A REDLINE STARBURST: CO(2–1) OBSERVATIONS OF AN EDDINGTON-LIMITED GALAXY REVEAL STAR FORMATION AT ITS MOST EXTREME

The source does not appear to be dominated by an AGN. Although SDSS J1506+54 has the most luminous [O iii] λ5007 and [Ne v] λ3426 emission of the DS12 sample (log(L_[O iii]/erg s^-1) = 42.1 and log(L_[Ne v]/erg s^-1) = 41.1), only four counts were detected with Chandra in the 2–10 keV X-ray band, corresponding to a luminosity of $L_X=2.9^{+4.4}_{-2.9}\times 10^{42}$ erg s⁻¹. Assuming typical ratios between X-ray, [O iii] λ5007, and [Ne v] λ3426 luminosities, and accounting for possible small ∼30% dust attenuation of the narrow-line region based on the reddening derived from our fit to the SED (DS12), the Chandra limits imply an X-ray absorption factor of ≈10–20 for a buried AGN (DS12; Heckman et al. 2005; Gilli et al. 2010). However, as DS12 argue, even if the X-ray attenuation factor was as high as 100, the AGN would contribute less than half of the observed 22 μm luminosity (Diamond-Stanic et al. 2009; Gandhi et al. 2009). Thus, star formation is likely to be the dominant power source of the bolometric emission of SDSS J1506+54.

The measured $L_{\rm IR}/L^{\prime }_{\rm CO}\approx 1500$ is one of the highest measured for any star-forming galaxy (Figure 3). Even if we accounted for an AGN contribution of 25% to our lowest estimate of L_IR, SDSS J1506+54 is at the extreme tail of the $L_{\rm IR}/L^{\prime }_{\rm CO}$ distribution, with $L_{\rm IR}/L^{\prime }_{\rm CO}\approx 550$ in this case. There are a handful of other star-forming galaxies with similar $L_{\rm IR}/L^{\prime }_{\rm CO}$ properties; for example, the z = 0.575 hyper-LIRG IRAS F00235+1024 (Combes et al. 2011) and the z = 0.633 ULIRG IRAS F10398+3247 (Combes et al. 2012) both have CO line luminosities measured in the 2–1 transition, and are also at the extreme end of the $L_{\rm IR}/L^{\prime }_{\rm CO}$ distribution (Figure 3). Several galaxies in the sample of Combes et al. (2012) have $L^{\prime }_{\rm CO}$ upper limits implying extreme $L_{\rm IR}/L^{\prime }_{\rm CO}$ and of these, two are obviously hosting AGN, as evident from their optical spectra.

Zoom In Zoom Out Reset image size

Like SDSS J1506+54, F00235+1024 is not dominated by an AGN, although it is thought to be a non-negligible (≈30%) AGN contribution to its total infrared emission (Verma et al. 2002; Farrah et al. 2002). F10398+3247 could also contain a deeply buried AGN, as it has a deep silicate absorption feature (τ_Si > 3; Dartois & Muñoz-Caro 2007), often associated with nuclei blanketed by a dense screen of carbonaceous and silicate grains. The key difference between these galaxies and SDSS J1506+54 is in optical morphology: F00235+1024 and F10398+3247 are interacting systems, where multiple UV/optical clumps can be identified in HST optical and NIR imaging (Stanford et al. 2000; Farrah et al. 2002; Combes et al. 2012), thus it is unclear whether these galaxies support such extreme $\dot{\Sigma }_\star$.

How do we interpret such extreme systems? Can $L_{\rm IR}/L^{\prime }_{\rm CO}$ be driven to such high values through star formation alone? In individual galaxies, isolated star-forming cores might be Eddington-limited, which is to say $\dot{\Sigma }_\star$ is capped by radiation pressure from the recently formed O and B associations (e.g., Scoville 2004; Shirley et al. 2003). However, when taken as a whole, galaxies form stars at sub-Eddington rates due to intermittency—the fact that only a small fraction of the cold gas reservoir is actually forming stars (Downes & Solomon 1998; Murray et al. 2005; Andrews & Thompson 2011). Under certain conditions, however, one could envision a scenario where the majority of the gas reservoir could be driven to form stars at the Eddington limit. SDSS J1506+54 is an excellent candidate for this phenomenon.

