to test the SB-QSO evolutionary sequence (e.g. Kennicutt & Evans 2012; Leroy et al. 2008). Molecular gas studies, mostly involving observations of carbon monoxide (CO) emission lines, have been carried out in SB-dominated galaxies (e.g. ultra-luminous infrared galaxies, ULIRGs, at low redshift, and submillimetre galaxies, SMGs, at higher redshift), unobscured QSOs, and in normal star-forming galaxies (SFGs; e.g. Carilli & Walter 2013; Daddi et al. 2010; Tacconi et al. 2013; Aravena et al. 2014). These studies show that optically luminous QSOs are characterised by low molecular gas content with respect to their current star formation rate (SFR), or alternatively by high star formation efficiency (SFE), defined as the rate of star formation per unit of molecular gas mass, SFE = SFR/Mgas or SFE = LIR/LCO 1, in the range between a few 100 L/(K km s−1 pc2 ) (units omitted from now on for simplicity) and ≈1000 (see e.g. Fig. 4 in Feruglio et al. 2014 for an early compilation). On the other hand, normal star-forming galaxies and SBs display SFE values in the range from ∼10 to 200, with the latter population generally lying at the upper end of this range (Feruglio et al. 2014). We note that the SFE of individual classes of sources is roughly constant with redshift at z > 1 (e.g. Sargent et al. 2014; Feruglio et al. 2014; Schinnerer et al. 2016; Tacconi et al. 2018; Yang et al. 2017). Therefore, the characteristic SFE-ranges for optically luminous QSOs and star-forming galaxies quoted above are effectively applicable at all z > 1 (see Sect. 4.1 for more detailed discussion).

According to the SB-QSO paradigm, the SFE of unobscured QSOs should be significantly higher than for main sequence (MS; Speagle et al. 2014) galaxies and SBs because of the AGN feedback which expelled the gas from their host galaxies during previous phases. In particular, during the initial stages of the merger, the SFR is expected to be a factor of a few higher (up to 10) with respect to normal, isolated galaxies (Di Matteo et al. 2005, 2007; Hopkins 2012). This translates to an enhancement of the SFE by the same amount (Di Matteo et al. 2007). An even higher SFE could be associated with the subsequent phases, i.e. the feedback phase and in particular the QSO phase, because the CO luminosity reflects the current (depleted) amount of molecular gas, while the IR emission responds to the associated decrease in SF more slowly than the outflow timescales (see Sect. 2; Fig. 1).

thumbnail Fig. 1.

Schematic diagram to illustrate the standard SB-QSO evolutionary sequence. In the first two panels, the blue ellipse and yellow star represent the galaxy and the central SMBH, respectively; the light and dark grey areas refer to inflowing gas and dust. Dark blue clouds with yellow outlines indicate outflows. In the last panel, the yellow star and ellipse represent the SMBH and the QSO emission. In the first phase (left), the central regions are fuelled by huge amounts of gas and dust, and the SMBH is highly obscured. These systems, originating from gas-rich galaxy mergers, are commonly identified as SMGs, and are associated with star formation efficiencies of ∼100 ÷ 200. The SFE of CT/highly obscured QSOs is mostly unknown. These systems, if tracing the same evolutionary stage, should have SFEs similar to those of SMGs. In the last phase (right), the system is revealed as an unobscured QSO (bright enough to easily outshine its host galaxy). In this phase its gas reservoir is expected to be strongly depleted. The SFE measured in optically luminous QSOs can span a wide range, from a few 100 to ∼2000. In the transition phase (middle), galaxy-wide outflows start to remove the dust and gas reservoirs (which only partially obscure the nuclear regions). We observe these systems as obscured QSOs (generally associated with signatures of outflows). For these systems, we naively expect to observe intermediate SFE between SMGs and unobscured QSOs.

Open with DEXTER

In the SB-QSO evolutionary sequence, obscured AGN are associated with a transitioning phase between the SB- and QSO-dominated stages so that the higher the obscuration, the larger the gas reservoirs in the host galaxy. In this framework, the obscuration is due to galaxy-scale absorbers rather than a nuclear dusty torus on the line of sight. Although this argument is still debated and might not be applicable to all the obscured systems, there is some observational evidence confirming the presence of galaxy-scale absorbers in obscured QSOs (see e.g. Lanzuisi et al. 2017 and references therein; Gallerani et al. 2017; Gilli et al. 2010; 2014; Rosario et al. 2018; Symeonidis 2017). This is suggested, for example, by the correlation between the nuclear column density (NH) and the amount of gas traced by the CO on kpc-scales (e.g. Rosario et al. 2018), and by the fact that in a few high-z CT AGN the absorber can be fully accounted for by the amount of CO gas and its compactness (e.g. Gilli et al. 2014). Moreover, numerical simulations by Hopkins et al. (2016) also show that typical nuclear conditions of post-merger objects are generally associated with isotropically obscured SMBHs with log(NH/cm−2) ≈ 24 − 26. Only when the AGN-driven winds start, a cavity is generated in the direction perpendicular to SMBH accretion disk and the nuclear regions can be observed as a mildly-to-modestly obscured AGN (see also e.g. Hopkins 2005).

The study of the molecular gas conditions in different classes of sources may play and important role to test the SB-QSO evolutionary sequence. In this regard, scarce observations have been carried out for heavily obscured AGN at the peak epoch of cosmic SF activity and SMBH growth (z∼ 2; Cimatti et al. 2006). The identification of distant, buried QSOs is particularly challenging since the high gas column densities absorb their emission below rest frame 10 keV, so that they might be largely missed even in the deepest hard X-ray surveys available today. Because of this limitation, only a few high-z highly obscured (NH > 1023 cm−2) quasars have been targeted in CO millimetre studies (e.g. Aravena et al. 2008; Coppin et al. 2008, 2010; Polletta et al. 2011).

The purpose of this work is to review our understanding of the molecular gas emission in highly obscured QSO host galaxies. We present the Jansky Very Large Array (JVLA) and the Plateau de Bure Interferometer (PdBI) observations of three obscured QSOs at z ∼ 2. We also take advantage of the available information for z > 1 buried systems in the literature. We test the aforementioned evolutionary scenario, i.e. whether these systems can be considered as the progenitors of optically luminous QSOs, comparing their star formation efficiencies (see Fig. 1). We will also compare the cold gas reservoir of obscured AGN with those of normal MS and SB galaxies with the same stellar mass and redshift to unveil the presence of potential feedback effects.

The paper is organised as follows. In Sect. 2 we present the targets analysed in this work, discussing their properties derived from available multiwavelength information and presenting the PdBI and JVLA observations and data analysis. Section 3 presents the samples of normal and submillimetre galaxies, unobscured and obscured QSOs collected from the literature. In Sect. 4 we show the SFE and gas fractions of the different classes of sources, and discuss these properties in the context of the galaxy evolution paradigm, also taking into account their host galaxy properties. Finally, we summarise our results in the Conclusion Section. A flat universe model with a Hubble constant of H0 = 67.7 km s−1 Mpc−1 and ΩM = 0.307 (Planck Collaboration XIII 2016) is adopted. We adopt a Chabrier initial mass function to derive host galaxy properties for the targets presented in this work and for the comparison samples. For the X-ray detected sources reported in this paper, we distinguish between mildly obscured (with 21 ≲ log(NH/cm−2)< 22), modestly obscured (22 ≲ log(NH/cm−2)< 23), highly obscured (23 ≲ log(NH/cm−2)< 24) and CT AGN.

2. The targets

The identification and classification of highly obscured and CT AGN are challenging tasks. In fact, X-ray spectroscopic analysis, the most robust method of identifying these systems, is only feasible for a small number of sources because it requires high-quality spectra. These sources have been detected using deep surveys by XMM-Newton (e.g. Comastri et al. 2011; Lanzuisi et al. 2015), Chandra (e.g. Brightman et al. 2014; Liu et al. 2017), and Swift (e.g. Ricci et al. 2017). The first publications from the NuSTAR survey programs are now further broadening the number of highly obscured sources in the Universe (e.g. Brightman et al. 2015; Farrah et al. 2016; Marchesi et al. 2018; Masini et al. 2016, 2018; Ursini et al. 2017).

Moreover, given the expected low spatial density of CT AGN in the 2–10 keV band, large area surveys are needed to collect sizeable samples. For this reason, alternative multiwavelength selection techniques based on high-ionisation, narrow optical emission lines (Vignali et al. 2010; Gilli et al. 2010) or on the ratio between mid-infrared and optical fluxes (Daddi et al. 2007; Eisenhardt 2012; Fiore et al. 2008, 2009) have been developed in recent years. They seem to represent a promising approach for selecting sizeable samples of highly obscured/CT AGN at z ∼ 1 − 3 (see e.g. Perna et al. 2015b; Piconcelli et al. 2015; Vito et al. 2017). However, they rely on indirect methods, and may still be prone to contamination by less obscured AGN or galaxies (e.g. Donley et al. 2008; see Brandt & Alexander 2015 for a recent review).

The sources analysed in this work were selected from the Feruglio et al. (2011, BzK4892 and BzK8608) and Comastri et al. (2011, CDF153) works on the basis of the presence of high equivalent width FeK emission in the X-ray (EW ∼ 1 − 2 keV), which constitutes unambiguous evidence for the presence of huge obscuring column densities along the line of sight. This distinguishes them from other samples of obscured AGN that have been studied in the literature, which were selected using multiwavelength selection techniques (e.g. Banerji et al. 2017; Brusa et al. 2018) or the X-ray hardness ratio (e.g. Polletta et al. 2011). The properties of the three CO follow-up targets analysed in this work are reported in Table 1.

Table 1.

Targets properties.

Our targets have also been observed as part of the 7 Ms Chandra Deep Field-South Survey (CDFS; Liu et al. 2017). The new X-ray spectral analysis2 has confirmed the presence of strong iron lines in two out of three sources (BzK4892 and CDF153; see Table 4 in Liu et al.) and their high obscuration with column density of NH ≈ 6 × 1023 cm−2 for BzK8608 and CDF153, and of NH ≈ 3 × 1024 cm−2 for BzK4892. We note that the CT QSO BzK4892 would also be selected by the Fiore et al. (2008) selection criteria as a candidate CT source (see Table 1).

