Abstract

We report on the results of observations in the CO(1–0) transition of a complete sample of Southern, intermediate-redshift (z= 0.2–0.5) ultraluminous infrared galaxies (ULIRGs) using the Mopra 22-m telescope. The 11 ULIRGs with LFIR > 1012.5 L south of δ=−12° were observed with integration times that varied between 5 and 24 h. Four marginal detections were obtained for individual targets in the sample. The ‘stacked’ spectrum of the entire sample yields a high significance, 10σ detection of the CO(1–0) transition at an average redshift of z= 0.38. The tightest correlation of LFIR and LCO for published low-redshift ULIRG samples (z < 0.2) is obtained after normalization of both these measures to a fixed dust temperature. With this normalization the relationship is linear. The distribution of dust-to-molecular hydrogen gas mass displays a systematic increase in dust-to-gas mass with galaxy luminosity for low-redshift samples, but this ratio declines dramatically for intermediate-redshift ULIRGs down to values comparable to that of the Small Magellanic Cloud. The upper envelope to the distribution of ULIRG molecular mass as function of look-back time demonstrates a dramatic rise by almost an order of magnitude from the current epoch out to 5 Gyr. This increase in maximum ULIRG gas mass with look-back time is even more rapid than that of the star formation rate density.

1 INTRODUCTION

The Universe has evolved dramatically with cosmic time. The star formation rate density, in particular, first rose in the first few Gyr and then declined by more than 1 order of magnitude in the past 10 Gyr (e.g. Hopkins & Beacom 2006). At the same time, the galaxies hosting star formation have systematically changed from being dominated by the highest mass systems in the distant past to a more equal mix of galaxy masses (Heavens et al. 2004) in the most recent 0.5 Gyr, where a representative local volume has been sampled. The processes driving this dramatic evolution will only be understood if there is a more complete documentation of the time evolution of the major baryonic constituents of the cosmos.

An important baryonic constituent to consider in the context of galaxies is the molecular gas mass; since this is a prerequisite for star formation and as such plays a key role in determining its rate and location. While ideally nature would have provided a direct tracer of molecular hydrogen mass under all physical conditions, this is unfortunately not the case; making it necessary to infer molecular gas mass from indirect measures. An important proxy for molecular hydrogen is the detection of carbon monoxide (CO) emission lines. Despite the fact that the CO emission lines are often highly self-opaque, high-resolution studies of individual molecular clouds and nearby external galaxies have permitted at least rough calibration of the total molecular hydrogen mass that is statistically associated with an observed CO line luminosity (e.g. Dickman et al. 1986). The situation is complicated by a significant statistical scatter as well as the likely dependencies of this association on many other factors. A good recent overview of such calibration factors for the CO(1–0) transition is given in Tacconi et al. (2008).

The relatively high brightness of the CO(1–0) transition has permitted its routine detection in emission at z∼ 0.1 (e.g. Chung et al. 2009) in a few cases out to redshifts as large as z∼ 0.4 (Geach et al. 2009, 2011), and most recently in very luminous galaxies at significantly higher redshifts as well (Riechers et al. 2010; Ivison et al. 2011). Higher J transitions of CO have been detected out to redshifts of 1 < z < 3.5 (e.g. Genzel et al. 2010). In this paper we address a major gap in our observational knowledge of the molecular gas mass associated with galaxies by targeting a complete Southern sample of the 11 most luminous far-infrared (FIR) detected galaxies at redshifts 0.2 < z < 0.5 for deep integrations with the Mopra 22-m telescope. Only a handful of detections have yet been made in this redshift range (Solomon et al. 1997; Geach et al. 2009, 2011). Our study complements a similar programme undertaken by Combes et al. (2011) directed at 30 Northern and equatorial targets. Together these studies provide some insights into the evolution of the molecular gas mass in ULIRGs during this pivotal era in cosmic history.

The current paper is organized as follows. We begin with a brief definition of the sample in Section 2, describe the observations and data reduction methods in Section 3, and present the results in Section 4. A more extensive discussion of the results in a cosmological context is deferred to a subsequent publication. In this paper we adopt a flat cosmological model with Ωλ = 0.73 and a Hubble constant of 71 km s−1 Mpc−1 (Hinshaw et al. 2009).

