A significant excess in major merger rate for AGNs with the highest Eddington ratios at z<0.2
V. Marian, K. Jahnke, I. Andika, E. Banados, V.N. Bennert, S. Cohen, B. Husemann, M. Kaasinen, A.M. Koekemoer, M. Mechtley, M. Onoue, J.T. Schindler, M. Schramm, A. Schulze, J.D. Silverman, I. Smirnova-Pinchukova, A. van der Wel, C. Villforth, R.A. Windhorst
DDraft version October 2, 2020
Typeset using L A TEX twocolumn style in AASTeX63
A significant excess in major merger rate for AGNs with the highest Eddington ratios at z < . Victor Marian,
1, 2
Knud Jahnke, Irham Andika,
1, 2
Eduardo Ba˜nados, Vardha N. Bennert, Seth Cohen, Bernd Husemann, Melanie Kaasinen,
1, 5, 2
Anton M. Koekemoer, Mira Mechtley, Masafusa Onoue, Jan-Torge Schindler, Malte Schramm, Andreas Schulze, John D. Silverman,
9, 10
Irina Smirnova-Pinchukova,
1, 2
Arjen van der Wel, Carolin Villforth, and Rogier A. Windhorst Max-Planck-Institut f¨ur Astronomie, K¨onigstuhl 17, 69117 Heidelberg, Germany International Max Planck Research School for Astronomy & Cosmic Physics at the University of Heidelberg Department of Physics, California Polytechnic State University, San Luis Obispo, CA 93407, USA School of Earth and Space Exploration, Arizona State University, P.O. Box 871404, Tempe, AZ 85287-1404, USA Universit¨at Heidelberg, Zentrum f¨ur Astronomie, Institut f¨ur Theoretische Astrophysik, Albert-Ueberle-Straße 2, D-69120 Heidelberg,Germany Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA Graduate school of Science and Engineering, Saitama Univ., 255 Shimo-Okubo, Sakura-ku, Saitama City, Saitama 338-8570, Japan National Astronomical Observatory of Japan, Mitaka, Tokyo 181-8588, Japan Kavli Institute for the Physics and Mathematics of the Universe, The University of Tokyo, Kashiwa, Japan 277-8583 (Kavli IPMU,WPI) Department of Astronomy, School of Science, The University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113-0033, Japan Sterrenkundig Observatorium, Universiteit Gent, Krijgslaan 281 S9, B-9000 Gent, Belgium University of Bath, Department of Physics, Claverton Down, BA2 7AY, Bath, United Kingdom
ABSTRACTObservational studies are increasingly finding evidence against major mergers being the dominantmechanism responsible for triggering AGN. After studying the connection between major mergers andAGN with the highest Eddington ratios at z = 2, we here expand our analysis to z < .
2, exploringthe same AGN parameter space. Using ESO VLT/FORS2 B − , V − and color images, we examinethe morphologies of 17 galaxies hosting AGNs with Eddington ratios λ edd > .
3, and 25 mass- andredshift-matched control galaxies. To match the appearance of the two samples, we add syntheticpoint sources to the inactive comparison galaxies. The combined sample of AGN and inactive galaxieswas independently ranked by 19 experts with respect to the degree of morphological distortion. Wecombine the resulting individual rankings into multiple overall rankings, from which we derive therespective major merger fractions of the two samples. With a best estimate of f m , agn = 0.41 ± f m , ina = 0.08 ± z < .
2, the origin of (cid:38)
50% of this specific AGN subpopulationstill remains unclear.
Keywords: galaxies: active — galaxies: evolution — galaxies: interactions — quasars: general INTRODUCTIONAn ever-growing number of empirical studies are find-ing that the properties of the black holes (BH) at thecenter of galaxies are closely correlated with the prop-erties of the host galaxy, i.e. BH mass, bulge velocity
Corresponding author: Victor [email protected] dispersion and mass, stellar host mass, velocity disper-sion or luminosity (e.g. Marconi & Hunt 2003; Hring &Rix 2004; Jahnke et al. 2009; Bennert et al. 2010, 2011;Beifiori et al. 2012; Graham & Scott 2013; McConnell& Ma 2013; Davis et al. 2018, 2019; de Nicola et al.2019; Sahu et al. 2019; Shankar et al. 2019; Ding et al.2020). These findings are complemented by state-of-the-art cosmological hydrodynamical simulations (Habouzitet al. 2019; Li et al. 2019a; Terrazas et al. 2019) thatattempt to capture the physics behind these relations. a r X i v : . [ a s t r o - ph . GA ] S e p Marian et al.
Combined with the widely-accepted assumption that ev-ery major galaxy hosts a supermassive BH in its center(Kormendy & Ho 2013), this strongly indicates that hi-erarchical structure formation applies to black holes inthe same way as it does to galaxies as a whole (Jahnke& Macci 2011).The potential feedback of the emitted radiation,winds, jets, or a combination thereof, when a BH be-comes active, (i.e. starts accreting matter) may havea broad range of effects on the host galaxy, dependingon the physical nature, geometry and/or size of thosedifferent outflow mechanisms (Silk & Rees 1998; Harri-son et al. 2018). These range from the total quenchingto the enhancement of star formation due to variousprocesses affecting the interstellar and circumgalacticmedium (Husemann & Harrison 2018; Weinberger et al.2018; Davies et al. 2019; Nelson et al. 2019; Truong et al.2019; Valentini et al. 2019; Oppenheimer et al. 2020), al-though the impact may also be negligible (Schulze et al.2019; O’Leary et al. 2020). In addition, individual AGNfeedback processes could even have an impact on largerscales by affecting satellite galaxies and the surroundingintracluster or intragroup medium (Blanton et al. 2010;Chowdhury et al. 2019; Dashyan et al. 2019; Li et al.2019b; Martin-Navarro et al. 2019).Considering this interplay between galaxies and theircentral BH in its active phase, it is imperative to un-derstand the mechanisms responsible for triggering theperiod of significant black hole accretion. For decadesit has been assumed that galaxies follow an evolution-ary path that includes at least one merging event withanother galaxy of a similar mass (i.e. a major merger).This gravitational encounter would strip part of the gasof its angular momentum, funneling it into the mostcentral regions where the BH(s) reside (Barnes & Hern-quist 1992; Sanders & Mirabel 1996). Such an inci-dent would ultimately lead to the active galactic nu-cleus (AGN) phase, in which the coalescing galaxy hostsat least one active BH in the center. This theoreti-cal scenario was comprehensively presented in the sem-inal work of Sanders et al. (1988), and further studiedwith numerous simulations (Springel et al. 2005; Hop-kins et al. 2006a, 2008; Somerville et al. 2008; McAlpineet al. 2018, 2020; Weigel et al. 2018) and observations(e.g. Yue et al. 2019; Gao et al. 2020). These causalconnections, between major mergers and the presenceof an active BH, have been found especially for partic-ular AGN populations at low redshift (Koss et al. 2010;Cotini et al. 2013; Sabater et al. 2013; Hong et al. 2015;Ellison et al. 2019), and high-luminosity AGNs at differ-ent cosmic epochs (Urrutia et al. 2008; Schawinski et al.2012; Treister et al. 2012; Glikman et al. 2015; Fan et al. 2016; Donley et al. 2018; Goulding et al. 2018; Urbano-Mayorgas et al. 2019).In recent years, however, a number of studies havefound that the fraction of major mergers amongst AGNhosts is < L X ≤ erg s − (Grogin et al. 2005; Allevato et al. 2011; Schawinski et al.2011; Kocevski et al. 2012; Bhm et al. 2013) or high X-ray luminosities with L X ≥ erg s − (Karouzos et al.2014; Villforth et al. 2014, 2017) found no significantconnection. Studies examining more specific samples ofAGNs have obtained similar results: neither sources thatpossess the highest BH masses (Mechtley et al. 2016) norheavily obscured AGNs (Schawinski et al. 2012; Zhaoet al. 2019) appear to be triggered predominantly by ma-jor mergers. Even AGNs assumed to be in an early evo-lutionary stage (Villforth et al. 2019), or those exhibitingthe highest Eddington ratios (Marian et al. 2019) showno signs of an enhanced merger fraction. Additionalstudies detected slight enhancements in the merger ratefor AGNs at different luminosities and redshifts; how-ever, the vast majority of AGNs were still not majormerger induced (Silverman et al. 2011; Rosario et al.2015; Hewlett et al. 2017). In contrast, recent work ex-amining secularly powered outflows (Smethurst et al.2019) and the dependence of local AGNs on environ-ment (Man et al. 2019) suggest that secular processesare the dominant mechanisms to trigger AGN activity.These studies, in which AGNs with a variety of differ-ent redshifts, brightnesses and masses have been exam-ined, have come to the unanimous conclusion that merg-ers should only be considered as one of several possiblemechanisms for initiating black hole growth. Therefore,it is necessary to consider alternative processes and/ordifferences in the lifetime of merger features and AGNs.Large-scale galactic bars (Cheung et al. 2015; Cister-nas et al. 2015; Goulding et al. 2017) and a time delaybetween a major merger event and the onset of an AGN(Cisternas et al. 2011; Mechtley et al. 2016; Marian et al.2019) appear to be an inadequate explanation for thesecontrary results regarding the relevance of large-scalemergers for triggering AGNs. Instead, Goulding et al.(2018) propose an intriguing alternative, which may easethis tension: although AGNs are indeed triggered bymajor mergers, their activity and therefore luminosity ajor merger rate for AGNs with the highest Eddington ratios at z < . λ edd = L/L edd ,i.e. the highest specific accretion rates at z < . z ∼
2. Contraryto z ∼
2, which marks the peak of cosmic black holeactivity (Boyle et al. 2000; Aird et al. 2015) and starformation rate (Madau & Dickinson 2014), the compa-rable population of local AGN host galaxies at z < . ∼
10 times lower black hole activity andstar formation rates (Aird et al. 2015). Moreover, onlya small fraction ( (cid:46) M ∗ / M (cid:12) ) >
10) may have undergone one or moremajor merger events since z ∼
1, with the majority ofsuch galaxies being undisturbed for the past ∼ z < . z ∼
2. Hence, we may expect different physicalprocesses to be dominant at such a low redshift, whichmakes it necessary to also examine the role of majormergers with respect to triggering AGNs at such a cos-mic time. Despite the expected small overall mergerrates at low redshifts, especially for the particular pop-ulation of AGNs showing the highest Eddington ratio,major mergers may still be the only viable option todeliver enough gas to the BH for it to reach such highspecific accretion rates.Like in almost all the aforementioned studies that re-ject major mergers as the dominant triggering mecha-nism of AGNs, we compare a specific sample of AGNhost galaxies to a sample of inactive comparison galax-ies, matched in redshift, stellar mass, observed wave-length, depth, and signal-to-noise ratio (S/N). We exam-ine 17 galaxies hosting AGNs with λ edd > . z < . Λ = 0 . , Ω =0 . h = 0 .
