HOT DUST OBSCURED GALAXIES WITH EXCESS BLUE LIGHT: DUAL AGN OR SINGLE AGN UNDER EXTREME CONDITIONS?

The most luminous infrared (IR) galaxies in the universe are thought to be massive galaxies during a key stage in their evolution (see, e.g., Hopkins et al. 2008). These galaxies are undergoing a combination of intense star formation and intense accretion onto their central super-massive black hole (SMBH), in both cases heavily enshrouded in dust. This makes these objects exceedingly luminous in the IR but typically faint at UV/optical wavelengths. There are several populations of galaxies that are known to be in such stages, both locally, such as ultra-luminous infrared galaxies (ULIRGs; e.g., Sanders & Mirabel 1996), and at high-z, such as submillimeter galaxies (SMGs; Blain et al. 2002; Casey et al. 2014), dust-obscured galaxies (DOGs; Dey et al. 2008) and hot dust-obscured galaxies (Hot DOGs; Eisenhardt et al. 2012; Wu et al. 2012), the latter of which were recently discovered by NASA's Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010).

Hot DOGs have characteristics that clearly separate them from other populations of luminous galaxies. One of their most distinctive properties is the hot dust temperatures from which the population draws its name (see Wu et al. 2012). Objects such as ULIRGs and SMGs have IR spectral energy distributions (SEDs) that peak at λ ∼ 100 μm and have typical dust temperatures of up to ∼40 K (e.g., Blain et al. 2003; Magnelli et al. 2012). This is consistent with their IR luminosities being powered primarily by star formation. DOGs can have somewhat warmer dust temperatures, as they are powered by a combination of active galactic nuclei (AGNs) and star formation (Melbourne et al. 2012). Hot DOGs have much hotter dust temperatures of T > 60 K (Wu et al. 2012), with shoulders to their SEDs of far-IR emission very significantly broader than single-temperature modified black-bodies, and with dust components with temperatures up to ∼450 K (Eisenhardt et al. 2012; Tsai et al. 2015). Because of these high temperatures, their SEDs peak at rest λ ∼ 20 μm, suggesting they are powered by extremely luminous, highly obscured AGNs.

Although the redshift distribution of Hot DOGs is similar to that of DOGs and SMGs, with most objects found at 1 < z < 4 (Assef et al. 2015; P. R. M. Eisenhardt et al. 2016, in preparation), their luminosities are significantly larger. Almost all Hot DOGs seem to have luminosities exceeding L_IR(8–1000 μm) = 10¹³ L_⊙ (Tsai et al. 2015), placing them in the category of hyper-luminous infrared galaxies (HyLIRGs, L_IR > 10¹³ L_⊙). Indeed, a significant fraction have L_IR > 10¹⁴ L_⊙ (Wu et al. 2012), making them extremely luminous infrared galaxies (ELIRGs, L_IR > 10¹⁴ L_⊙; Tsai et al. 2015). Hot DOGs rank among the most-luminous galaxies known (Tsai et al. 2015). In fact, Hot DOG W2246–0526 is the most luminous galaxy currently known (Diaz-Santos et al. 2016; Tsai et al. 2015). As expected for such objects, Hot DOGs are a rare population, with WISE only identifying one candidate every ∼30 deg² (Eisenhardt et al. 2012; Wu et al. 2012; Assef et al. 2015), with space densities comparable to those of similarly luminous QSOs (Assef et al. 2015). Hot DOGs also reside in significantly overdense environments (Jones et al. 2014; Assef et al. 2015) and a large fraction show extended Lyα emission features on scales of up to ∼100 kpc (Bridge et al. 2013).

The very red rest-frame optical through mid-IR SEDs of Hot DOGs imply that the hyper-luminous AGN that powers the IR is under significant obscuration ($\langle E(B-V)\rangle =6.4$, up to $E(B-V)\sim 20$; Assef et al. 2015), enough that the rest-frame UV-through-NIR SED is dominated by the underlying stellar emission, providing an uncompromised view of the host galaxy. Using Spitzer and ground-based NIR follow-up imaging observations to constrain the host-galaxy properties, Assef et al. (2015) studied their stellar masses and determined upper bounds of ${M}_{*}\lesssim {10}^{12}\;{M}_{\odot }$. They also showed this result is supported by the density of their environments as measured from Spitzer follow-up imaging (see Assef et al. 2015, for details). Although Hot DOG host galaxies may be among the most massive ones at their redshifts, the AGN is still unexpectedly luminous. Assef et al. (2015) show that in order to explain the AGN activity, it is necessary that either the SMBH is overly massive for its host galaxy or that the AGN is radiating significantly above the Eddington limit, or possibly a combination of both. The super-Eddington accretion scenario is also supported by the study of the most luminous Hot DOGs (L_IR > 10¹⁴ L_⊙) conducted by Tsai et al. (2015).

Being able to directly probe the host galaxy of these objects can also allow us to understand their rest-frame UV/optical properties. Previous studies of the SEDs of Hot DOGs have shown that most are well modeled by a regular star-forming galaxy at these wavelengths (see Eisenhardt et al. 2012; Assef et al. 2015, as well as Section 2). In this work, however, we focus on a small fraction of them that show exceedingly blue UV/optical SEDs. When modeling the SEDs of these Hot DOGs with excess blue emission, we find with a significant probability that a type 1 AGN is needed to explain the blue rest-frame colors, but with a bolometric luminosity that is ∼1% of that of the highly obscured AGN powering the hyper-luminous IR emission. These uncommon sources may provide important insights into the nature of the Hot DOG population and their role in the galaxy evolution paradigm. X-ray observations can provide key tests of explanations for them. As Hot DOGs are typically faint in soft X-rays, due to the large, possibly Compton-thick, absorbing H i column density (Stern et al. 2014; Piconcelli et al. 2015), deep X-ray observations are required to carry out such tests.

One of these uncommon sources, WISE J020446.13–050640.8 (W0204–0506 hereon) is located in the NOAO Deep Wide-Field Survey (Jannuzi & Dey 1999, NDWFS) Cetus field, within the footprint of the Large-Area Lyman Alpha survey (LALA; Rhoads et al. 2000), which has been observed to a depth of 174.5 ks by the Chandra X-ray Observatory with the ACIS-I instrument (Wang et al. 2007). This makes W0204–0506 the Hot DOG with the deepest X-ray observations to date. We use these X-ray observations in combination with multi-wavelength data to study W0204–0506. In Section 2 we discuss the selection and possible nature of objects with blue excess emission in general, while in Section 3 we focus on the UV through mid-IR SED of W0204–0506. In Section 4 we discuss the Chandra observations, and in Section 5 we discuss the nature of W0204–0506. Throughout this work we assume a flat ΛCDM cosmology with H₀ = 73 km s⁻¹ Mpc⁻¹, Ω_M = 0.3, and Ω_Λ = 0.7. We refer to all magnitudes in their standard photometric system, namely AB for griz and Vega for all other bands.

