PRIMUS: The relationship between Star formation and AGN accretion
Mojegan Azadi, James Aird, Alison Coil, John Moustakas, Alexander Mendez, Michael Blanton, Richard Cool, Daniel Eisenstein, Kenneth Wong, Guangtun Zhu
aa r X i v : . [ a s t r o - ph . GA ] M a y Accepted to
ApJ
Preprint typeset using L A TEX style emulateapj v. 12/16/11
PRIMUS: THE RELATIONSHIP BETWEEN STAR FORMATION AND AGN ACCRETION
Mojegan Azadi , James Aird , Alison L. Coil , John Moustakas , Alexander J. Mendez , Michael R.Blanton , Richard J. Cool , Daniel J. Eisenstein , Kenneth C. Wong , Guangtun Zhu Accepted to
ApJ
ABSTRACTWe study the evidence for a connection between active galactic nuclei (AGN) fueling and star formationby investigating the relationship between the X-ray luminosities of AGN and the star formation rates(SFRs) of their host galaxies. We identify a sample of 309 AGN with 10 < L X < erg s − at0 . < z < . L X . We do not find a significant correlation between SFR and the observed instantaneous L X for star forming AGN host galaxies. However, there is a weak but significant correlation betweenthe mean L X and SFR of detected AGN in star forming galaxies, which likely reflects that L X varieson shorter timescales than SFR. We find no correlation between stellar mass and L X within theAGN population. Within both populations of star forming and quiescent galaxies, we find a similarpower-law distribution in the probability of hosting an AGN as a function of specific accretion rate.Furthermore, at a given stellar mass, we find a star forming galaxy ∼ − Keywords: galaxies: active – galaxies: evolution – X-rays: galaxies INTRODUCTION
It has been several decades since the first observationsof galaxies with strong emission lines in their central re-gions (Seyfert 1943) and the classification of such sourcesas Active Galactic Nuclei (AGN). Since then, the rangeof observational phenomena associated with AGN hasexpanded to include sources classified based on a va-riety of X-ray, optical, infrared and radio criteria (e.g.Antonucci 1993) and there have been numerous inves-tigations into the physical nature of these AGN (for arecent review see Alexander & Hickox 2012). It is nowwidely accepted that AGN activity is due to the presenceof a supermassive black hole (SMBH) accreting gas anddust in the circumnuclear region, forming an accretiondisk that ultimately powers the AGN activity. Studiesfrom recent decades have established that SMBHs residein almost all galaxies with a bulge or spheroid compo-nent (e.g. Kormendy & Richstone 1995; Kormendy et al. Center for Astrophysics and Space Sciences, Department ofPhysics, University of California, 9500 Gilman Dr., La Jolla, SanDiego, CA 92093, USA Department of Physics, Durham University, Durham DH13LE, UK Institute of Astronomy, University of Cambridge, MadingleyRoad, Cambridge CB3 0HA, UK Department of Physics and Astronomy, Siena College, 515Loudon Road, Loudonville, NY 12211, USA Center for Cosmology and Particle Physics, Department ofPhysics, New York University, 4 Washington Place, New York,NY 10003, USA MMT Observatory, 1540 E Second Street, University of Ari-zona, Tucson, AZ 85721, USA Harvard College Observatory, 60 Garden St., Cambridge,MA 02138, USA Institute of Astronomy and Astrophysics, Academia Sinica,No.1, Sec. 4, Roosevelt Rd., Taipei 10617, Taiwan Department of Physics & Astronomy, Johns Hopkins Uni-versity, 3400 N. Charles Street, Baltimore, MD 21218, USA
Azadi et al. fraction of quasars in merging systems, several authorshave proposed that nuclear activity is also tightly con-nected to merger events (e.g. Sanders & Mirabel 1996;Canalizo & Stockton 2001; Ivison et al. 2010). However,recent studies find little to no connection between AGNactivity and the incidence of merger events in galaxieswith moderate luminosity AGN (e.g. Schawinski et al.2011; Kocevski et al. 2012), suggesting that secular pro-cesses such as turbulence and disk instabilities are moreeffective in enhancing nuclear accretion activity for AGNwith moderate luminosities (e.g. Mullaney et al. 2012b;Rosario et al. 2012).Studies of the locations of AGN host galaxies in the op-tical color-magnitude diagram are useful to investigatethe role of AGN in the evolution of their host galax-ies. Galaxies can be divided into two general populationsin this diagram: the blue cloud, consisting of predomi-nantly star forming galaxies, and the red sequence, com-prised mainly of quiescent, passively evolving galaxies(e.g. Blanton et al. 2003; Bell et al. 2003; Baldry et al.2004). There is also a third population, known asthe green valley, that includes galaxies in transitionbetween the other two populations (e.g. Mendez et al.2011). Many studies have demonstrated that X-ray de-tected AGN from 0 < z < ∼
2) enhancement in the probability of a galaxy of a given stellar mass hosting anAGN for galaxies with blue or green rest-frame opticalcolors. Bongiorno et al. (2012) used specific SFR (es-timated from fits to the optical–to–near-infrared spec-tral energy distributions) to split their galaxy sampleinto star forming and quiescent galaxies, finding nosignificant differences in the probability of hosting anAGN for galaxies in either population. More recently,Hern´an-Caballero et al. (2014) found that AGN hostshave similar distributions of rest frame optical colors toinactive galaxies of the same stellar mass but are morelikely to be hosted by galaxies with younger stellar pop-ulations. Georgakakis et al. (2014) also split AGN hostsinto star forming and quiescent populations based ontheir U − V versus V − J colors (which should be morerobust to dust extinction than rest-frame optical colors)and found that the space density of star forming hostsis higher than for quiescent hosts, with some weak evi-dence for differences in the shape of the accretion ratedistributions.A number of recent studies have used Herschel far in-frared data, which provides a more robust tracer of thetotal SFR and is not impacted by dust extinction, tocompare AGN hosts to the wider galaxy population. In
Herschel -detected populations, Mullaney et al. (2012b),Santini et al. (2012) and Rosario et al. (2013) all foundevidence for enhanced SFR in AGN hosts, compared tonon-active galaxies of the same stellar mass, and arguedthat the bulk of moderate-luminosity AGN are hostedby normal star forming galaxies. However,
Herschel isonly able to detect galaxies that are bright at far-infraredwavelengths and thus generally have high SFRs, mak-ing it difficult to measure the fraction of quiescent hostsand compare the SFRs of all AGN hosts to the full pop-ulation of star forming galaxies. However, given thatthe presence of dust can redden UV-optical colors, re-sults that rely solely on optical colors may be biased.Cardamone et al. (2010) found that dust reddening af-fects the colors of some star forming AGN host galaxies,pushing them to the green valley.In addition to determining whether AGN are morelikely to reside in star forming or quiescent host galaxies,several authors have studied whether there is an overallcorrelation between the level of star formation and thelevel of nuclear activity, as traced by the X-ray luminos-ity, in individual galaxies. Rovilos et al. (2012) foundno evidence for a correlation in AGN with L X < . erg s − at z < z >
1, using a sample of X-ray de-tected AGN in the
Chandra
Deep Field–South (CDFS).Mullaney et al. (2012b) use
Herschel -detected moderateluminosity ( L X = 10 − erg s − ) X-ray AGN in theCDFS and Chandra
Deep Field–North (CDFN) fields at0 < z < L AGN & erg s − at z < RIMUS: The relationship between Star Formation and AGN accretion z < .
5. Therefore, they suggestthat rather than violent mergers, secular processes are re-sponsible for both star formation and SMBH growth inmajority of the galaxies with moderate nuclear activity.There is some evidence that optically luminousAGN are found in galaxies with enhanced SFRs (e.g.Floyd et al. 2013). However, Page et al. (2012) foundthat star formation was suppressed in their sample of lu-minous X-ray detected AGN at 1 < z <
Herschel /SPIRE 250 µm detectionsin the CDFN. More recently, Harrison et al. (2012) foundno sign of star formation suppression in powerful AGNhosts using a larger sample. Moreover, Harrison et al.(2012) found that the average SFR in galaxies hostingAGN with luminosities in range of 10 < L X < erg s − at 1 < z < L X ,consistent with results from Mullaney et al. (2012b) andRosario et al. (2012).Studies of lower luminosity AGN have been carriedout in the local Universe. Using a sample of nearbySeyfert galaxies Diamond-Stanic & Rieke (2012) founda strong correlation between the AGN luminosity andthe SFR in the circumnuclear regions ( r < average accretion rate, averaging over both active andnon-active galaxies, and SFR. Chen et al. (2013) studiedstar forming galaxies detected by Herschel in the Bo¨otesfield at z <
1, including 34 X-ray and 87 MIR-detectedAGN, and found that the average accretion rate is cor-related with SFR. They also compared their result witha small sample of 20 X-ray detected AGN in FIR brightgalaxies from Symeonidis et al. (2011) at z ∼ z < instantaneous AGN luminosity influx-limited AGN samples.In this paper, we investigate the correlation betweenthe SFR and stellar mass of AGN host galaxies with the nuclear activity of their SMBHs, using a large sam-ple of galaxies with spectroscopic redshifts from thePRIsm MUlti-object Survey, PRIMUS, (Coil et al. 2011;Cool et al. 2013). We use X-ray data from
Chandra and
XMM-Newton surveys that cover ∼ of thePRIMUS area to identify a large sample of moderate-luminosity (10 < L X < erg s − ) AGN within ourgalaxy sample. We use X-ray luminosity as the tracer ofAGN activity and estimate SFRs and stellar masses byfitting the observed galaxy spectral energy distribution(SED) using UV and optical photometry of our sources.With this data, we are able to probe down to relativelylow SFRs and robustly separate our sample into qui-escent and star forming populations. We also measurethe fraction of AGN with star forming versus quiescenthost galaxies (compared to a stellar mass-matched galaxysamples) and quantify how this fraction is changing withredshift, which could potentially drive any observed cor-relations in the overall sample. Finally, we measure theprobability of a galaxy hosting an AGN and the distribu-tion of specific accretion rates for both the star formingand quiescent galaxy populations, updating the studyfrom Aird et al. (2012) using our more robust galaxyclassifications. We also further sub-divide the galaxypopulation to study the specific accretion rate distribu-tion as a function of the specific SFR.Section 2 briefly describes our data and the stellarmass completeness limits that we use to minimize ob-servational biases. In Section 3 we describe our resultson the correlation between SFR and AGN X-ray lumi-nosity in our full sample. We also investigate this cor-relation within sub-populations of star forming and qui-escent galaxies. In addition, we consider the connectionbetween galaxy star formation and AGN specific accre-tion rate and quantify the probability of a galaxy hostingan AGN as a function of specific SFR. We interpret ourresults in Section 4 and investigate the variation of theaverage X-ray luminosity of AGN with the star forma-tion activity of their host galaxies. We summarize ourresults in Section 5. Throughout the paper we adopt aflat cosmology with Ω Λ =0.7 and H =72 km s − Mpc − and all magnitudes are on AB system. DATA
In this study, we use multi-wavelength data from thePRIMUS survey covering four fields on the sky, includ-ing the
Chandra
Deep Field South (CDFS), COSMOS,ELAIS S1, and XMM-LSS fields. All of these fields havedeep UV, optical, and IR imaging as well as spectroscopicredshifts from PRIMUS. We also use X-ray imaging from
Chandra and
XMM-Newton to identify AGNs within thePRIMUS samples. We describe these datasets below, aswell as our method of estimating stellar masses and SFRsfor our sources by fitting their spectral energy distribu-tions (SEDs).