2 SAMPLE DEFINITION

Studies of molecular gas in the local Universe (e.g. Solomon et al. 1997) have demonstrated that the largest gaseous reservoirs are apparently associated with the galaxies most luminous in the FIR. This correlation provides an opportunity to sample the molecular content at higher look-back times by targeting ultraluminous infrared galaxies [ULIRGs, defined by log (LFIR) > 12] in the relevant redshift interval for deep integrations. Estimates of the total molecular gas mass can then be made on the basis of the molecular mass function (Keres, Yun & Young 2003), which may evolve, and the space density of the observed target population. While this approach is no substitute for complete sampling of the molecular mass function at each look-back time, it provides a first indication of possible trends.

Our sample was defined by considering all galaxies tabulated within the NASA/IPAC Extragalactic Database (NED)1 with FIR detections at 60 and 100 μm and spectroscopic redshifts in the range z = 0.2–0.5 that exceeded a limiting FIR luminosity log (LFIR) > 12.5. The FIR luminosity has been defined by LFIR = 4π D2L CC FFIR, in terms of the luminosity distance DL, a colour correction factor, CC = 1.42 (Sanders & Mirabel 1996), and a FIR flux density, FFIR = 1.26 × 10−14(2.58 S60+ S100) W m−2 (Sanders & Mirabel 1996). The 11 targets below a declination, δ < −12°, that satisfied these requirements are listed in Table 1. The declination cut-off of the sample was chosen to be complementary to the Northern sample of Combes et al. (2011).

Table 1

The Southern ULIRG sample.

Name RA (2000) Dec. (2000) z 
F00320−3307 00:34:28.5 −32:51:13.0 0.439 
00397−1312 00:42:15.5 −12:56:03.0 0.261 72 
00406−3127 00:43:03.2 −31:10:49.0 0.3424 
02262−4110 02:28:15.2 −40:57:16.0 0.493 37 
02456−2220 02:47:51.3 −22:07:38.0 0.296 
03538−6432 03:54:25.2 −64:23:45.0 0.3007 
F04565−2615 04:58:34.7 −26:11:14.0 0.490 
07380−2342 07:40:09.8 −23:49:58.0 0.292 
23515−2917 23:54:06.5 −29:01:00.0 0.3349 
F23529−2119 23:55:33.0 −21:03:09.0 0.428 56 
F23555−3436 23:58:06.5 −34:19:47.0 0.490 
Name RA (2000) Dec. (2000) z 
F00320−3307 00:34:28.5 −32:51:13.0 0.439 
00397−1312 00:42:15.5 −12:56:03.0 0.261 72 
00406−3127 00:43:03.2 −31:10:49.0 0.3424 
02262−4110 02:28:15.2 −40:57:16.0 0.493 37 
02456−2220 02:47:51.3 −22:07:38.0 0.296 
03538−6432 03:54:25.2 −64:23:45.0 0.3007 
F04565−2615 04:58:34.7 −26:11:14.0 0.490 
07380−2342 07:40:09.8 −23:49:58.0 0.292 
23515−2917 23:54:06.5 −29:01:00.0 0.3349 
F23529−2119 23:55:33.0 −21:03:09.0 0.428 56 
F23555−3436 23:58:06.5 −34:19:47.0 0.490 

3 OBSERVATIONS AND REDUCTION

Observations were carried out on the Mopra 22-m telescope2 near Coonabarabran, Australia, between 2008 May 15 and October 29. Centre frequencies for the target galaxies varied between 77 and 91 GHz. The centre frequencies were placed in the middle of one of four intermediate-frequency bands, each of 2-GHz width, and each sampled by 8192 spectral channels in two perpendicular polarizations. System temperatures were measured continuously by calibration against a noise diode and varied between about 160 and 350 K during useful observing conditions. Calibration of the antenna temperature for atmospheric attenuation was updated at 15 min intervals by the use of an absorbing paddle at ambient temperature placed over the feed [employing the method of Ulich & Haas (1976)]. The telescope pointing was updated each hour using the catalogued SiO maser of smallest angular separation with the target galaxy, yielding a pointing accuracy of 5–10 arcsec, relative to a full width at half-maximum (FWHM) beamwidth of about 36 arcsec.