7. At our sample’s median redshift of z ∼ . B - and V - approximately correspond to rest-frame U - and B -band. DATAWe base the sizes of our two samples on the goalto identify a potential predominant presence of majormerger signatures in AGN host galaxies with respect toa matched sample of inactive galaxies. As a fiducialinitial condition, we assume a merger fraction for ourcontrol sample of inactive sources of f m , ina = 0 .
15 withthe goal to be able to detect for an AGN host galaxymerger fraction of f m , agn ≥ . ∼
99% confidence. Asthe confidence of a detected difference in merger frac-tions can only increase for smaller values of f m , ina andto ensure we achieve this desired level of confidence, weuse this, when compared to literature results (e.g. Lotzet al. 2011; Man et al. 2016; Mundy et al. 2017), ratherlarge value for f m , ina . We expect this fiducial fractionto be an upper limit of the real merger rate for inactivegalaxies in our mass and redshift range.Since the number of available AGNs with high Edding-ton ratios at z < . AGN Host Galaxies
We construct our parent AGN sample by making useof the catalogs provided by the Hamburg/ESO survey(HES, Schulze & Wisotzki 2010), the Palomar GreenSurvey (PG, Vestergaard & Peterson 2006), and theSDSS DR7 (Shen et al. 2011). We constrain our selec-tion of potential targets to sources with a redshift of z < .
2. Since we require an estimate of the central BH mass
Marian et al. log( edd ) l o g ( M B H / M ) HESSDSS DR7PGSelected AGNs M acc (M yr ) Figure 1.
Left : Eddington ratio λ edd = L/L edd vs. black hole mass for the parent sample of AGNs at redshift z < . Right : Black hole mass accretion rate vs. black hole mass for the same sample indicating that our final selection consistsof AGNs possessing the highest specific accretion rates. and are interested in the AGNs with the highest specificaccretion rates, we only select unobscured broad-lineAGNs with an Eddington ratio λ edd = L/L edd > .
3. Toderive λ edd , we use the BH mass determinations basedon single epoch H β measurements and the bolometric lu-minosities, which, in turn, are based on the luminositiesat 5100 ˚A multiplied by a bolometric correction factor of k bol = 9 (Schulze & Wisotzki 2010; Netzer 2019). Boththe BH masses and luminosities at 5100 ˚A are takenfrom the respective catalogs.We apply a minimum BH mass threshold oflog( M BH / M (cid:12) ) = 7 .
7, which results in a median BHmass for our AGN sample of log( M BH / M (cid:12) ) ∼ .
0. Us-ing the M BH − M bulge scaling relation of Kormendy & Ho(2013) as a proxy to predict stellar host galaxy masses,the corresponding median stellar mass for our AGN hostgalaxies yields log( M ∗ / M (cid:12) ) ∼
11. This mass selec-tion results in feasible exposure times for our inactivegalaxies, which are required to be of equal stellar mass,and enables us to compare the results presented in thiswork with the findings of (Marian et al. 2019), which arebased on similar stellar host masses. Furthermore, weonly select targets with a declination of dec < +15 ◦ forbetter visibility with the VLT . All of these constraintsyield a total number of 19 suitable AGN host galax-ies, of which we observe 17 with
VLT FORS2 in V - and B -band (ESO programs 091.B-0672(A), 095.B-0773(A),and 098.A-0241(A), PI: Knud Jahnke). The median red-shift of these 17 sources lies at z = 0 . λ edd > . η with low accretion rates (Churazov et al. 2005; Wein-berger et al. 2018; Nelson et al. 2019) and can calculatethe BH mass accretion rates ˙ M acc (Fig. 1, right panel)as ˙ M acc = L/ηc , (1)where we define L as the derived bolometric luminositiesand assume an efficiency parameter η = 0 .
1. The rightpanel of Fig. 1 highlights that we target the AGNs withthe highest specific accretion rates, i.e. those with thehighest absolute mass accretion rates relative to theirBH masses.Each target has been observed for at least three longexposures, to detect large-scale distortion features downto B and V ∼ . / arcsec , and three short ex-posures, for an unsaturated image of the bright centralregion. The actual individual exposure times amountto 430 s and 14 s for B and 150 s and 8 s for V , respec-tively. In Table 1, we summarize the properties of ourAGN sample. We cite the corresponding catalog desig-nations, redshifts, apparent I -band magnitudes, as wellas the luminosities at 5100 ˚A, L , and the bolometricluminosities, derived by applying a correction factor of 9to L (Schulze & Wisotzki 2010; Netzer 2019). In ad-dition, we state the catalog values for the FWHM of thesingle-epoch measurements of the (broad) Hβ line, therespective BH masses M BH , along with the calculatedEddington ratios λ edd and mass accretion rates ˙ M acc .2.2. Inactive comparison sample
Given the size of the AGN sample and our assump-tions for the merger fractions for our AGN and controlsample ( f m , agn ≥ . f m , ina = 0 .
15) we need to ob-serve at least 25 inactive galaxies to meet our criterion ajor merger rate for AGNs with the highest Eddington ratios at z < . Table 1.
AGN sample propertiesAGN designation z m I L L bol FWHM M BH λ edd ˙ M acc mag erg s − log(L (cid:12) ) H β (km s − ) log(M (cid:12) ) M (cid:12) yr − (1) (2) (3) (4) (5) (6) (7) (8) (9)HE0119–2836 0.12 14.8 44.92 12.29 3363.00 8.2 0.36 1.3HE0132–0441 0.15 15.8 44.81 12.18 1719.00 8.0 0.44 1.0HE0157+0009 0.16 16.1 44.73 12.10 2369.00 7.8 0.60 0.9HE0444–3449 0.18 16.0 44.83 12.20 1714.00 8.1 0.35 1.1HE0558–5026 0.14 15.5 44.88 12.25 1583.40 8.0 0.51 1.2HE1201–2408 0.14 16.8 44.45 11.82 1820.86 7.8 0.33 0.4HE1226+0219 0.16 13.2 45.89 13.26 3835.03 8.8 0.82 12.1HE1228+0131 0.12 14.4 44.93 12.31 1866.19 8.1 0.43 1.4HE2011–6103 0.12 16.3 44.53 11.90 2862.51 7.9 0.32 0.5HE2152–0936 0.19 14.2 45.56 12.93 2183.42 8.7 0.52 5.8HE2258–5524 0.14 15.9 44.68 12.05 2419.42 7.8 0.54 0.8PG1001+054 0.16 16.3 44.74 12.11 1700.00 7.7 0.76 0.9PG1012+008 0.19 16.2 45.01 12.38 2615.00 8.2 0.45 1.6PG1211+143 0.09 14.3 45.07 12.44 1817.00 8.0 0.81 1.9SDSS-J032213.89+005513.4 0.18 16.1 44.72 12.09 2440.00 8.0 0.33 0.8SDSS-J105007.75+113228.6 0.13 15.7 44.57 11.94 1906.00 7.8 0.45 0.6SDSS-J124341.77+091707.1 0.19 16.8 44.41 11.78 1979.00 7.7 0.36 0.4 Note —Properties of the AGNs in our sample: Columns 1–3, 6, and 7 are taken from the respective catalogs(Vestergaard & Peterson 2006; Schulze & Wisotzki 2010; Shen et al. 2011). The bolometric luminosities L bol in column 5 are calculated by applying a bolometric correction factor of 9 to L (Schulze & Wisotzki 2010;Netzer 2019). Column 6 presents the FWHM of the broad component of H β . We calculate the Eddingtonratios λ edd and black hole mass accretion rates ˙ M acc in column 8 and 9 by using the bolometric luminosities L bol , the respective BH masses M BH , and a radiative efficiency parameter of η = 0 . of detecting a difference in those merger fractions with ∼
99% confidence. The comparison galaxies are ran-domly chosen from a parent sample of ∼ < ◦ and the redshift to z < .
2, resulting in amedian redshift of z ∼ .
13 for our control sample. Fur-thermore, we only choose sources that possess compara-ble stellar masses to our AGN host galaxies.As described in Section 2.1, we adopt the M BH − M bulge scaling relation of Kormendy & Ho (2013) to de-rive the median stellar host mass for the AGN samplefrom the inferred BH masses. We restrict the inactivegalaxies to a small range around the median derivedstellar mass of the AGN host galaxies, log( M ∗ / M (cid:12) ) =11 ± .
