Letter

# Large molecular gas reservoirs in ancestors of Milky Way-mass galaxies nine billion years ago

• Nature Astronomy 1, Article number: 0003 (2016)
• doi:10.1038/s41550-016-0003
Accepted:
Published online:

## Abstract

The gas accretion and star formation histories of galaxies like the Milky Way remain an outstanding problem in astrophysics 1,2 . Observations show that 8 billion years ago, the progenitors to Milky Way-mass galaxies were forming stars 30 times faster than today and were predicted to be rich in molecular gas 3 , in contrast to the low present-day gas fractions (<10%) 4,​5,​6 . Here we show the detection of molecular gas from the CO (J = 3–2) emission (rest-frame 345.8 GHz) in galaxies at redshifts z = 1.2–1.3, selected to have the stellar mass and star formation rate of the progenitors of today’s Milky Way-mass galaxies. The CO emission reveals large molecular gas masses, comparable to or exceeding the galaxy stellar masses, and implying that most of the baryons are in cold gas, not stars. The total luminosities of the galaxies from star formation and CO luminosities yield long gas consumption timescales. Compared to local spiral galaxies, the star formation efficiency, estimated from the ratio of total infrared luminosity (L IR) to CO emission, has remained nearly constant since redshift z = 1.2, despite the order of magnitude decrease in gas fraction, consistent with the results for other galaxies at this epoch 7,​8,​9,​10 . Therefore, the physical processes that determine the rate at which gas cools to form stars in distant galaxies appear to be similar to that in local galaxies.

Studies of the distribution of stellar ages and elemental abundances in the Milky Way and M31 have shown that most of their stars formed in the distant past, more than seven billion years ago 11,12 . This agrees with recent work showing that star formation in present-day galaxies with the mass of the Milky Way peaked more than 8 billion years ago 3 , at z > 1, with star formation rates (SFRs) that exceed 30 M  yr−1, compared with a present-day SFR of 1.7 ± 0.2 M  yr−1 for the Milky Way 13 .

Theoretical models explain periods with high SFRs as a result of rapid baryonic gas accretion from the intergalactic medium (IGM), which leads to high cold gas concentrations in galaxies at earlier times 14 . These models predict that the gas settles into rotationally supported, highly turbulent disks, which fragment to form stars 15 . Observations of star-forming galaxies at z > 1 (stellar masses, M > 2 × 1010M ) show evidence for gas-rich rotating disks 16,​17,​18,​19 , supporting these theories. However, the situation is far from settled for more common lower mass (M 1010M ) galaxies such as the progenitors to the Milky Way. Some models 20 predict that these galaxies should experience early, rapid star formation, leaving low gas fractions (<10%) at redshifts z ≈ 1. Others predict that the gas flows from the IGM can perturb and disrupt the formation of disk instabilities, thereby suppressing star formation in galaxies and extending star formation histories 21,22 . The first step to understanding star formation in galaxies like the Milky Way is to measure the amount of the cold gas in their progenitors at z > 1. As the gas is the fuel for star formation, the ratio of the SFR to gas mass could be used to test the physical processes in the models 23 .

With the greatly improved sensitivity offered by the Atacama Large Millimeter Array (ALMA), we are now able to explore the evolution of cold molecular gas in low-mass galaxies at redshifts z > 1. With ALMA, we observed the J = 3–2 transition of CO in four galaxies with the stellar mass and SFR expected of the main progenitors to present-day Milky Way-mass galaxies at redshifts z = 1.2−1.3 selected from deep imaging by the FourStar Galaxy Evolution (ZFOURGE) survey 24 (see the discussion in Methods). Figure 1 shows the integrated emission from the CO J = 3–2 transition in these galaxies, where the detections range in significance from 4.8σ to 13.7σ (root mean squared). The CO(J =3–2) emission coincides with the spatial positions of the galaxies in Hubble Space Telescope (HST) imaging (Fig. 1); the small offsets are consistent with astrometric calibrations and ALMA beam smearing. Table 1 gives the measured properties of these galaxies. The ALMA detections of CO emission probe the molecular mass in galaxies with the stellar mass and SFRs that the main progenitor of the Milky Way was expected to have 8.5 billion years ago. This provides an important extension of previous work, as the galaxies in our sample have lower stellar masses and SFRs than have been generally possible to study at these redshifts 10 .

CO is the most luminous tracer of molecular hydrogen (H2), the fuel for star formation. The CO specific intensity from the J to J−1 transition, I CO(J−[J−1]), is a function of both the gas density and temperature. In high redshift galaxies, studies have shown that the average excitation of CO(J =3−2) is similar to that of star-forming regions in the Milky Way 25 , and we assume an integrated Rayleigh–Jeans brightness temperature line ratio 26 , r 31 = I CO(3−2)/I CO(1−0) × (1/3)2 = 0.66. The total CO luminosity in the J = 1 to 0 transition is then $L CO ′ = 3.25 × 10 7 r 31 − 1 I CO ( 3 − 2 ) ν obs − 2 D L 2 ( 1 + z ) − 3$ , where ν obs is the frequency (in gigahertz) of the CO emission in the observed frame and D L is the luminosity distance in megaparsec. Table 1 displays the $L CO ′$ values. Using lower values of r 31 ≈ 0.4−0.5, as indicated in some other studies of star-forming galaxies at z ≈ 1−2 25,27 , would increase the $L CO ′$ values slightly, but would not change our conclusions.

