3. EVIDENCE FOR A RELATIONSHIP BETWEEN THE SFR AND THE GAS INFALL RATE

3.1. The gas-consumption timescale

The time-scale to consume the gas is too short for the observed stellar ages, therefore, a continuous gas supply is needed to explain why all types of galaxies have been forming stars for extended periods of time. This is an old idea (e.g., Roberts, 1963; Kennicutt, 1983) that has been reformulated in various ways along the years. A updated account is given next.

Figure 2. Observed SFR versus redshift for typical galaxies forming stars at present. The galaxies are divided into three bins according to the present stellar mass (see the inset). Galaxies of all masses have been forming stars over the whole history of the Universe. Figure adapted from Guglielmo et al. (2015, their Fig. 6).

Except for the most massive ones, galaxies have been forming star during the whole life-time of the universe. Figure 2 shows the observed SFR versus redshift for typical disk galaxies forming stars at present. The galaxies have been divided into three bins according to the present stellar mass. The figure shows how their SFR has been declining slightly with time but, overall, galaxies of all masses maintain a significant level of SF along the Hubble time (see also Heavens et al., 2004). If no gas supply is provided, the SF shuts off exponentially with a time-scale of

(10)

where we have plugged into Eq. (2) the values 2 Gyr, 2, and 0.3 for τ_g, η and R, respectively (Sect. 2). Since the SF has been active for much longer, a supply of external gas must have been feeding the SF process in all galaxies.

3.2. Relationship between stellar mass, SFR, and gas metallicity

In two independent papers, Mannucci et al. (2010) and Lara-López et al. (2010) found that the scatter in the well-known mass-metallicity relation correlates with the SFR. Such correlation had been suggested in previous studies (Ellison et al., 2008; Peeples et al., 2009; López-Sánchez, 2010). The fact that galaxies with higher SFR show lower metallicity at a given stellar mass is called fundamental metallicity relation (FMR). Figure 3 shows a recent account of the FMR by Salim et al. (2014). Galaxies are separated into two plots according to their SFR, and the difference is clear: high SFR objects (panel A) have lower metallicity than low SFR objects (panel B).

Figure 3. Gas-phase metallicity versus stellar mass for galaxies with extreme SFR. From a sample of ∼ 250,000 SDSS objects, the dots in the upper panel (A) show the galaxies with the highest SFRs. The lower panel (B) contains those galaxies with the lowest SFRs. The colored lines represent the median of the distribution of points for a given stellar mass. The green shaded region is the same in (A) and (B) and gives the 90 percentile range of the full sample, with the white line representing the median. Note the systematically lower metallicity for the sample of higher SFR. Figure adapted from Salim et al. (2014).

Neither metal-rich outflows nor the variation of the SF efficiency with M_⁎ explain the FMR. However, the stochastic feeding of the SF process with external metal-poor gas provides a natural explanation. The advent of external gas does not change M_⁎, but it decreases the mean metallicity of the star-forming gas while simultaneously triggering SF. As time goes on, the SF consumes gas and increases its metallicity, until new metal-poor gas arrives and the cycle re-starts. The process is illustrated in Fig. 4. It contains the temporal variation of the gas mass and metallicity predicted by the toy model in Sect. 1, assuming the gas accretion events to be stochastic. The figure shows how the arrival of gas at a particular time increases the gas mass and thus the SFR (Eq. [3]). This gas accretion event comes with a drop in metallicity. The gas is consumed by the SF, that rises the metallicity in a process that in the long run yields the stationary-state value given in Eq. (9). If a collection of galaxies with similar stellar masses are observed at different phases of the cycle triggered by a gas accretion event, they will show a dispersion in metallicity which anti-correlates with the instantaneous SFR, i.e., they will show the FMR. This explanation was already advanced by Mannucci et al. (2010) and it is generally accepted today.

Figure 4. Variation with time of gas mass (a) and metallicity (b) in a star-forming system where gas clumps are accreted randomly, i.e., at random times and with random masses. Time units are normalized to τin, the characteristic time-scale for the exponential fall-off of the gas content. The gas of mass and metallicity are normalized to their stationary-state value, indicated by horizontal dashed lines in the figures.

