Constraints on the nature of CID-42: recoil kick or supermassive black hole pair?

Abstract

The galaxy CXOC J100043.1+020637, also known as CID-42, is a highly unusual object. As an apparent galaxy merger remnant, it displays signatures of both an inspiraling, kiloparsec-scale active galactic nucleus (AGN) pair and of a recoiling AGN with a kick velocity of ≳ 1300 km s⁻¹. Among recoiling AGN candidates, CID-42 alone has both spatial offsets (in optical and X-ray bands) and spectroscopic offsets. In order to constrain the relative likelihood of both scenarios, we develop models using hydrodynamic galaxy merger simulations coupled with radiative transfer calculations. Our gas-rich, major merger models are generally well matched to the galactic morphology and to the inferred stellar mass and star formation rate. We show that a recoiling supermassive black hole (SMBH) in CID-42 should be observable as an AGN at the time of observation. However, in order for the recoiling AGN to produce narrow-line emission, it must be observed shortly after the kick while it still inhabits a dense gaseous region, implying a large total kick velocity (v_k ≳ 2000 km s⁻¹). For the dual AGN scenario, an unusually large broad-line offset is required, and the best match to the observed morphology requires a galaxy that is less luminous than CID-42. Further, the lack of X-ray emission from one of the two optical nuclei is not easily attributed to an intrinsically quiescent SMBH or to a Compton thick galactic environment. While the current data do not allow either the recoiling or the dual AGN scenario for CID-42 to be excluded, our models highlight the most relevant parameters for distinguishing these possibilities with future observations. In particular, high-quality, spatially resolved spectra that can pinpoint the origin of the broad-line and narrow-line features will be critical for determining the nature of this unique source.

1 INTRODUCTION

Recent breakthroughs in numerical relativity have allowed for fully general-relativistic simulations of black hole (BH) mergers, including calculations of gravitational waveforms and kick velocities from gravitational-wave (GW) recoil (e.g. Pretorius 2005; Campanelli et al. 2007a; Baker et al. 2008). Surprisingly, these simulations have found that GW recoil kicks may be quite large, up to ∼5000 km s⁻¹ (Campanelli et al. 2007b; Lousto & Zlochower 2011). Thus, the merged supermassive black hole (SMBH) may be ejected from its host galaxy in some cases, and kick velocities of even a few per cent of this maximum could be of astrophysical relevance. This discovery has prompted many studies of the possible electromagnetic (EM) signatures from recoiling SMBHs as well as observational searches for such objects.

While prompt EM signatures of SMBH mergers are of great interest for identifying counterparts to eventual GW detections, long-lived signatures such as offset active galactic nucleus (AGN) provide the best chance of finding recoiling SMBHs using purely EM observations. An AGN ejected from the centre of a galaxy can carry along its accretion disc as well as the broad-line (BL) region, which could produce an AGN spatially offset from its host galaxy or an AGN with offset broad emission lines (Madau & Quataert 2004; Loeb 2007; Blecha & Loeb 2008; Komossa & Merritt 2008). In either case the offset AGN lifetimes may be up to tens of Myr for a fairly wide range in kick speeds (Blecha et al. 2011). To date, several candidate recoiling SMBHs have been found with either spatial offsets (Batcheldor et al. 2010; Jonker et al. 2010) or kinematic offsets (Komossa, Zhou & Lu 2008; Shields et al. 2009; Robinson et al. 2010), though none has been confirmed (see Komossa 2012, for a review).

Recent years have also seen renewed energy in the search for inspiraling pairs of SMBHs. Systematic searches revealed that about 1 per cent of AGN have double-peaked narrow [O iii] lines, which could be a signature of SMBH orbital motion on kiloparsec (kpc) scales (Comerford et al. 2009a; Liu et al. 2010a; Smith et al. 2010). Follow-up observations using higher resolution imaging and slit or integral field unit (IFU) spectra have identified many strong dual AGN candidates (Liu et al. 2010b; Fu et al. 2011a, 2012; McGurk et al. 2011; Rosario et al. 2011; Shen et al. 2011a; Comerford et al. 2012), and in two cases, evidence for dual compact X-ray (Comerford et al. 2011) and radio (Fu et al. 2011b) sources, respectively, has been found. Shen et al. (2011a) estimate that at least 10 per cent of double-peaked narrow-line (NL) AGN may actually contain dual SMBHs.

Among these new data and interpretations of SMBHs in merging galaxies, the galaxy CXOC J100043.1+020637 (also known as CID-42; z = 0.359) is a fascinating case. Discovered serendipitously by Comerford et al. (2009b, hereafter JC09) based on the striking morphology evident in its COSMOS Hubble Space Telescope (HST)/Advanced Camera for Surveys (ACS) image, it is an apparent face-on spiral with a prominent tidal tail and two well-resolved brightness peaks in its centre, separated by 2.46 kpc (Fig. 1a). The disturbed morphology strongly indicates a recent galaxy merger. DEIMOS slit spectroscopy revealed dual emission components in three different NLs with a velocity separation of ∼150 km s⁻¹ and a spatial offset roughly consistent with the separation of the two brightness peaks in the image, albeit with low signal-to-noise ratio (S/N; JC09). No evidence for BLs is apparent in this spectrum. Thus, JC09 conclude that CID-42 is a good candidate for a kpc-scale, Type 2 AGN pair.