The combination of the CO luminosity and the luminosity from newly formed stars provides a crucial constraint on the star formation efficiency (SFE). We use the thermal infrared luminosity (L IR, measured over 8–1,000 μm in the rest frame), which originates from dust in dense molecular clouds heated by young stars, and is directly proportional to the total SFR. We measured L IR for galaxies in our study using model fits to fluxes measured from Spitzer Space Telescope and Herschel Space Observatory imaging covering 24–160 μm (see Methods). Table 1 displays these values. They span L IR = (1.5−2.7) × 1011L (corresponding to SFRs of 15–30 M  yr−1). Uncertainties are approximately 0.2 dex (60%) and are dominated by systematics from differences in the infrared model (see Methods).

Figure 2 shows the SFE, defined as $L IR / L CO ′$ , as a function of $L CO ′$ for the z = 1.2–1.3 galaxies in our sample compared with control samples. With ALMA we are now able efficiently to probe the CO luminosities of z > 1 star-forming galaxies at a factor of two lower than was previously possible. The galaxies in our sample have SFEs typical of the upper range of both local spiral galaxies and more massive, high-redshift star-forming galaxies. In such galaxies, star formation occurs in rotationally supported disks. In at least two of our galaxies, the CO(J =3−2) spectra show line profiles with a strong double peak (see Methods). This and the apparent presence of spatial velocity shear (see Methods) observed in our analysis of the CO data suggest that the same may be true for all of the z = 1.2−1.3 galaxies in our sample. Therefore, although both the SFRs and gas fractions are substantially higher in these distant galaxies, star formation probably occurs in rotating disks, where the physical processes governing the evolution of the gas appear to be similar to those of spiral galaxies in the local Universe. In contrast, the SFEs of more luminous, rarer objects, for example ultraluminous infrared galaxies (ULIRGs), quasi-stellar objects (QSOs) and submillimetre galaxies (SMGs), are significantly enhanced in the local and distant Universe. A prevailing theory is that ULIRGs, QSOs and SMGs are a result of increased gas densities from major gas-rich mergers 28 . These conditions seem to be inconsistent with the galaxies in our sample, suggesting that major mergers are not common among the main progenitors of Milky Way-mass galaxies at z 1.2–1.3.

The inverse of the SFE is proportional to the gas consumption timescale, which corresponds to a range of 200 to 700 Myr for the galaxies in our sample. In contrast, the consumption timescales for ULIRGs, QSOs and SMGs are less than 10 Myr 29 . Star formation in the average, main progenitor of Milky Way galaxies at z = 1.2–1.3 appears to be long-lasting, and comparable to findings for other star-forming disk galaxies at high redshifts 7,​8,​9,​10,18 .

The CO luminosities imply very high molecular gas fractions for the galaxies in our sample at z = 1.2–1.3, where we adopt the ratio of CO luminosity to mass in H2 Gas (M gas) for Galactic star-forming regions because the SFEs are similar (see Methods). Table 1 lists these values. Figure 3 shows the molecular gas fractions, f gas = M gas/(M gas + M), derived from CO observations as a function of M. While present day Milky Way-sized galaxies have low gas fractions, f gas < 10%, the results from our sample imply that the main progenitors to these galaxies at z = 1.2−1.3 have much higher values: in three of the galaxies in our sample the molecular gas mass was greater than or equal to the stellar mass (f gas 50%). This is consistent with the indirect gas fractions of galaxies at these redshifts that were inferred from the thermal dust emission 18,30 . These higher f gas values also argue against models with early, rapid gas consumption 20 and favour longer lasting, feedback-regulated star formation 21,​22,​23 .

The high molecular gas fractions and SFRs of the z = 1.2–1.3 galaxies in our sample imply that they will double their stellar mass within the gas consumption timescale. Therefore, at z 1.2, these galaxies have most, but not all, of the fuel needed to produce the M 5 × 1010M in stars in Milky Way-mass galaxies at present (Fig. 3). The average baryon accretion rate from the IGM must exceed 6 M  yr−1 at earlier times (z > 1.2) to account for the galaxies’ total stellar and molecular masses. In constrast, the galaxies only need to acquire 30–50% more baryonic mass from z 1 to the present (even accounting for losses from stellar evolution), which corresponds to an average gas accretion rate of only 1–2 M  yr−1. This reflects a dwindling supply of fresh baryonic gas. Therefore, Milky Way-mass galaxies appear to have accreted most of their gas at z > 1.2, during the first few billion years of history.