The FMR has received considerable attention in the literature ¹, both from an observational point of view, and from the point of view of its interpretation. Sometimes the claims seem to be contradictory, although the community is reaching a consensus in the sense that (1) the FMR is not an observational artifact, (2) it changes with redshift so that all metallicities decrease with increasing redshift, and (3) it is produced by metal-poor gas accretion triggering SF. The next paragraphs summarize the recent observational and theoretical work on the subject.

Observations of the FMR. The correlation between SFR and metallicity is weaker if systematic errors are taken into account (Telford et al., 2016). The correlation may disappear depending on the strong-line ratio used to estimate metallicities (Kashino et al., 2016). The FMR remains even if different metallicity and SFR indicators are used (Salim et al., 2014; Andrews & Martini, 2013).

The FMR disappears when using single H II regions rather than galaxy integrated parameters (Sánchez et al., 2013). There is no FMR in the local star-forming galaxies analyzed by Izotov et al. (2014), whereas it is present in the local sample discussed by Wu et al. (2016). Arguments against the existence of a FMR in star-forming galaxies with redshift smaller than one are presented by Izotov et al. (2015). There is no obvious FMR for galaxies with redshifts between 1 and 2, according to Divoy et al. (2014). There is not relationship at redshift 0.8 (de los Reyes et al., 2015).

At redshift around 1.4, the deviation from the MZR depends on the SFR, so that galaxies with higher SFR show lower metallicities at a given M_⁎ (Yabe et al., 2012). The FMR is still in place at redshift around 0.9, but the metallicities are systematically lower given M_⁎ and SFR (Ly et al., 2014; Ly et al., 2015). Galaxies with younger and more vigorous star formation tend to be more metal poor at a given M_⁎ (redshift between 0.3 and 0.9; Amorín et al., 2014). The FMR is in place at z ≥ 2 (Cullen et al., 2014). Galaxies at redshift 2 show evidence that the SFR is still a second parameter in the MZR, and are consistent with a non-evolving FMR (Maier et al., 2014). The FMR is in place at redshift 1.6, but it has evolved with respect to the FMR in the local universe so that metallicities are smaller (Zahid et al., 2014). At redshift 3.4, the metallicity generally anti-correlates with the distribution of SFR and with the gas surface density, although the relation differs from the FMR in the local universe (Troncoso et al., 2014). The evolution of the FMR previously reported in the literature may be an artifact introduced by the use of the different metallicity indicators at different redshifts (Cullen et al., 2014). There is a FMR at redshift 0.7 that seems to agree with the local one (Maier et al., 2015). There is no correlation at redshift 2.3 (Sanders et al., 2015). The FMR evolves with redshift (Brown et al., 2016). There is no significant dependence of the metallicity on SFR at fixed redshift and M_⁎ (objects with redshifts between 0.6 and 2.7; Wuyts et al., 2016).

There is also a more fundamental FMR where the SFR is replaced with the gas mass (Bothwell et al., 2013). The scatter of the FMR is reduced if H I mass is used instead of SFR (Jimmy et al., 2015). The central role assigned to the gas mass at the sacrifice of the SFR is also defended by Bothwell et al. (2016a) and Bothwell et al. (2016b). Moreover, Lian et al. (2015) claim that stellar age, rather than SFR or gas mass, is the third parameter in the FMR.

There is a correlation between the metallicity gradient along the radial distance in a galaxy and the SFR, in the sense that galaxies with high SFR tend to show flatter gradient (Stott et al., 2014).

Interpretations of the FMR. Most of the available explanations are based on simple analytical models very much in the spirit of the one described in Sect. 1. For example, Lilly et al. (2013) present a model galaxy whose properties self-regulate due to the short gas depletion time-scale. The model galaxy is near the stationary state, but the gas reservoir available to form stars is allowed to change in time. This drives the system out of the stationary state and provides a dependence of the metallicity on the SFR and mass gas. (The metallicity does not depend on the SFR in the stationary state; see Eq. [9]). The work by Lilly et al. successfully reproduce the FMR, allowing both w and τ_g to vary with stellar mass. It reproduces the overall drop of metallicity with increasing redshift by steadily increasing the gas infall rate. Other works with this type of interpretation are those by Davé et al. (2011); Brisbin & Harwit (2012); Dayal et al. (2013); Forbes et al. (2014b); Pipino et al. (2014); and Harwit & Brisbin (2015). Mergers are also able to reproduce the FMR according to Grønnow et al. (2015).

