ARlogo Annu. Rev. Astron. Astrophys. 2013. 51: 207-268
Copyright © 2013 by Annual Reviews. All rights reserved

Next Contents Previous


The improved sensitivity of millimeter and centimeter-wave instruments has allowed detection of CO emission from an increasingly diverse range of systems out to higher and higher redshifts. Most of the objects observed in CO to date represent the bright, rare end of the luminosity distribution: so-called submillimeter galaxies (SMGs) and QSO hosts. Some recent studies, however, target rotating disks that lie at the high-mass end of the star-forming blue sequence at their redshift (i.e., have typical star formation rates given their stellar mass — we will refer to them as "main sequence" galaxies). CO observations thus now begin to sample the regime of "main sequence" galaxies (Tacconi et al. 2010, Daddi et al. 2010), and will expand to lower luminosity systems over the next decade.

Estimating molecular gas masses is frequently the main goal of CO observations at high redshifts. Unfortunately, direct determination of XCO in high redshift objects remains tremendously challenging. An additional complicating factor is that most high-redshift observations do not measure the J = 1 → 0 transition of CO, but higher rotational transitions. Thus translating these measurements into molecular masses requires, at least in principle, understanding CO excitation in addition to the other physics governing XCO.

At this stage, the modeling of optically thin isotopologues (e.g., Papadopoulos et al. 2012) may offer the best opportunity for direct XCO measurements at high redshift. Such observations have been too costly to undertake, and even with ALMA may only be practical for the brightest objects. Dust continuum observations offer another route to estimate molecular masses, although their application at high redshift requires understanding dust-to-gas ratios in relation to metallicities — a complex problem that ultimately requires knowledge of the balance of ISM enrichment, dust production, and destruction processes in galaxies. In lieu of direct measurements, the best route is to understand the physical drivers of XCO and to apply knowledge acquired from local galaxies to systems at high redshift.

8.1. Observed CO Line Ratios

The ratio of the J = 3 → 2 to J = 1 → 0 transition, r31 ident T3 / T1, exhibits a wide range of values across galaxies and has been observed at both low and high redshift. Mauersberger et al. (1999) and Yao et al. (2003) consider r31 for large samples of nearby galaxies with infrared luminosities 9 ltapprox log(LFIR) ltapprox 12. They find a mean r31 approx 0.63-0.66 (median r31 ~ 0.5) and a broad distribution of values. No strong correlations between r31 and dust temperature or luminosity are evident, but they note that r31 increases with increasing concentration and star formation efficiency (measured as LFIR / M(H2), though note the CO intensity in the denominator). Iono et al. (2009) examine 15 luminous LIRGs and ULIRGs with log(LFIR) gtapprox 11.5 and find mean r31 approx 0.48 (median 0.4) integrated over whole galaxies, with a higher mean r31 approx 0.96 at the location of peak CO emission. Mao et al. (2010) discuss r31 in a sample of over 60 galaxies, finding that barred galaxies and starbursts have the higher averaged r31 values (r31 ~ 0.89 ± 0.11 for starbursts), followed by AGNs (see their Table 4). More recently, Papadopoulos et al. (2011) report the results of a large sample of LIRGs and ULIRGs composed of new observations and literature compilation, where they find mean r31 approx 0.67 and r21 approx 0.91. By contrast with these samples of active galaxies, the inner portion of the Milky Way has a ratio r31 approx 0.28 ± 0.17 (Fixsen, Bennett & Mather 1999), denoting a correspondingly lower excitation in a z = 0 "normal" star forming galaxy.

To first order r31 offers a tool to distinguish excited, star-forming gas similar to that seen in LIRGS and ULIRGS from the gas found in more normal systems like the Milky Way. As emphasized by a number of authors, however, these line ratios are hardly unambiguous indicators of local conditions. For example, Papadopoulos et al. (2011) show that for their sample of LIRGs and ULIRGs possible densities and temperatures are often poorly constrained, with data allowing high density (n ~ 104 cm-3), normal temperature (Tkin ~ 20 K) or low density (n ~ 300 cm-3) high temperature (Tkin ~ 150 K) solutions. Note as a limitation the assumption of similar filling fractions for galaxy-wide averages in all transitions, while high-resolution maps frequently find excitation gradients. It is thus key to appreciate that r31, although useful as a first order excitation indicator, combines the effects of both excitation and beam filling.

