NEAR-INFRARED IMAGING OF A z = 6.42 QUASAR HOST GALAXY WITH THE HUBBLE SPACE TELESCOPE WIDE FIELD CAMERA 3^*

Since their discovery by Schmidt (1963), quasars have been the best and most easily observed beacons to probe the distant universe. The unification model for active galactic nuclei (AGNs; e.g., Antonucci 1993 and references therein) views quasars as unobscured active nuclei, with a supermassive black hole (SMBH) as the central powerhouse behind the dominant nonstellar continuum.

Ground-based studies (e.g., Boroson & Oke 1982; McLeod & Rieke 1994) are limited by atmospheric seeing, and show that observing the underlying stellar populations is not trivial. Targett et al. (2012) have recently imaged z ≃ 4 quasar host galaxies using adaptive optics with the European Southern Observatory Very Large Telescope, but these required the best seeing conditions and targets with nearby (<40 arcsec) bright stars. Studies with the Hubble Space Telescope (HST) have concentrated on quasars at z ≃ 0.5–3, yielding important constraints on the morphology, luminosity, and stellar populations of quasar hosts (e.g., Disney et al. 1995; McLeod & Rieke 1995; Bahcall et al. 1997; McLure et al. 1999; Ridgway et al. 2001; Hutchings et al. 2002; Dunlop et al. 2003; Peng et al. 2006; Zakamska et al. 2006).

Goto et al. (2009) and Willott et al. (2011) used ground-based telescopes to detect extended Lyα emission out to 15 kpc around the z = 6.42 quasar CFHQS J232908−030158. This underscores the need to observe at longer, rest-frame near-ultraviolet wavelengths if emission from the host galaxy is to be differentiated from surrounding gas.

Imaging quasar host galaxies allows one to examine how mergers, starbursts, and other changes in environment affect the central nucleus. Observations of molecular gas and the far-infrared continua of z ≃ 6 quasars have suggested extremely high average star formation rates of ≳ 1000 M_☉ yr⁻¹ (Wang et al. 2010), implying potentially luminous hosts. As many quasars are hosted in massive galaxies, imaging z ≃ 6 quasar hosts may reveal the properties of the most massive first galaxies. The enormous SMBH mass (>10⁹ M_☉) associated with Sloan Digital Sky Survey (SDSS; York et al. 2000) quasars at z ≃ 6 represents a great theoretical challenge that heavily constrains possible formation methods and merger histories (Volonteri & Rees 2006; Li et al. 2007).

SDSS J114816.64+525150.3 (hereafter J1148+5251) is the best-studied member of the z ≃ 6 quasar population, having been extensively observed at multiple wavelengths since its discovery by Fan et al. (2003). Near-infrared spectroscopy by Willott et al. (2003) and Iwamuro et al. (2004) measured the Mg ii and Fe ii features, estimating a mass of 3 × 10⁹ M_☉ for the SMBH, an accretion rate near the Eddington limit, and an Fe ii/Mg ii ratio consistent with quasars at lower redshifts. Radio observations of CO lines (e.g., Walter et al. 2003; Riechers et al. 2009) indicate the presence of 2.2–2.4 × 10¹⁰ M_☉ of high-excitation molecular gas extending to ≃ 2.5 kpc (r = 045). Studies of the [C ii] line at 158 μm (Maiolino et al. 2005; Walter et al. 2009) provide evidence that the quasar host galaxy is undergoing a vigorous starburst, with an estimated star formation rate density of ≃ 1000 M_☉ yr⁻¹ kpc⁻² extending over kiloparsec scales. Studies of the continuum emission at far-infrared (FIR) wavelengths indicate a warm dust component with an AGN-corrected FIR luminosity of 9.2 × 10¹² L_☉ (Wang et al. 2010) and corresponding dust mass of 4.2–7.0 × 10⁸ M_☉ (Bertoldi et al. 2003; Robson et al. 2004; Beelen et al. 2006). Locally, this is in the range of ultraluminous infrared galaxies (ULIRGs; galaxies with L_FIR > 10¹² L_☉). Near- and mid-infrared Spitzer Space Telescope observations by Jiang et al. (2006) show clear evidence for prominent hot dust within the galaxy. All of these observations argue for a host galaxy with a significant stellar mass component undergoing an extreme episode of star formation.

