Draft version June 2, 2023

Typeset using LATEX twocolumn style in AASTeX62

A Candidate Dual QSO at Cosmic Noon

Eilat Glikman, Rachel Langgin,2, 3 Makoto A. Johnstone,1 Ilsang Yoon,4 Julia M. Comerford,5
1

Brooke D. Simmons,6 Hannah Stacey,7 Mark Lacy,4 and John M. O’Meara8, 9

1 Department of Physics, Middlebury College, Middlebury, VT 05753, USA
2 Department of Astrophysics, Haverford College, Haverford, PA 19041, USA
3 Bryn Mawr College, Bryn Mawr, PA 19010, USA
arXiv:2306.00068v1 [astro-ph.GA] 31 May 2023

4 National Radio Astronomy Observatory, 520 Edgemont Road, Charlottesville, VA 22903, USA
5 Department of Astrophysical and Planetary Sciences, University of Colorado, Boulder, CO 80309, USA
6 Department of Physics, Lancaster University, Bailrigg, Lancaster LA1 4YB, UK
7 Max Planck Institute for Astrophysics, Karl-Schwarzschild Str 1, D-85748 Garching bei München, Germany
8 Department of Physics, Saint Michael’s College, One Winooski Park, Colchester, VT, 05439, USA
9 W.M. Keck Observatory 65-1120 Mamalahoa Highway, Kamuela, HI 96743, USA

ABSTRACT
We report the discovery of a candidate dual QSO at z=1.889, a redshift that is in the era known as
“cosmic noon” where most of the Universe’s black hole and stellar mass growth occurred. The source
was identified in Hubble Space Telescope WFC3/IR images of a dust-reddened QSO that showed two
closely-separated point sources at a projected distance of 0.′′ 26, or 2.2 kpc. This red QSO was targeted
for imaging to explore whether red QSOs are hosted by merging galaxies. We subsequently obtained
a spatially-resolved STIS spectrum of the system, covering the visible spectral range, and verifying
the presence of two distinct QSO components. We also obtained high-resolution radio continuum
observations with the VLBA at 1.4 GHz (21-cm L band) and found two sources coincident with the
optical positions. The sources have similar black hole masses, bolometric luminosities, and radio
loudness parameters. However, their colors and reddenings differ significantly. The redder QSO has a
higher Eddington ratio, consistent with previous findings. We consider the possibility of gravitational
lensing and and find that it would require extreme and unlikely conditions. If confirmed as a bona-
fide dual QSO, this system would link dust-reddening to galaxy and supermassive black hole mergers,
opening up a new population in which to search for samples of dual AGN.

Keywords: Quasars (1319), Double quasars (406)

1. INTRODUCTION et al. 1988). It is expected, therefore, that at some

The next generation gravitational wave experiment, point during this process both SMBHs will be simul-
LISA, will detect the signal from the coalescence of su- taneously active and therefore discoverable as a pair of
permassive black holes (SMBHs) in the 105 − 107 M⊙ active galactic nuclei (AGNs).
range. Since every large galaxy hosts a nuclear SMBH, While theoretical investigations into the physics of
understanding the black hole merger process into the SMBH binaries (e.g., mass ratio, coalescence time scale,
supermassive regime is essential for a full picture of AGN activity) have been making steady progress, ob-
galaxy evolution. Galaxy mergers have also been in- servational constraints are still lacking due to the small
voked to explain the many scaling relations seen be- numbers of confirmed dual AGNs. These simulations do
tween galaxies and their nuclear supermassive black find that late-stage major mergers are the most likely
holes (SMBHs) suggesting a co-evolution between the to produce dual AGNs (i.e., separations ≤ 10 kpc; Van
two systems (Magorrian et al. 1998; Gebhardt et al. Wassenhove et al. 2012; Blecha et al. 2013; Steinborn
2000; Marconi & Hunt 2003). In addition, gas-rich (i.e., et al. 2016) suggesting that those are the best systems
“wet”) mergers are understood to trigger the most lumi- in which to search.
nous QSOs through the funneling of gas and dust into Dust-reddened (or, red) QSOs, represent a short-lived
to the nucleus fueling accretion onto the SMBHs, which phase of QSO evolution driven by the “wet” merger sce-
are also being brought together by the merger (Sanders nario described above. During such a merger, much of
2