2.1. Host galaxy properties from SED fitting

In order to derive the host galaxy properties of the three targets, we collected the photometric information from the UV to mid-infrared catalogue in the CANDELS/GOODS-S field, published in Guo et al. (2013), and from a new Herschel catalogue (Wang et al. in prep.) which utilises an optimised algorithm for photometry in crowded Herschel maps (see Appendix A). ALMA continuum observations at 250 GHz (Ueda et al. 2018) and 340 GHz (Elbaz et al. 2018) are also used to better constrain the mm emission of BzK4892. ALMA band 7 observations obtained as part of the programs 2013.1.00884.S (PI: Alexander D.) and 2012.1.00869.S (PI: Mullaney J.R.) are used instead to derive 3σ upper limits on the host galaxy dust emission at 340 GHz for BzK8608 and CDF153, respectively. As the listed uncertainties on broad-band photometry in published catalogues are usually underestimated (i.e. only telescope uncertainties are usually quoted), we increase the nominal uncertainties by a factor of 2 (see also e.g. Shangguan et al. 2018) in order to take into account systematic errors due to confusion limit or calibrations issues for example (e.g. Nguyen et al. 2010).

To model the observed spectral energy distribution (SEDs), we adopted the publicly available AGNfitter algorithm (Calistro Rivera et al. 2016), which implements a Bayesian Markov chain Monte Carlo (MCMC) method to disentangle the different physical components contributing to the observed SED. The data points have been fitted with a combination of (1) a stellar component, to account for host galaxy contribution; (2) two black hole components, referred to as the “mid-IR torus” and big-blue bump (BBB); and (3) a far-IR to sub-mm emission associated with dust-obscured star formation activity in the host galaxy.

We refer the interested reader to Calistro Rivera et al. (2016) for details on the modelling of the main components described above and the description of the AGNfitter algorithm. Here, we briefly mention that the stellar emission is modelled with stellar population models of Bruzual & Charlot (2003) convolved with exponentially declining star formation histories (SFH), and that stellar templates can be reddened according to the Calzetti et al. (2000) reddening law. The AGN dust emission models are taken from Silva et al. (2004) and consider different templates for unobscured/mildly obscured AGN, modestly obscured, highly obscured and CT AGN. The torus models associated with higher obscurations are characterised by more absorbed near-IR emission and mild silicate absorption at 9.7 μm (see Fig. 1 in Calistro Rivera et al. 2016). The BBB emission is modelled using a modified version of the empirical template for type 1 SDSS QSOs constructed by Richards et al. (2006) and considering the possible extinction by applying the SMC reddening law (Prevot 1984) to the BBB template. Both the reddening values E(BV)gal and E(BV)BBB range from 0 to 1. Finally, the cold dust emission is modelled using the semi-empirical starburst template libraries by Chary & Elbaz (2001) and Dale & Helou (2002).

Spectroscopic redshifts are available, and are used as input for the SED decomposition with AGNfitter, for BzK4892 and CDF153. They are based on VLT/FORS follow-up (Szokoly et al. 2004). In the case of BzK8608 we adopt a photometric redshift value of (Appendix B).

The SED fitting was performed imposing log(NH) > 23, based on the available information from the X-ray spectroscopic analysis. From the best-fit models shown in Fig. 2, we obtained stellar masses of log(Mstar/M) ≈11.0 − 11.3, star formation rates in the range 70 − 620 M yr−1, and total (8 − 1000 μm) stellar infrared luminosities LIR ≈ (2 − 60)×1011 L (see Table 1 for individual measurements).

thumbnail Fig. 2.

SED fitting of BzK4892 (top left panel), BzK8608 (top right panel), and CDF153 (bottom left panel). The black dots represent the observed data points from the Guo et al. (2013) and Wang et al. (in prep.) catalogues; 5σ upper limits are shown with black triangles. The data points at log(ν) ∼12 are from ALMA continuum observations at 250 GHz and 340 GHz (see Sect. 2). The best-fit SED obtained with AGNfitter (Calistro Rivera et al. 2016) and the different components used to model the SED are shown as labelled in the figure. Ten randomly picked realisations from the posterior distribution function are overplotted in the figures in order to visualise the dynamic range of the parameter values.

Open with DEXTER

We note that the BzK4892 SED best fit is clearly reproducing the far-IR and the optical emission from which LIR and Mstar measurements are derived. The presence of a BBB emission at the bluest wavelengths is instead not well constrained and is responsible for the 0.15 dex uncertainty on the stellar mass estimate. The stellar mass of BzK8608 and CDF153 are instead better constrained (with ∼0.1 dex uncertainties); on the other hand, their IR luminosities are associated with larger uncertainties (a few ∼0.1 dex), being these sources much fainter and more affected by blending in the far-IR (Appendix A).

Combining the stellar mass and SFR values derived with AGNfitter, we can compare the specific SFRs (sSFR = SFR/Mstar) of our targets with typical values of main sequence galaxies. BzK4892 lies slightly above the MS in the SFR-Mstar plane at z ∼ 2.5 (Speagle et al. 2014) and can be classified as a normal MS galaxy (see Table 1; in this work, we classified as starburst all the sources whose sSFR is 0.6 dex above the MS).

High-resolution continuum images at 870 μm obtained with ALMA showed that the dust in BzK4892 host galaxy is highly concentrated, i.e. confined inside the galaxy extent seen in the HST rest frame V band (Barro et al. 2016; Elbaz et al. 2018, with a V band effective radius of ∼0.5 − 0.8 kpc). Coupling the information on the spatial extent observed in the dust continuum image and the SFR inferred from the SED fit, we derive a star formation rate surface density ΣSFR ∼ 140 M yr−1 kpc−2, similar to those observed in other SMGs (see e.g. Gilli et al. 2014; Yang et al. 2017). Therefore, BzK4892 is plausibly experiencing a very efficient and compact mode of star formation, as expected for merger-driven SB systems (e.g. Hopkins et al. 2006).

BzK8608 and CDF153 have specific star formation rates compatible (within the errors) with the values expected for MS galaxies at their stellar mass and redshift, according to the relation of Speagle et al. (2014), and are therefore associated with MS host galaxies.

2.2. Millimetre observations

BzK4892 and BzK8608 were observed with JVLA (10–12 January 2012), in the DnC configuration, as part of the VLA/11B-060 program (PI: Daddi E.). Observations were carried out with receivers tuned to ∼32.5 GHz, corresponding to the expected frequency of the redshifted CO(1-0) emission line, for a total time on source of 4.4 h. An additional window at ∼28.3 GHz was observed in order to attempt a continuum detection in combination with line-free channels in the upper spectral window. Both spectral windows cover a bandwidth of 1 GHz each. We reduced the data with the CASA VLA Calibration pipeline.

CDF153 was observed with PdBI in 2012 with the array set to the C configuration as part of the program W041 (PI. Feruglio C.); the receivers were tuned to the frequency of 91.1 GHz, expected for the CO(2-1) line. The spectral set-up covered a total bandwidth of 2 GHz. The PdBI data were calibrated and imaged with the GILDAS CLIC and MAPPING3 software. The on-source time was 4.8 h (equivalent to six antennas).

The QSO J0329−2357 (for BzK sources) and 0338-214 (CDF153) have been used as phase calibration sources. We used 3C 48 (for BzK sources) and MWC349 (for CDF153) for absolute flux calibrations, which yield an absolute flux accuracy of ∼10%. The clean beam of the VLA and PdBI observations are reported in Table 2. We note that, due to the low declination of CDF153, the associated PdBI beam is highly elongated.

Table 2.

JVLA and PdBI observations.

After flagging poor visibilities, the reduced data sets reached noise levels of 0.04 mJy beam−1 per 40 MHz channel for BzK4892, 0.03 mJy beam−1 per 40 MHz for BzK8608 and 0.29 mJy beam−1 per 120 MHz channels for CDF153 (corresponding to ≈400 km s−1, for a“natural” imaging scheme).

These sensitivities were not sufficient to detect the CO line or continuum emission in our targets. However, we were able to infer a stringent 3σ upper limit to the LCO of 1.6 × 1010 K km s−1 pc2 for BzK4892, following Carilli & Walter (2013), and considering typical AGN FWHMCO of 400 km s−1 (e.g. Carilli & Walter 2013; see also Decarli et al. 2018). This upper limit is also consistent with the CO(4-3) luminosity derived from ALMA observations, LCO(4 − 3) = (1.3 ± 0.2)×1010 K km s−1 pc2, which will be presented in an upcoming paper (D’Amato et al. in prep.). From the LCO upper limit we derive a gas fraction fgas = Mgas/(Mstar + Mgas)< 10%, assuming that Mgas = αCOLCO = 0.8LCO (e.g. Bothwell et al. 2013; Carilli & Walter 2013) and that the interstellar medium (ISM) is dominated by molecular gas. We note that the upper limit on the gas fraction we derived from the JVLA observations of the CO(1-0) line is a factor of 4 lower than the values reported by Elbaz et al. (2018) and Barro et al. (2016), derived from dust continuum emission assuming dust-to-gas and mass-metallicity relations (Magdis et al. 2012; Scoville et al. 2016). We suggest that the discrepancy between the different estimates of fgas is due to the large uncertainties still affecting the derivation of gas mass from dust continuum observations. Our CO-based estimate of the molecular gas mass is not dependent on any excitation ratio correction, while the luminosity-to-gas-mass conversion factor αCO = 0.8 was chosen on the basis of the remarkable compactness of the dust continuum emission in the system (Barro et al. 2016; Elbaz et al. 2018). On the other hand, any dust-based estimate of Mgas will strongly depend on the assumed metallicity and dust-to-gas ratio.

An alternative interpretation is given by the fact that CO- and dust-based molecular mass measurements can be associated with different timescales: while the CO luminosity is related to the current amount of molecular gas in the host, the production of dust is linked to stellar processes acting over timescales of the order of ∼100 Myr (e.g. Gall et al. 2011). Therefore, the discrepancy between the two estimates can also be explained assuming a rapid decrease in the gas reservoirs in the host galaxy, i.e. assuming recent/ongoing outflow episodes.