In the absence of nuclear heating, and assuming the model of momentum-driven feedback (e.g., Murray et al. 2005) the upper limit of $L_{\rm IR}/L^{\prime }_{\rm CO}$ for star-forming galaxies is set by the Eddington luminosity. For optically thick dust emission (τ_100 μm > 1),

$$\begin{equation} L_{\rm Edd} = \frac{4\pi G c}{\kappa }X_{\rm CO}L^{\prime }_{\rm CO} \end{equation} \tag{ 1 }$$

(Andrews & Thompson 2011), here we set X_CO = 3 (see Section 5.3) and κ is the Rosseland mean dust opacity. In Figure 3 we show the Eddington luminosity predicted for $\kappa = 5\hbox{--}10\,{\rm cm}^2\,{\rm g}^{-1}\,(f_{\rm dg}/f^{\rm MW}_{\rm dg})$, where $f^{\rm MW}_{\rm dg} = {1}/{150}$ is the Milky Way dust-to-gas mass ratio (see Andrews & Thompson 2011). We assume a dust-to-gas ratio of f_dg ≈ 1/50, appropriate for dusty star-forming galaxies (e.g., Kovács et al. 2006). Increasing either f_dg (as might be appropriate in highly obscured galaxies), or κ, or decreasing X_CO, lowers L_Edd for a fixed $L^{\prime }_{\rm CO}$.

Our chosen values aim to be conservative in this respect, such that no star-forming galaxies can be described as super-Eddington, but we caution the reader that in this parameter space the Eddington limit is not expected to be a hard edge. A simple framework that explains the intrinsic range in $L_{\rm IR}/L^{\prime }_{\rm CO}$ is the model of Geach & Papadopoulos (2012) which predicts the minimal dense gas mass for a given L_IR, and a range of gas mass ratios ξ_SF = M_dense/M_total. Typically, ξ_SF ≈ 0.02 for quiescent discs and ξ_SF ≈ 0.2 for ULIRG-like systems (Figure 3). The Eddington limit in Equation (1) is approached as ξ_SF → 1, indicating that the majority of the gas in SDSS J1506+54 is in a dense, active phase and the intermittency is close to unity.

Recent studies of the local (U)LIRG population where well-sampled CO and ¹³CO ladders are available (Papadopoulos et al. 2012b) show that, when high-J CO or heavy rotor (HCN, CS) line transitions are included in radiative transfer models, the $M_{\rm H_2}/L^{\prime }_{\rm CO}=X_{\rm CO}$ factor increases by a factor ∼3–10 compared to the widely applied X_CO = 0.8 conversion⁸ usually adopted for ULIRG-like systems (Papadopoulos et al. 2012a). The explanation is that in the supersonic turbulent discs of ULIRGs, a significant fraction of the gas mass is expected to be in a dense (n > 10⁴ cm⁻³) phase (Padoan & Nordlund 2002; Papadopoulos et al. 2012a, 2012b; Papadopoulos & Geach 2012). This is not well traced by single low-J CO tracers, despite the fact that these were originally used to calibrate the value of $M_{\rm H_2}/L^{\prime }_{\rm CO}$ in ULIRGs (Downes et al. 1993; Solomon et al. 1997; Downes & Solomon 1998).

