Limits to Rest-frame Ultraviolet Emission from Far-infrared-luminous z ≃ 6 Quasar Hosts

Guided by our initial experience with SDSS J1148+5251 (Mechtley et al. 2012), we determined that high-redshift quasars with weaker UV emission (M_1450Å > −26.5 mag) but secure submillimeter detections, i.e., with large rest-frame FIR to UV flux ratios (F_FIR/F_UV ≳ 100), are the best candidates for successful detection of host emission.

The rationale behind this selection is that a high FIR luminosity—and associated high star formation rate—coupled with a lower nuclear UV luminosity results in a less extreme nuclear-to-host contrast ratio and thus improved detectability of host UV emission. At the time of selection (February 2012), there were only five such z ≃ 6 quasars known that met these criteria: CFHQS J0033−0125, SDSS J0129−0035, SDSS J0203+0012, NDWFS J1425+3254, and SDSS J2054−0005 (see Figure 1). We note that while these quasars are UV-"faint" relative to the observed high-redshift quasar sample, they are still very luminous in the UV with −26.5 mag < M_1450Å < −23.9 mag.

Zoom In Zoom Out Reset image size

Although the FIR emission suggests the presence of significant dust in the host galaxies, the quasar discovery spectra (rest-frame UV) do not show anomalous features compared to the rest of the population—i.e., they are otherwise normal z ≃ 6 quasars, rather than showing significant spectral reddening or absorption features such as present in the FIRST/Two Micron All Sky Survey sample at lower redshifts (Urrutia et al. 2008; Glikman et al. 2015). Furthermore, more than ∼25% of z ≃ 6 quasars have similarly high FIR luminosities (Willott et al. 2007; Wang et al. 2008, 2010, 2011), so these FIR-luminous quasars are broadly representative of a significant subpopulation, rather than atypical objects.

In addition to the FIR-luminous quasars described above, we also analyze data from the prototype hot-dust-free quasar SDSS J0005–0006 (Fan et al. 2004; Jiang et al. 2010), which also lacks cold dust (Wang et al. 2008). With a lower luminosity and no evidence for significant dust content, this quasar was selected as a counterpoint to SDSS J1148+5251. This source is representative of a smaller but still important subpopulation. At 5.8 < z < 6.4, Jiang et al. (2010) found two apparently dust-free quasars in a sample of 21 quasars, or ≈10% of the population. Leipski et al. (2014) also found that ≈15% of their sample of 69 quasars at z > 5 are deficient in (but not devoid of) hot dust, and there is evidence of a trend toward higher dust-poor fraction with increasing redshift (Jun & Im 2013).

5.1.1. Magnitude Limits

The formal signal-to-noise ratio (S/N) of the residual flux in the core of each quasar after PSF subtraction is presented in Figure 5. This central region includes residual flux from the core of the quasar PSF, caused by an imperfect match of profiles of the quasar and empirical PSF used for subtraction, alongside any potential host galaxy flux. While the total flux is subtracted correctly, with a median residual consistent with zero, the pixel-to-pixel variation of the PSFs results in pixels with residual flux that is significantly larger than expected by the noise map, with formal signal-to-noise ratios of up to $\left|S/N\right|\simeq 30$. In other words, the subtraction technique produces considerable residuals in the inner region. We note that significant quasar over- or undersubtraction is unlikely, as this would produce negative or positive residuals in the diffraction spikes, respectively, which are not visible.

Zoom In Zoom Out Reset image size

To estimate the flux of the host without including this contaminated inner region, we instead measure the surface brightness in annuli from 7 to 10 pixels, or 042–060. We choose an inner radius of 7 pixels, as this is where the pixel-to-pixel variations of the S/N first reach the expected/background level. This approach ignores the central core, while including enough pixels to make a reasonable detection if any flux was present. For all quasars in both filters, no significant flux detection could be made in these annuli. The 2σ surface brightness limits in these annuli, from the noise of each image, are given in Table 2.