Around the same time, CID-42 was also identified as an unusual object during visual inspection of the optical counterparts to Chandra-COSMOS sources (Elvis et al. 2009; Civano et al. 2012a). Additionally, a broad (Hβ) line was detected in the Sloan Digital Sky Survey (SDSS) spectrum of CID-42 (Shen et al. 2011b) that was not seen in the DEIMOS spectrum. Civano et al. (2010, hereafter FC10) and Civano et al. (2012b, hereafter FC12) also found a prominent broad Hβ line in IMACS/Magellan (Trump et al. 2007, 2009) and VIMOS/VLT (zCOSMOS; Lilly et al. 2007) spectra. More interestingly, the broad and narrow Hβ components are offset by ∼1360 km s⁻¹. The double-peaked NLs seen in the DEIMOS spectrum would be unresolved in these spectra, which have higher S/N but lower spectral resolution (R = 700 on IMACS and VIMOS versus R = 3000 on DEIMOS). There is also no measurable spatial offset between the broad and narrow emission components, though this may be attributable to seeing conditions (FC10).

A BL/NL offset of 1360 km s⁻¹ is much too large to be explained by orbital motion of dual AGN on kpc scales. However, this is a signature expected from a recoiling AGN, which should retain only the most tightly bound gas and stars (including the BL region) and leave behind gas and stars at larger radii (including the NL region) (Merritt et al. 2006). Accordingly, FC10 proposed that CID-42 may contain a single, rapidly recoiling SMBH rather than two inspiraling SMBHs. In this scenario, one of the optical brightness peaks (the NW source in the HST image, Fig. 1a) corresponds to the central stellar cusp left behind by the recoiling SMBH, and the other (SE source) corresponds to the accretion luminosity of the recoiling AGN. A surface brightness decomposition of this image performed by FC10 is consistent with this picture, in that the SE source (the purported recoiling AGN) is best fitted as a point source, while the NW source (the purported empty stellar cusp) is best fitted with an extended Sérsic component.

Even more intriguing evidence as to the nature of CID-42 comes from high-resolution Chandra observations (FC12), which reveal that the X-ray emission detected in CID-42 arises from only the SE source, with a 3σ upper limit of 4 per cent of the observed X-ray flux from the NW source. This finding supports the recoil scenario for CID-42, in which only one of the optical brightness peaks contains an AGN, but it does not rule out the possibility of a quiescent or Compton thick AGN in the NW source.

Another possibility is a gravitational slingshot recoil resulting from a triple-SMBH interaction. This may occur if the binary SMBH inspiral time is long enough that the host galaxy undergoes a subsequent merger in the meantime, bringing a third SMBH into the system. We focus instead on GW recoil in our modelling, on the grounds that a triple-SMBH interaction is most likely in a gas-poor merger remnant, unlike CID-42. Nonetheless, a slingshot recoil cannot be ruled out by the observations, and this possibility is discussed further in Section 3.1.

Because CID-42 displays signatures of both dual and recoiling AGN, and because it is the only recoil candidate with evidence for both spatial and kinematic offsets, this object is an ideal case study for theoretical models. Blecha et al. (2011) have conducted a large parameter study of recoiling SMBHs in hydrodynamic simulations of galaxy mergers, which allowed them to estimate observable lifetimes of spatially and kinematically offset recoiling AGN for a wide range of kick velocities and merging galaxy parameters. Here, we use hydrodynamic simulations to constrain the possible parameters of a recoiling or dual AGN in CID-42 and to gain insight about the relative plausibility of these scenarios. We provide models for both a GW-recoiling AGN and for an AGN pair, based on the methods of Blecha et al. (2011) and Blecha, Loeb & Narayan (2012).

The details of our simulations are outlined in Section 2.1. We describe the properties of our best-fitting recoiling SMBH simulations in Section 2.2, and those of our best-fitting dual SMBH simulation in Section 2.3. In Sections 3.1–3.3, we discuss the plausibility of the recoiling and dual AGN scenarios, and we briefly summarize in Section 3.4.

2 MODELS

2.1 Initial conditions and merger simulations

To model the possible origins of CID-42, we begin by simulating a major galaxy merger using the smoothed particle hydrodynamics (SPH) code gadget-3 (Springel 2005). Each progenitor galaxy contains dark matter, star and gas particles, as well as an accreting, central SMBH. gadget-3 also contains subresolution models for star formation and for supernova and AGN feedback.

We have designed simulations for both the GW recoil and dual SMBH scenarios that match as closely as possible the observed properties of CID-42. Specifically, the progenitor galaxies have a mass ratio of q = 0.5, and the primary galaxy has a baryonic (total) mass of 1.9 × 10¹⁰ M_⊙ (4.5 × 10¹¹ M_⊙). Because star formation occurs throughout the simulation, the initial baryonic mass is chosen such that by the time the system would be observed, near the time of SMBH merger, the stellar mass matches that inferred for CID-42 (∼2.5 × 10¹⁰ M_⊙; FC10). The galaxies are disc dominated with 10 per cent of the initial stellar mass in a bulge component. For the recoiling (dual) SMBH model, the initial gas fraction in the disc (f_gas) is 0.5 (0.4). The slightly lower gas content in the latter results in a more relaxed galaxy morphology prior to the SMBH merger, consistent with observations.

Each SMBH particle has an initial dynamical mass of 10⁻⁴ the total mass. The initial ‘seed’ mass (on to which accretion is calculated smoothly via the Bondi–Hoyle–Lyttleton formula) is 7 × 10⁵ M_⊙ for the recoil model and 10⁶ M_⊙ for the dual SMBH model. The BH and star formation prescriptions are described in more detail in Springel, Di Matteo & Hernquist (2005).

