7. INFLOWS, OUTFLOWS, AND RECYCLING

7.1. The Problems: Galaxy Fueling and “Missing” Metals

Recent findings show that the CGM possesses a significant budget of baryons, but how are they feeding galaxies across the spectrum of galaxy masses (Figure 2)? Accreting gas passes through the CGM on its journey from the IGM to galaxies, where it presumably leaves some observable signatures that we can use to characterize the inflows. The rates of accretion onto galaxies and of outflow out of galaxies are crucial parameters in most models of galaxy evolution (Tinsley, 1980). However, there is not agreement about where and how a galaxy's fuel source is regulated. It is often assumed gas inflow from the IGM is balanced by the sum of star formation, gas ejection as outflows, and any net buildup of gas in the ISM (Lilly et al., 2013, Dekel & Mandelker, 2014, Somerville & Davé, 2015). This formulation completely omits the role of the CGM, even at the phenomenological level, but this “bathtub” model appears to nonetheless describe the many broad trends in galaxy scaling relations with redshift (Dekel & Mandelker, 2014).Col These models, though they do not explicitly address the CGM's composition or physical state, nonetheless have specific implications for its content and evolution (e.g., Shattow, Croton & Bibiano, 2015). Conversely, models that use physical principles to describe the regulation of flows between the CGM and ISM (Voit et al., 2015b) can reproduce the same phenomenological galaxy scaling relations without detailed treatments of star formation inside galaxies. With observations of the CGM and its dynamics, we can potentially assess whether its role in regulating star formation is trivial, as the former models assume, or essential, as the latter models assume. Ideally, CGM observations would not only answer this question, but also reveal how it fuels star formation and manages outflows as a function of galaxy mass.

The observations we have discussed up to this point reveal the CGM (at low z) as a massive gaseous medium with a rich internal kinematic structure that is, in bulk, consistent with being bound to the host galaxies. Yet the degeneracy between kinematics and the physical location of absorbing gas can easily get lost in transverse sightline observations.

In simulations the CGM can appear to have obvious and well-ordered large-scale structure, with accreting and outflowing gas occupying physically distinct regions such as filaments and biconical outflows (Shen et al., 2012, Corlies & Schiminovich, 2016, see also Figure 3), but at low redshift, circumgalactic gas tends to be more well mixed, with instantaneous velocities having little bearing on the origin or fate of a particular pocket of gas (Ford et al., 2014, Muratov et al., 2015, Christensen et al., 2016), though this is also seen at z = 3 (van de Voort et al., 2012). In light of the observational projection effects, and theoretical cautions, we will now consider what can be learned from observing inflow and outflow directly in down-the-barrel observations, in which we interpret gas blueshifted relative to the galaxy as outflowing and redshifted gas as inflowing. These observations are better at probing gas in or near the disk-halo interface rather than the “proper” CGM out in the halo. Considering them in conjunction with CGM finding from transverse sightlines promises insights into the dynamics of the CGM that are not otherwise available.

7.2. Empirical Signs of Fueling and Inflows

Gas accretion is perhaps the most fundamental process in their formation (Fox & Davé, 2017), as they must acquire gas, but feedback is optional. In the prevailing theoretical paradigm, gas flowing into galaxies at ≲ 10¹² M_⊙ should be dynamically and thermally cold, while more massive halos receive most of their baryons as hotter (T > 10⁵) gas (Dekel & Woo, 2003, Dekel & Birnboim, 2006, Kereš et al., 2005, Kereš & Hernquist, 2009, Stewart et al., 2011, though see Nelson et al., 2013). Thus cold, dense, metal-poor CGM gas is often interpreted as direct evidence of accretion. First, cool, dense CGM gas is abundant in the form of LLSs. A large fraction of these are metal-poor at all redshifts Lehner et al. (2013), Glidden et al. (2016), Cooper et al. (2015). Metal-poor LLSs are evident as tracers of accretion in high resolution simulations (Fumagalli et al., 2011, Hafen et al., 2016). The cool, bound H i seen in the CGM of z ∼ 0.25 galaxies (Tumlinson et al., 2013) should have a short cooling time. Finally, the finding from COS-GASS that there is a correlation between interstellar and circumgalactic H i (Borthakur et al., 2015) implies a connection between circumgalactic fuel and star forming fuel. Though sub-L^* and dwarf galaxies have not yet had their “cool” CGM masses measured directly, the widespread presence of Lyα at similar strength suggests they too possess significant budgets of cold halo gas.