With the improving sensitivity of the available instrumentation, measurements of line ratios including CO J = 1 → 0 at high redshift are rapidly increasing. For SMGs, the presumed high redshift simile of local ULIRGs, the measured line ratios resemble those found for local LIRGs and ULIRGs. Harris et al. (2010), Swinbank et al. (2010), and Ivison et al. (2011) report mean r31 approx 0.68, 0.66, and 0.55 respectively for a handful of SMGs. Carilli et al. (2010) and Bothwell et al. (2012) note that fitting observed SMG spectral line energy distributions requires a cool and a warm component, with most of the mass in the former. Riechers et al. (2011b) find low integrated r41 approx 0.25-0.6 in two SMGs, noting the existence of excitation gradients, with CO J = 1 → 0 in a more extended distribution than the higher transition. At the brightest end of the galaxy distribution Riechers et al. (2011) report r31 ~ 0.96-1.06 for four high-redshift SMGs hosting quasars, suggesting that the excitation in these sources is such that the emission is thermalized to higher rotational transitions (see also Weiss et al. 2007 for a CO excitation discussion of a number of bright SMG and QSO-host high-redshift galaxies).

The lowest galaxy luminosities currently probed by high redshift CO observations correspond to the high luminosity end of the star-forming galaxy main sequence. Unlike SMGs or local ULIRGS, these systems are not major mergers. Excitation data remain scarce for these "main sequence" disk galaxies. Dannerbauer et al. (2009) observed a BzK galaxy at z approx 1.5 report r32 ~ 0.5 and r21 ~ 0.6. Aravena et al. (2010) report r31 approx 0.61 and r21 ~ 0.4-1.2 in a few BzK galaxies. These observations have large uncertainties, but suggest that the conditions in high-z disk galaxies are not as extreme at those in local ULIRGs. As a counterpoint, Riechers et al. (2010) discuss the CO J = 1 → 0 detection of two highly magnified (µ ~ 30) low mass galaxies where they find r31 approx 0.72 and 0.78 with errors of ~ 0.2, suggesting higher excitation than the previous examples if there is no differential lensing. For further discussion see the review by Carilli & Walter in this same issue.

8.2. Estimates of XCO in High Redshift Systems

A number of observational studies have considered the problem of XCO in high-redshift sources, combining line ratio measurements, consistency arguments, and scaling relations to argue for plausible values.

Tacconi et al. (2008) discuss the values of XCO applicable to a sample of high-z star-forming galaxies dominated by SMGs but also including lower mass galaxies. Allowing for a dark matter contribution of 10%-20% within a 10 kpc radius, they minimize the chi2 of the difference between the dynamical mass and the sum of the stellar, gas, and dark matter. They find a Galactic XCO to be strongly disfavored, while a smaller, ULIRG-like XCO,20 approx 0.5 produces much more satisfactory results for SMGs (Fig. 10).

Figure 10

Figure 10. Minimization of the difference between dynamical mass and the sum of the gaseous, stellar, and dark matter components, for a sample of 9 high-z galaxies with high quality data, dominated by SMGs (Tacconi et al. 2008). chi2 and reduced chi2 (left and right side axes, respectively) are shown here as a function of chosen Initial Mass Function, for different gas (alpha) and dark matter (fdark) parameters. The analysis distinguishes between alpha in the SMG subsample (alphaSMG) and in the UV selected subsample (alphaUV) of lower mass galaxies.

In their SMG sample, Ivison et al. (2011) find XCO,20 ~ 0.9-2.3 to be compatible with the observed star formation histories and dynamical and stellar masses. They argue for lower XCO,20 ltapprox 0.5 in most cases, however, based on star formation efficiency considerations. They decompose the molecular gas into a warm, star-forming phase (with intrinsic r31 ~ 0.9) and a cold, quiescent phase (r31 ~ 0.3), and invoke an approximately maximal starburst for the star-forming phase. The maximal starburst, LIR / M(H2) approx 500 Lodot Modot-1, represents the largest allowed ratio of star formation to H2 before radiation pressure disperses a starburst (e.g., Thompson, Quataert & Murray 2005).