For the other two targets, the derived upper limits on the CO(1-0) luminosity are LCO < 2.8 × 1010 K km s−1 pc2 for CDF153, assuming a 0.8 excitation correction CO(2-1)/CO(1-0), and LCO < 1.5 × 1010 K km s−1 pc2 for BxK8608. Both sources are associated with gas fractions fgas ≲ 50%. Despite the presence of an AGN, for these two targets we conservatively assumed αCO = 3.6, the common value adopted for MS galaxies (e.g. Carilli & Walter 2013). The constraints on the gas fractions decrease to ≲20% if αCO = 0.8 is assumed instead (see e.g. Brusa et al. 2018; Popping et al. 2017 for AGN-hosting MS galaxies with αCO < 1).

We note that for BzK8608, the derived quantities should be considered tentative. Given that no unambiguous spectroscopic redshift has been obtained for this target (see Feruglio et al. 2011; Appendix B), all our measurements were inferred after adopting the photometric redshift derived in Appendix B. We also note that it is possible that the CO(1-0) might fall outside the receiver tuning range, which does not cover the entire 1σ interval associated with our photometric redshift estimate (see Fig. B.1).

3. Literature samples

In order to explore the conditions of molecular gas in high-z galaxies and to test the SB-QSO evolutionary scenario, we combined our new observations with data from the literature, considering different classes of objects at z > 1, from normal galaxies and SMGs to obscured and unobscured QSOs. In particular, we consider:

In Appendix C we tabulate the information collected separately for each individual class of sources and provide stellar mass and MS-offset (sSFR/sSFRMS) distributions for all the sources for which stellar masses are available. These systems are associated with massive host galaxies, with Mstar ≈ 1011 M (with a 1σ dispersion of 0.4 dex). We note that the collected SMGs have been selected in the different papers according to their infrared luminosity > 1012 L, and that for many of them the Mstar is unknown; therefore, this sample can include a heterogeneous collection of SB and normal MS galaxies. In the next sections we combine their CO(1-0) luminosities with total IR luminosities and stellar masses to investigate the ISM conditions of the different types of galaxies, also taking into account the presence and the obscuration of accreting SMBHs.

While the ground-state transition CO(1-0) is the preferred emission line for tracing host galaxy molecular gas reservoirs, in many cases we only have access to observations of CO emission lines with Jup > 1 (Tables C.1, C.2, C.3). Converting these to an estimate of the CO(1-0) luminosity requires knowledge of the spectral line energy distribution (SLED), from which it is possible to derive the luminosity ratios between CO states with Jup > 1 and the CO(1-0). However, multi-J CO observations reveal a significant diversity of SLED shapes (e.g. Bothwell et al. 2013; Brusa et al. 2018) caused by varying ISM conditions in different systems (e.g. Bothwell et al. 2013); as a result, the conversion to CO(1-0) emission can be highly uncertain when the host galaxy conditions are unknown.

To avoid assumptions for the excitation correction, we used the CO(1-0) measurements reported in Sharon et al. (2016) for the five sources previously observed in Jup > 1 CO lines by Coppin et al. (2008) and Bothwell et al. (2013). For all the other objects, the CO(1-0) luminosities used in this work are reported as inferred and tabulated in the different papers, therefore without uniform assumptions about the excitation correction. In fact, as our collected sample is based on different classes of sources, the use of a unique factor would introduce large uncertainties.

Apart from the above-mentioned issues related to the inferred CO(1-0) luminosities, caution is required here also for the (8 − 1000 μm) infrared luminosities since for most of the collected targets the LIR determination is not self-consistent. Generally, IR luminosity is obtained from SED fitting analysis, distinguishing between AGN and galaxy contributions when complete multiwavelength data are available (e.g. Brusa et al. 2018; Kakkad et al. 2017) or when using a modified greybody model for given temperature and dust emissivity (e.g. Banerji et al. 2017). The reliability of the derived estimates also depends on the availability of (deep) infrared and submillimetre observations of individual targets (see Notes in Table C.2).

Finally, we note that a small fraction of optically luminous QSOs and SMGs are associated with lensed systems (13 SMGs from Yang et al. 2017 and Fudamoto et al. 2017, and 15 QSOs from Carilli & Walter 2013). All lensed galaxy luminosities have therefore been corrected for lens magnification factors, which may be affected by large uncertainties (see e.g. Feruglio et al. 2014; Perna et al. 2018). We are confident, however, that the LIR we report in this work are generally a good approximation of the dust-emission related to SF. In Sect. 4.1.2, we nevertheless discuss what impact systematics errors in LIR measurements could have on our results by i) considering only a subsample of sources whose IR luminosities have been derived distinguishing between AGN and galaxy contributions, and ii) exploring a scenario where IR luminosity is significantly contaminated by AGN-related emission, in keeping with Duras et al. (2017).

4. Results

4.1. Star formation efficiencies and AGN obscuration

In Fig. 3 (left) we report the CO(1-0) and infrared luminosities for all four classes of targets mentioned above, indicating with star symbols the sources associated with mildly obscured to highly obscured and CT QSOs (the symbols are scaled so that the largest size corresponds to the highest column density; see inset, right panel). To our knowledge, the hyper-luminous galaxies hosting QSOs from (Polletta et al. 2011, SW022550 and SW022513), the SMG/QSO from Coppin et al. (2010, XID403), and the MS galaxy GMASS953 (Popping et al. 2017; Talia et al. 2018) are the only z > 1 CT sources for which molecular gas has been studied5. The figure highlights the differences between normal galaxies, which follow the LIRLCO relation for MS galaxies (solid line, as reported by Sargent et al. 2014), and the obscured QSO population, whose sources are generally below the locus of MS galaxies. In general, optically luminous QSOs are further below the relation. From this plot we note that the majority of SMGs in our sample follow the relation of MS sources.

thumbnail Fig. 3.

Left panel: observed correlation between measures of CO(1-0) luminosity, LCO, and (8 − 1000 μm) IR luminosity, LIR, for the samples of normal galaxies, SMGs, and QSOs at z > 1. For the three obscured QSOs described in this paper, we report the 3σ upper limits, as labelled. Open symbols refer to main sequence galaxies and SMG, filled symbols refer to QSOs. Where column density measurements are available, scaled star symbols indicate mild to high NH values, as labelled in the right panel inset. Olive green symbols are the galaxies from Sargent et al. (2014); green symbols refer to SMGs from Bothwell et al. (2013), Coppin et al. (2008), and Silverman et al. (2015); red symbols refer to obscured targets. Optically bright QSOs are represented with blue dots. The solid curve represents the best-fit relation derived by (Sargent et al. 2014, 1σ scatter indicated with the grey area), for the MS galaxy sample they collected in the redshift range 0< z ≲ 3. Right panel: SFE vs. redshift for the same sample. In this panel, we labelled the names of the four sources from the literature associated with CT QSOs. The solid curve shows the 2-SFM predicted evolutionary trend of MS galaxies of Mstar = 1011 M (shaded area: illustration of the scatter around the mean evolutionary trend for galaxies with SFR values 1σ above/below the MS). The long-dashed curve is located at 10-fold higher SFE than the mean MS evolutionary trend. Representative 1σ errors on individual data points are shown in the bottom left corner of the plots. For the two CT QSOs SW022513 and SW022550, Polletta et al. (2011) indicated a wide range for their IR luminosities (see Notes in Table C.2); these sources are therefore plotted in the left panel at the centroid of the IR range, while horizontal dotted lines show the IR luminosity interval. In the right panel we similarly reported the two sources at the centroid positions and indicated the SFE ranges with vertical dotted lines.

Open with DEXTER

The three obscured AGN presented in this paper are also shown in the figure. Non-stringent upper limits on the CO luminosities are associated with BzK8608 and CDF1536; the upper limit for BzK4892 is instead located well below the LIRLCO relation for MS galaxies.

In Fig. 3 (right) we show the SFE as a function of redshift for the same compilation of MS galaxies, SMGs, and AGN at z > 1. Unobscured AGN show SFE in the range between 100 and 2000, while MS galaxies (SMGs) have SFE ≲ 100 (≲ 200). The figure also shows the obscured AGN (red symbols) and the SMGs with AGN activity (filled green symbols), with SFE in the range between ≈100 and 1000. The figure shows no clear trend with redshift for the SFE of individual classes of targets; consistent the slow SFE-evolution at z > 1 already discussed in the Introduction. In particular, it has been observed that the SFE of MS galaxies increases in the redshift range z ∼ 0 − 1, and then flattens at z > 1 (e.g. Sargent et al. 2014; Santini et al. 2014). This trend has been confirmed up to z ≲ 3; at higher redshifts, CO observations of normal MS galaxies are still scarce (e.g. Dessauges-Zavadsky et al. 2017; Magdis et al. 2017), and the SFE evolution is thus primarily constrained by continuum-based gas mass estimates (e.g. Schinnerer et al. 2016) which suggest merely slow evolution also beyond z ∼ 3. SFE measurements of more luminous star-forming galaxies, i.e. starburst ULIRGs and SMGs (with LIR > 1012L), have instead been derived for large samples of targets both at low-z and in the distant Universe (out to z ∼ 6) and, similarly to MS galaxies, do not display a strong variation with redshift at z > 1 (e.g. Aravena et al. 2016; Feruglio et al. 2014; Magdis et al. 2012; Sargent et al. 2014; Yang et al. 2017). Instead, the SFE evolution of unobscured AGN populations is not well known, with CO follow-up focusing mostly on lensed targets at z ∼ 2 − 3 and on very distant (z > 4) sources. As unobscured AGN display a wider range of SFE values than MS galaxies and SMGs, the available CO spectroscopy does not reveal any clear trends of SFE with redshift for this class of objects.

In the figure we also show the expected average evolutionary trend of MS galaxies of Mstar = 1011 M from the 2-SFM (two-star formation mode) framework of Sargent et al. (2014). The 2-SFM was introduced in Sargent et al. (2012), and takes advantage of basic correlated observables (e.g. the MS relation and its evolution with z; Speagle et al. 2014) to predict the SFE (and molecular gas content; see Sect. 4.2) of MS galaxies at different redshifts. We used a CO-to-gas conversion factor αCO = 3.6 (e.g. Daddi et al. 2010) and the LIR−SFR relation of Kennicutt (1998) to adapt Eq. (22) from Sargent et al. (2014) to our empirical definition of SFE. We also considered the redshift evolution of the MS relation derived by Speagle et al. (2014). The weak 2-SFM trend further emphasises that the redshift evolution of the SFR (and of the SFE) of MS galaxies is not responsible for the heterogenous distributions shown in Fig. 3 (right) for our different target populations.