As discussed earlier, the fact that the AGN powering the IR emission of Hot DOGs is under heavy dust obscuration allows for the possibility of studying in depth the underlying host galaxy. The rest-frame UV through NIR SED of a galaxy can provide interesting constraints on the underlying stellar population, such as its age and mass, as well as the unobscured star formation rate (SFR) (e.g., Kennicutt 1998) and the star formation history. Eisenhardt et al. (2012) showed that the first Hot DOG studied in depth, WISE J181417.29+341224.9, had an SED dominated by a strongly star-forming galaxy (∼300 M_⊙ yr⁻¹) at λ_rest ≲ 1 μm. This is consistent with the fact that this object has a rest-frame UV spectrum similar to a Lyman break galaxy, with no discernible emission lines other than Lyα, and several interstellar absorption lines typical of the Lyman break population (e.g., Shapley et al. 2003). Wu et al. (2012) showed that although a fraction of Hot DOGs have similar rest-frame UV spectra, most show high-ionization narrow emission lines indicative of type 2 AGN and a small subset even show broad emission lines such as C iv. The expectation is, then, that these objects have a significant range of rest-frame UV through NIR SEDs.

To study their SEDs we rely here on the subset of sources with photometry provided by the Sloan Digital Sky Survey (SDSS; York et al. 2000). Note that although Hot DOGs are selected over the entire extragalactic sky, they are typically faint in the optical, with r ≳ 23. This implies that although a significant fraction of our objects lie within the SDSS footprint, only a fraction are detected. Specifically we find that 433 out of 934 Hot DOG candidates fall within the area covered by SDSS DR12 (Ahn et al. 2014), of which 153 (35%) are detected with signal-to-noise ratio (S/N) ≥ 3 in at least one band and 114 are detected at the same level in at least three passbands. To complement this, we use the NIR imaging from Assef et al. (2015) and the deeper r-band observations from the Hot DOG imaging program presented by P. R. M. Eisenhardt et al. (2016, in preparation). A brief description of the latter is also provided by Assef et al. (2015).

Assef et al. (2015, also see Eisenhardt et al. 2012) recently studied the rest-frame optical through mid-IR SEDs of a large sample of Hot DOGs, and showed that the combination of a composite host galaxy SED template and a single obscured AGN SED template typically does a good job of modeling the photometry. Assef et al. (2015) model their SEDs using the algorithm and templates of Assef et al. (2010). The best-fit SED models consist of a non-negative combination of three empirically derived galaxy templates, approximately corresponding to E, Sbc and Im type galaxies, and a single AGN template. They also fit for the reddening of the AGN component, parametrized by the color excess $E(B-V)$, considering values from 0 to 10^1.5. The assumed reddening-law corresponds to an SMC-like extinction for λ < 3300 Å, and a Galactic extinction curve at longer wavelengths. Additional details are provided in Assef et al. (2010, 2015). From here on we will refer to this SED model as the "1AGN" model, as a single AGN component is used. Assef et al. (2015) modeled the SED of 96 Hot DOGs with spectroscopic redshifts z > 1 and for which follow-up imaging had been obtained by Spitzer (Griffith et al. 2012), using WISE W3 and W4 bands from the WISE All-sky data release (Cutri et al. 2012), [3.6] and [4.5] imaging from the Spitzer follow-up, and J, H and Ks imaging from the NIR follow-up program presented by Assef et al. (2015) whenever available. The suggested correction to the WISE W3 and W4 band photometry for red sources by Wright et al. (2010, also see Brown et al. 2014) was applied before modeling the SEDs.

Here, we recalculate these fits but modify the sample to encompass only the 36 Hot DOGs with (a) z > 1 (following Assef et al. 2015) and (b) available ugriz modelMag ¹² photometry in the SDSS DR12 database with S/N > 3 in at least one band. Note that of these, 28 have S/N ≥ 3 in at least three SDSS bands. Unlike Assef et al. (2015), we also consider here Hot DOGs not covered by the Spitzer follow-up program, using for them the lower S/N W1 and W2 fluxes from the AllWISE data release (Cutri et al. 2013)¹³ instead. We note that Assef et al. (2015) decided to not use AllWISE fluxes due to concerns over the selection function modeling, but these are not important for our current study. Additionally, we add the r-band photometry from the follow-up program described by P. R. M. Eisenhardt et al. (2016, in preparation) for 25 sources.

When including the observed optical photometry we find that the "1AGN" model no longer provides a good fit for a fraction of Hot DOGs which present significant excess emission at rest-frame UV/optical wavelengths compared to what is allowed by the SED templates of Assef et al. (2010). This excess is similar to what would be expected from an unobscured AGN, although much less luminous than the one powering the IR emission. To properly identify these sources, we refit all 36 objects in our sample using the same algorithm described earlier, but adding an additional AGN component with the same template for which we fit an independent normalization and reddening value. We refer to this SED model as the "2AGN" model. Note that the "1AGN" and "2AGN" models have four and six parameters, respectively. An F-test shows that 8 of the 36 sources (22%) have good fits and a probability (P_ran) below 5% that the improvement due to the additional AGN component is spurious. These sources are shown in Table 1.¹⁴ The four sources with the highest probability of needing a secondary AGN component are shown, as examples, in Figure 1. The rest-frame UV optical emission is clearly consistent with that of a type 1 AGN. Because of their significant excess rest-frame UV/optical emission, we refer to these objects as Blue Excess Hot DOGs (BHDs).

Zoom In Zoom Out Reset image size

While P_ran = 5% could be considered a large enough probability to be wary of the F-test interpretation, we note that the constraints on the presence of the secondary AGN are much tighter, as the F-test cannot take into account the fact that the non-negative requirement for the combination of our templates provides additional constraints. In order to further assess the need for the secondary AGN component in the best-fit SED model of these objects, we have created 1000 realizations of their observed SEDs by re-sampling the photometry according to its uncertainties. We find that the secondary AGN component is needed (which we define as the best-fit secondary AGN component providing at least 2/3 of the flux in one or more of the SDSS bands) in 98% of the realizations. For comparison, objects with P_ran > 30% use the secondary AGN component in only 66% of the realizations.