## ZFOURGE dataset

We selected the galaxies in our sample from the ZFOURGE survey 24 . The main ZFOURGE survey obtained very deep near-infrared imaging in five medium-band filters (J 1, J 2, J 3, H s, H l) from the FourStar instrument 31 on the Magellan Baade 6.5 m telescope in the three southern fields covered by the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS) HST imaging 32,33 . The ZFOURGE catalogues combine the FourStar images with ancillary ground-based imaging (spanning 0.3–2.5 μm), the CANDELS HST Advanced Camera for Surveys (ACS) and Wide Field Camera 3 imaging, and Spitzer Infrared Array Camera imaging (spanning 3.6–8.0 μm). For this study, we selected targets from the earlier version 2.1 ZFOURGE catalogues. These include photometric redshifts and stellar masses estimated from the multiwavelength catalogues as described elsewhere 3,34 . Of interest here, the catalogues are complete for objects with limiting stellar masses log(M /M ) ≥ 9.0−9.2 in the redshift range 1.0 < z < 1.5, well below the stellar masses of the typical progenitor of a Milky Way-mass galaxy 3 .

## Selection of Milky Way-mass galaxy progenitors

We selected galaxies as targets for ALMA observations of the CO(J = 3–2) transition that have the typical stellar mass and SFR of progenitors to Milky Way-mass galaxies at z = 1.2−1.3. We identified progenitors of galaxies with the present-day stellar mass of a Milky Way-mass galaxy (M = 5 × 1010M at z = 0) 13,35 using abundance-matching techniques 36 . The progenitors to such galaxies had a median stellar mass of log(M /M ) = 10.21 at z = 1.1−1.4 (ref. 3). These abundance matching methods give a stellar mass ~0.2 dex lower than those selected at a constant co-moving number density at these redshifts 37 . More recent work has shown that progenitors of Milky Way-mass galaxies span a range of stellar mass at z = 1.2, with a 30th to 70th percentile range of log(M /M ) = 10.05–10.34, and a median value consistent with the median above 38 . While observations of CO in z > 1 galaxies have probed stellar masses down to log(M /M ) > 10.4 (ref. 10), these correspond to the more massive progenitors of present-day Milky Way-mass galaxies. Our sample extends studies of the CO emission to the median stellar mass of progenitors of present-day Milky Way galaxies.

We also selected galaxies with the typical SFRs of the Milky Way-mass progenitors for observations with ALMA. In our previous work we used deep Spitzer and Herschel imaging to measure an average total L IR, L IR = (2.0 ± 0.1) × 1011L , for all Milky Way-mass progenitor galaxies in this redshift and stellar mass range in ZFOURGE 3 . This corresponds to a SFR of 21 ± 2 M  yr−1.

In summary, we used the following criteria to select targets for ALMA:

1. Photometric redshift, 1.1< z <1.4

2. Stellar mass, −0.15< log(M/M) − 10.2< +0.15

3. SFR, −0.15< log(SFR/M)yr−1 − 1.3< +0.15

4. Measured spectroscopic redshift

The restrictions on photometric redshift, stellar mass and SFR results in the selection of galaxies with stellar mass and SFR within 0.15 dex (that is, within 40%) of the expected median values of the progenitors to Milky Way-mass galaxies.

The final selection criterion requires that the galaxies have a redshift measured from spectroscopy. This ensures that the redshift accuracy is sufficient for the redshifted CO(J =3–2) emission line to fall within the frequency range of an ALMA spectral window. While the ZFOURGE photometric redshifts are good ( $σ z /(1 + z )<1$ %) 24 , they are not sufficient for this purpose.

Of the 24,690 galaxies in the full ZFOURGE catalogue, 39 satisfied the first three selection criteria. At the time of our proposal for ALMA for cycle 2 observations (2013 December), seven galaxies satisfied all of our selection criteria (including having a published spectroscopic redshift in the literature) 39 . From these, we selected four objects that offered some contrast in SFR (spanning nearly 0.3 dex). For the analysis presented here, we re-derived stellar masses and uncertainties using the FAST code 40 with an extended stellar population library (including a broader metallicity range of 0.2−1.0 Z and a finer grid spacing of star formation histories) compared with the one used for the ZFOURGE catalogues 24 . These stellar masses with 68% likelihood ranges are listed in Table 1, and are consistent with those in the v2.1 and v3.4 ZFOURGE catalogues.

All four of the galaxies selected for ALMA observations have properties that are typical of progenitors to a Milky Way-mass galaxy at z = 1.2−1.3. Supplementary Figure 1 shows the SFR–stellar mass relationship for galaxies from ZFOURGE with 1.1 < z<1.4. Galaxies with the stellar mass and SFR of the Milky Way-mass progenitors lie on the star-forming ‘main sequence’ 41 , and this includes the four galaxies we observed with ALMA. Therefore, they correspond to a typical star-forming galaxy at these redshifts. Previous studies of CO emission in galaxies at these redshifts have been limited to higher SFRs (30 M  yr−1) and/or stellar masses (M > 2.5 × 1010M ) 9,10 . As illustrated in Supplementary Fig. 1, the observations of our sample with ALMA provide an important extension compared with previous studies. Furthermore, previous studies required up to 25 h of integration with the Institut de Radioastronomie Millimétrique Plateau de Bure interferometer to detect galaxies at these masses and SFR limits 10 . Our ALMA observations required only about 40 min, demonstrating the efficacy of ALMA for this science.

