A Candidate Kiloparsec-scale Quasar Pair at $z=5.66$

We report the discovery of a close quasar pair candidate at $z=5.66$, J2037--4537. J2037--4537 is resolved into two quasar images at the same redshift in ground-based observations. Followup spectroscopy shows significant differences in both the continuum slopes and emission line properties of the two images. The two quasar images have a projected separation of $1\farcs24$ ($7.3\text{~kpc}$ at $z=5.66$) and a redshift difference of $\Delta z\lesssim0.01$. High-resolution images taken by {\em Hubble Space Telescope} do not detect the foreground lensing galaxy. The observational features of J2037--4537 strongly disfavor the lensing hypothesis. If J2037--4537 is a physical quasar pair, it indicates a quasar clustering signal of $\sim10^5$ at a separation of $\sim10$ proper kpc (pkpc), and gives the first observational constraint on the pair fraction of $z>5$ quasars, $f_\text{pair}(r<30\text{~pkpc})>0.3\%$. The properties of J2037--4537 are consistent with those of merger-triggered quasar pairs in hydrodynamical simulations of galaxy mergers.


INTRODUCTION
Galaxy mergers are natural consequences of hierarchical structure formation of the universe (e.g., Cole et al. 2000). It has been proposed that galaxy mergers can trigger powerful quasar activity which regulates the star formation of their host galaxies by significant feedback (e.g., Di Matteo et al. 2005). In some rare cases, the supermassive black holes (SMBHs) in both progenitor galaxies are ignited by the merging event, forming a close pair of quasars (for a recent review, see De Rosa et al. 2019). High-resolution hydrodynamical simulations suggest that close quasar pairs correspond to a special, short-lived phase of galaxy mergers, which most frequently appear when the SMBHs have a mass ratio close to one and a separation smaller than 10 kpc (e.g., Capelo et al. 2015Capelo et al. , 2017. Quasar pairs are also predicted in cosmological simulations (e.g., Steinborn et al. 2016;Volonteri et al. 2016;Rosas-Guevara et al. 2019), and their statistics (e.g., the fraction of pairs among all quasars) constrain the evolution of SMBHs and their host galaxies. In addition, close quasar pairs trace overdensities (e.g., Onoue et al. 2018), and are unique probes of the small-scale structure of intergalactic medium (e.g., Rorai et al. 2017).
Searches of quasar pairs have been carried out using wide-area optical and infrared (IR) sky surveys (e.g., Hennawi et al. 2010;Silverman et al. 2020). These studies have discovered several tens of close quasar pairs (with projected separation ∆d 10 kpc) out to z ∼ 3 (e.g., Shen et al. 2021;Tang et al. 2021). Due to the rapid decline of quasar number densities (e.g., Kulkarni et al. 2019) and the relatively poor physical resolution for most observations, it becomes difficult to identify close quasar pairs at z 3. Finding high-redshift quasar pairs is critical to the understanding of the evolution and environment of high-redshift quasars and galaxy mergers.
In this letter, we report a close quasar pair candidate at redshift z = 5.66 with a projected proper distance of ∆d = 7.3 kpc, J2037-4537. Following Hennawi et al. (2010), in this Letter, we define a quasar pair as two quasars with ∆d < 1 Mpc and a radial velocity difference of ∆v < 2000 km s −1 . We further define close quasar pairs to be those with ∆d < 10 kpc. The most distant quasar pair previously known is at z = 5.1 with ∆d = 135 kpc (McGreer et al. 2016), and for close quasar pairs, the highest-redshift one is at z = 3.1 with ∆d = 7.4 kpc (Tang et al. 2021). J2037-4537 significantly extends the frontier of studies of quasar pairs. This letter is organized as follows. We describe the ground-based observations and the Hubble Space Tele-scope (HST) imaging in Section 2. We discuss the interpretation of the data in Section 3 and summarize in Section 4. We use a flat ΛCDM cosmology with H 0 = 70 km s −1 Mpc −1 and Ω M = 0.3.