Dekel & Mandelker (2014) use one of these simple toy models to study the redshift dependence of the FMR, finding problems to reproduce some of the observational constrains, in particular, the ratio SFR/M_g. The need to go beyond simple models because they do not reproduce the observed variation with redshift of SFR/M_g is also argued by Peng & Maiolino (2014).

Yates et al. (2012) use thousands of galaxies from dark-matter numerical simulations to interpret the FMR. Baryons that follow the dark-matter are added, generating a non-stationary clumpy gas accretion that drives the evolution of the model galaxies. The numerical simulation reproduces the main observational trends, including an apparent turnover of the mass-metallicity relationship at very high M_⁎. The temporal evolution of the gas mass and metallicity of individual galaxies is qualitatively similar to the variations displayed in Fig. 4.

Romeo Velonà et al. (2013) present SPH-cosmological simulations of hundreds of galaxies. Surprisingly, more active galaxies in terms of SFR are also metal-richer (see their Fig. 12). The reason of this contradictory result is not properly understood.

De Rossi et al. (2015) employ hydrodynamical zoom-in cosmological simulations of 500 galaxies to study the scaling relations. The model galaxies show the trends corresponding to the observed FMR. They also find that satellite galaxies have higher metallicity for the same stellar mass, as it is indeed observed (Pasquali et al., 2012).

Kacprzak et al. (2016) find galaxies at redshift around 2 following the FMR. They show that the gas masses and metallicities required to reproduce the observed FMR are consistent with cold-accretion predictions obtained from their hydrodynamical simulations.

3.3. Relationship between lopsidedness and metallicity

Surprisingly, the extremely metal poor (XMP) galaxies of the local universe turn out to show a particular morphology consisting of a bright head and a faint tail, which is commonly referred to as cometary or tadpole. This correspondence between low metallicity and shape was first noted by Papaderos et al. (2008), and then it has been confirmed in other studies (e.g., Morales-Luis et al., 2011; Filho et al., 2013; Sánchez Almeida et al., 2016). These morphologies represent 80 % of the objects in the XMP catalog used by Filho et al. (2013). The tadpole morphology is not unusual at high redshift, where galaxies tend to be clumpy and elongated (Elmegreen et al., 2005), however it is rather uncommon in the local universe where XMPs reside. For reference, only 0.2 % of the star-forming galaxies in the Kiso survey are cometary (Elmegreen et al., 2012). Figure 5 displays several of these XMP galaxies with the characteristic morphology.

Figure 5. Typical set of galaxies selected only because they are XMP. Surprisingly, they tend to show cometary or tadpole morphology, with a bright blue head and a faint redder tail. The images are composite color images from SDSS broad-band filters, therefore, they trace stellar light. Adapted from Fig. 5 in Morales-Luis et al. (2011).

Even though XMP galaxies are a very particular type of galaxy, the morphology-metallicity relation that they exhibit is only the extreme case of a common behavior followed by many star-forming galaxies. Reichard et al. (2009) quantify the lopsidedness of 25,000 star-forming galaxies from SDSS using the amplitude of the m = 1 azimuthal Fourier mode. At a fixed mass, the more metal-poor galaxies turn out to be more lopsided, extending the morphology-metallicity relation to the full population of star-forming galaxies.

This property of the XMPs and the other galaxies can be naturally understood within the gas accretion triggering scenario. The actual characteristics of the starbursts induced by gas accretion are far from being properly understood and modeled (e.g., Verbeke et al., 2014; Casuso et al., 2006; see also Sect. 4). However, a few general trends seem to be clear. The accreted gas is metal-poor (e.g., Dekel et al., 2009; van de Voort & Schaye, 2012), and it induces off-center giant star-forming clumps that gradually migrate toward the disk centers (Ceverino et al., 2010; Mandelker et al., 2014). The giant star-forming clumps may be born in-situ or ex-situ. In the first case, the accreted gas builds up the gas reservoir in the disk to a point where disk instabilities set in and trigger SF. In the second case, already formed clumps are incorporated into the disk. They may come with stars and dark matter, and thus, they are often indistinguishable from gas-rich minor mergers (Mandelker et al., 2014). In any case, a significant part of the SF in the disks occurs in these giant clumps. As a result of the whole process, the gas accretion produces bright off-center starbursts increasing the lopsidedness of the host disk. This increase and the decrease of metallicity come hand-to-hand together, giving rise to a relation between morphology and metallicity qualitatively similar to the observed one.