In this Letter, we report on near-infrared imaging of J1148+5251 with the HST Wide Field Camera 3 (WFC3) and our methods for characterizing and subtracting the instrument and telescope point-spread function (PSF). In Section 2, we describe our near-infrared WFC3 imaging of J1148+5251 and a nearby star, used for PSF characterization. In Section 3, we describe our method for subtracting the central point source from the quasar images. We describe our simulations for assessing the reliability of our subtraction method in Section 4. Finally, in Section 5, we discuss the implications of these results and our plans for future investigation. For this Letter, we adopt a ΛCDM cosmology with H₀ = 70.3 km s⁻¹ Mpc⁻¹, Ω_M = 0.271, and Ω_Λ = 0.729 (Komatsu et al. 2011). Unless otherwise stated, all magnitudes use the AB system (Oke 1974) and have been corrected for Galactic extinction using the map of Schlegel et al. (1998).

The HST observations of J1148+5251 were performed on 2011 January 31 (HST Program ID 12332, PI: R. Windhorst) using the WFC3 IR channel with the F125W (Wide J) and F160W (WFC3 H) filters. Previous programs (e.g., Hutchings et al. 2002) found that the quality of empirical quasar point-source subtractions was significantly affected by PSF time variability. This variability is mostly due to so-called spacecraft breathing effects, i.e., thermally induced defocus of the HST secondary mirror due to movement of the Optical Telescope Assembly as the telescope goes into and out of Earth shadow (e.g., Hershey 1998). Examining these studies and instrument reports, we identified two primary sources of thermal variation that we expected to affect our observations—the spacecraft attitude and the orbital day–night cycle.

To minimize thermal variations due to spacecraft attitude, we constrained our PSF star to be within 5° of the target quasar, and required that the quasar and PSF star observations be collected in contiguous orbits. We further requested (and were granted) that our observations be scheduled immediately following another observation near the same celestial coordinates, thus minimizing thermal equilibration time in our first orbit. We selected the PSF star to match the quasar near-infrared colors ((J − H) = 0.55 Vega mag, (H − K') = 0.72 Vega mag) as closely as possible, to minimize differences in wavelength-dependent PSF features. Many stars with colors similar to the quasar also have SDSS spectra, since they were targeted by SDSS as high-redshift quasar candidates. We examined these spectra where available, to reject obvious spectroscopic binaries or other contaminants. After checking the resulting candidate stars for HST guide stars, we selected the star 2MASS J11552259+4937342 (spectral type K7, m_J = 16.272 ± 0.079 Vega mag, (J − H) = 0.488 ± 0.158 Vega mag, (H − Ks) = 0.651 ± 0.176 Vega mag; Skrutskie et al. 2006) as our final PSF target.

Unfortunately, there is no way to eliminate thermal variation due to the orbital day/night cycle. Thus, we constructed our orbits to ensure that we fully sampled this cycle, and that equal fractions of the final, combined quasar and PSF star images would come from any given location in orbital phase. Each quasar exposure was matched by several shorter PSF star exposures, taken at the same sub-pixel dither point at the same phase within the orbit. This phase-space sampling is summarized in the bottom panel of Figure 1. Any given detector location was never exposed beyond half-well depth in a single orbit, thus avoiding saturation or detector persistence. The quasar was observed in this pattern for three orbits in total, and the PSF star was observed for a single orbit.¹² The total exposure times for each object and filter combination are summarized in Table 1. Analysis was performed on Multidrizzle-combined images (Koekemoer et al. 2011) with an output pixel scale of 0065 to achieve Nyquist sampling of the PSF core in both filters, enabling accurate spatial shifting. The cosmic-ray rejection step of Multidrizzle was disabled, since the WFC3 IR MULTIACCUM readout mode provided sufficient cosmic ray rejection.

Zoom In Zoom Out Reset image size

To subtract the quasar point source, we used the program GalFit (Peng et al. 2010) to fit a PSF single-component model to the quasar image. We transformed the Multidrizzle-generated weight maps into uncertainty maps as in Dickinson et al. (2004), including the effects of correlated noise and shot noise from the quasar, and supplied these to GalFit as the pixel-to-pixel uncertainty ("sigma") image. We then subtracted the best-fit model from the original image and inspected the residual.