DATA AND ANALYSIS
2.1. Ground-Based Imaging and Spectroscopy J2037-4537 was discovered in our on-going survey for high-redshift gravitationally lensed quasars (Yue et al. in prep) in the Dark Energy Survey (DES, e.g., Abbott et al. 2018) field. It is not detected in the DES g−band, and has the DES r = 22.8, i = 20.4, z = 19.6 and Y = 19.5, exhibiting typical colors of a quasar at redshift z 5.5 (e.g., Yang et al. 2019). In the DES images ( Figure 1, upper left), J2037-4537 appears to be two point sources separated by 1. 24, where the fainter one (object B) is redder than the brighter one (object A). Table 1 lists the photometric properties of the two objects. The magnitudes are obtained by fitting the images as two point sources using galfit (Peng et al. 2002), where the point spread function (PSF) models are built by the IRAF task psf.
The optical spectrum of J2037-4537, taken by the Low Dispersion Survey Spectrograph (LDSS-3) on Magellan/Clay telescope, reveals that both objects are quasars at redshift z = 5.66 ( Figure 1). We use the VPH-Red grism and the 1. 0 slit which deliver a resolution of R = 1350. The spectrum is reduced with the standard IRAF pipeline. The two objects are clearly resolved in the spectrum and have the same Lyα-break wavelength, suggesting a redshift difference ∆z 0.01. Object B has a redder spectral energy distribution (SED) shape than object A, which agrees with the broad-band photometries. We also obtained near-IR spectra of the two objects using the Folded-port InfraRed Echelle (FIRE) on Magellan/Baade telescope. The spectra are reduced with PypeIt (Prochaska et al. 2020).
J2037-4537 could either be a physical quasar pair, or two images of a strongly lensed quasar. Both the image and the spectra do not show signs of a third object (e.g., a foreground lensing galaxy). This is further confirmed by the HST image (Section 2.2). In addition, the two objects have different emission line profiles. The difference is the most obvious for the C iii] line and can also be seen in the A/B flux ratio (Figure 1, lower panel). These features disfavor the strong-lensing scenario and indicate that J2037-4537 is a physical quasar pair. Figure 2 illustrates the C iv broad emission lines of the two objects, which are fitted as a power-law continuum plus a Gaussian profile. Using the C iv line width and the empirical relation in Vestergaard & Peterson (2006), we estimate the SMBH masses to be log M BH /M =  Although the Mg ii line is a better SMBH mass indicator for high-redshift quasars, it falls in the wavelengths with strong sky emission lines at z = 5.66. The results are listed in Table 1.

HST Imaging
Gravitational lensing can also generate two close quasar images at the same redshift. To fully explore the lensing hypothesis, we observed J2037-4537 using HST ACS/WFC in the F850LP filter, aiming at detecting or ruling out the foreground lensing galaxy (Program GO-16507). The F850LP filter has a wavelength coverage of 0.8µm < λ < 1.1µm, where the two objects show significantly different SEDs in the ground-based spectroscopy. Figure 3 presents the HST image and the two-PSF fitting residual. J2037-4537 is well-described by the two-PSF model and there is no evidence for a third object in this field. The extracted 1-D spectrum of the two objects, which is a combination of the Magellan/Clay LDSS-3 spectra (λ < 1µm) and the Magellan/Baade FIRE spectra (λ > 1µm). Both objects exhibit features of a z = 5.66 quasar, and object B is redder than object A. The dashed line shows the F850LP filter used in HST ACS/WFC imaging. Lower Panel: The flux ratio between the two objects. The shaded area marks wavelengths bluer than the Lyα break, where the quasars have little flux due to the IGM absorption. There are complex features around the emission lines, which suggest that the two objects have different line profiles, and disfavor the strong lensing hypothesis.
To further set a flux limit for the possible lensing galaxy, we measure the flux in the region between the two point sources (region C, marked by the red circle in the right panel of Figure 3). Since PSF models of ACS/WFC usually have non-negligible errors in PSF wings (Jee et al. 2007), we measure two additional regions (C1 and C2, marked by the yellow circles) to correct this systematic uncertainty. The yellow regions have the same size as the red one, and the distance from region C1 (C2) to object A (B) equals to the distance from region C to object A (B). The flux in region C1 (C2) thus estimates the contribution of the PSF wing residuals from object A (B) to the flux in region C. We estimate the flux of the possible foreground galaxy as This gives a non-detection and rules out the existence of a foreground lensing galaxy with m F850LP < 26.7 at 3σ level. Assuming a typical redshift of z l ∼ 1 for the lensing galaxy (e.g., Hilbert et al. 2008;Wyithe et al. 2011), this limit gives an absolute magnitude of M −17.6, which is about four magnitudes fainter than the breaking magnitude of the galaxy luminosity function at z ∼ 1 (M * ∼ −21.5; e.g., Faber et al. 2007).