01. Finally, we vet all potential sources againsthard X-ray AGN signatures (Baumgartner et al. 2013),to remove any galaxies with a hidden, obscured AGN.In Table 2 we provide the coordinates, redshifts, k - corrected and dereddened I -band magnitudes, and me-dian stellar masses from the MPA-JHU catalog for ourcomparison galaxies.With the exception of one source , all of the 25 galax-ies in our final sample were observed in the B - and V -band with a comparable observational setup as forour AGN host galaxies. Each target has been observedwith at least three individual, 470 s and 180 s long, ex-posures in B and V , respectively. This selection and ob-servational approach enables us to analyze two distinctsamples of AGN host galaxies and inactive comparisongalaxies, which are nonetheless matched in redshift, stel-lar (host) mass, depth, spatial resolution, filter band andS/N. Thus, we can directly compare potential relativedifferences in the merger fractions of both populations. Due to weather losses one target was only observed in V -band Marian et al.
Table 2.
Comparison galaxy sample propertiesGalaxy designation α (J2000) δ (J2000) z m I M ∗ deg deg mag log(M (cid:12) )(1) (2) (3) (4) (5) (6)Gal000232 0.164 − − − − − − − − − − − − − Note —Our designations (column 1), coordinates (columns 2 and 3),redshifts (column 4) k -corrected and dereddened I -band magnitudes(column 5), and photometric median stellar masses for the inactivegalaxies in our comparison sample taken from the MPA-JHU catalog(Kauffmann et al. 2003; Brinchmann et al. 2004). Data reduction and preparation
We require a seeing of 1 (cid:48)(cid:48) or better to diagnose large-scale merger signatures at a minimum required spatialresolution of ∼ Astropy package photutils (Bradley et al. 2019),and calculating the corresponding median FWHM of allsources. We visually check and re-measure every sin-gle exposure with a median FWHM > (cid:48)(cid:48) and discard individual exposures with a median FWHM above thisthreshold. Out of a total of ∼
450 individual frames,we reject 22 from the subsequent reduction process andanalysis. Despite the exclusion of these images, we endup with at least three individual exposures per band forevery object.To execute all the initial data reduction steps, i.e. thebias and flat-field correction, sky background subtrac-tions, astrometry and aligning, and combination of in-dividual exposures, we use the data processing pipeline ajor merger rate for AGNs with the highest Eddington ratios at z < . THELI (Erben et al. 2005; Schirmer 2013). The re-sulting pixel scale of 0 . (cid:48)(cid:48)
252 corresponds to ∼ B - and V -band observations to create color images using Multi-ColorFits (Cigan 2019).To ensure that the samples are directly comparable,we mimic the appearance of the AGN host galaxies inthe images of the inactive galaxies by adding a syntheticpoint source on top of the respective flux centers. Tothis end, we first detect the 15 brightest, unsaturatedstars within the central image regions around each in-active galaxy with the help of the DAOStarFinder algo-rithm within the photutils package. For each galaxy,we then visually select and cut out one of the detectedstars, and upscale the brightness correspondingly, suchthat they possess a central brightness comparable toHE2152–0936, our second brightest AGN source. In thecourse of this procedure we also downscale noise in theouter parts. Since an upscaling with a constant factorwould lead to a noticeable discrepancy in flux betweenthe galaxy and the edge of the artificially enhanced pointsource, we fit the original point sources with a 2D Gaus-sian and determine a circular region centered around thebrightest pixel with a radius of 5 σ . We divide this re-gion into five bins and upscale the pixel values dependingupon which bin they lie in. For the innermost region,i.e. within 1 σ , we upscale with the total scaling factor,whereas for the outermost region, i.e. between 4 and5 σ we apply a scaling factor lower by 5 × − . Forthe intermediate bins we choose a multiple of the scal-ing factor such that the distribution of the scaling factorwith radius follows a Gaussian function. Using this ap-proach we create point sources that resemble the centralregions of our AGN host galaxies, but also blend in un-recognizably and smoothly into the respective galaxies.We add these point sources randomly at the centroid ofeach inactive galaxy, mimicking the appearance of ourAGN host galaxies. Our point sources have a similarsize to the upper limit of ∼ (cid:48)(cid:48) set on the seeing, whereasthe typical diameter of our sample galaxies, both AGNand inactive, is of the order of 5–6 (cid:48)(cid:48) . Thus, in contrastto our study of highly-accreting AGNs at z ∼ V - and B -band images, re- https://multicolorfits.readthedocs.io spectively, whereas the right column (c) shows the colorimages. To optimize the visibility of large-scale struc-tures and possible merger signatures, while blending outthe brightest inner regions, we chose different parame-ters for the color cuts and color map for the single bandimages as well as the color images. However, within oneset, i.e. V -, B -band or color images, the parameters areconstant. In addition, we adopted a Gaussian 2-pixelsmoothing for the color images only. Due to the dif-ferent visualization of the sources, we can test for anysystematic differences in the subsequent distortion rank-ings or the resulting merger fractions (see Section 4). MORPHOLOGICAL ANALYSIS & MERGERFRACTIONSWe join both processed samples (for which the galax-ies can no longer be visually separated as AGN or not)resulting in a final sample of 42 sources in V - and 41in B -band and color, respectively. To derive the mergerfractions, 19 experts , proficient in working with imag-ing data of galaxies, perform a visual assessment of thetargets, ranking them from most to least distorted withrespect to the appearance of large scale distortion fea-tures. These features are indicative of ongoing or recentmajor merger events. Each set of V -, B -band and colorimages is ranked independently by each expert. We notethat there are an increasing number of machine learningalgorithms that can classify galaxies, based on their mor-phologies and possibly merger state (e.g. Bottrell et al.2019; Cheng et al. 2019; Snyder et al. 2019). However,we rely on the human interpretation and judgment dueto the manageable sample size and the extensive timeand logistic requirement to teach an automatic classifi-cation routine with a matching ‘external’ training set.Since the sources in the joint sample are indistinguish-able with respect to whether or not they are active, ev-ery expert’s individual bias regarding the classificationof a major/minor merger applies equally to AGN hostgalaxies and comparison galaxies. Thus, in our subse-quent analysis any personal subjectivity in classificationwill have the same impact on either of the two subsam-ples. To further reduce any systematic bias, the datasetprovided to each of the 19 ranking experts is random-ized. As an additional task we request every classifierto choose a “cut-off” rank below which they deem allsources to be in a merging state, or, to at least show signsof a recent gravitational disturbance, like asymmetries, The rankings were done by the coauthors Andika, Ba˜nados, Ben-nert, Cohen, Husemann, Jahnke, Kaasinen, Koekemoer, Mar-ian, Onoue, Schindler, Schramm, Schulze, Silverman, Smirnova-Pinchukova, van der Wel, Villforth and Windhorst
Marian et al. (a) (b) Gal510223 (c) HE2011-6103 Figure 2.
Two sources representative for our targets. On the top row we show one of the comparison galaxies and on thelower row one of our AGNs is displayed. From left to right we present a postage stamp in (a) V-band, (b) B-band and (c) color,respectively. Note: In order to enhance the visibility the images are not shown with the same cuts and color map parameters. tidal tails or double nuclei. Every galaxy with a rankhigher than the cut-off is interpreted to be completelyfree of major disturbances stemming from interactions.In our ensuing analysis we will use this property to deter-mine the merger fractions of our two samples and alsodiscuss the dependence of those fractions on differentcut-off ranks (see Sect. 4).We combine the 57 individual rankings (19 expertstimes three sets) into three consensus sequences for eachrespective set. We apply the same methods as in Marianet al. (2019) to combine the individual rankings and re-peat this task for each set, i.e. separately for B -, V -band,and color images. For our first approach we calculateand weigh the average ranks of each galaxy, whereas forour second and third approach we use the Borda count(Emerson 2013) and Schulze algorithm (Schulze 2011,2018), respectively. More information on the differentmethods and on how we implement them are providedin Appendix A. Ultimately, by applying all three meth-ods to all three sets we obtain nine overall rankings. We select various cut-off ranks and split the combinedrankings back into AGN host and comparison galaxies.Subsequently, we derive the merger fractions for eachchosen cut-off rank by counting how many active and in-active galaxies are above and below this threshold. Themerger fraction is then simply defined as, f m = aa + b , (2)where a represents the number of merging galaxies,whereas b counts the sources that are undisturbed. How-ever, since we only examine samples of limited size, weneed to quantify the probability densities and uncertain-ties introduced by the shot noise for our resulting mergerfractions. Based on those two parameters a and b we canquantify the probability densities for a continuous rangeof merger fractions in the feasible interval [0 ,
1] by us-ing the beta distribution (see also Mechtley et al. 2016;Marian et al. 2019), f ( x ) = ( a + b + 1)! a ! b ! x a − (1 − x ) b − . (3) ajor merger rate for AGNs with the highest Eddington ratios at z < . AverageBordaSchulze ±0.10 ±0.09 ±0.09 ±0.10 ±0.09 ±0.09 ±0.10 ±0.09 ±0.09 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12 ±0.12
B V ColorAverageBordaSchulze ±0.00 ±0.04 ±0.04 ±0.00 ±0.04 ±0.04 ±0.00 ±0.04 ±0.04
B V Color ±0.06 ±0.05 ±0.06 ±0.06 ±0.05 ±0.06 ±0.06 ±0.05 ±0.06
B V Color ±0.09 ±0.08 ±0.09 ±0.09 ±0.08 ±0.09 ±0.09 ±0.08 ±0.09
B V Color ±0.10 ±0.10 ±0.10 ±0.10 ±0.10 ±0.10 ±0.10 ±0.10 ±0.10 M e r g e r f r a c t i o n Cut-off rank = 5 Cut-off rank = 10 Cut-off rank = 15 Cut-off rank = 20AGN host galaxiesComparison galaxies
Figure 3.