Determining the fraction of Hot DOGs that show this blue excess is challenging given the selection effects. As discussed by Assef et al. (2015), the spectroscopic follow-up program has mostly focused on the brightest objects in W4, namely the 252 Hot DOG candidates with W4 < 7.2. Of the 36 objects in our sample, 22 fall in this category and six are selected as BHDs. However, there is a strong bias for BHDs to be detected by SDSS so we instead need to consider all objects within the SDSS area to assess their true fraction. There are 51 Hot DOGs with z > 1 and W4 < 7.2 within the SDSS DR12 area. As this sample is limited by W4, we posit that objects that are faint at observed optical wavelengths, and hence reddest in optical–W4, are unlikely to have blue excesses, and, in turn, be classified as BHDs. While this is likely a reasonable assumption, this is not guaranteed since the BHD selection criteria is based primarily on the UV/optical SED shape, and further observations are needed to properly assess this. Regardless, assuming that SDSS undetected Hot DOGs are not BHDs implies a true fraction of 6/51 (12%). However there is a natural bias toward a higher success rate in measuring redshifts for the optically brighter objects. Considering that 30% of objects did not successfully yield a spectroscopic redshift (Assef et al. 2015; P. R. M. Eisenhardt et al. 2016, in preparation) and that these are unlikely to be BHDs for the same reason as discussed above, this implies that the true fraction is closer to 6/73, or about 8% in the W4 < 7.2 sample. Extrapolating this to the entire Hot DOG sample at fainter W4 is not trivial due to the observational biases discussed. We note that only 2 BHDs are identified among Hot DOGs with W4 > 7.2, suggesting the true fraction of BHDs might be significantly lower when considering the entire population.

We suggest three potential scenarios to explain these BHDs.

• Leaked AGN Light: A single AGN is present in the system and is responsible for the IR emission by heating the dust that surrounds it in almost all directions. However, a small fraction of the rest-frame optical/UV emission of this luminous AGN is leaked out of the high-obscuration region. One possibility is that a fraction of the AGN light is scattered off into our line of sight. This effect has been observed in local Seyfert 2 galaxies through polarimetry (Antonucci & Miller 1985; Antonucci 1993), and the AGN powering the IR emission in Hot DOGs may be luminous enough for its scattered component to dominate the rest-frame UV/optical fluxes. Alternatively, a small opening between the dust clouds may allow us a partial direct line of sight toward the accretion disk. The opening would have to be small enough that only ∼1% of the accretion disk is directly visible. While in these scenarios one would naively expect to see other targets with partially obscured UV AGN emission, the SED selection method would not be able to identify them as it is only sensitive to those with the largest UV excesses.
• Dual Quasar: The secondary AGN emission component may come from an additional accreting SMBH in the system that is much less luminous than the highly obscured one powering the IR emission. Since major mergers are thought to play an important role in the evolution of massive galaxies and, in particular, in the triggering of intense, obscured AGN and star formation activity (see, e.g., Hopkins et al. 2008; Koss et al. 2011b), Hot DOGs could correspond to merger stages during which the SMBHs are not yet gravitationally bound (e.g., Comerford et al. 2009). The combination of a dual AGN with a luminous, obscured component and a less luminous but unobscured one has been previously observed in one candidate and one confirmed dual AGN, studied respectively by Barrows et al. (2012) and Fu et al. (2011), although the disparity between the component luminosities was much smaller than for BHDs. Because dust is likely abundant in Hot DOGs even on large scales, we expect the scenario in which only one nucleus is unobscured to be more likely if the nuclei are separated by scales ≳1 kpc, as this is the typical size of the dusty region in ULIRGs (Díaz-Santos et al. 2011).
• Extreme Star Formation: Alternatively, enhanced UV/optical emission could be produced by a young, massive starburst. A young starburst can have a broad-band power-law shaped SED, similar to that of an accretion disk, in the UV/optical regime (see, e.g., Leitherer et al. 1999). The SFR needed to power such a luminous emission through a young starburst would be very substantial. Such an intense starburst would be coeval with a hyper-luminous obscured quasar, with important implications for galaxy evolution scenarios.

Additional observations can help disentangle some of these scenarios. For example, light leaked from the hyper-luminous AGN through reflection on the dust grains could be directly probed through spectropolarimetry, as the characteristic broad emission lines of type 1 AGN should be more apparent in polarized light (Antonucci & Miller 1985; Antonucci 1993). A young coeval starburst could be excluded through optical variability, as this would only be expected for light from an accretion disk. Furthermore, since the typical amplitude of the variability for a given timescale correlates with the accretion disk luminosity (Vanden Berk et al. 2004; MacLeod et al. 2010), long-term variability monitoring could allow us to differentiate whether the excess blue emission light is due to light leaked from the hyper-luminous AGN or comes from a secondary, lower luminosity active SMBH. Also, high spatial resolution UV/optical imaging could identify the two nuclei of a dual AGN, or confirm the presence of a galaxy-wide massive young starburst. The latter might also be probed by high resolution ALMA observations of the far-IR continuum and CO emission.

X-ray observations can also help test these scenarios. Stern et al. (2014) reported on combined NuSTAR and XMM-Newton observations of three Hot DOGs without blue excesses, showing that they have heavily absorbed (possibly Compton-thick) X-ray emission. If BHDs are due to partial coverage or a secondary unobscured accreting SMBH, we would expect commensurate unabsorbed X-ray emission coming from the fainter AGN component. On the other hand, if the excess rest-frame UV/optical were due to scattered light from the hyper-luminous, obscured AGN or from extreme star formation, we would only expect to see strongly absorbed X-ray emission from the highly obscured AGN powering the infrared emission as X-ray photons are less scattered by dust grains or free electrons than UV ones. In principle very deep X-ray observations can also test the star formation scenario through a significant soft X-ray excess (see Mineo et al. 2014, and Section 5.3).

As mentioned in Section 1, one of the Hot DOGs that we have identified as BHDs, namely W0204–0506, is located in the NDWFS Cetus field, within an area observed by the Chandra X-ray Observatory ACIS-I instrument. In the following sections we study its X-ray-through-IR SED to gain more insight into its nature, which may also help provide insight into the nature of excess blue light from the other BHDs.

3.1. Observations

W0204–0506 was selected as a Hot DOG candidate through the criteria of Eisenhardt et al. (2012). These candidates (or "W12-Drops") are selected purely by their WISE magnitudes, corresponding to objects with W1 > 17.4 mag, high S/N detections in either W3 or W4, and with very red W2–W3 or W2–W4 colors (see Eisenhardt et al. 2012, for details). Table 2 shows the WISE fluxes for W0204–0506 from the AllWISE Data Release.