Standard target observing sessions lasted 1 hour and consisted of alternate target and reference spectra, each of 1 min duration, using a pair of reference locations offset by 15 arcmin to both the east and west of each target galaxy. Each target spectrum was calibrated with the average of the two adjoining offset reference spectra. Only those observing sessions with stable system temperatures below about 350 K were retained for further processing.

Individual 1 min integrations on the target were subjected to Fourier filtering. Peaks in the Fourier spectrum exceeding 10 times its rms fluctuation level were replaced with zero. A more rigorous Fourier clipping was then applied to narrow features in the Fourier spectrum. Localized Fourier peaks, exceeding the average Fourier amplitude over a sliding 10-pixel window by five times the rms fluctuation level, were clipped at this 5σ excursion level. Tests of this procedure verified that injected signals of the anticipated amplitude and linewidth of plausible detections of our target galaxies were not significantly degraded by this filtering. The typical improvement in the rms fluctuation level provided by this filtering was 50 per cent. The target integrations of individual observing sessions were averaged and a second order polynomial baseline was fitted to the central 1.75 GHz of the band and subtracted from the entire spectrum. All observing sessions for individual targets were averaged together with an inverse-variance weighting. No further baseline fitting of any type was applied to the averaged spectrum.

Total observing times per target varied between 5 and 24 h. Spectral smoothing of the final combined spectra provides a decreasing rms fluctuation level that scales approximately with the square root of bandwidth through about 30 km s−1. Further spectral smoothing, from 30 to 90 km s−1, decreases the fluctuation level by only about 50 per cent, rather than the expected 70 per cent, implying that residual systematic bandpass effects begin to dominate. The resulting rms sensitivity at a spectral resolution of about 90 km s−1 varied between 0.4 and 0.7 mK in terms of calibrated antenna temperature, T*A. Main beam brightness temperature is given by TB=T*AMB, and the main beam efficiency at 77–91 GHz is estimated to be ηMB∼ 0.49 (Ladd et al. 2005). The assumed telescope gain that relates calibrated antenna temperature to flux density in this frequency range3 is 22 Jy K−1.

4 RESULTS

Final spectra for each of the observed targets are shown in Fig. 1 at a spectral resolution of about 90 km s−1. Vertical lines in this figure mark an interval of 1000 km s−1 centred on the reference redshift listed in Table 2. The uncertainty in the published redshifts is indicated in each panel. These can be quite substantial. No high significance detections of the redshifted CO(1–0) line were achieved. Upper limits to the integrated line strength can be calculated by assuming a representative linewidth. Detections of CO (1–0) in lower redshift samples of ULIRGs (e.g. Solomon et al. 1997; Chung et al. 2009) suggest that the typical observed linewidth is about 300 km s−1. We therefore assumed a more conservative 500 km s−1 FWHM for the determination of an appropriate noise level. (Flux and mass limits become less restrictive the larger the assumed linewidth.) Since we are searching for the counterpart of a target with known position and velocity, the probability for a false positive detection is given by the tail probability of a standard normal distribution under the assumption of approximately Gaussian noise. A false positive exceeding 2σ then has a probability of 0.023. Even for our entire sample size of 11 targets, the false positive rate at this level of significance is only 0.25, which is still significantly less than unity. The 2σ upper limits to integrated line strength were determined from the measured fluctuation level over the entire spectrum (including any possible emission signature) after spectral smoothing to 500 km s−1 FWHM. The four cases (00397−1312, 03538−6432, F04565−2615 and 23515−2917) where the actual line integral within the central 1000 km s−1 of the spectrum exceeds this 2σ limit are noted as marginal detections in the Table, together with their 1σ errors. As noted above, some of the target redshifts have large uncertainties, particularly for F00320−3307 and F23555−3436. In both these cases, there may be evidence for CO emission at an offset velocity that may well be consistent with the current uncertainties in the systemic velocity.