In the event of a high-velocity GW recoil kick, the Bondi accretion rate (∝ρ_gas/v³_BH) drops precipitously. Accretion on to the recoiling SMBH instead becomes dominated by the gas that remains bound to the SMBH. We adopt the ‘ejected-disc’ model described in Blecha et al. (2011), in which the gas ejected along with the SMBH is modelled as an isolated accretion disc with a monotonically decreasing accretion rate normalized to the Bondi rate at the time of the kick.

In order to compare our hydrodynamic models more directly with observations, we have employed the 3D, polychromatic dust radiative transfer code sunrise (Jonsson, Groves & Cox 2010). The HST/ACS (F814W) image of CID-42 is shown in Fig. 1(a), along with sunrise images in the same band for our GW recoil model (Fig. 1b) and dual SMBH model (Fig. 1b). Recall that in the GW recoil scenario, only one (post-merger) SMBH is present in CID-42, as opposed to the two SMBHs present in the latter model. Figs 1(b) and (c) show the mergers at the time of observation (t_obs), i.e. when the spatial offset [and for the recoil, the line-of-sight (LOS) velocity offset] of the SMBH(s) matches the observed values. Although the recoiling SMBH position is marked in Fig. 1(b), the ejected-disc luminosity is not accounted for in the image, so there is no observable excess of emission from this point. In the dual SMBH model (Fig. 1c), the SMBH positions correspond to the two central brightness peaks.

The galactic morphology of each model is broadly similar to the observed morphology. While these models are not unique solutions, they have some generic properties. For example, the prominent asymmetric tidal feature can arise in the late stages of a major but unequal-mass merger, such as the q = 0.5 mergers shown here. Fainter tidal features can also be seen in the lower right (SW) edge of both the observed and simulated images. Some differences between the two models can be seen as well. The merger remnant with dual SMBHs appears to be somewhat ‘rounder’ than does the remnant with a recoiling SMBH. However, the bolometric luminosity of the recoil model (L_bol ∼ 2.5 × 10¹¹ L_⊙) is closer to that inferred for CID-42 from spectral energy distribution (SED) fitting (L_bol ∼ 2.9 × 10¹¹ L_⊙; FC10). In the dual SMBH model, L_bol is only ∼6.5 × 10¹⁰ L_⊙. Table 1 summarizes the relevant physical parameters of our models in comparison to those inferred from observations of CID-42.

Fig. 2 shows the merger evolution of various quantities for both the recoiling and dual SMBH models. Because the star formation rate (SFR) of CID-42 inferred from SED fitting is ∼25 M_⊙ yr⁻¹, the models require a starburst that more or less coincides with t_obs. Similarly, the estimated bolometric (L_24μm-UV) luminosity of the SE source, assuming that it is a Type 1 AGN, is 7.3 × 10¹⁰ L_⊙, or ∼4 per cent L_Edd for an estimated SMBH mass of 6.5 × 10⁷ M_⊙ (FC10). Thus, the SMBHs are also near their peak luminosity at the time of the SMBH merger. The presence of a stellar bulge in the progenitor galaxies acts to stabilize the gas and stellar discs against perturbations, such that the strongest gas inflow is delayed until the galaxies’ final coalescence (e.g. Mihos & Hernquist 1996). A high initial gas content in the progenitor galaxies (40–50 per cent of the disc mass) is also required for fuelling rapid star formation and SMBH growth.

The dual and recoiling SMBH models have very similar evolution, but because t_obs in the dual SMBH model necessarily occurs some time prior to the SMBH merger (t_mrg−50 Myr in our case), the system is observed before the peak SFR and |$\dot{M}_{\rm BH}$|⁠. Specifically, the SFR at t_obs in the dual SMBH model is ∼6 M_⊙ yr⁻¹versus ∼21 M_⊙ yr⁻¹ for the recoil model. At t_obs, ∼50–70 per cent of the gas has been consumed. The remaining gas mass, ∼4–5 × 10⁹ M_⊙, might be detectable, and an estimate of its neutral fraction obtained, via CO or 21-cm observations.

Our simulated merger remnants are consistent with being luminous infrared galaxies (LIRGs), as is CID-42. However, our sunrise spectra are more IR-dominated than the observed SED; we find that at the time of observation, L_24 μm/L_bol ∼ 0.2 and 0.4 for the dual and recoiling SMBH models, respectively, compared with L_24 μm/L_bol ∼ 0.1 for the Seyfert 1.8 SED fit to CID-42 (FC10). In the recoiling SMBH case, this disparity owes partly to the fact that the post-recoil AGN luminosity is calculated analytically after the simulation, and thus does not contribute to the calculated SED. Similarly, in the dual SMBH case, the (smaller) SED mismatch arises partly because L_bol for the brighter AGN is a factor of ∼7 lower than L_{bol, BH} inferred for CID-42 (FC10), such that both AGN contribute relatively little to the calculated SED. At the same time, the SFR at t_obs in the dual SMBH model is more than a factor of 3 lower than both the SFR in the recoil model and that inferred for CID-42 (FC12). However, the results presented here do not depend on the shape of the simulated SED.