PRIMUS
We use data from PRIMUS, the largest faint galaxy,intermediate-redshift survey completed to date. ThePRIMUS survey used the IMACS spectrograph on Mag-ellan I Baade 6.5 m telescope at Las Campanas obser-vatory, with a slitmask and a low-dispersion prism. Thesurvey has a spectroscopic resolution of R ∼
40 and cov-ers a total of 9.1 deg of sky, spread over seven extra- Azadi et al. galactic fields with deep multiwavelength data. Objectswere targeted to i ∼
23 using well-understood targetingweights. Four additional fields were targeted that had alarge number of prior, high-resolution spectroscopic red-shifts; these data were used for calibration purposes. Inthese calibration fields higher priority was given to tar-gets with prior spectroscopic redshifts. The full detailsof the survey, targeting and data summary are presentedin Coil et al. (2011).In PRIMUS, objects are classified as stars, broad-lineAGNs (BLAGNs) and galaxies based on their spectra. Ineach class, low-resolution spectra and multi-wavelengthphotometry of the objects are simultaneously fit with anempirical library of templates. For this study we restrictour sample to the sources with robust redshifts ( Q ≥ ∼ ,
000 robust redshifts at z ∼ − .
2, with aredshift precision of σ z / (1 + z ) ∼ . Chandra and
XMM-Newton , UV imaging fromGALEX, deep optical imaging from a range of telescopes,and infrared imaging from the Infrared Array Camera(IRAC) and the Multiband Imaging Photometer (MIPS)on
Spitzer . In this paper, we restrict our analysis tospectroscopic sources targeted within the area with jointGALEX UV, optical, and
Spitzer
IRAC imaging. Wethus restrict our sample to the COSMOS, ELAIS-S1 andXMM-LSS science fields in PRIMUS. In Sections 3.1–3.4we also use the PRIMUS calibration field CDFS-CALIB(hereafter CDFS), which overlaps with the deep
Chandra
X-ray coverage. Taken together, these four fields cover4.96 deg of the sky. We exclude the X-ray AGN andgalaxy samples in the CDFS field below in the analysisof Section 3.5, as it does not provide a uniformly targetedsample of galaxies, which is required for the analysis inthat section. X-ray data
We use X-ray data to identify AGN within thePRIMUS galaxy sample. We have compiled X-ray cata-logs in the CDFS, COSMOS, ELAIS S1, and XMM-LSSfields based on published
Chandra and
XMM-Newton surveys. In the CDFS field, we use the Lehmer et al.(2005) and Luo et al. (2008) X-ray catalogs correspond-ing to the 2 Ms observations of the central region (reach-ing depths of f − ∼ . × − erg s − cm − )and the flanking 250 ks observations (reaching depths of f − ∼ . × − erg s − cm − ). The entire COS-MOS field was observed with XMM-Newton to depths of f − ∼ × − erg s − cm − (Hasinger et al. 2007);additionally the central ∼ . was observed with Chandra to depths of f − ∼ × − erg s − cm − (Elvis et al. 2009). In the ELAIS-S1 field we use the cat-alog of Puccetti et al. (2006), based on the XMM-Newton observations that reach f − ∼ × − erg s − cm − . Finally, in XMM-LSS we use X-ray data availablefrom the both the Pierre et al. (2007) catalog and fromthe XMM-Newton
Deep Survey (Ueda et al. 2008) downto the f − ∼ × − erg s − cm − . We use thelikelihood ratio technique (e.g. Sutherland & Saunders 1992; Ciliegi et al. 2003; Brusa et al. 2007; Laird et al.2009) to identify reliable optical counterparts (in the i or R band) to the X-ray sources; for objects with mul-tiple counterparts, the match with the highest ratio ischosen. For full details of the construction of our X-raycatalogs and matching procedure see Aird et al. (2012)and Mendez et al. (2013).In this study, we consider a sample of obscured AGNdetected in the hard (2–10 keV) X-ray band, where wehave excluded objects that were classified as BLAGN inthe PRIMUS spectra. These BLAGN constitute about12% of our AGN sample. As their optical emission candominate over the optical light of the host galaxy, forsuch sources we are unable to estimate the stellar massand SFR of the host. Restricting to hard X-ray (2–10 keV) detections ensures that we can estimate theX-ray luminosity with reasonable accuracy and are notstrongly biased against the selection of moderately ob-scured sources. Although hard X-ray emission passesthrough regions with moderate hydrogen column den-sities, it can not penetrate Compton-thick regions withheavy obscuration; thus our sample lacks this potentiallyimportant population.We also restrict our analysis to sources with moderateX-ray luminosities in the range 10 < L X < ergs − resulting in a final sample of 309 AGN. The lowerlimit of 10 erg s − ensures that the observed X-ray lu-minosity is dominated by light from an AGN rather thanfrom star formation activity in the host galaxy. The up-per limit ensures that the optical light of the host is notstrongly contaminated by the presence of an AGN, suchthat we can estimate stellar masses and SFRs. It alsoensures that our sample is not strongly biased by ourexclusion of BLAGN, which constitute a higher fractionof the X-ray selected AGN population at high luminosi-ties. To summarize, our AGN sample consists of 309sources with hard-band X-ray detections and robust red-shifts from PRIMUS (in the range 0 . < z < .
2) that arenot classified as BLAGN (thus the host galaxy dominatesthe optical light) and have X-ray luminosities within therange 10 < L X < erg s − . Stellar Mass and SFR Estimates
SED fitting is a widely adopted method for estimatingthe physical properties of galaxies. We estimate stel-lar masses and SFRs of the galaxies by fitting the ob-served SED based on the UV and optical photometry ofour sources. As we exclude BLAGN and do not includeIRAC photometry in our SED fits, we do not include anAGN contribution in the SED fitting. We fit the SEDsusing the iSEDfit code (Moustakas et al. 2013), whichis a Bayesian fitting code that compares the observedphotometry for each source to a large Monte Carlo gridof SED models which span a wide range of stellar pop-ulation parameters (e.g. age, metallicity, dust, and starformation history) to estimate the stellar mass and SFRof a galaxy.With iSEDfit we find the posterior probability distri-bution of stellar mass and SFR of a galaxy by marginal-izing over all of the other parameters. We then take themedian of the probability distribution functions as thebest estimate of the stellar mass or SFR of each galaxy.The uncertainty on each parameter is calculated as onequarter of the 2.3–97.7 percentile range of the probability
RIMUS: The relationship between Star Formation and AGN accretion -1 0 1 2 3(iSEDfit) Log SFR [M Ο • yr -1 ]-10123 ( H e r sc he l ) Log S F R [ M Ο • y r - ] COSMOS
Herschel detected AGN
Figure 1.
