Accretion and star formation rates in low redshift type-II active galactic nuclei
aa r X i v : . [ a s t r o - ph . GA ] J u l Mon. Not. R. Astron. Soc. , 1–16 (2002) Printed 8 November 2018 (MN LaTEX style file v2.2)
Accretion and star formation rates in low redshift type-IIactive galactic nuclei
Hagai Netzer ⋆ School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel
ABSTRACT
Accretion and star formation (SF) rates in low redshift SDSS type-II active galacticnuclei (AGN) are critically evaluated. Comparison with photoionization models indi-cates that bolometric luminosity ( L bol ) estimates based on L([O iii ] λ L bol in low ionization sources such as LINERs. An alternative methodbased on L(H β ) is less sensitive to ionization level and a novel method, based on a com-bination of L([O iii ] λ i ] λ L/L
Edd with no indication fora change in accretion mechanism, or mode of mass supply. There are very few, if any,LINERs in all type-I samples which results in a much narrower
L/L
Edd distributioncompared with type-II samples. 3. There is a strong correlation between SF luminos-ity, L SF , and L bol over more than five orders of magnitude in luminosity. This leads toa simple relationship between bulge and BH growth rates, g ( bulge ) /g ( BH ) ∝ L − . bol ,where g ( bulge ) /g ( BH ) ≃
115 for L bol =10 ergs s − . Seyfert 2s and LINER 2s followthe same L SF - L bol correlation for all sources with a stellar age indicator, D n L bol , L SF , L/L
Edd and the specific SF rate follow D n Key words:
Galaxies: Active – Galaxies:Seyferts – Galaxies: Black holes – Galaxies:Nuclei – Galaxies: star formation
Black hole (BH) masses in thousands of type-I active galacticnuclei (AGN) can now be obtained from optical-UV spec-troscopy. The method is based on the known relationshipbetween the luminosity of the non-stellar continuum at somewavelength (e.g. 5100˚A) and the mean broad line region(BLR) size derived from reverberation mapping (RM; e.g.Kaspi et al. 2000; Kaspi et al. 2005; Vestergaard and Peter-son 2006; Bentz et al. 2009). This size is combined with amean gas velocity estimated from broad emission line pro-files (e.g. H β ), and the virial assumption about the gas mo-tion, to obtain M BH . The accuracy of the method has beendiscussed in various papers (e.g. Bentz and Peterson 2006)and is estimated to be a factor of about 2. Complications dueto the effect of radiation pressure force on the motion of theBLR gas (Marconi et al. 2008) are probably not very impor- ⋆ E-mail: [email protected] tant, at least for low redshift low luminosity AGN (Netzer2009, hereafter N09; see also Marconi et al. 2009). Estimatesof the normalized accretion rate,
L/L
Edd , are obtained bycombining the derived M BH with estimates of the bolometricluminosity, L bol .Different methods are required to obtain mass and ac-cretion rates in type-II AGN, where the non-stellar contin-uum is not directly observed. In low redshift type-II sourcessituated in bulge dominated hosts, the stellar velocity dis-persion in the bulge of the galaxy, σ ∗ , can be transformedto BH mass (e.g. Tremaine et al. 2002). The bolometric lu-minosity is usually estimated from known relationships be-tween L bol and the luminosity of certain narrow emissionlines. The assumption is that the line luminosity representsthe same fraction of L bol in all AGN. A line that is com-monly used is [O iii ] λ L bol is obtained from observations of type-I sources, whereboth the line and the non-stellar continuum are directly ob-served. Such scaling is discussed and explained in several c (cid:13) Hagai Netzer papers, e.g. Heckman et al. (2004), Netzer et al. (2006),Kewley et al. (2006; hereafter K06), Kauffmann and Heck-man (2009; hereafter KH09) and N09. A related methodwhich is less sensitive to reddening correction is based onthe luminosity of the mid-IR line [
OIV ] 25 . µ m (see discus-sion and suggested conversion factors in Dasyra et al. 2008).Unfortunately, the number of sources with such mid-IR mea-surements is very small. Yet another method is based on theluminosity of the hard X-ray continuum which is directly ob-served in most type-I and type-II AGN. While potentiallyinsensitive to reddening and other complications, the exactbolometric correction factor required to convert the 2–10keV luminosity to L bol depends on the global SED and isstill debatable (see e.g. the differences between Marconi etal. 2004 and Vasudevan and Fabian 2007).The emission line scaling method was used by Heckmanet al. (2004) to investigate mass and accretion rates in a largesample of data release one (DR1) Sloan digital sky survey(SDSS; York et al. 2000) type-II AGN. L([O iii ] λ L bol ≃ iii ] λ iii ] λ L bol = 600L([O iii ] λ L bol and SFluminosity ( L SF ). This requires re-evaluation of the meth-ods used to estimate L bol in type-II AGN which is presentedin §
2. A new, improved L bol indicator is used, in §
3, to re-evaluate M BH and L/L
Edd in LINERs and in Seyfert 2s.I also use the same indicator to test various AGN and SFcorrelations. § =70 km/sec/Mpc, Ω m = 0 . Λ = 0 . iii ] λ ) L([O i ] λ ) AND L(H β ) ASBOLOMETRIC LUMINOSITY INDICATORSIN TYPE-II AGN2.1 Theoretical line and continuum luminosityratios The current analysis applies to narrow emission lines ingalaxies that contain both an active AGN with a narrowemission line region (NLR), and starburst (SB) ionized gas.The term SF is perhaps more appropriate to discuss the var-ious processes addressed below but SB has been used in thepast to describe such galaxies and I keep to this terminol-ogy when discussing those issues where the name SB is com-monly used. While the NLR properties have been studied,extensively, observationally and theoretically, the combina-tion of AGN and SB excited gas in the same host presentsa real challenge, especially in sources observed with a largeentrance aperture. A main tool for distinguishing the twotypes of excitation is based on BPT (Baldwin, Phillip and Terlevich 1981) line ratio diagrams where AGN and SB re-gions are well separated. Detailed explanation, and criticaldiscussion of such methods can be found in numerous papersincluding Veilleux and Osterbrock and (1987), Kauffmannet al. (2003a, hereafter K03), K06, Brinchmann et al. (2004,hereafter B04), and Groves et al. (2006a; 2006b).Photoionization by a non-stellar radiation field can ex-plain the large observed range of ionization and excita-tion conditions in AGN. In particular, it can explain theL([O iii ] λ β ) (hereafter [O iii ]/H β ) line ratio in thehighest ionization type-II Seyfert galaxies ([O iii ]/H β ∼ iii ]/H β ∼ iii ] λ β ) and L([O i ] λ µ m continuum. 2. The bolometriccorrection factor, BC, which specifies the ratio of the totalluminosity to the optical continuum luminosity. The BC isnot directly observed and its value is estimated from anal-ysis of various emission lines. 3. The normalization of theoptical and X-ray fluxes, e.g. the value of α ox . 4. The shapeof the X-ray continuum.Experimenting with a range of possible continuumshapes shows that the two that are shown in the left panel ofFig. 1 bracket most realistic SEDs. The first of these (SED-1)is a combination of a disc-like optical-UV continuum, com-bined with a L ν ∝ ν − . X-ray (0 . < E <
50 keV) power-law. In this case, BC, which is defined here as L bol / L , c (cid:13) , 1–16 ccretion and star formation rates in type-II AGN .1 1 1010 −11 −10 −9 −8 Rydberg no r m a li z ed L ν SED − 1SED − 2 .1 1 10RydbergSuppressed blue bump
V1V2V3V4 V5
Figure 1.
Left panel: Two assumed SEDs representing the rangeof continuum shapes in AGN. The top curve (SED-1) is made ofa disk-like optical -UV continuum and a powerlaw F ν ∝ ν − . X-ray (E > α ox =1 .
35. SED-2 has the same optical and X-ray shapes and the twoare combined with a F ν ∝ ν − . powerlaw. Here BC=14 and α ox = 1 .
05. Right panel: High accretion rate SED (V1) andfour assumed low accretion rate SEDs obtained by truncatingthe “blue bump” in different ways keeping α ox unchanged (1.05). where L is λL λ at 5100˚A, is 7 and α ox = 1 .