## Far-infrared data and infrared luminosities

The ZFOURGE fields include imaging at far-infrared wavelengths from the Spitzer Multiband Imaging Photometer (24 μm) and the Herschel Photoconductor Array Camera and Spectrometer (PACS) (70, 100 and 160 μm). Fluxes in these data are measured using source detections based on previous locations of sources in HST Wide Field Camera 3 F160W filter (1.6 μm) data using methods identical to those described elsewhere 42,​43,​44 . We measured flux uncertainties and evaluated source completeness through extensive artificial object simulations following the same procedures discussed elsewhere 44,45 . The 24–160 μm flux densities for the four objects studied here are listed in Supplementary Table 1. Note that one source, ZFOURGE Chandra Deep Field South (CDFS) 4409, has no coverage by PACS 70 μm.

In all cases, the infrared flux densities and flux uncertainties that we measure for our sources are consistent with other published values 42,46 available online (http://irsa.ipac.caltech.edu/data/Herschel/PEP). In many cases, our measured flux densities at 70 μm and 160 μm have a signal-to-noise ratio (SNR) <3. While formally undetected, we include this information in our analysis as it provides important constraints on the total infrared emission from these galaxies.

To measure total infrared luminosities, L IR we fit models of the infrared spectral energy distribution 47,​48,​49 to the flux densities shown in Supplementary Table 1. Because the data sample the Wein side of the thermal emission comprehensively, the constraints on L IR are quite robust. Supplementary Fig. 2 shows the fits using the published models 47,48 which bracket the range of values. The slight differences in the shapes of the infrared spectral energy distributions lead to systematically different values of L IR, where L IR values from the templates of Rieke et al. are higher by Δ(logL IR) = 0.1–0.2 dex. We have also calculated L IR ignoring data where objects are detected at <2σ, but this produces changes in L IR by <15% in most cases. We therefore adopt the L IR from the fits to the model of Rieke et al. for all the infrared data, which we report in Table 1. If we instead adopt the results from the fits to the models of Chary & Elbaz, the SFEs would decline, and gas consumption timescales would increase for the z = 1.2−1.3 galaxies in our sample studied here. This would bring the SFEs further in line with local spiral galaxies, strengthening that conclusion.

The total L IR for the z = 1.2–1.3 galaxies in our sample span L IR = (1.5−3.2)  × 1011L , as listed in Table 1. In these galaxies, most of the bolometric emission from star formation is emitted in the thermal infrared range. In contrast, according to our measurements, the rest-frame ultraviolet emission (uncorrected for dust extinction) contributes only 4–6% to the total SFR implied by the L IR in these galaxies. This is consistent with mean values measured in local luminous infrared galaxies 50 .

## ALMA observations and data reduction

Our cycle 2 ALMA observations were taken between 2015 April 6 and 2015 May 2 in Band 4 with 36 antennas in the C34-2 configuration, and these observations provided a maximum baseline of 348.5 m. For each source, we configured ALMA to observe in four spectral windows, 1.875 GHz per window, spanning the frequency range 134.48–156.90 GHz (depending on the expected frequency of the CO(J =3–2) transition for each source). We centred the CO(J =3–2) line in one of the spectral windows assuming the optical spectroscopic redshift obtained from the literature 39 . The ALMA integrations ranged between 37.3 and 41.8 min on source. One source (ZFOURGE CDFS 6497) was erroneously observed twice, and received double the exposure time. The other spectral windows probe the continuum of the line. Flux, phase and band-pass calibrators were also obtained. Supplementary Table 2 provides details about the observations for each source.

We reduced the data with Common Astronomy Software Applications 51 (CASA) version 4.5.0-REL with the calibration script supplied by the National Radio Astronomy Observatory. We then ran the cleaning algorithm with natural weighting. For the spectral window containing the CO(J =3–2) line, we reduced the data with channels of 25 km s−1 and 75 km s−1 with a cell size of 0.2 arcsec. The angular sizes of the cleaned beam full-width at half-maximum are given in Supplementary Table 2, and this cell size gives six or seven cells along the semi-minor axis of the beam.

We also attempted to measure the continuum for each galaxy by cleaning and combining the spectral windows excluding channels expected to have CO emission. We failed to detect any signal of the continuum; we also therefore made no correction for the continuum to the CO line fluxes.