The SFR derived from
Herschel versus from iSEDfit for sources in the COSMOS field. We use
Herschel deep 100 µ m ob-servations and convert the FIR luminosity to a SFR using Equation(1). Contours show the distribution of PRIMUS galaxies detectedby Herschel , while filled blue circles indicate AGN detected by
Herschel . The bulk of the AGN sample is not
Herschel -detected.The green dashed line indicates the 1:1 relation. While the ma-jority of
Herschel -detected galaxies lie close to this 1:1 line, thereis a clear population that scatters above the line indicating thatwe underestimate the SFR using iSEDfit , although the shape ofthis distribution will be due to
Herschel only detecting dusty star-forming galaxies with high SFRs. The
Herschel -detected X-raysources span a similar space to the galaxies; upper limits on the
Herschel
SFRs for X-ray sources without
Herschel detections placethem in a similar space, confirming that the shape of the distri-bution is primarily driven by the limited depths of the
Herschel data. distributions, which would be equivalent to a 1 σ uncer-tainty in the case of a Gaussian distribution. For detailson iSEDfit see Moustakas et al. (2013).For the SED fitting used in this paper, we adoptthe Flexible Stellar Population Synthesis (FSPS) mod-els (Conroy et al. 2009; Conroy & Gunn 2010) withChabrier (2003) initial mass function (IMF) from 0.1 to100 M ⊙ and stellar metallicity in the range of 0 . 05. We consider exponentially declining star for-mation histories Ψ ∝ τ exp ( − tτ ), allowing for τ withinthe range of 0 . < τ < iSEDfit on our UV and optical photometry. Totest the accuracy of our SFRs, we compare them with Herschel Space Observatory deep far infrared observa-tions of the COSMOS field. We use deep 100 µ mobservations of the PACS Evolutionary Probe, PEP12(Lutz et al. 2011), reaching a 3 σ limit of 5 mJy at 100 µ m(Berta et al. 2011). We then use the Kennicutt (1998) re-lation, given here as Equation (1), using Chabrier IMFto convert the FIR luminosity to SFR: SF R M ⊙ yr − = 1 . × − L IR L ⊙ (1)Figure 1 compares the Herschel derived SFR with ourestimate from iSEDfit for PRIMUS sources in the COS-MOS field. Contours show the distribution of PRIMUSgalaxies that are detected by Herschel , while blue circlesshow PRIMUS X-ray AGN that are detected by Her-schel . The dashed line represents the 1:1 relation. Ascan be seen in the figure, the error bars on the Her-schel estimated SFRs are much smaller than those from iSEDfit and are likely underestimated, as a single tem-plate is used to calculate the total IR luminosity.While many AGN and galaxies lie near the 1:1 relation,there is a sizable portion of the sample well above theline, with the SFR estimated from Herschel much higherthan the SFR from iSEDfit . This is not surprising, asthe IR luminosity is a more accurate probe of the SFRin dusty galaxies, while the inferred SFR from fitting theUV and optical SED primarily reflects unobscured starformation (though iSEDfit does fit and account for dustobscuration). Most PRIMUS X-ray AGN in COSMOSare not detected by Herschel , and for these sources wecalculate the 3 σ upper limits on SFR as estimated fromthe Herschel imaging. These AGN not detected by Her-schel show a similar overall offset as the detected AGN inthis figure, though we only have upper limits, such thatthe true values may lie close to the 1:1 line. A histogramof the Herschel to iSEDfit SFR differences in the Her-schel -detected galaxy sample peaks at ∆(log SF R ) = 0but has a median offset of 0.6 dex. Within this sample,42% of the sources have a Herschel SFR that is morethan a factor of three higher than the iSEDfit SFR,although the shape of this distribution will be stronglyskewed due to the limited depths of the Herschel data.We note that, based on the KS test, the distribu-tions of ∆(log SF R ) for the Herschel -detected AGN and Herschel -detected galaxies are not significantly different.Thus, the overall SFR estimated by iSEDfit for dustygalaxies may be systematically low, which appears tobe due to iSEDfit underestimating the dust extinctionin some of the Herschel -detected galaxies (and thus un-derestimating the SFR). However, what we are inter-ested in here is whether there is a correlation betweenSFR and L X . A systematic offset will not affect our re-sults, although additional scatter in our SFR estimatescould wash out any underlying correlations. Addition-ally, as Herschel detects warm dust heated by star for-mation, the Herschel -detected sample includes only themost dusty star forming galaxies, such that the SFR dif-ferences in the full PRIMUS galaxy and AGN sample willbe less pronounced.Below in Section 3.2 we split the PRIMUS sample intostar forming versus quiescent galaxies using the iSEDfit stellar mass and SFR values of each source. Here we esti-mate the contamination of our quiescent sample by starforming galaxies, by finding that in the COSMOS field7% of our quiescent sample (defined using iSEDfit out-puts) is detected by Herschel ; these galaxies are thereforestar forming galaxies and are misclassified. Stellar mass completeness limits Azadi et al. 41 42 43 44−2−1012 Log S F R [ M Ο • y r − ] cc=0.32 (p=0.00) 41 42 43 44Log L X [erg s −1 ] cc=0.24 (p=0.00) 41 42 43 44 cc=0.02 (p=0.89) Figure 2. The SFR versus L X for a sample of non-broad line AGN in four PRIMUS fields, including CDFS, COSMOS, ELAIS S1 andXMM-LSS, for three redshift bins spanning 0 . < z < . 2. Orange crosses show individual AGN, while green circles illustrate the medianSFR in bins of L X . The error bars show the uncertainty on our calculation of the median points, measured from bootstrap resampling.The blue dotted line shows the SFR expected if the X-ray emission is from HMXBs; the fact that our sources are all well below this lineindicates that the X-ray emission is from AGN. The correlation coefficients and correlation significance of the individual points in eachpanel are calculated from Spearman’s rank correlation. A small p value denotes it is unlikely for a correlation to have been occurred byaccident and the correlation is considered significant if the p value is less than 0.05. The lower two redshift panels show a weak trendbetween SFR and L X , which is not apparent in the highest redshift panel, where our data probe a narrower range of L X . As PRIMUS is a flux-limited survey, targeting objectsto i ∼ 23, this introduces a bias into our sample where weare unable to detect low-mass galaxies at higher redshifts,unless they have high SFRs (increasing the amount ofblue light from the galaxy). To minimize this bias wedefine a stellar mass limit above which we can detect allgalaxies, regardless of their SFR. This stellar mass limitis a smooth function of redshift and is slightly differ-ent in each field, depending on the band used for targetselection (see also Aird et al. 2012). Briefly, we definea template for a maximally old simple stellar popula-tion at z ∼ . < z < . 2, of which 283 are AGN detectedin the hard X-ray band with 10 < L X < erg s − . RESULTS In this section, we investigate the relationship betweenSFR and X-ray luminosity, L X , in our AGN sample. Inorder to uncover whether stellar mass could be an under-lying variable, we also investigate the stellar mass depen-dence of L X . We further divide our full AGN sample intothose with star forming and quiescent host galaxies, toconsider how SFR and stellar mass vary with L X withineach host population. We also measure the probabilitythat a galaxy of a given stellar mass and redshift hostsan AGN, as a function of the specific SFR of the galaxy. The relationship between SFR and stellar masswith L X Figure 2 shows the SFR of AGN host galaxies plottedas a function of L X , in three redshift bins spanning 0 . 2. Orange crosses show individual AGN, whilegreen circles show the median SFR in bins of L X , whereeach bin contains at least 15 sources, to visually highlightany correlations. X-ray emission at these luminositiescan arise not only from AGN but potentially from highmass X-ray binaries (HMXBs), which are tracers of starformation activity and generally have lower luminosities, L X ∼ − erg s − . The dotted blue line in Figure 2 isfrom Ranalli et al. (2003) and shows the relation betweenSFR and L X for HMXB: SF R M ⊙ yr − = 2 . × − L X(2 − 10 Kev) erg s − (2)As this line is well above our sources, it indicates thatthe X-ray emission seen for our sample is from AGN anddoes not suffer from contamination by HMXBs.For the median points shown in Figure 2, we estimateerror bars using bootstrap resampling. The uncertaintyshown reflects the variance among the median SFR ineach of 1000 bootstrap samples. These errors are similarto the standard errors calculated in each L X bin. Thisfigure clearly shows that there is a wide spread in SFR atany given value of L X , such that the standard deviationof the points is typically 3–4 times greater than the errorshown.We use the r correlate routine in IDL to find the cor-relation coefficients and correlation significance of the in-dividual points in Figure 2. This routine computes theSpearman’s rank correlation coefficient and the signifi-cance of its deviation from zero, p . This value indicatesthe probability of obtaining a desired event under the nullhypothesis that the event happened purely by chance. Asmall p value denotes it is unlikely for the correlation to RIMUS: The relationship between Star Formation and AGN accretion 41 42 43 449101112 Log [ M * / M Ο • ] cc=-0.01 (p=0.93) 41 42 43 44Log L X [erg s -1 ] cc=-0.21 (p=0.01) 41 42 43 44 cc=0.12 (p=0.31) Figure 3. Stellar mass versus L X for our sample of non-broad AGN in three redshift bins spanning 0 . < z < . 