35. SuchSEDs are typical of high luminosity AGN. The second case(SED-2) combines the above X-ray powerlaw and the opti-cal disc-like continuum with a Lyman continuum powerlawgiven by L ν ∝ ν − . . In this case BC=14 and α ox = 1 . α ox are more typical of low luminos-ity AGN. The two continua shown in Fig. 1 are normalizedto have the same ionizing luminosity to emphasize the factthat SED-1 produces relatively more photons close to theLyman edge.The second possibility is that different phases of activityare associated with different SEDs. For example, a luminous,fast accreting phase may be associated with an SED similarto one of the continua shown in the left panel of Fig. 1 whilea less luminous, slower accreting rate phase, might show adifferent SED. This may be the result of a change in thespectral properties of the central accretion disc, in particu-lar a significant reduction in the “blue bump” flux. Such apossibility is illustrated by the various curves shown in theright panel of Fig. 1. The top SED, marked V1, representsthe high accretion rate phase. It is similar to SED-1 exceptthat a relatively stronger X-ray continuum, typical of lowluminosity AGN, has been used. In this case BC=12.4 and α ox = 1 .
1. The additional curves, marked V2–V5, repre-sent several possibilities where the accretion disc continuumis significantly reduced. Such spectra have been speculatedfor LINERs and the ones shown here were motivated bythe recent work of Maoz (2007). According to Maoz (2007),there is a strong UV (2500˚A) point-source continuum in sev-eral nearby LINERs. In those cases, α ox ∼
1, very similarto the values observed in Seyfert 1 galaxies. It thus seemsthat either there are no significant changes in the near UV-SED between low and high accretion rate phases, or elsesuch changes only affect the part of the continuum between ∼ ∼ . cm − , enough to make the cloudoptically thick to the ionizing continuum radiation, and hy-drogen number density of n H = 10 cm − , low enough toavoid strong collisional de-excitation of the [O iii ] λ U = Q ( H ) / πr n H c , where r is thecloud central source distance and Q ( H ) the rate of emissionof Lyman continuum photons. This was done for two generictypes of models, one with ISM-type depletion and one fordust-free gas. The gas composition was changed between oneand four times solar and the depletion fraction is assumed tobe independent of abundance. The models shown here areall for dusty clouds which are thought to be more appro-priate for NLR conditions. I have included several modelssimilar to the cases discussed in Groves et al. (2004; 2006b)where radiation pressure force operating mostly on grains,determines the internal pressure in the cloud. Experimentingwith the various parameters confirms that U , the dust con-tent and the SED shape are the most important parametersthat determine the interesting line ratios.The results of some of the calculations are given inFig. 2. The figure shows the fraction of L bol emitted bythe three lines in question, [O iii ] λ i ] λ β .The ratios are calculated assuming a covering factor of unity(Ω = 4 π ). The left panel of Fig. 2 shows a noticeable dif-ference between the behaviors of [O iii ] λ β . Theintensity of [O iii ] λ iii ])/L bol varies by about 1.7 dex overthe relevant range of 10 − < U < × − . In contrast,L(H β )/L bol is basically constant (less than 0.3 dex) over thesame range which reflects the constant fraction of the Ly-man continuum photons absorbed by the optically thick gas.Such results are well known and well documented in numer-ous publications. In dust-free large U models (not shownhere), L(H β )/L bol and L([O iii ])/L bol are larger by abouta factor two compared with the dusty models shown here.Dusty gas clouds produce less [O iii ] λ β photondue to the absorption of part of the ionizing radiation by thedust and the destruction of locally emitted line photons bygrains. The first process is more important in high ionizationgas. The latter affects H β more than [O iii ] λ iii ] λ iii ])/L bol since the spectralshapes of the two are quite similar at high energies. This isnot the case for L(H β )/L bol because of the larger fraction ofionizing photons just beyond the Lyman edge in SED-1 (seeFig. 1). All this, again, is well known from earlier calculations(e.g. Netzer and Laor 1993). c (cid:13) , 1–16 Hagai Netzer .0001 .001 .01.0001.001.01 Ionization parameter L ( li ne ) / L bo l .1 1[OI]/[OIII] Figure 2.
Left panel: Calculated L([O iii ])/L bol (black)L([O i ])/L bol (blue) and L(H β )/L bol (red) ratios for various pho-toionized, dusty solar composition clouds with full covering of theradiation source. Thin lines: SED-1. Thick lines: SED-2. Thickdashed line: SED-2 and radiation pressure dominated clouds.Right panel: Similar calculations for the V1 continuum. Samecolour coding, same ratios as left panel but vs. [O i ]/[O iii ]. Thinlines, solar abundances. Intermediate width lines, 3 × solar abun-dances. Thick lines, 3 × solar, constant pressure model. I also calculated L([O i ])/L bol curves and show themon the same scale. This ratio is somewhat more sensi-tive to U than L(H β )/L bol but is much less sensitivethan L([O iii ])/L bol . The differences between [O i ] λ iii ] λ i ]/[O iii ]. This line ratiovaries with U in a way which is different from [O iii ]/H β . Allthis is the basis for the new calibration method discussed in § L bol re-radiated by thenarrow [O iii ] λ L bol re-radiated by H β and [O i ] λ U . L(H β ) seems tobe the most reliable L bol indicator for pure AGN excitation.Similar line-to-continuum luminosity ratios were alsocalculated for the scenario illustrated in the right panel ofFig. 1. Table 1 lists some of the results. The first row inthe table lists luminosity ratios for a high S2 accretion ratephase assuming the V1 SED. I fixed the value of U to give[O iii ]/H β = 10, typical of many such sources. The other fourrows represent low accretion rate phases with the V2–V5SEDs. Given an SED, I fixed U to produce [O iii ]/H β ≃ iii ])/L bol drops much more (factors of 10–20) com-pared with the drop in L([O i ])/L bol and L(H β )/L bol (factors1.5–2). The reason is that the difference in ionization poten-tial between hydrogen and O +1 is small and changes in theflux of the O +1 ionizing photons are associated with sim-ilar changes in the number of Lyman continuum photons.Most of the reduction in [O iii ]/H β is due to changes in U and cannot be explained by changes in continuum shape, atleast not for the SEDs considered here.To summarize, large changes in [O iii ]/H β are associ-ated with large changes in L([O iii ])/L bol . In contrast, thechanges in L(H β )/L bol and L([O i ])/L bol following a compa-rable change in [O iii ]/H β are much smaller even for dra-matic changes in SED shapes like the ones considered here. L bol estimators The most reliable estimates of L([O iii ])/L bol andL(H β )/L bol are obtained from observations of type-IAGNs. In such sources, L , L([O iii ] λ β ) (due to blending with the broad H β ),are directly observed and luminosity dependent bolometriccorrections are known with reasonable accuracy (Marconiet al. 2004; Netzer et al. 2007; Shen et al. 2008; Vestergaard2008; see summary in N09). This allows a direct conversionof line to continuum luminosity.To quantify this, I used type-I data from the SDSS/DR5sample described in Netzer and Trakhtenbrot (2007, here-after NT07). As explained in N09, the sample is incompleteat z . z > .
2. The sample containsabout 9000 radio quite (RQ) type-I AGN with z . iii ] λ β ) and L , for mostof these objects. The number of sources in the redshift in-terval 0.1–0.2 is 1333 and the number of sources with re-liable narrow [O iii ]/H β measurements is 1172. The fittingprocedure used to measure the lines, and the other detailsof the analysis are described in NT07. The bolometric cor-rection factor I used is somewhat different than the oneused by NT07 and is taken to be BC=9 − log L , where L = L /10 erg s − (see N09).Several type-II samples are used in the present analy-sis. The first was extracted from the SDSS/DR4 (Adelman-McCarthy et al., 2006) archive which is available on theMPA URL site . The archive includes redshifts, stellar ve-locity dispersion measured through the 3 ′′ SDSS fibre, linefluxes, the D n (4000) index (luminosity weighted mean stel-lar age) and various other properties. The stellar velocitydispersion was used to derive BH mass. This is a reliableprocedure for elliptical and bulge galaxies but gives poorerresults for disk and pseudobulge systems. In this work I amstudying mass and accretion rate distributions in type-IIAGN and I also make a comparison with type-I sources. Forthe first part I will try to avoid galaxies with less reliable BHmass estimates. This requires the removal of blue galaxiesfrom the sample and is explained in § (cid:13) , 1–16 ccretion and star formation rates in type-II AGN Table 1. L bol estimators for full covering dusty NLRs with solar composition and various SEDsshown in Fig. 1SED log U [O iii ]/H β L bol /L([O iii ]) L bol /L(H β ) L bol /L([O i ] λ fluxes were obtained in two ways, using the observed H α /H β line ratios. The first assumes standard galactic reddeningcurve combined with the assumption that the intrinsic ratiois H α /H β =2.86. This value is somewhat smaller than the oneexpected for S2s (3–3.1) but is a very good approximationfor L2s. The second is aimed to duplicate the values usedin K03, B04 and in Groves et al. (2006a, 2006b). Here theextinction is given by A λ ∝ λ − . . Part of the motivation touse such extinction is related to the issue of foreground vs.internal dust. Here it is only taken as one of two possible red-dening laws. The λ − . reddening law is very flat comparedwith the galactic law and results with significantly larger A V and intrinsic line luminosities. For example, the median red-dening for the high S/N (see below) sample of L2s and S2s,assuming galactic reddening, is A V = 1 .