We note, however, that the same trend can also be reproduced by gas-rich metal-poor mergers (e.g., Kazantzidis et al., 2009; Pawlik et al., 2016). As we pointed out above, gas-rich minor mergers and gas accretion events are often impossible to distinguish, both observationally and from the point of view of the numerical simulations. On the one hand, it is unclear how to define galaxy at the low-mass end of the galaxy mass function. If the presence of stars is essential (see Forbes & Kroupa, 2011), whether a gas dominated system is or is not a galaxy ultimately depends on the sensitivity of the observation (e.g., Cannon et al., 2014; Serra et al., 2015; Janowiecki et al., 2015). On the other hand, the presence or absence of stars in a particular dark-matter halo of a numerical simulation depends on details of the assumed sub-grid physics, which may or may not be adequate to describe the formation of stellar systems in objects with sizes and masses at the resolution of the simulation.

This remarkable association between SFR and lopsidedness has been observed in the H I morphology too. Lelli et al. (2014) find that dwarfs with active SF have more asymmetric outer H I envelopes than typical irregulars. Moreover, galaxies hosting an old burst (≥ 100 Myr) have more symmetric H I morphology than those with a young one (≤ 100 Myr).

3.4. Metallicity drops in starbursts of local star-forming galaxies

According to the conventional wisdom, the gas of the local gas-rich dwarf galaxies has uniform metallicity (Kobulnicky & Skillman, 1996; Croxall et al., 2009; Pilyugin et al., 2015). However, there is mounting evidence that some particular objects do show metallicity inhomogeneities (Papaderos et al., 2006; Izotov et al., 2009; Werk et al., 2010; Levesque et al., 2011; Izotov et al., 2012; Haurberg et al., 2013; Sánchez Almeida et al., 2013; Thöne et al., 2014; Sánchez Almeida et al., 2014b; Richards et al., 2014). They are often associated with star-forming regions, so that a drop in metallicity occurs in regions of intense SF.

Sánchez Almeida et al. (2015) carried out a systematic study of the variation of gas metallicity along the major axis of a representative sample of XMP galaxies. Metallicities were inferred using HCM (Pérez-Montero, 2014), a code that compares the observed optical emission lines with photoionization models and which provides metallicity measurements in agreement with direct-method within 0.07 dex. Figure 6 contains the result for one of the galaxies. It shows a clear drop in metallicity at the peak surface SFR. The pattern is the same in 9 out of the 10 studied galaxies. The XMP star-forming clumps are immersed in a host galaxy which is several times more metal-rich. Figure 7 summarizes these results. Independent observations proof that the XMP galaxies rotate, and that the star-forming clumps of low metallicity are dynamically decoupled from the underlying disk (Olmo-García et al., 2016).

Figure 6. Left: SDSS image of the XMP galaxy J1132+57, with the red bar indicating the position of the slit during observation. The arrows indicate north and east, and the scale on the top left corner corresponds to 10 arcsec. Right: variation of surface SFR (blue solid line) and oxygen abundance (red symbols with error bars) along the slit. Note the drop in abundance associated with the peak SFR. Figure adapted from Sánchez Almeida et al. (2015).

Figure 7. Summary plot with the oxygen abundance of the starburst (square symbols) and the host galaxy (asterisks) for the XMP galaxies studied by Sánchez Almeida et al. (2015). They are represented versus the surface SFR inferred from Hα. The star-forming clumps are 0.5 dex metal-poorer than the host, and have a SFR between 10 and 20 times larger. The axis on top gives the gas surface density. The lines show the chemical evolution of a clump at the position of the red bullet depending on whether it evolves as a closed-box (red line) or as an open box (blue line). Taken from Sánchez Almeida et al. (2015).

