Although ALMA is not a large survey facility, it is of primordial importance to gather the properties (gas content and excitation, star formation rate, etc.) of a large number of objects, to gain a statistical significance, and to be able to split the samples in several categories, to explore the influence of parameters. Essential in the value of surveys are the selection criteria, for them to be representative even if flux-limited. Surveys have been done with carefully chosen criteria from multi-wavelength studies: these are pointed surveys when sources are already well-known (position, z, stellar mass, SFR). Another type of surveys is the deep field method, unbiased, but sometimes less successful, according to the choice of sensitivity required (shallow, deep), and the choice of tuning (continuum, lines, etc.). Blind surveys have however the immense advantage to be completely unbiased by other wavelengths, with the hope to detect brand new objects, obscured in the optical and UV. Both are now described in turn.
Optical/IR surveys have shown that the cosmic star formation density had a peak at about z = 2, and then has dropped by a factor ∼ 20 (e.g. Madau and Dickinson, 2014). It was also discovered that, although luminous and ultra-luminous infrared galaxies (ULIRGs) tend to dominate more and more the star formation as z approaches 1 (Le Floc’h et al, 2005), the starburst mode is not the dominant star forming mode at z ∼ 2, but contributes only by 10%. The confusion arose at the start because local ULIRGs are all starbursts, due to galaxy interactions and mergers (e.g. Sanders and Mirabel, 1996). Starbursts can be defined as transient states, with an elevated star formation rate (SFR) which cannot be sustained during the "normal" time-scale to consume the gas content of a galaxy, which is 2 Gyr (Bigiel et al, 2008). This time-scale is also called the depletion time tdep. When the redshift increases, the average SFR increases, and LIRGs or ULIRGs do not require anymore the starburst mode for their interpretation. It is then possible to define a Main Sequence (MS) of “normal” star formation (Whitaker et al, 2012, Whitaker et al, 2014, Speagle et al, 2014). The specific SFR, divided by the stellar mass, i.e. sSFR = SFR/M* increases strongly with redshift, in (1 + z)3 up to z ∼ 2, and decreases only slightly with mass, as M*−0.1,−0.4 (Lilly et al, 2013). About 90% of the star formation in the Universe occurs on the main sequence, essentially in exponential galaxy disks (e.g. Wuyts et al, 2011).
The evolution of galaxies along the main sequence (MS), and the evolution of the main sequence itself with redshift, have been the object of many models, implying both gas accretion to re-fuel galaxies after the depletion times of the order of a few Gyr, and the moderation of star formation due to the feedback. Some propose a quasi-stationnary state (Bouché et al, 2010, Lilly et al, 2013), and others a violent evolution across the MS, passing through a starburst phase or a quiescent one, via violent instabilities and compaction (Dekel et al, 2009, Tacchella et al, 2016).
It is crucial to investigate through the abundance of the molecular component for galaxies on the MS, as a function of redshift, what are the main processes regulating the star formation, and galaxy evolution. A large number of surveys have been carried out, for instance the PHIBSS survey with NOEMA, or the COSMOS with ALMA from dust emission, which demonstrate a large increase of the molecular gas fraction with redshift (Tacconi et al, 2010, Tacconi et al, 2013, Scoville et al, 2014, Scoville et al, 2016), but also a slight decrease of the depletion time, i.e. an increase of star formation efficiency.
In these surveys, two different tracers for the interstellar gas have been used: the most direct one, the CO lines, depending on the CO-to-H2 conversion factor, and the dust emission in the Rayleigh-Jeans domain, which traces both atomic and molecular gas, but depends also on metallicity, and on the assumed dust temperature, albeit in a linear way (Scoville et al, 2014, Schinnerer et al, 2016). The proportionality factor between dust emission and gas mass is established from nearby galaxies and the Milky Way (through the Planck satellite data), but several parameters may vary, as the slope of the dust opacity with frequency, the metallicity and dust abundance, or the nature of dust. In the literature, it was found that the gas masses derived at high redshift from the dust emission are somewhat larger than that from the CO lines, at least by a factor 2. Part of the explanation could be that dust emission traces both atomic and molecular gas. Although it is very difficult to have direct estimation of the HI gas at high z, the best estimation comes from the Lyα absorbers along the line of sight towards remote quasars (e.g. Prochaska et al, 2005), and it appears that the cosmic density of HI gas is roughly constant. Given that the molecular gas strongly increases with z, it is assumed that it will dominate as soon as z > 0.5 (e.g. Lagos et al, 2012).