07 and the equiva-lent number for the λ − . law is A V = 1 .
80. Fortunately, aconsistent use of the two reddening laws (see § L bol .Two additional complications related to reddening mustbe considered. First, the emission line reddening in type-IIAGN is thought to be larger than in type-I AGN. This hasbeen discussed in various papers, e.g. Netzer et al. (2006),and Melendez et al. (2008). Any reddening correction basedon average values obtained from type-I samples must in-clude this effect. Second, the average H α /H β in previousoptically selected and X-ray selected samples (e.g. the Bas-sani et al 1999 sample) is larger than the one measured here.This is possibly explained by K03 who noted the large rangein H α /H β and the tendency to detect the highest valuesin galaxies containing more SF gas (e.g. smaller D n iii ] λ α /H β in the SDSS S2sub-sample in order to choose the correct scaling betweentype-I and type-II sources.The total number of L2s and S2s in the sample con-sidered here is about 85,000 and the maximum redshift isabout 0.25. The main purpose is to obtain the most reli-able L bol by using various emission lines. Therefore, most ofthe results pertain to sources with S/N > α , H β ,[O iii ] λ ii ] λ i ] λ § ii ]/H α vs. [O iii ]/H β K03 criteria for separating AGNfrom SB galaxies. This group is referred to here as the “Ka03sample”. The second and third groups are obtained by us- ing the Ke06 division lines based on the [O i ]/H α and the[N ii ]/H α vs. [O iii ]/H β line ratios (see K06). These crite-ria are aimed to isolate, as much as possible, pure AGNfrom starburst and composite sources. The objects that arethought to best represent this group are in the region of theBPT diagram defined by the combination of the two K06 cri-teria. There are 11,803 such sources and they are referred tohere as the “Ke06 sample”. The S2 sub-group contains 6641sources and the L2 group 5162 sources. Obviously, the ratioof the populations depends on the assumed S/N since L2shave, on the average, weaker emission lines. For example, as-suming S/N=1 in the Ka03 sample, I get N(L2)/N(S2)=2.5.The samples were selected by their reddening corrected lineratios. This introduces only a small difference compared withthe selection by uncorrected line ratios.An additional small sample is taken from Ho et al.(1997). These are high quality observations of nearby sourcesobserved through small apertures. Here the contributionfrom SB regions is small and can be more easily removed. Us-ing the Ho et al. definitions, I choose from the tables only L2sand S2s and avoided transition sources (e.g. those markedas L2/T2). BH mass estimates for most of these sources canbe obtained from σ ∗ measurements listed in the literature.The archive used for this purpose is LEDA (Paturel et al.2003) Every narrow emission line is a potential L bol indicator.However, lines like [N ii ] λ
1. The H β indicator . In principle, this is the best indi-cator because of the very flat dependence of L(H β )/L bol oncontinuum shape and ionization parameter (Fig. 2 and Ta-ble 1). The normalization of L(H β )/L bol can be obtainedfrom type-I observations where the optical AGN continuumis directly observed. A major source of uncertainty is the dif-ficult deblending of the relatively weak narrow componentfrom the strong, broad H β line. The fitting procedure takesthis into account by imposing several criteria, such as assum-ing the same profile for the narrow [O iii ] λ β linesand forcing lower and upper limits on [O iii ]/H β (see NT07).The typical uncertainty in this procedure is estimated to be0.2 dex.The values of L(H β )/L bol obtained from the NT07 sam-ple are shown in Fig. 3 as a function of [O iii ]/H β . Alsoshown are means and standard deviations in bins of 0.1 c (cid:13) , 1–16 Hagai Netzer β L ( li ne ) / L bo l Figure 3.
Estimated L([O iii ])/L bol (black) and L(H β )/L bol (red) in 0 . z . iii ]/H β .The smooth curves are the theoretical models of Fig. 2 (thin linesSED-1, thick lines SED-2) where the ionization parameter is re-placed by [O iii ]/H β . The lines are shifted down relative to Fig. 2by 1.4 dex. See text for more explanation. dex in [O iii ]/H β . The typical standard deviation is 0.25–0.3dex and is a combination of the uncertainties in the L(H β )measurements and the scattering in covering factor (see be-low). Measurements of L([O iii ])/L bol in the NT07 sampleare also shown. As expected from the photoionization cal-culations (Fig. 2), L([O iii ])/L bol increases almost linearlywith [O iii ]/H β while L(H β )/L bol is only weakly dependenton [O iii ]/H β .To make a direct comparison between models and ob-servations, I used the calculations shown in Fig. 2 to con-vert ionization parameter to [O iii ]/H β . For the case shownhere I chose the calculated L(H β )/L bol and L([O iii ])/L bol in the SED-1 and SED-2 dusty, solar composition clouds.This enables me to plot the theoretical curves of Fig. 2 ontop of the data in Fig. 3 by shifting the curves down tofit the observations. The required shift is 1.4 dex and theagreement between model and observations is very good.The difference between the vertical scales of Fig. 2 andFig, 3 is a manifestation of the fact that the NLR cover-ing fraction is much smaller than unity. The data shown inFig. 3 suggest that for narrow H β lines in low ionization([O iii ]/H β
4) type-I AGN, L bol ∼ , β ). The ra-tio increases smoothly to about 20,000 in high ionizationparameter sources ([O iii ]/H β ∼ α /H β in type-II sources. As explained inK03, and noted earlier, the amount of reddening depends onthe stellar population and is higher in smaller D n β redden-ing correction factors for S2s in the Ke06 sample are ∼ ∼ . λ − . reddeninglaw. The corresponding factors for L2s are ∼ . ∼ . λ − . approximation. Forcomparison, the median correction factor in the Bassani etal. (1999) sample, for galactic-type reddening, is close to 8.The S2 factors are more appropriate since there are very fewLINERs in the type-I sample I used for the calibration (see § β .Given all these considerations, including the differencesin extinction between type-I and type-II sources, result inthe following relation for reddening corrected L(H β ) and the λ − . extinction law,log L bol = log L ( H β )+3 . max (cid:20) ., . OIII ] H β − . (cid:21) . (1)For galactic reddening, the factor 3.48 should be replaced by3.75. The uncertainty on this ratio depends on the accuracyof the scaling shown in Fig. 3 and is estimated to be at leastas large as the size of the error bars shown in the diagram,i.e. about 0.3–0.4 dex.The above expression enables me to calculate the meanNLR covering fraction. The estimate depends on the amountof line photon destruction by internal dust which is alreadyincluded in the photoionization calculations of Fig. 2. As-suming this accounts for a factor of 1.5–2 out of the totalextinction, and a reddening correction factor of about 6 forH β in S2s, gives a mean NLR covering fraction of 7–15%.For galactic type reddening, the covering fraction is smallerby a factors of 1.5–2. Obviously, the covering factor changesfrom one object to the next which is probably the main rea-son for the large scatter in L(H β )/L bol seen in Fig. 3. Allthese estimates are based on the assumption of little or noextinction of the optical continuum in type-I AGN which, initself, is subjected to some uncertainty (e.g. Netzer 1990).The obvious limitation of the L(H β ) method is the dif-ficulty in estimating the SB contribution to the observedBalmer lines, especially in large aperture observations. Thishas been discussed, extensively, in K03, K06, Groves et al.(2006a), and KH09. K06 define, for each source, a quantitywhich describes its distance from the SB region in the rel-evant BPT diagram. Groves et al. (2006a) and KH09 sug-gested somewhat different procedures that serve the samepurpose. The H β method is more reliable in cases wherethe SB contributions are small, e.g. extreme S2s with large[O iii ]/H β or pure L2s. It is also the preferred method forsmall aperture observations, like the Ho et al. (1997) sam-ple.