2. Orange crosses showindividual AGN while green circles illustrate the median stellar mass in bins of L X . The error bars show the uncertainty on median points,measured from bootstrap resampling. The solid dark green line indicates the PRIMUS stellar mass completeness limit. The grey dashedline is a prediction from a model presented in Aird et al. (2013) and shows the predicted median stellar mass as a function of L X in eachredshift range. The blue dotted line shows the prediction of this model for sources above the PRIMUS mass completeness limit. There isno significant trend between stellar mass and L X in the first and last panel but there is a negative correlation in the middle panel that ismainly due the sources below the mass completeness limit. have been occurred by accident. A correlation is consid-ered significant if the p value is less than 0.05. We quote p values to an accuracy of two decimal places, thus p = 0 . > p =0.00), 0.24 ( p =0.00), and 0.02 ( p =0.89), respectively,from the lowest to the highest redshift bin.In this figure, there is a large scatter in SFR in bins of L X in all three redshift ranges. At a given L X , the av-erage SFR increases at higher redshifts, consistent withthe overall increase in SFR seen in the galaxy popula-tion (e.g. Bell et al. 2005; Elbaz et al. 2007; Noeske et al.2007). In the two lowest redshift ranges we see a weakpositive correlation in the median points that is con-firmed by the significance of the correlation coefficientsmeasured above. We find no correlation between SFRand L X in the highest redshift range, but we note thatat higher redshifts we are probing a more limited rangeof X-ray luminosity.To determine whether the observed trends in thefirst two panels are actually being driven by redshift-dependent selection effects, we measure the median red-shift in the each bin of L X . As the median redshift doesnot systematically vary with L X , redshift is not drivingany trends in Figure 2.However, the weak correlation between SFR and L X seen in the lower two redshift panels could be due to anunderlying trend between stellar mass and L X . Giventhat within the galaxy population there is a positive cor-relation between SFR and stellar mass (e.g. Elbaz et al.2007; Karim et al. 2011), it is possible that the observedcorrelation between SFR and L X could actually be due toan underlying correlation between stellar mass and L X .In Figure 3 we show the stellar mass of AGN hostgalaxies as a function of L X . As in Figure 2, orangecrosses indicate individual sources while green circlesshow median values in bins of L X , and errors on the me- dian points are calculated using bootstrap resampling.The standard deviation of the data points is larger thanthe error shown by a factor of 3–5. The correlation co-efficients of the individual points in this figure are -0.01( p =0.93), -0.21 ( p =0.01) and 0.12 ( p =0.31), respectively,for the three redshift ranges shown. We note that wehave not applied the stellar mass completeness limits dis-cussed above in Section 2.4 and we show our full X-rayAGN sample in both Figures 2 and 3. In Figure 3, weonly find a significant correlation in the middle redshiftrange; however, the correlation is negative and appearsto be driven by a small number of sources below thePRIMUS mass completeness limit shown with the soliddark green line. If we only consider sources above thestellar mass completeness limits, we do not find any sig-nificant correlation between stellar mass and L X for X-ray AGN in any of the three redshift ranges. We thusconclude that the observed (weak, but significant) pos-itive correlations between SFR and L X in Figure 2 are not due to a positive correlation between stellar massand L X within our X-ray AGN sample.The grey dashed line in Figure 3 is a prediction from amodel presented in Aird et al. (2013). This model takesthe stellar mass function of galaxies and populates thegalaxies with AGN using a universal power-law distri-bution of specific accretion rates (the rate of accretionscaled relative the host stellar mass, see Section 3.3 be-low). The specific accretion rate distribution itself doesnot depend on stellar mass but has a normalization thatevolves with redshift (motivated by the observational re-sults of Aird et al. 2012). The grey dashed line showsthe predicted median stellar mass as a function of L X ,using this model, and in particular shows that the me-dian stellar mass of AGN host galaxies should not varysignificantly with L X , over the luminosity range wherewe have data. The blue dotted line shows the predictionof this model for sources above the PRIMUS mass com-pleteness limits, and the median points in our sample lie Azadi et al. close to this line, confirming this lack of a correlation.There is a large scatter in the stellar mass at anygiven X-ray luminosity. We find that the average stel-lar mass of the AGN host galaxies in all three red-shift ranges is higher than 10 M ⊙ . This is consistentwith prior literature, including Kauffmann et al. (2003);Xue et al. (2010) and Aird et al. (2012), who find thatobserved AGN are predominantly hosted by moderatelymassive galaxies. While some recent studies have foundevidence for AGN activity in much lower mass, dwarfgalaxies, such sources represent a small fraction of theX-ray selected population and have X-ray luminositiesbelow our limit (e.g. Moran et al. 1999; Barth et al. 2004;Reines et al. 2011, 2013; Secrest et al. 2015). Nonethe-less, we note that Aird et al. (2013) attribute the domi-nance of moderate-mass host galaxies in X-ray–selectedAGN samples to the shape of the stellar mass function ofgalaxies, combined with the wide power-law distributionof AGN accretion rates, rather than an enhancement ofAGN activity in galaxies of a particular stellar mass.Overall, Figure 3 demonstrates that over the X-rayluminosity that we probe, 41 < log L X < 44, there isno correlation between AGN luminosity and host stellarmass within the AGN sample (although we note that ourAGN sample is dominated by moderately massive galax-ies, M ∗ > M ⊙ at all luminosities and redshifts).The weak correlation seen above between SFR and X-ray luminosity is therefore not due to an underlying cor-relation between L X and stellar mass within the AGNsample. The relationship between SFR and stellar masswith L X for star forming and quiescent galaxies Our results above indicate that there is a weak but sig-nificant correlation between SFR and L X for AGN hostgalaxies for the lowest two redshift bins probed here,spanning 0 . < z < . 8. We do not find a significantpositive correlation between stellar mass and L X for thesame sample, indicating that stellar mass is not driv-ing the observed SFR- L X relation. We further investi-gate possible effects of the host population by splittingthe AGN host galaxies into star forming and quiescentpopulations and determining whether a SFR- L X relationexists within either of these populations alone.We classify each AGN host galaxy as star forming orquiescent using its specific star formation rate, sSF R ,which is defined as the SFR per unit stellar mass, SF R M ∗ .Histograms of log sSF R of our full galaxy sample withinrelatively narrow redshift ranges show two prominentpeaks: one corresponding to the “main sequence” of starforming galaxies (Noeske et al. 2007) and the other toquiescent galaxies. The locations of the peaks evolvewith redshift, in that sSF Rs are higher (on average) athigher redshift. To classify galaxies into star formingor quiescent, we wish to use the sSF R correspondingto the minimum between these two populations in thesSFR histogram. This minimum is not always clear inall of the (six) redshift bins used, but the peak of thestar forming main sequence can be traced easily at allredshifts. We therefore first fit for the evolution of thispeak, sSF R max , as a linear function of redshift, whichis given by:log( sSF R max ( z )) = 0 . × z − . 62 (3) We then normalize the sSF R of each galaxy relativeto this sSF R max ( z ), which we call the epoch normal-ized specific star formation rate, EN sSF R (see alsoStott et al. (2012)) :log( EN sSF R ) = log( sSF R ) − . × z + 9 . 62 (4)Finally, we plot a histogram of log EN sSF R for allof galaxies, which exhibits a clear bimodality. We findthe minimum between the two peaks of the bimodaldistribution at log EN sSF R min = − . 2. We divideour sample into star forming and quiescent galaxies ac-cording to whether their EN sSF R is above or belowlog EN sSF R min , respectively. Ultimately, we apply thesame classification scheme to our X-ray AGN sample.Figure 4 shows the location of PRIMUS galaxies andAGN within the SFR-stellar mass plane, in the samethree redshift ranges as used above. The contours indi-cate the location of PRIMUS galaxies, while red crossesshow individual AGN host galaxies. The blue line showsthe separation defined above between the star formingand quiescent populations (shown at the median redshiftof each redshift range); galaxies above the line are con-sidered to be star forming while galaxies below are clas-sified as quiescent. The vertical dashed green line showsthe stellar mass completeness limit for PRIMUS, at themedian redshift of that panel. Above this stellar masslimit we are complete for both star forming and quies-cent galaxies (see Section 2.4 above). Figure 4 shows thatAGN are present in both the star forming and quiescentgalaxy populations. However, the most massive galaxies(and therefore AGN host galaxies) in the sample tend tobe quiescent, especially at z > . L X of AGN host galaxies for each populations inFigure 5. The blue points show host galaxies on the starforming main sequence, while the red points show quies-cent host galaxies. We find the median SFR in bins ofX-ray luminosity, where we require a minimum of 12 and10 AGN per bin for the star forming and quiescent pop-ulations, respectively. We note that these median pointsare for the purpose of illustrating the trends in data andthe numbers are chosen to have at least three L X binsfor each population in each panel. As above, error barson the median points are from bootstrap resampling andthe standard deviation in each population is larger thanthe errors shown by a factor of 3–4. We find a fairly highscatter in the SFR at a given L X within each host galaxypopulation. However, due to the flux limit of the surveywe probe a smaller range of L X in the highest redshiftpanel.Within the star forming population, the lack of a trendin the median points and the correlation coefficients(none of which are significant, as seen in the figure) con-firm the absence of any significant correlation betweenSFR and L X in all three redshift ranges probed here.Within the quiescent population there is a weak but sig-nificant trend in the lowest redshift range only; the corre-lation coefficients are 0.44 ( p =0.00), -0.04 ( p =0.81) and0.15 ( p =0.43), respectively from lowest to highest red-shift.We further show in Figure 6 the stellar mass versus L X of the AGN host galaxies, split into the star form- RIMUS: The relationship between Star Formation and AGN accretion Log S F R [ M Ο • y r − ] O • ] Figure 4. SFR versus stellar mass for PRIMUS galaxies and AGN, shown in three redshift ranges. Contours show the distribution ofgalaxies in this space; there are two distinct populations: star forming galaxies, with a relatively high SFR at a given stellar mass, andquiescent galaxies, with a low SFR at a given stellar mass. Red crosses show X-ray AGN, which reside in both star forming and quiescenthost galaxies. The blue solid line is the classification used to define galaxies as being either star forming (above the line) or quiescent(below the line). This line evolves with redshift. The vertical dotted green line shows the PRIMUS stellar mass completeness limit, whichis also a function of redshift. To create samples that are complete in stellar mass, we exclude sources to the left of this line. 41 42 43 44−2−1012 Log S F R [ M Ο • y r − ] cc=0.07 (p=0.61)cc=0.44 (p=0.00) QuiescentStar−Forming 41 42 43 44Log L X [erg s −1 ] cc=0.17 (p=0.14)cc=−0.04 (p=0.81) QuiescentStar−Forming 41 42 43 44 cc=−0.07 (p=0.70)cc=0.15 (p=0.43) QuiescentStar−Forming Figure 5. The SFR versus L X of AGN host galaxies, in three redshift bins, where the host galaxies are split into star forming (blueplus) or quiescent (red triangle). Similar to Figure 2, the errors are measured using the bootstrap resampling. Blue and red circles showthe median SFR in bins of L X , for the star forming and quiescent host populations, respectively. There is a significant positive trend inquiescent host population in the lowest redshift panel that vanishes at higher redshifts. There