the black hole growth occurs in a heavily enshrouded cused on those same redshifts (13 and 11 objects, respec-
environment followed by a relatively brief transitional tively), using ACS and WFC3/IR, respectively (PID
phase in which the obscuring dust is cleared by outflows 14706, PI Glikman). The images were observed with
and radiation-driven winds and is seen as a moderately a four-point box dither pattern and were reduced us-
reddened, Type 1, luminous QSO. After feedback pro- ing the Astrodrizzle package with a final pixel scale of
cesses clear the dust, the canonical blue QSO shines 0.′′ 06. One source, J122016.9+112627.092 , appeared as
through and dominates (Sanders et al. 1988; Hopkins two closely separated point sources (left and middle pan-
et al. 2005, 2008). Objects in the transitional phase, i.e, els of Figure 1) in both the F105W and F160W filters.
moderately obscured, red QSOs, are farther along the The WFC3/IR observations were designed to be identi-
merger timeline, and are thus ideal systems for finding cal to those in Glikman et al. (2015), with W2M J1220
dual AGNs. having exposure times of 797 s and 1597 s and reaching
Samples of red quasars1 have been identified through 3σ surface brightness limits of 23.67 mag arcsec−2 and
radio plus near-infrared selection (e.g., the FIRST- 23.87 mag arcsec−2 in F105W and F160W, respectively.
2MASS, or F2M, red quasar survey; Glikman et al. From the ground, this source appears as a single object
2004, 2007, 2012) and, more recently, mid- plus near- with r = 18.13 in the Sloan Digital Sky Survey (SDSS)
infrared selection (e.g., the WISE-2MASS, or W2M, red and H = 16.01 in 2MASS.
QSO survey; Glikman et al. 2018, 2022). These red QSO This source possesses an optical spectrum in SDSS and
samples span a broad range of redshifts (0.1 < z < 3) was assigned a redshift of z = 1.871 (see §3 for details
and reddenings (0.2 ≲ E(B − V ) ≲ 1.5); have very on the corrected redshift), shown in the right panel of
high accretion rates (L/LEdd > 0.1; Kim et al. 2015), Figure 1. The spectrum is well-fit by a QSO composite
sufficient to blow out the obscuring material (Glikman spectrum, constructed by combining the UV template
2017). Their spectra often show broad absorption lines of Telfer et al. (2002) with the optical-to-near-infrared
(BALs) that are associated with outflows and feedback template from Glikman et al. (2006) and reddened with
(Urrutia et al. 2009). Crucially, Hubble Space Telescope the SMC dust law of Gordon & Clayton (1998), by
(HST) imaging at z ≃ 0.7 and z ≃ 2 reveals that ≳ 80% E(B − V ) = 0.246. We also obtained a near-infrared
of F2M red quasars are hosted by merging galaxies (Ur- spectrum with the TripleSpec near-infrared spectro-
rutia et al. 2008; Glikman et al. 2015) making them graph (Wilson et al. 2004) on the 200-inch Hale tele-
more likely to harbor dual AGNs (or, more luminous, scope at the Palomar Observatory, also shown in Fig-
dual QSOs). ure 1. The Balmer lines are shifted into the atmo-
In this paper, we present the discovery of a candi- spheric absorption bands and cannot be studied from
date dual QSO in HST imaging of a sample of W2M red the ground. Due to the seeing-limited resolution of ∼ 1′′ ,
QSOs from Glikman et al. (2022). The QSO’s redshift this optical-through-near-infrared spectrum represents
of z ∼ 1.9 probes the epoch of peak AGN and star for- the combined light of both components and is therefore
mation in the universe. Throughout this work we quote not well-fit by a single reddened QSO across the full
magnitudes on the AB system, unless explicitly stated wavelength range.
otherwise. When analyzing spectra for extinction prop- W2M J1220 is also detected in the FIRST survey
erties, we first correct them for Galactic extinction, us- with an integrated flux density of Fint,20cm = 2.33 mJy
ing the Fitzpatrick (1999) extinction curve. When com- (Fpk,20cm = 1.50 mJy/beam), which corresponds to a
puting luminosities and any other cosmology-dependent total radio power of P1.4GHz = 5.7 × 1025 W Hz−1 . Ta-
quantities, we use the ΛCDM concordance cosmology: ble 1 lists the optical through near-infrared photometry
H0 = 70 km s−1 Mpc−1 , ΩM = 0.30, and ΩΛ = 0.70. for this source.
2.1. Morphological modeling
2. SOURCE CHARACTERISTICS
To determine the separation of the two sources and
A Cycle 24 HST program imaged the host galaxies of measure their respective magnitudes we modeled the
11 W2M red QSOs (5 QSOs at z ∼ 0.7 and 6 QSOs WFC3/IR images in both filters using Galfit (Peng et al.
z ∼ 2) to compare with the F2M imaging studies of 2002). We used a point spread function (PSF) con-
Urrutia et al. (2008) and Glikman et al. (2015) that fo- structed by combining a few dozen bright stars in each
HST filter from observations that used the same dither
1 In this letter, we adopt the canonical nomenclature that distin- pattern following the same procedure as in Glikman
guishes quasars, radio-detected luminous AGN whose radio emis-
sion is essential to their selection, from QSOs, the overall class of
luminous AGN. 2 Hereafter, W2M J1220
 3

Figure 1. Left – Color combined WFC3/IR image showing the presence of two closely-separated central peaks. The red
layer is the F160W image, the green layer is an average of the F160W and F105W images, and the blue layer is the F105W
image. Middle – Surface plot of the image counts in the F160W image where two distinct sources are visible. Right – Optical
through near-infrared spectrum of W2M1220+1126 (black line). A reddened QSO template, made out of the UV composite
QSO template of Telfer et al. (2002) combined with the optical-to-near-infrared composite spectrum from Glikman et al. (2006),
with E(B − V ) = 0.246 is overplotted with a red line and an unreddened QSO template is shown in blue. We see that the
Balmer lines are shifted into the atmospheric absorption bands. The STIS G750L transmission curve used in this work is shown
with a gray dot-dash curve.
Table 1. Photometric prop- the positions of the F160W and F105W components to
erties of F2M J1220 each other resulted in poorer fits and yielded residuals
with strong negative/positive flux asymmetries.
Band AB mag
We first fit a model consisting of two PSFs and a back-
g 19.07±0.01 ground sky component. While both sources are consis-
r 18.130± 0.008 tent with a point spread function, the residual image
i 17.452±0.007 showed excess flux in need of additional model com-
z 17.12±0.01 ponents. We added a Sérsic component to the model
J 16.44±0.07 resulting in a better fit, verified by an F-test whose
H 16.01±0.06 probability was consistent with 0, strongly suggesting
K 15.67±0.06 that we can reject the null hypothesis. However, the
Northern Componenta best-fit Sérsic component, situated next to the southern
F105W 18.020±0.003 PSF, had an effective radius, Re , of 0.06 pixels, which
F160W 16.9739±0.0002 is not physically meaningful. Although the addition of
Southern Componenta this Sérsic component improved the fit statistic and ac-
F105W 17.411±0.002 counted for flux not captured by the PSFs, it is unclear
F160W 16.8656±0.0004
how much of this added component was accounting for
PSF mismatches3 . Because the added Sérsic did not
a The magnitudes reported
model the extended emission seen in the residual image
from the HST images are to the east of the two PSFs, we added a second Sérsic
based on the PSF compo-
component, which does account for this excess flux and
nents determined by Galfit
as described in §2.1. whose inclusion is supported by an F-test with proba-
bility consistent with 0.
The best-fit Galfit model therefore is composed of two
PSFs and two Sérsic components. The locations of the
PSFs, in both filters, indicate a projected separation of
et al. (2015) using stars whose images were were ob- 0.2680 ± 0.0003′′ , which corresponds to ∼ 2.2 kpc at the
tained within 12 months of W2M 1220. The stars used QSO’s redshift. Figure 2 shows the residual images from
to construct the PSF were chosen to lie in the central this fit where different model components have been sub-
region of the WFC3/IR detector to minimize distortion
effects. All archival observations were re-reduced us-
3 Extensive investigation into the PSF subtraction did not re-
ing the same Astrodrizzle parameters as W2M 1220.
veal any systematic effects when fit to archival point sources lo-
When fitting, all parameters were allowed to be free in
cated at the same pixel position as W2M 1220. However, we find
both filter images, which produced the best fits (i.e., that the PSF is not able to capture all the flux from very bright
smallest χ2ν ) and cleanest residuals. Attempts at fixing point sources resulting in significant, yet symmetric residuals.
4