The merger fractions for every set ( B − , V − band, and color images) and ranking combination method (Average,Borda, Schulze) for four distinct cut-off ranks. In the top row we show the corresponding fractions of disturbed AGN hostgalaxies, the bottom row depicts analogously the inactive comparison galaxies. The smaller numbers below the actual mergerfraction values give the standard deviations (i.e. 1 σ ) of the corresponding beta distributions. The respective standard deviations and means of theassociated merger fraction probability distributions arethen derived by, σ ( x ) = (cid:115) ab ( a + b ) ( a + b + 1) , (4)and Eq. 2, respectively.In Figure 3, we present the corresponding means andstandard deviations of the various probability distribu-tions for every combination of method and set for fourdistinct cut-off ranks at 5, 10, 15, and 20. The mergerfractions increase with cut-off rank, because a highercut-off rank means that more galaxies are below thislimit and are thereby considered to exhibit merger fea-tures. We find no evidence that the choice of combina-tion method or the choice of B -, V -, or color image-setaffect the resulting merger fractions. For all combina-tions the results for a given sample and cut-off rank arewell within the errors of each other or even equal. How-ever, it is also evident that for cut-off ranks (cid:46)
15 themerger fractions for the AGN host galaxies (Fig. 3, up-per row) are significantly larger then the fraction of dis-turbed inactive galaxies (Fig. 3, bottom row). This isnot the case for larger cut-off ranks. We discuss the im-plications of the chosen cut-off ranks on our recoveredmerger fractions and the potential causal connection be- tween major mergers and the triggering of AGNs in thefollowing section.3.1.
Constraining the absolute merger fractions
We have calculated the merger fractions for two sam-ples of 17 AGN host galaxies and 25 inactive comparisongalaxies. As mentioned in the preceding section, the fi-nal merger fractions depend on the choice of cut-off rank.In Appendix B we present the continuous evolution ofmerger fractions with cut-off rank for all combinationsof set and method, while in this section we describe thetwo approaches we used to analyze and interpret our re-sults. Firstly, we base the cut-off rank on our experts’opinions, and secondly, we construct this limit so thatthe resulting merger fraction of our inactive control sam-ple is consistent with the merger rates presented in theliterature. To obtain a valid first estimate, we calculatedthe means of the individual cut-off ranks chosen by eachclassifying expert for each set. The average cut-off ranksare 21 ±
8, 22 ±
9, and 18 ± B , V , and color sets,respectively.We suspect that the reason for such high cut-off ranks,which are almost bisecting our joint samples, lies in thevisual determinations of our experts. Since our galax-ies are well-resolved, any minor asymmetries (which donot need to stem from a recent major merger event,0 Marian et al. but can be of a minor merger or secular origin) can beeasily identified. This leads our experts to put thoseparticular sources into the ‘merger bin’, i.e. below thecut-off rank, increasing the percentage of galaxies clas-sified as merging. With a corresponding cut-off rank =20, the merger fractions range between f m , agn = 0.41 ± f m , agn = 0.53 ± f m , ina = 0.40 ± f m , ina = 0.50 ± major mergers on the formation of AGNs,without considering the effects of minor gravitationalencounters or other processes shaping the morphologyof a galaxy. Thus, we have to correct our recoveredmerger fractions for the contamination by sources withminor asymmetries. Such a high merger rate of ∼ R m ∼ .
05 [Galaxy − Gyr − ]. This number repre-sents the number of galaxies currently in a merger state,divided by the timescale of the visibility of merger sig-natures. In order to obtain an absolute merger frac-tion, representative of our comparison sample, we needto multiply this rate with the timescale T m in which amajor merger is observable. This property not only de-pends strongly on the mass ratio, individual masses, andgas fractions of the two progenitor galaxies, but also onthe depth of the observations. Considering our targets’low redshifts and surface brightness limits we choose acomparatively conservative value of t m ∼ . f m ∼ .
08 for galax-ies in our mass bin and at our sample’s redshift.Such a value for the merger fraction for our compar-ison galaxies corresponds to a cut-off rank = 10. Co-incidentally, at this cut-off rank the respective mergerfractions are equal over all sets and methods for eachof the two samples (Fig. 3) and yield f m , agn = 0.41 ± f m , ina = 0.08 ± f m , ina is notonly in excellent agreement with the major merger ratesfound in the 3DHST survey by Man et al. (2016) forall five fields (AEGIS, COSMOS, GOODS-N, GOODS- P r o b a b ili t y d e n s i t y AGNsComparison galaxies
Figure 4.
Probability distributions for the derived mergerfractions of our z < .
2, high-accretion AGN host galaxies(blue) and inactive galaxies (red) at a cut-off rank = 10.The solid and dotted lines show the means and modes ofthe respective merger fractions, while the dashed lines andshaded regions depict the central 68% confidence intervals.At this particular cut-off rank the respective merger fractionsare identical, independent of method and set.
S, UDS) in CANDELS (Grogin et al. 2011; Koekemoeret al. 2011), but also for the major merger fractions re-covered by MUSE deep observations (Ventou et al. 2017,2019) as well as studies by Duncan et al. (2019) in CAN-DELS, and in GAMA by Mundy et al. (2017).The two corresponding probability distributions areshown in Figure 4, with blue and red denoting the prob-ability distributions for the AGN sample and the com-parison sample, respectively. The shaded regions rep-resent the 1 σ intervals and the solid and dotted linesdepict the corresponding means and the modes. Dueto the low number of merging comparison galaxies theassociated probability distribution appears considerablyskewed with the corresponding mean not coinciding withthe peak position. Thus, we also report the merger frac-tion associated with the mode of the distribution, whichyields f m , ina ∼ DISCUSSION4.1.
Robustness of results
For a cut-off at rank 10, the resulting merger fractionstranslate to,– 7/17 AGN host galaxies showing merger features,and, ajor merger rate for AGNs with the highest Eddington ratios at z < . C and D ). The same sourcesthat occupy the first eight ranks in the nine initial con-sensus sequences, populate the highest positions in thisfinal ranking as well. Therefore, we obtain an unchangedresult for both merger fractions after again applying acut-off at rank 10.Considering the appearance of seven AGN host galax-ies among the eight highest-ranked sources and the clearexcess in merger fractions for the AGN host galaxieswith respect to the inactive sample with a significantdifference of > . σ , we conclude that major mergersare an essential triggering mechanism for AGNs with thehighest Eddington ratios at z < .
2. However, based onthe mean of our recovered probability distribution forthe AGN merger fraction we only find a ∼
22% proba-bility that the merger fraction is above the threshold of f m , agn = 0 .
5. This means that although major merg-ers are indeed a non-negligible mechanism in triggeringour specific population of AGNs, more than half of theBHs must be activated by different means, like secularprocesses or minor mergers. We discuss the role of thelatter in triggering AGNs with the highest specific ac-cretion rates at low redshifts in more detail in Sect. 4.5.4.2.
Comparison to previous studies
Our result, which shows an excess in AGN mergerfraction compared to a matched control sample, standsin contrast to recent simulations (Steinborn et al. 2018;Ricarte et al. 2019) and several previous empirical stud- ies examining the potential causal connection betweenmajor mergers and the triggering of different popula-tions of AGNs. Villforth et al. (2014) found no increasein merger signatures with luminosity and also reportedconsistent disturbance fractions between the AGNs andcomparison galaxies for their sample of observed low-and moderate-luminosity AGNs (41 (cid:46) L X [erg s − ] (cid:46) .
5) at 0 . (cid:46) z (cid:46) .
8. In contrast, Silverman et al.(2011) found an enhanced merger rate for AGNs of mod-erate X-ray luminosities in spectroscopic pairs at z < . +8 . − . % is still significantlylower than what we find here.AGNs and host galaxies at comparable redshifts andluminosities as our sample were explored by Bhm et al.(2013) and Grogin et al. (2005). They assessed theneighboring counts, asymmetries, and various morpho-logical indices (concentration, Gini coefficient and M index) to characterize the respective host galaxies, butfound no significant causality between major mergersand AGNs. Likewise, Allevato et al. (2011), Schawin-ski et al. (2011), and Rosario et al. (2015) detected noredshift evolution of morphological properties for similarAGNs up to z ∼ . +5 . − . % of comparable AGNs at z ∼ (cid:46) L X [erg s − ] (cid:46) . (cid:46) z (cid:46) . ∼ z ∼ M BH = 10 − M (cid:12) ) and a sampleof 84 matched inactive galaxies. Similarly, in our previ-ous work (Marian et al. 2019) in which we examined 21AGNs with the highest Eddington ratios ( λ edd > . z ∼ z < .
05 Koss et al. (2010) reported an enhanced mergerfraction of 18% when compared to a matched controlsample, in which only 1% of the sources display mergerfeatures. However, they speculated that their AGNsmay not be classified correctly via means of optical di-agnostics due to superimposing features of ongoing star2
Marian et al. formation and optical extinction. In fact it appears that,independent of redshift, obscured and luminous AGNsare more likely to be connected to major merger events.Albeit, it should be noted that this is expected sinceby focusing on obscured sources a bias towards mergingsystems is most likely introduced as that obscurationmay be due to dust within a merging (U)LIRG-like host.With this caveat in mind, Glikman et al. (2015), Fanet al. (2016), and Donley et al. (2018) detected mergerfractions >
50% for such reddened or obscured AGNssources at z ∼ z ∼
3, and 0 < z <
5, respectively.Also at low redshifts ( z (cid:46) .
2) Koss et al. (2018) andEllison et al. (2019) presented comparable results. Inaddition, in the latter study the authors described an in-crease of merger fraction with AGN luminosity, with themost luminous AGNs exhibiting the highest merger inci-dence. Corresponding findings have also been reportedby Treister et al. (2012), Hong et al. (2015) and Gould-ing et al. (2018), who have analyzed luminous AGNs(log( L bol [erg s − ]) >
45) at various redshifts. Especiallywith a merger fraction of ∼
44% for luminous AGNs at z < . < z < .