Figure 1

Final spectra of CO(1–0) for the target galaxies in the sample shown at a spectral resolution of about 90 km s−1. Vertical lines mark an interval of 1000 km s−1 centred on the expected systemic velocity. The published uncertainty in the systemic velocity is indicated in each panel. The average stacked spectrum is shown at lower right with the average redshift indicated.

Figure 1

Final spectra of CO(1–0) for the target galaxies in the sample shown at a spectral resolution of about 90 km s−1. Vertical lines mark an interval of 1000 km s−1 centred on the expected systemic velocity. The published uncertainty in the systemic velocity is indicated in each panel. The average stacked spectrum is shown at lower right with the average redshift indicated.

Table 2

Source attributes and results.

Name z LFIR[log (L)] TD (K) MD [log (M)] τ (min) ΔT*A (mK)a SCO (Jy km s−1)b LCO[log (L)] LCO[log (K km s−1 pc2)] forumla[log (M)] 
F00320−3307 0.439 12.68 46 8.29 340 0.63 <7.9 <6.58 <10.89 <11.10 
00397−1312 0.261 72 12.67 52 8.02 285 0.61 12.6 ± 4.1 6.32: 10.63: 10.79: 
00406−3127 0.3424 12.58 50 8.03 283 0.69 <7.7 <6.35 <10.66 <10.84 
02262−4110 0.493 37 12.69 56 7.82 667 0.60 <7.8 <6.69 <11.00 <11.13 
02456−2220 0.296 12.50 46 8.17 560 0.51 <5.8 <6.10 <10.41 <10.63 
03538−6432 0.3007 12.58 49 8.07 377 0.50 11.1 ± 3.0 6.39: 10.70: 10.89: 
F04565−2615 0.490 12.62 49 8.11 915 0.64 7.8 ± 3.1 6.68: 10.99: 11.18: 
07380−2342 0.292 12.78 37 9.05 1485 0.40 <3.8 <5.90 <10.21 <10.52 
23515−2917 0.3349 12.54 46 8.17 455 0.49 8.5 ± 2.8 6.37: 10.68: 10.89: 
F23529−2119 0.428 56 12.52 47 8.11 440 0.48 <4.1 <6.28 <10.59 <10.80 
F23555−3436 0.490 12.67 46 8.31 552 0.65 <7.5 <6.66 <10.97 <11.18 
Average 0.38 ± 0.09 12.63 47 8.29   4.84 ± 0.48 6.45 10.76 10.96 
Name z LFIR[log (L)] TD (K) MD [log (M)] τ (min) ΔT*A (mK)a SCO (Jy km s−1)b LCO[log (L)] LCO[log (K km s−1 pc2)] forumla[log (M)] 
F00320−3307 0.439 12.68 46 8.29 340 0.63 <7.9 <6.58 <10.89 <11.10 
00397−1312 0.261 72 12.67 52 8.02 285 0.61 12.6 ± 4.1 6.32: 10.63: 10.79: 
00406−3127 0.3424 12.58 50 8.03 283 0.69 <7.7 <6.35 <10.66 <10.84 
02262−4110 0.493 37 12.69 56 7.82 667 0.60 <7.8 <6.69 <11.00 <11.13 
02456−2220 0.296 12.50 46 8.17 560 0.51 <5.8 <6.10 <10.41 <10.63 
03538−6432 0.3007 12.58 49 8.07 377 0.50 11.1 ± 3.0 6.39: 10.70: 10.89: 
F04565−2615 0.490 12.62 49 8.11 915 0.64 7.8 ± 3.1 6.68: 10.99: 11.18: 
07380−2342 0.292 12.78 37 9.05 1485 0.40 <3.8 <5.90 <10.21 <10.52 
23515−2917 0.3349 12.54 46 8.17 455 0.49 8.5 ± 2.8 6.37: 10.68: 10.89: 
F23529−2119 0.428 56 12.52 47 8.11 440 0.48 <4.1 <6.28 <10.59 <10.80 
F23555−3436 0.490 12.67 46 8.31 552 0.65 <7.5 <6.66 <10.97 <11.18 
Average 0.38 ± 0.09 12.63 47 8.29   4.84 ± 0.48 6.45 10.76 10.96 

aThe spectral rms is listed for a resolution of about 90 km s−1.

bUpper limits are 2σ for a 500 km s−1 linewidth.