tracted from the data. We mark in the rightmost panel in the WFC3/IR image is apparent in the spectrum as
the positions of the four model components. Table 2 lists well.
the best-fit parameters for this model, noting that the
2.3. VLBA Imaging
first Sérsic component may not represent a physically
meaningful model. W2M J1220 is detected in the FIRST catalog (Becker
et al. 2003) with a 20 cm integrated flux density of 2.33
mJy. The peak flux density is 1.50 mJy/beam with an
2.2. HST follow-up with STIS
rms of 0.146 mJy/beam. The source’s deconvolved ma-
We obtained STIS spatially-resolved spectroscopy in jor and minor axes are 5.′′ 57 and 2.′′ 26, respectively, in-
the G750L mode covering a wavelength range from 5240- dicating that the image is slightly resolved4 .
10270 Å (dot-dash line in Figure 1; Cycle 29, PID Aiming to detect two distinct radio components at
16794). The 52′′ × 0.′′ 2 slit was oriented at a po- their optical positions, we obtained 257 minutes of on-
sition angle of 177.286◦ to capture both components source integration with the Very Long Baseline Array
in a single observation. The STIS CCD has a plate (VLBA) split into two equal-length dual polarization ob-
scale of 0.05078”/pixel such that the two components servations on 19 August 2021 and 03 December 2021 in
are separated by ∼ 5 pixels. The standard STIS the L-band (1.4 GHz or 20 cm). We used J1218+1105
reduction pipeline was used to remove detector sig- as a phase calibrator, which we measure to have an in-
natures and defringe the spectra using the STIStools tegrated flux density of 0.177 Jy in L-band located only
defringe.defringe command to remove the fringing 0.58 deg away from our target.
pattern. The observations were flagged, calibrated, cleaned,
We use the x1d routine to extract each spectrum, ad- and imaged with the Common Astronomy Software Ap-
justing the parameters to minimize overlap between the plications (CASA; CASA Team et al. 2022) package Ver-
two. We constrain the search region for finding a peak sion 6.5, following the approach described in VLBA Sci-
in the extraction profile by setting MAXSRCH to 1.5 for ence Memo #38 (Linford 2022). Fort Davis (FD) was
each source and A2CENTER to 506 and 511 pixels, respec- selected as the reference antenna. To ensure that the
tively. We set the extraction box size to 3 pixels and use amplitude scaling accounted for the wide bandpass, the
a 10 pixel offset from the peak, which is far from both task ACCOR was run twice – first for the initial calibra-
source profiles, for the background subtraction region. tions and again after the bandpass correction. A phase-
Two distinct spectra were extracted at 511.383 pixels referenced (Stokes I) image of the target was produced
and 506.324 pixels, respectively. by applying the TCLEAN task with natural weighting.
To evaluate the the impact of blending on our spectral The final calibrated image spans 320 x 0.001” along each
extraction, we sum the cosmic-ray cleaned, normalized, axis with an rms noise level of 0.017 mJy and is shown
and defringed science spectrum along the wavelength in Figure 5.
axis and plot the spatial profile of the two spectra in Fig- The calibrated VLBA image shows two distinct point
ure 3. We fit two Gaussian distributions to this summed sources oriented at a position angle of 172.205◦ with a
profile, keeping the width of the Gaussians tied to each separation of 0.′′ 26 (2.2 kpc). We performed 2D Gaus-
other, and fixing the centers to the positions found by sian fitting on each source with CASAviewer and found
x1d. The extracted region for each spectrum is shown in that the northern source has an integrated flux density
light and dark shaded pink. Using the best-fit σ of 1.04 of 0.502 ± 0.066 mJy and a peak flux density of 0.165
pixels, we calculate that the southern spectrum overlaps ± 0.016 mJy/beam. The deconvolved major and mi-
the northern spectrum by ∼ 0.6%, ensuring that the in- nor axis are 0.′′ 0234 and 0.′′ 0087. The southern source
dividual spectra are not contaminated by blending. We has an integrated flux density of 0.330 ± 0.044 mJy, a
also determine that the 3-pixel aperture loses 14.8% of peak flux density of 0.146 ± 0.014 mJy/beam, and de-
the total flux. We correct the fluxes of our spectra by convolved major and minor axes of 0.′′ 0166 and 0.′′ 0072.
this amount. In both sources, the major axis is slightly larger than
Figure 4, left, shows the resultant extracted spectra the CLEAN beam which has FWHM of 0.′′ 01 along both
for both QSO components. The Mg II line that is seen axes.
in the SDSS spectrum (Fig. 1) is visible in both the We overlay in contours the HST WFC3/IR F160W
southern and northern components at 2800Å. However, fluxes. The two VLBA point sources and their position
the C III] line at ∼ 2000Å is only seen in the southern angles are consistent with the HST position, though out-
component, where the signal-to-noise ratio is sufficiently
high. The two spectra have different continuum shapes
and the redder color of the northern component seen
4 The FIRST survey has an angular resolution of 5′′ .
 5

Table 2. Galfit parameters

Component F160W (χ2ν = 58.2) F105W (χ2ν = 28.8)

R.A.a Decl. a
Mag n Re R.A.a Decl.a Mag n Re

(J2000) (J2000) (mag) (kpc) (J2000) (J2000) (mag) (kpc)

North PSF +0.8800 +0.2805 16.97 ... ... +0.8783 +0.2875 18.02 ... ...
South PSF +0.8785 +0.0097 16.87 ... ... +0.8781 +0.0232 17.41 ... ...
Sérsic (central) +0.8922 +0.0668 20.31 1.32 ± 6.21 0.03 ± 0.23 +0.8926 +0.0471 21.00 1.76b 0.02b
Sérsic (eastern) +0.9379 +0.1756 20.89 1.94b 1.5b +0.9383 +0.1469 20.31 1.50 ± 0.37 1.01 ± 0.04
a The positions reported are shifts in seconds with respect to R.A. 12:20:16 and shifts in arcseconds with respect to Decl. +11:26:28.
b These parameters are flagged by Galfit as being outside the range of acceptable values. However, the fit resulted in an acceptable
χ2ν enabling a capture of the residual flux in the components. We do not report errors for these parameters.