6, who detected amerger fraction of ∼
40% and a general increase of dis-tortion incidence with stellar mass. Finally, McAlpineet al. (2018, 2020) reported for the
EAGLE simulationthat major mergers – while of no great importance athigh redshifts – play a significant role at low redshiftsand present a consistent major merger fraction of ∼ z ∼ Physical interpretation and comparison to AGNcounterparts at z ∼ In light of our previous work at z ∼ z ∼ z < .
2, we have to factor in the impact of surface bright-ness dimming on detecting possible faint morphologi-cal distortion features. With a drop in surface bright-ness of ∼ at z ∼
2, we miss at this redshiftmost definitely merger features we otherwise would seeat z ∼ .
2. This effect can be enhanced by the factthat galaxies at z ∼ z ∼ z ∼ z ∼ .
2. In the latter casethe starburst may happen as much in the galaxy’s outerspiral arms and tidal streams, whereas the starburst in agalaxy at z ∼ z ∼ z ∼ .
2, a more complex situation is possiblewhere the visibility of an AGN host galaxy at z ∼ z ∼ f m , agn = 0.24 ± z ∼ f m , agn = 0.41 ± z ∼ f m , ina = 0.19 ± f m , ina =0.08 ± z ∼ z ∼ . z ∼ z ∼
2, which would implicitly include such sources,but found no significant effect on the resulting mergerrates. In addition, just as in this work, we deliberatelyhave only selected type-1 AGNs, minimizing the proba- ajor merger rate for AGNs with the highest Eddington ratios at z < . z ∼
2. A rigorous analysis would require a larger sam-ple and data at longer wavelengths, as the James WebbSpace Telescope (JWST) will be able to provide at z ∼ z ∼
2, but a clearexcess of the AGN merger fraction when compared toinactive galaxies at z < . z ∼ z < z ∼ z ∼ z < EAGLE sim-ulations, which sees major mergers in a negligible rolefor triggering AGNs at high redshifts, but shows thatsuch galaxy encounters play a substantial role at lowredshifts, yielding comparable major merger fractions(McAlpine et al. 2018, 2020). However, it should benoted that despite the excess in major merger fractionfor our AGN host galaxies, (cid:38)
50% of our sample appear not to be not triggered by such an event, requiring analternative explanation for the existence of such AGNs.4.4.
AGN merger fraction and luminosity
Although our AGN sources can be considered lu-minous for sources at z < L bol [erg s − ]) > . . (cid:46) log( L bol [erg s − ]) (cid:46)
46, but feature the smallest BHmasses in that luminosity bin (7 . < log( M BH / M (cid:12) ) < . ∼
10 more luminous AGNs in our threeinitial parent catalogs would have been selectable. Un-like other studies, which detect an enhanced merger ratefor luminous AGNs we see no trend of the strength ofthe merger features – i.e. rank – with either BH massor BH mass accretion rate/luminosity within our AGNsample (Fig. 5). In fact HE1226+0219 and HE2152–0936, both with distinctly higher absolute mass accre-tion rates with respect to our other sample AGNs, onlyoccupy the ranks ∼
30 and ∼
25 in all the consensusrankings and show clearly no significant merger features.However, due to our selection of AGNs being based on acombination of BH mass and Eddington ratio, we notethat apart from the two aforementioned most luminousAGNs our sources sample a relatively narrow luminosityrange. Still, because of the lack of an obvious correla-tion of merger fraction with AGN luminosity, our resultsrequire an alternative explanation – especially consider-ing that the existence of such a trend is still inconclu-sive. Despite some studies have found evidence of such alink between merger rate and luminosity (Treister et al.2012; Fan et al. 2016; Goulding et al. 2018) others didnot (Villforth et al. 2014, 2017; Hewlett et al. 2017).4.5.
The (un)importance of minor mergers
In Sect. 3.1 we argue that the initial high merger frac-tion of our sample of control galaxies, is the result ofour experts including galaxies in the merger category,which show features that are only the consequence ofminor merger events. Lotz et al. (2011) state that theminor merger rate is ∼ < M sat /M primary (cid:46) total merger fraction of f m , ina = 0.33 ± Marian et al. l o g ( M B H / M ) M a cc ( M y r ) l o g ( L b o l [ e r g s ]) Figure 5.
Overall consensus rank vs. black hole mass (top)and black hole mass accretion rate and bolometric luminos-ity (bottom) for our sample of AGNs. The vertical dashedline visualizes a cut-off at rank 10, which was used in ourdiscussion. cut-off at a rank of 17 and in turn in a total merger frac-tion of f m , agn = 0.47 ± Considering AGN and merger timescales
With major mergers only triggering at most ∼
50% ofour AGNs and minor mergers playing a subdominantrole the question still remains which process(es) are re-sponsible for triggering high Eddington rate AGNs at z < .
2. With that question in mind and a diminishingnumber of alternative mechanisms we consider a possi-ble impact of the different timescales. Previous studies,which have found no enhancement in distortion fractionsbetween AGNs and a matched sample of control galax-ies, analyzed a potential disparity in AGN and mergerlifetimes to be an explanation for their results (Cister-nas et al. 2011; Mechtley et al. 2016; Marian et al. 2019).The unanimous conclusion is that the difference in lifecycles is not sufficient to explain the lack of excess inmerger rates, since the timescale of merger features be-ing observable is much longer than the lifetime of therespective AGNs.We consider a scenario in which some of the galax-ies that host no visible AGN and feature only minordistortions are actually the result of a major mergerevent which also lead to a past phase of active blackhole growth. However, since the lifetime of AGNs canbe significantly shorter when compared to that of majormerger features, the only detectable remains of such agravitational encounter would be in the form of minorasymmetries. This implies that if we utilize the total merger fractions we derived in the previous subsection,a part of the 33 ±
9% inactive galaxies that show distor-tions of various strength have actually hosted a majormerger triggered AGN in the past. As a result the AGNmajor merger fraction with f m , agn = 0.41 ± z < . ajor merger rate for AGNs with the highest Eddington ratios at z < . f m , ina & agn by adoptingthe formula presented in Marian et al. (2019): f m , ina & agn = f agn × f m , agn × t m t agn . (5)Here, f agn and t agn represent the fraction and life-time of AGNs with an Eddington ratio >
30% withrespect to the total galaxy population at our redshiftand mass bin. The timescale in which the merger fea-tures are observable is given by t m , while f m , agn de-scribes the total merger fraction of our specific AGNpopulation. We derive f agn by utilizing the numberdensities provided by stellar mass and quasar bolomet-ric luminosity functions at z ∼ I -band magnitudes this yieldslogΦ ∼ − . − mag − for the total galaxy pop-ulation (Hirschmann et al. 2014; Henriques et al. 2015;Furlong et al. 2015; Lacey et al. 2016; Pillepich et al.2018) and logΦ ∼ − . − mag − for our particu-lar population of AGNs (Hopkins et al. 2007; Fanidakiset al. 2012; Hirschmann et al. 2014; Sijacki et al. 2015),resulting in f agn ∼ . × − , which is in excellent agree-ment with the value for the active fraction reported bySchulze & Wisotzki (2010) for BHs at a redshift z < . M BH / M (cid:12) ) ∼
8. For f m , agn we use ourreported value of f m , agn = 0 . ± .
12, but also repeatour calculations for f m , agn = 0 .
30 and 0 .
70. Besides ourinitial estimate of t m = 1 . × yr (see Sect. 3.1), inaddition, we use t m = 10 yr for comparison. Finally, inaccordance to previous studies we constrain our AGNlifetime t agn to a range between 10 and 10 yr (Mar-tini 2004; Hopkins et al. 2005; Shen et al. 2007; Hopkins& Hernquist 2009; Conroy & White 2013; Cen & Sa-farzadeh 2015). Since we can not distinguish between in-active merging galaxies that already went through theirAGN phase, are yet to host an AGN or are currentlyhosting an intermittent AGN, it is not necessary for usto consider any time lag (Hopkins et al. 2006b; Wildet al. 2010; McAlpine et al. 2020) between the onset ofthe actual phase of active black hole growth and thebeginning/coalescence of the merger. Hence, from theperspective of timescales our result solely depends onthe relative difference between the AGN and merger life-times and thus we have to consider our fraction of in-active merging galaxies, which have hosted an AGN tobe an upper limit. However, a visual re-examination re-turned only a low number of galaxies with asymmetries actually having a close companion. Therefore, we con-clude that most of the distorted galaxies are already inthe late stages of their merging process, indicating that,if at all, they already experienced a potential AGN phasewith a low chance of an intermittent AGN becoming ac-tive again.The total merger fraction of our inactive galaxies,which amounts to f m , ina ∼ .
35, serves as an upperbound for f m , ina & agn . Both parameters being equalwould imply that all distorted, inactive galaxies havehosted (or will host) an AGN. Conversely, f m , ina & agn =0 would correspond to no such galaxy ever hosting anAGN. In Fig. 6 we present the results of our compu-tations for different f m , agn and t m = 10 yr (left) and1 . × yr (right). The blue lines and the shaded re-gions denote the results for our retrieved AGN mergerfraction and the corresponding 1 σ intervals, while the vi-olet and yellow lines display the trend for f m , agn = 0 . .