Table 2.  Quasar Host Galaxy Detection Limits

Quasar Name	SBJ, 2σ Limit (042–060)	mJ, 2σ Limit (Sérsic Fit)	SBH, 2σ Limit (042–060)	mH, 2σ Limit (Sérsic Fit)
	(AB Mag/''2)	(AB Mag)	(AB Mag/''2)	(AB Mag)
CFHQS J0033–0125	24.6	22.9	24.4	22.7
SDSS J0129–0035	24.5	22.8	24.3	22.6
SDSS J0203+0012	24.4	22.7	24.2	22.4
NDWFS J1425+3254	24.7	22.9	24.4	22.7
SDSS J2054–0005	24.6	22.9	24.4	22.7
SDSS J0005–0006	24.8	23.1	24.6	22.9

Note. Photometry of the residual image (after PSF subtraction) in the regions surrounding each quasar. The left columns for each filter give the surface brightness in the annulus 042–060 surrounding each quasar, as shown in Figure 5. As no significant signal is detected, these are 2σ upper limits calculated from the noise in each image. The right column for each filter gives magnitude limits estimated from these surface brightness limits. These are calculated by considering a range of Sérsic profiles with reasonable n and R_e (Shibuya et al. 2015) that are constrained to have the measured surface brightness in 042–060 and determining the most likely magnitude limit using a Monte Carlo approach (see the Appendix).

Download table as:  ASCIITypeset image

To obtain the total magnitude limit of a given host, we consider a range of Sérsic profiles, with a distribution of n and R_e guided by z ≃ 6 observations (Shibuya et al. 2015), and take a Monte Carlo approach to determine the most likely magnitude limit given the surface brightness limit in the annulus (see the Appendix). The 2σ magnitude limits obtained by this method are given in Table 2, with the J- and H-band limits in the range of 22.7–23.1 mag and 22.4–22.9 mag, respectively.

5.1.2. Stellar Mass Limits

Measuring the redshift evolution of the black hole–stellar mass relation is of key importance for understanding the coevolution of black holes and their host galaxies. Relative to the well-studied and accurately measured local relation (see, e.g., the review of Kormendy & Ho 2013), at higher redshifts observations suggest that black holes are more massive compared to their hosts. For example, at z ≃ 1.5 Ding et al. (2020) find a black hole–stellar mass ratio that is 2.7 times larger than the local relation, while at z ≃ 2 Peng et al. (2006) find a black hole–bulge mass relation 3–6 times larger. At higher redshifts, existing observations of luminous z ≃ 6 quasars with ALMA generally find ratios of black hole to dynamical mass that are significantly larger than the local relation (e.g., Maiolino et al. 2007; Riechers et al. 2008; Venemans et al. 2012; Wang et al. 2013). However, many studies claim that high observed relations are a result of selection effects (Lauer et al. 2007; Schulze & Wisotzki 2011, 2014; DeGraf et al. 2015; Willott et al. 2017; Ding et al. 2020). ALMA observations of lower-luminosity z ≃ 6 quasars indeed find these to lie on or below the local relation (Willott et al. 2017; Izumi et al. 2018, 2019).

To investigate the black hole–stellar mass relation using our HST observations, we convert our host magnitude limits to limits on stellar mass. We calculate UV slopes β of the hosts by fitting the relation $m=-2.5\mathrm{log}({\lambda }^{\beta +2})+{m}_{0}$, equivalent to f_λ ∝ λ^β, to the two host magnitude limits m_J and m_H at λ = 1.25 and 1.6 μm, respectively. Using this relation and the determined β and m₀, we calculate the UV apparent magnitude limit m_UV as that at rest-frame 1500 Å, or λ =(1 + z) × 0.15 μm. We convert this to an absolute magnitude using ${M}_{\mathrm{UV}}={m}_{\mathrm{UV}}-{DM}+2.5\mathrm{log}(1+z)$, where DM is the distance modulus.