2. The [O iii ] λ method . Estimates of L bol based onthe observed intensity of the [O iii ] λ iii ] λ L bol isabout 3000–3500. Netzer et al. (2006) show the clear depen-dence of this scaling on source luminosity and/or redshift.This effect is not taken into account in the present work be-cause the type-I and type-II SDSS populations are differentin terms of the fraction of low ionization (L2) sources andI prefer to use, instead, average values for the S2s (see dis-cussion below). Using any [O iii ] λ c (cid:13) , 1–16 ccretion and star formation rates in type-II AGN includes, therefore, another uncertainty which is related tothe source luminosity.The use of reddening corrected line intensities reducethe conversion factor by a factor of 3–6, depending on theextinction law used. For the λ − . extinction law I find L bol ≃ iii ] λ L bol ≃ iii ] λ iii ] λ iii ])/L bol onthe level of ionization. For a given L bol , the range inL([O iii ] λ iii ])/L bol which is cal-ibrated from S2 observations severely under-estimate L bol in L2s. This issue is illustrated in the simulations describedbelow.
3. The [O i ] λ method. The [O i ] λ L bol indicator. The total observed range in [O i ]/H α is about1.2 dex, similar to the observed range in [O iii ]/H β . Thus, theexpected uncertainty in L([O i ])/L bol , assuming the Balmerlines are the best L bol indicators, is large. However, the cal-culations (Fig. 2) show that the [O i ]/H α ratio is not verysensitive to the level of ionization of the gas, or its metallic-ity, and most of the range in [O i ]/H α is likely to be due tothe changes in SEDs (e.g. Groves 2006a) and/or the columndensity of the NLR gas. There are clearly some trends in[O i ]/H α that are not entirely understood.Given the theoretical calculations, and the mean ob-served line ratios in the Ke06 sample, I estimate L bol ≃ , i ] λ L bol ≃ i ] λ λ − . reddening law. For galactic extinction,the numbers are larger by a factor of ∼
4. The [O i ]/[O iii ] method. This is a new indicator whichis based on the theoretical calculations, the relative weaknessof [O i ] λ i ]/[O iii ] line ratio decreaseslinearly with the ionization parameter. It can thus be usedto estimate the changes in the level of ionization of the gas.This can provide the needed correction for the [O iii ] λ iii ]/H β vs. [O i ]/[O iii ] for the high S/N Ka03 and Ke06samples. The clear strong trend confirms the similar depen-dences of the two line ratios on the level of ionization ofthe gas. As explained, the Ke06 sample was chosen to be re-moved, as much as possible, from the SB region. This sampleis shown by red points and the Ka03 sample by black points.I have also added the Ho et al. (1997) S2s (green triangles)and L2s (blue triangles). As expected, these small apertureobservations lie in the Ke06 zone away from the SB region.To test the agreement with the theoretical calculations,I draw three lines corresponding to three calculated modelswith various assumptions about metallicity, SED and therole of radiation pressure (see figure caption). All modelspass through the Ke06 and the Ho et al. (1997) points ver-ifying that the red colored part of the diagram is indeeddominated by the AGN contribution. The curves that rep- .1 1.1110 [OI]/[OIII] [ O III]/ H β Figure 4. [O iii ]/H β vs. [O i ]/[O iii ] for sources in the Ke06 (redpoints) and Ka03 (black points) samples. The Ho et al. (1997)S2s are shown as green triangles and the L2s as blue triangles.The black curves represent three theoretical dusty cloud models:1. V1 SED from Fig. 1 with 3 × solar abundance (thick solid line).2. V1 SED and constant pressure dominated clouds (thick dashedline). 3. SED-1 solar composition constant pressure clouds (thindashed line). resent models with different abundances also show that thegas metallicity does not affect much the above line ratios.The observations and the models suggest the following linearrelationship in the AGN dominated part of the diagram,log [ OIII ] H β = − .
75 log [ OI ][ OIII ] − . . (2)The uncertainties on the two terms, from the fit procedureonly, is about 10%. Since [O iii ]/H β is an ionization parame-ter indicator, and since this parameter determines the exactL([O iii ])/L bol (Fig. 2), the two can be combined to derivethe following expression for the case of a λ − . reddeninglaw,log L bol = 3 . .
25 log L ([ O III ] λ .
75 log L ([ OI ] λ . (3)For galactic reddening, the constant 3.53 should be replacedby 3.8. The normalization of L bol is based on eqn. 2 at thelimit of the largest [O iii ]/H β and the agreement betweentheory and observations in type-I sources (Fig. 2). The un-certainties for individual measurements are due mainly tothe uncertainty in the slope of eqn. 2, the range in the NLRcovering factor and the assumed bolometric correction fac-tor, BC. The combined uncertainty cannot be smaller than ∼ . To further investigate these issues, I ran simple simulationsdesigned to mimic the distributions of SBs, AGN and com-posite sources in the [O iii ]/H β vs. [O i ]/H α plane. The sim-ulations are shown in Fig. 5. The starting points are the dis-tributions shown in K06 Fig. 1c and Ho (2008) Fig. 2. Theseare used to define “pure SB” (blue points) and “pure AGN” c (cid:13) , 1–16 Hagai Netzer .01 .1.1110 [OI]/H α [ O III]/ H β AGNCompositeSB .01 .1 1[OI]/H α Figure 5.
Simulated BPT diagrams. Pure AGN are marked withred points and pure SB with blue points. Left panel: The assumedlocations of SBs and AGN and observed sources from the Ke06sample (small black points). Right panel: Simulated compositesources (large black squares) and observed Ke06 sources (smallblack points). The green curve marks the Ke06 AGN-SB divisionline. (red points) regions in the left panel of the diagram. Alsoshown are all sources from the Ke06 sample (small blackpoints) and the Ke06 division line separating SB from AGN.These simulated SBs and AGN were used to create a sampleof composite sources that represent different combinationsof the two in terms of their L bol (represented by the H β luminosity) and different line ratios.I start by randomly mixing the two populations us-ing their H β luminosity, i.e. for a total luminosity L(H β ),a fraction f L(H β ) is contributed by the AGN and a fraction(1 − f )L(H β ) by the SB, where f is distributed uniformlyover the range 0–1.. I then chose, randomly, using the blue(SB) and the red (AGN) areas, values for [O iii ]/H β and[O i ]/H α for the simulated SB and AGN. These are combinedto form a composite spectrum with composite [O iii ]/H β and[O i ]/H α line ratios. These simulated sources are shown asblack squares in the right panel of Fig. 5. The compositesources are the only ones analyzed here and the points rep-resenting pure AGN and pure SBs in the left diagram areonly shown to demonstrate their locations. Clearly, there isa good agreement between the areas occupied by real andsimulated sources in the right panel of Fig. 5.It is important to note that the density of the simulatedsources in Fig. 5 is not the same as the density of observedsources. Instead, the simulations assume evenly spread val-ues of f and evenly spread [O iii ]/H β and [O i ]/H α line ratiosin the SB and AGN regimes. The diagrams emphasizes theproblem of identifying the composite sources whose locationsoverlap with the pure SB and pure AGN regions.For each of the simulated composite spectra I calcu-lated the “real” L bol of the AGN (i.e. the one obtained from f L(H β )) and compared it with the “estimated” L bol , i.e. Note that B04 assigned a single location to all AGN and KH09two representing points, one for S2s and one for L2s. Here I usea much large range determined by the observations
11 H β (AGN)/H β (total) E s t i m a t ed L B o l / R ea l L B o l E s t i m a t ed L B o l / R ea l L B o l β (composite) Figure 6.
Estimated L bol relative to the real (i.e. assumed bythe simulation) L bol using the [O iii ] λ i ]/[O iii ] (blue points) methods. The upper panels assumethat the SB contributions to the [O iii ] λ i ] λ iii ] λ i ] λ β flux emitted bythe AGN. The vertical line marks the 1/3 division line, i.e. theSB fraction in left to the line is greater than 2/3. the one that would have been obtained by using the var-ious L bol indicators. Four examples are plotted in Fig. 6.The top two panels compare estimated-over-real L bol us-ing the [O iii ] λ i ]/[O iii ] (bluepoints) indicators. The two are plotted against the ob-served [O iii ]/H β and the AGN fraction in the compositespectrum. This simulation is done under an optimistic as-sumption that the SB contribution to each of the lines cancompletely and accurately be removed. As expected, the[O iii ] λ L bol in sourceswith large [O iii ]/H β and underestimates it in sources withsmall [O iii ]/H β . The [O i ]/[O iii ] indicator is distributedmore uniformly, in a narrower band, with a mean of 1.0 anda scatter of about 0.15 dex. The bottom panels show similarratios based on the “observed” [O iii ] λ i ] λ i ] λ i ]/[O iii ] method but superior to the [O iii ] λ iii ] λ c (cid:13)000
Estimated L bol relative to the real (i.e. assumed bythe simulation) L bol using the [O iii ] λ i ]/[O iii ] (blue points) methods. The upper panels assumethat the SB contributions to the [O iii ] λ i ] λ iii ] λ i ] λ β flux emitted bythe AGN. The vertical line marks the 1/3 division line, i.e. theSB fraction in left to the line is greater than 2/3. the one that would have been obtained by using the var-ious L bol indicators. Four examples are plotted in Fig. 6.The top two panels compare estimated-over-real L bol us-ing the [O iii ] λ i ]/[O iii ] (bluepoints) indicators. The two are plotted against the ob-served [O iii ]/H β and the AGN fraction in the compositespectrum. This simulation is done under an optimistic as-sumption that the SB contribution to each of the lines cancompletely and accurately be removed. As expected, the[O iii ] λ L bol in sourceswith large [O iii ]/H β and underestimates it in sources withsmall [O iii ]/H β . The [O i ]/[O iii ] indicator is distributedmore uniformly, in a narrower band, with a mean of 1.0 anda scatter of about 0.15 dex. The bottom panels show similarratios based on the “observed” [O iii ] λ i ] λ i ] λ i ]/[O iii ] method but superior to the [O iii ] λ iii ] λ c (cid:13)000 , 1–16 ccretion and star formation rates in type-II AGN .1 1.1110 [OI]/[OIII] [ O III]/ H β .1 1[OI]/[OIII] Figure 7.