70, respectively. The fraction of merging inac-tive galaxies hosting an AGN at some point during themerging process increases with shorter AGN lifetimes.In addition, for a given period of AGN activity thisshare grows with longer merger timescales and largerAGN merger fractions, both due to an enhanced prob-ability to find a distorted galaxy actually hosting anAGN. Depending on the merger timescale and assum-ing the lower limit of our AGN merger fraction is cor-rect, we can deduce a lower bound for the AGN lifetimeby considering every inactive distorted galaxy to hostan AGN, i.e. f m , ina & agn ≡ f m , ina . The life span of anAGN corresponds then to a minimum of 1 . × yr and1 . × yr for merger timescales of 10 yr and 1.5 × yr,respectively (Fig. 6, dotted lines).However, based on the best estimates for accretionrate histories we have today (Di Matteo et al. 2005; Jo-hansson et al. 2009a,b; Hopkins & Quataert 2010; Junget al. 2018), we fix the time period in which an AGNaccretes above λ edd > . t agn = 10 yr. The in-ferred fractions of inactive merging galaxies that alsohost an AGN at any given time yield then f m , ina & agn =0 . +0 . − . and 0 . +0 . − . for t m = 10 yr and 1.5 × yr,respectively (Fig. 6, dashed lines). So, adding even theupper limit of this fraction onto the AGN major mergerrate we derived in Section 3.1 this only results in a re-vised AGN major merger fraction, which is barely abovethe threshold of 0.5, which in turn would indicate thatthe majority of AGNs is triggered by major mergers.This result still leaves ∼
50% of AGNs to be of un-known origin. Only by assuming a significantly lowerAGN duty cycle of t agn ∼ yr and thus regarding al-most every distorted inactive galaxy hosting an AGN,we can obtain AGN major merger fractions of ∼ Marian et al. t agn f m , i n a & a g n t m = 10 yrf m , agn = 0.30f m , agn = 0.47f m , agn = 0.70 t agn t m = 1.5 × 10 yr t m / t agn f m , i n a & a g n / f m , i n a t m / t agn Figure 6.
Total fraction of merging inactive galaxies that hosted an AGN in the recent past f m , ina & agn in dependence of theAGN life time t agn for a merger timescale t m of 10 yr (left) and t m =1.5 × yr (right). The blue line including the shadedregion represents our result of the AGN total merger fraction of f m , agn = 0.47 ± f m , agn = 0.30 and 0.70, respectively. The dotted line corresponds to a lower limit of t agn , the dashed lines display the resulting f m , ina & agn ∼ .
09 for an assumed t agn = 10 . which would then leave no doubt about the role of ma-jor mergers and the triggering of high Eddington rateAGNs at z < .
2. Hence, we conclude that neither adifference in AGN and merger timescales nor the poten-tial presence of intermittent AGNs affect significantlyour derived AGN merger rate. In order to better con-strain our inferred estimates, more detailed simulationspredicting especially AGN timescales in dependence ofaccretion rate are imperative. SUMMARY & CONCLUSIONSWe examined a potential direct connection betweenAGNs specifically exhibiting the highest Eddington ra-tios and major mergers at z < V , B and colorimages. We adjusted our control galaxies by adding ar-tificial point sources on top of their flux centers, whichyielded two indistinguishable samples, that were joinedto create a randomized overall sample of 42 targets. Thisoverall sample was ranked according to the presence ofmerger features (from most to least distorted) by 19 ex-perts. We combined the individual rankings of each set,i.e. V , B and color, by applying three different methods,resulting in a total number of nine consensus rankings.This allowed us to determine any bias, which might beintroduced by visually classifying the galaxies at differ-ent wavelengths or the algorithm to combine the individ-ual classifications. Finally, we also created one overallsequence by combining the nine initial consensus rank- ings. We divided all rankings into; 1) galaxies show-ing distinct merger features and 2) galaxies showing nosigns of a gravitational disturbance, by choosing specificcut-off ranks. As a final step we derived the respectivemerger fractions by counting the numbers of active andcontrol galaxies above and below these particular limitsand applying those quantities to a beta distribution.Our findings depend heavily on the choice of distinc-tion between merging and undisturbed systems. To an-alyze how the selection of the cut-off rank affected ourresult, we; (1) selected it based on the visual interpreta-tions by the experts and (2) chose it such that the mergerrate of our comparison sample was consistent with theoverall major merger fraction of galaxies in our mass andredshift range. When we considered the average deter-minations of the classifiers, approximately half of bothpopulations showed signs of a current or recent mergerevent, suggesting no causal connection between majormergers and the triggering of this particular populationof AGNs.Since our first approach also considers asymmetries orsignatures that stem from processes other than a majormerger event, we adjust the major merger fraction ofthe inactive galaxies to be consistent with recent sim-ulations and observations. As a result we find a sub-stantial excess in the major merger fraction of the AGNsample with respect to the inactive galaxies. Coinci-dentally, with a separation at the corresponding cut-offrank we also found a clear distinction between strongly- ajor merger rate for AGNs with the highest Eddington ratios at z < . f m , agn = 0.41 ± f m , ina =0.08 ± z < .
2, major mergers are an essential mecha-nism to trigger black hole growth.– We rule out that minor mergers play a consider-able role in the triggering of our subpopulation ofAGNs.– Considering AGN and merger lifetimes as well asAGN variability induced by an ongoing mergerevent, our best estimate results in ∼
50% of ourAGN population still being of unknown origin.Extending our study to include IFU-observations anda larger number of sources would enable us to ana-lyze the AGN host galaxies in more detail. By assess-ing the strength of potential past merger events by ex-amining the kinematics and stellar populations, whilelarger number provides better statistics we can deter-mine, which processes are responsible for the triggeringof the remaining ∼
50% and whether major mergers areindeed the dominant mechanism.ACKNOWLEDGMENTSWe thank the referee for the constructive feedback,which improved the quality of this work.We also thank Mischa Schirmer for his helpful guid-ance in using
THELI
Facility:
ESO-VLT(FORS2)
Software: astropy (Astropy Collaboration et al.2013, 2018), Matplotlib (Hunter 2007), MultiColorFits(Cigan 2019), SAOImageDS9 (Joye & Mandel 2003),THELI (Erben et al. 2005; Schirmer 2013)8
Marian et al.
APPENDIX A. DETAILS ON COMBINATION METHODSEvery method to combine individual votes into a combined consensus sequence violates at least one of three criteriadescribed by Arrow’s Impossibility Theorem (Arrow 1950). It states that no existing method, which combines twoor more individual votes satisfies the following three axioms: (1) non-dictatorship, such that all individual votes areconsidered to be equal; (2) unanimity or the weak Pareto principle, stating that if all voters agree on
X > Y , thisalso holds true for the overall ranking; and (3) the independence of irrelevant alternatives, such that the consensusrelation between X and Y only depends on the individual preferences between those two entities and not any additionaloption(s). As additional conditions we introduce the Condorcet paradox and the Condorcet criterion (Condorcet 1785;Condorcet et al. 1989). The first one states that an overall sequence can be cyclic – e.g. X wins over Y , whichwins over Z , which in turn wins over X – although the individual votes are not. The latter explains that an overalltop-ranked candidate wins in every pairwise comparison with every other candidate.Below we present the methods we apply to create the overall rankings. As stated in Section 3 we use three differentalgorithms to construct those combined rankings to determine any potential bias introduced by the method. However,in addition all of our three methods also satisfy or infringe the above mentioned criteria differently, which gives useven more detailed insights in any potential introduction of differences in the merger fractions.For our first method to combine the individual expert rankings we adopt the same method applied in Mechtleyet al. (2016) and Marian et al. (2019). We start with calculating the mean rank for each galaxy from the individualrankings and discard every individual expert classification of each galaxy, if it differs more than 2 σ from the respectiveaverage rank. Out of the 798 individual assessments in V -band we reject 25 votes, while out of the total 779 ratings,17 are discarded for the sets in B -band and color, respectively. However, since we weigh individual votes this methodobviously violates the non-dictatorship criterion.Our second method, the Borda count approach (Emerson 2013), satisfies this condition, but violates in exchangethe independence of irrelevant alternatives. We adapt the original version of this method in which the first rankedoption receives n points, the second one n − n being the total number of candidates, by applyingthe Dowdall system (Reilly 2002). With that approach the candidates receive the reciprocal value of their respectiveranks, i.e. the first ranked option is rewarded 1 /n = 1 point, the next one 0.5 points and so on. As low rank galaxiesmay be ranked more randomly due to a lack of significant merger features, we can decrease the impact those sourcesmight have on our overall ranking by using this variant of the Borda count.This approach avoids the Condorcet paradox, but only our third method, the Schulze method (Schulze 2011, 2018),also satisfies the Condorcet criterion. With this method all pairwise comparisons between two candidates X and Y for all individual rankings are calculated and put into relation to each other, resulting in an overall ranking, wherethe top-ranked candidate, wins indeed over all other candidates, being the so-called Condorcet winner. Going to lowerranks within the resulting consensus sequence the second-placed candidate only loses to the first-ranked option and soon (for more details and examples please see Schulze 2018). B. DEPENDENCE OF MERGER FRACTIONS ON CUT-OFF RANKIn Sections 3 and 4 we describe how the choice of cut-off rank can influence the resulting merger fractions and alsopresent for four selected cut-off ranks the corresponding merger fractions. In Figure 7 we now present the continuousdependence of merger fractions on cut-off rank for all combinations of method and set. The AGN host galaxies andinactive galaxies are shown in blue and red, respectively. The shaded regions denote the 1 σ confidence interval fromshot- and classification-noise. As already indicated in Figure 3 and described in Section 3, it is also shown in Figure 7that first, neither the choice of method to combine the individual rankings nor the selection of set has any significantimpact on the resulting absolute merger fractions or the relative differences between them. Second, compared to theinactive comparison sample and for cut-off ranks (cid:46)
15 the AGN host galaxies show a clear excess in merger fractions.This clearly indicates that our conclusions rely considerably on the choice of cut-off rank, which is extensively discussedin the main text. ajor merger rate for AGNs with the highest Eddington ratios at z < . M e r g e r f r a c t i o n Average / B-band
AGNsComparison galaxies
Average / V-band Average / Color0.00.20.40.60.81.0 M e r g e r f r a c t i o n Borda / B-band Borda / V-band Borda / Color10 20 30Cut-off rank0.00.20.40.60.81.0 M e r g e r f r a c t i o n Schulze / B-band 10 20 30Cut-off rankSchulze / V-band 10 20 30Cut-off rankSchulze / Color
Figure 7.