We first attempted this subtraction using the image of our PSF star as the model PSF. The results are shown in Figure 1. We measured the residuals using a 05 radius aperture, obtaining upper limits of m_J > 22.8 mag and m_H > 23.0 mag (2 σ). This includes the total noise contribution from both the quasar and the empirical PSF, measured by scaling the PSF uncertainty map by the same factor as in the fit, and adding it in quadrature to the quasar uncertainty map. The noise contribution from the subtracted PSF is comparable to that of the quasar since the two images have comparable signal-to-noise ratio (S/N; see Table 1), which leads to cosmetic defects (holes) in the subtraction, despite the net positive residual.

We also generated a TinyTim¹³ (Krist et al. 2011) model of the PSF, which we calibrated to the PSF star observations. A 5 × spatially oversampled TinyTim model was generated for each WFC3 exposure to allow for subpixel shifting. We used the spectrum of J1148+5251 obtained by Iwamuro et al. (2004) as our model spectrum. We used the HST Focus Model¹⁴ (Cox & Niemi 2011) to estimate the secondary mirror despace for each exposure, and included the field-dependent coma and astigmatism measurements built into TinyTim. The individual HST detectors have different mean focus offsets in the Focus Model, but the offset for the WFC3 IR channel has not been characterized. We therefore allowed the mean Z4 Zernike coefficient in TinyTim (R⁰₂ in the original formulation of Zernike 1934) to float as a free parameter in our optimization, adding this single mean value to the Focus Model estimate for each exposure. These models for individual exposures were then combined, weighted by exposure time, to produce a composite PSF for each Multidrizzle-combined science image. GalFit also accepts a pixel response convolution kernel for oversampled PSFs. We generated this quantity in a similar manner, by drizzling copies of the empirical WFC3 pixel response convolution kernel (modeling inter-pixel capacitance and jitter; see Hartig 2008) using the same shifts applied to the real images.

The result of the TinyTim PSF subtraction, with a significantly reduced noise floor compared to the direct subtraction, is shown in Figure 2. No host galaxy is detected, to a limiting surface brightness from r = 03 to 05 radius of μ_J > 23.5 and μ_H > 23.7 mag arcsec⁻² (2 σ). The inner 03 was excluded from the fit, as the best-fit TinyTim models produce PSF cores that are consistently narrower than those observed. Visible residuals (diffraction spikes and spots) are also seen when subtracting this model from the PSF star observations.

Zoom In Zoom Out Reset image size

Having established no host galaxy detection using the TinyTim PSF, we sought to quantify this subtraction method's ability to recover the host galaxy flux as a function of host galaxy parameters. To do so, we used GalFit to simulate a point source along with a Sérsic profile host galaxy, with total flux adding up to m_J = 19.1 mag, the measured flux in F125W. This simulated image contained no noise, so we added both shot noise from the object and a Gaussian noise field drizzled in the same manner as the real quasar image, to match its correlated noise properties. We then performed the same analysis that we used on the real quasar image, using GalFit to subtract a TinyTim-generated point source and measuring the surface brightness from r = 03 to 05 in the residual image.

We ran a grid of 256 models using this technique, varying the total integrated flux of the host galaxy from m_J = 20 to 26 mag, the effective radius from r_e = 01 to 09, and Sérsic indexes n = 1.0 and 4.0. The magnitude range represents host galaxies with luminosities from ≃ 1/2 to 1/500 of the total quasar luminosity. Fainter host galaxies than this are undetectable due to shot noise from the point source. The range in effective radius corresponds to r_e ≃ 0.6–5.0 kpc at z = 6.42.

Figure 3 summarizes the results of these simulations, plotting the measured surface brightness from r = 03 to 05 and contours representing the 1σ, 2σ, and 5σ detection limits. Inspecting the residuals of these model subtractions, we also found that bright (m_J < 22.5 mag), compact (r_e < 03) host galaxies would cause the method to significantly oversubtract the PSF. This would show negative residuals from the diffraction spikes, which are not seen in Figure 2.

Zoom In Zoom Out Reset image size

The model surface brightness from r = 03 to 05 reaches the 2σ upper limit of μ_J > 23.5 mag arcsec⁻² for a host galaxy of m_J > 22–23 mag, depending upon Sérsic index and effective radius.