Figure 2. Color combined residual images from the best-fit Galfit model, as described in Table 2. Left – HST WFC3/IR image
with just the two PSF components subtracted. Middle – Both PSF components and central Sérsic component subtracted; bright
extended emission is seen to the east. Right – Full residual with all model components subtracted. In this frame, the best-fit
model parameters are marked. White circles are at the PSF positions. The cyan cross is the central Sérsic parameter located
slightly to the east of the southern PSF. The cyan circle is the position of the Sérsic component that best-fits the extended
emission farther to the east.

side the 0.′′ 03 Gaia-based astrometric errors, confirming correct the observed F160W photometry, corresponding
their associations as the two components of W2M J1220. to rest-fame 5320Å, by these extinction values and, ap-
plying a bolometric correction of 9.2 (Richards et al.
3. RESULTS 2006), compute Lbol,south = 3.06 × 1044 erg s−1 and
With our near-infrared imaging, spatially-resolved op- Lbol,north = 4.84 × 1044 erg s−1 . This means that
tical spectroscopy, and 20 cm radio imaging in hand, we the northern source, which appears fainter, is intrin-
are able to analyze the properties of the individual QSO sically more luminous after correcting for its substan-
components. We noticed that the SDSS-assigned red- tially higher amount of extinction. We note that the
shift of z = 1.871 was based on the C IV line, which intrinsically more luminous component coincides with
did not align with the Mg II line in the STIS spec- the brighter radio source.
tra. Since C IV is known to be blushifted relative to We fit a Gaussian profile to the Mg II line in both
QSOs’ systemic redshifts (Richards et al. 2011), we up- spectra to measure the vFWHM values as 2830 ± 650 km
date the source redshift to z = 1.889 based on the Mg II s−1 and 3920 ± 260 km s−1 in the northern and south-
line center derived from these new observations. We fit ern sources, respectively. The errors are computed by
a reddened QSO template to each spectrum and find perturbing the best fit model using the spectrum’s er-
that the southern source has E(B − V ) = 0.179 ± 0.001 ror array and re-fitting 10,000 times. We compute the
while the northern source has E(B −V ) = 0.458±0.009. standard deviation of the Gaussian σ parameter found
These fits are shown with a pink line in Figure 4. We in each fit iteration, shown in the right panels Figure 4.
6

accretion rates than their unobscured counterparts (Ur-

rutia et al. 2012; Kim et al. 2015).
From the definition, R ≡ f (1.4GHz)/f (B), we calcu-
lated the radio loudness parameter of the two sources.
The optical flux is determined by the method described
above, where the QSO template, scaled to the de-
reddened F160W flux, is passed through a Johnson B
filter curve. The radio flux is not K-corrected, given that
the radio spectral index for each source is not known.
We find that both sources have nearly identical R val-
ues, at the boundary of the radio-quiet regime5 , with
Rnorth ≈ 0.46 and Rsouth ≈ 0.48. Table 3 lists all the
Figure 3. Spatial profile of the source spectrum collapsed
derived properties for the two QSOs in this dual system.
along the x-direction. Black points are the summed counts
at each spatial pixel position, and the red line is a double 4. DISCUSSION
Gaussian fit to the data. The shaded areas represent our
3′′ extraction regions, chosen to minimize blending. Two
The confirmation of two distinct QSO spectra, at the
distinct peaks are shown with the southern spectrum over- same redshift, separated by 0.′′ 26, and coincident with
lapping the northern spectrum by ∼ 0.6%. two compact radio sources provides strong evidence that
W2M J1220 is a dual QSO. Most of the known and con-
For the two QSOs, we apply the single-epoch virial firmed dual AGNs are at low redshifts (z < 0.7; e.g. Koss
black hole mass estimator (MBH ) following the formal- et al. 2012; Comerford et al. 2012; Müller-Sánchez et al.
ism of Shen & Liu (2012), 2015; Fu et al. 2015; Rubinur et al. 2019), which is not
yet probing the epoch of peak QSO activity in the uni-
verse (z ≃ 2; Madau & Dickinson 2014) when merger
     