We adopt the z = 6 M_*−M_UV relation derived by Song et al. (2016) using a large sample of galaxies from the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS)/Great Observatories Origins Deep Survey (GOODS) fields and the Hubble Ultra Deep Field (HUDF):

$$\begin{eqnarray}&&\mathrm{log}({M}_{* })=9.53\pm 0.02-(0.50\pm 0.03)\times ({M}_{\mathrm{UV}}+21),\end{eqnarray} \tag{ 1 }$$

with a scatter of 0.36 dex. We use a Monte Carlo technique, sampling from a uniform distribution of magnitudes ranging from M_UV = −20 mag to our 2σ upper limit for each quasar host. The lower luminosity limit of M_UV = −20 mag was chosen as this is as faint as high-redshift quasar hosts are expected to be from the BlueTides simulation (Feng et al. 2015; Marshall et al. 2019, see Figure 12). We assign a stellar mass to each sampled magnitude using Equation (1), given a normal distribution with σ = 0.36 dex, to determine the resulting probability distribution of stellar masses. These stellar masses are normally distributed, so we adopt the 2σ upper limit from this relation as our host mass limit. We note that this results in a lower, less pessimistic limit than simply taking the 2σ upper mass limit at the 2σ magnitude limit, which is very conservative.

We present the black hole–stellar mass relation from the stellar mass limits of our z ≃ 6 quasar hosts in Figure 6. Our limits are consistent with the black hole–stellar mass relation from existing submillimeter observations of z ≃ 6 quasars (Willott et al. 2017; Izumi et al. 2018, 2019; Pensabene et al. 2020), which measure the dynamical mass of the host using gas dynamics as probed by the [C ii] line, and assume M_* = M_dyn. Three of our quasars have dynamical mass measurements (Wang et al. 2010; Pensabene et al. 2020), which we also show in Figure 6. Our stellar mass upper limit for SDSS J2054–0005 lies above the measured dynamical mass of ${M}_{{\rm{dyn}}}\,={0.7}_{-0.3}^{+4.5}\times {10}^{10}{M}_{\odot }$ (Pensabene et al. 2020), suggesting that either our limit is significantly larger than the true stellar mass, with much deeper observations required to detect the underlying stellar emission, or the measured dynamical mass underestimates the total mass of the Galaxy. Our stellar mass upper limit for SDSS J0129–0035 is consistent with the lower dynamical mass limit of M_dyn > 7.8 × 10¹⁰M_⊙ (Pensabene et al. 2020). The lower limit on the dynamical mass for NDWFS J1425+3254 of M_dyn > 1.56 × 10¹¹M_⊙ (Wang et al. 2010) is larger than our stellar mass upper limit, suggesting that we are close to detecting the stellar component of this quasar host galaxy.

Zoom In Zoom Out Reset image size

The stellar mass limits of five of our six quasars are consistent with the local Kormendy & Ho (2013) M_* − M_BH relation. SDSS J0203+0012, however, has a stellar mass of M_* < 1.89 × 10¹¹M_⊙, lower than expected by the local relation, given its extremely large black hole mass of M_BH = 5.2 × 10¹⁰M_⊙ (Table 1; Shen et al. 2019). However, we note that this black hole mass is determined by the C iv line, as the more robust Mg ii line is not covered by ground-based observations. SDSS J0203+0012 is a broad absorption line (BAL) quasar (Mortlock et al. 2009), and so the dynamics probed by the C iv line are likely affected by the outflows. Hence, we place no significance on our black hole–stellar mass relation limit for this object.

The M_*−M_UV relation is derived from observations of UV-selected galaxies and might not apply if our hosts were dusty star-forming or quiescent galaxies; as these galaxies may be particularly dusty, our mass limits would be underestimates. Mid-infrared observations to allow for detailed SED fitting, using the upcoming James Webb Space Telescope (JWST), for example, are necessary to accurately determine the stellar masses of these potentially dusty host galaxies.

Figures 2–4 reveal that all of our six quasars have neighboring galaxies within the surrounding 65 × 65. For SDSS J0129–0035, NDWFS J1425+3254, and SDSS J0005–0006, some companions overlap with the quasar PSF, highlighting the need for the quasar PSF subtraction in order to fully understand the local quasar environment.