2.2 Recoiling SMBH Model

In our recoil models, the central galactic escape speed at the time of the SMBH merger is 773 km s⁻¹. Thus, we expect v_k ≫ v_esc for a recoil kick of ∼1300 km s⁻¹ in CID-42, such that the recoil event must have occurred recently. In order to match the position of the SE relative to the NW source, we direct the kick at an angle of 225° in the plane of the sky. The kick velocity then has two free parameters: the fraction of the total kick directed in the LOS and the amount of the measured 1360 km s⁻¹ BL/NL offset that is due to the motion of the SMBH. Specifically, if the NL emission originates from the NW source and the BL emission from the SE, then a spatial-offset correction must be applied to the measured velocity offset, such that the actual BL/NL velocity offset would be 1180 km s⁻¹ (FC10).

We consider models for two different kick speeds. The ‘low’ kick, v_k = 1428 km s⁻¹, corresponds to [v_LOS(t_mrg)/v_k]² = 0.9 and v_LOS(t_obs) = 1180 km s⁻¹, and the ‘high’ kick, v_k = 2470 km s⁻¹, corresponds to [v_LOS(t_mrg)/v_k]² = 1/3 and v_LOS(t_obs) = 1360 km s⁻¹ (see Table 1). Note that although v_k ≫ v_esc, there is a small difference between v_LOS(t_mrg) and v_LOS(t_obs) owing to deceleration. We choose these values of v_k because velocities lower than ∼1400 km s⁻¹ would require the kick to be oriented almost entirely along the LOS, and velocities much higher than ∼2500 km s⁻¹ would have a very small probability of occurring. These probabilities are discussed further in Section 3.1, but we note that both the ‘low’ and ‘high’ kick velocities are quite large in absolute terms.

For the low kick, t_obs occurs 5.9 Myr after the recoil. The mass of the ejected gas disc is ∼1.4 per cent M_BH, and L_{bol, BH} = 56 per cent L_Edd at the time of the kick. At t_obs, the AGN luminosity has declined to 1 per cent L_Edd, or 3.2 × 10⁹ L_⊙ (assuming a radiative efficiency of 0.1). For the high kick, a gas mass of only ∼0.5 per cent M_BH remains bound to the recoiling SMBH, but because the system would be observed sooner (t_obs = t_mrg + 1.3 Myr), the AGN luminosity is slightly higher. These values are summarized in Table 1. While these AGN luminosities are lower than that estimated for CID-42 via SED fitting (FC10), the Eddington ratios of the order of ∼1 per cent are consistent.

The relationships between t_obs, t_AGN and v_k are illustrated in Fig. 3. For the low-luminosity AGN definition (L_{bol, BH} > 3 × 10⁹ L_⊙), the AGN lifetimes of the recoiling BHs (as defined in equation 1) are 2.3 and 6.4 Myr for the low and high kicks, respectively, i.e. both have t_AGN > t_obs. In fact, for the observed parameters of CID-42, virtually all allowed values of v_k and t_obs result in a low-luminosity recoiling AGN with t_AGN > t_obs.

Moreover, we expect the recoiling AGN lifetimes to be even longer than these estimates, because the AGN luminosities and SMBH masses calculated self-consistently from our recoil simulations are lower than the values inferred for the SE source. We can use the fully analytic ejected-disc model to calculate the post-recoil AGN luminosities and lifetimes for arbitrary values of M_BH(t_mrg) and |$\dot{M}_{\rm BH}$|(t_mrg). Specifically, we can match |$\dot{M}_{\rm BH}$|(t_obs) to the observed luminosity of the SE source in CID-42 and extrapolate backwards to obtain its value at t_mrg for each recoil model. The estimates of t_AGN using these values (along with M_BH inferred for the SE source) are indicated by the dashed lines in Fig. 3. Here, the SMBH in each recoil model has L_{bol, BH}/L_Edd > 3 per cent at t_obs. By the lowest luminosity AGN definition, t_AGN is 4.9 and 18 Myr for the low- and high-kick models, respectively. Therefore, the fuel supply carried along with the ejected SMBH should not be a limiting factor for the recoiling AGN scenario.

Another key consideration is whether the recoiling AGN is able to ionize gas and produce NL emission as it leaves the galaxy. Based on the current data, it is unclear whether the observed NL emission is coming from the SE or NW source. However, this will be constrained in the near future via individual HST spectra of each source (Civano et al., in preparation), so it is important to know whether a recoiling AGN might be capable of producing the observed emission lines.

It is clear from Fig. 4 that the recoiling SMBH is ejected to a low-gas-density region within at most a few Myr. In both recoil models, all of the gas near the SMBH at t_obs is warm and diffuse, with an average temperature of a few ×10⁵ K and densities of 10^−1.5 and 10^−2.5 cm⁻³ for the high- and low-kick models, respectively. None of this gas is condensed into the ‘cold’ phase of the multiphase model for the interstellar medium (ISM) used in gadget-3 (Springel & Hernquist 2003). Thus, we would not a priori expect any cold clouds to be locally ionized by the recoiling SMBH. However, sufficiently luminous AGN may ionize gas from outside the galaxy, similar to what is observed in some merging galaxies containing quasars (e.g. da Silva et al. 2011).

Blecha et al. (2012) have developed a semi-analytic model for estimating the NL emission around AGN in hydrodynamic simulations of galaxy mergers. We have applied this model to our simulations of recoiling SMBHs in CID-42, including accretion from the ejected gas disc. In order to estimate the lifetime of observable NL emission, we assume a ratio of the [O iii] λ5007 luminosity to the bolometric AGN luminosity of L_[O iii]/L_{bol, BH} = 10^{− 3}. This is the ratio of the total [O iii] luminosity to the luminosity of the SE source estimated by FC10. In reality, this ratio may be higher (such that the estimated NL lifetime would be an upper limit), because L_[O iii] is measured from the total continuum rather than just the AGN continuum. However, additional uncertainty comes from the fact that L_[O iii] is measured from a slit spectrum, whereas L_{bol, BH} is measured in a circular aperture; the direction of this effect is less clear. We therefore consider the ratio 10⁻³ to be a reasonable guess, but not necessarily an upper limit.