Supplementary Fig. 3 shows the spectra of the CO(J =3–2) emission for the four galaxies. For each galaxy, there is positive emission in the channels at the expected location of the CO(J =3–2) line. To determine the peak of emission we fitted the spectra to models using single and double gaussian distributions. The velocity offsets between the CO redshift and the redshift from the optical spectroscopy range from −30 to +100 km s−1. This is consistent with uncertainties in the redshift measurements.

We created total intensity maps of the CO(J =3–2) lines in each galaxy by combining the channels showing positive emission around the expected position of each line. We also created first moment (velocity) and second moment (velocity dispersion) maps to study galaxy gas dynamics. From the total intensity maps corrected for the primary beam, we measured integrated flux densities for the CO(J =3–2) transition, I CO(3−2), for each galaxy in our sample using the two-dimensional profile fitting tool in CASA. These are presented in Table 1, and are in the range I CO(3−2) = 0.11–0.33 Jy km s−1. While the SNRs of the integrated values in Table 1 are in the range 2.3–7.7, the detection significance (measured from the peak of the emission) is much higher, with SNRs in the range 4.8–13.7.

As discussed in the next section, the ALMA spectra in Supplementary Fig. 3 show evidence for complex velocities, except for ZFOURGE CDFS 467, where the data quality is lower. The spectrum of this object does show a tentative, weak emission to the blue side of the systemic redshift (at velocities in the range −400 to −200 km s−1). However, the integrated emission from these channels is not significant, having a SNR of ~2.0 at the peak. When the emission from channels on the blue side is summed with that from channels on the red side (where the object is detected), the overall significance of the detection is reduced from 4.8 to 3.0. We therefore do not include this emission in the analysis of this object. However, including it would increase the CO luminosity by a factor of 1.6, making it more consistent with the other objects in the sample.

## Velocity shear in CO emission

Our analysis of the spectra of the CO(J =3–2) emission in the four z = 1.2−1.3 galaxies in our sample in Supplementary Fig. 3 shows that in many cases a model with a double gaussian distribution fits the data better than a model with a single gaussian distribution. This is consistent with the expected signature of rotation. Further evidence comes from observations of velocity shear in the galaxies: some galaxies show spatial variations in their velocity components. For three of the galaxies with the strongest emission, (ZFOURGE CDFS 4409, 6497 and 8193) we measured total CO(J =3–2) intensity maps separately from the channels towards the blue side (approaching) and towards the red side (receding) relative to the velocity with the minimum emission between the peaks. Two of these galaxies (ZFOURGE 6497 and 8193) showed velocity shear (in the third object the SNR was too low to robustly determine the centroid of the two separate components). Supplementary Figure 4 shows that there are spatial offsets in the centroids of the emissions in the red and blue components in these galaxies. While the beam size precludes accurate modelling of the velocity shear, the spatial variations are consistent with rotation.

Taken together, the CO spectra and spatial separation of the approaching and receding velocity components provide reasonable evidence for rotation in the z = 1.2−1.3 galaxies in our sample. While the presence of velocity structures with double peaks in the CO spectra could be expected for merging systems, we consider this unlikely for two reasons. First, the HST morphologies of the galaxies in our sample (Fig. 1) show no indications of double nuclei, which would be expected if a merger was responsible for the CO velocity structure. Second, the galaxies in our sample lie on the main sequence of the SFR–M relationship (Supplementary Fig. 1), and they show no indications of merger-induced starbursts on their SFEs, L IR/ $L CO ′$ (Fig. 2). Direct confirmation of rotation in these galaxies would require kinematic data with a higher spatial resolution, which would be possible to obtain from ALMA in larger configurations, but with considerably more exposure time.

## Molecular gas mass from CO emission

The observed CO luminosity is proportional to the total cold molecular gas mass, M gas. At the temperatures and pressures of the interstellar medium in our galaxies, we expect that most of the gas exists in the molecular phase 52 , and therefore the molecular gas accounts for the majority of baryons in the gas phase. The constant of proportionality (the ratio of the gas mass to light), α CO, is given by $(1) α CO = M gas / L CO ′$

Based on the L IR/ $L CO ′$ ratios, the conditions in the z = 1.2–1.3 galaxies in our sample appear to be similar to those in normal star-forming regions and star-forming disk galaxies, which show values of α CO ≈ 4 M (K km s−1 pc 2 )−1, and trace the total amount of molecular gas including a correction for helium 29,53 . The factor for converting CO to molecular gas is α = 4.3 M (K km s−1 pc2)−1 for star-forming regions in the Milky Way galaxy and in ‘normal’ star-forming galaxies 53 . The galaxies in our sample have L IR/ $L CO ′$ ratios consistent with those of other normal star-forming galaxies (Fig. 2). We therefore adopt α CO = 3.6 M (K km s−1 pc2)−1 that has been found to apply to normal star-forming (more massive) galaxies at these redshifts, and contains a calibration uncertainty of 22% 7 .