On the other hand, values of SFE > 100 can be easily explained by assuming the presence of starburst activity, i.e. considering SFR higher than for MS galaxies (thus enhancing LIR), or by invoking feedback phenomena which depleted the cold gas reservoirs (thus reducing LCO). As an example, we reported in the figure the predicted trend obtained by assuming a SFR (LCO) a factor of 10 higher (lower) than a typical MS galaxy.

4.1.1. SFE-contrast between star-forming galaxies and AGN

As already noted by several authors, according to the SB-QSO evolutionary scenario, when an obscured QSO is detected at sub-mm/mm wavelengths, it could be in the transition phase from SMG to an unobscured QSO (e.g. Coppin et al. 2008). The high SFE values reported in Fig. 3 for these sources are consistent with this scenario. In order to better visualise this result, we report in Fig. 4 the cumulative distributions of SFE for three different subsets of our sample: (1) the MS and submillimetre galaxies without evidence of AGN, combined into a single sample (hereafter “SFG sample”), as MS and SMGs are associated with similar SFE (see Fig. 3); (2) the SMGs with AGN activity (from Bothwell et al. 2013; Coppin et al. 2008) and the obscured AGN (hereafter referred to as “dusty AGN” for simplicity); and (3) the optically luminous AGN. We also divided the sources into three redshift intervals, 1 < z < 2, 2 < z < 3, and z > 3, to test the presence of a possible redshift evolution in the properties of AGN targets. The cumulative distributions reported in Fig. 4 highlight the stark differences between star-forming systems and unobscured AGN, with values of SFE > 200 for the large majority (∼80%) of luminous AGN and only for a small fraction (∼20%) of star-forming galaxies. Instead, the SFE distribution of the population of dusty AGN is similar to that of unobscured AGN.

thumbnail Fig. 4.

First three panels: cumulative distributions of SFE for the samples of star-forming galaxies (green line), unobscured (blue), and dusty (red) QSOs in the three redshift intervals specified along the upper edge of the figures. All targets collected from the literature (for which SFE measurements can be derived) are used to construct the cumulative distributions; in the first panel, we do not report the cumulative distribution of unobscured AGN (our sample only includes two of these objects at 1≤ z ≤ 2). The figures show the clear separation in SFE between AGN and SFGs. Right panel: cumulative distribution of SFE for the same three samples, considering the entire redshift range covered by our targets. Short-dashed red line: cumulative SFE distribution of dusty AGN obtained by reducing by 60% all their IR luminosities to test for the impact of cool, AGN-heated dust (e.g. Duras et al. 2017; see Sect. 4.1.2 for details). Long-dashed red line: distribution of dusty AGN for which a high-quality LIR has been measured through an SED-decomposition into AGN and galaxy emission.

Open with DEXTER

In order to quantitatively compare the three subsamples, we use the Kolmogorov–Smirnov test (KS test; Press et al. 1992). The non-parametric KS test measures the probability that two data sets are drawn from the same parent population (“null hypothesis”). The KS statistic for the three redshift intervals is D ≈ 0.6, which corresponds to a null hypothesis probability rejected at level ∼90 − 99.8% for the comparisons between the SFE of SFGs and unobscured AGN and also between SFGs and dusty AGN7. Overall, the KS tests thus confirm the difference between star-forming systems and AGN. Instead, unobscured and dusty AGN may be associated with the same parent population (D ≈ 0.4, with a null hypothesis rejected at level ∼75%), at least at z > 2 where we have enough unobscured AGN to perform a statistical test.

The cumulative distributions of the SFGs show a clear evolution with redshift; this can be due to both the slow SFE evolution highlighted in Fig. 3 and to an observational bias (at higher redshifts the sample is dominated by SMGs rather than normal galaxies). The medians of the SFG cumulative distributions evolve from ∼80 to ∼130 between the first and last redshift bin shown in Fig. 4. This is the joint effect of the slow SFE evolution of MS galaxies (see Fig. 3 and associated discussion in the text) and an observational bias towards moderately high-SFE starburst SMGs (which represent a higher fraction of our SFG sample relative to MS galaxies at the highest redshifts). Our measurement of the significance of the SFE-gap between the AGN and SFG populations is thus a conservative estimate, as it would widen further if (at the highest redshift) our SFG sample contained similar fractions of MS and SB galaxies as at z ∼ 1. Dusty and unobscured AGN samples instead suffer from poor statistics and no clear trend can be observed. Given the similar SFE-offsets between AGN and SFGs at all redshifts, we constructed the total cumulative distributions of SFGs, dusty AGN and unobscured AGN in the entire redshift range covered by our literature sample (Fig. 4, right). Also in this case, the KS statistic is D ≈ 0.6 (with a null hypothesis probability rejected at level ≫99.9%) for the comparisons between the SFE of SFGs and unobscured AGN and also between SFGs and dusty AGN, and D = 0.25, with a null hypothesis rejected at level ∼75%, for the comparison between unobscured and dusty AGN.

The position of the dusty AGN in the SFE cumulative distributions and in the SFE-z plane could suggest that their gas reservoirs have already been affected by feedback. In fact, the systems with low to moderate column densities could be associated with sources in which the feedback processes have already started to clean the line of sight toward the nuclear regions. This argument can be confirmed for the submillimetre-detected AGN in Coppin et al. (2008), for which evidence of ionised outflows have been obtained by the authors from the analysis of near-infrared spectra. Strong, multiphase outflows have also been detected in the obscured QSO at z = 1.59, XID2028 (Brusa et al. 2015; Cresci et al. 2015a; Perna et al. 2015a), with SFE ≈ 250 and atypical gas consumption conditions (i.e. very low molecular gas fraction, fgas ∼ 4%; see Brusa et al. 2018).

Figure 3 also shows that two out of the five CT sources in our sample (BzK4892 and XID403) are strongly shifted to low values of CO luminosity (and high SFE). As discussed in the Introduction, however, we would expect such systems to lie on the LIRLCO relation of MS galaxies (and have SFE ≲ 100 − 200) according to the “classical” SB-QSO evolutionary sequence, where at this stage the molecular gas content should not yet have been affected by feedback according to model predictions (see Fig. 1). Before discussing viable interpretations of these results, we take into account the possible effects of biases in the determination of the SFE values in QSO systems.

4.1.2. Possible biases in SFE derivation for dusty AGN

The empirical SFE = LIR/LCO measurements are a useful instrument for testing the model predictions as they do not depend on the assumption that it is necessary to convert CO luminosity in the molecular gas mass. The SFE is also the best “reference frame” for comparing obscured and unobscured QSOs: comparing gas masses does not provide a meaningful test as they correlate with the mass of the systems which may differ systematically between the two samples; it is also not practical to compare gas fractions as optically luminous QSOs suffer from poorly constrained masses due to AGN-related emission overpowering the light from stars (which also prevents the direct comparison of cold gas properties in mass-matched samples).

On the other hand, the LIR and LCO measurements (involved in estimating the SFE) are both subject to systematic uncertainties. In particular, CO luminosities may be affected by excitation factor uncertainties (Sect. 4.1). Moreover, the infrared luminosities of our sample have been derived via a variety of approaches and in some cases are poorly constrained due to non-detection in the far-IR bands (see Notes in Table C.2). In this section, we test the impact of these biases on our results, considering the total samples of SFGs, dusty and unobscured AGN (i.e. without considering distinct redshift bins; see Sect. 4.1.1).

In order to probe whether incorrect IR luminosity measurements affect our dusty AGN SFE cumulative distribution, we consider only the subsample of dusty AGN whose LIR has been measured through an SED-decomposition into AGN and galaxy emission. Moreover, we also discarded all AGN for which IR emission is not well constrained because of the lack of data above 24 μm (see Notes in Table C.2). The obtained cumulative distribution is shown in Fig. 4, right. A KS test demonstrated that even in this case the populations of SFGs and dusty AGN are different (D = 0.57 and null hypothesis rejected at level > 99.9%).

It has also been suggested that quasar-heated dust on kpc scales can contribute significantly to the far-IR luminosity (e.g. Symeonidis 2017). This contribution, which would not be taken into account with standard SED decomposition involving a mid-IR torus component, can be as high as about 60% for the most luminous sources (Lbol ∼ 1048 erg s−1; Duras et al. 2017; see also Symeonidis 2017). In order to test whether the observed difference in the SFE distribution of dusty AGN and star-forming galaxies is due to this bias in LIR measurements, we applied a correction to the SFE values by reducing the IR luminosity by a factor of 60% for all the dusty AGN. We note that this is a very conservative approach: with average absorption-corrected 2–10 keV luminosities of 3 × 1044erg s−1 (see Table C.2; this corresponds to an average bolometric luminosity of 1046erg s−1, assuming a bolometric correction of 30 following Lusso et al. 2012), our dusty AGN are generally much fainter than the WISSH QSOs presented by Duras et al. (2017) and the sources discussed in Symeonidis (2017).

The distribution derived with rescaled LIR is shown in Fig. 4, right (red dashed line). Even by considering this extreme correction, the SFG and buried AGN cumulative distributions are significantly different (null hypothesis rejected at level ∼93%; an overlap of the two distributions only becomes possible if we assume that > 75% of the IR luminosities reported in Table C.2 are associated with quasar emission). We can therefore reasonably exclude that the difference between dusty AGN and SFGs cumulative distributions is due to an overestimation of the AGN host galaxy infrared luminosities.

The excitation factors generally used to derive CO luminosities from Jup > 1 states of dusty AGN are 0.8 − 1.0 (see Notes in Table C.2). We note that smaller values have also been proposed in the literature for the excitation corrections, in particular for high Jup states (Jup > 2; e.g. Bothwell et al. 2013; Yang et al. 2017; Sharon et al. 2016; up to a factor of 2 smaller); therefore, possible biases in the CO luminosity determination could be responsible for a slight shift to the left in Fig. 4. We therefore consider the SFE distribution of only those sources whose Mgas is derived from CO(1-0) and CO(2-1) line fluxes, which are less affected by excitation factor uncertainties, and perform a KS test between the SFGs sample and the subsample of dusty AGN associated with low J transitions. We prove that even in this case the two populations are different (D = 0.72; and null hypothesis rejected at level ≫99.9%).