is no significant correlation between SFRand L X in star forming galaxies in any redshift bins. ing and quiescent populations. As in Figure 5, the blueand red symbols represent star forming and quiescentgalaxies, respectively, and the error bars are calculatedusing bootstrap resampling. The horizontal green solidline shows the PRIMUS stellar mass completeness limitat the median redshift of each panel, and the grey dottedline is from the model of Aird et al. (2013), as describedin Section 3.1, for sources above this completeness limit.Both galaxy populations have a range of stellar massesbut on average the stellar masses of the quiescent galaxiesare higher in all three redshift ranges. Applying the stel-lar mass completeness limit clearly narrows the dynamicrange of our sample in stellar mass at higher redshifts,particularly for the star forming population. As with thefull galaxy population, we do not find any significant cor- relation between stellar mass and L X for either the starforming or quiescent host galaxy populations in any ofthe three redshift ranges. We also find that applying thestellar mass completeness limit and splitting our sam-ple to star forming and quiescent, the weak trend foundabove in Figure 3 for the middle redshift range now van-ishes.Overall, it appears that the weak positive trend in thefirst panel of Figure 2 is due to a correlation betweenSFR and L X in the quiescent host population. Figure6 further shows that this trend is not due to the stellarmass. For the other redshift ranges, we no longer finda significant correlation between SFR and L X after afterapplying the stellar mass completeness limit and split-ting the sample into the star forming and quiescent host0 Azadi et al. 41 42 43 449101112 Log [ M * / M Ο • ] cc=0.06 (p=0.69)cc=0.09 (p=0.57) QuiescentStar-Forming 41 42 43 44Log L X [erg s -1 ] cc=0.02 (p=0.88)cc=-0.15 (p=0.30) QuiescentStar-Forming 41 42 43 44 cc=0.20 (p=0.26)cc=0.00 (p=0.98) QuiescentStar-Forming Figure 6. The average stellar mass versus L X , in three redshift ranges. The host galaxies are split into star forming (blue plus) orquiescent (red triangle). The errors are measured using the bootstrap resampling. Similar to Figure 3, the grey dashed line is a predictionof stellar mass as a function of L X from a model presented in Aird et al. (2013) for sources above PRIMUS mass completeness limit. Thegreen solid line shows the PRIMUS stellar mass completeness limit, which is also a function of redshift. There is no significant correlationbetween stellar mass and L X in either of populations above the mass completeness limit. 41 42 43 44 F r a c t i on o f A G N w i t h S F ho s t Mass-matched galaxiesGalaxiesAGN 41 42 43 44Log L X [erg s -1 ] Mass-matched galaxiesGalaxiesAGN 41 42 43 44 Mass-matched galaxiesGalaxiesAGN Figure 7. Variation of the fraction of AGN in star forming host galaxies with L X , for stellar mass complete samples, shown with bluelines. The errors are calculated from the binomial distribution using the Bayesian method of Cameron (2011). The green lines show thefraction of all galaxies above the stellar mass completeness limits that are star forming, while the pink lines show this fraction for galaxysamples that have the same stellar mass distributions as the AGN host galaxies. While the star forming host fraction increases withincreasing L X in the lower two redshift ranges, given the error bars neither trend is significant. In the middle redshift range there is a 2 σ difference between the total fraction of AGN host galaxies that are star forming and the fraction of stellar mass matched galaxies that arestar forming. galaxies (although we note that splitting up our sample,in itself, could eliminate any weakly significant trendsseen in the full sample). The fraction of star forming AGN host galaxies In Section 3.1 we found that in the two lowest red-shift panels of Figure 2 there is a weak but significantpositive correlation between SFR and L X in AGN hostgalaxies. In Section 3.2 we demonstrated that when split-ting the sample of these host galaxies into star formingand quiescent, this trend disappeared except for quies-cent AGN hosts in the lowest redshift bin. We now in-vestigate whether a change in the fraction of star forminghost galaxies, as a function of L X , could be driving the observed correlation between SFR and L X for the fullAGN host sample. For example, a higher fraction of starforming host galaxies at the most luminous end in Figure2 could create a positive trend in the full sample, as isseen.We calculate the fraction of X-ray selected AGN witha star forming host galaxy as a function of L X , where weconsider only galaxies above the stellar mass complete-ness limits. The results are shown in Figure 7 in binsof 0.5 dex in L X , in three redshift ranges. The errorson the fractions are calculated assuming a binomial dis-tribution using the Bayesian method of Cameron (2011)and are equivalent to 1 σ uncertainties (68.3% equal-tailconfidence intervals). To compare the AGN host galax- RIMUS: The relationship between Star Formation and AGN accretion −4 −3 −2 −1 0−2−1012 Log S F R [ M Ο • y r − ] cc=−0.37 (p=0.03)cc=0.36 (p=0.13)cc=−0.37 (p=0.03)cc=0.36 (p=0.13)cc=−0.37 (p=0.03)cc=0.36 (p=0.13)cc=−0.37 (p=0.03)cc=0.36 (p=0.13) QuiescentStar−Forming −4 −3 −2 −1 0Log λ cc=−0.05 (p=0.71)cc=−0.11 (p=0.60)cc=−0.05 (p=0.71)cc=−0.11 (p=0.60)cc=−0.05 (p=0.71)cc=−0.11 (p=0.60)cc=−0.05 (p=0.71)cc=−0.11 (p=0.60) QuiescentStar−Forming −4 −3 −2 −1 0 cc=−0.10 (p=0.64)cc=−0.18 (p=0.45)cc=−0.10 (p=0.64)cc=−0.18 (p=0.45)cc=−0.10 (p=0.64)cc=−0.18 (p=0.45) QuiescentStar−Forming Figure 8. The average SFR versus specific accretion rate, in three redshift bins, where the host galaxies are split into star forming (blueplus) or quiescent (red triangle). The errors are measured using bootstrap resampling. Blue and red circles show the median SFR in bins of λ , for the star forming and quiescent host populations, respectively. The grey regions show the area below the specific accretion rate limit;sources in these regions are excluded from the sample. There is no significant correlation between SFR and λ within the quiescent hostpopulation. There is a 2 σ correlation in the star forming host population in the lowest redshift range that is not seen at higher redshifts. ies with inactive galaxies in the same redshift range, weshow the fraction of entire PRIMUS galaxies above thestellar mass completeness limit that are star forming withgreen solid lines. Although we find AGN in galaxies witha wide range of stellar mass, they are mainly found in rel-atively massive galaxies. Furthermore, the median stellarmass of the AGN host galaxies increases from 10 . to10 . and 10 . M ⊙ respectively over the three red-shift ranges shown in Figure 7, primarily due to the in-creasing stellar mass limit of the PRIMUS sample. Foreach redshift range we therefore make a sample of stel-lar mass-matched galaxies that has the same stellar massdistribution as the AGN host galaxies. Then we comparethe fraction of AGN within star forming host galaxieswith a sample of inactive galaxies with a similar stellarmass and redshift distribution. To construct this sam-ple, we weight each galaxy in a redshift bin such thatthe weighted distribution of stellar masses matches thestellar mass distribution of the X-ray AGN host galaxies.The fraction of the stellar mass-matched galaxy samplethat is star forming is shown with pink lines in Figure 7.In the lowest redshift range, 0 . < z < . 5, there isa strong apparent trend such that the fraction of X-rayAGN in star forming hosts increases with increasing X-ray luminosities, from ∼ 20% at L X = 10 . erg s − to 100% at L X = 10 . erg s − . This trend is gener-ally consistent with other studies (e.g. Kauffmann et al.2003; Heckman & Kauffmann 2006) in a similar red-shift regime that found that low luminosity AGN typ-ically reside in early type galaxies with low star for-mation activity, while powerful AGN reside in galaxieswith young stellar populations. In the middle redshiftrange, 0 . < z < . 8, the fraction also increases with in-creasing X-ray luminosity, though not as strongly as atlower redshift. However, in the highest redshift range,0 . < z < . 2, the fraction declines with X-ray lumi-nosity, over the more limited range spanned by the dataat these higher redshifts. However, considering the largeerror bars, none of these trends are significant.In an attempt to decrease the error bars, in each red- Table 1 The fraction of X-ray selected AGN with star forming hostgalaxies, and the fraction of stellar mass-matched galaxies thatare star forming.Redshift AGN with SF hosts SF mass-matched galaxies0 . < z < . ± 14% 46%0 . < z < . ± 12% 38%0 . < z < . ± 17% 43% shift range we split the AGN sample into just two equal-width bins in L X . In the lowest redshift range, for41 < log L X < ± 19% and increasesto 63 ± 20% at higher luminosities of 42.5 < log L X < ± 30% to 67 ± 13% in the two luminosity bins. Consid-ering the error bars, neither of these increases is signifi-cant.Table 1 lists the overall fraction of AGN host galax-ies that are star forming, as well as the fraction of stellarmass-matched galaxies that are star forming, in each red-shift range. The variation of the star forming fractionfor the stellar mass-matched galaxy sample with red-shift is due to the combination of the PRIMUS stellarmass limits (which restrict us to higher stellar massesat higher redshifts), the preference for observed X-rayAGN to be found in more massive galaxies (across allredshifts), and the intrinsic changes in the star formingfraction for galaxies of a given stellar mass with increas-ing redshift (e.g. Moustakas et al. 2013). The differencein the star forming fraction of AGN host galaxies com-pared to that of stellar mass-matched galaxies is less than1 σ and therefore is not significant in the lowest and high-est redshift bins; however, these fractions are different atthe 2 σ level in the middle redshift range. The fraction ofAGN with star forming host galaxies across our full red-shift range of 0 . < z < . ± Azadi et al. -4 -3 -2 -1 0 F r a c t i on o f A G N w i t h S F ho s t Mass-matched galaxiesGalaxiesAGN −4 −3 −2 −1 0Log λ Mass−matched galaxiesGalaxiesAGN −4 −3 −2 −1 0 Mass−matched galaxiesGalaxiesAGN Figure 9. Variation of the fraction of AGN in star forming hosts with λ , for the specific accretion rate complete sample, shown withblue lines. The errors are calculated from the binomial distribution using the Bayesian method of Cameron (2011). The green lines showthe fraction of all galaxies above the specific accretion rate limits that are star forming, while the pink lines show this fraction for galaxysamples that have the same stellar mass distributions as the AGN host galaxies. The fraction of star forming hosts increases with increasing λ in the lowest redshift. The difference between the fraction of star forming hosts and mass-matched galaxy sample is less than 1 σ in thefirst and last panel and is less than 2 σ in the middle panel, confirming that these fractions are not significantly different. range, there is not a significant difference in the the frac-tion of AGN hosted by star forming galaxies and thefraction of star forming galaxies with the same stellarmass distribution.We find that the fraction of AGN with star forminghost galaxies appears to increase with L X in the twolowest redshift ranges, spanning 0 . < z < . 