All this evidence taken together strongly indicates that galaxies possess large reservoirs of CGM gas eligible for accretion. Yet evidence for fuel does not automatically constitute evidence for fueling: bound, cold gas has turned up in halos where its presence is surprising, such as the CGM of passive galaxies (Thom et al., 2012). The actual fate of this material is unclear: how can we claim the bound cold gas is fueling star forming galaxies but not the passive galaxies? We therefore seek direct signatures of gas accretion onto galaxies. Yet these signatures are notoriously difficult to observe as incoming material may be metal poor, ionized, and obscured by outflowing material. Once gas is near the disk, proving empirically that it is accreting can be extremely difficult when it is seen in projection and its kinematics are easily confused with disk material.

The Milky Way itself provides direct and unambiguous evidence for inflow in the form of its blueshifted high-velocity clouds (HVCs) and the striking Magellanic Stream. The HVCs arise in many complexes of clouds lying within ∼ 10 kpc of the disk and have 100-300 km s⁻¹ blueshifted radial velocities that indicate they will reach the disk within 10⁷⁻⁸ yr. Their mass inflow rate falls between 0.1–0.5 M_⊙ yr⁻¹, compared to the 1–2 M_⊙yr⁻¹ of star formation (Putman, Peek & Joung, 2012b). These clouds are all detectable in 21 cm emission, meaning that they occupy the tip of the column density distribution of CGM gas seen around other galaxies. The inflow rate inferred for ionized gas is much larger than for the classical HVCs, ≃ 0.8 − 1.4 M_⊙ yr⁻¹ (Lehner & Howk, 2011), more comparable to the Milky Way's star formation rate. The Magellanic Stream is estimated to contain around 2 × 10⁹ M_⊙ of gas in neutral and ionized form Bland-Hawthorn et al. (2007), Tepper-García, Bland-Hawthorn & Sutherland (2015), and could provide ∼ 5 M_⊙ yr⁻¹ of gas to the Milky Way disk as it accretes (Fox et al., 2014). Unfortunately, HVCs both above and below the radio-detection threshold are difficult to detect in external galaxies, despite intensive searches (Putman, Peek & Joung, 2012b), and satellites like the Magellanic Clouds and their Stream are not very common in L* galaxies. So we cannot generalize this result to mainstream galaxy populations.

Down-the-barrel spectroscopy provides complementary information on inflows. Using this technique on z ∼ 0.5 galaxies with Keck spectroscopy and HST imaging, Rubin et al. (2012) detected clear signs of inflow at 80−200 km s⁻¹ in star forming galaxies of log M_⋆ / M_⊙ = 9.5 − 10.5, inferring mass inflow rates of ≳ 0.2–3M_⊙ yr⁻¹. It seems likely that these estimates significantly undercount inflow, since inflowing (redshifted) gas is often obscured by outflows (blueshifted) or by emission from the galaxy's ISM (this problem is esepcially noticeable at higher redshift, Steidel et al. (2010)). Even if outflow is not present, the profiles are not sensitive to accretion from the lower half of the bimodal LLS metallicity distribution (Lehner et al., 2013), which could make up a large fraction of the available cold CGM gas. Recently, Zheng et al. (2017) reported the detection of enriched, accreting gas at the disk-halo interface of M33 via COS observations of SiIV absorption along several sightlines to bright O stars in the disk. Their kinematic modeling of the observed absorption features implies an accretion rate of 2.9 M_⊙ yr⁻¹. While these results provide evidence for accretion of cold, metal-enriched gas directly into galaxy disks, evidence for more metal-poor “cold-mode” accretion, and for gas entering further out in the disk (“on-ramp”, Figure 1), is still lacking (though see Bouché et al., 2013), as is empirical characterization of how accretion rates vary with galaxy mass.

7.3. The Preeminence of Outflows

By consensus, outflows are an accomplice if not the perpetrator in each of the problems outlined in Section 2. The existence of outflows is not in question: the large share of metals outside galaxies provides incontrovertible evidence for them (Section 6). COS-Halos found widespread O vi around star-forming galaxies — extended to ∼ 300 kpc by Johnson, Chen & Mulchaey (2015b) — but could not show that this ion becomes more prevalent with SFR. Even so, simulations found that robust outflows were necessary to produce the observed reservoir of metals e.g., Hummels et al. (2013), Ford et al. (2013), Suresh et al. (2015), such that the high metal ions provide a significant constraint on the time-integrated effects of outflows even if it does not show the effects of recent or ongoing outflows directly. After that, the important questions concern how they transport baryons, metals, momentum, energy, and angular momentum. There is empirical evidence and strong theoretical suggestions that the physical drivers and properties of galaxy winds — their velocity, mass loading, metal content, and likelihood of escape — depends on galaxy mass, circular velocity (v_circ), star formation rate, and metallicity. Many investigators pursue CGM observations in the hope that they can help to constrain these outflows and how they scale with galaxy properties.