There is evidence that the ratio of CO to molecular gas, α CO, varies with metallicity, Z, where the values in the discussion above correspond to solar values, Z = Z . Theoretical work 54,55 predicts that α COZ −0.5, while empirical measurements at redshifts z > 1 suggest a possible steeper relationship 56 , α COZ −1.2 to Z −1.8. If the galaxies in our ALMA sample have Z < Z , one may expect an increase in α CO. Work on the relationship between stellar mass and metallicity (MZ) 57,​58,​59 at 0.9 < z < 1.3 shows that star-forming galaxies in the mass range of our sample should have metallicities in the range 0.6–1.0 Z . This implies a higher α CO (and higher gas masses) by at most a factor of 2. This would correspond to even more dramatic evolution in the gas fraction from z 1.3 to the present for Milky Way-mass galaxies. For this reason, we adopt the (more conservative) ratio of CO to molecular gas for solar metallicity, α CO = 3.6 M (K km s−1 pc2)−1, as stated in the discussion above.

## Conventions

Throughout, we assume a Chabrier initial mass function 60 when deriving stellar masses and SFRs. For all cosmological calculations, we assume Ω m  = 0.3, Ω Λ = 0.7, and H 0 = 70 km s−1 Mpc−1, consistent with the recent constraints from the Planck Collaboration 61 and the local distance scale 62 .

## Data Availability

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request. Data from the ZFOURGE survey can be obtained from http://zfourge.tamu.edu.

How to cite this article: Papovich, C. et al. Large molecular gas reservoirs in ancestors of Milky Way-mass galaxies nine billion years ago. Nat. Astron. 1, 0003 (2016).

## References

1. 1.

, , & Forming realistic late-type spirals in a ΛCDM universe: the Eris simulation. Astrophys. J. 742, 76 (2011).

2. 2.

, , , & A diversity of progenitors and histories for isolated spiral galaxies. Astrophys. J. 756, 26 (2012).

3. 3.

et al. ZFOURGE/CANDELS: on the evolution of M* galaxy progenitors from z =3 to 0.5. Astrophys. J. 803, 26 (2015).

4. 4.

Molecular gas in nearby galaxies. I: CO observations of a distance-limited sample. Astron. Astrophys. 272, 123–136 (1993).

5. 5.

et al. Heracles: The HERA CO line extragalactic survey. Astron. J. 137, 4670–4696 (2009).

6. 6.

et al. COLD GASS, an IRAM legacy survey of molecular gas in massive galaxies. I: relations between H2, H i, stellar content and structural properties. Mon. Not. R. Astron. Soc. 415, 32–60 (2011).

7. 7.

et al. Very high gas fractions and extended gas reservoirs in z = 1.5 disk galaxies. Astrophys. J. 713, 686–707 (2010).

8. 8.

et al. The molecular gas content of z = 3 Lyman break galaxies: evidence of a non-evolving gas fraction in main-sequence galaxies at z > 2. Astrophys. J. 758, L9 (2012).

9. 9.

et al. Dust temperature and CO–H2 conversion factor variations in the SFR-M plane. Astron. Astrophys. 548, A22 (2012).

10. 10.

et al. Phibss: molecular gas content and scaling relations in z1-3 massive, main-sequence star-forming galaxies. Astrophys. J. 768, 74 (2013).

11. 11.

et al. The dominant epoch of star formation in the Milky Way formed the thick disk. Astrophys. J. 781, L31 (2014).

12. 12.

et al. The spatially-resolved star formation history of the M31 outer disc. Mon. Not. R. Astron. Soc. 453, L113–L117 (2015).

13. 13.

& Improved estimates of the Milky Way’s stellar mass and star formation rate from hierarchical Bayesian meta-analysis. Astrophys. J. 806, 96 (2015).

14. 14.

et al. Elemental abundances in Milky Way-like galaxies from a hierarchical galaxy formation model. Mon. Not. R. Astron. Soc. 445, 970–987 (2014).

15. 15.

et al. Cold streams in early massive hot haloes as the main mode of galaxy formation. Nature 457, 451–454 (2009).

16. 16.

et al. The rapid formation of a large rotating disk galaxy three billion years after the Big Bang. Nature 442, 786–789 (2006).

17. 17.

et al. High molecular gas fractions in normal massive star-forming galaxies in the young Universe. Nature 463, 781–784 (2010).

18. 18.

et al. Combined CO and dust scaling relations of depletion time and molecular gas fractions with cosmic time, specific star-formation rate, and stellar mass. Astrophys. J. 800, 20 (2015).

19. 19.

et al. The KMOS3D survey: design, first results, and the evolution of galaxy kinematics from 0.7 <= z <= 2.7. Astrophys. J. 799, 209 (2015).

20. 20.

, & High-redshift clumpy discs and bulges in cosmological simulations. Mon. Not. R. Astron. Soc. 404, 2151–2169 (2010).

21. 21.

, & On the effect of cosmological inflow on turbulence and instability in galactic discs. Mon. Not. R. Astron. Soc. 425, 788–800 (2012).

22. 22.