8, but con-sidering the large error bars the trends observed are notsignificant. However, the star forming host fraction doesincrease from ∼ 30% to ∼ L X increases, at leastin the lowest redshift range, such that the correlation be-tween SFR and L X observed at 0 . < z < . L X . The relationship between SFR and specificaccretion rate In Section 3.2 we showed that there is a wide rangein the stellar masses of AGN host galaxies for both starforming and quiescent populations. This large scattercould potentially hide an underlying correlation betweenSFR and specific accretion rate. Therefore we furtherinvestigate the dependence of SFR and the star forminghost fraction as a function of specific accretion rate, in or-der to remove any stellar mass dependence. The specificaccretion rate, λ ∝ L bol / M ∗ , traces the rate of accretionscaled relative to the host stellar mass. We calculate thespecific accretion rate as λ = L bol . × erg s − × . M ∗ M ⊙ . (5)Thus, λ is a rough tracer of the Eddington ra-tio, under the assumption that M ∗ ≈ M bulge and M BH ≈ . M bulge (Hunt 2003).Figure 8 shows the SFR versus specific accretion rateof our sample, split into star forming and quiescent hostgalaxies. Both populations host AGN with a wide rangeof specific accretion rate, though the average specific ac- cretion rate in star forming host galaxies is higher thanin quiescent host galaxies in all three redshift panels ofFigure 8. However, the average stellar mass of the starforming galaxies is lower than that of the quiescent galax-ies, and we can not detect lower specific accretion ratesources in lower stellar mass host galaxies, because ofthe X-ray flux limit. To minimize this bias, we first es-timate an approximate X-ray luminosity limit in eachredshift bin by taking the X-ray luminosity that 90% ofthe sources exceed ( L ). This luminosity limit gives arough indication of the luminosity below which we are nolonger sampling the X-ray AGN population (for a givenredshift) and is primarily determined by our deepest field(i.e. the CDFS) . This luminosity limit corresponds to adifferent limit in specific accretion rate, depending on themass of the host. To convert to a specific accretion ratelimit, we find the stellar mass above which 90% of ourX-ray sources lie ( M ) in each redshift bin. We con-vert L and M to a limit in specific accretion rate, λ , using Equation (5). We note that M is higherthan our nominal stellar-mass-completeness limits (seeSection 2.4) due to the fact that our X-ray sources arepredominantly found in moderately massive hosts; thusour specific accretion rate limits are lower than if oursample all had hosts with masses at the nominal stellar-mass-completeness limit. Above our specific accretionrate limit, we should have a sample that is representa-tive of the X-ray AGN sample. However, we will missany lower mass galaxies with the same accretion rate.The shaded region in Figure 8 illustrates the range ofspecific accretion rates below this limit, where we willnot have a representative sample. Above this limit wecalculate the correlation coefficients of individual AGN,both the star forming and quiescent host populations We note this does not correspond to the X-ray completenesscorrections calculated with the full X-ray sensitivity curves thatare described and applied in Section 3.5 to accurately recover thefraction of galaxies that host an AGN of a given specific accretionrate RIMUS: The relationship between Star Formation and AGN accretion Figure 10. Left: Epoch-normalized specific star formation rate ( ENsSF R ) versus stellar mass ( M ∗ ) for the galaxy sample (blackcontours) and X-ray AGN (pink crosses) samples considered in Section 3.5. The horizontal dashed lines indicate the dividing lines betweenour four ENsSF R bins, as labeled. Right: Normalized distributions of ENsSF R for the galaxy and X-ray AGN samples, restricted toa stellar mass limit of log M ∗ / M ⊙ > . ENsSF R sources. Table 2 The fraction of X-ray selected AGN with star forming hostgalaxies, and the fraction of stellar mass-matched galaxies thatare star forming, above the specific accretion rate limit.Redshift AGN with SF hosts SF mass-matched galaxies0 . < z < . ± 18% 54%0 . < z < . ± 14% 45%0 . < z < . ± 21% 48% separately, and find that the correlation coefficients arenegligible for quiescent host galaxies in all three redshiftranges. Within the star forming host population, in thelowest redshift range there is a weak negative correlationwith a coefficient of -0.37 ( p =0.03), such that an increasein the specific accretion rate corresponds to a decrease inthe SFR. This trend disappears in the two higher redshiftpanels.Figure 9 shows the fraction of AGN with a star form-ing host galaxy as a function of AGN specific accretionrate. Similar to Figure 7, this fraction is compared forX-ray AGN host galaxies, all galaxies above the stellarmass limit of the PRIMUS survey, and for galaxy sam-ples with the same stellar mass distribution as the AGNhost galaxies. The comparison is done only above thespecific accretion rate limit at a given redshift where wewill have a representative sample. As before, the errorbars on the fractions are from the binomial distributionand are equivalent to 1 σ uncertainties. Table 2 lists thefraction of AGN host galaxies that are star forming (overall values of L X ), as well as the fraction of stellar mass-matched galaxies that are star forming, in each redshiftrange, above our specific accretion rate limit.We find that the fraction of AGN with star forminghost galaxies at 0 . < z < . . < z < . < σ ) to the fraction of star forming galaxies inthe mass-matched galaxy sample. In the middle redshiftrange (0 . < z < . σ difference. Figure9 shows that the fraction of AGN in star forming host galaxies may increase with specific accretion rate in thelower redshift range. In the higher two redshift rangesthis fraction appears to be roughly constant. Thesetrends are generally consistent with what was found inFigure 7, as a function of L X , but our limited sample sizemeans we do not have a high significance result and thusthese trends should be treated with caution.To summarize, when we re-examine trends with SFRand the specific accretion rate (rather than L X ) withinour X-ray AGN sample, we do not find any highly sig-nificant ( > σ ) correlations that were previously missed.Furthermore, the 3 σ correlation between SFR and L X forquiescent host galaxies that was seen in the first panel ofFigure 5 is no longer found when we renormalize L X bythe stellar mass (Figure 8). We note that applying thespecific accretion rate limits has reduced our sample sizeand thus could be why we no longer find a significantcorrelation. However, overall we conclude that there isno evidence for a correlation between SFR and specificaccretion rate for either star forming or quiescent hostgalaxies within the X-ray AGN population. The probability of a galaxy hosting an AGN as afunction of star formation rate In the above analysis we consider a sample of AGNsselected based on their X-ray emission, investigate thecorrelation between the SFR of the host galaxy and theAGN luminosity, and measure the fraction of these X-rayAGNs with hosts that are star forming versus quiescent.In this section we take an alternative approach (basedon the approach of Aird et al. 2012, hereafter A12): weselect samples of galaxies with a specified range of prop-erties (in our case, a particular range of SFR) and deter-mine the probability of finding an AGN in such galaxies.Our galaxy sample consists of all PRIMUS galaxies withstellar masses above the (redshift-dependent) complete-ness limits defined in Section 2.4. For this analysis wedo not include galaxies or X-ray AGNs from the CDFSfield; this field was a PRIMUS calibration field and thetargeted galaxy sample is not defined in the same, con-sistent manner as the PRIMUS science fields. Thus, we4 Azadi et al. Figure 11. The probability density for a galaxy of given stellar mass, M ∗ , and redshift, z , to host an AGN of specific accretion rate, λ ,here dividing the galaxy sample into star forming and quiescent populations according to Equation (4). The thick black line correspondsto the best-fit power-law relationship measured by A12 for the overall galaxy population, evaluated at the center of the given redshift bin.Blue (red) points are estimated using the N obs /N mdl method considering the star forming (quiescent) galaxy populations only, but withreference to this overall model that allows us to account for the underlying redshift evolution, stellar-mass-dependent selection effects, andthe X-ray completeness. We see that at all redshifts the probability of a galaxy hosting an AGN depends strongly on specific accretion forboth star forming and quiescent galaxy populations (rising towards low values of λ ) but is a factor ∼ − λ . are unable to accurately determine the probability of agalaxy hosting an AGN for the CDFS field.We divide the galaxy sample according to the epoch-normalized specific SFR (ENsSFR), or the ratio of thesSFR to that of the main-sequence of star formation atthe redshift of the galaxy, that we defined in Section3.2 (see Equation (4)). By working with the EN sSF R we can remove both the dependence of SFR on stel-lar mass and the overall evolution of the star formingmain sequence over our redshift range. Figure 10 (leftpanel) shows the distribution of EN sSF R (as a func-tion of stellar mass) for the galaxy and X-ray AGN sam-ples we consider here, along with dividing lines betweenour populations. The horizontal green dashed line di-vides the quiescent and star forming galaxy populations,corresponding to the same EN sSF R cuts used in Sec-tion 3.2 (logENsSFR= − . EN sSF R = 0, by defi-nition) and the peak of the quiescent galaxy population(at log EN sSF R = − . EN sSF R for the galaxy and X-ray AGN samples, abovea mass limit of log M ∗ / M ⊙ > . . < z < . EN sSF R than that of the galaxies. This appears con-sistent with the findings of Section 3.2, where we foundweak evidence for a higher fraction of X-ray AGNs to befound in star forming galaxies (i.e. at higher ENsSFR)compared to galaxies of equivalent stellar mass.To accurately determine the probability of a galaxy hosting an AGN we must correct for a number of sourcesof incompleteness in our X-ray selected AGN popula-tion. These effects are ultimately due to the (varying)flux limit of the X-ray observations. The X-ray fluxlimit means that lower luminosity sources will not beidentified at higher redshifts over the entire survey area,thus we must upweight any sources we do detect. A12also showed that probability of a galaxy hosting AGNis given by a power-law distribution of specific accretionrate ( λ ), where lower λ sources are more common thanhigh λ sources in galaxies of all stellar masses (see alsoBongiorno et al. 2012). However, as λ scales with thehost stellar mass, AGNs with the same λ in a lower stel-lar mass host would have a lower observed X-ray lumi-nosity, and thus may fall below our flux limits and beunder-represented in our X-ray selected sample.To account for these effects, we take the global relationfrom A12 for the probability density of a galaxy of given M ∗ , and redshift, z , hosting an AGN of specific accretionrate, λ , per unit logarithmic λ , given by p ( λ | M ∗ , z ) d log L X = Aλ γ E (cid:18) z z (cid:19) γ z d log λ (6)where z = 0 . N mdl = N samp X k =1 Z p ( λ | M k , z k ) p det ( f X ( λ, M k , z k )) d log λ (7)where the summation is performed over all N samp galax-ies in our sub-sample, M k and z k are the stellar massesand redshifts of each galaxy, and p det ( f X ( λ, M k , z k )) isthe probability of detecting an X-ray source with flux f X .We calculate the X-ray flux, f X , by converting λ into a RIMUS: The relationship between Star Formation and AGN accretion Figure 12. The probability density for a galaxy to host an AGNof specific accretion rate λ , dividing the galaxy sample into fourpopulations according to their epoch-normalized specific star for-mation rates (ENsSFRs). We consider a sample spanning the entire0 . < z < . p ( λ | M ∗ , z ) according to the best-fit model of A12. The solid blackline shows this best fit model evaluated at the center of the redshiftbin( z = 0 . 