Here we assume that the majority of the gas reservoir is indeed in a dense, turbulent phase. Setting X_CO = 3 and r₂₁ = 0.8 (Papadopoulos et al. 2012b; Geach & Papadopoulos 2012), we estimate $M_{\rm H_2}=(1.9\pm 0.3)\times 10^{10}\,M_\odot$. This is in agreement with the minimum dense gas mass expected if L_IR is powered by Eddington-limited star formation, since one expects a maximum _⋆ = L_IR/M_dense ≈ 500 (L_☉/M_☉) (Scoville 2004; Thompson et al. 2005; Thompson 2009). The _⋆ expected in feedback models is indeed close to the value actually measured both for resolved star-forming cores in spiral disks and even entire starbursts (Scoville 2004; Shirley et al. 2003). Similar values of _⋆ are obtained when one considers the mass-to-light ratio of a deeply dust-enshrouded (τ_IR > 1) zero-age main sequence (ZAMS) population, with L_ZAMS ≈ L_IR (Downes & Solomon 1998). It is important to note that, regardless of the exact gas mass calibration, assuming that the majority of the infrared emission originates from enshrouded star-forming regions, the empirical scalings in Figure 3 clearly indicate that SDSS J1506+54 is close to the Eddington limit.

The compact starburst in SDSS J1506+54 should be effective in the launching of a powerful momentum-driven galactic wind, powered by radiation pressure on dust grains intermingled with the ISM, and the ram-pressure of supernovae detonations (Murray et al. 2005). The strongly blueshifted Mg ii lines are excellent evidence that this is the case (Tremonti et al. 2007; DS12).

The 2 mm spectrum of SDSS J1506+54 exhibits a marginally significant (3.2σ) narrow (σ < 40 km s⁻¹) spike close to the outflow velocity traced by Mg ii (no RFI or telluric contaminant is expected at this frequency). We can draw no conclusions on such a low-significance feature, but we remark that if cold gas has become swept up in the outflow traced by Mg ii absorption, and has survived shock-heating, then we might expect to detect entrained cold gas in this way.

We speculate that SDSS J1506+54 is the remnant of a major merger. In this picture angular momentum losses caused by the dissipation of energy in shocks and gravitational torques cause the collapse of disk-gas to the central regions in each parent, and subsequently into a single merged core. This is a classic picture of galaxy growth that has been supported for many years by both theories, numerical simulations, and observations (Toomre & Toomre 1972; Sanders et al. 1988; Hernquist 1989; Barnes & Hernquist 1991; Sanders & Mirabel 1996; Schweizer 1996; Mihos & Hernquist 1996; Wuyts et al. 2010; Hopkins et al. 2013).

Tidal interactions in earlier stages of the merger might well be responsible for the faint extended (over several kiloparsecs) optical light (Figure 1; P. H. Sell et al. 2013, in preparation). Episodes of massive star formation must have occurred in the recent past (0.5–1 Gyr), with the A stars formed in (or moved to) relatively unobscured regions, in order to produce strong Hδ absorption observed in the optical spectrum. Indeed, the presence of post-starburst spectral features is entirely consistent with the typical timescales for equal-mass mergers (Lotz et al. 2008).

In the absence of additional cooling of gas, SDSS J1506+54 must be close to the cessation of star formation. At the current rate the gas will be consumed within a few tens of Myr, increasing the stellar mass by just ∼15%. This will place the galaxy in the exponential tail of the stellar mass function at z ≈ 0.5 (Moustakas et al. 2013), but the current starburst contributes only a modest fraction of the total mass. Nevertheless, this small addition can have important consequences for the properties of the descendant (Robaina et al. 2009; Hopkins et al. 2010, 2013; Hopkins & Hernquist 2010).

Signatures of compact Eddington-limited starburst episodes could be found in the "fossil record" of local spheroids. Integral field observations of the cores of local elliptical galaxies reveal examples of compact (∼100 pc), "kinematically decoupled," relatively young (≲5 Gyr) stellar components in a subset of the population (e.g., McDermid et al. 2006). The origin of these structures could well be found in powerful, compact starbursts such as the one we present. Star formation driven at extremely high $\dot{\Sigma }_\star$ could also have a critical impact on the form of the stellar initial mass function, due to the corresponding high cosmic-ray densities in such environments (Papadopoulos et al. 2011).