& Delayed star formation in high-redshift stream-fed galaxies. Mon. Not. R. Astron. Soc. 437, L56–L60 (2014).

23. 23.

& On the interplay between star formation and feedback in galaxy formation simulations. Astrophys. J. 804, 18 (2015).

24. 24.

et al. The FourStar Galaxy Evolution Survey (ZFOURGE): ultraviolet to far-infrared catalogs, medium–bandwidth photometric redshifts with improved accuracy, stellar masses, and confirmation of quiescent galaxies to z3.5. Astrophys. J. 830, 51 (2016).

25. 25.

et al. CO excitation of normal star-forming galaxies out to z = 1.5 as regulated by the properties of their interstellar medium. Astron. Astrophys. 577, A46 (2015).

26. 26.

et al. High-resolution imaging of PHIBSS z 2 main-sequence galaxies in CO J = 1–0. Astrophys. J. 809, 175 (2015).

27. 27.

et al. CO(1–0) line imaging of massive star-forming disc galaxies at z=1.5–2.2. Mon. Not. R. Astron. Soc. 442, 558–564 (2014).

28. 28.

, & Dusty star-forming galaxies at high redshift. Phys. Rep. 541, 45–161 (2014).

29. 29.

& Cool gas in high-redshift galaxies. Annu. Rev. Astron. Astrophys. 51, 105–161 (2013).

30. 30.

et al. ISM masses and the star formation law at z=1 to 6: ALMA observations of dust continuum in 145 galaxies in the COSMOS survey field. Astrophys. J. 820, 83 (2016).

31. 31.

et al. FourStar: the near-infrared imager for the 6.5 μm Baade telescope at Las Campanas Observatory. Publ. Astron. Soc. Pacif. 125, 654–682 (2013).

32. 32.

et al. CANDELS: the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey. Astrophys. J. Suppl. Ser. 197, 35 (2011).

33. 33.

et al. CANDELS: the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey — the Hubble Space Telescope observations, imaging data products, and mosaics. Astrophys. J. Suppl. Ser. 197, 36 (2011).

34. 34.

et al. Galaxy stellar mass functions from ZFOURGE/CANDELS: an excess of low-mass galaxies since z = 2 and the rapid buildup of quiescent galaxies. Astrophys. J. 783, 85 (2014).

35. 35.

, & The mid-life crisis of the Milky Way and M31. Astrophys. J. 736, 84 (2011).

36. 36.

, & Galactic star formation and accretion histories from matching galaxies to dark matter haloes. Mon. Not. R. Astron. Soc. 428, 3121–3138 (2013).

37. 37.

et al. The assembly of Milky-Way-like galaxies since z2.5. Astrophys. J. 771, L35 (2013).

38. 38.

& An improved probabilistic approach for linking progenitor and descendant galaxy populations using comoving number density. Preprint at (2016).

39. 39.

et al. The Great Observatories Origins Deep Survey. VLT/FORS2 spectroscopy in the GOODS-South field: part III. Astron. Astrophys. 478, 83–92 (2008).

40. 40.

et al. An ultra-deep near-infrared spectrum of a compact quiescent galaxy at z = 2.2. Astrophys. J. 700, 221–231 (2009).

41. 41.

et al. The SFR-M* relation and empirical star-formation histories from ZFOURGE* at 0.5 < z < 4. Astrophys. J. 817, 118 (2016).

42. 42.

et al. GOODS-Herschel: an infrared main sequence for star-forming galaxies. Astron. Astrophys. 533, A119 (2011).

43. 43.

et al. Evolution of the dusty infrared luminosity function from z = 0 to z = 2.3 using observations from Spitzer. Astron. Astrophys. 528, A35 (2011).

44. 44.

et al. The deepest Herschel-PACS far-infrared survey: number counts and infrared luminosity functions from combined PEP/GOODS-H observations. Astron. Astrophys. 553, A132 (2013).

45. 45.

et al. PEP Full Public Data Release (DR1): PACS Data (Max Planck Institute, 2013);

46. 46.

et al. PACS Evolutionary Probe (PEP) — a Herschel key program. Astron. Astrophys. 532, A90 (2011).

47. 47.

et al. Determining star formation rates for infrared galaxies. Astrophys. J. 692, 556–573 (2009).

48. 48.

& Interpreting the cosmic infrared background: constraints on the evolution of the dust-enshrouded star formation rate. Astrophys. J. 556, 562–581 (2001).

49. 49.

et al. Infrared spectral energy distributions of nearby galaxies. Astrophys. J. 633, 857–870 (2005).

50. 50.

et al. The Great Observatories All-sky LIRG Survey: comparison of ultraviolet and far-infrared properties. Astrophys. J. 715, 572–588 (2010).

51. 51.

, & In Astronomical Data Analysis Software and Systems XVI (eds Shaw, R. A., Hill, F. & Bell, D. J ) 127 (Astronomical Society of the Pacific Conference Series, 2007).