MBH,vir L3000 vFWHM
log = a+b log +c log , rates were significantly higher (Conselice et al. 2003;
M⊙ 1044 erg/s km/s
(1) Rodriguez-Gomez et al. 2015). Therefore, the identifica-
adopting the values a = 0.740, b = 0.620, c = 2.00 tion of a dual QSO system at this epoch is noteworthy,
for single-epoch measurements of FWHMMgII and L3000 , especially given that red QSOs are predominantly found
based on the calibration of Shen et al. (2011). For this in merging hosts.
calculation, we estimate L3000 two different ways. We W2M J1220 is comparable to LBQS 0103−2753
measure it directly from the STIS spectra by applying (Shields et al. 2012), which is a confirmed dual QSO,
an aperture correction to the observed flux, and de- separated by 0.′′ 3, at z = 0.858. LBQS 0103−2753
reddening each spectrum. We then apply an artificial was identified in HST imaging and verified with a STIS
30 Å-wide box-car filter centered on 3000 Å to measure spectrum, similar to W2M J1220. Deep HST imaging of
the source flux, from which we determine luminosity. LBQS 0103−2753 reveals tidal features and morphologi-
The second method starts with the F160W source mag- cal evidence of a recent merger. The WFC3/IR imaging
nitudes (Table 1), which are far less sensitive to uncer- for W2M J1220 is not deep enough to show these fea-
tainties in E(B − V ). We de-redden these magnitudes tures. The two-component spectra of LBQS 0103−2753
and use spectrophotometry to scale a QSO composite are quite distinct, with one of the components show-
template to match the F160W flux. From the scaled ing BAL features indicative of outflows. There is also
template, we measure the 3000 Å flux using the box-car a velocity offset of ∼ 1500 km s−1 between the two
filter as in the previous method. This results in a range components. Although the black hole masses of LBQS
of black hole masses, listed in Table 3. The BH masses 0103−2753 are ∼ 1 − 1.5 orders of magnitude higher
differ for each method by ≲ 0.5 dex but are extremely than W2M J1220, they are similar to each other (both
similar (≲ 0.2 dex) between the two components. These have MBH ∼ 108.5−9 M⊙ ).
masses are at the high end of the range accessible to In the cosmic noon era (z ∼ 2 − 3), Shen et al.
LISA. (2021) report two dual QSO candidates using the novel
Combining MBH and Lbol allows us to estimate the
Eddington ratio. We find that the more obscured north- 5 Objects with R > 2 are categorized as ”radio-loud”, while ob-
ern source has L/LEdd ≃ 0.1 − 0.3, while the less ob- jects are generally considered ”radio-quiet” when R < 0.5. Radio-
scured, southern source has L/LEdd ≃ 0.04 − 0.1. This intermediate sources are those that fit neither category 0.5 < R <
is consistent with findings that red QSOs have higher 2 (Stocke et al. 1992).
 7

Figure 4. Left – Individual spectra of the two QSO components plotted at rest wavelengths. The pink curves represent the
best-fit reddened QSO template. The northern source (gray line) is reddened by E(B − V ) = 0.432 while the southern source
(black line) is reddened by E(B − V ) = 0.184. Mg II and C III lines are labeled. Right – Gaussian fits to the Mg II emission
line in the southern (top) and northern (bottom) spectra showing 10,000 iterations determined by perturbing the best-fit line
using the error arrays. The range of fits reflects the uncertainty in the derived Gaussian parameters.

Table 3. Individual QSO characteristics

Source R.A.a Decl.a E(B − V ) vFWHM log MBH b log Lbol L/LEdd Spk,20cm R
−1 −1
(J2000) (J2000) (mag) (km s ) (M⊙ ) (erg s ) (mJy)
North 12:20:16.87176 +11:26:28.344 0.458 ± 0.009 3140 ± 800 (7.2 − 7.7) ± 0.2 44.76 0.08 − 0.22 0.502 ± 0.066 0.460
South 12:20:16.87420 +11:26:28.082 0.179 ± 0.001 3800 ± 230 (7.40 − 7.70) ± 0.04 44.46 0.05 − 0.1 0.330 ± 0.044 0.478
a The celestial coordinate positions are from the VLBA data and have the uncertainties δ ′′
R.A. = 0.00003s and δDecl. = 0.001 .

b The range of log M

BH is based on the two methods for estimating L3000 while the quoted uncertainties are based on the propagation of the
vFWHM uncertainties and are therefore lower limits.

technique of ‘varstrometry’ which identifies sources with Mangat et al. (2021) to re-interpret the system as two
high astrometric variability in Gaia suggestive of two QSOs that are lensed to form four point source images
distinct, closely-spaced sources with randomly varying based on HST optical and IR observations as well as
fluxes (Hwang et al. 2020). One source, J0841+4825, is VLA observations.
at z = 2.95 and is separated by 0.′′ 46, though its ground- Yue et al. (2021) report a candidate QSO pair at z =
based spatially-resolved spectroscopy shows highly sim- 5.66 separated by 1.′′ 24, or 7.3 kpc. Spatially resolved
ilar spectra which could be explained by gravitational spectroscopy reveal two spectra with similar line char-
lensing. Both components of J0749+2255, at z = 2.17 acteristics but different reddenings, as in W2M J1220.
and also separated by 0.′′ 46, are detected by VLBA ob-
servations at 15 GHz, has a spatially resolved STIS 4.1. Lensing considerations
spectrum, and HST imaging showing merger signatures Some of the properties between the two components of
in the host, putting this system on solid footing for a W2M J1220, such as the derived black hole masses, ra-
dual QSO (Chen et al. 2023). There are 45 additional dio loudness parameters, as well as near-identical emis-
varstrometry-selected dual QSO candidates extending sion line centers and profile shapes, may be explained by
out to z ≃ 3 awaiting confirmation (Chen et al. 2022). gravitational lensing. Here we consider that possibility.
The gravitational lens PSJ1721+8842, initially thought Figure 6 shows the ratio of the northern to south-
to be a quadruple lens at z = 2.37, was analyzed by ern spectrum, with the ratio of the best-fit reddened
templates (pink curves in Figure 4) over-plotted. While
8