Testing the Strong-Lensing Hypothesis
The distinct SEDs of the two quasar images and the non-detection of the lensing galaxy strongly disfavor the Quasar B Figure 2. The C iv emission lines of the two quasars (black) and the best-fit model (red). We use a power-law continuum and a Gaussian profile to describe the emission lines. The absorption features (gray dashed area) are masked during the fitting. lensing hypothesis. Although microlensing and/or differential reddening can lead to different spectral features in lensed quasar images (e.g., Sluse et al. 2012), it is difficult for these effects to explain the observational features of J2037-4537. If object A and B are lensed images of the same quasar, the extinction of image B must be much stronger than A given their colors. At long wavelengths (λ obs 1.5µm), image B is brighter than image A, meaning that image A is intrinsically fainter. Meanwhile, in most galaxy-scale lensing systems, the fainter image is closer to the deflector (e.g., Mason et al. 2015) and should have a stronger extinction. This conflicts with the fact that image B is redder than A, suggesting that the lensing scenario is unlikely.
In principle, a deflector galaxy with highly irregular mass profiles and/or dust distributions might be able to generate a lensing system like J2037-4537. We thus report J2037-4537 as a high-confidence close quasar pair candidate. The definitive test of the lensing sce-  There is no sign of a third object besides the two point sources. We measure the flux in the area C (the red circle, with a radius of 0. 35) to estimate the flux limit of the possible foreground lensing galaxy. We also measure the fluxes in two additional areas, C1 and C2 (yellow circles), which correct the contributions from the PSF-subtraction residual (see text for details). Our estimation gives a 3σ flux limit of mF850LP > 26.7 for region C.
nario can be achieved with the Atacama Large Millimeter/submillimeter Array (ALMA). ALMA will measure the [C ii] redshifts of the two quasar images to an accuracy of < 0.001 (e.g., Decarli et al. 2018), and will reveal the lensed arcs of the quasar host galaxy. In the following discussions, however, we will treat J2037-4537 as a physical quasar pair. pair fraction, f pair (r < 30 pkpc), is about 1% with little redshift evolution.
J2037-4537 allows us to constrain the quasar pair fraction at high redshift for the first time. This is done by estimating the expected number of quasars we can find in our survey. J2037-4537 is discovered serendipitously in our survey for high-redshift lensed quasars in the DES field, which covers ∼ 5000 deg 2 . We identify J2037-4537 as a lens candidate with quasar-dominated flux, for which we request a non-detection in DES−g, a DES−z magnitude of m z < 21, and a projected separation of ∆d < 3. 0. The g− and z−band flux cuts restrict our targets to 5 z 6 quasars (e.g., Wang et al. 2016;Yang et al. 2017Yang et al. , 2019. As the survey is not completed yet, we simply assume a selection function of one for 5 < z < 6 quasars, and provide the lower limit of the quasar pair fraction. We use SIMQSO (McGreer et al. 2013) to generate a mock catalog of quasars in the DES field at 5 < z < 6 that are brighter than m z = 21. SIMQSO is a Pythonbased package which generates mock catalogs of quasars with simulated spectra, and has been shown to accurately reproduce the observed colors and magnitudes of quasars (e.g., Ross et al. 2013). We use a broken power-law to parameterize the quasar luminosity function (QLF): We adopt the z = 5 QLF from Kim et al. (2020) and the z = 6 QLF from Matsuoka et al. (2018), and linearly interpolate the QLF parameters (log Φ * , M * , α, β) for other redshifts at 5 < z < 6. SIMQSO suggests that there are N all = 485 quasars brighter than m z = 21 at 5 < z < 6 in the DES field. J2037-4537 directly constrains the number of quasar pairs at z > 5 with 0. 9 < ∆d < 3. 0. The lower boundary of the projected separation corresponds to the spatial resolution of the DES survey. To avoid confusion, we always use r to denote the physical separation of a quasar pair, and use ∆d to denote the projected separation. By converting the angular separations to proper distances, we find f pair [5.3 < ∆d (pkpc) < 17.7] ≥ 1/485 = 0.2%. As a comparison, the expected number of quasar pairs is 7 × 10 −6 if quasars are randomly distributed. We estimate this number by assigning random positions to the quasars in the mock catalog, and counting pairs of mock quasars with 0. 9 < ∆d < 3. 0 and ∆v < 2000 km s −1 , following Hennawi et al. (2006). The clustering signal for z > 5 quasars is thus W p (0. 9 < ∆d < 3. 0) ∼ 10 5 . Hennawi et al. (2006) measure the quasar correlation function at 1 z 3 and ∆d ∼ 0.1 pMpc. Extrapolating the correlation function in Hennawi et al. (2006) to small scales gives a clustering signal of W p ∼ 10 3 . This comparison indicates that close quasar pairs are regulated by processes (likely galaxy mergers) that are different from quasar clustering at larger scale.
In order to make direct comparisons to the simulated pair fractions, we convert f pair (0. 9 < ∆d < 3. 0) to f pair (r < 30 pkpc) as follows. We assume that the smallscale quasar correlation function can be described by a power-law, i.e., ξ(r) = (r/R 0 ) −γ . Following the analysis in Hennawi et al. (2006) and McGreer et al. (2016), the expected number of quasar pairs with separation r < R can be calculated as (3) where the sum goes through all the mock quasars and n(z j ) is the number density of quasars at redshift z j that have m z < 21. We use the quasar spectrum template from Vanden Berk et al. (2001) to convert the absolute magnitude M 1450 to the z−band apparent magnitude. Similarly, the expected number of quasar pairs with projected separation d min < ∆d < d max is where a = (1 + z) −1 is the scale factor, H(z j ) is the Hubble constant at redshift z j , and v max = 2000 km s −1 is the maximum velocity difference of the quasar pair. Note that all the quantities in Equation 3 and 4 are in comoving units. We adopt a fiducial power-law index of γ = 2, which gives a good description of quasar pairs down to a separation of 100 pkpc (e.g., Shen et al. 2010). At small separations, we have ξ(r) 1, and γ = 2 gives a flat distribution of dNpair(r<R) dR , in agreement with the simulated AGN pairs in Rosas-Guevara et al. (2019).
By applying N pair (0. 9 < ∆d < 3. 0) > 1 to the above equations, we get R 0 > 254h −1 comoving Mpc (cMpc), N pair (r < 30 pkpc) > 1.5 and f pair (r < 30 pkpc) > 0.3%. Figure 4 shows the comparison between our results and the predictions from the cosmological simulations. The lower limit is several times lower than the predicted pair fraction. Note that the simulations count both obscured and unobscured AGNs. Since our survey only targets unobscured type-I quasars, we expect that the observed quasar pair fraction in our survey should The simulated AGN samples have L bol > 10 43 erg s −1 , and include both obscured and unobscured AGNs. The observed lower limit is several times lower than the simulated sample. Note that the two quasars in J2037-4537 have L bol ∼ 10 47 erg s −1 (Table 1). Observations at lower redshift indicate that the pair fraction decreases with AGN luminosity, which could explain the difference between the simulations and the lower limit set by J2037-4537 (see text for details).
be lower than the simulated AGN pair fractions. In addition, the AGN pair fraction depends on the luminosity and the SMBH mass, as discussed in De Rosa et al.
. Silverman et al. (2020) give an observed pair fraction of ∼ 0.26% for z 4 quasars, and their sample has a luminosity cut (log L bol > 45.3) that is comparable to our survey. Silverman et al. (2020) show that the observed fraction is consistent with a luminosity-matched sample in the cosmological simulations, which indicates that the pair fraction might be lower for luminous AGNs. Current cosmological simulations do not have sufficient volumes to produce a quasar pair like J2037-4537 at z > 5. Future simulations with larger volumes will enable direct comparisons with observations and provide unique constraints on the SMBH evolution models at high redshift.