The existence of localized metallicity drops suggests a recent metal-poor gas accretion episode. The time-scale for the azimuthal mixing of the gas in turbulent disk is short, of the order of a fraction of the rotational period (e.g. de Avillez & Mac Low 2002; Yang & Krumholz 2012; Petit et al. 2015), equivalent to a few hundred Myr. Therefore, the metal-poor gas forming stars must have arrived to the disk recently, as naively expected for SF episodes driven by external metal-poor gas accretion. As we discussed in Sect. 3.3, the triggering of SF feeding from external gas is a complex process not properly understood yet (see also Sect. 4). There is a significant degree of gas mixing in the CGM (Sect. 4) and the naive interpretation may not be correct. However, the cosmological numerical simulations of galaxies analyzed by Ceverino et al. (2016) are reassuring. The model galaxies produce off-center star-forming clumps with a metallicity lower than the metallicity of their immediate surroundings. Figure 8, taken from Ceverino et al. (2016), shows metallicity versus surface SFR for a number of clump intra-clump pairs. Each pair is joined by a dotted line. They follow a clear pattern where the point of lower metallicity coincides with the point of larger SFR. Qualitatively, the figure resembles the behavior of the star-forming clumps in XMPs (see Fig. 7).

Figure 8. Metallicity [12 + log(O/H)] versus surface SFR (ΣSFR) for a number of clumps and their nearby intra-clump medium in the cosmological numerical simulations of galaxies by Ceverino et al. (2016). Each pair is joined by a dotted line. In all but one case (the one shown in black), the region of large SFR (the clump) has lower metallicity than the nearby region of small SFR (the intra-clump medium). The small black dots represent 100 randomly chosen apertures in one of the model galaxies. Adapted from Fig. 3 in Ceverino et al. (2016).

3.5. The traditional G-dwarf problem

This observation is included here both for historical reasons, and because the accepted interpretation is easy to understand in terms of the toy model described in Sect. 1. The so-called G-dwarf problem was the first clear indication that an external metal-poor gas supply was needed to explain the observed properties a stellar population.

Figure 9. Fraction of stars in each 0.1 dex metallicity bin. [Fe/H] denotes the Fe abundance referred to the solar metallicity in a decimal logarithm scale. The dashed and solid lines represent observed uncorrected and corrected data, respectively. The dotted line is the distribution predicted by a closed-box chemical evolution model, and it largely deviates from the observed one. The observed distribution, from Woolf & West (2012), corresponds to M dwarf stars in the solar neighborhood, but is very similar to the distribution of G dwarfs discussed in the text (see, e.g., Rocha-Pinto & Maciel 1996).

If a system of pure metal-poor gas evolves as a closed box, each new generation of stars must be less numerous and more metal-rich than the previous one. Therefore, in such a system, the number of stars is expected to decrease with increasing metallicity. However, the distribution of metallicities of the G dwarf stars in the solar neighborhood does not show the fall-off expected in a closed box. There is a deficit of sub-solar metallicity G dwarf stars in the solar neighborhood (van den Bergh 1962; Schmidt 1963; Lynden-Bell 1975) – see Fig. 9. This problem has deserved careful attention in the literature, with solutions going from variations of the IMF (Carigi 1996, Martinelli & Matteucci 2000) to inhomogeneous chemical evolution and star formation (Malinie et al. 1993). Among them, a continuous metal-poor gas inflow sustaining the formation of the G stars seems be the preferred mechanism (Pagel 2001; Edmunds 2005). The explanation was first proposed by Larson (1972), who discovered that the SF maintained by constant metal-poor gas accretion reaches a constant value set only by the stellar yield (see Eq. [9]), which implies a value around the solar metallicity. In the context of this explanation, the apparent deficit of sub-solar metallicity G dwarfs is actually an excess of solar metallicity G dwarf stars formed over time out of an ISM always near equilibrium at approximately the solar metallicity.

The G dwarf problem has also been observed in K dwarfs (e.g. Casuso & Beckman 2004) and in M dwarfs (e.g., Woolf & West 2012, and it exists in other galaxies as well (e.g., Worthey et al. 1996. Current chemical evolution models resort to metal-poor gas inflow to reproduce the spatial distribution of stellar metallicities observed in the disk of spirals (e.g., Chiappini et al. 2001; Magrini et al. 2010; Mollá et al. 2016; Pezzulli & Fraternali 2016). Such gas inflow is needed for the same reasons invoked to solve the G dwarf problem.