The NL model of Blecha et al. (2012) calculates the Hβ rather than the [O iii] luminosity, because the former is less sensitive to the conditions in the ISM. Thus, we must also assume an [O iii]/Hβ ratio. The ratio estimated from the spectra of CID-42 is ∼6 (JC09; FC10), so we use a ratio of 10 as a conservative upper limit. This implies a minimum observable L_Hβ/L_{bol, BH} of 10⁻⁴.

In this analysis of the NLR, we must account for the difference between the AGN luminosities calculated self-consistently in our simulations and the AGN luminosity of the SE source estimated from its F814W magnitude. Taking the values from the simulation, we find that the recoiling AGN can produce NL emission above the estimated minimum value of 3–5 × 10⁵ L_⊙ for only about |$1\text{--}2$| Myr after the kick, depending on v_k. In each case the NL lifetime is shorter than t_obs. Note that any observed photoionization of NLRs must be ongoing; otherwise, NL AGN emission would essentially last for only a light crossing time ( ≲ 10³ yr for a typical NLR). After this, [O iii] would disappear on a time-scale of ∼years via charge-transfer recombinations (destroying the large [O iii]/Hβ ratio characteristic of AGN), followed by H recombination on a time-scale of ≲ 10³ yr.

If instead we assume that the recoiling AGN has M_BH(t_mrg) = 6.5 × 10⁷ M_⊙ and L_{bol, BH} = 7.3 × 10¹⁰ L_⊙ (the estimates for the SE source in CID-42 assuming that it is a Type 1 AGN; FC10), the NL lifetimes are longer, despite the stricter minimum Hβ luminosity of 7.3 × 10⁶ L_⊙. Specifically, the NL lifetime for the low kick is 4.0 Myr. This is still shorter than t_obs = 5.9 Myr, so the recoiling AGN in this model should not produce observable NL emission. In the high-kick model, however, the NL lifetime is 1.6 Myr, compared to t_obs = 1.3 Myr. This suggests that a recoiling AGN in CID-42 could to produce the observed NLs, provided that the total kick velocity is sufficiently large ( ≳ 2000 km s⁻¹).

If the NL emission arises from stellar photoionization in the NW source, the recoiling AGN scenario does not require that the kicked SMBH produce observable NLs. However, the current spectra of CID-42 have AGN-like NL flux ratios (JC09). This suggests that the recoil scenario requires either a large total kick velocity or an unusual star-forming region in the NW source. Alternatively, a recoiling AGN might be able to produce NL emission more locally if the SMBH happened to be ejected along a dense tidal stream or if the galactic LOS gas distribution differed substantially from that of our model.

Finally, we note that the total LOS gas column density to the position of the BHs is low; log N_gas ≈ 20.8–20.9 cm⁻². Thus, attenuation is insignificant along the LOS to the recoiling AGN, even though it is recoiling away from the observer (as inferred from the redshifted broad Hβ line).

2.3 Dual SMBH model

In the dual SMBH model, the projected SMBH separation of 2.5 kpc occurs at t_obs = t_mrg−50 Myr. The slightly less dissipative merger in this model is key to reproducing the observed morphology prior to the SMBH merger. If we simply examine the pre-merger snapshot from the recoil model in which the SMBHs have a projected separation of 2.5 kpc, we find that the galaxy is substantially more disturbed, with multiple prominent tidal features and a more compact and asymmetric core.

The SMBH masses at t_obs are 4.5 × 10⁶ and 2.1 × 10⁶ M_⊙. Thus, each SMBH is more than a factor of 10 smaller than the estimated mass of the SE source in CID-42. The less massive SMBH is actually more luminous, with |$\dot{M} = 13$| per cent |$\dot{M}_{\rm Edd}$| and L_{bol, BH} = 8.9 × 10⁹ L_⊙. The more massive SMBH is accreting at only 0.8 per cent |$\dot{M}_{\rm Edd}$|⁠, which corresponds to a bolometric luminosity of L_{bol, BH} = 1.2 × 10⁹ L_⊙. Notably, this lower luminosity AGN is consistent with the upper limit on the X-ray luminosity for the NW source of CID-42 (L_X < 10⁴² erg s⁻¹), so this alone does not place constraints on whether it may contain an SMBH.

However, because the more luminous AGN in our model is about eight times fainter than the luminosity of the SE source in CID-42, the AGN luminosity ratio in our model may be more informative than the absolute luminosities. In CID-42, the measured F814W luminosity ratio of the NW and SE sources is ∼2.5, and the NW source is brighter (FC10). Given the uncertainty as to whether an AGN exists at all in the NW source, this tells us little about the possible AGN luminosity ratio. Instead, we compare the limit on X-ray emission from the NW source with the bolometric luminosity ratio of our simulated AGN. Luminosity ratios greater than 25 correspond to the 4 per cent upper limit on the fraction of hard X-ray flux originating in the NW source (FC12). In the kpc-scale phase of our simulation (defined loosely as SMBH separations <10 kpc), the AGN luminosity ratio is above 25 for ≲ 1 per cent of the time, and not at all for separations <5 kpc. At t_obs, the luminosity ratio of the two AGN is 7.5. This suggests that an AGN pair with an intrinsically large luminosity ratio is a possible but relatively unlikely explanation for the lack of X-ray emission from the NW optical source of CID-42.