In the lower-right panel of Fig. 1, we present the spectrum obtained by aligning all of the measured spectra to relative velocity and forming the simple unweighted average. This yields a high signal-to-noise ratio detection of the mean CO(1–0) line from our sample, as noted in Table 2.

The corresponding line luminosity is given by  

1
formula
or  
2
formula
for the luminosity distance, DL, in Mpc, the rest frequency, ν0, in GHz and the integrated line strength, SV, in Jy km s−1. Some authors make use of the quantity  
3
formula
or  
4
formula
with DL, ν0 and SV as defined above. For the CO(1–0) transition, these measures are related simply by LCO/LCO = 2.05 × 104 (K km s−1 pc2) L−1. The utility of the LCO measure is that it permits more direct comparison of different line transitions since it is formulated as a product of brightness temperature with surface area (Solomon et al. 1997). Line emission of the same brightness temperature originating from the same region will yield the same LCO, independent of transition or observing frequency.

Calculation of an associated total hydrogen mass requires an assumed relationship between this quantity and the CO(1–0) line luminosity. A good compilation of such conversion factors is given in fig. 10 of Tacconi et al. (2008). There is clearly a large scatter in the derived conversion factors, and there may well be underlying dependencies on many physical factors, including metallicity, gas surface density and particularly gas excitation temperature. A plausible conversion factor appropriate for ULIRGs is likely to be about 1 M (K km s−1 pc2)−1 or about 2 × 104 M L−1.

We plot the distribution of CO(1–0) and FIR luminosity in Fig. 2, including both our own results and those of Combes et al. (2011) as well as previously published studies of less distant and less luminous systems by Solomon et al. (1997), Chung et al. (2009) and Gao & Solomon (2004), all of which have been undertaken in the CO(1–0) transition. Although the low-redshift data lie on a well-defined locus in the plot, the intermediate redshift data from both the current study and that of Combes et al. (2011) show divergence from this simple correlation. A subset of the detections has a significant excess CO luminosity, while another subset consisting primarily of upper limits in CO is significantly underluminous. The difference in CO luminosity between these two populations is about 1 order of magnitude. A power law with slope of 1.37 ± 0.06 has been fitted by least squares to the three low-redshift samples and is overlaid in the figure. The correlation coefficient of the data points is 0.891. As noted by previous authors (e.g. Chung et al. 2009), this relationship is steeper than linear. A single power-law slope provides a reasonable representation of this correlation for FIR luminosities 9.3 < log (LFIR) < 12.2. As noted previously, the associated gas mass for a given CO(1–0) line luminosity is likely to vary with the gas kinetic and excitation temperatures, while LFIR is expected to be very sensitive to the dust temperature, TD, possibly varying as T4−6D (Soifer et al. 1989).

Figure 2

The relationship between CO(1–0) and FIR luminosity. Our results are indicated with the filled circles and 2σ upper limit symbols. The large filled pentagon represents our ‘stacked’ result. A power law of slope 1.37 ± 0.06 represents a least-squares fit to the low-redshift samples and has a correlation coefficient 0.891.

Figure 2

The relationship between CO(1–0) and FIR luminosity. Our results are indicated with the filled circles and 2σ upper limit symbols. The large filled pentagon represents our ‘stacked’ result. A power law of slope 1.37 ± 0.06 represents a least-squares fit to the low-redshift samples and has a correlation coefficient 0.891.