7) and points are calculated using the N obs /N mdl relativeto this model, for each ENsSF R bin. The shape of p ( λ | M ∗ , z )remains roughly the same but the normalization increases movingfrom the quiescent populations to galaxies below or above the starforming main sequence (see also Figure 13). bolometric luminosity (given the stellar mass, M k , of thegalaxy under consideration), which we then convert intoan X-ray luminosity (using the bolometric corrections ofHopkins et al. 2007), and finally convert to an X-ray fluxgiven the redshift of the galaxy, z k . The dependence of p det on flux is determined by the X-ray sensitivity curves,calculated in Section 4.1 of A12. We use the ratio of thispredicted number of X-ray AGNs to the actual observednumber, N obs , in a given galaxy sub-sample to rescalethe prediction for p ( λ | M ∗ , z ) from the A12 model atthe centre of bin of given redshift and λ (based on themethod of Miyaji et al. 2001). Errors on the binned esti-mates are based on the Poisson error given the observednumber of X-ray AGNs (taken from Gehrels 1986). Thismethod allows us to account for underlying variations ofthe model, the X-ray completeness, and the stellar massand redshift distribution of the galaxy sample within agiven bin.In Figure 11 we present binned estimates of p ( λ | M ∗ , z ) using the method described above for threeredshift bins, dividing the galaxy sample into the quies-cent (red crosses) and star forming (blue circles) popu-lations based on our EN sSF R cut. The black line indi-cates the underlying power-law model of A12. We see aclear difference between the estimates for the star form-ing and quiescent populations at all redshifts, wherebythe probability of finding an AGN (of fixed λ ) is higherin a star forming galaxy than for a quiescent galaxy. Thisdifference is more significant at higher redshifts ( z > . Figure 13. The probability density of a galaxy hosting an AGN, p ( λ | M ∗ , z ), evaluated at λ = 0 . 01 and z = 0 . 7, as a functionof epoch-normalized specific star formation rate ( ENsSF R ). Col-ored points correspond to bins defined relative to the star form-ing main sequence (as used in Figure 12), whereas the light greypoints are for evenly spaced, 0.5 dex wide bins of ENsSFR. Theestimates use data from the entire 0 . < z < . − < log λ < − 1, but the plotted values are estimated at λ = 0 . 01 and z = 0 . N obs /N mdl method (relative tothe A12 model). We see that the probability of a galaxy hostingan AGN rises with ENsSF R over the quiescent galaxy populationbut appears to flatten off for galaxies around the star forming mainsequence. accreting sources. The overall increase in the probabilityof finding an AGN as redshift increases is also seen forboth populations.In Figure 12 we combine our data over the entire0 . < z < . EN sSF R shownin Figure 10. The underlying redshift evolution acrossthis wide bin is corrected for by our N obs /N mdl usingthe A12 model. We again see that p ( λ | M ∗ , z ) has afairly consistent power-law shape across the populationsbut the overall normalization increases moving across thequiescent population (from low to high SFR).However, the normalization appears roughly constantacross the star forming population. To investigate thesetrends in more detail, we use our N obs /N mdl method toestimate p ( λ | M ∗ , z ) for a wide − < log λ < − λ = 0 . EN sSF R as the colored points in Fig-ure 13. The light grey points also show estimates for 8fixed width bins of log ENsSFR. We confirm an increase(at 4 σ significance) in the normalization of p ( λ | M ∗ , z )with EN sSF R across the quiescent population, whichincreases further (by a further factor 1.7 ± . 3) to higher EN sSF R (into the star forming galaxy population),where it appears to reach a plateau.Overall, our results show that the probability of agalaxy hosting an AGN is higher for galaxies on the starforming main sequence. The probability of hosting anAGN drops for galaxies below the main sequence, wherethe star formation rates are lower. Nonetheless, AGNsare found in all types of galaxies and appear to have asimilar overall distribution of accretion rates. Our re-6 Azadi et al. sults indicate that about 1–2% of quiescent galaxies at z ∼ . ∼ DISCUSSION Is there a correlation between SFR and AGNluminosity? In this paper we investigate the relationship betweenSFR and AGN luminosity, using L X , and generally findthat these two quantities are not strongly correlatedwithin moderate-luminosity X-ray selected AGN sam-ples. There is a large scatter in SFR at any given L X andmore powerful AGN are not necessarily in more highlystar forming galaxies. We find evidence for a weak cor-relation between SFR and L X in our full sample of bothstar forming and quiescent host galaxies at z < . L X in low redshift quiescent host galaxies; we do notfind a similar trend in star forming host galaxies at theseredshifts, and/or 2) the fact that AGN with higher L X are more likely to have a star forming host galaxy (e.g.Kauffmann et al. 2003; Heckman & Kauffmann 2006),though our error bars are too large to measure this withsignificance. We note that the SFR- L X correlation ob-served in the full sample at 0 . < z < . L X , rather than a correlation between SFRitself and L X . We confirm that none of the observedtrends between SFR and L X are driven by an underlyingcorrelation between L X and stellar mass within our AGNsample. While we find that the mean AGN host stellarmass is log M ∗ M ⊙ ∼ L X range, con-sistent with (e.g. Aird et al. 2012; ? ) We also find alarge scatter in specific accretion rate for both star form-ing and quiescent host galaxies, which could explain thelack of a direct correlation between SFR and L X . In fact,for the low redshift quiescent host galaxies where we dofind a significant correlation between SFR and L X , we donot find a correlation between SFR and specific accretionrate.Overall, within the sample of AGN with star form-ing host galaxies, we do not find significant correla-tions between SFR and AGN luminosity. This is consis-tent with an emerging picture where the instantaneousBH accretion rate (e.g. Chen et al. 2013; Hickox et al.2014) is decoupled from the current rate of star for-mation, at least in galaxies hosting moderate luminos-ity AGN (e.g. Mullaney et al. 2012b; Rosario et al. 2012;Harrison et al. 2012; Rovilos et al. 2012). A possible sce-nario for BH fueling in low to moderate luminosity AGNis that it is driven by stochastic infall of cold gas from cir-cumnuclear regions (e.g. Kauffmann & Heckman 2009).In fact star formation in most “main sequence” star form-ing galaxies appears to be driven by internal processes,e.g. disk instabilities and turbulence (e.g. Elbaz et al.2007; Daddi et al. 2010). Therefore, rather than there being a direct connection between star formation andAGN activity, these two processes likely share a commongas supply. However, there are studies that find a corre-lation between SFR and L X in star forming galaxies thathost powerful AGN, both at at z < z > L X .Our sample at 0 . < z < . Her-schel , as quiescent galaxies are not detected at these red-shifts. Interestingly, we find a significant correlation be-tween SFR and L X in quiescent host galaxies at z < . L X in quiescent galax-ies, for example due to a common cold gas supply. Itcould illustrate a different triggering mechanism (e.g. mi-nor mergers or other secular processes) than that in starforming galaxies, which channels cold gas into the centralregions.We note that as our results rely on UV-optical SEDfits, we could underestimate the level of dust (and there-fore SFR) in a fraction of our galaxies. This could po-tentially introduce a bias that might hide an underlyingcorrelation. Additionally, our SED templates do not in-clude an AGN component; however, we have removedsources with broad lines in their optical spectra, and forour sample the light is clearly dominated by the hostgalaxy. If there was any residual blue light from an AGN,this would decrease our estimated SFRs. Unaccountedfor, this could potentially introduce an observed correla-tion between SFR and L X that does not exist; this couldbe happening in our low redshift quiescent host galaxies.Finally, our sample is limited to some extent by statisticsand a larger and deeper sample would be helpful. Average X-ray luminosity versus SFR While we do not find a direct correlation between SFRand instantaneous X-ray luminosity in star forming hostgalaxies, these processes could still be connected througha common triggering and fueling mechanism. Due toa stochastic fueling process, accretion activity onto theSMBH is time variable (e.g. Ulrich et al. 1997; Peterson2001) and AGN luminosities may drop more than 10 in less than 10 years (Keel et al. 2012b). Addition-ally, recent studies found [OIII] emission from ionizedclouds in the outskirt of galaxies with little or no evidenceof on-going AGN activity (e.g. Schawinski et al. 2010;Keel et al. 2012a,b), confirming that AGN illuminatingthese clouds were much brighter in the past. In contrast,star formation remains stable for a longer period of time;even in starburst galaxies this process can last ∼ 100 Myrs(e.g. Wong 2009; Hickox et al. 2012). Due to the rapidvariability of AGN, using an instantaneous AGN lumi-nosity to compare with SFR may hide an underlying con-nection between AGN and star formation. Instead, usingan average AGN luminosity may be more appropriate to RIMUS: The relationship between Star Formation and AGN accretion -1 0 1 2 3Log SFR [M Ο • yr -1 ]404142434445 Log L X [ e r g s - ] X-ray detected AGN (this paper)X-ray and MIR detected AGN (Chen et al. 2013)X-ray detected AGN (Symeonidis et al. 2011)Stacked (Chen et al. 2013) AGN with star forming hosts at 0.2< z <1 Figure 14. X-ray luminosity versus SFR in star forming hostgalaxies at redshift 0 . < z < . 