Table 2.  Photometry of W0204–0506

Band	Observed λ	Rest λ	Magnitudea	Flux	Telescope/Instrument	Ref
	(μm)	(μm)		(μJy)
g'	0.46	0.15	22.94 (0.05)	2.4 (0.1)	MMT/Megacam	(1)
r'	0.61	0.20	22.53 (0.06)	3.5 (0.2)	MMT/Megacam	(1)
i'	0.75	0.24	22.09 (0.09)	5.2 (0.4)	MMT/Megacam	(1)
z'	0.89	0.29	21.69 (0.06)	7.4 (0.4)	MMT/Megacam	(1)
u	0.35	0.13	23.00 (0.60)	2.2 (1.3)	SDSS	(2)
g	0.46	0.15	22.66 (0.17)	3.1 (0.5)	SDSS	(2)
r	0.61	0.20	22.49 (0.23)	3.6 (0.8)	SDSS	(2)
i	0.75	0.24	21.80 (0.18)	6.9 (1.2)	SDSS	(2)
z	0.89	0.29	22.03 (0.67)	4.3 (4.3)	SDSS	(2)
r	0.62	0.20	22.36 (0.17)	4.1 (0.6)	SOAR/SOI	(3)
J	1.2	0.40	20.77 (0.22)	8.0 (1.6)	WIYN/WHIRC	(4)
[3.6]	3.6	1.15	17.18 (0.05)	37.2 (1.9)	Spitzer/IRAC	(5)
[4.5]	4.5	1.45	16.34 (0.03)	52.2 (2.6)	Spitzer/IRAC	(5)
W1	3.4	1.06	17.34 (0.12)	35.4 (3.8)	WISE	(6)
W2	4.6	1.47	16.10 (0.16)	61.8 (9.0)	WISE	(6)
W3b	12	3.41	10.25 (0.06)	2714 (140)	WISE	(6)
W4b	22	7.07	7.07 (0.09)	11414 (946)	WISE	(6)
P70	70	22.58	⋯	71000 (3000)	Herschel/PACS	(7)
P160	160	54.84	⋯	113000 (5000)	Herschel/PACS	(7)
S250	250	80.65	⋯	46000 (11000)	Herschel/SPIRE	(7)
S350	350	112.90	⋯	27000 (11000)	Herschel/SPIRE	(7)
S500	500	161.29	⋯	17000 (15000)	Herschel/SPIRE	(7)

Notes.

^aMagnitudes are presented in their standard photometric system, namely AB for g'r'i'z' and ugriz, and Vega for the rest. For the SDSS bands we present asinh magnitudes, while we present Pogson magnitudes for all others. ^bAs discussed in Section 2, we have corrected the W3 and W4 fluxes according to the prescription suggested by Wright et al. (2010) for objects with red WISE colors.

References. (1) Finkelstein et al. (2007), (2) Ahn et al. (2014), (3) P. R. M. Eisenhardt et al. (2016, in preparation) (4) Assef et al. (2015), (5) Griffith et al. (2012), (6) Wright et al. (2010), (7) Tsai et al. (2015).

Download table as:  ASCIITypeset image

By selection, Hot DOGs have low S/N W1 and W2 fluxes. These wavelengths, however, provide significant information about the host, and hence are crucial for modeling the SEDs of these sources. To this end, we obtained Spitzer imaging of W0204–0506 in the [3.6] and [4.5] bands on UT 2011 February 28, as part of a comprehensive survey of the Hot DOG population (Griffith et al. 2012). We also obtained J-band imaging of W0204–0506 using the WHIRC camera on the WIYN telescope on the night of UT 2012 January 1 as part of the NIR follow-up campaign (Assef et al. 2015). Although optical photometry is available for this source from SDSS DR12 and the follow-up program mentioned in the previous section, we use instead deeper g'r'i'z' imaging obtained by Finkelstein et al. (2007) using the Multiple Mirror Telescope/Megacam in 2005 November and 2006 January. The fluxes of all these bands are listed in Table 2. The profile of this object is dominated by a central point source but shows an extended component that accounts for up to 30% of the integrated flux. We note that all three measurements of r-band photometry are consistent within the error-bars, so we cannot attempt to use optical variability to test the proposed scenarios (see Section 2). Table 2 also shows the fluxes of W0204–0506 in the Herschel/PACS 70 and 170 μm bands and in the Herschel/SPIRE 250, 350 and 500 μm bands, obtained as part of a Hot DOG follow-up program (PID: OT2_peisenha_2, PI: Eisenhardt; see Tsai et al. 2015, for details).

An optical spectrum of W0204–0506 was obtained with the GMOS-S spectrograph on the Gemini South telescope on UT 2011 November 27, using the 15 × 108'' longslit and the B600_G5323 disperser with a 2 × 600 s exposure time as part of an optical spectroscopic follow-up campaign of Hot DOGs (P. R. M. Eisenhardt et al. 2016, in preparation). Figure 2 shows the reduced spectrum of W0204–0506, which displays the C iv λ1549 Å, He ii λ1640 Å and C iii] λ1909 Å emission lines at a redshift of z = 2.100 ± 0.002. Each of the three emission lines is statistically well modeled by a single Gaussian component given the S/N of our observations. For C iv, He ii and C iii] we find that the best-fit Gaussian components have rest-frame FWHM of 1630 ± 220, 950 ± 200 and 550 ± 100 km s⁻¹ respectively. Note that the He ii emission line is comparable in strength to C iv and C iii], which is atypical of broad-lined quasars. For example, in the SDSS composite quasar spectrum of Vanden Berk et al. (2001), He ii has 2%–3% the strength of C iv and C iii] though the comparison is unfair since He ii is narrow and the carbon lines are dominated by broad emission in unobscured sources. For high luminosity narrow-line AGN such as radio galaxies, He ii is often comparable in strength to the carbon lines, as seen in the composite radio galaxy spectra of McCarthy (1993) and Stern et al. (1999). Furthermore, the FWHM we find for C iv and He ii are consistent with those measured in the composite spectrum of Stern et al. (1999, respectively 1540 and 1150 km s⁻¹), although C iii] is significantly narrower than the 1260 km s⁻¹ found by Stern et al. (1999). As another example of the similarities between Hot DOGs and radio galaxies, comparing Spitzer imaging of Hot DOGs to those of radio galaxies as reported by Wylezalek et al. (2013), Assef et al. (2015) showed that Hot DOGs reside in similarly overdense environments, suggestive of high-redshift proto-clusters in the process of formation.

Zoom In Zoom Out Reset image size

Combining the redshift with the Spitzer, WISE and Herschel observations we estimate an infrared luminosity of L_IR = 4.4 × 10¹³ L_⊙ using the approach of Tsai et al. (2015), which puts W0204–0506 well into the HyLIRG category. W0204–0506 is not detected by the VLA FIRST survey at 1.4 GHz, with a catalog detection limit of 1 mJy/beam. Assuming the object is a point source, this translates into a specific luminosity limit of ${L}_{\nu }(4.34\;\mathrm{GHz})\;\lt 9.7\times {10}^{31}\;\mathrm{erg}\;{{\rm{s}}}^{-1}\;{\mathrm{Hz}}^{-1}$ and a limiting flux density ratio between rest-frame 4.34 GHz and rest-frame 4400 Å of ${f}_{\nu }(4.34\;\mathrm{GHz})/{{\rm{f}}}_{\nu }(4400\;\mathring{\rm A} )\lt 1000$, neither of which is stringent enough to classify W0204–0506 as radio-loud nor confirm it as radio-quiet (Stern et al. 2000).