The properties of these 20 neighboring galaxies from the maximum posterior model found by psfMC are listed in Table 3. We calculate their UV magnitudes and slopes following the same procedure as for the host galaxies (see Section 5.1.2). The magnitudes and colors of these galaxies are displayed in Figure 7, along with samples of star-forming galaxies at z ≃ 6. Four of these companions have colors/UV slopes that are too red (β > 0) or too blue (β < −4) to be consistent with z ≃ 6 galaxies. In addition, seven galaxies are too bright to be likely at this redshift, with magnitudes brighter than M_UV = −22.1 mag, the magnitude of the brightest spectroscopically confirmed z ≃ 6 galaxy in the sample of Finkelstein et al. (2015). The remaining nine companion galaxies—surrounding five of our six quasars—have UV magnitudes and slopes consistent with those of star-forming galaxies at z ≃ 6. The majority of these are brighter than M* (−20.79 mag at z = 6; Finkelstein 2016; see Table 3). Unfortunately, existing observations at submillimeter to radio wavelengths do not resolve and/or detect the individual sources (e.g., Wang et al. 2013), so morphological comparisons with existing data are not possible.

Zoom In Zoom Out Reset image size

These nine potential companion galaxies are separated from the quasars by 14–32, corresponding to projected distances of 8.4–19.4 kpc. Simulations show that galaxies with companions at similar separations have higher active galactic nucleus (AGN) fractions (McAlpine et al. 2020) and also enhanced star formation rates (Patton et al. 2020). This suggests that these companions, if their true 3D distance is of order their projected distance, could be interacting with the quasar host galaxies and may potentially have triggered or enhanced the observed AGN activity. However, tidal features from any such interaction would likely be rendered invisible owing to the (1 + z)⁴ surface brightness dimming at z ≃ 6.

We examine the relationship between size, Sérsic index, and magnitude for these neighboring galaxies in Figure 8, in comparison to z ≃ 6 galaxies in the CANDELS GOODS-South sample (Brammer et al. 2012; van der Wel et al. 2012; Skelton et al. 2014; Momcheva et al. 2016). This shows that our companion galaxy sizes are as large as the largest z ≃ 6 CANDELS field galaxies but are larger than the median size of z ≃ 6 field galaxies by an average of 01. Their Sérsic profiles are as steep as the steepest Sérsic indexes of CANDELS z ≃ 6 field galaxies but are larger than the median Sérsic index of z ≃ 6 field galaxies by an average of 2. Thus, our potential quasar companion galaxies have morphological parameters that are consistent with the larger and steeper-Sérsic CANDELS z ≃ 6 field galaxies. Similar conclusions are made when comparing our potential companion galaxies to neighbors within 3'' of z ≃ 6 galaxies in the CANDELS GOODS-South sample (Figure 8, van der Wel et al. 2012), of which the majority (95%) are foreground objects at z < 5.5. Our potential companions have sizes and Sérsic indexes that are larger than the median of the neighbors of z ≃ 6 CANDELS galaxies, although their properties are reasonably consistent with the more massive neighbors. Hence, based on their size and Sérsic distributions, we cannot determine whether our observed neighbors are more likely to be z ≃ 6 galaxies than foreground interlopers.

Zoom In Zoom Out Reset image size

We show the relationship between size and UV absolute magnitude for these potential companion galaxies in Figure 9, assuming that they are at the same redshift as the quasar. Our objects that have UV magnitudes and slopes consistent with z ≃ 6 galaxies lie on a relatively tight size–luminosity relation. This relation is fairly consistent with, but somewhat higher than, that of z ≃ 6 Lyman break galaxies measured by Shibuya et al. (2015) (for galaxies with −22 mag ≲ M_UV ≲ −18 mag) and Kawamata et al. (2018) (for galaxies with −21.5 mag ≲ M_UV ≲ −12 mag), with our objects having larger sizes at the same luminosities.