While Tacconi et al. (2008) and Ivison et al. (2011) point to similarities between ULIRGs and SMGs, note that Bothwell et al. (2010) find that several SMGs have gas distributions (imaged in J ≥ 3 transitions) more extended than those of local ULIRGs. This may suggest that local ULIRGs may not be the best analogues of SMGs, and that some SMGs may not arise from major mergers. The excitation gradients seen by Ivison et al. (2011) and Riechers et al. (2011) also point to extended quiescent gas reservoirs in SMGs that are not frequently observed in local ULIRGs.

Daddi et al. (2010b) and Genzel et al. (2010) study high-redshift galaxies selected from optically identified objects at z ~ 1.2-2.2. These are still massive systems, but drawn from the main sequence at their redshifts rather than major merger-driven starbursts. Kinematically, most of these systems are extended rotating disks, although with larger velocity dispersions than local disks. Daddi et al. (2010), Genzel et al. (2010), and Tacconi et al. (2010) argue that, because their CO emission likely arises from collections of self-gravitating GMCs (as suggested by local estimates of Toomre Q, Genzel et al. 2011) with modest temperatures T ~ 20-35 K, these systems have Galactic XCO. Additional constraints come from dynamical measurements that, together with simulations and modeling of the stellar populations, yield a gas mass resulting in XCO,20 approx 1.7 ± 0.4, a value considerably higher than observed in ULIRGs or estimated for SMGs (Daddi et al. 2010b). Note that these authors also provide a calibration for the dynamical mass estimation in this type of galaxy.

A different approach to quantify the amount of gas in a galaxy is to try to use the continuum dust emission. Magdis et al. (2011) use this approach in two very well studied high-z galaxies, a massive SMG at z ~ 4 and a "main sequence" disk star-forming galaxy at z ~ 1.5. They take advantage of the well characterized SEDs of these objects to calculate their dust masses employing different approaches to dust modeling (traditional modified blackbody as well as the type of models used by Draine et al. 2007). This dust mass is converted into a gas mass (assumed to be dominated by molecular gas) using a metallicity derived through the corresponding SFR-mass-metallicity relation (e.g., Mannucci et al. 2010) and assuming a Galactic dust-to-gas ratio corrected (approximately linearly) by metallicity. The result is XCO,20 ~ 0.4 ± 0.2 for the SMG, and XCO,20 ~ 1.9 ± 1.4 for the disk galaxy. Magnelli et al. (2012) follow a similar procedure in a mixed sample of 17 high-z galaxies, including "merger-like" and "main sequence" disks, finding values of XCO consistent with Galactic for the "main sequence" disks and and a factor of several lower for the "merger-like" galaxies. They also find XCO propto Tdust-0.8, similar to the behavior expected from Eq. 12.

Although this new use of dust observations at high-redshift is promising, we caution that going from dust SEDs to gas masses involves a large chain of assumptions that is fraught with potential problems. At a fundamental level, we are far from understanding the dust creation-destruction balance which sets the dust-to-gas ratio in galaxies (e.g., Draine 2009, Dwek & Cherchneff 2011 and references therein). Moreover, the translation of a SED into a dust mass relies almost entirely on dust grain properties (composition, mass emissivity, size distribution) that have been derived for the Milky Way (Draine & Li 2001, 2007b), and are extremely uncertain even for local galaxies (e.g., Galliano et al. 2011). Note that the self-consistent dust-based XCO estimates discussed Section 5.2 are emphasized precisely because they are free to a large degree from these problems. Nonetheless, this is an interesting approach that reinforces what seems to be the outstanding trend discussed above. ULIRG-like conversion factors for SMG (merger) galaxies, and Milky Way-like conversion factors for disks at high redshift.

Genzel et al. (2012) present 44 low-mass high-redshift galaxies, showing that observations are starting to probe the low metallicity regime at high redshift. The authors find the effect already discussed for local low metallicity galaxies in Section 6, where CO is disproportionally faint for the star formation activity (Fig. 11). By assuming that these galaxies obey the same gas surface-density to star formation relation observed in local disks (e.g., Leroy et al. 2008), Genzel et al. (2012) estimate H2 masses and derive a metallicity-dependent calibration for XCO at high redshift. They find XCO propto Z-1.3 - Z-1.8, in approximate agreement with local measurements (Leroy et al. 2011, Schruba et al. 2012).