3.2. SED Modeling

The top panel of Figure 3 shows the best-fit "1AGN" model to the UV-through-mid-IR photometry of W0204–0506. The fit to the optical photometry is poor, as reflected by the large value of χ² = 46.07 with only four degrees of freedom for the fit (${\chi }_{\nu }^{2}=11.5$). The bottom panel of this figure shows the best-fit "2AGN" model, which provides a significantly better χ² value of 2.38 (${\chi }_{\nu }^{2}=1.2$). Comparing both models through an F-test yields P_Ran = 5 × 10⁻², implying that the addition of a secondary unobscured AGN component is justified (see discussion in Section 1). The best-fit "2AGN" model consists of a highly obscured luminous AGN that dominates the infrared luminosity with $E(B-V)=9.7\pm 1.2$ and a 6 μm luminosity of L_{6 μm} = 1.9 ± 0.2 × 10¹³ L_⊙. The rest-frame UV/optical is, on the other hand, dominated by a lightly reddened AGN component with $E(B-V)=0.13\pm 0.02$ and a significantly lower 6 μm luminosity of L_{6 μm} = 2.7 ± 0.5 × 10¹¹ L_⊙, of order 1% of the luminosity of the highly obscured AGN.

Zoom In Zoom Out Reset image size

SED modeling can also allow us to constrain the stellar mass of this object. Unfortunately, we do not have enough information to constrain the M/L ratio, so instead we use the approach of Assef et al. (2015) and estimate an upper bound. The maximal stellar mass estimate analysis for Hot DOGs done by Assef et al. (2015) yields ${M}_{*}^{{\rm{Max}}}=7\times {10}^{11}\;{M}_{\odot }$ for this object, assuming the "1AGN" model. The "2AGN" best-fit will yield a lower upper-bound on the stellar mass, as the rest-frame K-band luminosity is less dominated by the stellar emission, so we use the estimate of the "1AGN" model as it is the most conservative upper-bound.

While the "2AGN" model fits the data very well, we cannot immediately interpret this as proving the existence of the secondary unobscured AGN component in the SED of W0204–0506 expected for the dual AGN and reflection scenarios. Instead, the blue excess emission may be described by an extremely luminous and young starburst, as described in Section 2, which falls beyond the parameter space covered by the empirical host galaxy templates we considered. In the next section we study the deep X-ray observations available for this object to gain further insight into the nature of the unusual SED of W0204–0506.

W0204–0506 was serendipitously observed by the Chandra X-ray Observatory as part of the LALA Cetus field observations (PID: 04700805, PI:Malhotra) presented by Wang et al. (2007). The field was observed during Cycle 4 with the ACIS-I (Garmire et al. 2003) instrument for 160 ks starting on UT 2003 June 13 and then again for 15 ks more starting on UT 2003 June 15. For simplicity we only use the 160 ks observation. The Chandra observation was taken in the Timed Event mode, and we extracted spectra from the ACIS-I detector using the standard pipeline in CIAO v4.6.¹⁵ The source spectrum was obtained from a circular region of radius ∼2'', while the background was extracted from a larger nearby circular region that was free from any other contaminating sources. The source is detected with 94 counts and is visually consistent with a point source. Owing to the low count statistics, the source spectrum was only lightly grouped with a minimum of 1 count per bin. We therefore carry out the parameter estimation by minimizing the Cash statistic (Cash 1979), modified through the W-statistic provided by XSPEC¹⁶ to account for the subtracted background.

Figure 4 shows the unfolded spectrum from the Chandra observations. The emission is clearly hard, implying it is powered by a highly absorbed AGN, as expected from the multi-wavelength SED presented in the previous section. While the figure may show a tentative Fe Kα line, the counts are too low to determine whether the line is real or simply a statistical fluctuation. The figure also shows the best-fit absorbed AGN obtained using the models of Brightman & Nandra (2011). These models predict the X-ray spectrum as observed through an optically thick medium with a toroidal geometry, as posited by the AGN unified scheme. The models employ Monte-Carlo techniques to simulate the transfer of X-ray photons through the optically thick neutral medium, self-consistently including the effects of photoelectric absorption, Compton scattering and fluorescence from Fe K, among other elements. Treating these effects self-consistently rather than separately has the advantage of reducing the number of free parameters and of gaining constraints on the spectral parameters. It is therefore particularly useful for low count spectra such as the one we are fitting here.

Zoom In Zoom Out Reset image size

The best-fit model yields a photon-index of ${\rm{\Gamma }}={1.6}_{-0.0}^{+0.8}$, a neutral hydrogen column density of ${N}_{{\rm{H}}}={6.3}_{-2.1}^{+8.1}\times {10}^{23}\;{\mathrm{cm}}^{-2}$ and an absorption-corrected 2–10 keV luminosity of $\mathrm{log}{L}_{2-10\mathrm{keV}}/\mathrm{erg}\;{{\rm{s}}}^{-1}={44.9}_{-0.14}^{+0.86}$. The best-fit model has a Cash statistic of C = 66.08 for 77 degrees of freedom. The uncertainties quoted correspond to the 90% confidence interval. Note that the photon-index is poorly constrained by the data, so we require Γ ≥ 1.6 since lower values of the photon-index are only observed in SMBHs accreting at low Eddington rates (e.g., Shemmer et al. 2006, 2008; Risaliti et al. 2009; Brightman et al. 2013).