Zoom In Zoom Out Reset image size

We note that the measured sizes of galaxies may be affected by systematic differences between these studies. For example, the treatment of the sky background in the fitting can have a significant impact on the resulting Sérsic fit parameters (see, e.g., Guo et al. 2009; Bruce et al. 2012). We include the background level as a free parameter in the MCMC fitting, which can result in larger values of R_e and Sérsic index n than if the background level is fixed (Bruce et al. 2012), potentially contributing to some of the discrepancy between our results and those of Shibuya et al. (2015), who assume a fixed background level. Bruce et al. (2012) show that there is a positive correlation between measured R_e and Sérsic index n. Shibuya et al. (2015) assume n = 1.5 for Lyman break galaxies, and Kawamata et al. (2018) fix n = 1. As our fitting method finds larger Sérsic indexes for our potential z ≃ 6 companion galaxies, with a mean of n = 4.3, this correlation may explain, at least in part, our larger measured sizes.

Using the number of galaxies observed in HST data of comparable depth (WFC3 ERS2 field, m_H < 26.5; Windhorst et al. 2011, see their Figure 12), the average number of galaxies observed is ≈373,000 per square degree, or 0.0288 objects per square arcsecond, if galaxies are uniformly distributed on the sky. We thus expect to find on average 7.3 random foreground objects within our six ≈65 × 65 quasar images (total area ≈250 square arcseconds; Figures 2–4), at z ≪ 6 and unrelated to our quasars. The surface density variations in these numbers due to foreground cosmic variance and photometric zero-point errors are expected to be ≲15% in the H band (see, e.g., Figure 3 of Driver et al. 2016), so we would on average expect ≲8.4 random foreground objects at z ≪ 6.

Assuming that the Galaxy neighbor distribution follows a Poisson distribution, the probability of observing a total of 20 galaxies within the ≈250 square arcsecond area, 4σ above the expected value of 8.4 foreground galaxies, is ≲0.0003. Hence, given the unlikelihood that so many galaxies would be found by chance, it is probable that some of these objects are physically associated with the quasars. Of the 20 close neighbors, 11 have UV magnitudes and slopes that suggest that they are unlikely to be at z ≃ 6. Finding these 11 foreground galaxies is consistent within 1σ of these Poisson expectations. The remaining nine neighbors have UV magnitudes and slopes consistent with known z ≃ 6 galaxies. We will adopt this number of nine potential quasar companion galaxies in our discussion below; however, we note that the true number of m_J < 26.5 z ≃ 6 companion galaxies in our images could be between 0 and 20, with 9 a more reasonable upper limit on the expected number based on the above arguments.

In Figure 10 we plot the average number of potential z ≃ 6 companion galaxies found in our quasar fields, compared to expectations for number counts of z = 6 galaxies in random fields (Finkelstein et al. 2015). We see significant excess compared to expectations of random pointings, with our average of ∼1.5 galaxies within 0.13 cMpc of the quasars (32 at z = 6) higher than the expected number of 0.0056 galaxies, corresponding to an overdensity of a factor of ∼270. This large excess is similar to the overdensity found by Decarli et al. (2017, see their Figure 3(b)), who detected four companion galaxies in [C ii] with ALMA around 4 of 25 z ≳ 6 quasars.

Zoom In Zoom Out Reset image size

Decarli et al. (2017) find that their measured overdensity is consistent with measurements of quasar–Lyman break galaxy clustering at z ≃ 4 (García-Vergara et al. 2017), when applied to the [C ii] luminosity function. We consider the models of the z ≃ 4 García-Vergara et al. (2017) quasar–galaxy clustering and the z = 5.9 galaxy–galaxy clustering of Qiu et al. (2018) to account for this effect, and we find that our measurement is consistent with both clustering models applied to the Finkelstein et al. (2015) luminosity function. Thus, as with the Decarli et al. (2017) sample, our observed potential z ≃ 6 companion counts can be explained by expectations for high-redshift galaxy clustering. Hence, while we find that our quasars are in environments that are denser than the average field density, with an overdensity of a factor of ∼270, they are in similar environments to that expected for a typical luminous z ≃ 6 field galaxy. In fact, if we have overestimated the number of true companion galaxies at 9, then our quasars may be in somewhat less dense regions than the typical luminous z ≃ 6 galaxy.