0. Gray points are PRIMUSindividual X-ray detected AGN, while blue circles show the average L X in bins of SFR. The error bars on each median point reflect thestandard deviation. The blue solid line shows the best fitted lineto the median points. We find a weak positive correlation with thecoefficient of 0.24 ( p = 0.00) between the average L X and SFR inour AGN sample. The median redshift in each bin of SFR indicatesour correlation is not due to evolution of SFR with redshift. Thepink stars are the detected AGN in Chen et al. (2013), while thepink solid line indicates their average sample which also includesX-ray stacks of all galaxies. The green stars are X-ray detectedAGN from Symeonidis et al. (2011), showing LIRG and ULIRGgalaxies at z ∼ explore relationships between AGN and their host galax-ies (e.g. Chen et al. 2013; Hickox et al. 2014). Recently,Chen et al. (2013) used the average AGN luminosity andfound that the SMBH accretion rate is directly linked tothe SFR in star forming galaxies at z ∼ . . < z < L X and SFR. Wealso compare our results with Chen et al. (2013) (hear-after C13). Since we use SED fits with UV and opticalbands to estimate SFRs and C13 use Herschel /SPIRE250 µm measurements, we use Equation (1) to converttheir estimated IR luminosity to SFR for comparison.Figure 14 shows L X for our sample plotted as a functionof SFR. Gray points indicate individual X-ray detectedAGN, while blue circles show the average L X of the graypoints in bins of SFR. Error bars on each median pointreflect the standard deviation in that bin. The blue solidline shows the best fitted line to our median points, usinga non-linear least square fit. We emphasize that this lineis for X-ray detected AGN only.Pink stars show detected AGN in the C13 sample,which includes 34 X-ray detected AGN (using Chandra ACIS-I 5 ks observations with L X . − kev ) and 87 MIRdetected AGN (using 4.5 µm observation from Spitzer ) instar forming galaxies in the Bo¨otes field at 0 . < z < . 8. From AGN detected in both X-rays and the MIR,C13 used the median L X L IR to estimate the L X for MIRdetected AGN. The solid pink line shows the average L X found for both detected and undetected AGN, where C13 stacked the X-ray emission around active and non-activegalaxies, removing the contribution from XBs in the X-ray stack. The green stars show X-ray detected AGNin Symeonidis et al. (2011), which includes 17 LIRG andULIRG galaxies at z ∼ 1, from the 2 Ms observations inthe CDFN.In our detected AGN sample, there is a slight but sig-nificant trend here with a correlation coefficient of 0.24( p = 0.00) between the average L X and SFR. This issomewhat shallower than the trend in the C13 detectedAGN sample, but where the samples overlap there isgood agreement. C13 are not able to detect AGN with L X . erg s − due to their shallower X-ray data,therefore their points do not probe to as low SFR as oursample. Since the total X-ray area of PRIMUS is smallerthan the area probed by C13, and BLAGN are not in-cluded in this study, our sample lacks the highly starforming galaxies above log( SF R M ⊙ yr − ) ∼ L X and SFR found in C13, although this sample only in-cludes AGN in higher SFR sources (LIRGs and ULIRGs).Furthermore deeper X-ray data may provide greater dy-namic range in L X revealing a steeper trend. Our sampleincludes lower SFR sources but it does not probe as deepin X-ray luminosity and therefore is difficult to comparedirectly with the Symeonidis et al. (2011) sample.Overall, our sample of X-ray detected AGN shows alarge scatter in L X at a given SFR, though we find a weakbut significant correlation between them. This correla-tion indicates that the rate of black hole growth is relatedto the SFR in star forming galaxies, when effectively av-eraging black hole growth over long timescales, consistentwith stochastic fueling of the AGN from the same ulti-mate fuel supply as that for star formation. While herewe averaged only our X-ray detected sources, ideally wewould want to take the average over the entire galaxysample. To do this properly we would need to stack ourgalaxy sample, which is beyond the scope of this paper.We also note that the X-ray flux limit of our survey im-pacts the correlation that we find and that with deeperX-ray data we would be able to investigate this correla-tion more precisely. Where do AGN live? In our study, we find a large scatter between SFR and L X , with little evidence of a direct correlation, when con-sidering X-ray selected AGN with either star forming orquiescent host galaxies. However, our results from Sec-tion 3.5 indicate that, when considering the entire galaxypopulation, one is more likely to find an AGN in a starforming galaxy. Within either the star forming or quies-cent galaxy populations, we find AGNs with a wide rangeof specific accretion rates, described by a roughly power-8 Azadi et al. law distribution. However, for a given λ , the probabilityof a star forming galaxy hosting an AGN is higher thanfor a quiescent galaxy. Enhanced AGN activity in starforming galaxies has also been seen in several recent stud-ies (e.g. Silverman et al. 2009; Mullaney et al. 2012b;Aird et al. 2012; Rovilos et al. 2012; Rosario et al. 2012).The differences in the distributions can either be inter-preted as an increased probability of AGN activity be-ing triggered in galaxies with large reservoirs of cold gas(that also fuel star formation), or that AGNs in suchgalaxies are accreting, on average , at higher rates (seealso Georgakakis et al. 2014).These findings are consistent with the picture discussedabove, where the level of AGN accretion in a given galaxycan vary substantially over short time periods (relativeto the star formation timescales), and could explain thatlack of a strong, direct correlation between SFR and L X :while the overall probability of hosting an AGN is higherfor higher SFRs, the instantaneous accretion rate that weobserve from a single galaxy can vary over many orders ofmagnitude, washing out any direct correlation betweenthe SFR and L X .While this is appealing, it is important to note whereour results do not fit in with this simple picture. Firstly,we do find AGNs in quiescent galaxies that may havevery low levels of star formation and the distribution hasa similar power-law shape (albeit shifted to lower λ ), in-dicating that the underlying physical processes that reg-ulate AGNs may be similar in quiescent galaxies to thosetaking place in star forming galaxies (although ultimatelythe large scale fueling processes may be different). Sec-ondly, we do not find a rise in the probability of host-ing an AGN with SFR within the star forming galaxypopulation itself, indicating that increased star forma-tion does not go hand-in-hand with increased (average)BH growth. Conversely, for quiescent galaxies with re-duced SFRs –placing them below the main sequence ofstar formation– we find that the probability of hostingan AGN is decreased by a factor ∼ − 3. Furthermore,as SFRs decrease within the quiescent population we findthat the probability of hosting an AGN also decreases, in-dicating that as star formation is shut down there is alsoa reduction of AGN activity in quiescent galaxies. Never-theless, as emphasized above, AGNs are still widespreadwithin quiescent galaxies, with a wide range of specificaccretion rates.We note that the number of bins we used to classify ourhost galaxies is limited by our sample size, and with thecurrent binning we do not have a sufficiently large sampleto directly measure the shape of p ( λ | M ∗ , z ) within eachbin. Larger samples would allow us to accurately trackchanges in the distribution of specific accretion rate asa function of the host galaxy properties and would shedlight on any change in the underlying physical processes. SUMMARY In this paper we study the relationship between AGNX-ray luminosity and their host galaxies’ SFR and stel-lar mass. We use a sample of 309 X-ray selected AGNwith spectroscopic redshifts from the PRIMUS survey at0 . < z < . 2. We exclude BLAGN to minimize thecontribution of AGN light when estimating host galaxyproperties, and we include AGN with 10 < L X < erg s − . Our main conclusions are as follows: • Star formation rate and AGN luminosity are notstrongly correlated within our X-ray AGN sampleat 0 . < z < . 2. There is a wide range of SFRs ata given L X , and a higher L X does not necessarilyimply a higher SFR. We do not find any significantcorrelation between SFR and L X in star forminghost galaxies, though we do find a weak but signif-icant correlation between the mean L X of detectedAGN and SFR. This correlation implies an under-lying connection that may exist due to a commongas supply but the variability of AGN accretion onrelatively short timescales makes it hard to observe. • AGN with a wide range of L X reside in both starforming and quiescent galaxies with a wide rangeof stellar masses, although are generally found inmoderately massive ( & M ⊙ ) galaxies. How-ever, we do not find any correlation between stellarmass and L X within our X-ray AGN sample for ei-ther the star forming or quiescent host populations. • We find a wide range of specific accretion rates, λ ( L X normalized by host stellar mass), across thestar forming and quiescent host populations, whichcould explain the lack of a stronger correlation be-tween SFR and L X . • The fraction of AGN residing in star forming hostgalaxies increases with increasing AGN X-ray lu-minosity, indicating that more powerful AGN aremainly hosted in star forming galaxies at z < • Finally, we consider the fraction of AGN withinthe entire galaxy population. The probability thata galaxy of a given stellar mass, M ∗ , and redshift, z hosts an AGN as a function of specific accretionrate, p ( λ | M ∗ , z ), is roughly a power law for bothstar forming and quiescent host galaxies. The prob-ability of hosting an AGN at a given specific accre-tion rate is higher for star forming galaxies thanquiescent galaxies. Furthermore, this probabilityincreases with SFR within the quiescent galaxypopulation, though within the star forming popula-tion there is no change across the “main sequence”of star formation.Within star forming galaxies, known to contain abun-dant cold gas, we find no direct correlation between SFRand instantaneous AGN activity, although the overallprobability of hosting an AGN is higher than in qui-escent galaxies. Conversely, in quiescent host galaxies,where the overall probability of finding an AGN is some-what lower, we do find evidence for a correlation betweenSFR and AGN instantaneous luminosity which may sug-gest different triggering and fueling processes (e.g. minormergers, secular processes) drive both star formation andAGN activity in such galaxies. However, the distribu-tion of accretion rates in both star forming and quiescentgalaxies has a similar approximately power-law form, in-dicating that AGN accretion is ultimately a stochasticprocess and that the same physical processes may regu-late AGN activity once gas is funneled to the central fewparsecs.We thank the referee for their positive comments andconstructive advice which helped improving this paper. RIMUS: The relationship between Star Formation and AGN accretion