We must also consider that a highly obscured AGN could be present in the NW source. Given the lack of detection of X-rays from this location, an obscured AGN here would have to be Compton thick (log N_H > 24 cm⁻²). The total LOS gas column densities to the SMBHs in our model are log N_gas ≈ 23.8 and 22.1 cm⁻² for the more and less luminous AGN, respectively. Notably, the column density of the more luminous AGN is near the Compton thick limit, though these numbers should not be taken at face value. The Compton thick limit refers to the equivalent neutral hydrogen column density of the gas, which depends on the ionic fraction of each element, whereas we obtain the total gas column density from our simulations. Our column densities therefore represent an upper limit on N_H. Further, the fact that the more intrinsically luminous AGN is hosted in the more obscured density cusp works against the need for a large emergent luminosity ratio in the hard X-ray band.

Although we do not find strong evidence for an AGN obscured by the galactic environment of CID-42, many AGN are thought to contain dense, obscuring material on small scales, well below the resolution limit of our simulations. If Compton thick clouds fell along our LOS to an AGN in the NW source, this could cause significant attenuation through the hard X-ray band. In this case, much of the AGN luminosity would be reprocessed as IR emission, creating an IR excess relative to that expected for a star-forming galaxy with an SFR of ∼25 M_⊙ yr⁻¹. Thus, the fact that the SED of CID-42 is well fitted by a template for a star-forming galaxy (FC10) argues against the presence of a Compton thick AGN.

Strong Fe Kα emission (EW ∼ 1–2 keV; Matt, Guainazzi & Maiolino 2003) is also expected in heavily obscured X-ray sources (those with log N_H ≳ 23 cm⁻²). A Fe Kα emission feature with EW ∼ 100–500 eV is indeed observed in both the Chandra and XMM–Newton data for CID-42 (FC10). However, because this equivalent width could easily be produced by a Type 1 AGN in the SE source, we cannot determine whether the Fe Kα line originates from an obscured AGN in the NW source.

As with the recoil scenario, we have applied the NL model of Blecha et al. (2012) to our dual SMBH simulation. In this case, we wish to know whether the system should appear as an NL AGN and whether the emission lines may be double peaked, as indicated by the spectra in JC09. We find that at t_obs, the lower luminosity AGN has a narrow Hβ luminosity too low to be observable. The more luminous AGN has a higher ionizing photon flux, but as it also resides in a higher density region than the first AGN, its ionization parameter is marginally too low by the criteria of Blecha et al. (2012). The lack of clear evidence for strong NL emission from these AGN reflects their low luminosities relative to the SE source of CID-42. Regardless of the NL luminosities, their relative LOS velocity in this snapshot is only 64 km s⁻¹. Because this is smaller than the typical FWHM of NLs, it is unlikely that double-peaked emission lines would be resolvable in this case. Our model for CID-42 is not a unique solution, however, and the LOS velocities for projected SMBH separations of 2–3 kpc range from ∼0–150 km s⁻¹. Accordingly, we cannot rule out the possibility that a system like CID-42 could produce double-peaked NLs. Civano et al. (in preparation) have obtained a new DEIMOS spectrum of CID-42 with much higher S/N, which will reveal definitively whether the NLs are double peaked.

3 DISCUSSION

3.1 GW recoil scenario

Several observed features of CID-42 support the scenario that it contains a single, rapidly recoiling SMBH. The offset between the broad and narrow emission lines is expected to be characteristic of recoiling AGN (Merritt et al. 2006). The single X-ray source coincident with the SE optical brightness peak is also consistent with the GW recoil scenario, in which the SE source corresponds to a recoiling AGN that has left the central stellar cusp behind (FC12). Additionally, the best-fitting surface brightness decomposition performed by FC10 consists of an extended (Sérsic) component for the NW source and a point-like component for the SE source, consistent with a stellar cusp and a recoiling AGN, respectively.

We note that because CID-42 is actively star forming, implying a high gas content, we expect the merger remnant to have a stellar cusp even after the displacement of stars via binary SMBH inspiral (e.g. Milosavljević & Merritt 2001; Yu 2002; Merritt et al. 2006) and the recoil event itself (Boylan-Kolchin, Ma & Quataert 2004; Gualandris & Merritt 2008). Gas-rich major mergers trigger a rapid inflow of cold gas and subsequent star formation (e.g. Barnes & Hernquist 1996; Mihos & Hernquist 1996; Mayer et al. 2007; Hopkins et al. 2008), thereby maintaining a steep central density profile. Stellar interactions with the SMBHs should be relatively unimportant in such cases, because hydrodynamic processes drive the efficient formation and inspiral of the binary SMBH (Mayer et al. 2007). Furthermore, Blecha et al. (2011) found that recoil kicks may actually enhance the central stellar density by displacing AGN feedback and delaying the quenching of star formation. We accordingly refer to a stellar ‘cusp’ when discussing the NW source, though the observations do not exclude the presence of a stellar core on scales below the HST image resolution of 0.03 arcsec (150 pc).

Our recoiling SMBH models of CID-42 are able to reproduce the morphology and stellar mass of CID-42 in the late stages of a gas-rich, major galaxy merger. The peak SFR and |$\dot{M}_{\rm BH}$| occur near the time of SMBH merger, and their values at t_obs are consistent with estimates from the observations (SFR ∼ 25 M_⊙ yr⁻¹ and L_{bol, BH}/L_Edd ∼ few per cent; FC10; FC12).