Evolution of the merger fractions for the AGN host galaxies (blue) and inactive galaxies (red) in dependence ofcut-off rank for each combination of set and method. The shaded regions give the 1 σ confidence interval. Marian et al. C. VISUAL OVERALL CONSENSUS RANKINGTo have a ‘meta’ singular consensus sequence we apply the Schulze method (see Section 3 and Appendix A) to ourfinal nine overall rankings, which we calculated for each combination of set and method. We show all sources in theresulting order, and include for completeness also the sources already shown in Figure 2. The respective rank for eachobject is given in parentheses besides its designation. It should be noted that Gal176221 is only ranked last, becauseit was only observed in V -band and therefore only appears in the three corresponding consensus rankings. In thosethree respective rankings it is always positioned at rank 14. Clearly visible is the drop-off in strong merger features ata cut-off rank (cid:38) (a) (b) (c)(1) - SDSSJ105007.75+113228.6 (a) (b) (c)(2) - Gal030481 Figure 8.
From left to right we present a postage stamp in (a) V-band, (b) B-band and (c) color, respectively. Note: In orderto enhance the visibility the images are not shown with the same cuts and color map parameters. ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(3) - HE0157+0009 (a) (b) (c)(4) - HE2011-6103 (a) (b) (c)(5) - HE2258-5524 Figure 8. (Continued.) Marian et al. (a) (b) (c)(6) - HE0132-0441 (a) (b) (c)(7) - HE0558-5026 (a) (b) (c)(8) - PG1012+008 Figure 8. (Continued.) ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(9) - Gal458007 (a) (b) (c)(10) - Gal079769 (a) (b) (c)(11) - Gal270096 Figure 8. (Continued.) Marian et al. (a) (b) (c)(12) - Gal698144 (a) (b) (c)(13) - Gal782980 (a) (b) (c)(14) - HE0444-3449 Figure 8. (Continued.) ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(15) - Gal534882 (a) (b) (c)(16) - Gal510223 (a) (b) (c)(17) - Gal050873 Figure 8. (Continued.) Marian et al. (a) (b) (c)(18) - Gal419090 (a) (b) (c)(19) - Gal676011 (a) (b) (c)(20) - SDSSJ124341.77+091707.1 Figure 8. (Continued.) ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(21) - Gal498251 (a) (b) (c)(22) - Gal286443 (a) (b) (c)(23) - HE2152-0936 Figure 8. (Continued.) Marian et al. (a) (b) (c)(24) - Gal185580 (a) (b) (c)(25) - Gal003114 (a) (b) (c)(26) - Gal204260 Figure 8. (Continued.) ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(27) - Gal347112 (a) (b) (c)(28) - Gal557614 (a) (b) (c)(29) - HE1226+0219 Figure 8. (Continued.) Marian et al. (a) (b) (c)(30) - Gal095873 (a) (b) (c)(31) - Gal210148 (a) (b) (c)(32) - Gal221730 Figure 8. (Continued.) ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(33) - Gal000232 (a) (b) (c)(34) - HE1228+0131 (a) (b) (c)(35) - HE1201-2408 Figure 8. (Continued.) Marian et al. (a) (b) (c)(36) - Gal391560 (a) (b) (c)(37) - PG1001+054 (a) (b) (c)(38) - HE0119-2836 Figure 8. (Continued.) ajor merger rate for AGNs with the highest Eddington ratios at z < . (a) (b) (c)(39) - PG1211+143 (a) (b) (c)(40) - SDSSJ032213.89+005513.4 (a) (b) (c)(41) - Gal656010 Figure 8. (Continued.) Marian et al. (a) (b) Not available (c) Not available(42) - Gal176221 Figure 8. (Continued.)D.
TABULAR OVERALL CONSENSUS RANKINGSComplementary to Appendix C we present in this section for referential use the consensus ranks for each target forall sets and combination methods. As in Appendix C the sources are sorted by rank of the ‘meta’ consensus ranking,i.e. the combined ranking of the nine overall rankings (see Sect. 4.1).
Table 3 . Final consensus ranksTarget Borda Average SchulzeV-band B-band Color V-band B-band Color V-band B-band ColorSDSS-J105007.75+113228.6 1 3 3 1 4 4 1 3 5Gal030481 4 6 1 2 5 1 2 5 1HE0157+0009 3 1 4 3 1 6 3 1 6HE2011-6103 2 2 5 4 2 5 4 2 4HE2258-5524 5 4 2 6 6 2 5 4 2HE0132-0441 7 7 6 5 7 3 6 7 3HE0558-5026 6 5 8 7 3 8 7 6 8PG1012+008 8 8 7 8 8 7 8 8 7Gal458007 10 11 9 9 10 9 9 10 10Gal079769 12 10 10 13 9 13 10 9 9Gal270096 11 12 12 10 11 12 11 11 11Gal698144 18 13 13 15 12 10 18 12 12Gal782980 9 9 15 12 13 16 12 13 16HE0444-3449 13 21 11 11 22 11 13 21 13Gal534882 15 16 16 20 19 18 17 17 15Gal510223 19 14 21 17 15 19 22 14 19Gal050873 20 17 17 19 20 17 19 18 17Gal419090 22 22 18 18 18 15 20 16 18
Table 3 continued ajor merger rate for AGNs with the highest Eddington ratios at z < . Table 3 (continued)
Target Borda Average SchulzeV-band B-band Color V-band B-band Color V-band B-band ColorGal676011 23 15 20 23 16 21 23 15 20SDSS-J124341.77+091707.1 17 19 22 21 14 22 15 22 24Gal498251 21 18 26 22 17 25 21 19 22Gal286443 16 20 27 16 21 26 16 20 23HE2152-0936 24 23 37 24 24 35 24 23 35Gal185580 26 30 14 25 29 14 25 28 14Gal003114 28 24 29 28 23 27 27 24 26Gal204260 31 26 19 29 26 20 30 29 21Gal347112 30 29 23 26 27 23 26 26 27Gal557614 27 27 24 30 28 24 28 27 25HE1226+0219 25 28 25 27 31 29 33 33 29Gal095873 29 31 32 31 30 31 29 30 32Gal210148 33 25 35 33 25 33 31 25 33Gal221730 34 33 28 32 33 28 35 32 31Gal000232 36 34 30 37 34 32 34 31 30HE1228+0131 35 32 33 34 32 37 36 34 36HE1201-2408 37 38 34 35 36 34 32 36 34Gal391560 38 39 31 36 39 30 38 35 28PG1001+054 32 36 41 38 35 40 37 37 39HE0119-2836 40 37 39 39 37 38 39 38 38PG1211+143 39 35 38 40 38 39 40 39 41SDSS-J032213.89+005513.4 41 40 40 41 40 41 42 40 40Gal656010 42 41 36 42 41 36 41 41 37Gal176221 14 N/A N/A 14 N/A N/A 14 N/A N/A
Note —The final ranks for each source depending on combination method (Borda, Average or Schulze) and set ( B , V orcolor images. The targets are sorted by a repeated use of the Schulze method on this nine overall rankings resulting in asingular consensus sequence. Since we have for Gal176221 only observations in V -band it is ranked last by the algorithm. Marian et al.