To summarise, we find no evidence that the observed differences between dusty AGN and star-forming galaxies SFEs are caused by observational biases.

4.2. Low gas fractions in dusty AGN

Figure 5 (left) shows the gas fractions of the 26 dusty AGN for which stellar mass estimates can be derived from multiwavelength data, as a function of redshift. The gas fractions of optically luminous QSOs, instead, are mostly unknown from the observational point of view, and are not reported in the figure8.

thumbnail Fig. 5.

Left panel: redshift evolution of the gas fraction fgas for our sample of dusty AGN. The targets are colour-coded according to the distance from the main sequence of SFGs (see colour bar in the bottom right corner). For BzK8608 and CDF153 we also show (with open symbols) the upper limits we would obtain by assuming αCO = 0.8 instead of 3.6 (see Sect. 2). The predicted evolutionary trend of MS galaxies of Mstar = 1011 M from the 2-SFM model is shown with a solid curve (shaded area: 1σ scatter around average evolutionary trend line). We also show the predicted trends for MS galaxies of Mstar = 2 × 1010 M and Mstar = 5 × 1011, as labelled in the figure. The error bar in the bottom left corner of the plot shows the representative 1σ error for the data points (we assume a 0.2 dex uncertainty for Mstar estimates; see Table C.2). Right panel: redshift evolution of the normalised gas fraction μgas for our sample of dusty AGN. The predicted evolutionary trend of “gas depleted” MS galaxies of Mstar = 1011 M is shown with the black curve (the shaded area indicates 1σ scatter around the average trend). This trend has been obtained in the framework of the 2-SFM model (Sargent et al. 2014) by considering the depletion of gas due to AGN-driven outflows (Mout = 5 × 1010 M). We also display the predicted trends for Mout = 1010 and 1011 M. The blue symbols refer instead to the predicted μgas of high-z unobscured QSOs from Valiante et al. (2014). The star symbols are associated with X-ray detected AGN and are identical to those used in Fig. 3. Essentially all obscured AGN hosts show a gas fraction deficit compared to normal galaxies, with CT QSOs (large star symbols) displaying some of the strongest deficits.

Open with DEXTER

In Fig. 5 we colour-coded the sources according to their distance from the MS (from ∼1 dex below to +1.2 dex above the sequence; see accompanying colour bar). Most galaxies in our SFGs sample and dusty AGN lie on the MS, with the latter population showing slightly higher MS offsets sSFR/sSFRMS (see Fig. C.1).

In the figure the gas fractions are compared with the expected average evolutionary trend of MS galaxies of Mstar = 1011M from the 2-SFM framework of Sargent et al. (2014). The 2-SFM trend reproduces well the behaviour of massive (Mstar ≈ 1010 − 1011 M) MS galaxies (e.g. Schinnerer et al. 2016) and, in this mass range, is also consistent with other gas fraction evolutionary trends in the literature (e.g. Dessauges-Zavadsky et al. 2017; Genzel et al. 2015; Tacconi et al. 2018).

It is still debated whether SBs are expected to have gas fractions that are similar to (e.g. Sargent et al. 2014) or higher than those of MS galaxies (e.g. Genzel et al. 2015; Tacconi et al. 2018). Nevertheless, the 2-SFM trend reproduces well the distribution of the SFGs presented in this work. These targets are not reported in the figure in order not to overpopulate the plot.

Figure 5 (right) also shows that our dusty AGN are preferentially located below the gas fraction relation expected for MS galaxies. Moreover, many of the AGN data points that currently lie on top or above the 2-SFM trend are upper limits. This finding is in line with the idea that dusty AGN are associated with the feedback phase, where the molecular gas content has been depleted by powerful outflows.

Recent cosmological simulations (e.g. Costa et al. 2018) and observations (Brusa et al. 2018; Fiore et al. 2017) both suggest that AGN-driven outflows can have mass outflow rates of the order of 102 − 104 M yr−1 and that the typical timescales associated with AGN winds are of the order of ∼10 Myr (e.g. Barai et al. 2018, Costa et al. 2018, Fiore et al. 2017, Hopkins et al. 2005, Hopkins et al. 2016, Perna et al. 2015a and Valiante et al. 2014 for both observational and theoretical arguments). There is also evidence that ejected material is multiphase and predominantly composed of cool gas (e.g. Costa et al. 2018, Zubovas & King 2014; see also Brusa et al. 2018 and Fiore et al. 2017). Cosmological simulations show that the outflow masses can reach ≈(5 − 80)×109 M during the feedback phase (Costa et al. 2018, and references therein). Interestingly, this estimate is also consistent with observational findings: an outflow mass of 1010 M can be derived assuming a typical outflow mass rate of out ≈ 103 M yr−1 and an outflow timescale of 10 Myr.

We consider all these arguments to derive some predictions for the gas fraction of dusty AGN associated with Mstar = 1011 M host galaxies. If we assume that dusty AGN are caught 10 Myr after the initial outflow, their gas reservoirs would be depleted by ≈5 × 1010 M; moreover, during this time a given amount of gas is converted into stars, so that , with from Sargent et al. (2014). Assuming a constant SFR (from the Mstar−SFR relationship), the term SFR × Δt is negligible with respect to Mout. In Fig. 5 (right) we show the predicted trend for “gas depleted galaxies” with the associated 1σ confidence interval. Such a large interval is mostly due to the intrinsic dispersion in the SFR-Mstar MS relation (0.2 dex; Speagle et al. 2014). We also note that at lower redshifts the outflow mass could be overestimated as the cosmological simulations are generally performed at higher redshift (z ∼ 6), where host galaxies are associated with more rich gas reservoirs (and the AGN winds can affect larger amounts of gas; see e.g. Fig. 5, left).

In the figure, we also show the normalised gas fractions predicted by Valiante et al. (2014) for optically luminous QSOs at 5≤ z ≤6.4. The authors presented a semi-analytical model for the formation and evolution of high-z QSOs and their host galaxies according to the SB-QSO scenario. They studied the evolution of the stellar and gas masses in order to explain the observed properties of 5≤ z ≤ 6.4 QSOs. We considered the predicted stellar and gas masses at the end of the evolutionary tracks they presented for five distinct high-z QSOs (Figs. 6 and 7 in Valiante et al.) to derive fgas estimates, and reported in the figure the ratios of these values to the . Finally, depleted gas reservoirs in z ∼ 6 galaxies experiencing AGN feedback are also found in the hydrodynamical simulations by Barai et al. (2018), with fgas ∼ 20 − 30% lower with respect to the prelude of the feedback phase (private communication).

An accurate comparison between different predictions and theoretical feedback prescriptions is beyond the purpose of this paper. We refer the reader to (Harrison et al. 2018, and references therein) for a relevant recent review. Here, we simply note that the low gas fractions (and μgas) we derived for dusty AGN could be explained by recent outflows which depleted their host galaxy cold gas reservoirs.

An alternative interpretation of our results could be that such low μgas are due to an incorrect estimate of the gas mass because of an incorrect assumption on αCO. In fact, a difference of a factor of 4.5 in Mgas easily emerges when we consider αCO = 3.6 instead of 0.8, the value normally used for high-z AGN and SMGs. We note, however, that even assuming αCO = 3.6 for all the sources in our sample (i.e. SFGs and AGN) the population of dusty AGN is still associated with lower μgas values with respect to SFGs. Moreover, for several dusty AGN the αCO = 0.8 value has been chosen on the basis of the observed compactness and/or temperature of the cold gas reservoirs in the AGN host galaxies (e.g. BzK4892, XID2028, XID403, GMASS953).

We also note that for the few dusty AGN associated with αCO = 3.6 in Fig. 5, i.e. BzK8608, CDF153, and the Kakkad et al. (2017) targets, the CO-to-gas factor has been assumed consistently with other MS galaxies (because of their sSFR), ignoring the presence of the AGN and the possibility that smaller amount of cold gas (by a factor of 4.5) could be present in these systems (Kakkad et al. 2017).

In Fig. 5 (right) we also indicate with variably-sized star symbols the column densities associated with individual AGN (as in Fig. 3). We note that all the sources with lowest μgas (fgas ≲ 0.2) are associated with high obscuration (log(NH) ≳ 23 cm−2), with the exception of XID2028 (log(NH)∼21.84 cm−2)9. In particular, all CT QSOs in our sample have strongly depleted molecular gas reservoirs, with fgas ≲ 0.2 and μgas ≲ 0.4.

We also derived the ratios between the (2–10) keV and IR luminosities per individual target with known absorption-corrected X-ray luminosity. In the framework of a simultaneous evolution between SMBH and host galaxy (Kormendy & Ho 2013), this luminosity ratio can be used to distinguish between SF- and AGN-dominated systems, using as a reference threshold the ratio log(LX /LIR) = − 2.2 (see Sect. 4.4 in Ueda et al. 2018). We found that all but a few sources in our sample are associated with dominant AGN activity (i.e. log(LX /LIR) > − 2.2) and that there is no obvious correlation between μgas and LX /LIR. The high fraction of AGN-dominated starbursting systems in the sample can be plausibly due to observational bias: the multiwavelength information required to recover stellar masses, and infrared and X-ray luminosities could be, in fact, only available for the most extreme sources at z > 1.

In summary, we conclude that the gas fraction of dusty AGN and of CT systems in particular strongly differ from MS and SB galaxies without evidence of AGN activity.

4.3. SB-QSO evolutionary sequence vs. orientation-based unification scheme

It has been proposed that gas-rich galaxy major mergers funnel a significant amount of matter toward the nuclear regions, triggering SMBH and starburst activity. Evidence supporting a merger-driven scenario has recently been proposed by Fan et al. (2018), who report CO line emission from a hyper-luminous, dust-obscured QSO at z ≳ 3 suggesting the presence of gas-rich major mergers in the system (see also Polletta et al. 2011). Gas-rich mergers related to high-z SMG/QSOs have also been recently presented by Trakhtenbrot et al. (2017) and Fogasy et al. (2017, see also Banerji et al. 2017; Vignali et al. 2018). In the SB-QSO framework, these systems would be associated with the initial evolutionary stages before feedback phenomena start to deplete the cold gas.