most closely-separated lensed QSO known6 Such small-

separation lenses (on the order of ∼ 100 milli-arcseconds
or, so-called, milli-lenses; Frey et al. 2010; Spingola et al.
2019; Casadio et al. 2021) are rare and have been diffi-
cult to find, but probe supermassive (106 −109 ) compact
objects as putative lenses. Recent systematic searches
using VLBA data have resulted in few viable candidates.
This is because a lens with an unusually high surface
density is needed to yield such small separations.
We explore the range of possible lens masses that
could result in a separation of 0.′′ 26 as a function
of redshift to determine the plausibility of lensing in
W2M1220. We employ the relation among distances,
lens mass, and source separation, and a point mass lens,
 2  
c 2 Dd Ds
M (θE ) = θE , (2)
4G Dds
Figure 5. Calibrated and cleaned image of VLBA L-band
observations of W2M J1220 produced with CASA. Overplot- and for a singular isothermal sphere (SIS) mass model,
ted contours indicate the flux from the WFC3/IR F160W r
image at 5σ levels. Two point sources are detected at the σ2 R θE Ds
HST position with a separation of 0.′′ 26 and a position angle M (θE ) = v with σv = c . (3)
G 4π Dds
of 172.205◦ . Beam size of 0.′′ 005 is shown in red in the bot-
tom left. Here, M (θE ) is the mass enclosed within the angular
radius θE , Dd and Ds are the angular diameter distance
to the lens, source, respectively, and Dds is the angular
diameter distance between the lens and source. Given
the redshift of W2M J1220, Ds is known we can compute
M (θE ) as a function of lens redshift. We find that even
the smallest lens masses, residing at z ∼ 0.4−0.6, would
require > 109 M⊙ to be confined to the innermost kpc2 .
Assuming lensing as a possibility, we model the system
with GRAVLENS (Keeton 2001) and determine the lens
position that would yield the two VLBA positions find-
ing that a lens would need to be situated 0.′′ 021695 to the
west and 0.′′ 159543 to the south of the northern source.
An attempt to fix a Sérsic component at this position
Figure 6. Flux ratio of the northern to southern spectra relative to fixed PSF components in our morphological
showing the disappearance of the C IV and Mg II emission fitting (§2) resulted in failure of Galfit to converge. In
lines (vertical dashed lines), motivating an exploration of fact, the central Sérsic component found by Galfit, while
gravitational lensing as the cause for the pair of QSO images. possessing unphysical properties, is situated to the east
The pink line represents a ratio of the best-fit reddening of both PSF components.
curves shown in Figure 4. Identifying intrinsic differences, such as in the spec-
tral slope, between the two QSOs would rule out lens-
the C IV and Mg II emission lines, marked by vertical ing. The E(B − V ) values that we find are determined
dashed lines, disappear, the shape of the ratio is broadly with respect to a composite QSO spectrum that has
consistent with the difference in reddening. The undu- a spectral slope of αν = −0.47 at λ < 5000. We
lating features that deviate from a QSO template seen in vary the slope of the QSO template (following Section
both the ratio spectrum and in the southern component
are not consistent with known QSO spectral features,
6 J0439+1634 has a separation of 0.′′ 2 (Fan et al. 2019). This
such as iron emission in the ultraviolet (UV; Vestergaard
source is of the rare class of “naked” cusp lenses involving three
& Wilkes 2001). images, which is ruled out for our system by the VLBA image.
With a separation of 0.′′ 26, corresponding to an B0218+357, with a separation of 0.′′ 34 is currently the next most
Einstein radius of 0.′′ 13, W2M J1220 would be the closely-separated lens (Patnaik et al. 1993).
 9

5.3 of Glikman et al. 2007) and recomputed E(B − V ) densities at other radio frequencies and compare their
for a template with αν = −0.25 (a bluer slope) and radio spectral indices which, if different for each source,
αν = −0.76 (a redder slope), which represent the range would also rule out lensing. And, if a lens is responsi-
intrinsic to unreddened QSOs (Richards et al. 2003). ble, its nature as highly compact and with extreme dust
While the bluer template does yield slightly higher val- gradients, is worthy of its own study.
ues by ∆E(B − V ) ∼ 0.02 (and vice-versa for the redder
template, yielding lower values by ∆E(B − V ) ∼ 0.03), 4.2. dual statistics for red QSOs
the differences are not sufficient to account for the dif- The serendipitous discovery of a QSO pair in HST
ference in E(B − V ) between the two sources as seen in imaging of a red QSO raises the question of their fre-
the continuum fits (Figure 4), the flux ratios (Figure 6) quency compared to unreddened QSOs. Shen et al.
and the F 105W − F 160W colors. We therefore cannot (2023) investigate statistically the incidence of QSO
attribute the different reddenings to intrinsic differences pairs using Gaia detections of known SDSS QSOs with
in spectral slopes between the two spectra and the best Lbol > 1045.8 erg s−1 at 1.5 < z < 3.5 with separa-
explanation for the E(B −V ) values is different amounts tions of 0.′′ 4 − 3′′ and find an integrated pair fraction
of dust reddening along the two lines of sight. of ∼ 6 × 10−4 . Assuming this fraction is constant at
To achieve the observed amount of reddening would the 0.′′ 26 separation of W2M 1220 and is a factor of
require a lensing galaxy with a significant amount of ∼ 10 higher given its lower luminosity (Lbol ∼ 1044.8
gas and dust. While we cannot rule out lensing with erg s−1 ), the pair fraction is estimated to be ∼ 10−3
this spectral slope investigation, it does rule out the ex- (Shen et al. 2023). However, only 17 red QSOs have
otic possibility that the lens might be a freely floating been imaged with HST at z ∼ 2 (11 F2M quasars re-
SMBH. ported in Glikman et al. 2015, and 6 W2M QSOs which
Finally, the similarities between the black hole proper- include W2M 1220). This fraction of 0.06 (1/17) would
ties in W2M J1220 are seen in previously confirmed dual be an order of magnitude higher than that found for the
QSOs, such as LBQS 0103−2753 (Shields et al. 2012). luminous, unobscured QSOs investigated in Shen et al.
Simulations of SMBH binaries predict major mergers (2023). Given that red QSOs are known to be hosted by
more often produce dual AGNs with similar BH masses major mergers, this population may be the most likely
(Blecha et al. 2013; Steinborn et al. 2016). And, since for finding dual QSOs although with only a single sys-
red QSOs are known to be associated with major merg- tem, we cannot draw broad conclusions.
ers, this discovery may reflect a selection effect towards
similar mass BHs. 5. CONCLUSIONS
The similarity in radio loudness may reflect the en-
We report the discovery of a dual QSO candidate,
hanced low-level radio emission seen in W2M red QSOs,
separated by 0.′′ 26 corresponding to 2.2 kpc at z =
which has been interpreted as coming from either a
1.889. The sources are confirmed as QSOs with a
dusty wind or nascent jets (Glikman et al. 2022). Like-
spatially-resolved STIS spectrum and high-resolution
wise, the differences between the two QSOs, such as the
VLBA imaging at 1.4 GHz which reveal two point
higher accretion rate seen in the redder source, is con-
sources consistent with the positions in the HST im-
sistent with what has been seen in red QSOs elsewhere
ages. The two components are reddened by different
(Urrutia et al. 2012; Kim et al. 2015).
amounts of dust-extinction. When corrected for this ex-
Finally, as some gravitational lenses show intervening
tinction, the properties of the QSOs are similar, includ-
absorption features in the individual component spectra
ing black hole masses ∼ 107.5 M⊙ and radio loudness
(e.g., Rubin et al. 2018), which may reveal the puta-
of ∼ 0.5 (though their Eddington ratios differ). These
tive lens redshift, we explored the absorption features
similarities mean we cannot rule out gravitational lens-
in the STIS spectra and could not identify any evidence
ing, though the lens is not detected in the imaging and
for such coherent features. This further weakens the
extended features seen in the HST imaging may indi-
possibility of a gas rich galaxy as a gravitational lens.
cate merging hosts. The features of these two QSOs are
Therefore, although we cannot definitively rule it out,
consistent with previous findings in dual AGNs.
we consider lensing to be the less likely explanation for
A dual QSO discovered at cosmic noon in a survey
this system.
for red QSOs, which is a population known to be hosted
One way to rule out lensing would be to obtain deeper
by major mergers, can provide a unique population in
imaging with HST that may reveal evidence of a merg-
which to search for such systems where both black holes
ing system, as is seen in the dual QSO found by Chen
are active at the same time. Given that only ∼ 30 red
et al. (2023). Another way would be to measure the flux
quasars have been observed with HST, finding a candi-
10