From a compilation of all literature data on molecular gas at high z, around the main sequence and slightly above, scaling relations have now been derived for about 1400 objects (Genzel et al, 2015, Tacconi et al, 2018). The main results are a quantification of the increase of gas fraction with z, and decrease of depletion time, as shown in Fig. 3. These scaling relations take into account all the various parameters (distance from the MS, defined as δMS = sSFR / sSFR(MS, z, M*), redshift, stellar mass), exploiting the fact that the dependence of gas fraction and depletion time (or SFE, the Star Formation Efficiency = 1 / tdep) on these parameters is uncorrelated to some extent, i.e. the variables are separable.
 |
Figure 3. Scaling relations of µgas = Mmolgas / M* with redshift (left), and depletion time tdep = Mmolgas / SFR (right), for the binned data sets (large symbols) and the individual data points (colored distributions), from Tacconi et al (2018). All available data from NOEMA and ALMA have been taken into account, with zero point corrections. Image reproduced with permission from Tacconi et al (2018), copyright by AAS. |
The variation of the depletion time with δMS is clear, at a given z and M*, tdep is decreasing for galaxies above the MS in the starburst phase, and increasing below in the quenching phase. The main results found by all surveys and analysis is that the gas content in galaxies increases significantly with z, as ∼ (1 + z)2, but still lower than the SFR on the MS, which is increasing as ∼ (1 + z)3 up to z = 3-4 (Scoville et al, 2017, Tacconi et al, 2018). In addition, the star formation efficiency (SFE) is increasing with z, i.e. the depletion time varies as (1 + z)−0.6 to (1 + z)−1. There is no dependency of SFE with stellar mass, on the main sequence. The gas fraction, or the gas-to-stellar mass ratio decreases with stellar mass. This explains also the variation of the slope of the MS with stellar mass: if the SFR is almost linear with M* for small masses, it then saturates and the slope is lower than 1. Since high mass galaxies have a more massive bulge, and bulges do not participate to star formation, it is tempting to subtract the bulge mass, to check the SFR variation of disk only. Abramson et al (2014) performed the bulge-disk decomposition for large samples of low-z galaxies in the Sloan survey, and indeed, the MS slope is almost vanishing (and vanishes completey in some samples). The sSFR of disks only is quasi independent of their stellar mass. The remaining dependency could be related to the central bulge concentration (Pan et al, 2016).
All these results are supporting models where galaxy evolution and star formation are mainly driven by external gas accretion. Several interpretations have been elaborated (Berta et al, 2013, Scoville et al, 2017). It is possible to trace the evolution of galaxies, their star formation rate and gas content, assuming continuity, neglecting in a first step the contribution of starbursts (5 - 10%), and the quenched galaxies, which must be of very high mass, and in dense environments (Peng et al, 2010). To maintain the evolution of the MS, it is necessary that galaxies are continuously accreting gas, to refuel their SFR, given their low depletion time-scales, lower than 1Gyr. This refueling can be mostly due to cold gas accretion (Dekel et al, 2013), but also may include minor mergers. The major mergers are considered to be exceptional events, making galaxies to exit the MS from above for a transient period. Using the empirically determined relations between gas content and SFE with redshift, stellar mass and offset from MS, and assuming dM* / dt = 0.7 SFR (30% of the gas is returned to the interstellar medium through stellar mass loss), it is possible to trace the evolution of individual galaxies on the MS diagram (cf Figure 4). The set of equations can be closed, and the net accretion rates can be derived as a function of z and the main parameters. For the average considered mass, the accretion rate increases as ∼ (1 + z)3.5, which may explain the high SFR in early galaxies, and is justified by the high gas density in the early universe.
Adopting these simplifying hypotheses, and summing over the stellar mass function (e.g. Ilbert et al, 2013), the equations can yield the evolution of the cosmic density of the total gas in galaxies (H2 + HI, traced by dust emission, result given in Figure 4).
 |
Figure 4. Left: Evolution of galaxies on the main sequence (MS), assuming they evolve continuously on the MS. The full lines show the MS relations at 5 different redshifts (colours) from the observed consensus compiled by Speagle et al (2014). The dash lines show the time evolution of 1, 5, and 10 × 1010 M⊙ galaxies in between two redshifts. They evolve downward and rightward with time, at a rate 0.7 SFR, taking into account 30% of stellar mass loss. Right: The gas and stellar mass cosmic densities versus redshift for galaxies in the range M* = 1010 to 1012 M⊙. These computations have used the observed scaling relations between gas mass and the 3 parameters (z, δMS, M*), and the stellar mass functions from Ilbert et al (2013). Images reproduced with permission from Scoville et al (2017), copyright by AAS. |
Lensed galaxies allow to explore a lower mass regime (M* < 2.5 × 1010 M⊙), with lower SFR (< 40 M⊙ / yr) (Dessauges-Zavadsky et al, 2015). It is now possible to see the star formation efficiency decrease with stellar mass. This low mass regime reveals the same increase of gas fraction and SFE with redshift. With both CO lines and dust continuum emission, it is possible to see large variations of the dust-to-gas ratio among the various types of star forming galaxies, even at a given metallicity.