Simulated [O iii ]/H β vs. [O i ]/[O iii ] diagrams obtainedfrom the data shown in Fig. 5. The left panel shows the entiresimulated sample and the right panel only sources above the Ke06division line. The green points in both diagrams are sources wherethe estimated L bol obtained by the [O i ]/[O iii ] method deviatesfrom the “real” (i.e. assumed in the simulation) L bol by morethan a factor two. better, and the [O i ] λ i ]/H α .A critical issue is the fraction of SB contribution tothe [O iii ] λ i ] λ β line is due to the AGN emission. At this mixing, the[O i ]/[O iii ] method is still producing L bol estimates that arewithin a factor 2 of the real values. This number seems tobe a practical limit when estimating the contamination ofthe [O iii ] λ L bol , inoptically selected samples.Finally I show in Fig. 7 the simulated source distri-bution in the [O iii ]/H β vs. [O i ]/[O iii ] plane. This plot isvery similar to the one obtained from the real observations(Fig. 4). There are two parts to the diagram. The left handside shows all composite sources (black points), pure SBs(blue) and pure AGN (red). It also shows in green all com-posite sources where the [O i ]/[O iii ] indicator results in L bol which deviates by more than a factor of two from the real L bol (12% of the sources). As expected, most of these sourceslie close to the border line between SBs and AGN. The righthand side is a similar diagram for sources in the Ke06 part ofthe diagram. The fraction of green sources here is only 2.5%.Thus, using the [O i ]/[O iii ] estimator one expects more than85% of the sources in the Ka03 AGN region to be within afactor of two of the intrinsic L bol of the AGN. In the Ke06 re-gion, the fraction is more than 95%. Note again that the frac-tions quoted depend on the assumed source density acrossthe entire plane. Hence, they only apply to samples that aredistributed with equal density over the assumed AGN andSB regions. .0001 .001 .01 .1.1 L/L Edd F r a c t i on BH < 7.3 .0001 .001 .01 .1L/L Edd BH < 8.37 < log M BH < 81.7 < D n .1 F r a c t i on BH < 81.2 < D n Figure 8.
A comparison of the various methods for estimating
L/L
Edd in several BH mass and galaxy type (D n z < . L/L
Edd . Black curves are values obtainedwith the [O iii ] λ β methodand blue curves with the[O i ]/[O iii ] method. The bottom panelsshow L/L
Edd distributions in two BH mass groups for all D n n M BH , as marked. L bol indicators I compared several
L/L
Edd distributions calculated with thevarious methods. The bottom panels of Fig. 8 show the dis-tributions in two BH mass groups, one with 10 M(BH) . M ⊙ and one with 10 M(BH) . M ⊙ . As dis-cussed in KH09, lower mass BHs are found, primarily, inhosts with younger stellar populations and are, on the aver-age, faster accretors. The large BHs are, typically, in hostswith older stellar population and lower L/L
Edd . The dia-gram confirms this finding but shows that the [O iii ] λ L/L
Edd by a factor of ∼
4. The bottom left panel shows also that the H β indica-tor over-estimates L/L
Edd in lower M BH sources, becauseof the SB contribution to the line. The bottom right panelshows the good agreement of the H β and [O i ]/[O iii ] indi-cators in high mass BH systems where the SB contributionis negligible.The upper panels of Fig. 8 compare L/L
Edd in differentgroups of the stellar age indicator, D n iii ] λ L/L
Edd is underestimated by a factor ∼ ∼ i ]/[O iii ] method. The recent KH09 paper shows apowerlaw distribution of L([O iii ])/M BH in old, large D n L/L
Edd in those galaxies. This change in the shape ofthe accretion rate distribution is explained by KH09 as dueto a different mechanism (stellar mass loss) of mass sup-ply to the central BH. The distribution of
L/L
Edd in the1.7 < D n < . i ]/[O iii ] method does not confirm this idea. Its shapecannot be fitted by a powerlaw and its mean L/L
Edd islarger by a factor of ∼ iii ])/M BH ) L/L
Edd found by KH09. This is a manifes- c (cid:13) , 1–16 Hagai Netzer .0001 .001 .01 .1.1 L/L
Edd F r a c t i on all S2 .0001 .001 .01 .1L/L Edd all L2
Figure 9.
Same as Fig. 8 for all Ke06 AGN at all redshifts dividedinto S2s and L2s. tation of the failure of the [O iii ] λ L bol in low ionization AGN.Fig. 9 shows a comparison of L/L
Edd distributions forall S2s and L2s in the Ke06 sample. The two groups, com-bined, cover a large range in BH mass, 10 . − M ⊙ , and havedifferent L/L
Edd . The diagram shows that the [O iii ] λ L/L
Edd in the slower accreting L2s,where the two other indicators agree very well. Again, thehistograms are not meant to show the real distribution of
L/L
Edd in all accreting BHs but rather to point to the mostreliable L bol indicators. Having established [O i ]/[O iii ] as the best L bol indicator fortype-II AGN, I now address two important physical issues:the comparison of mass and accretion rates in low redshiftL2s and S2s and the correlation of L bol with SFR and thespecific SFR (SSFR). Unlike the Heckman et al. (2004) andthe KH09 papers that address the combined AGN-SB pop-ulation, the present work discusses only AGN-dominatedsources, those where L bol is larger than the SF luminos-ity. I made no attempt to take into account accreting BHsin sources classified as SB galaxies. Such sources fall insidethe SB region on the Ka03 BPT diagrams and play an im-portant role in the KH09 analysis where, in several of thediagrams, they outnumber AGN. A major purpose of this work is to study mass and accre-tion rate distributions, and their various correlations, forL2s and S2s in the Ka03 and Ke06 samples. Since σ ∗ -basedestimates of BH mass are known to be reliable in bulge andelliptical galaxies, and unreliable in disks and pseudobluges,I made an attempt to remove the less reliable estimates. Thesimplest approach is to remove blue galaxies thus avoidingdisks and pseudobulges (e.g. Drory & Fisher 2007). Since u .0001 .001 .01 .1 1.1 L/L Edd F r a c t i on Red galaxies BH < 7.37.5 < log M BH < 7.88 < log M BH < 8.3 .001 .01 .1 1L/L Edd
Blue galaxies
Figure 10.