Moreover, massive, powerful multiphase outflows have been discovered in several local (Feruglio et al. 2010; Perna et al. 2017; Sturm 2011; Veilleux et al. 2017) and high-redshift galaxies hosting obscured AGN (Brusa et al. 2018; Cresci et al. 2015a; Fan et al. 2018; Feruglio et al. 2017; Nesvadba et al. 2016; Popping et al. 2017), demonstrating that luminous AGN are capable of expelling large amounts of gas from the host galaxies, thereby potentially explaining the low gas fraction (and high SFE) of optically luminous QSOs. We note, however, that from an observational point of view, there is not a clear separation between the two classes of dusty AGN: the first is associated with the prelude of the feedback phase (and therefore with more massive gas reservoirs and higher obscuration) and the second is associated with the feedback stage (and moderate column densities). This is the reason why CO-follow up of CT AGN, more reasonably associated with the prelude of the blow-out phase, are essential in order to test the SB-QSO scenario.

We find that the molecular gas reservoirs of dusty AGN are significantly different from those of SFGs (i.e. MS and submillimetre galaxies without evidence of AGN): Figs. 3 and 4 show that SFE in the dusty AGN sample are higher than those of SFGs. This difference cannot be removed considering the biases in CO and infrared luminosity determinations. In Fig. 5 we also show that the gas fraction of dusty AGN is generally lower than that expected for MS and SB galaxies. However, the SFEs of dusty AGN and unobscured QSOs are similar (and clearly higher than those of SFGs), counter to what we would naively expect for sources in a transition phase between SBs and optically bright QSOs. This result could suggest that the similarly enhanced SFEs in optically bright and dusty AGN do not imply a temporal sequence, but that similar mechanisms are in act in these systems.

Considering only the CT QSOs, we find that two out of the five observed systems (i.e. BzK4892 and XID403) are associated with high SFE (≥300). This finding is also at odds with the SB-QSO evolutionary paradigm, as we would expect that the star formation efficiency of CT sources should be more similar to that of starburst galaxies (i.e. ≲200; see Fig. 4). Moreover, all the CT AGN in our sample have μgas ≲ 0.4 (Fig. 5, right).

One possible explanation for the high SFE values and low gas fractions in CT sources is that these powerful dusty AGN might be already able to quickly clear the gas from the host disk through feedback in this early stage. This hypothesis is confirmed by the presence of strong ionised outflows in the other CT QSOs in our sample, SW022550 and SW022513 (Polletta et al. 2011), for which deep spectroscopic information have been used to study the gas kinematics in the host galaxies, and GMASS953, for which there are several indications of the presence of AGN-driven large-scale outflows (Talia et al. 2018; see also Perna et al. 2015b for another example of powerful outflow in a CT QSO at high-z). We note, however, that these three CT AGN are also associated with lower SFEs with respect to XID403 and BzK4892 and, therefore, that more extreme conditions might be present in the latter AGN (e.g. the concomitant presence of strong SF and AGN activity).

An alternative scenario is given by the possible presence of positive feedback in the host galaxies: AGN could in fact enhance the SF by outflow-induced pressure (e.g. Cresci et al. 2015a,b; Cresci & Maiolino 2018; Silk 2013; see also Maiolino et al. 2017). In this case, the high SFR observed in the host galaxy may be associated with residual star formation activity in an already gas depleted host galaxy. The available information does not allow us to discriminate between positive and negative feedback scenarios; with deep, spatially resolved CO spectroscopy it would be possible to map the molecular gas reservoirs and study the effects of outflows (e.g. Brusa et al. 2018; Carniani et al. 2017).

Finally, it is worth noting that our results could be discussed in the context of the unified model (Antonucci 1993). According to the orientation-based unification scheme, orientation effects relative to an obscuring medium which hides the view of the inner SMBH are responsible for the differences between obscured (type 2) and unobscured (type 1) QSOs10. Therefore, these systems should be associated, on average, with similar host galaxy properties (i.e. molecular gas content and infrared luminosities). There is evidence suggesting that obscured quasars have higher average SFR and infrared luminosities than the unobscured systems (e.g. Chen et al. 2015; Hiner et al. 2009; Zakamska et al. 2016). Given this observed diversity in IR luminosity and assuming that both QSO classes have similar molecular gas reservoirs, we should observe the highest SFE in dusty systems, in contradiction with our results. This could be interpreted as evidence that the unified model cannot explain the observed SFE of high-z QSOs.

Further investigation is required, however, to probe the absence of biases in actual observations, in particular to discriminate between galaxy-scale absorbers and the presence of a torus in dusty AGN, although gas rich-galaxies could be preferentially selected within type 2 AGN samples (see e.g. the discussion in Zakamska et al. 2016). Under the assumption that cold gas is isotropically distributed in the host, we also note that the high column densities observed in all CT QSOs but SW022513 (for which we do not have information regarding CO extension) can be accounted for by the presence of cold gas on a kp scale rather than, or in addition to, the nuclear (pc-scale) torus (see Gilli et al. 2014 for details). A further caveat is the small number of CT sources with available CO spectroscopy. In the future, a larger sample of this class of sources will be helpful in confirming our results.

5. Conclusions

We presented JVLA and PdBI observations for two highly obscured (BzK8608 and CDF153) and one Compton-thick (BzK4892) QSOs. We targeted the CO(1-0) line in BzK4892 and BzK8608, both at redshift z ∼ 2.5, and the CO(2-1) transition in CDF153, at redshift z = 1.536. These observations allowed us to place upper limits on the strength of the CO emission lines, corresponding to LCO < (1.5 − 2.8)×1010 K km s−1 pc2 for the three targets. We used AGNfitter (Calistro Rivera et al. 2016) to fit their SEDs and decompose AGN and galaxy emission, obtaining estimates of the infrared luminosity and the stellar mass of the host galaxies.

When we place these targets in the LCOLIR plane (Fig. 3) and compare their positions with respect to the relation expected for MS galaxies, we find that BzK4892 is the only target for which we were able to derive a physically stringent upper limit on the CO luminosity. This source displays a high star formation rate surface density, ΣSFR ∼ 130 M yr−1 kpc−2 (for an effective radius re = 0.8 kpc; Barro et al. 2016), a low gas fraction (fgas < 0.06), and a star formation efficiency SFE > 410, pointing to an advanced level of gas depletion in this CT QSO. Instead, for BzK8608 and CDF153 the derived upper limits on the gas fractions (fgas ≲ 0.5) and lower limits on the SFE of ≈20 are compatible with available measurements for normal MS and SMGs. Lower gas fractions would be inferred from our CO observations if these sources can be shown to have ISM conditions generally associated with low CO-to-gas conversion factors (e.g. high ΣSFR and/or high Tdust). We also reiterate here that for BzK8608, all measurements should be regarded as tentative measurements because of the adopted photometric redshift (Sect. 2.2).

In the second part of the paper, we combined our measurements with a sample comprising normal and submillimetre galaxies, plus dusty (see Sect. 4.1.2) and unobscured AGN at z > 1 from the literature. Our results suggest that submillimetre detected, dusty AGN are associated with star formation efficiencies very similar to those of unobscured QSOs, and higher than those of SFGs.

Furthermore, we found that CT QSOs may behave more similarly to optically luminous QSOs and obscured AGN than to SMGs: indication of outflows in the ionised and molecular phases of the ISM (SW022550, SW022513, and GMASS953) and indirect evidence of depleted gas (BzK4892) point to the possibility that, in the context of the SB-QSO paradigm, highly obscured systems are affected by feedback phenomena right from the early phases of the evolutionary sequence. We would like to note that evidence of outflows even in early- or intermediate-stage merger systems has been collected for several nearby ULIRG/AGN galaxies (e.g. Feruglio et al. 2013, 2015; Rupke & Veilleux 2013, 2015; Saito et al. 2018) and in the 3C 298 radio-loud QSO at z = 1.439 (Vayner et al. 2017; see also Sect. 4.3; footnote 9). These results, together with the fact that SFE of dusty and unobscured QSOs are comparable on average, could be consistent with the SB-QSO evolutionary scenario assuming that feedback effects occur early in the evolutionary sequence.

Our findings are not in tension with recent observational and theoretical work showing that the SFR of galaxies hosting AGN is not strongly affected by feedback (e.g. Mullaney et al. 2015; Roos et al. 2015): the high SFE in our dusty AGN is reasonably due to depleted cold gas reservoirs (see Fig. 5) rather than significant SFR variations. We also note that, with respect to the Mullaney et al. (2015) sample mostly comprising MS galaxies hosting AGN, in this paper we consider starbursting, dusty AGN, which are expected to be affected by strong outflows (e.g. Hopkins et al. 2008).

We also want to emphasise that particular chemical and kinematic conditions could be at the origin of the observed LIR/LCO ratios (e.g. Blitz & Rosolowsky 2006; Saintonge et al. 2016; Shi et al. 2011), and the similarly enhanced SFE (and depleted gas reservoirs) in optically bright and dusty AGN could not necessarily imply a temporal sequence for the two samples. For example, powerful outflows, revealed both in highly obscured and optically bright QSOs (e.g. Bischetti et al. 2016; Carniani et al. 2017; Perna et al. 2015b), could peculiarly alter the cold gas conditions and be responsible for the enhanced SFE values. The exact interpretation of our results needs further exploration.

We also emphasise that the small number of highly obscured sources and the uncertainties still affecting the measurements of both CO and IR luminosities do not allow us to draw any strong conclusion about the physical conditions in these distant buried QSOs and the possible evolutionary connection between unobscured and highly obscured AGN. Further observations of high-z SMG/QSOs with the JVLA, ALMA, and/or NOEMA facilities will help shed light on the exact nature of the host galaxies of rapidly growing SMBHs, both in terms of assembling larger samples, and also of obtaining higher spatial resolution detections to determine the direct effects of feedback phenomena on the host galaxy gas reservoirs.