date dual QSO in such a small sample suggests an ele- scopes (MAST) at the Space Telescope Science Institute.
vated incidence of dual activity in red QSOs. Because The specific observations analyzed can be accessed via
W2M J1220 was found serendipitously, a targeted high 10.17909/5ydb-ex84 and 10.17909/s2sz-t252.
resolution imaging effort of red QSOs at z = 2 − 3 may This research is based on observations made with the
be the most fruitful place to find dual quasars during a NASA/ESA Hubble Space Telescope obtained from the
crucial phase of SMBH/Galaxy co-evolution. Space Telescope Science Institute, which is operated by
the Association of Universities for Research in Astron-
omy, Inc., under NASA contract NAS 5–26555. These
We thank Marianne Vestergaard for sharing the Fe observations are associated with program(s) PID 16794.
UV template which we used to look for features in our The National Radio Astronomy Observatory is a fa-
spectra. E.G. acknowledges the generous support of the cility of the National Science Foundation operated un-
Cottrell Scholar Award through the Research Corpora- der cooperative agreement by Associated Universities,
tion for Science Advancement. E.G. is grateful to the Inc. This work made use of the Swinburne University
Mittelman Family Foundation for their generous sup- of Technology software correlator (Deller et al. 2011),
port. We gratefully acknowledge the National Science developed as part of the Australian Major National Re-
Foundation’s support of the Keck Northeast Astron- search Facilities Programme and operated under licence.
omy Consortium’s REU program through grant AST-
1950797. BDS acknowledges support through a UK Re-
Facilities: HST(STIS), VLBA, SDSS, Palomar(TripleSpec)
search and Innovation Future Leaders Fellowship [grant
number MR/T044136/1]. Software: astropy (Astropy Collaboration et al. 2013,
Some/all of the data presented in this paper were 2018), astroconda, CASA
obtained from the Mikulski Archive for Space Tele- 10.17909/5ydb-ex84 10.17909/s2sz-t252

REFERENCES
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., Fitzpatrick, E. L. 1999, PASP, 111, 63, doi: 10.1086/316293
et al. 2013, A&A, 558, A33, Frey, S., Paragi, Z., Campbell, R. M., & Moór, A. 2010,
doi: 10.1051/0004-6361/201322068 A&A, 513, A18, doi: 10.1051/0004-6361/200913864
Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., Fu, H., Wrobel, J. M., Myers, A. D., Djorgovski, S. G., &
et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f Yan, L. 2015, ApJL, 815, L6,
Becker, R. H., Helfand, D. J., White, R. L., Gregg, M. D., doi: 10.1088/2041-8205/815/1/L6
& Laurent-Muehleisen, S. A. 2003, VizieR Online Data Gebhardt, K., Bender, R., Bower, G., et al. 2000, ApJL,
Catalog, VIII/71 539, L13, doi: 10.1086/312840
Blecha, L., Loeb, A., & Narayan, R. 2013, MNRAS, 429, Glikman, E. 2017, Research Notes of the American
2594, doi: 10.1093/mnras/sts533 Astronomical Society, 1, 48,
CASA Team, Bean, B., Bhatnagar, S., et al. 2022, PASP, doi: 10.3847/2515-5172/aaa0c0
134, 114501, doi: 10.1088/1538-3873/ac9642 Glikman, E., Gregg, M. D., Lacy, M., et al. 2004, ApJ, 607,
Casadio, C., Blinov, D., Readhead, A. C. S., et al. 2021, 60
MNRAS, 507, L6, doi: 10.1093/mnrasl/slab082 Glikman, E., Helfand, D. J., & White, R. L. 2006, ApJ,
Chen, Y.-C., Hwang, H.-C., Shen, Y., et al. 2022, ApJ, 925, 640, 579, doi: 10.1086/500098
162, doi: 10.3847/1538-4357/ac401b Glikman, E., Helfand, D. J., White, R. L., et al. 2007, ApJ,
Chen, Y.-C., Liu, X., Foord, A., et al. 2023, Nature, 616, 667, 673, doi: 10.1086/521073
45, doi: 10.1038/s41586-023-05766-6 Glikman, E., Simmons, B., Mailly, M., et al. 2015, ApJ,
Comerford, J. M., Gerke, B. F., Stern, D., et al. 2012, ApJ, 806, 218, doi: 10.1088/0004-637X/806/2/218
753, 42, doi: 10.1088/0004-637X/753/1/42 Glikman, E., Urrutia, T., Lacy, M., et al. 2012, ApJ, 757,
Conselice, C. J., Bershady, M. A., Dickinson, M., & 51, doi: 10.1088/0004-637X/757/1/51
Papovich, C. 2003, AJ, 126, 1183, doi: 10.1086/377318 Glikman, E., Lacy, M., LaMassa, S., et al. 2018, ApJ, 861,
Deller, A. T., Brisken, W. F., Phillips, C. J., et al. 2011, 37, doi: 10.3847/1538-4357/aac5d8
PASP, 123, 275, doi: 10.1086/658907 —. 2022, ApJ, 934, 119, doi: 10.3847/1538-4357/ac6bee
Fan, X., Wang, F., Yang, J., et al. 2019, ApJL, 870, L11, Gordon, K. D., & Clayton, G. C. 1998, ApJ, 500, 816,
doi: 10.3847/2041-8213/aaeffe doi: 10.1086/305774
 11