L/L
Edd distributions of Ka03 z . M BH groups as indicated. The distributions are shown separatelyfor red and blue galaxies. and r magnitudes are available for the entire sample (SDSS“model magnitudes”), I used the u − r method describedby Baldry et al.,(2004) to separate red and blue galaxies inthe Ka03 and Ke06 samples. About 75% of the AGN in theKa03 sample and about 85% in the Ke06 sample are foundto be in red hosts. As expected, the fraction of S2s in bluegalaxies is larger than their fraction in the entire popula-tion. The method is probably too conservative since thereare bulge galaxies that fall into the “blue” category, espe-cially the larger σ ∗ systems. Also, the border line betweenthe groups is not as sharp as the expression suggested byBaldry et al. All remaining analysis that refers to BH massand L/L
Edd in S2s and L2s is based on the red sub-sampleonly. For those cases where I compare type-I and type-IIproperties, I use the entire sample since no red sub-sampleis available for type-I AGN.Fig. 10 shows
L/L
Edd distributions for several M BH groups in z . L/L
Edd in the red sub-sample. As seen, the largerBHs in the local universe are the slower accretors, similar tothe behaviour in low and high redshift type-I samples. The
L/L
Edd distributions are broader than observed in type-Isources. This is mostly due to the lack of L2s in type-I sam-ples. There is no strong indication for a change in the shapeof the distribution when going to larger mass BHs. The rightpanel shows the same distributions for the blue sub-sample.Here there is no difference between the mass groups reflect-ing, most probably, the unreliable M BH estimates in suchhosts. Mixing the two sub-samples will cause an increase inthe intrinsic scatter and will tend to over-populate the larger L/L
Edd part of the diagram.The dependences of
L/L
Edd on [O i ]/[O iii ] and L bol for AGN in red hosts in the Ke06 sub-sample are shown inFig. 11. The S2s and L2s are marked with different coloursto demonstrate the smooth transition between the groups.Plotted also are L2s from Ho et al. (1997) that probe a muchlower L bol range. The left hand side shows that, without theHo et al. L2s, there is a strong apparent correlation thatmay be interpreted as an indication that the [O i ]/[O iii ] ra- c (cid:13) , 1–16 ccretion and star formation rates in type-II AGN .1 110 −6 −5 .0001.001.01.11 [OI]/[OIII] L / L E dd L Bol [erg/sec]
Figure 11.
Ke06 AGN in red galaxies divided into S2s (redpoints) and L2s (blue points). The left panel shows
L/L
Edd vs.[O i ]/[O iii ] and the right panel L/L
Edd vs. L bol . Large blue pointsdenote L2s from the Ho et al. (1997) sample. The two diagonalstrips in the right panel are two subclasses of the S2s and L2s pop-ulations. The top one contain AGN with 7 < log M BH < . M ⊙ and the bottom one AGN with 8 < log M BH < . M ⊙ . tio is a good accretion rate indicator. Adding the Ho et al.L2s change this conclusion and illustrates the double-naturedistribution of L/L
Edd with respect to [O i ]/[O iii ].The right hand side of the diagram explores the lumi-nosity dependence of L/L
Edd . As found in several earlierstudies, L bol and L/L
Edd are strongly correlated over morethan three orders of magnitudes in the SDSS sample, andover more than five orders of magnitude when including theHo et al. L2s. The
L/L
Edd range in the S2 class is smaller, ∼ L/L
Edd and L bol for L2s would drop by about a factor 5 if the [O iii ] λ M BH can be considered asan additional dimension of the L/L
Edd vs. L bol correlation.This is illustrated by the two parallel bands where the up-per stripe show L2s and S2s with M BH =10 − . M ⊙ and thelower one M BH =10 − . M ⊙ . There are clear and smoothtransitions between S2s and L2s and between all BH massgroups, over more than five orders of magnitude in L bol andno indication for a different sources of mass supply.Fig. 12 shows a comparison of 0 . < z < . L/L
Edd vs. L bol (left panel) and L/L
Edd vs. M BH (right panel). The divi-sion between S2s and L2s is the same as in Fig. 11 and thetype-I sources are shown by small black squares. L bol fortype-Is is measured as explained in N09 (the [O iii ] λ L/L
Edd range ofsuch sources for all mass BHs.The distribution of
L/L
Edd in high redshift AGN sam-ples has been studied, extensively, in several recent publi-cations. Kollmeier et al. (2006), Netzer et al. (2007), Shen .0001.001.01.11 L Bol [erg/sec] L / L E dd M BH /M sun Figure 12.
Left: Same as right panel of Fig. 11 but for 0 . Edd vs. M BH . et al. (2007), and Gavignaud et al. (2008), used various se-lected samples, with different redshifts and depths, to ad-dress this issue. The results of the various studies are quitedifferent due to, among other things, the selection methodsand the way used to estimate M BH . All those samples, aswell as other low redshift type-I SDSS samples, clearly donot contain LINERs. The reason is the much fainter AGNcontinuum and broad emission lines in such sources whichmakes their detection against the stellar continuum very dif-ficult. The L2 fraction in the present low redshift sample isas large, or even larger than the S2 fraction yet all the abovestudies completely ignored this population. Obviously, theremay well be fewer L2s at high redshifts yet it is very unlikelythat such sources are completely missing. This suggests thatthe overall range of L/L Edd in type-I AGN of all BH massis likely to be much larger than assumed so far. Most sources in the DR4 sample show clear indications forstarburst activity and long SF history. This has been dis-cussed in K03, K06, Groves et al. (2006a; 2006b), KH09 andvarious other papers. Very detailed discussions, with variousprescriptions of how to deduce the SFR, are given in B04and in Salim et al. (2007; hereafter S07).According to B04, the SSFR in galaxies hosting type-IIAGN can directly be obtained from the SDSS spectroscopy.This is achieved by calibrating the SB luminosity in “pure”SB systems against the stellar age indicator, D n α ) (see details inB04). This can then be applied to all AGN where D n α emission is produced in very young( few × yr) HII regions. This results in a big spread of c (cid:13) , 1–16 Hagai Netzer SSFR for a given D n − M ⊙ /yr/ M ∗ , the 68 per cent confidencelevel approaches 1.2 dex. The error is significantly reducedat higher SFRs. In addition, the SFRs for the entire galaxywere deduced by assuming that the SFR within fiber has thesame dependency on colour as the SFR outside of it. TheB04 recommendation for estimating SFRs for AGN hosts isto use the fibre-based values for both the SSFR and M ∗ .These numbers can be found in the DR4 archive and wereadopted here by using the median-based estimates (see gen-eral explanation in the DR4 archive).The more recent GALEX-based work of S07 providesan independent test of the B04 method. This work uses thetwo GALEX UV bands to provide UV-based SFRs for alarge number of SDSS galaxies. According to S07, the agree-ment between the attenuation corrected (using the Charlotand Fall 2000 two component method) H α -based and UV-based SFRs is extremely good for the younger SF systems.On the other hand, large deviations can be found in older,more massive galaxies, where the B04 method tends to over-estimate the SFR. It is however clear from their results (e.g.Fig. 19) that some SF activity, at a level of 0.1 M ⊙ /yr oreven lower, is observed in many AGN, especially the veryslow accretors. Inspection of the S07 results (e.g. their Figs.3, 4 and 19) suggest that the deviations from the B04 es-timates are large and the H α -based estimates are poor forgalaxies with 4. 000 > . 8. Also, the likelihood of no SF (their“no H α ´’ sources) is much larger for those galaxies with totalstellar mass exceeding about 10 . M ⊙ . A possible reasonfor this over-estimation of the SFR in large mass, old stel-lar hosts is the identification of a LINER excited H α line asdue to SF activity. Given all these uncertainties, I chose toavoid AGN-hosts with 4. 