Figure 11

Figure 11. Metallicity effects in high redshift "main sequence" galaxies (Genzel et al. 2012). The top panels show rest-frame Halpha images of zC406690 and Q2343-BX610, two z ~ 2 galaxies with similar rotational velocities (vrot approx 224 and 216 km s-1 respectively), star formation rates (SFR approx 480 and 212 Modot yr-1, including extinction correction), and stellar masses (M* approx 4.3 and 17 × 1010 Modot). The bottom panels show the corresponding rest frame optical spectrum containing the Halpha and [NII] transitions at 6563 and 6585 Å (red), and the CO J = 3 → 2 emission (blue). The galaxy zC406690 has a low metallicity, indicated by a low [NII] / Halpha ratio, and a correspondingly low CO emission, despite its large Halpha flux and star formation activity.

8.3. Synthesis: XCO at High Redshift

Estimates of the appropriate XCO for high redshift galaxies rely largely on scaling and consistency arguments. Because they probe the physical state of the molecular gas, line ratio measurements represent a particularly powerful tool for such arguments. Given the difficulties in observing the J = 1 → 0 line, they also represent a critical measurement to understand how to translate observations of high-J transitions into CO J = 1 → 0. Studies of local LIRGs and ULIRGs show r31 gtapprox 0.5 globally in these sources, and r31 ~ 1 when focusing on the active regions. Similar ratios are observed in very massive galaxies at high redshifts (SMGs), where r31 ~ 0.5-0.7 for integrated fluxes, and r31 ~ 1 is observed in the very compact starbursts surrounding QSOs. CO J = 1 → 0 data for main sequence star-forming galaxies at high-z remain scarce, making it is difficult to assert whether they possess large low excitation reservoirs of molecular gas. As the data quality, resolution, and availability improves, we expect that the simple single-component models used today will evolve into more realistic multi-component models for the molecular ISM of these sources.

A number of authors have explored the value of XCO applicable to samples of high-redshift galaxies. The consensus, validated to first order by numerical modeling (e.g., Narayanan et al. 2011), is that massive merger-driven starbursts such as SMGs are most consistent with a low XCO similar to local ULIRGs, while blue-sequence galaxy disks most likely have higher XCO, similar to local disks. This is reasonable in terms of the physics that drive the value of XCO. High density environments with an extended warm molecular phase not contained in self-gravitating clouds will result in low XCO, while molecular gas contained in collections of self-gravitating GMCs will have XCO close to the Galactic disk value. We expect the most heavily star-forming of these disks to have higher H2 temperatures, and consequently somewhat lower XCO values. Because of the opposing effects of density and temperature in self-gravitating GMCs (for example, Eq. 12), however, it appears that values as low as those observed in ULIRGs will only occur when the CO emission is dominated by an extended warm component that is not self gravitating (e.g., Papadopoulos et al. 2012).

Multi-line studies including both low- and high-J transitions allow more rigorous constraints on the density and kinetic temperature of the gas (Papadopoulos et al. 2012). In the near future, multi-line studies of CO isotopologues or paired observations of CO and optically thin sub-mm dust emission will offer additional, more direct constraints on XCO, though with their caveats and shortcomings.

An exciting development at high redshift is the emergence of main sequence "normal" galaxy surveys that begin to sample the lower metallicity regime prevalent at early times. In these samples we see XCO effects that are consistent with those already discussed for local galaxies (Section 6): metal-poor galaxies are underluminous in CO for their star formation activity (Genzel et al. 2012). This situation will be increasingly common, and more extreme, as CO surveys sample lower galaxy masses and earlier times: by z ~ 3-4 galaxy metallicities drop by 0.5-0.7 dex for a fixed stellar mass, and the characteristic galaxy mass also becomes lower (Mannucci et al. 2009). Ultimately the best probe of the molecular gas content of these low metallicity young galaxies may be the fine structure line of [CII], or perhaps the dust continuum.

Next Contents Previous