Hopkins, P. F., Hernquist, L., Cox, T. J., et al. 2005, ApJ, Rubinur, K., Das, M., & Kharb, P. 2019, MNRAS, 484,
630, 705, doi: 10.1086/432438 4933, doi: 10.1093/mnras/stz334
Hopkins, P. F., Hernquist, L., Cox, T. J., & Kereš, D. 2008, Sanders, D. B., Soifer, B. T., Elias, J. H., et al. 1988, ApJ,
ApJS, 175, 356, doi: 10.1086/524362 325, 74, doi: 10.1086/165983
Hwang, H.-C., Shen, Y., Zakamska, N., & Liu, X. 2020, Shen, Y., & Liu, X. 2012, ApJ, 753, 125,
ApJ, 888, 73, doi: 10.3847/1538-4357/ab5c1a doi: 10.1088/0004-637X/753/2/125
Keeton, C. R. 2001, arXiv e-prints, astro, Shen, Y., Richards, G. T., Strauss, M. A., et al. 2011,
doi: 10.48550/arXiv.astro-ph/0102340 ApJS, 194, 45, doi: 10.1088/0067-0049/194/2/45
Shen, Y., Chen, Y.-C., Hwang, H.-C., et al. 2021, Nature
Kim, D., Im, M., Glikman, E., Woo, J.-H., & Urrutia, T.
Astronomy, 5, 569, doi: 10.1038/s41550-021-01323-1
2015, ApJ, 812, 66, doi: 10.1088/0004-637X/812/1/66
Shen, Y., Hwang, H.-C., Oguri, M., et al. 2023, ApJ, 943,
Koss, M., Mushotzky, R., Treister, E., et al. 2012, ApJL,
38, doi: 10.3847/1538-4357/aca662
746, L22, doi: 10.1088/2041-8205/746/2/L22
Shields, G. A., Rosario, D. J., Junkkarinen, V., et al. 2012,
Linford, J. 2022, VLBA Scientific Memos, 38
ApJ, 744, 151, doi: 10.1088/0004-637X/744/2/151
Madau, P., & Dickinson, M. 2014, ARA&A, 52, 415,
Spingola, C., McKean, J. P., Lee, M., Deller, A., & Moldon,
doi: 10.1146/annurev-astro-081811-125615
J. 2019, MNRAS, 483, 2125, doi: 10.1093/mnras/sty3189
Magorrian, J., Tremaine, S., Richstone, D., et al. 1998, AJ,
Steinborn, L. K., Dolag, K., Comerford, J. M., et al. 2016,
115, 2285, doi: 10.1086/300353
MNRAS, 458, 1013, doi: 10.1093/mnras/stw316
Mangat, C. S., McKean, J. P., Brilenkov, R., et al. 2021,
Stocke, J. T., Morris, S. L., Weymann, R. J., & Foltz, C. B.
MNRAS, 508, L64, doi: 10.1093/mnrasl/slab106
1992, ApJ, 396, 487, doi: 10.1086/171735
Marconi, A., & Hunt, L. K. 2003, ApJL, 589, L21, Telfer, R. C., Zheng, W., Kriss, G. A., & Davidsen, A. F.
doi: 10.1086/375804 2002, ApJ, 565, 773
Müller-Sánchez, F., Comerford, J. M., Nevin, R., et al. Urrutia, T., Becker, R. H., White, R. L., et al. 2009, ApJ,
2015, ApJ, 813, 103, doi: 10.1088/0004-637X/813/2/103 698, 1095, doi: 10.1088/0004-637X/698/2/1095
Patnaik, A. R., Browne, I. W. A., King, L. J., et al. 1993, Urrutia, T., Lacy, M., & Becker, R. H. 2008, ApJ, 674, 80,
MNRAS, 261, 435, doi: 10.1093/mnras/261.2.435 doi: 10.1086/523959
Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, Urrutia, T., Lacy, M., Spoon, H., et al. 2012, ApJ, 757,
AJ, 124, 266, doi: 10.1086/340952 125, doi: 10.1088/0004-637X/757/2/125
Richards, G. T., Hall, P. B., Vanden Berk, D. E., et al. Van Wassenhove, S., Volonteri, M., Mayer, L., et al. 2012,
2003, AJ, 126, 1131, doi: 10.1086/377014 ApJL, 748, L7, doi: 10.1088/2041-8205/748/1/L7
Richards, G. T., Lacy, M., Storrie-Lombardi, L. J., et al. Vestergaard, M., & Wilkes, B. J. 2001, ApJS, 134, 1,
2006, ApJS, 166, 470, doi: 10.1086/506525 doi: 10.1086/320357
Richards, G. T., Kruczek, N. E., Gallagher, S. C., et al. Wilson, J. C., Henderson, C. P., Herter, T. L., et al. 2004,
2011, AJ, 141, 167, doi: 10.1088/0004-6256/141/5/167 in Proc. SPIE, Vol. 5492, Ground-based Instrumentation
Rodriguez-Gomez, V., Genel, S., Vogelsberger, M., et al. for Astronomy, ed. A. F. M. Moorwood & M. Iye,
2015, MNRAS, 449, 49, doi: 10.1093/mnras/stv264 1295–1305, doi: 10.1117/12.550925
Yue, M., Fan, X., Yang, J., & Wang, F. 2021, ApJL, 921,
Rubin, K. H. R., O’Meara, J. M., Cooksey, K. L., et al.
L27, doi: 10.3847/2041-8213/ac31a9
2018, ApJ, 859, 146, doi: 10.3847/1538-4357/aaaeb7