While Decarli et al. (2017) had a secure number of z ≃ 6 companions from ALMA [C ii] redshifts, we present a consistent upper limit from the number of possible z ≃ 6 companions following the above arguments. Clearly, JWST integral field redshifts, ALMA [C ii] redshifts, or VLT MUSE redshifts would be needed to determine the real number of companion galaxies around our quasars and their overdensity compared to the field at z ≃ 6.

5.2.1. Additional Observations of NDWFS J1425+3254

The quasar NDWFS J1425+3254 shows further evidence for having close companions. The discovery spectrum of Cool et al. (2006) shows a significant absorption feature at roughly 8350 Å, 20 Å redward of Lyα. This line could potentially be caused by H I absorption from a companion galaxy infalling at 720 km s⁻¹ (Mechtley 2014). By assuming that the system is virialized and that the companion is at a projected distance of 4.8 kpc, Mechtley (2014) find that this corresponds to a dynamical mass of ∼5.8 × 10¹¹M_⊙. Using the Song et al. (2016) M_* − M_UV relation (Equation (1)), from our detection limits the host of NDWFS J1425+3254 has a stellar mass of M_* < 2.0 × 10¹¹M_⊙, with the two companions having masses of ∼7.8 × 10⁸M_⊙ and ∼1.3 × 10⁹M_⊙. The properties of the CO (6−5) line (Wang et al. 2010) provide independent evidence for a group-like gravitational potential; the line fit gives an FWHM of 690 ± 180 km s⁻¹, and the peak of the emission is redshifted (z = 5.89) from the reported Lyα redshift (z = 5.85; Cool et al. 2006).

To further investigate this system, we obtained observations with the Large Binocular Camera (LBC) on the Large Binocular Telescope (LBT) in the g, r, and i bands (Figure 11). No LBC g-band flux is detected in a 20 aperture to a limit of m_g ≳ 28.3 mag, with Lyman–Werner flux from the quasar detected in the r band at m_r = 24.7 mag. Even in decent seeing conditions (≈08–10) the ground-based PSF of the quasar has broad wings that significantly affect the detection limit of close companions out to 20 or greater (see, e.g., Ashcraft et al. 2018). A best-effort point-source subtraction results in an upper limit of m_r ≳ 25.7 mag for the more distant of the two companions. The J- and H-band detections but faint g- and r-band limits are sufficient to exclude the possibility that these companions are blue foreground galaxies, but not that they could be red luminous galaxies at z ≃ 1.1. Additional observations are therefore required to confirm that these "companion" galaxies are indeed at z ≃ 6, and not foreground interlopers.

Zoom In Zoom Out Reset image size

To understand the magnitude limits and dust properties of our quasar host galaxies, we consider the sample of z = 7 quasars in the BlueTides simulation (Feng et al. 2015). BlueTides is a large-scale cosmological hydrodynamical simulation, which models the evolution of 2 × 7040³ particles in a cosmological box of volume (400/h cMpc)³ from initial conditions at z = 99 to z = 7, the lowest published redshift to date (Marshall et al. 2019; Ni et al. 2020). Figure 12 presents the relation between galaxy and quasar UV luminosity, for our observations and the BlueTides quasars. Both the intrinsic and dust-attenuated galaxy magnitudes for the BlueTides galaxies are shown, with the dust attenuation of galaxies modeled in BlueTides using the density of metals along a line of sight (for full details, see Marshall et al. 2019). From the BlueTides simulation, we find that hosts of M_UV < −23 mag quasars at z = 7 have between 1.4 and 3.8 mag extinction in the UV, with an average of A_UV = 2.6 mag, corresponding to A_V = 1.0 (Calzetti et al. 2000).

Zoom In Zoom Out Reset image size

From Figure 12, we see that the majority of intrinsic UV magnitudes for host galaxies of similar-luminosity quasars in BlueTides are brighter than our host galaxy upper limits, with only a small percentage consistent with our limits. Given that we make no detections of all six quasar hosts, our upper limits are sufficient to rule out the possibility that our quasars are generally hosted by dust-free galaxies. Instead, our limits favor host galaxies with significant dust attenuation, consistent with the $\langle {A}_{\mathrm{UV}}\rangle =2.6$ mag that is seen for the BlueTides quasar hosts (Figure 12, Marshall et al. 2019). If the BlueTides sample including dust obscuration is representative of the true z ≃ 6 quasar population, our upper limits are brighter than the host magnitudes that are expected, and future observations would need to probe at least ∼1 mag fainter to begin to detect the host emission. We will focus on integrating BlueTides with our observational techniques to make specific predictions for upcoming JWST observations in future work.