Gao & Solomon (2004) have demonstrated a higher degree of correlation of LFIR with LCO once a correction of the FIR luminosity for a varying dust temperature has been accounted for. The dust temperature can be estimated from the FIR photometry if an emmissivity law is assumed. This is often taken to be that of a Planck function multiplied by νβ, for frequency, ν, and power-law exponent, β. Lisenfeld, Isaak & Hills (2000) have found that the spectral energy distributions (SEDs) in their sample of FIR luminous galaxies could be fitted with 0.85 < β < 1.9. The dust temperature, in the Wien approximation, can be estimated from the ratio, RS=S1/S2, of flux densities at two fixed observing frequencies, ν1 and ν2, from  

5
formula
where Rν12. For the 60- and 100-μm IRAS bands, this yields  
6
formula
Since the integrated FIR luminosity is expected to scale as Tβ+ 4D, we have considered normalizations of the FIR luminosity down to a reference temperature of 25 K using the form:  
7
formula
The case has also been made that the CO(1–0) line luminosity would vary approximately linearly with TD (Solomon et al. 1997) if the dust and gas temperatures are reasonably coupled, suggesting a similar normalization of the form:  
8
formula
We find empirically that the values (α, β) = (1, 1) minimize the dispersion in the temperature-normalized CO(1–0) versus FIR luminosity relationship. The resulting relation is illustrated in Fig. 3, where a power law of slope 0.99 ± 0.04 has been fitted by least squares to the three low-redshift samples. The correlation coefficient with this choice of (α, β) is 0.902. A very similar relation, with a slope 1.03 ± 0.05, is obtained for (α, β) = (1, 1.5), which may be more representative of the FIR SEDs (Lisenfeld et al. 2000). The distinction between the CO-luminous and CO-poor populations is undiminished in the temperature corrected relationship.

Figure 3

The relationship between temperature-normalized CO(1–0) and FIR luminosity. LCO is normalized by (TD/25)1 and LFIR by (TD/25)5, for a reference dust temperature, TD = 25 K. A power law of slope 0.99 ± 0.04 represents a least-squares fit to the low-redshift samples and has a correlation coefficient 0.902. A similar fit is achieved for LCO normalized by (TD/25)1 and LFIR by (TD/25)5.5.

Figure 3

The relationship between temperature-normalized CO(1–0) and FIR luminosity. LCO is normalized by (TD/25)1 and LFIR by (TD/25)5, for a reference dust temperature, TD = 25 K. A power law of slope 0.99 ± 0.04 represents a least-squares fit to the low-redshift samples and has a correlation coefficient 0.902. A similar fit is achieved for LCO normalized by (TD/25)1 and LFIR by (TD/25)5.5.

We can estimate the dust mass associated with our targets from  

9
formula
where Sν0 is the observed flux density in an FIR band, DL is the luminosity distance, κνz is the absorption cross-section per unit dust mass at the target rest-frame frequency νz0(1 +z), and Bνz(TD) is the Planck function at this frequency for a dust temperature, TD (Hildebrand 1983). The dust absorption cross-section can be approximated by  
10
formula
in units of cm2 g−1 for wavelengths expressed in μm between 40 and 1000 μm. This analytic form fits the tabulated data of Draine (2003) to better than 5 per cent over the indicated wavelength range. We calculate the dust masses for our sample and the comparison samples noted previously using the observed IRAS 100-μm fluxes and the dust temperatures calculated with equation (6) and β = 1.5, listing these in Table 2.

Once the CO luminosity has been temperature-corrected, it seems more likely that a single conversion factor to total molecular hydrogen mass, appropriate to the 25 K reference temperature, might apply. From Tacconi et al. (2008), this might correspond to about 3 M (K km s−1 pc2)−1 or about 6 × 104 ML−1. Even so, there may well be other systematic dependencies and a large intrinsic scatter, which add significant uncertainty to this conversion factor. The relationship between molecular hydrogen mass and dust mass is shown in Fig. 4. A power-law fit to the low-redshift samples has slope 1.09 ± 0.05 and a correlation coefficient of 0.893, and is plotted in the figure. Also shown are diagonal lines corresponding to fixed dust-to-hydrogen mass ratios of 0.001, 0.003 and 0.01, which are labelled sub-Small Magellanic Cloud (sub-SMC), Large Magellanic Cloud (LMC) and Milky Way (MW), respectively (cf. Draine et al. 2007). The sub-SMC designation is used for the lowest curve since the SMC is estimated to have a mass ratio of 0.002 (e.g. Weingartner & Draine 2001). The correlated locus in Fig. 4 defined by the low-redshift samples systematically increases in dust-to-gas mass ratio with increasing gas mass, from SMC-like values at the low end, through LMC and up to MW values for forumla.