REFERENCES
Aird, J., Coil, A. L., & Georgakakis, A. 2019, MonthlyNotices of the Royal Astronomical Society, 484, 4360,doi: 10.1093/mnras/stz125Aird, J., Coil, A. L., Georgakakis, A., et al. 2015, MonthlyNotices of the Royal Astronomical Society, 451, 1892,doi: 10.1093/mnras/stv1062Allevato, V., Finoguenov, A., Cappelluti, N., et al. 2011,The Astrophysical Journal, 736, 99,doi: 10.1088/0004-637X/736/2/99Arrow, K. J. 1950, Journal of Political Economy, 58, 328,doi: 10.1086/256963Astropy Collaboration, Robitaille, T. P., Tollerud, E. J.,et al. 2013, \ aap, 558, A33,doi: 10.1051/0004-6361/201322068Astropy Collaboration, Price-Whelan, A. M., Sip \ Hocz,B. M., et al. 2018, \ aj, 156, 123,doi: 10.3847/1538-3881/aabc4fBarnes, J. E., & Hernquist, L. 1992, \ araa, 30, 705,doi: 10.1146/annurev.aa.30.090192.003421Baumgartner, W. H., Tueller, J., Markwardt, C. B., et al.2013, The Astrophysical Journal Supplement Series, 207,19, doi: 10.1088/0067-0049/207/2/19Beifiori, A., Courteau, S., Corsini, E. M., & Zhu, Y. 2012,Monthly Notices of the Royal Astronomical Society, 419,2497, doi: 10.1111/j.1365-2966.2011.19903.xBennert, V. N., Auger, M. W., Treu, T., Woo, J.-H., &Malkan, M. A. 2011, The Astrophysical Journal, 742,107, doi: 10.1088/0004-637X/742/2/107Bennert, V. N., Treu, T., Woo, J.-H., et al. 2010, TheAstrophysical Journal, 708, 1507,doi: 10.1088/0004-637X/708/2/1507Blanton, E. L., Clarke, T. E., Sarazin, C. L., Randall,S. W., & McNamara, B. R. 2010, Proceedings of theNational Academy of Sciences, 107, 7174,doi: 10.1073/pnas.0913904107Bottrell, C., Hani, M. H., Teimoorinia, H., et al. 2019,Monthly Notices of the Royal Astronomical Society, 490,5390, doi: 10.1093/mnras/stz2934Boyle, B. J., Shanks, T., Croom, S. M., et al. 2000,Monthly Notices of the Royal Astronomical Society, 317,1014, doi: 10.1046/j.1365-8711.2000.03730.xBradley, L., Sipcz, B., Robitaille, T., et al. 2019,astropy/photutils: v0.7.2, Zenodo,doi: 10.5281/zenodo.3568287Bridge, C. R., Carlberg, R. G., & Sullivan, M. 2010, TheAstrophysical Journal, 709, 1067,doi: 10.1088/0004-637X/709/2/1067 Brinchmann, J., Charlot, S., White, S. D. M., et al. 2004,Monthly Notices of the Royal Astronomical Society, 351,1151, doi: 10.1111/j.1365-2966.2004.07881.xBhm, A., Wisotzki, L., Bell, E. F., et al. 2013, Astronomy& Astrophysics, 549, A46,doi: 10.1051/0004-6361/201015444Casteels, K. R. V., Conselice, C. J., Bamford, S. P., et al.2014, Monthly Notices of the Royal AstronomicalSociety, 445, 1157, doi: 10.1093/mnras/stu1799Cen, R., & Safarzadeh, M. 2015, The Astrophysical JournalLetters, 798, L38, doi: 10.1088/2041-8205/798/2/L38Cheng, T.-Y., Conselice, C. J., Aragn-Salamanca, A., et al.2019, arXiv e-prints, 1908, arXiv:1908.03610.http://adsabs.harvard.edu/abs/2019arXiv190803610CCheung, E., Trump, J. R., Athanassoula, E., et al. 2015,Monthly Notices of the Royal Astronomical Society, 447,506, doi: 10.1093/mnras/stu2462Chowdhury, R. K., Chatterjee, S., Lonappan, A. I.,Khandai, N., & DiMatteo, T. 2019, arXiv:1911.07824[astro-ph]. http://arxiv.org/abs/1911.07824Churazov, E., Sazonov, S., Sunyaev, R., et al. 2005,Monthly Notices of the Royal Astronomical Society:Letters, 363, L91, doi: 10.1111/j.1745-3933.2005.00093.xCigan, P. 2019, MultiColorFits: Colorize and combinemultiple fits images for visually aesthetic scientific plots.https://ascl.net/1909.002Cisternas, M., Sheth, K., Salvato, M., et al. 2015, TheAstrophysical Journal, 802, 137,doi: 10.1088/0004-637X/802/2/137Cisternas, M., Jahnke, K., Inskip, K. J., et al. 2011, TheAstrophysical Journal, 726, 57,doi: 10.1088/0004-637X/726/2/57Condorcet, J.-A.-N. d. C. .-. . m. d. A. d. t. 1785, Essai surl’application de l’analyse la probabilit des dcisionsrendues la pluralit des voix ([Reprod.]) / par M. lemarquis de Condorcet,... (de l’Impr. royale (Paris)).https://gallica.bnf.fr/ark:/12148/bpt6k417181Condorcet, J.-A.-N. d. C., Sommerlad, F., McLean, I.,University of Oxford, & Faculty of Social Studies. 1989,The political theory of Condorcet (Oxford: University ofOxford, Faculty of Social Studies)Conroy, C., & White, M. 2013, The Astrophysical Journal,762, 70, doi: 10.1088/0004-637X/762/2/70Cotini, S., Ripamonti, E., Caccianiga, A., et al. 2013,Monthly Notices of the Royal Astronomical Society, 431,2661, doi: 10.1093/mnras/stt358Dashyan, G., Choi, E., Somerville, R. S., et al. 2019,Monthly Notices of the Royal Astronomical Society, 487,5889, doi: 10.1093/mnras/stz1697 ajor merger rate for AGNs with the highest Eddington ratios at z < . Davies, J. J., Crain, R. A., Oppenheimer, B. D., & Schaye,J. 2019, arXiv e-prints, 1908, arXiv:1908.11380.http://adsabs.harvard.edu/abs/2019arXiv190811380DDavis, B. L., Graham, A. W., & Cameron, E. 2018, TheAstrophysical Journal, 869, 113,doi: 10.3847/1538-4357/aae820—. 2019, The Astrophysical Journal, 873, 85,doi: 10.3847/1538-4357/aaf3b8de Nicola, S., Marconi, A., & Longo, G. 2019, MonthlyNotices of the Royal Astronomical Society, 490, 600,doi: 10.1093/mnras/stz2472Delvecchio, I., Lutz, D., Berta, S., et al. 2015, MonthlyNotices of the Royal Astronomical Society, 449, 373,doi: 10.1093/mnras/stv213Di Matteo, T., Springel, V., & Hernquist, L. 2005, Nature,433, 604, doi: 10.1038/nature03335Ding, X., Silverman, J., Treu, T., et al. 2020, TheAstrophysical Journal, 888, 37,doi: 10.3847/1538-4357/ab5b90Donley, J. L., Kartaltepe, J., Kocevski, D., et al. 2018, TheAstrophysical Journal, 853, 63,doi: 10.3847/1538-4357/aa9ffaDuncan, K., Conselice, C. J., Mundy, C., et al. 2019, TheAstrophysical Journal, 876, 110,doi: 10.3847/1538-4357/ab148aEllison, S. L., Viswanathan, A., Patton, D. R., et al. 2019,MNRAS, 487, 2491, doi: 10.1093/mnras/stz1431Emerson, P. 2013, Social Choice and Welfare, 40, 353,doi: 10.1007/s00355-011-0603-9Erben, T., Schirmer, M., Dietrich, J. P., et al. 2005,Astronomische Nachrichten, 326, 432,doi: 10.1002/asna.200510396Fan, L., Han, Y., Fang, G., et al. 2016, The AstrophysicalJournal, 822, L32, doi: 10.3847/2041-8205/822/2/L32Fanidakis, N., Baugh, C. M., Benson, A. J., et al. 2012,Monthly Notices of the Royal Astronomical Society, 419,2797, doi: 10.1111/j.1365-2966.2011.19931.xFurlong, M., Bower, R. G., Theuns, T., et al. 2015,Monthly Notices of the Royal Astronomical Society, 450,4486, doi: 10.1093/mnras/stv852Gabor, J. M., Impey, C. D., Jahnke, K., et al. 2009, TheAstrophysical Journal, 691, 705,doi: 10.1088/0004-637X/691/1/705Gao, F., Wang, L., Pearson, W. J., et al. 2020,arXiv:2004.00680 [astro-ph].http://arxiv.org/abs/2004.00680Georgakakis, A., Coil, A. L., Laird, E. S., et al. 2009,Monthly Notices of the Royal Astronomical Society, 397,623, doi: 10.1111/j.1365-2966.2009.14951.x Glikman, E., Simmons, B., Mailly, M., et al. 2015, TheAstrophysical Journal, 806, 218,doi: 10.1088/0004-637X/806/2/218Goulding, A. D., Matthaey, E., Greene, J. E., et al. 2017,The Astrophysical Journal, 843, 135,doi: 10.3847/1538-4357/aa755bGoulding, A. D., Greene, J. E., Bezanson, R., et al. 2018,Publications of the Astronomical Society of Japan, 70,doi: 10.1093/pasj/psx135Graham, A. W., & Scott, N. 2013, The AstrophysicalJournal, 764, 151, doi: 10.1088/0004-637X/764/2/151Grogin, N. A., Conselice, C. J., Chatzichristou, E., et al.2005, The Astrophysical Journal, 627, L97,doi: 10.1086/432256Grogin, N. A., Kocevski, D. D., Faber, S. M., et al. 2011,The Astrophysical Journal Supplement Series, 197, 35,doi: 10.1088/0067-0049/197/2/35Habouzit, M., Genel, S., Somerville, R. S., et al. 2019,Monthly Notices of the Royal Astronomical Society, 484,4413, doi: 10.1093/mnras/stz102Harrison, C. M., Costa, T., Tadhunter, C. N., et al. 2018,Nature Astronomy, 2, 198,doi: 10.1038/s41550-018-0403-6Henriques, B. M. B., White, S. D. M., Thomas, P. A., et al.2015, Monthly Notices of the Royal AstronomicalSociety, 451, 2663, doi: 10.1093/mnras/stv705Hewlett, T., Villforth, C., Wild, V., et al. 2017, MonthlyNotices of the Royal Astronomical Society, 470, 755,doi: 10.1093/mnras/stx997Hirschmann, M., Dolag, K., Saro, A., et al. 2014, MonthlyNotices of the Royal Astronomical Society, 442, 2304,doi: 10.1093/mnras/stu1023Hong, J., Im, M., Kim, M., & Ho, L. C. 2015, TheAstrophysical Journal, 804, 34,doi: 10.1088/0004-637X/804/1/34Hopkins, P. F., & Hernquist, L. 2009, The AstrophysicalJournal, 698, 1550, doi: 10.1088/0004-637X/698/2/1550Hopkins, P. F., Hernquist, L., Cox, T. J., et al. 2006a, TheAstrophysical Journal Supplement Series, 163, 1,doi: 10.1086/499298Hopkins, P. F., Hernquist, L., Cox, T. J., & Kere, D. 2008,The Astrophysical Journal Supplement Series, 175, 356,doi: 10.1086/524362Hopkins, P. F., Hernquist, L., Martini, P., et al. 2005, TheAstrophysical Journal, 625, L71, doi: 10.1086/431146Hopkins, P. F., & Quataert, E. 2010, Monthly Notices ofthe Royal Astronomical Society, 407, 1529,doi: 10.1111/j.1365-2966.2010.17064.xHopkins, P. F., Richards, G. T., & Hernquist, L. 2007, TheAstrophysical Journal, 654, 731, doi: 10.1086/509629 Marian et al.