Less clear is whether a recoiling AGN could produce the observed NL emission in CID-42. If the total kick velocity is large, such that the system is observed soon after the kick, the recoiling AGN may still be close enough to the dense gaseous region to produce observable NL emission. Specifically, we find that the recoiling AGN in our high-kick model (v_k = 2470 km s⁻¹) could produce the observed NLs in CID-42, but that it probably could not do so with a lower kick of 1428 km s⁻¹.

In the latter case, special configurations may be required for the recoiling AGN to produce the NL emission, such as ejection of the SMBH along a dense tidal stream. If the observed NLs originate from star formation in the NW source rather than from the recoiling SMBH, then the recoil scenario requires unusual line flux ratios for a star-forming region. The origin of the NL emission is therefore an important consideration in evaluating the likelihood that CID-42 contains a recoiling AGN, and underscores the need for corroborating evidence. Individual spectra of the NW and SE sources will provide important constraints in this regard (Civano et al., in preparation), and IFU data could also provide useful information about the spatial distribution of narrow (and broad) emission-line regions in CID-42.

IFU spectra, or HST imaging in multiple bands, would also constrain the colours of the NW and SE sources, and of the galaxy as a whole. Specifically, this would indicate whether the galaxy is indeed blue and actively star forming, as predicted by SED fitting (FC10; FC12) and our simulations. Such observations would also reveal whether the starburst is concentrated within the NW source, and they would provide an independent check on the starburst age of a few Myr estimated from SED fitting (FC12).

Another consideration is that a merger remnant with a recoiling AGN should contain, in principle, one central brightness peak (the stellar cusp) and one offset brightness peak (the AGN). An AGN pair, in contrast, should have both brightness peaks offset from the centre. However, this is an inconclusive diagnostic for CID-42, because its centre of stellar light is not well constrained. The surface brightness decomposition of FC10 places the centre between the NW and SE sources, but it is consistent with being at the position of the NW source within the errors. The central stellar cusp could also be intrinsically asymmetric in the unrelaxed merger remnant, and tidal features seen in projection could contribute light to the apparent galactic disc.

As noted in Section 2.2, even the lowest possible recoil velocity for CID-42 is quite large ( ≳ 1300 km s⁻¹). The intrinsic probability of such a large kick is not clear. Recoil kick probability distributions have been derived from numerical relativity simulations (e.g. Baker et al. 2008; van Metre et al. 2010; Lousto et al. 2012). However, the distributions of progenitor mass ratios, spins and spin orientations are still quite uncertain. In particular, if a circumbinary gas disc forms around the inspiraling SMBHs, it may align or partially align their spins (Bogdanović, Reynolds & Miller 2007; Dotti et al. 2010). The maximum recoil velocity resulting from the merger of BHs with aligned spins is only ∼200 km s⁻¹. Partial alignment may also occur via GR precession effects (Kesden, Sperhake & Berti 2010; Berti, Kesden & Sperhake 2012), though this mechanism can also result in anti-alignment of the BH spins, yielding substantially larger kicks. At any rate, it is clear that the GW recoil scenario for CID-42 requires that spin alignment prior to BH merger is inefficient, at least in some cases.

3.2 Gravitational slingshot recoil scenario

A rapidly recoiling SMBH could also result from the gravitational slingshot mechanism. As described in Section 1, this requires a long binary SMBH inspiral time, such that a subsequent galaxy merger introduces a third SMBH before the initial SMBH binary has coalesced. A three-body encounter may result in the ejection of the lightest SMBH at high speeds, and may also cause the rapid merger of the remaining SMBHs (Hoffman & Loeb 2007). Thus, in the slingshot-recoil scenario, two or three SMBHs would be present in CID-42: the SE source would be the ejected SMBH, and the NW source would contain the SMBH(s) left behind.

While we have not constructed a model for this scenario, much of our analysis for the GW recoil model can be applied here. If the ejected SMBH retains a gas disc after the three-body encounter (such that it appears as an AGN), the accretion rate will evolve as described in Section 2.1. The relationship between v_k and t_obs would also be the same as in Fig. 3. If the NW source contains an obscured, radio-loud AGN, this could be detected with very long baseline interferometry (VLBI). If it contains a binary AGN that survived the three-body encounter, the binary might be spatially resolvable with VLBI imaging, provided that its separation were at least a few parsec (pc) and the AGN were both radio loud. However, binary radio-detected AGN are intrinsically quite rare, with only one example known (Rodriguez et al. 2006) despite a dedicated search of thousands of archival VLBI images (Burke-Spolaor 2011).

If SMBH binaries do not merge efficiently, the probability of such triple SMBH systems forming is non-negligible, especially at high redshift (z ≳ 2; Kulkarni & Loeb 2012). In our simulations, efficient merging of the SMBHs is merely an assumption owing to resolution limits. However, smaller scale simulations of inspiraling SMBHs indicate that gas drag (e.g. Escala et al. 2005; Dotti et al. 2007) or non-axisymmetry of the stellar potential (e.g. Berczik et al. 2006; Khan et al. 2012) can prevent SMBH binaries from ‘stalling’ (the so-called ‘final pc problem’; e.g. Begelman et al. 1980; Yu 2002). CID-42 is clearly a gas-rich system with a disturbed stellar distribution. Thus, while the possibility of gravitational slingshot recoil cannot be excluded, we consider it to be unlikely. Put another way, if new evidence points towards a slingshot recoil in CID-42 (for example, if compact radio emission were detected in the NW source and consistent velocity offsets were found in multiple BLs in the SE source), the implications for binary SMBHs would be far-reaching. This would indicate that many unmerged SMBH binaries may be lurking below resolution limits even in gas-rich environments, to the detriment of SMBH merger event rates for future GW observatories.