The properties derived for the luminous, highly absorbed AGN emission dominating the X-rays are in good agreement with those derived for the luminous, highly obscured AGN component that dominates the IR SED. Assuming the median dust-to-gas ratio observed by Maiolino et al. (2001) in AGN, namely $E(B-V)/{N}_{{\rm{H}}}=1.5\times {10}^{-23}\;{\mathrm{cm}}^{2}\;\mathrm{mag}$, the best-fit accretion disk obscuration found through the SED modeling for the highly obscured AGN corresponds to a neutral hydrogen column density of N_H = 6.5 ± 0.8 × 10²³ cm⁻², a value which is consistent with that measured from the X-ray spectrum. Similarly, we can compare the estimated luminosities, as the intrinsic L_{2–10 keV} luminosity of an AGN correlates well, although with considerable scatter, with its monochromatic luminosity at 6 μm, L_{6 μm} (e.g., Fiore et al. 2009; Gandhi et al. 2009; Bauer et al. 2010; Mateos et al. 2015; Stern 2015). These relations are typically linear (with the exception of Fiore et al. 2009, who used a broken power-law) and have been derived for limited energy regimes, but Stern (2015) has recently shown that a quadratic relation does a better job at describing the entire AGN luminosity range. In particular, this relation has been derived considering objects with luminosities as high as those of Hot DOGs, making it the most appropriate one to use in this context. From the relation presented by Stern (2015), adding its scatter of 0.37 dex to the uncertainty budget, we get that the best-fit L_{6 μm} for the highly obscured, luminous AGN component corresponds to an intrinsic X-ray luminosity of $\mathrm{log}{L}_{2-10\mathrm{keV}}/\mathrm{erg}\;{{\rm{s}}}^{-1}=45.36\pm 0.37$, which is consistent with the estimate from the X-ray spectrum. Figure 5 shows the confidence contours for log L_{2–10 keV} and N_H obtained from the X-ray spectrum, as well as the estimates based on the multi-wavelength SED described above. The latter are consistent with the X-ray spectrum estimates, although the agreement is better once the scatter of the L_{2–10 keV} – L_{6 μm} is taken into account. This consistency implies that the same AGN component is responsible for both the IR SED and the X-ray emission, as expected.

Zoom In Zoom Out Reset image size

In Section 2 we suggested three mechanisms to explain the blue excess component that dominates the rest-frame UV/optical SED of BHDs. For the presence of a secondary active SMBH with little or no obscuration, we would expect to see the corresponding unabsorbed/lightly absorbed X-ray emission. However, the X-ray spectrum in Figure 4 is strongly inconsistent with the presence of an additional AGN component with little or no absorption. If we use the L_{6 μm}–L_{2–10 keV} relation of Stern (2015), we find that the corresponding X-ray luminosity for the secondary AGN that best-fits the rest-frame UV/optical SED would be log L_{2–10 keV}/erg s⁻¹ = 44.32 ± 0.37. Assuming the median gas-to-dust ratio of Maiolino et al. (2001), the best-fit obscuration of $E(B-V)=0.13$ corresponds to a low absorption of N_H = 8.9 × 10²¹ cm⁻²; such a component would have significant power at soft energies. Figure 4 shows the expected X-ray signal from a power-law spectrum with the above luminosity and a photon-index Γ = 1.9 under this light amount of absorption in comparison to the observed X-ray spectrum, illustrating that the observations are highly inconsistent. If we allow for the luminosity of this component to be fit to the spectrum (keeping N_H and Γ fixed) simultaneously with the parameters for the heavily absorbed power-law described earlier, we find that with 90% confidence $\mathrm{log}{L}_{2-10\mathrm{keV}}/\mathrm{erg}\;{{\rm{s}}}^{-1}\lt 43.3$, over an order magnitude below the expectation from the luminosity of the component that dominates the rest-frame UV/optical SED. Figure 5 shows the confidence intervals for the luminosity of this component as well as the expectation from the UV/optical SED signal. The confidence intervals for the primary component are qualitatively unaffected by the additional unobscured component and hence are not shown. Fixing the X-ray luminosity of the component to the expected value of $\mathrm{log}{L}_{2-10\mathrm{keV}}/\mathrm{erg}\;{{\rm{s}}}^{-1}=44.32$ yields a Cash statistic of C=194.46. This implies a ΔC = 128.38 above the best-fit, which means we can rule this scenario out with >99.9% confidence.

As stated in Section 2, there are three scenarios that can naturally explain the SED of W0204–0506. Below we discuss each of them in light of the observations described in the previous sections.

5.1. Leaked AGN Light

5.2. Dual Quasar

5.3. Extreme Star Formation

It is possible that the UV/optical excess observed in BHDs is not due to accretion onto an SMBH, but to intense, unobscured star formation. Over the first ∼100 Myr, a pure starburst has a UV/optical SED rising strongly toward the blue, up to the Lyman break. This is similar to the SED of unobscured QSOs. Hence, it is possible to model the SED of W0204–0506 as a mildly obscured young starburst instead of as a mildly obscured AGN in Section 3. Detailed modeling is not feasible with the low number of photometric bands available for W0204–0506 in the rest-frame UV/optical, so we concentrate here in determining the lowest possible amount of star formation that would be needed to power its UV/optical SED. We use SED models generated with the Starburst99 v7.0.0 code (Leitherer et al. 1999, 2010, 2014; Vázquez & Leitherer 2005) in combination with the EzGal package of Mancone & Gonzalez (2012), and take the following approach.

For W0204–0506, J-band corresponds to rest-frame 4000 Å, so its blue ${z}^{\prime }-J$ color precludes the presence of a strong Balmer-break, implying a very young age for the starburst. We assume models with a constant SFR generated by Starburst99, although our results are qualitatively similar if we instead use a simple stellar population. We assume the latest Geneva models available for this version of Starburst99 (see Leitherer et al. 2014, for details). Since the ${g}^{\prime }-{r}^{\prime }$ color is significantly redder than would be expected for a young starburst, we also allow for obscuration to be fit to this component, assuming the same reddening-law as in Section 3. Assuming a reddening law with a strong 2175 Å feature, as is observed in the Milky Way (e.g., Cardelli et al. 1989), results in a poorer description of the SED. The large amount of dust present in the system is indicative of a considerable mean metallicity, so we assume a solar metallicity for our models, although we discuss the effects of this choice later in this section. We assume a Salpeter initial mass function (IMF; Salpeter 1955) with mass ranges between 0.08 M_⊙ and 120 M_⊙. For the rest of the input parameters of the Starburst99 code we assume the standard recommended options.