SDSS J0005–0006 was found to be a dust-poor quasar by Jiang et al. (2013), as it was undetected with the Spitzer Space Telescope at 15.6 and 24 μm. Further observations by Leipski et al. (2014) detected the quasar with Spitzer at these wavelengths; however, they did not detect emission at ≥100 μm with Herschel, and so they also conclude that the quasar is deficient of hot dust compared to the majority of quasars in their sample. Nondetection of the quasar with the Max Planck Millimeter Bolometer Array (MAMBO) at 250 GHz (Wang et al. 2008) results in upper limits of the dust mass of the host of M_dust < 1.9 × 10⁸M_⊙ (Calura et al. 2014). While these observations do not detect significant amounts of dust in this system, it is still possible for some dust to be present, resulting in low-level dust attenuation of the host galaxy. Thus, while our magnitude limit for the host of SDSS J0005–0006 is fainter than the magnitudes expected of quasar hosts with no dust attenuation from the BlueTides simulation, our nondetection of the host can be reasonably explained by some minor dust attenuation in the generally dust-deficient system. Additionally, if the simulation was run to z ≃ 6, we would expect to see a larger sample of luminous quasars, and potentially more with host luminosities fainter than our magnitude limits, which could explain our observations. Thus, the "dust-free" nature of SDSS J0005–0006 is not in significant tension with our overall dust predictions.

Our six quasars were selected in the z band as i-band dropouts with rest-frame UV luminosities at z ≃ 6 of −26.5 mag ≲ M_UV ≲ −24 mag (Table 1). Hence, their UV accretion disks are still, by selection, remarkably well visible. Five of our z ≃ 6 quasars were also selected to have significant FIR emission, and as a consequence the young stellar populations in their host galaxies are not visible in the best high dynamic range J- and H-band images that HST can produce.

We inferred that their host galaxies are likely considerably dusty ($\langle {A}_{\mathrm{UV}}\rangle =2.6$ mag), yet the embedded quasars are easily detected in the rest-frame UV and thus not significantly obscured (see, e.g., Vito et al. 2019). This must have significant consequences for the geometry of the small- and large-scale dust distribution. One possible explanation is that the embedded rapidly accreting supermassive black hole produced a significant outflow that vacated a sufficiently large cone on scales of 10–100 pc—fortuitously aligned in our direction—that the quasar has become clearly visible at rest-frame UV wavelengths, while the host galaxy is not. Significant outflows have indeed been observed from high-redshift quasars (e.g., Alexander et al. 2010; Nesvadba et al. 2011; Maiolino et al. 2012) and are expected to be able to carve a window for observing the quasar through otherwise high-density gas (Ni et al. 2020). Thus, it seems likely that such outflows are present in these systems. These objects are therefore high-priority targets for JWST, which will add 1.6–29 μm wavelength imaging coverage to our 1.2–1.6 μm HST images, and so is expected to much better constrain the dust extinction and geometry that UV images alone cannot capture.

Five z ≃ 6 quasars for this HST program were selected as those UV-faint quasars with confirmed submillimeter detections, and thus the greatest rest-frame L_FIR/L_UV ratios. This selects host systems with the greatest non-AGN contribution to the FIR flux, with inferred ultraluminous infrared galaxy (ULIRG) class FIR luminosities (>10¹² L_⊙) and implied star formation rates of ≈500M_⊙ yr⁻¹ (Wang et al. 2011). Locally, ULIRGs are gas-rich with high inferred star formation rates, and most are undergoing major mergers or at least strong interactions (e.g., Howell et al. 2010; Kim et al. 2013). This suggests that this sample of z ≃ 6 quasars are a distinct quasar subpopulation, which may be biased toward quasars with nearby interacting galaxies. This potential selection bias may mean that while our six quasars are in environments typical of luminous z ≃ 6 galaxies, the overall z ≃ 6 quasar population may reside in somewhat underdense environments. However, note that Trakhtenbrot et al. (2017b) observed three FIR-bright and three FIR-faint quasars with ALMA, finding spectroscopically confirmed submillimeter companion galaxies interacting with three quasars—one FIR-bright and two FIR-faint. We also find a potential companion around SDSS J0005–0006, which is not detected in the FIR (Wang et al. 2008; Jiang et al. 2013; Leipski et al. 2014). Hence, companion galaxies are not necessarily a feature of only FIR-bright quasars.