Observations of some gas-rich galaxies below the MS suggests that quenching does not require the total removal or depletion of molecular gas, as many quenching models propose (Suess et al, 2017). Spilker et al (2018) detected with ALMA the CO line in 4 out of 8 z ∼ 0.7 passive galaxies 3-10 times below the MS. Their gas fraction is below 10%, small enough that the depletion time is rather short. The gas rotation axis is aligned on the stellar one, implying no recent gas accretion. Even though the samples are still not enough to draw firm conclusions, it appears that the quenching towards forming massive red and dead galaxies is rather slow, and due to the cessation of gas accretion.
One of the very successful surveys of ALMA in its first cycle (16 antenna) was to search for CO lines in the high-z (z >1) sample of dusty continuum sources, assembled over 1300 square degrees by the South Pole Telescope (SPT). The survey benefitted from the negative K-correction, and therefore all redshifts were expected with minimum bias. Most of the highest flux sources are lensed. Out of 26 sources, ALMA detected 23 in one CO line (among them 12 with multiple lines, so that the redshift is clearly determined). In the continuum at 870 µm, with spatial resolutions 0.5 - 1.5”, only one minute integration per source was sufficient to reveal the arc and ring morphology of the lensed background objects, as displayed in Figure 5. Star formation rates larger than 500 M⊙ / yr imply that the sources are ULIRGs (Vieira et al, 2013). The spectroscopic survey in Band 3 more than doubled the known redshifts at this epoch, and had a median redshift of z = 3.5. The fraction of dusty and luminous starbursts at high z appears higher than previously thought (see Figure 6).
 |
Figure 5. ALMA 870 µm images of 10 SPT (South Pole Telescope) sources (red contours), superposed on near-infrared images (NIR, grey-scale) from HST, VLT or SOAR telescopes. The NIR indicates the starlight from ths foreground lensing galaxies. The high-z galaxies are only seen by ALMA. Their spectroscopic redshifts were determined by ALMA CO line observations, and are shown in red in each panel, of size 8" × 8". Image reproduced with permission from Vieira et al (2013), copyright by Springer. |
 |
Figure 6. The cumulative redshift distribution of luminous, dusty starburst galaxies: the SPT galaxies, with ALMA determined redshifts, are shown in black. The blue sample objects have redshifts determined from rest-frame ultraviolet spectroscopy. The orange sample galaxies in the COSMOS survey have only photometric redshifts from optical/IR. ALMA detected a large fraction of high-z dusty starburst galaxies, and previous surveys were biased to lower redshift than the underlying population. Image reproduced with permission from Vieira et al (2013), copyright by Springer. |
In the UV/optical/IR domains, considerable knowledge on galaxy evolution has come from the study of blank fields, integrating deeply in selected regions of the sky minimizing foregrounds, with the HST (HDF, UDF, XDF, Illingworth et al 2013), and also with a multitude of instruments at all wavelengths, from the X-ray (Chandra/XMM), to the far-infrared (Spitzer, Herschel) and radio (VLA). From the 11 HST filters, it has been possible to obtain nearly 10 000 photometric redshifts (e.g. Rafelski et al, 2015). Follow-up from the ground has obtained also spectroscopic redshifts, namely with the VLT (Le Fèvre et al, 2004, Bacon et al, 2017), although the latter spectro-z still amount to less than 2% of the total. These surveys have allowed precious knowledge on galaxy properties (sizes, stellar masses, star formation rates), and their evolution with redshift (e.g. Madau and Dickinson, 2014). However, to understand galaxy evolution, the fuel of star formation, the molecular gas, has to be observed. Also optical surveys are biased against the most obscured and dusty star forming galaxies, and sub-mm surveys are needed. Already ponctual observations have shown that indeed dusty starbursts exist up to z = 6 (Riechers et al, 2013), and surveys with Herschel (Elbaz et al, 2011), or SCUBA-2 (Coppin et al, 2015) have used priors to tackle blending, and stacking to explore just below the sensitivity limit of their instruments.
With ALMA gaining a large factor in sensitivity and spatial resolution, deep surveys are now eagerly expected. The first survey of the Subaru-XMM (SXDF) deep field (1.5 arcmin2 with ALMA) reported by Tadaki et al (2015) has observed in 1.1mm continuum, with a sensitivity of σ = 55 µJy/beam. They targetted 12 Hα-selected star-forming galaxies (SFG) at z = 2-2.5, but detected only 3 of them. The frequency corresponds to 300-400 µm in the rest-frame, so that dust emission should be easy to detect. It also corresponds to 100 µm, the peak of dust emission, for z = 10, and objects with the same mass should be even more easy to detect with the K-correction, so that their absence indicates a drop in the luminosity function with z. One of the object detected is very compact (Re = 0.7 kpc), with a high gas fraction of 44%. In the same ALMA field Hatsukade et al (2016) conclude from the possibly detected 23 sources above 0.2 mJy that the source count is typical, and comparable to all previous ALMA serendipitous detections.