Figure 4

The relationship between molecular hydrogen mass and dust mass. A power law of slope 1.09 ± 0.05 is shown as a dashed line and represents a least-squares fit to the low-redshift samples and has a correlation coefficient 0.893. Dotted lines labelled with sub-SMC, LMC and MW represent increasing dust-to-molecular hydrogen mass ratios of 0.001, 0.003 and 0.01, respectively.

Figure 4

The relationship between molecular hydrogen mass and dust mass. A power law of slope 1.09 ± 0.05 is shown as a dashed line and represents a least-squares fit to the low-redshift samples and has a correlation coefficient 0.893. Dotted lines labelled with sub-SMC, LMC and MW represent increasing dust-to-molecular hydrogen mass ratios of 0.001, 0.003 and 0.01, respectively.

What is striking in Fig. 4 is that all of the intermediate-redshift detections in both the current and Combes et al. (2011) studies depart from the low-redshift trend in the direction of extremely low dust-to-molecular gas mass ratio. Our well-defined sample average point of 0.0021 ± 0.0002 is comparable to the dust-to-hydrogen mass ratio of the SMC. In contrast, there are several upper limits in the figure which appear to be dust-rich (or gas-poor) by about an order of magnitude. The most extreme point from the current sample is the source IRAS 0738−2342, while in the Combes et al. (2011) sample it is IRAS 19104 + 8436. Two additional sources in the Northern sample, [HB89] 1821 + 643 and F 00415−0737, are also discrepant but less extreme. While very little is currently known about F 00415−0737, Ruiz et al. (2010) have recently presented SED fits for IRAS 0738−2342 and [HB89] 1821 + 643. They estimate relative active galactic nucleus (AGN):starburst contributions to the bolometric luminosity of about 50:50 and 80:20 for these two sources. Both IRAS 19104 + 8436 and [HB89] 1821 + 643 are classified as Seyfert 1/quasi-stellar objects, suggesting that LFIR for IRAS 19104 + 8436 may also be dominated by non-thermal rather than dust emission.

It seems plausible that the apparent ‘gas-poor’ subpopulation seen at intermediate redshift in Figs 2, 3 and 4 may simply be a manifestation of AGN contamination, with the star formation dominated population demonstrating a well-defined trend relative to the low-redshift samples, consistent with a dramatic decline in the dust-to-molecular gas mass ratio.

We consider the distribution of ULIRG molecular hydrogen mass with look-back time, τ, in Fig. 5. Since ULIRGs are identified by their extreme luminosity in the FIR band, where self-opacity and foreground extinction effects are minimal, they represent a population that is easily recognized in unbiased surveys of the sky. Both our own study and the others shown in the figure were drawn from the most luminous sources in the IRAS all-sky survey. As such they are not expected to suffer from incompleteness at high luminosity, and despite the fact that they are a rare population, they permit a useful assessment of the associated molecular gas mass that is required to feed the ULIRG phenomenon. Although the lowest detected gas masses at each look-back time simply reflect the sensitivity of the observation and the cut-off LFIR of the sample, the upper envelope to mass in Fig. 5 is a significant attribute of the population, at least once the sample extends over a representative volume of about 107 Mpc3, since the most luminous sources are clearly the easiest to detect at any redshift. This condition should be met for redshifts greater than about 0.03 or look-back times exceeding 0.4 Gyr. This local volume saturation effect is apparent in the figure for τ < 0.5 Gyr. The indicative curve drawn in the figure,  

11
formula
demonstrates the order of magnitude decline in maximum molecular gas mass of the ULIRG population during the past 5 Gyr. Such a decline is comparable to, but even more dramatic than, that seen in the star formation rate density (Hopkins & Beacom 2006) which can be well described by  
12
formula
for a star formation rate density forumla in M yr−1 Mpc−3 for τ < 8 Gyr. This contrast suggests significant changes in the molecular gas mass function at even these modest look-back times.

Figure 5

The relationship between look-back time and ULIRG molecular hydrogen mass. An indicative upper envelope to molecular gas mass, forumla, is plotted demonstrating the order of magnitude decline in maximum ULIRG gas mass over the past 5 Gyr.