3.3 Dual SMBH scenario

The observations of CID-42 are also consistent with an SMBH pair rather than a recoiling SMBH. In this scenario, the NW and SE sources are two inspiraling SMBHs with a 2.5 kpc separation. However, the ∼1300 km s⁻¹ offset of the broad Hβ line is too large to be associated with orbital motion on kpc scales. Many AGN spectra show offset BLs that are often attributed to outflows, but the offset line in CID-42 is atypically large. CID-42 constitutes 1/323, or 0.3 per cent, of BL AGN in the Chandra-COSMOS survey with spectroscopic redshifts (Elvis et al. 2009), and less than 1 per cent of SDSS quasars have broad Hβ lines offset by >1000 km s⁻¹ (Bonning, Shields & Salviander 2007; Eracleous et al. 2012).

Theoretically, a projected SMBH separation of 2.5 kpc could also occur during an early close passage of the merging galaxies, or even during a flyby encounter. However, this is an unlikely explanation for CID-42. Its tidal features are indicative of a late-stage merger, and the highest |$\dot{M}_{\rm BH}$| and SFR typically occur near coalescence (see Fig. 2). An unbound encounter would be required to explain the 1300 km s⁻¹ BL offset via relative galactic motion, whereas the maximal relative speeds in our galaxy merger simulations are ∼500–600 km s⁻¹. In such a flyby, only very specific configurations and timing would allow a projected separation of 2.5 kpc to be observed.

Our dual SMBH model is well matched to the morphology and stellar mass of CID-42, but a slightly less dissipative merger (lower f_gas) is required in order to match the observed morphology prior to the SMBH merger. t_obs also occurs slightly before the peak SFR and SMBH accretion rate. As a consequence, the SFR, |$\dot{M}_{\rm BH}$| and luminosities are lower in the dual SMBH model than the estimates for CID-42.

Owing to these low AGN luminosities of a few ×10⁹ L_⊙, their NL emission (as determined from the model of Blecha et al. 2012) may be too faint to be observable. Also, their relative LOS velocity of only 64 km s⁻¹ is unlikely to be spectroscopically resolvable. This is consistent with the supposition that the double-peaked NL structure in the DEIMOS spectrum (JC09) does not arise from dual SMBH orbital motion.

The non-detection of X-ray emission in the NW source (FC12) places some constraints on whether it may host an AGN. The NW source could contain a quiescent SMBH with L_X < 10⁴² erg s⁻¹, but our models indicate that two SMBHs in a late-stage, gas-rich major merger are unlikely to have a large intrinsic luminosity ratio. Specifically, an X-ray faint AGN in the NW source requires a luminosity ratio >25, which occurs for <1 per cent of the kpc-scale phase in the dual SMBH simulation.

The NW source could also contain a Compton thick AGN. The upper limits on the gas column densities along the LOS to each SMBH (log N_gas ≈ 22.1 and 23.8 cm⁻²) indicate that these AGN should be absorbed and highly obscured, respectively, but not Compton thick. The hard X-ray-detected AGN pair in NGC 3393, for example, has estimated LOS column densities of a few ×10²³ cm⁻² (Fabbiano et al. 2011). It is possible that obscuration on subresolution scales could hide an AGN in the NW source, but in this case, we expect the SED to be more IR-dominated than what is observed. Finally, while a strong iron emission line can indicate the presence of a heavily obscured AGN, this diagnostic is not informative for CID-42. A fairly strong Fe Kα emission line is detected, but it is impossible to say how much, if any, of this emission is contributed by the NW source, which has a weaker continuum by a factor of at least 25.

The detection of compact radio emission from either or both optical brightness peaks in CID-42 would provide strong evidence for the presence of one or two AGN. As mentioned above, however, radio-detected AGN pairs appear to be very rare, with only one radio-loud kpc-scale quasar pair (3C 75; Owen et al. 1985) and one pc-scale AGN pair (Rodriguez et al. 2006) known. Regardless, given the ambiguous nature of the NW optical source and the X-ray non-detection there, the possibility of a radio detection is enticing, as this would provide the clearest evidence for an obscured AGN in the NW source.

3.4 Summary

We find that neither the recoiling nor dual SMBH scenarios for CID-42 can be ruled out based on current data. It is clear that whatever its true nature, CID-42 is not a ‘typical’ galaxy even among the rather exotic subsets that host dual or recoiling AGN. Both scenarios require some extraordinary features, such as a dual AGN with an unusually large BL offset or a rapidly recoiling AGN with an atypical NL region. Through our analysis we have identified and constrained the requisite properties of CID-42 in either case. These constraints combined with further observations in the coming months and years, such as spatially resolved and higher quality optical spectra, radio imaging and CO or 21-cm observations, should rapidly narrow the possible origins of this intriguing system.

We would like to thank Julie Comerford for useful discussions and Chris Hayward, Patrik Jonsson and Greg Snyder for helpful guidance with using sunrise. This work was supported in part by NSF grant AST-0907890 and NASA grants NNX08AL43G and NNA09DB30A.

REFERENCES