Figure 6 shows the best-fit SED to the observed UV/optical SED. When fitting the SED, we impose a minimum age of 1 Myr and find the best-fit model has χ² = 10.1, with an age of 1 Myr, $E(B-V)=0.20$ and $\mathrm{SFR}=5200\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$. Note that the fit is much poorer than for the AGN model, particularly as we only consider five data points. As we fit for the amplitude, reddening and age, this implies ${\chi }_{\nu }^{2}=10.1/2=5.05$. If we lift the minimum age requirement we find the best-fit SED requires an enormous SFR of approximately $50,000\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ but with an almost indistinguishable χ² = 9.9. This implies there is a strong degeneracy between age and SFR in our models, which is made clearer by the contours shown in Figure 7. Considering this degeneracy, we can determine that, with 90% confidence, $\mathrm{SFR}\gtrsim 1000\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$. While $\mathrm{SFR}\gt 1000\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ is routinely observed in SMGs and ULIRGS (e.g., Barger et al. 2014), it is always heavily dust obscured and hence only detected in the far-IR. Barger et al. (2014) shows, based on the measurements of van der Burg et al. (2010), that UV measured SFRs in Lyman break galaxies at z > 3 cut off at $\sim 300\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$, suggesting that an unobscured ${\mathrm{SFR}}_{{\rm{Min}}}\;\sim 1000\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ is very unlikely.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

A somewhat better fit to the data can be obtained by assuming a lower metallicity. If we consider the lowest possible metallicity provided for the Geneva models by Starburst99, namely Z = 0.001, we find that the best-fit SED has χ² = 7.8 (${\chi }_{\nu }^{2}=3.9$) and SFR = 5400 M_⊙ yr⁻¹. Yet, as shown in Figure 7, this significantly relaxes the requirement on the minimum SFR, implying only $\mathrm{SFR}\gtrsim 250\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ with 90% confidence. This would be consistent with the upper envelope of what has been found for Lyman break galaxies. However this model would require that the gas feeding the starburst has far lower metallicity than that feeding and surrounding the nucleus, so we do not consider it any further.

Note that we have not included the WISE W1 band flux to constrain these fits despite the fact that W1 would be dominated by the host galaxy emission if the blue-excess is due to star formation (see Figure 3). Adding the W1 flux as a constraint results in a much poorer fit by our constant SFR model (χ² = 28) that severely underestimates the observed W1 flux. This is not surprising, as it simply highlights the need for an older stellar population in the system. However, it is worth noting that adding an older stellar component would only make the best-fit SFR larger and the starburst younger, as it would imply an even stronger inverse Balmer-break for the starburst. Nevertheless, we expect the changes to be negligible based on the W1 amplitude and the large J-band flux uncertainty.

An independent constraint on the SFR can be obtained from Herschel/SPIRE observations of W0204–0506, as the cold dust emission from star-forming regions is a good tracer of their activity (e.g., Kennicutt 1998). As discussed by previous studies (Eisenhardt et al. 2012; Wu et al. 2012, 2014; Jones et al. 2014; Tsai et al. 2015), the far-IR SEDs of Hot DOGs are dominated by the hot dust emission powered by the AGN, and show no evidence for a significant cold component associated with star formation. Assuming that the Herschel/SPIRE fluxes of W0204–0506 are powered solely by star formation, we can thus place a conservative limit on the SFR. We fit the 250, 350 and 500 μm fluxes with a modified blackbody and then estimate the SFR limit using the ${L}_{{\rm{FIR}}}\mbox{--}\mathrm{SFR}$ relation of Kennicutt (1998), assuming a Salpeter IMF. We assume β = 1.5 and a dust temperature of T = 40 K for the modified blackbody, similar to what was done by Wu et al. (2012) to fit ground-based submm observations of Hot DOGs. This temperature is chosen as it is representative of the hottest dust emission associated with star formation (Magnelli et al. 2012); lower temperatures will result in lower SFR estimates. The best-fit modified blackbody, shown by the solid gray line in Figure 3, yields ${\mathrm{SFR}}_{{\rm{IR}}}^{{\rm{Lim}}}=2500\pm 500\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ if we assume the limit case where all luminosity of this component is powered by star formation. Using the 90% confidence level, shown by the dotted gray line in Figure 3, we can place the conservative limit of SFR_IR < 3350 M_⊙ yr⁻¹.

Considering the lower limit derived earlier from the optical SED of SFR_Min ≳ 1000 M_⊙ yr⁻¹, we conclude it is possible that the UV/optical SED is powered by star formation. However, we consider it less likely than the reflection scenario because (a) the SED fit is much poorer than with our 2AGN model, and (b) as discussed earlier, it would be a level of unobscured star formation that has not been observed before (Barger et al. 2014). If the UV/optical SED is dominated by star formation, the upper bound on the stellar mass of ${M}_{*}^{{\rm{Max}}}=7\times {10}^{11}\;{M}_{\odot }$ combined with the lower bound on SFR of 10³ M_⊙ yr⁻¹ implies a specific SFR > 1.4 × 10⁻⁹ yr⁻¹, i.e., that the host galaxy would be doubling its stellar mass in a timescale shorter than 700 Myr.

For completeness, we also estimate the maximum SFR allowed by the X-ray data in addition to the primary AGN component. We used the relation between SFR and integrated 0.5–8 keV X-ray luminosity outlined in Mineo et al. (2014): ${L}_{{\rm{X}}}/\mathrm{SFR}=4\times {10}^{39}\;(\mathrm{erg}\;{{\rm{s}}}^{-1})/({M}_{\odot }\;{\mathrm{yr}}^{-1})$. These authors found the total 0.5–8 keV X-ray emission from star formation was ∼2/3 from X-ray binaries (XRBs), and ∼1/3 from diffuse plasma emission from the ISM. For the plasma emission, we adopted a Mekal plasma model (Mewe et al. 1985) with a rest-frame temperature set to 0.25 keV (the average found by Mineo et al. 2012). Additionally, at the SFRs relevant here, we assumed that the integrated spectrum from the XRB would be dominated by ultraluminous X-ray sources (ULXs; Feng & Soria 2011, for a recent review), and adopted a simple model based on recent NuSTAR observations of ULXs (Bachetti et al. 2013; Walton et al. 2013, 2014, 2015a, 2015b; Mukherjee et al. 2015; Rana et al. 2015), approximating the spectrum with a cutoff powerlaw model with Γ = 1.5 and E_cut = 7 keV, scaled to z = 2.1. Both components are then modified by the neutral absorption column of $1.3\times {10}^{22}\;{\mathrm{cm}}^{-2}$ indicated by the optical emission (also assumed to be at z = 2.1). Setting the XRB and diffuse plasma contributions to 2/3 and 1/3 of the total SFR X-ray emission in the rest-frame 0.5–8 keV bandpass, respectively, and allowing the total emission to vary, we obtain a 90% upper limit of $\sim 5500\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$, somewhat less constraining than, but consistent with, the upper limit obtained from the Herschel observations. We also note for completeness that the star formation constraints from the FIRST radio observations are poorer than for the X-rays, as the upper bound of 1 mJy at observed 1.4 GHz flux implies ${\mathrm{SFR}}_{1.4\mathrm{GHz}}\lt 7600\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ according to the relation of Murphy et al. (2011) and assuming F_ν ∝ ν^−0.2.

We have introduced an interesting subsample of Hot DOGs that show UV/optical broad-band emission significantly in excess of what is typically expected for star formation. Based on UV through mid-IR photometry and spectroscopy for a large sample of objects, we find that BHDs constitute ∼8% of Hot DOGs, although this number is considerably uncertain due to the complex selection function.