Finally, we also want to emphasise that CO(1-0) measurements in high-z galaxies are of enormous utility to the study of galaxy evolution as they are not affected by excitation correction uncertainties and can be used to compare galaxies with unknown and possibly different physical conditions. The CO(1-0) luminosity can be used to derive molecular gas content in the host galaxy, assuming a luminosity-to-gas-mass conversion factor. This estimate is arguably less affected by the uncertainties than measurements based on dust continuum observations, for which assumptions on the dust-to-gas ratios and/or metallicity and dust temperature are required (e.g. Magdis et al. 2012; Scoville et al. 2016). Existing CO(1-0) observations, however, have so far been limited to a small number of sources (see e.g. Table C.2 for the sample of dusty AGN); the measurements reported in this paper for BzK4892 would therefore be useful for establishing the context for the next studies of highly obscured QSOs in the high-z Universe.


1

In this paper, we use the “empirical” SFE definition, SFE = LIR/LCO, rather than the “physical” SFE in units of yr−1. Here LIR refers to the 8 − 1000 μm integrated stellar infrared luminosity, while LCO indicates the CO(1-0) luminosity. The advantage of using these empirical measurements is that they are not subject to the uncertainties related to the use of conversion factors needed in the physical formalism for the derivation of gas masses (see Sect. 4.5 in Carilli & Walter 2013).

2

At the time of observations, X-ray properties were derived from the 4 Ms Chandra survey (for the BzK sources; Feruglio et al. 2011) and the ultra-deep (≈3.3 Ms) XMM-Newton survey (CDF153; Comastri et al. 2011).

4

The sources MIPS506 and MIPS16144 have been excluded from the sample of obscured AGN on the basis of the presence of strong PAH features in their mid-IR spectra (Yan et al. 2010) and, for MIPS506, the absence of AGN signatures in optical spectra (Bauer et al. 2010). These sources are therefore associated with the SMG sample, as per their LIR > 1012 L.

5

Source C92, from the sample of Kakkad et al. (2017), previously proposed as candidate CT QSO by (Bongiorno 2014, XID 60053), is here associated with a modestly obscured QSO, with log(NH) ∼22.5 cm−2, on the basis of deeper Chandra observation (Marchesi et al. 2016). Source C1148, from Kakkad et al. (2017), is instead a candidate CT source; however, the target is detected with 20 net counts and deeper observations are required to confirm its high obscuration (Marchesi et al. 2016).

6

At the time of observations, the infrared luminosities of BzK8608 and CDF153 were overestimated by a factor of a few (as they were derived on the basis of radio 1.4 GHz observations), and the required depths in CO were settled assuming a SFE = 100.

7

Lower limit values are not included in the distributions of dusty AGN and unobscured QSOs. The inclusion of these sources, representing ∼20% of the samples of dusty AGN and unobscured QSOs, would further increase the SFE-constrast between SFGs and AGN. KS test results are obtained employing the FR/RSS method (Peterson et al. 1998); the values reported in the text correspond to the average results from 100 iterations.

8

The difficulties in deriving unobscured QSO host galaxy stellar masses from SED fitting (see Sect. 4.1.2) make gas fraction estimates for these sources unreliable. An alternative approach is to infer Mstar via a dynamical mass based on molecular line observations, using the relation MstarMdynMgas (e.g. Venemans et al. 2017; Wang et al. 2010). However, these measurements are affected by very large uncertainties.

9

Similar gas fractions are also associated with AMS12, MRC0152-209, and SMMJ030227, for which we do not have X-ray information. HST/ACS observations of the latter target show that this SMG/QSO is a complex system surrounded by a diffuse halo of material in a ring-like structure, with a compact companion at ∼1.3″ (∼11 kpc; Swinbank et al. 2006). Spectroscopic observations of this peculiar target also suggest the presence of complex kinematics between the nuclei (Harrison et al. 2012; Menéndez-Delmestre et al. 2013).

10

AGN with discordant optical (type1/type2) and X-ray (unobscured/obscured) classification have been reported in the literature. In this work we do not take into account this discrepancy since these sources represent a small fraction of the AGN population (see e.g. Merloni et al. 2014; Ordovás-Pascual et al. 2017 for details).

Acknowledgments

We thank the anonymous referee for a useful and detailed report which helped improve the paper. MP thanks the Marco Polo Project of Bologna University and the University of Sussex for the hospitality during his stay in Brighton, where part of this work was carried out. MP also thanks S. Salvadori, S. Gallerani, A. Pallottini, M. C. Orofino, and L. Graziani for fruitful discussions. GC acknowledges the support by INAF/Frontiera through the “Progetti Premiali” funding scheme of the Italian Ministry of Education, University, and Research, and the INAF PRIN-SKA 2017 programme 1.05.01.88.04. Much of this research was carried out at the Munich Institute for Astro- and Particle Physics (MIAPP) of the DFG cluster of excellence “Origin and Structure of the Universe” during the program “In & Out: What rules the galaxy baryon cycle?” (July 2017). MP, MB, and GL acknowledge support from the FP7 Career Integration Grant “eEASy” (“SMBH evolution through cosmic time: from current surveys to eROSITA-Euclid AGN Synergies”, CIG 321913). MTS was supported by a Royal Society Leverhulme Trust Senior Research Fellowship (LT150041). EL is supported by a European Union COFUND/Durham Junior Research Fellowship (under EU grant agreement no. 609412). This work is based on observations carried out under project number W041 [PI: C. Feruglio], with the IRAM PdBI Interferometer [30m telescope], and under project number VLA/11B-060 [PI: E. Daddi], with the JVLA telescope. IRAM is supported by INSU/CNRS (France), MPG (Germany), and IGN (Spain). The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. This work has made use of the Rainbow Cosmological Surveys Database, which is operated by the Universidad Complutense de Madrid (UCM), partnered with the University of California Observatories at Santa Cruz (UCO/Lick, UCSC).

References

Appendix A: Far-IR photometry and SED fit comparison with previous works

The three obscured AGN presented in this work, BzK4892, BzK8608, and CDF153, have been observed as part of the GOODS-Herschel (Elbaz et al. 2011) and PEP (Magnelli et al. 2013) surveys. However, their PACS 70, 100, and 160 μm, and SPIRE 250, 350, and 500 μm photometric data may be affected by source confusion. To derive their host galaxy properties, we therefore used a new Herschel catalogue presented by T. Wang et al. (in prep.), which utilises an optimised algorithm (T-PHOT; Merlin et al. 2015) for photometry in crowded Herschel maps. T-PHOT uses minimum chi-square estimation to generate flux density estimates using galaxy positions extracted from shorter wavelength images.

In Table A.1 we report the T-PHOT fluxes used in this work and, when available, those from the Rainbow Cosmological Surveys database11 for comparison.

Table A.1.

Far-IR photometric data.

BzK4892, BzK8608, and CDF153 multiband SED fits are also presented in the Rainbow database. However, for CDF153 and BzK8608 the SED fits were performed excluding the un-deblended, and hence potentially biased, λ > 24 μm photometry in the Rainbow database. Furthermore, in the Rainbow SED fits only stellar SED templates are used to reproduce the photometry. As a result, the LIR and SFR measurements of these two sources cannot be compared with our measurements (Sect. 2.1). We also note that the Rainbow photometric redshift of BzK8608 is z= 2.63, while in this work we used z = 2.51. The Rainbow BzK4892 SED is instead also sampled in the far-IR, and their LIR is compatible with our SED fit measurement within a factor of 2 (see also Elbaz et al. 2018). The stellar masses derived for the three targets are also compatible with ours within a factor of 2.

Appendix B: BzK8608 photometric redshift

We derived a photometric redshift for BzK8608 by running AGNfitter on a grid of redshifts (Δz = 0.01) spanning the redshift range 2.4< z< 3, corresponding to the redshifts reported for BzK8608 in previous studies (e.g. Feruglio et al. 2011; Hsu et al. 2014). The normalised redshift likelihood distribution is shown in Fig. B.1; the adopted redshift corresponds to the peak of the distribution (redshift errors define the likelihood ratio limits at a 68% confidence interval; D’Agostini 2003). This redshift estimate is also consistent with the photometric redshift reported in Hsu et al. (2014).

Tentative spectroscopic redshifts from the iron Kα line in X-ray spectra have been reported by (Feruglio et al. 2011, zKα = 2.9 ± 0.1) and (Buchner et al. 2014, zKα = 2.83 ± 0.02). However, this feature is not detected in the deeper Chandra observations (see Sect. 2).

The frequency coverage of our JVLA observations maps is in the two redshift intervals [2.5–2.6] and [2.95–3.08]. We searched for line emission through the accessible redshift range, finding no significant features. Given that an unambiguous spectroscopic redshift has not yet been obtained for this target, we used the derived photometric redshift for the estimate of the CO luminosity upper limit of BzK8608 and the host galaxy properties. We note that the IR luminosity and stellar mass derived with AGNfitter do not show a significant dependence on redshift in the range 2.4 < z < 3.

thumbnail Fig. B.1.

Likelihood distribution as a function of redshift, obtained repeating the SED fit computation of BzK8608 for fixed z value (the redshift binning is 0.01). The maximum corresponds to the redshift z = 2.51. Shaded areas represent the frequency coverage of our JVLA observations.

Open with DEXTER

Appendix C: SMG and QSO samples

In Table C.1, C.2, and C.3 we reported the information collected from the literature for SMGs, obscured and unobscured AGN, respectively. Stellar masses, X-ray luminosities, and column densities are reported for the only sources for which these quantities have been reported in literature. All luminosities are re-derived using the values tabulated in the papers mentioned in Sect. 3 and considering Planck Collaboration XIII (2016) Cosmological parameters; the host galaxy quantities, instead, have been corrected for the adopted Chabrier IMF. Moreover, we tabulated measurement uncertainties (at the 1σ confidence level) for the only sources for which these quantities were previously reported in the literature. The stellar mass measurements of unobscured AGN are not reported in Table C.3 (see Sect. 4.2). Gas masses are tabulated for SMGs and dusty AGN targets, and are used to derive the gas fractions shown in Fig. 5.