000 > . M ∗ > . whencomparing SFRs and SSFRs to L bol and L/L Edd .Having obtained L SF and SSFR for all AGN (excludingthose removed by the S07-based criteria), I can now examinethe correlations of these properties with L bol and L/L Edd .In the following I use the same M BH groups and the samedivision into S2s and L2s as in the previous sections. Theanalysis is meant to show various correlated properties but itis not adequate for estimating global population quantities,such as the total mass accretion onto massive BHs in thelocal universe. L SF and L bol correlations L bol in type-I AGN is readily obtained from optical andnear-IR spectroscopy. Measuring L SF in those sources ishampered by the strong non-stellar continuum and the in-tense emission lines that prevent reliable measurements ofthe narrow Balmer lines and/or D n L SF in luminous, high redshift sources isbased on infrared (IR) indicators, mostly PAH features inthe mid-infrared (MIR) and cold dust emission in the far-infrared (FIR). This has been discussed in numerous papers(see Sanders and Mirabel 1996 for the earlier work and Lutzet al. 2008 for references to more recent publications).PAH-based and FIR-based estimates of L SF in low andhigh redshift, high luminosity type-I AGN are described inSchweitzer et al. (2006), Netzer et al. (2007), Schweitzer et al. (2008) and Lutz et al. (2008). These papers de-scribe Spitzer /IRS spectroscopy of two type-I samples. TheQUEST sample (Schweitzer et al. 2006) contains 28 low red-shift ( z ∼ . 1) PG QSOs. About half of the sources showclear PAH 7.7 µ m emission that is interpreted as a SF sig-nature and converted to L SF using the correlation betweenL(PAH 7.7 µ m) and L(FIR). The stacked spectrum of allother QUEST QSOs shows the same feature with indicationof L SF which is only a factor of ∼ z = 2 − 3, using Spitzer /IRS observations of strong sub-mmQSOs. This provided L SF for sources of much higher lumi-nosity. Lutz et al. (2008) contains also estimates of L SF foradditional high redshift AGN observed by Spitzer /IRS. Theassumption in both studies is that L SF = νL ν at 60 µ m. Asshown in Netzer et al. (2007) for the QUEST sample, andin Lutz et al. (2008) for the high redshift QSOs, there is astrong correlation between L SF obtained from the FIR and L5100 over almost four orders of magnitude in L bol .I combined the present measurements of low redshifttype-II AGN with the above L(IR)-based estimates for type-I sources to show the combined L SF - L bol diagram for AGNdominated sources (Fig. 13). The error bars on L SF for in-dividual sources (yellow lines) are adopted from B04. Theyrepresent a 34 per cent confidence level. The low luminositytype-II AGN fall on the continuation of the same correlationas the high luminosity, type-I AGN. The slope of the cor-relation is about 0.8 ( L SF ∝ L bol . ) and L SF = L bol at about L bol =10 ergs s − which is outside the range of the sample.I note that the line shown here is not a fit to the dataand there was no attempt to estimate the error on its slope.Such a fit could not be reliably obtained not only becauseof the upper limits, but also because the SDSS sources, allat the low luminosity end, outnumber the higher luminositysources by a huge factor. It is only meant to show that a lineof this slope seems to connect all AGN-dominated sourceswith real detections over a large luminosity range. Note alsothat S2s (red points) and L2s (blue points) are well sepa-rated but follow the same general correlation. This suggeststhat L SF follows L bol in low as well as in high accretion rateAGN-dominated sources.The correlation in Fig. 13 may be the result of two se-lection effects. One is the omission of SF-dominated systemsabove the straight line and the other the absence of pureAGN below the line. The missing sources above the line arenot due to observational limitations. There are many strongSB galaxies in the SDSS sample that show weak or no AGNactivity and hence are not included in the present sample.There are also ULIRGs and SMGs with small or no AGNactivity. These are also not included in this AGN-dominatedcorrelation. The bottom part of high L bol and very low L SF is more problematic and quite typical of flux limited sam-ples. Given the exclusion of AGN hosts with large 4. 000 andlarge stellar mass, such sources are not expected to be miss-ing at low redshifts since the SDSS type-II AGN sample isthought to be complete to about z=0.1. However, the highluminosity AGN samples, especially those at the highest lu-minosity end, are known to be incomplete and the fractionof missing sources with very small SFR is not known. A sce-nario which is described below gives at least one possibilitythat does not require such sources.The left side of Fig. 14 shows a suggested time evolution c (cid:13) , 1–16 ccretion and star formation rates in type-II AGN L Bol L S F o r L µ m Figure 13. L SF vs. L bol for AGN. The type-II Ke06 AGN arethe small points with yellow error bars. The red points are S2sand the blue points are L2s. QUEST QSOs from Netzer et al.(2007) are shown as large black squares and high redshift QSOsfrom Lutz et al. (2008) as large red squares. Empty symbols withlines represent upper limits. The slope of the straight line is 0.8. from pure SB (thick line) to a combined powerful SB-AGNto to a combined weak SB-AGN. This is a “single event”that can occur, in principle, several times in the history ofa galaxy. The long SF episode starts at time t . Some of thecold gas finds its way to the center which results in the onsetof BH accretion at time t . The AGN rise time is short andlasts until t . This is followed by an intense AGN phase until t . The diminishing supply of cold gas to the SF regions, andthe central BH, cause a period where the two fade in paralleluntil t . The two processes may terminate together or, al-ternatively, the AGN fading may last a little longer. In bothcases, L SF and L bol are much smaller at large t comparedwith their peak values. This schematic evolution resemblesthe S07 suggestion of a smooth sequence that begins withSF galaxies without AGN and extends at its massive endto AGN hosts with different levels of SF. In this scenario,weak AGN fall are associated with lower SFR relative tothe peak activity of both. The strong AGN phase has a tailthat extends into the domain of quiescent galaxies. AGN inearly type galaxies that do not show any SF activity arenot shown. The diagram shows two such scenarios with thesame SFR. One for a high luminosity AGN (top solid line,large L bol / L SF ) and one with a weaker AGN (dotted line,smaller L bol /Lsf).The above scenario is translated to an L SF vs. L bol curve in the right side of the diagram. Here the straightline represents the correlation of Fig. 13. Two pure SF-dominated systems with different supply of gas are shownas two rising and horizontal lines during times t − t . Thefading parts, t − t , representing the decreasing branchesof both L SF and L bol , are shown as lines going down par-allel to the main correlation. The regions t − t on bothcurves are obviously more complex. They may involve con-stant SF and BH accretion rates (i.e. one point on the curve)or periods where both L SF and L bol increase along the maincorrelation, perhaps up to a point where L/L Edd ∼ 1. In thisscenario, there are no missing objects below the correlation Figure 14. A schematic SF-AGN evolution sequence. The leftpart shows the relevant times for the SF (thick line) and AGN(thin solid line - luminous AGN, dashed line - low luminosityAGN) evolution. The time t − t precedes the onset of the AGNaccretion. The peak (Seyfert) AGN activity occurs between t and t and the long decay, up to t , is the LINER, low accretionrate phase. The right part translates this scenario to an L SF vs. L bol relationship illustrating two tracks for law luminosity(bottom dashed line) and high luminosity (top solid line) events(time flags are only marked for the latter). The diagonal line onthe right represents the observed relationship from Fig. 13. line in Fig. 13 unless the AGN accretion continues into thequiescent galaxy part with no SF activity. The left upperpart of the diagram will be filled by SF galaxies that are notincluded in Fig. 13. These are found in the SDSS sample andin several, high redshift LIG, ULIRG and SMG samples.The correlation in Fig. 13 can be translated to a ra-tio between the bulge growth rate, g ( bulge ), assumed to beproportional to L SF , and BH growth rate, g ( BH ), assumedto be proportional to L bol . For BH radiation conversion ef-ficiency of η BH , and SF radiation conversion efficiency of η SF , the ratio is g ( bulge ) g ( BH ) ≃ (cid:20) η BH / . η SF / × − (cid:21) (cid:20) L Bol erg s − (cid:21) − . . (4)In bulge dominated systems, the time integrated ratio ofthese quantities must equal the local measured value of M ∗ /M BH . This number is larger by a factor of at leastsix compared with the value inferred from eqn. 4. It sug-gests that even the slowest accreting BHs in AGN domi-nated sources, those at the bottom left part of Fig. 13, withSFR ∼ . M ⊙ /yr, are still growing at a relative rate whichis about six times faster than their integrated cosmic growthrate. For the fastest accreting BHs in type-II AGN the ratiois larger than 20. The constant (115) in the above equationis similar to the one found by Silverman et al. (2009) in theiranalysis of the BH growth rate and SFR in zCOSMOS galax-ies (see their Fig. 13). These numbers can be translated, ina simplistic way, to the ratio of duty cycles between BHand SF activity. It it difficult to assign similar numbers tothe high redshift population since M ∗ /M BH is not known athigh redshift.Heckman et al. (2004) studied the M ∗ /M BH ratio inSDSS sources and found good agreement between the to- c (cid:13) , 1–16 Hagai Netzer tal accumulated M ∗ and M BH . Their population average isconsistent with g ( bulge ) /g ( BH ) ∼ L SF L bol and stellar ages A possible way to test the simple theoretical scenario ofFig. 