3.6. Existence of a minimum metallicity for the star-forming gas

XMPs are defined as galaxies where the gas that produces stars has a metallicity smaller than 10 % of the solar metallicity. They turn out to be quite rare. Systematic searches, such as that carried by Sánchez Almeida et al. (2016), render a few hundred objects in catalogs containing of the order of one million galaxies (XMP represent ≪ 0.1 % of the known galaxies). Interestingly, their metallicity, and therefore the metallicity of all local galaxies, seems to have a lower limit at around 2% of the the solar metallicity. Figure 10 displays the distribution of metallicities found by Sánchez Almeida et al. (2016) in the search for XMP candidates in SDSS. It has a sharp cut-off at low metallicity, with no galaxy with 12 + log(O/H) ≤ 7.0. The current record-breaking object is AGC 198691, with a metallicity around 2.1 % times the solar metallicity (Hirschauer et al. 2016). It is also included in Fig. 10 for reference.

Figure 10. Distribution of oxygen abundance [12 + log(O/H)] for all the objects found in the search for low-metallicity galaxies by Sánchez Almeida et al. (2016). The solar metallicity is set at 12 + log(O/H)⊙ = 8.69, therefore, the value 12 + log(O/H) = 7.69 separates the XMPs (the thick red line) and the failed XMP candidates (the thin blue line). There is a clear drop in the distribution towards low metallicity, with no object metal-poorer than 2 % of the solar metallicity. AGC 198691 is the star-forming galaxy with the lowest metallicity known (Hirschauer et al. 2016), and the lower limit it sets is marked with an arrow. Figure adapted from Sánchez Almeida et al. (2016).

The existence of this metallicity threshold is not an artifact. Observers have been looking for record-breaking galaxies during the last 45 years, after the discovery of the prototypical XMP galaxy I Zw 18 (Sargent & Searle 1970). These efforts led to enlarging the number of known XMPs (Terlevich et al. 1991; Guseva et al. 2009; Guseva et al. 2015; Morales-Luis et al. 2011; Izotov et al. 2012; Sánchez Almeida et al. 2016), but the lower limit metallicity set by I Zw 18 remains almost unchanged (the metallicity of I Zw 18 is about 3% times the solar metallicity; see, e.g., Thuan & Izotov 2005).

Several explanations have been put forward to account for the existence of a minimum metallicity in the gas that forms stars. Kunth & Sargent (1986) point out the self-enrichment of the H II region used for measuring. However, the time-scale for the SN ejecta to cool down and mix is of the order of several hundred Myr (e.g., Legrand et al. 2001) and, thus, longer than the age of the H II regions. This fact eliminates the possibility of self-contamination. Self-enrichment by massive star winds seems to be negligible too (Kröger et al. 2006). Kunth & Lebouteiller (2011) suggest the pre-enrichment of the proto-galactic clouds, but it is unclear why there should be a minimum metallicity for such clouds, except perhaps the value set by the metal contamination produced by Pop III stars. Pop III star contamination has been suggested too (Audouze & Silk 1995; Thuan & Izotov 2005), but the expected level of metal enrichment is of the order of 10⁻⁴ times the solar metallicity (Bromm & Larson 2004; Muratov et al. 2013), and so much lower than the observed threshold.

Alternatively, if the SF is feeding from gas of the IGM, there is a minimum gas-phase metallicity set by the metallicity of the local IGM. This possibility provides a natural explanation for the long-lasting puzzle. Numerical simulations predict the local cosmic web gas to have a metallicity of the order of 1 % times the solar value (e.g., van de Voort & Schaye 2012, Rahmati et al. 2016). The metal content of the IGM has been rising over time contaminated by galactic winds, and now it happens to be at the level of the observed metallicity threshold.

Other independent observations also support the existence of a minimum metallicity in the CGM of galaxies at the level of 1 % times the solar metallicity. Lehner et al. (2013) measure the distribution of metallicity of Lyα absorbing clouds around galaxies with redshift up to one. The clouds are observed in absorption against background QSOs. The distribution of metallicity turns out to be bimodal with typical values around 2.5 % solar and 50 % solar. The high-metallicity branch is expected to represent galaxy outflows, whereas the low-metallicity branch corresponds to inflows. The lowest measured metallicity turns out to be 1 % times the solar metallicity. In a completely independent type of work, Filho et al. (2013) studied the H I gas around XMP galaxies. The XMPs happen to be extremely gas-rich, with gas fractions typically in excess of 10. Assuming that all the metals produced by the observed stellar populations have been diluted in their huge H I reservoirs, the metallicity of the H I is again around a few percent of the solar value. The same conclusion has been recently reached by Thuan et al. (2016). Sometimes the metallicity of the H I gas can be measured directly using UV lines in absorption against the stellar light of the galaxy. In the case of I Zw 18, Lebouteiller et al. (2013) find that H I region abundances are also around 1 % of the solar value.