We have performed point-source subtraction on the z = 6.42 quasar J1148+5251, with both empirical and modeled PSFs. Using direct subtraction, we measure an upper limit of m_J > 22.8 mag and m_H > 23.0 mag (2 σ). With the modeled PSF subtraction, we measure a limiting surface brightness measured from 03 to 05 of μ_J > 23.5 mag arcsec⁻² and μ_H > 23.7 mag arcsec⁻² (2 σ). Performing the same subtraction method on simulated quasars, we found that this surface brightness limit corresponds to a host galaxy of m_J > 22–23 mag, consistent with the direct subtraction limit.

Using the direct subtraction limits, the upper limits on the rest-frame monochromatic luminosity (λL_λ) at 1700 Å and 2200 Å are L₁₇₀₀ < 8.4 × 10¹¹ L_☉ and L₂₂₀₀ < 5.4 × 10¹¹ L_☉, assuming a flat spectrum within each band when applying the K-correction (Oke & Sandage 1968). This is comparable to the most luminous Lyman break galaxies at z ≃ 2–3 (Hoopes et al. 2007).

Using our upper limits and Equation (1) from Kennicutt (1998), which relates L_ν to star formation rate, we estimate a star formation rate of SFR < 210–250 M_☉ yr⁻¹. This estimate ignores dust attenuation and assumes a continuous star formation rate over 10⁸ years or longer. A younger population would decrease this upper limit, while dust would allow for a higher (absorption-corrected) rate. The star formation rate estimated from the AGN-corrected FIR luminosity by Wang et al. (2010) is 2380 M_☉ yr⁻¹. Since J1148+5251 would be classified as a ULIRG locally, this discrepancy is likely due to significant UV absorption by dust.

We can constrain the infrared excess (IRX) of the host galaxy, defined as the infrared to far-ultraviolet (FUV) luminosity ratio L_IR/L_FUV (e.g., Howell et al. 2010), usually expressed in logarithmic units. Using our upper limit for L₁₇₀₀ and an AGN-corrected infrared luminosity L_IR = 9.2 × 10¹² L_☉ (Wang et al. 2010), we calculate log (IRX) > 1.0, consistent with local luminous infrared galaxies (LIRGs) and ULIRGs (Howell et al. 2010), but greater than local starburst galaxies and high-redshift Lyman break galaxies (Overzier et al. 2011).

Figure 4 plots broadband measurements for J1148+5251, as well as our upper limits for the host galaxy flux at 1700 Å and 2200 Å and the AGN-corrected FIR measurements of Wang et al. (2010). Also plotted are the spectral energy distributions (SEDs) of four local galaxy systems—three LIRGs (Arp 220, IRAS 22491−1808, and IC 1623), representing the range in IRX from Howell et al. (2010), and the star-forming spiral NGC 4631, representing a galaxy with log (IRX) < 1.0, which is thus excluded as a potential host. Photometric points for the local galaxies are taken from NED¹⁵ and the SEDs have been normalized to match the AGN-corrected emission of J1148+5251 between 40 and 200 μm.

Zoom In Zoom Out Reset image size

We estimate A_FUV > 2.1 mag of UV absorption using the relation between the IRX flux ratio and A_FUV (e.g., Overzier et al. 2011, Equation (1)). Using the empirical relation IRX_{M99, inner} (A_FUV = 4.54 + 2.07β ± 0.4) from Overzier et al. (2011), we obtain a limit of β > −1.2 ± 0.2. This matches local (U)LIRGs (Howell et al. 2010), but is redder than almost all local star-forming galaxies and z ≃ 6 Lyman break galaxies (Overzier et al. 2011; Bouwens et al. 2012).

The TinyTim-based subtraction may be improved in the future with more accurate WFC3 IR PSFs. Since uncertainties introduced by the PSF model scale with PSF brightness, our further WFC3 observations will target quasars where the contrast ratio between point source and host galaxy is expected to be smaller, such as optically faint z ≃ 6 quasars with large FIR luminosities. The James Webb Space Telescope will enable us to use the PSF subtraction method at rest-frame ultraviolet and optical wavelengths with better-sampled empirical PSFs in a more stable thermal environment.

Special thanks are due to K. Long, J. MacKenty, and A. Roman at STScI for their advice and expertise when planning the observations, and the anonymous referee for very useful suggestions. We are grateful to J. Krist and C. Peng for the public availability of TinyTim and GalFit, respectively. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. Support for this HST program was provided by NASA through grant GO-12332.*.A from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA Contract NAS 5−26555. We owe a deep debt of gratitude to the entire WFC3 instrument team and the astronauts of STS-125 for this marvelous instrument on HST.