Direct observational evidence for outflows is readily available at all redshifts (see Veilleux, Cecil & Bland-Hawthorn, 2005, for a review). In the nearby universe, large-scale complex multiphase outflows are seen in starbursts (e.g., M82) and from the Milky Way's central regions (Fox et al., 2015). Down-the-barrel spectroscopy of the Na i D in local starbursts (Martin, 2005) found that outflow velocities depend linearly on v_circ. Rubin et al. (2012) and Bordoloi et al. (2014a) characterized similar flows using Mg ii at z ∼ 1. At z > 2, where the FUV-band ions used at z ∼ 0 appear at visible wavelengths, Steidel et al. (2010) used down-the-barrel spectroscopy to detect nearly “ubiquitous” outflows in rapidly star-forming LBG galaxies, with no clear indications for redshifted inflow. While these results help constrain the mass loading and covering fraction of outflows, they do not show how far these winds propagate into the CGM. It may be that the bulk of the energy is transported out in the hot gas while the bulk of the mass leaves in the cold phase, but this is still an open question (Strickland & Heckman, 2009).

Absorbers on transverse sightlines can directly constrain the impact of winds on the CGM. Cross-correlations of Mg ii absorbers with the orientation of galaxies on the sky at z ≲ 1, from both samples of individual galaxies (Kacprzak et al., 2012, Mathes et al., 2014) and stacked spectroscopy (Bordoloi et al., 2011, Zhu & Ménard, 2013b) find that the strongest absorbers prefer the semi-minor axis of disk galaxies, as expected for biconical outflows emerging from the disk. The preference for the semi-minor axis disappears by ∼ 60−80 kpc, indicating that winds propagate at least that far, or merge into the general medium near that radius (e.g., the z = 2 example in Figure 3). Studies of outflow covering fractions at z ∼ 1 reinforce a picture of outflows being roughly biconical, with little surface area (∼ 5%) solely dedicated to inflow (Martin et al., 2012, Rubin et al., 2014). Another strong clue about outflows comes from examining the CGM of starburst and post-starburst galaxies. Using an SDSS-selected sample, Heckman & Borthakur (2016) found unusually strong H i and multphase ions at 100−200 kpc compared with the COS-Halos and COS-GASS samples of galaxies at lower SFRs. These studies collectively show that SFR is a factor in determining the content of the CGM, perhaps as far out as R_vir.

Down-the-barrel measurements tell us that outflows are ubiquitous, and sightline measurements tell us that they reach 100 kpc scales. Together these findings suggest that a large part of the CGM is made of outflows, and to examine one is to illuminate the other. The open questions concern not only the basic scaling of velocity and mass loading with galaxy v_circ — which has received much attention — but just as importantly the distribution of outflow temperatures, metallicities, and fate. These cannot (yet) be simulated from first principles but can be constrained by the combination of CGM and down-the-barrel observations. The former constrain the radial extent and the velocity fields of multiphase gas far from the disk, while the latter constrain the initial velocities, mass loading, and (possibly) metallicities.

A recent goal of models and simulations has been to discriminate between winds that are “momentum-driven” (Murray, Quataert & Thompson, 2005), which appear to improve the match of simulations to the galaxy mass-metallicity relation (Finlator & Davé, 2008) and the metal content of the IGM (Oppenheimer & Davé, 2006, Oppenheimer & Davé, 2008), and those that are “energy-driven” (Murray, Ménard & Thompson, 2011), which appear to better match the galaxy stellar mass function (Davé, Finlator & Oppenheimer, 2012a) and new COS data (Ford et al., 2016). A momentum-driven outflow has a velocity v_w ∝ v_circ⁻¹, while an energy-driven flow has much faster outflows for low-mass galaxies with v_w ∝ v_circ⁻²; with a fiducial wind speed of ∼ 100 km s⁻¹, an unimpeded flow reaches 100 kpc in only 1 Gyr, i.e., the scales on which metals are seen in the CGM (Section 6). Thus understanding the history of CGM metals and the velocities and mass flow rates of galactic flows go hand-in-hand. Real winds may depend less on the local potential well and more on the local star formation rate surface density (Kornei et al., 2012, Heckman et al., 2015). New hydrodynamic simulations of galaxies that resolve the multiphase ISM and explicitly include radiation pressure and thermal pressure (Hopkins, Quataert & Murray, 2012) support this picture. Like essentially every other simulation suite on the market, however, models with this feedback scheme have too little O vi in the CGM while retaining too many metals in stars (Muratov et al., 2015).