14 is to use the measured D n n × Mysfor D n yrs for D n n L bol and L SF vs. D n n n L SF at various 4. 000 as given in B04. As explained, L SF inhosts with 4. 000 > . L bol and L SF in each 4. 000 group.The dependence of both L bol and L SF on the age ofthe stellar population age has been known for some time(K03, S07, KH09). However, the great similarity in the shapeof the two is a new feature. It indicates, yet again, a pro-portionality of L bol and L SF over a large range of stellarpopulations. This is another manifestation of the L SF - L bol correlation shown in Fig. 13. The general behaviour resem-bles the schematic scenario of Fig. 14 since D n n . L/L Edd and the SSFR with D n L/L Edd (i.e. multiplying M ∗ by the Eddington lu-minosity of a solar mass BH). The overall shape of the uppercurve is similar to the previous correlation. It shows a slightdependence of L/L Edd on M BH (the smaller BHs are thefastest accretors). The lower curve is only shown for com-parison since its shape simply reflect the SSFR-D n n D n L S F , L B o l ( e r g s − L Bol BH < 7.8 L SF S2+L2 (Ke06)L Bol L SF S2+L2 (Ka03) Figure 15. L bol (black symbols) and L SF (blue symbols) vs.D n . < log M BH < . M ⊙ z < . L SF . −7 −6 −5 .0001.001.01.11 D n SS F R L / L E dd Ka03 AGN z<0.17.5 < log M BH < 7.87.0 < log M BH < 7.38.0 < log M BH < 8.3 Figure 16. L/L Edd and SSFR (normalized to the same units as L/L Edd ) vs. D n z < . observed range of L SF and L/L Edd in a given L bol bin. Thefirst may reflect the combination of early and late type stel-lar populations for the same luminosity AGN and the sec-ond the combination of BHs with different accretion rates,all having the same L bol . This is a natural consequence ofthe schematic evolution between times t and t shown inFig. 14. A critical evaluation of the various ways used to estimate L bol in type-II AGN suggests that some of the previousmethods, in particular the one based on L([O iii ] λ c (cid:13) , 1–16 ccretion and star formation rates in type-II AGN estimates L/L Edd by factors of ∼ 5. The paper suggests sev-eral alternative methods for estimating L bol and use them toinvestigate the properties of type-I and type-II low redshiftAGN from the SDSS sample. The main conclusions are:1. The best method for estimating L bol in type-IIAGN is based on a combination of L([O iii ] λ i ] λ L/L Edd distributions of S2s resemble thoseof similar luminosity type-I sources. The comparison be-tween the two types is based on different ways of measuring M BH and care must be taken in using the σ ∗ method inblue galaxies, especially disk dominated and pseudobulgegalaxies. Most type-II AGN are L2s with very small val-ues of L/L Edd and there are very few, if any, such sourcesin type-I samples. This results in a much narrower L/L Edd distribution for type-I AGN.3. S2s and L2s form a continuous sequence of L/L Edd withno indication for a change in the mode of mass supply. Theoverall range in L/L Edd for the entire population is aboutfive orders of magnitude while for S2s it is only three ordersof magnitude.4. There is a clear and strong correlation between L SF and L bol over more than five orders of magnitude in luminosityin AGN-dominated systems. The low luminosity, low red-shift type-II sources fall on the same correlation as the moreluminous, high redshift, type-I sources. There is no distinc-tion between S2s and L2s all the way down to a SFR of 0.1 M ⊙ /yr and both follow the same L SF - L bol correlation.5. L bol , L SF , L/L Edd and the SSFR all follow, in a similarway, the D n n L bol - L SF correlation, thismay give a clue to the sequence of events that leads fromSF to AGN activity in individual sources. ACKNOWLEDGMENTS I am grateful to Benny Trakhtenbrot for help in using theDR4 archive and for useful discussions. Dan Maoz providedimportant insights about LINERs and their properties. Ihave benefited from discussions and exchange of informa-tion with G. Kauffmann and T. Heckman. Useful discussionswith Kristen Shapiro, Amiel Sternberg, John Kormendy,Niv Drory, Karl Gebhardt and Yan-Mei Chen are gratefullyacknowledged. I thank MPE and his director, R. Genzel,for their hospitality during several visits to their institute.Funding for this work has been provided by the Israel Sci-ence Foundation grant 364/07 and by the Jack Adler Chairfor Extragalactic Astronomy. REFERENCES Adelman-McCarthy, et al., 2006, ApJS, 162, 38.Baldry, I. K., Glazebrook, K., Brinkmann, J., Ivezi´c, ˇZ.,Lupton, R. H., Nichol, R. C., & Szalay, A. S. 2004, ApJ,600, 681Baldwin, J., Phillip, M, and Terlevich, R, 1981, PASP93,5 (BPT)Baldwin, J., Ferland, G., Korista, K., & Verner, D. 1995,ApJL, 455, L119 Baskin, A., & Laor, A. 2005, MNRAS, 358, 1043Bassani, L, Dadina, M., Maiolino, R. et al., 1999, ApJS,121, 473Bentz, M. C., Peterson, B. M., Netzer, H., Pogge, R. W.,& Vestergaard, M. 2009, ApJ, 697, 160Brinchmann, J., Charlot, S., White, S. D. M., Tremonti,C., Kauffmann, G., Heckman, T., & Brinkman, J. 2004,MNRAS, 351, 1151 (B04)Charlot, S., & Fall, S. M. 2000, ApJ, 539, 718Dasyra, K. M., et al. 2008, ApJL, 674, L9Drory, N., & Fisher, D.,B., 2007, ApJ, 664, 640Ferguson, J. W., Korista, K. T., Baldwin, J. A., & Ferland,G. J. 1997, ApJ, 487, 122Ferland, G. J., & Netzer, H. 1983, ApJ, 264, 105Groves, B. A., Dopita, M. A., & Sutherland, R. S. 2004,ApJS, 153, 75Groves, B., Kewley, L., Kauffmann, G., & Heckman, T.2006a, New Astronomy Review, 50, 743Groves, B. A., Heckman, T. M., & Kauffmann, G. 2006b,MNRAS, 371, 1559Heckman, T. M., Kauffmann, G., Brinchmann, J., Charlot,S., Tremonti, C., & White, S. D. M. 2004, ApJ, 613, 109Ho, L.i C., Filippenko, A. V., & Sargent, W. L., 1997,ApJS, 112, 315Ho, L. C. 2008, ARA&A, 46, 475Kaspi, S., & Netzer, H. 1999, ApJ, 524, 71Kaspi, S., Smith, P. S., Netzer, H., Maoz, D., Jannuzi,B. T., & Giveon, U. 2000, ApJ, 533, 631 (K00)Kaspi, S., Maoz, D., Netzer, H., Peterson, B. M., Vester-gaard, M., & Jannuzi, B. T. 2005, ApJ, 629, 61 (K05)Kauffmann, G., et al. 2003, MNRAS, 346, 1055 (K03)Kauffmann, G., et al. 2003b, MNRAS, 341, 54Kauffmann, G, Heckman, T, 2009 (accepted by MNRAS;arXiv:0812.1224) (KH09)Kewley, L. J., Dopita, M. A., Sutherland, R. S., Heisler,C. A., & Trevena, J. 2001, ApJ, 556, 121Kewley, et al, 2006, MNRAS, 372, 961 (K06)Korista, K., Baldwin, J., Ferland, G., & Verner, D. 1997,ApJS, 108, 401Maoz, D. 2007, MNRAS, 377, 1696Marconi, A., Risaliti, G., Gilli, R.,Veilleux, S., & Osterbrock, D. E. 1987, NASA ConferencePublication, 2466, 737Hunt, L. K., Maiolino, R., & Salvati,M. 2004, MNRAS, 351, 169Marconi, A., Axon, D. J., Maiolino, R., Nagao, T., Pas-torini, G., Pietrini, P., Robinson, A., & Torricelli, G. 2008,ApJ, 678, 693,Veilleux, S., & Osterbrock, D. E. 1987, NASA ConferencePublication, 2466, 737Marconi, A., Axon, D., Maiolino, R., Nagao, T., Pietrini,P., Robinson, A., & Torricelli, G. 2009, arXiv:0809.0390Mel´endez, M., et al. 2008, ApJ, 682, 94Netzer, H. 1990, Active Galactic Nuclei, 57Netzer, H., & Laor, A. 1993, ApJL, 404, L51Netzer, H. 2009, ApJ, 695, 793 (N09)Netzer, H., & Trakhtenbrot, B. 2007, ApJ, 654, 754 (NT07)Netzer, H., Mainieri, V., Rosati, P., & Trakhtenbrot, B.2006, A&A, 453, 525Paturel, G., Petit, C., Prugniel, P., Theureau, G.,Rousseau, J., Brouty, M., Dubois, P., & Cambr´esy, L. 2003,A&A, 412, 45Peterson, B. M., & Bentz, M. C. 2006, New Astronomy c (cid:13) , 1–16 Hagai Netzer Review, 50, 796Reyes, R., et al. 2008, AJ, 136, 2373Salim, S., et al., 2007, ApJS, 173, 267 (S07)Sanders, D. B., & Mirabel, I. F. 1996, ARA&A, 34, 749Shen, Y., Greene, J. E., Strauss, M. A., Richards, G. T., &Schneider, D. P. 2008, ApJ, 680, 169Tremaine, S., et al. 2002, ApJ, 574, 740Silverman, J. D., et al. 2009, ApJ, 696, 396Vasudevan, R. V., & Fabian, A. C. 2007, MNRAS, 381,1235Veilleux, S., & Osterbrock, D. E. 1987, NASA ConferencePublication, 2466, 737Vestergaard, M., & Peterson, B. M. 2006, ApJ, 641, 689York, et al., 2000, AJ, 120, 1579This paper has been typeset from a TEX/ LaTEX file preparedby the author. c (cid:13)000