This selection bias is also likely to affect the measured black hole–stellar mass relation, as these quasars are not necessarily representative of the overall z ≃ 6 quasar population. For example, the most highly star-forming quasar host galaxies in BlueTides show a significantly steeper black hole–stellar mass relation, with such quasars lying on the main relation for the most massive black holes, and below the relation for lower-mass black holes. This result suggests that our ULIRG-type hosts, which are also selected to be UV-faint and thus potentially have lower-mass black holes, may lie below the black hole–stellar mass relation of the full quasar population.

The bias to selecting ULIRG-class host galaxies may also affect our stellar mass limits, as these galaxies generally lie significantly above the SFR–stellar mass main sequence. Their extreme star formation rates may indicate the presence of large amounts of dust extinction, as discussed in Section 6.1, which could further bias our measurements.

Companion galaxies have been discovered around a range of high-redshift quasars, with the majority seen only in observations at submillimeter wavelengths (e.g., Wagg et al. 2012; Decarli et al. 2017). For example, Trakhtenbrot et al. (2017a, 2017b) found companions physically associated with three of six z ≃ 4.8 quasars observed with ALMA, at separations of 14–45 kpc. Those companions have dynamical masses M_dyn = (2.1–10.7) × 10¹⁰M_*, compared with the quasar hosts that have M_dyn = (3.7–7.4) × 10¹⁰M_*, indicative of major galaxy interactions. These companion galaxies are not detected in Spitzer data, and so Trakhtenbrot et al. (2017b) conclude that there must be significant dust obscuration. This result may explain the lack of companions observed in rest-frame UV observations (e.g., Willott et al. 2005); this is supported by simulations (Marshall et al. 2019).

Our potential companion galaxies are detected in the rest-frame UV, suggesting that these companions may have less dust attenuation than those observed in the submillimeter. Other studies have also observed companions in the rest-frame UV; for example, McGreer et al. (2014) discovered a companion galaxy around both a z = 4.9 and a z = 6.25 quasar, at 5 and 12 kpc projected separations. While the companion of the z = 4.9 quasar is spectroscopically confirmed, the z = 6.25 companion is presumed to be at that redshift based on imaging in two HST filters, as in our study. While identifying these two companions, McGreer et al. (2014) reported that bright companions around high-redshift quasars are uncommon, with an incidence of ≲2/29 for ≳5L* galaxies and ≲1/6 for 2 ≲ L ≲ 5L* galaxies.

Finding quasar companion galaxies is consistent with the scenario that the growth of high-redshift quasars is triggered by galaxy mergers. While simulations predict that galaxy mergers can fuel quasar activity, it is unclear whether these are the dominant cause of high-redshift black hole growth. For example, using the EAGLE simulation, McAlpine et al. (2018) reported that at z = 0 ∼60% of black holes undergoing a rapid growth phase do so within ±0.5 dynamical times of a galaxy–galaxy merger, and McAlpine et al. (2020) found an overabundance of AGNs within merging systems relative to control samples of inactive or isolated galaxies. However, while galaxies experiencing mergers have two to three times higher accretion rates than isolated galaxies, the majority of black hole mass growth does not occur during the merger periods (McAlpine et al. 2020). Thus, if the potential companions are confirmed to be associated with our z ≃ 6 quasars, this result may be due to our biased sample of FIR-luminous quasars (see discussion in Section 6.2) and not necessarily indicative that nearby companions are common around high-redshift quasars.