The SMG/QSO sources presented in Coppin et al. (2008) and Bothwell et al. (2013) have been included in the obscured (dusty) AGN sample table (see Sect. 4.1). The dusty AGN are characterised by red spectra/colours (e.g. Banerji et al. 2017; Brusa et al. 2018; Coppin et al. 2008; Fan et al. 2018; Kakkad et al. 2017; Yan et al. 2010) and/or high column densities (e.g. Bothwell et al. 2013; Gilli et al. 2014; Polletta et al. 2011; Stefan et al. 2015; Vayner et al. 2017).

All the non-lensed luminous AGN are characterised by bright UV emission (with rest-frame absolute magnitude at 1450Å of M1450 < −25; Bañados et al. 2016; Priddey & McMahon 2001), while lensed galaxies are all associated with negligible obscuration (E(BV)≲0.112) and the presence of BLR emission lines in their rest-frame UV spectra (e.g. Anguita et al. 2008; Mazzucchelli et al. 2017; Sharon et al. 2016; Shields et al. 2006; Wang et al. 2013). The sample of unobscured AGN do not include the bright sources for which the IR luminosities are not available (e.g. Carniani et al. 2017).

We would like point out that only a few optically bright QSOs have been studied through their X-ray emission and that their characterisation requires long-exposure observations. Therefore, we cannot confidently exclude the presence of nuclear obscuration in the individual targets reported in Table C.3. We note, however, that there are several indications suggesting that the population of optically luminous QSOs are associated with X-ray spectra characterised by low column densities (NH ≲ 1022 cm−2; Nanni et al. 2017; Shemmer et al. 2006; Vignali et al. 2005). Therefore, we can be reasonably confident about the separation between the dusty and unobscured QSO populations presented in this work.

In Fig. C.1 we report the distribution of the stellar masses of star-forming galaxies and dusty AGN; these sources are associated with massive host galaxies, with an average ⟨Mstar⟩=1011 M. In Fig. C.1 (bottom panel) we also report the distribution of specific star formation ratio sSFR/sSFRMS (according to the relation of Speagle et al. 2014) for the same sources, showing that the majority of SFGs and dusty AGN collected in this work are associated with MS galaxies, with the latter population showing slightly higher MS-offsets sSFR/sSFRMS.

thumbnail Fig. C.1.

Top panel: stellar mass distribution for sample collected in this work. Blue- and red-shaded histograms refer to the samples of SFGs (i.e. MS and SMGs) and dusty AGN. Bottom panel: specific star formation rate sSFR/sSFRMS distributions for the same samples in the top panel. The sSFRMS is derived according to the relation of Speagle et al. (2014). In the figure, we also show the separation between normal MS galaxies and SBs at log(sSFR/sSFRMS)=0.6.

Open with DEXTER

Table C.1.

SMG sample.

Table C.2.

obscured QSO sample.

Table C.3.

Optically bright QSO sample.

All Tables

Table 1.

Targets properties.

Table 2.

JVLA and PdBI observations.

Table A.1.

Far-IR photometric data.

Table C.1.

SMG sample.

Table C.2.

obscured QSO sample.

Table C.3.

Optically bright QSO sample.

All Figures

thumbnail Fig. 1.

Schematic diagram to illustrate the standard SB-QSO evolutionary sequence. In the first two panels, the blue ellipse and yellow star represent the galaxy and the central SMBH, respectively; the light and dark grey areas refer to inflowing gas and dust. Dark blue clouds with yellow outlines indicate outflows. In the last panel, the yellow star and ellipse represent the SMBH and the QSO emission. In the first phase (left), the central regions are fuelled by huge amounts of gas and dust, and the SMBH is highly obscured. These systems, originating from gas-rich galaxy mergers, are commonly identified as SMGs, and are associated with star formation efficiencies of ∼100 ÷ 200. The SFE of CT/highly obscured QSOs is mostly unknown. These systems, if tracing the same evolutionary stage, should have SFEs similar to those of SMGs. In the last phase (right), the system is revealed as an unobscured QSO (bright enough to easily outshine its host galaxy). In this phase its gas reservoir is expected to be strongly depleted. The SFE measured in optically luminous QSOs can span a wide range, from a few 100 to ∼2000. In the transition phase (middle), galaxy-wide outflows start to remove the dust and gas reservoirs (which only partially obscure the nuclear regions). We observe these systems as obscured QSOs (generally associated with signatures of outflows). For these systems, we naively expect to observe intermediate SFE between SMGs and unobscured QSOs.

Open with DEXTER
In the text
thumbnail Fig. 2.

SED fitting of BzK4892 (top left panel), BzK8608 (top right panel), and CDF153 (bottom left panel). The black dots represent the observed data points from the Guo et al. (2013) and Wang et al. (in prep.) catalogues; 5σ upper limits are shown with black triangles. The data points at log(ν) ∼12 are from ALMA continuum observations at 250 GHz and 340 GHz (see Sect. 2). The best-fit SED obtained with AGNfitter (Calistro Rivera et al. 2016) and the different components used to model the SED are shown as labelled in the figure. Ten randomly picked realisations from the posterior distribution function are overplotted in the figures in order to visualise the dynamic range of the parameter values.

Open with DEXTER
In the text
thumbnail Fig. 3.

Left panel: observed correlation between measures of CO(1-0) luminosity, LCO, and (8 − 1000 μm) IR luminosity, LIR, for the samples of normal galaxies, SMGs, and QSOs at z > 1. For the three obscured QSOs described in this paper, we report the 3σ upper limits, as labelled. Open symbols refer to main sequence galaxies and SMG, filled symbols refer to QSOs. Where column density measurements are available, scaled star symbols indicate mild to high NH values, as labelled in the right panel inset. Olive green symbols are the galaxies from Sargent et al. (2014); green symbols refer to SMGs from Bothwell et al. (2013), Coppin et al. (2008), and Silverman et al. (2015); red symbols refer to obscured targets. Optically bright QSOs are represented with blue dots. The solid curve represents the best-fit relation derived by (Sargent et al. 2014, 1σ scatter indicated with the grey area), for the MS galaxy sample they collected in the redshift range 0< z ≲ 3. Right panel: SFE vs. redshift for the same sample. In this panel, we labelled the names of the four sources from the literature associated with CT QSOs. The solid curve shows the 2-SFM predicted evolutionary trend of MS galaxies of Mstar = 1011 M (shaded area: illustration of the scatter around the mean evolutionary trend for galaxies with SFR values 1σ above/below the MS). The long-dashed curve is located at 10-fold higher SFE than the mean MS evolutionary trend. Representative 1σ errors on individual data points are shown in the bottom left corner of the plots. For the two CT QSOs SW022513 and SW022550, Polletta et al. (2011) indicated a wide range for their IR luminosities (see Notes in Table C.2); these sources are therefore plotted in the left panel at the centroid of the IR range, while horizontal dotted lines show the IR luminosity interval. In the right panel we similarly reported the two sources at the centroid positions and indicated the SFE ranges with vertical dotted lines.

Open with DEXTER
In the text
thumbnail Fig. 4.

First three panels: cumulative distributions of SFE for the samples of star-forming galaxies (green line), unobscured (blue), and dusty (red) QSOs in the three redshift intervals specified along the upper edge of the figures. All targets collected from the literature (for which SFE measurements can be derived) are used to construct the cumulative distributions; in the first panel, we do not report the cumulative distribution of unobscured AGN (our sample only includes two of these objects at 1≤ z ≤ 2). The figures show the clear separation in SFE between AGN and SFGs. Right panel: cumulative distribution of SFE for the same three samples, considering the entire redshift range covered by our targets. Short-dashed red line: cumulative SFE distribution of dusty AGN obtained by reducing by 60% all their IR luminosities to test for the impact of cool, AGN-heated dust (e.g. Duras et al. 2017; see Sect. 4.1.2 for details). Long-dashed red line: distribution of dusty AGN for which a high-quality LIR has been measured through an SED-decomposition into AGN and galaxy emission.

Open with DEXTER
In the text
thumbnail Fig. 5.

Left panel: redshift evolution of the gas fraction fgas for our sample of dusty AGN. The targets are colour-coded according to the distance from the main sequence of SFGs (see colour bar in the bottom right corner). For BzK8608 and CDF153 we also show (with open symbols) the upper limits we would obtain by assuming αCO = 0.8 instead of 3.6 (see Sect. 2). The predicted evolutionary trend of MS galaxies of Mstar = 1011 M from the 2-SFM model is shown with a solid curve (shaded area: 1σ scatter around average evolutionary trend line). We also show the predicted trends for MS galaxies of Mstar = 2 × 1010 M and Mstar = 5 × 1011, as labelled in the figure. The error bar in the bottom left corner of the plot shows the representative 1σ error for the data points (we assume a 0.2 dex uncertainty for Mstar estimates; see Table C.2). Right panel: redshift evolution of the normalised gas fraction μgas for our sample of dusty AGN. The predicted evolutionary trend of “gas depleted” MS galaxies of Mstar = 1011 M is shown with the black curve (the shaded area indicates 1σ scatter around the average trend). This trend has been obtained in the framework of the 2-SFM model (Sargent et al. 2014) by considering the depletion of gas due to AGN-driven outflows (Mout = 5 × 1010 M). We also display the predicted trends for Mout = 1010 and 1011 M. The blue symbols refer instead to the predicted μgas of high-z unobscured QSOs from Valiante et al. (2014). The star symbols are associated with X-ray detected AGN and are identical to those used in Fig. 3. Essentially all obscured AGN hosts show a gas fraction deficit compared to normal galaxies, with CT QSOs (large star symbols) displaying some of the strongest deficits.

Open with DEXTER
In the text
thumbnail Fig. B.1.

Likelihood distribution as a function of redshift, obtained repeating the SED fit computation of BzK8608 for fixed z value (the redshift binning is 0.01). The maximum corresponds to the redshift z = 2.51. Shaded areas represent the frequency coverage of our JVLA observations.

Open with DEXTER
In the text
thumbnail Fig. C.1.

Top panel: stellar mass distribution for sample collected in this work. Blue- and red-shaded histograms refer to the samples of SFGs (i.e. MS and SMGs) and dusty AGN. Bottom panel: specific star formation rate sSFR/sSFRMS distributions for the same samples in the top panel. The sSFRMS is derived according to the relation of Speagle et al. (2014). In the figure, we also show the separation between normal MS galaxies and SBs at log(sSFR/sSFRMS)=0.6.

Open with DEXTER
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.