7.4. Following the Metals: The Role of Recycling

Inflow and outflow are necessary processes in galaxy and CGM evolution; can one become the other by the recycling of outflows into fresh accretion of ejected gas? We have already established that, at least at low-redshift, galaxies require a long-term source of fuel, and that their CGM gas and metals are massive and bound. Recycling is a natural consequence; this gas “should” reaccrete onto the galaxy if the cooling time is short. Indeed, the predominance of metal-enriched accretion is supported by essentially all cosmological simulations where the origins of gas joining the ISM has been tracked: significant fractions at gas accreting onto galaxies has previously been ISM gas — and often through multiple cycles (Ford et al., 2014, Christensen et al., 2016, Muratov et al., 2016), with the majority of star formation at late times fueled by recycled gas (Oppenheimer et al., 2010). Ford et al. (2014) found 60% of all star formation at z = 0 is powered by gas that was in the CGM a billion years before. This idea has the intriguing implication that a substantial fraction of all heavy elements on Earth once cycled through the Milky Way's halo at 100 kpc scales. The timescales are unclear: Christensen et al. (2016) find that half of outflow mass is recycled on timescale of 1 Gyr with a logarithmic tail, independent of halo mass, while Oppenheimer & Davé (2008) find that t_rec ∝ M_halo^−1/2 ∼ 10^{9 ± 0.5} yr, a timescale so short for massive galaxies that it is like not having an outflow at all, and so long for dwarfs that it essentially escapes forever.

Thus the idea of recycling is well-motivated, but the details are still murky. Is it a simple process in which gas launched at v < v_esc encounters hydrodynamic resistance and eventually succumbs to gravity to fall back into the galaxy as part of a large-scale halo fountain? Or is the CGM well-mixed but multi-phase, with metal-rich gas precipitating out of the hot halo and raining onto the galaxy (Voit et al., 2015a, Fraternali et al., 2015, Thompson et al., 2016)? Here too can metals help disentangle the ins and outs. Intriguingly, dense CGM gas (Lehner et al., 2013, Wotta et al., 2016; Section 6.2) is roughly equally divided between gas at a few percent solar (metal-poor IGM accretion) and 40% solar (recycling ejecta?).

While gas “accreting” from the IGM generally has (or is assumed to have) very low metallicity (Lehnert et al., 2013, Cooper et al., 2015, Glidden et al., 2016), cases with metallicity well below the IGM (Lyα forest) at the same redshift are rare (Fumagalli, O’Meara & Prochaska, 2011, Crighton, O’Meara & Murphy, 2016). That is, either pristine cosmic accretion entrains metal-enriched circumgalactic gas on its way into the galaxy (e.g., Fraternali et al., 2015), or that even at the highest redshifts where accretion is potentially observable, it is at least partially comprised by material that has previously been in the ISM, i.e., that recycled mode accretion is critical to galaxy evolution even at early cosmic times. Yet most formulations of the “bathtub model” assume that the accreting gas is pristine (e.g., Lu, Blanc & Benson, 2015, though see Davé, Finlator & Oppenheimer, 2012b). Entrainment is a commonly invoked phenomenon for galaxy outflows, where it refers to the wind fluid sweeping up ambient ISM and mixing it with the fresh supernova ejecta powering the outflow. (It is important to note that the metallicity of the outflowing material is necessarily higher than that of the ambient ISM, contrary to what is assumed in some popular simulation recipes, e.g., Vogelsberger et al. 2014.) Does “recycled accretion” behave in a simular way but in the opposite direction, with pristine inflows sweeping up metal-polluted CGM material on its way from the IGM to the ISM? Or do galaxy winds preferentially re-accrete, sweeping up more pristine cosmic accretion?

Taking all this evidence into account, we can see the outlines of an emerging picture of galaxy inflows, at least at low redshift. They arise in the massive reservoir of cold, metal enriched gas bound to a galaxy's potential well, and enter the disk in HVC-like clouds but also in smooth flows of ionized gas. There may be a metal-poor component that comes more directly from the IGM without spending much time in the CGM, or otherwise acquiring metals. All these aspects of the CGM — cold, bound, metal enriched, and accreting — align better with the phenomenon of “recycled accretion” better than the bimodal “hot / cold” accretion. Recycled accretion arises from the ejection of metal-enriched galactic winds that lack the energy to escape the halo entirely, or which encounter the CGM itself and lose energy to radiation from shocks and then eventually cool and re-enter the galaxy. It may be that “recycling”, rather than “accretion and feedback” is the more accurate way of viewing how galaxies acquire their gas.