52. 52.

et al. Different star formation laws for disks versus starbursts at low and high redshifts. Astrophys. J. 714, L118–L122 (2010).

53. 53.

, & The CO-to-H 2 conversion factor. Annu. Rev. Astron. Astrophys. 51, 207–268 (2013).

54. 54.

, & The X-factor in galaxies. I: dependence on environment and scale. Astrophys. J. 747, 124 (2012).

55. 55.

, , & A general model for the CO–H 2 conversion factor in galaxies with applications to the star formation law. Mon. Not. R. Astron. Soc. 421, 3127–3146 (2012).

56. 56.

et al. The metallicity dependence of the CO−H 2 conversion factor in z > 1 star-forming galaxies. Astrophys. J. 746, 69 (2012).

57. 57.

et al. Physical properties of galaxies and their evolution in the VIMOS VLT Deep Survey. II. Extending the mass–metallicity relation to the range z 0.89–1.24. Astron. Astrophys. 495, 73–81 (2009).

58. 58.

, & The mass–metallicity and luminosity–metallicity relations from DEEP2 at z 0.8. Astrophys. J. 730, 137 (2011).

59. 59.

et al. A fundamental metallicity relation for galaxies at z = 0.84–1.47 from HiZELS. Mon. Not. R. Astron. Soc. 436, 1130–1141 (2013).

60. 60.

Galactic stellar and substellar initial mass function. Publ. Astron. Soc. Pacif. 115, 763–795 (2003).

61. 61.

Planck Collaboration, Planck 2015 results. XIII: Cosmological parameters. Astrophys. Astron. 594, A13 (2016).

62. 62.

et al. A 24% determination of the local value of the Hubble constant. Astrophys. J. 826, 56 (2016).

63. 63.

& The star formation rate and dense molecular gas in galaxies. Astrophys. J. 606, 271–290 (2004).

64. 64.

et al. Galaxy evolution and star formation efficiency at 0.2 < z < 0.6. Astron. Astrophys. 528, A124 (2011).

65. 65.

et al. Gas fraction and star formation efficiency at z < 1.0. Astron. Astrophys. 550, A41 (2013).

## Acknowledgements

The authors thank their colleagues on the CANDELS and ZFOURGE surveys for providing high quality data. The authors also thank the ALMA staff for facilitating the observations and aiding in the calibration and reduction process. The authors acknowledge generous support from the Mitchell Institute for Fundamental Physics and Astronomy at Texas A&M University. This paper makes use of the following ALMA data: ADS/JAO.ALMA#2011.0.01234.S. ALMA is a partnership of the European Southern Observatory (representing its member states), the National Science Foundation (USA) and the National Institutes of Natural Sciences (Japan), together with the National Research Council (Canada), the National Science Council and the Academia Sinica Institute of Astronomy and Astrophysics (Taiwan) and the Korea Astronomy and Space Science Institute (Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by the European Southern Observatory, Associated Universities, Inc./National Radio Astronomy Observatory and the National Astronomical Observatory of Japan. The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.

## Affiliations

1. ### George P. and Cynthia Woods Mitchell Institute for Fundamental Physics and Astronomy, Texas A&M University, College Station, Texas 78743, USA

• C. Papovich
•  & K.-V. Tran
2. ### Department of Physics and Astronomy, Texas A&M University, 4242 TAMU, College Station, Texas 78743, USA

• C. Papovich
•  & K.-V. Tran
3. ### Leiden Observatory, Leiden University, PO Box 9513, Leiden 2300 RA, The Netherlands

• I. Labbé
•  & C. Straatman
4. ### Centre for Astrophysics & Supercomputing, Swinburne University, Hawthorn, Victoria 3122, Australia

• K. Glazebrook
• , G. Bekiaris
•  & D. Fisher
5. ### National Optical Astronomy Observatory, Tucson, Arizona 85719, USA

• M. Dickinson
•  & H. Inami
6. ### Department of Astronomy, The University of Texas at Austin, 2515 Speedway, Stop C1400, Austin, Texas 78712, USA

• S. L. Finkelstein
•  & R. C. Livermore

• H. Inami

• L. Spitler

• L. Spitler

## Contributions

C.P. led the ALMA observing programme, handled the data reduction and led the writing of the manuscript. I.L., K.G., R.Q., L.S., C.S. and K.-V.T. contributed extensively to the ZFOURGE data set, used in much of the analysis. S.L.F., D.F. and R.C.L. contributed to the design of the ALMA observing programme and assisted in the reduction and interpretation of the ALMA data. G.B. and K.G. assisted in the interpretation of the ALMA data. M.D. and H.I. carried out the data analysis of the Spitzer and Herschel imaging. All coauthors contributed to the writing of the manuscript and to the ALMA observing programme.

## Competing interests

The authors declare no competing financial interests.

## Corresponding author

Correspondence to C. Papovich.

## PDF files

1. 1.

### Supplementary Information

Supplementary Tables 1-2, Supplementary Figures 1–4