Dunlop et al (2017) reports about the 4.5 arcmin2 ALMA survey of the HUDF at 1.3mm at σ = 30 µJy sensitivity. The extraction of reliable sources in continuum is difficult. About 50 sources are first found above 3.5σ, but around 30 are also found in negative, i.e. below -3.5σ. Therefore most of the 50 sources must be spurious. Comparing with other data, 16 detections are then secured, through counterparts with HST, infrared and/or radio-cm, 13 of them having a spectroscopic redshift in the optical. The average redshift is z = 2.15 and only one source has z > 3. The lack of high-redshift detections confirms the rapid drop-off of high-mass galaxies in the field, above z = 3. Figure 7 shows clearly that the ALMA detections are among the most UV-obscured objects in the HUDF.
 |
Figure 7. Left: the UV absolute magnitude of all galaxies in the HUDF, as a function of redshift, with the ALMA 1.3mm detections in red. Right: when stellar masses are considered, now the ALMA detections are at the top, meaning that they are indeed the most obscured in UV. The bright blue box emphasizes the ALMA detection of 80% of the most massive galaxies (M* > 2 × 1010 M⊙) at z > 2. Below z = 1, the detection rate of massive galaxies drops, which indicates possible quenching. The grey-blue box gathers the galaxies which have been stacked and lead to an ALMA global detection. The absence of very massive galaxies above z = 3 is clearly visible. Image reproduced with permission from Dunlop et al (2017). |
Aravena et al (2016) have carried out a deeper 1.2mm ALMA survey of the HUDF in a restricted region of 1 arcmin2, with σ = 13 µJy sensitivity. They detect 9 sources at 3.5σ with average z = 1.6, and only one source above z = 2, which is significantly lower than the shallower survey of Dunlop et al (2017). The detections correspond to 55% of the extragalactic background light (EBL) at 1.2mm measured by the Planck satellite; when stacking all the sources optically known in this region, it is possible to recover 80% of this EBL.
In addition to this continuum survey of 1 arcmin2 of the HUDF, the same team carried out an ALMA spectroscopic survey (ASPECS, CO and [CII] lines) in two frequency bands at 3mm and 1mm, covering the frequency ranges 84-115 GHz, and 212-272 GHz (Walter et al, 2016). A blind search for lines have found 10 candidates in the 3mm band and 11 at 1mm. The identification of the sources is then done searching for optical/NIR counterparts, with a known redshift. This occurs in 9 out of the 21 candidates. In one or two cases, other CO lines at higher J are also detected in the same survey, and confirm the identification. Most of the times, the lack of other lines suggest that the redshift of the object is large and/or the upper level J of the CO line is large. In addition, stacking has been done for all sources with known redshifts for the first 4 CO lines, but with no detection (Decarli et al, 2016b). Molecular masses were derived for each of the identified sources, and found compatible with previous results for main sequence galaxies, with a large scatter (Decarli et al, 2016a). All results and constraints on the derived cosmic H2 density are gathered in Figure 8.
 |
Figure 8. Comoving mass density of molecular gas in galaxies ρ(H2) as a function of redshift. The ALMA spectroscopic survey (ASPECS) constraints are plotted in pink boxes, with vertical sizes corresponding to the uncertainties. The blue boxes represent the IRAM interferometer constraints (Walter et al, 2014). The predictions of semi-analytical models are superposed as a yellow line (Obreschkow et al, 2009), a blue line (Lagos et al, 2012) and a green line (Popping et al, 2014). The compilation of literature data on MS galaxies is the grey area (Sargent et al, 2014). The circle symbol at z = 0 is from Keres et al (2003), and the losange from Boselli et al (2014). Keating et al (2016) have computed an upper limit (orange triangle) from CO intensity mapping at z ∼ 3. The global behaviour of the H2 cosmic density is very similar to the star formation density, with a peak around z = 2. Image reproduced with permission from Decarli et al (2016b), copyright by AAS. |
From the ASPECS survey, it is now possible to estimate the expected signal from CO lines during an intensity mapping experiment. Based on individual detections only, Carilli et al (2016) estimate the mean surface brightness to 0.94 µK at 3mm and 0.55 µK at 1.3mm, these values being lower limits to take into account all the possible lines below detection.