3.7. Origin of α-enhanced gas forming stars in local galaxies

The gas forming stars in some of the local dwarf galaxies has elemental abundances which do not scale with the solar abundances. The gas is α−enhanced, using the terminology employed when studying stellar populations in massive galaxies and in the MW halo (e.g., Adibekyan et al. 2011; Vazdekis et al. 2015). The star-forming gas often shows log(N/O) ≃ −1.5 (e.g., James et al. 2015; Vincenzo et al. 2016) which is much smaller than the solar value (of the order of -0.86; Asplund et al. 2009). Stars and gas with α−enhanced composition are expected to be formed from the ejecta of young stellar populations (e.g., Izotov & Thuan 1999), so that low mass stars have not had the time to explode as SN, and so, elements like Fe are underrepresented compared to the solar composition. N is one of these elements, even though N is also produced in intermediate-mass stars (e.g., Henry et al. 2000). Figure 11 shows the time evolution of N/O in several different model starbursts with very different SF efficiencies (i.e., with different τ_g in the parlance used in Sect. 1). Vincenzo et al. (2016) compute them to model the relationship between N/O and O/H observed in the galaxies of the local universe. Independently of the SF efficiency, when a starburst is 2 Gyr old it has already reached N/O ≃ N/O_⊙ (see Fig. 11). Something equivalent is shown by Köppen & Hensler (2005) when modeling the evolution of systems that accrete large amounts of metal-free gas. After a transient phase that lasts around 2 Gyr, the system returns to the original N/O_⊙. The fact that the gas is α-enhanced is consistent with the thorough study on the heavy element abundances in local star-forming galaxies carried out by Izotov & Thuan (1999). The α-elements Ne, Si, S and Ar, produced by SN explosions of massive (> 10 M_⊙) stars that also produce O, show the same α-element/O ratio as the Sun. However, O turns out to be overabundant with respect to N, C and Fe by factors of around three, which are typical of MW halo stars.

Figure 11. Time evolution of N/O for chemical evolution models aimed at reproducing the relationship between N/O and O/H exhibited by the galaxies of the local universe. The different curves correspond to different τg going from 2 Gyr (solid magenta line) to 0.3 Gyr (red dashed line). The figure includes the N/O found in the Sun and the characteristic value observed in metal-poor galaxies of the local universe. By the time that the burst is 2 Gyr old, N/O has already reached the solar value in all models. Adapted from Vincenzo et al. (2016).

It is well known that all galaxies, including the local dwarf galaxies, have their mass dominated by old stars with ages extending all the way back to the origin of the universe (e.g., Heavens et al. 2004; Sánchez Almeida et al. 2012, see also Fig. 2). The fact that N/O is low in the star-forming gas of some local dwarfs implies that their evolved stellar populations are not the source of the metals. If the metals are not produced by the observed stellar populations, where do they come from? They likely come together with the accreted gas from the IGM. The IGM gas is expected to have α-enhanced chemical composition since the metals its contains were produced by dwarfs galaxies in the early universe when the stellar populations were young (e.g., Adelberger et al. 2005; van de Voort & Schaye 22012; Yates et al. 2013).

The presence of α-enhanced gas forming stars at high redshift seems to be common. Steidel et al. (2016) argue that the unusually high excitation of the gas forming stars in galaxies at redshift around 2 is due to their extreme α−enhanced composition, with O/Fe ∼ 5 times the solar ratio. Such extreme conditions yield Fe-poor stars in an ISM of moderate-high metallicity as traced by O. Similar physical conditions in the local universe may be responsible for the presence of high excitation narrow He II lines in the spectra of metal-poor galaxies of the local universe (Shirazi & Brinchmann 2012).