The Triggering Mechanism of the Quasar Pair
The tangential separation between the two quasars in J2037-4537 is 7.3 kpc, and the spectra suggest a redshift difference of ∆z 0.01. As such, this quasar pair must reside in a galaxy merger. Although current data cannot rule out the possibility that the two quasars are triggered independently and coincidentally before the merging event, we argue that J2037-4537 is likely triggered by the galaxy merger. This is because the correlation length indicated by J2037-4537 (R 0 > 254h −1 cMpc) is much larger than the correlation length of z ∼ 5 quasars (R 0 ∼ 20h −1 cMpc, e.g., McGreer et al. 2016). The significant difference in R 0 at large and small scales indicates that close quasar pairs and quasar pairs with separation 100 pkpc might have different origins. The former are more likely results of galaxy mergers, while the latter are related to structure formation at sub-Mpc scale.
The observational features of J2037-4537 is consistent with merger-triggered quasar pairs in hydrodynamical simulations. Capelo et al. (2017) explore the factors that control the emergence of AGN pairs in galaxy mergers, and conclude that close quasar pairs usually appear in galaxy mergers with SMBH mass ratios close to one and separations less than 10 pkpc. It is thus plausible that the two quasars are triggered by the merging process. Future observations (likely with ALMA or the James Webb Space Telescope) will reveal more details of the gas kinematics and will allow investigating how the gas feeds the SMBHs in this system. Besides, we notice that the two quasars show similar profiles for a few broad emission lines (e.g., the Lyα, N v, O i and Si iv lines). If J2037-4537 is a physical quasar pair, these similarities might be causally connected to the physical association of the two quasars and encode critical information about how the quasar activities are triggered.

SUMMARY
We report the discovery of a close quasar pair candidate at z = 5.66, J2037-4537, which has a projected separation of 1. 24 (7.3 kpc). The ground-based spectroscopy shows that the resolved two objects are two quasars at the same redshift, and the high-resolution HST imaging does not detect the foreground lensing galaxy. Given the features in the spectra and the high-resolution images, it is highly unlikely that J2037-4537 is a lensed quasar. J2037-4537 sets the first constraint on the pair fraction of quasars at 5 < z < 6, f pair (r < 30 pkpc) > 0.3%. Future observations of the gas kinematics in J2037-4537 will reveal more information about the triggering mechanism of the quasar pair.