Figure 5

The relationship between look-back time and ULIRG molecular hydrogen mass. An indicative upper envelope to molecular gas mass, forumla, is plotted demonstrating the order of magnitude decline in maximum ULIRG gas mass over the past 5 Gyr.

We will present further analysis of the time evolution of molecular gas and other baryonic constituents of galaxies in a subsequent publication.

1
The NED is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
2
The Mopra radio telescope is part of the Australia Telescope which is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO.
3
Calibration factors are documented in the Mopra guide (http://www.narrabri.atnf.csiro.au/mopra/mopragu.pdf).

We thank an anonymous referee for constructive suggestions of improvements to the manuscript, including formation of a ‘stacked’ spectrum for the sample. The Mopra radio telescope is part of the Australia Telescope National Facility which is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO.

REFERENCES

Chung
A.
Narayanan
G.
Yun
M. S.
Heyer
M.
Erickson
N. R.
,
2009
,
AJ
 ,
138
,
858
Combes
F.
Garcia-Burillo
S.
Braine
J.
Schinnerer
E.
Walter
F.
Colina
L.
,
2011
,
A&A
 ,
528
,
124
Dickman
R. L.
Snell
R. L.
Schloerb
F. P.
Solomon
P. M.
,
1986
,
ApJ
 ,
309
,
326
Draine
B. T.
,
2003
,
ARA&A
 ,
41
,
241
Draine
B. T.
et al.,
2007
,
ApJ
 ,
663
,
866
Gao
Y.
Solomon
P. M.
,
2004
,
ApJ
 ,
606
,
271
Geach
J. E.
Smail
I.
Coppin
K.
Moran
S. M.
Edge
A. C.
Ellis
R. S.
,
2009
,
MNRAS
 ,
395
,
L62
Geach
J. E.
Smail
I.
Moran
S. M.
MacArthur
L. A.
Del
P.
Lagos
C.
Edge
A. C.
,
2011
,
ApJ
 ,
730
,
L19
Genzel
R.
et al.,
2010
,
MNRAS
 ,
407
,
2091
Heavens
A.
Panter
B.
Jimenez
R.
Dunlop
J.
,
2004
,
Nat
 ,
428
,
625
Hildebrand
R. H.
,
1983
,
QJRAS
 ,
24
,
267
Hinshaw
G.
et al.,
2009
,
ApJS
 ,
180
,
225
Hopkins
A. M.
Beacom
J. F.
,
2006
,
ApJ
 ,
651
,
142
Ivison
R. J.
Papadopoulos
P. P.
Smail
I.
Greve
T. R.
Thomson
A. P.
Xilouris
E. M.
Chapman
S. C.
,
2011
,
MNRAS
 ,
412
,
1913
Keres
D.
Yun
M. S.
Young
J. S.
,
2003
,
ApJ
 ,
582
,
659
Ladd
N.
Purcell
C.
Wong
T.
Robertson
S.
,
2005
,
PASP
 ,
22
,
62
Lisenfeld
U.
Isaak
K. G.
Hills
R.
,
2000
,
MNRAS
 ,
312
,
433
Riechers
D. A.
Carilli
C. L.
Walter
F.
Momjian
E.
,
2010
,
ApJ
 ,
714
,
L153
Ruiz
A.
Miniutti
G.
Panessa
F.
Carrera
F. J.
,
2010
,
A&A
 ,
515
,
A99
Sanders
D. B.
Mirabel
F.
,
1996
,
ARA&A
 ,
34
,
749
Soifer
B. T.
Boehmer
L.
Neugebauer
G.
Sanders
D. B.
,
1989
,
AJ
 ,
98
,
766
Solomon
P. M.
Downes
D.
Radford
S. J. E.
Barrett
J. W.
,
1997
,
ApJ
 ,
478
,
144
Tacconi
L. J.
et al.,
2008
,
ApJ
 ,
680
,
246
Ulich
B. L.
Haas
R. W.
,
1976
,
ApJS
 ,
30
,
247
Weingartner
J. C.
Draine
B. T.
,
2001
,
ApJ
 ,
548
,
296