3.8. The metallicity of the quiescent BCD galaxies

The blue-compact dwarf (BCD) galaxies are going through an intense starburst phase that cannot be maintained for long. Consequently, there must be many dormant galaxies in a pre or a post BCD phase. They are called quiescent BCDs (QBCD). BCDs possess one or a few bright starburst regions on a low-surface host. Masking out the high-surface knots, Amorín et al. (2007, 2009) characterized the photometric properties of the host galaxy, which likely corresponds to the underlying QBCD. Using these properties as reference, Sánchez Almeida et al. (2008) found out that the SDSS catalog contains as many as 30 QBCDs per BCD. BCDs and QBCDs seem to be drawn from a single galaxy population that metamorphose with a cycle where the quiescent phase lasts 30 times longer than the star-forming phase. However, this interpretation presents a difficulty. The gas metallicity of the QBCDs is systematically higher than the metallicity of the BCDs, which cannot happen if the transformation between QBCD and BCD occurs through closed-box evolution, where the precursor always has lower metallicity than the offspring. The problem naturally goes away if the BCD phase is triggered by the accretion of external metal-poor gas that feeds the observed SF episode. The external driving of the BCD phase also explains why the stellar metallicity of BCDs and QBCDs agrees, even though their gas metallicity differs (Sánchez Almeida et al. 2009). The stellar populations of BCDs and QBCDs are statistically the same because only a small fraction of the galaxy stellar mass is built up in each new burst.

3.9. Direct measurement of inflows in star-forming galaxies

To the best of our knowledge, there is no direct measurement of metal-poor gas inflows in the CGM of galaxies. Finding whether a particular Doppler shift corresponds to inflows or outflows is tricky because the same signal is provided by inflows or outflows depending on whether the source is in the foreground or the background. Fortunately, the sense of motion can be disambiguated when the gas is absorbing or emitting against the spectrum of the galaxy. In this cases the gas is in the foreground, which breaks the degeneracy.

Thus, gas inflows have been detected in absorption against stellar spectra (e.g., Rubin & MaNGA Team 2016), but they are probably associated with metal rich gas returning to the galaxy after being ejected by a starburst or an AGN (e.g., Sancisi et al. 2008; Marasco et al. 2012; Fraternali 2014). Neutral gas has been studied in absorption in several metal-poor objects (Aloisi et al. 2003; Cannon et al. 2005; Lebouteiller et al. 2009; 2013). However, the relative velocity between the gas absorption and the systemic velocity of the galaxy is too small to tell whether the gas is falling in or going out. If anything, the gas seems to be flowing out with a mild velocity between 10 and 20 km s⁻¹ (Lebouteiller et al. 2013). The origin of this putative metal-poor outflow is puzzling.

Recent observations by Fathivavsari et al. (2016) show signs of infall in DLAs eclipsing QSOs, i.e., dense H I gas so close in redshift to the QSO that it blocks the Lyα emission of the source. Their metallicities are relatively low, and most of them have redshifts larger than that of their background QSO, strengthening the idea that they could be associated with some low metallicity infalling material accreting onto the QSO host galaxies.

Even though the works by Kacprzak et al. (2012, 2015) do not have kinematical information, they are very suggestive of gas inflows in star-forming galaxies. They study the azimuthal distribution of Mg II and O VI absorbers around the galaxy responsible for the absorption. These absorption systems trace gas in the CGM and the IGM, depending on the distance to the galaxy. Kacprzak et al. (2012, 2015) find that the absorption preferentially occurs along the directions pointed out by the minor and the major axes of the central galaxy. The absorbers aligned with the major axis are expected to show gas inflows, with the absorbers in the direction of the minor axis corresponding to gas outflows. In order to secure the whole scenario, one would need to have measurements of the velocity and metallicity of the Mg II systems. Unfortunately, they are not available. However, two independent measurements support the above interpretation. Firstly, outflows prefer the direction perpendicular to the plane of the galaxy because Mg II outflows are faster in face-on galaxies (Bordoloi et al. 2014, Rubin et al. 2014). Secondly, the metallicity of Lyman limit systems (i.e., gas clumps of moderate H I column density) is observed to be bimodal, with one peak at low metallicity and the other at high metallicity (Lehner et al. 2013). This bimodality is to be expected if the observed absorption is produced by metal-poor inflows and metal-rich outflows.

¹ The discovery papers have about 600 citations in the ADS. Back.