We argue that the blue excess can most naturally be explained by three different scenarios, namely: (i) light leaked from the hyper-luminous, highly obscured AGN that dominates the IR emission, either by reflection off dust grains or free electrons, or by an opening between dust clouds allowing a direct line of sight to a fraction of the accretion disk; (ii) a second, less luminous and largely unobscured AGN in the system; or (iii) a young massive coeval starburst. While our current data does not allow us to generally differentiate between these scenarios in most objects, we argue additional observations can help disentangle them. In particular we note that the detection of rest-frame UV/optical variability could confirm the AGN nature of the blue excess, and that spectropolarimetry could confirm the reflection scenario. Similarly, high spatial resolution UV/optical imaging could identify the two nuclei of a dual AGN, or confirm the presence of a galaxy-wide massive young starburst. We also argue that X-ray observations would allow us to at least partially differentiate between these scenarios with some confidence. One of the BHDs found on this article, namely W0204–0506, was serendipitously imaged in the Chandra/ACIS-I 174.5 ks observations of the LALA field by Wang et al. (2007), and we use these data to analyze the different scenarios outlined above. With this depth, W0204–0506 is the Hot DOG with the best X-ray coverage to date.

We find that the X-ray spectrum of W0204–0506 is dominated by the higher energy emission. Of the 94 photons detected in the 0.3–8 keV energy range, 80 have energies between 2 and 8 keV. Using the models of Brightman & Nandra (2011), we find that the X-ray spectrum is well fit by an absorbed AGN with ${\rm{\Gamma }}={1.6}_{-0.0}^{+0.8}$, ${N}_{{\rm{H}}}={6.3}_{-2.1}^{+8.1}\times {10}^{23}\;{\mathrm{cm}}^{-2}$ and an intrinsic luminosity of $\mathrm{log}{L}_{2-10\mathrm{keV}}/\mathrm{erg}\;{{\rm{s}}}^{-1}\;={44.9}_{-0.14}^{+0.86}$. We show that the values of N_H and log L_{2–10 keV} are consistent with those expected from the IR properties of the AGN through correlations established in the literature.

Furthermore, we find that the ACIS-I observations strongly limit the contribution from a hypothetical secondary, lightly absorbed AGN, and show that a component consistent with the AGN derived from the UV/optical SED is ruled out with >99.9% confidence. Hence, we conclude that the excess blue emission in W0204–0506 is highly unlikely to be contributed by a secondary, less luminous AGN in the system. The lack of detection of a secondary component in the X-rays is consistent with the scenario of a single partially covered AGN, but leaked AGN light due to partial coverage implies an unlikely dust geometry, and we consider reflected light from a single AGN to be a more likely explanation. The X-ray observations are also consistent with an extreme, coeval starburst. Using observations from Herschel/SPIRE, we show that the IR emission puts a robust upper limit of SFR_IR < 3350 M_⊙ yr⁻¹, while the rest-frame UV/optical SED requires $\mathrm{SFR}\gtrsim 1000\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ to be powered by the star formation, showing that it is possible the UV/optical emission is powered by star formation. However, we consider this scenario less likely the than the reflection one because (a) the fit to the optical data is much worse than for the AGN model, and (b) the required unobscured SFR would be much larger than the highest observed in Lyman-break galaxies (Barger et al. 2014). Hence, the reflection scenario either by dust grains or free electrons is the most likely one to explain the nature of the blue excess in W0204–0506, although key details of the model and an in-depth analysis must be done to fully ascertain its likelihood. Further testing and constraints of this scenario can be obtained through deep spectropolarimetric observations to determine the polarization fraction of the UV/optical emission. Additionally, high-spatial resolution UV/optical imaging can offer insight into whether there is a strong galaxy-wide star-burst or if the excess blue emission is concentrated in the nucleus, as would be expected for the reflection scenario. A recently approved Chandra/HST observing program (PI: Assef, Proposal ID: 17700696) will obtain X-ray and multi-wavelength UV/optical imaging observations for two additional BHDs, as well as UV/optical imaging for W0204–0506, allowing us to probe this. These two additional BHDs are also being observed as part of an ALMA program aimed at studying the [C ii] and far-IR continuum of Hot DOGs (PI: Assef, Proposal ID:2013.1.00576.S and 2015.1.00612.S). The first results of this program are reported by Diaz-Santos et al. (2016). Additional, high spatial-resolution ALMA observations of the CO emission lines and longer wavelength far-IR continuum in these objects would determine the extension of the possible starburst. The combination of these observations will further probe the nature of these intriguing objects.

We thank Sangeeta Malhotra and James Rhoads for providing us with optical imaging data for the Cetus field. We also thank the anonymous referee for useful comments and suggestions. R.J.A. was supported by Gemini-CONICYT grant number 32120009 and FONDECYT grant number 1151408. FEB acknowledges support from CONICYT-Chile (Basal-CATA PFB-06/2007, FONDECYT 1141218, "EMBIGGEN" Anillo ACT1101), and the Ministry of Economy, Development, and Tourism's Millennium Science Initiative through grant IC120009, awarded to The Millennium Institute of Astrophysics, MAS. T.D.-S. acknowledges support from ALMA-CONYCIT project 31130005 and FONDECYT 1151239. This material is based upon work supported by the National Aeronautics and Space Administration under Proposal No. 13-ADAP13-0092 issued through the Astrophysics Data Analysis Program. The scientific results reported in this article are based to a significant degree on data obtained from the Chandra Data Archive. This publication makes use of data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. This work is based in part on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. The WIYN Observatory is a joint facility of the University of Wisconsin-Madison, Indiana University, Yale University, and the National Optical Astronomy Observatory. Funding for SDSS-III has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, and the U.S. Department of Energy Office of Science. The SDSS-III web site is http://www.sdss3.org/. SDSS-III is managed by the Astrophysical Research Consortium for the Participating Institutions of the SDSS-III Collaboration including the University of Arizona, the Brazilian Participation Group, Brookhaven National Laboratory, Carnegie Mellon University, University of Florida, the French Participation Group, the German Participation Group, Harvard University, the Instituto de Astrofisica de Canarias, the Michigan State/Notre Dame/JINA Participation Group, Johns Hopkins University, Lawrence Berkeley National Laboratory, Max Planck Institute for Astrophysics, Max Planck Institute for Extraterrestrial Physics, New Mexico State University, New York University, Ohio State University, Pennsylvania State University, University of Portsmouth, Princeton University, the Spanish Participation Group, University of Tokyo, University of Utah, Vanderbilt University, University of Virginia, University of Washington, and Yale University. Some of the observations reported here were obtained at the MMT Observatory, a joint facility of the Smithsonian Institution and the University of Arizona.