Triggering optical AGN: the need for cold gas, and the indirect roles of galaxy environment and interactions
aa r X i v : . [ a s t r o - ph . GA ] N ov Mon. Not. R. Astron. Soc. , 1–8 (2014) Printed 28 August 2018 (MN L A TEX style file v2.2)
Triggering optical AGN: the need for cold gas, and theindirect roles of galaxy environment and interactions
J. Sabater ⋆ , P. N. Best and T. M. Heckman Institute for Astronomy (IfA), University of Edinburgh, Royal Observatory, Blackford Hill, EH9 3HJ Edinburgh, U.K. Center for Astrophysical Sciences, Department of Physics & Astronomy, The Johns Hopkins University, Baltimore, MD 21218, USA
Accepted XXXX Month XX. Received XXXX Month XX; in original form XXXX Month XX
ABSTRACT
We present a study of the prevalence and luminosity of Active Galactic Nuclei (AGN;traced by optical spectra) as a function of both environment and galaxy interactions.For this study we used a sample of more than 250000 galaxies drawn from the SloanDigital Sky Survey and, crucially, we controlled for the effect of both stellar mass andcentral star formation activity. Once these two factors are taken into account, theeffect of the local density of galaxies and of one-on-one interactions is minimal in boththe prevalence of AGN activity and AGN luminosity. This suggests that the level ofnuclear activity depends primarily on the availability of cold gas in the nuclear regionsof galaxies and that secular processes can drive the AGN activity in the majority ofcases. Large scale environment and galaxy interactions only affect AGN activity in anindirect manner, by influencing the central gas supply.
Key words: galaxies: evolution – galaxies: interaction – galaxies: active – radiocontinuum: galaxies – surveys
Active Galactic Nuclei (AGN) are closely linked to galaxyevolution and may play a fundamental role in the feedbackprocesses that both quench the growth of massive galax-ies (see reviews by Cattaneo et al. 2009; Heckman & Best2014, and references therein) and establish the strong corre-lations between a galaxy’s black hole mass and the velocitydispersion of its stellar bulge (e.g. Marconi & Hunt 2003;Ferrarese & Merritt 2000; Gebhardt et al. 2000). The pres-ence and characteristics of an AGN are tightly related tothose of their parent galaxies, and nuclear activity has alsobeen found to be linked to galaxy environment in severalstudies. Nuclear star formation (SF) activity is enhanced byinteractions (see Li et al. 2008a, and references therein) asis expected from numerical simulations, but the dependenceof AGN activity on environment is not so clear. Many ap-parently contradictory results are still found in the literature(see discussion in Sabater et al. 2013, hereafter SBA13).An important point to consider is that the word “envi-ronment” is used to refer to at least two relevant but distinctaspects of galaxy environment: (a) the large scale environ-ment, which effectively controls the gas supply to galaxiesand, (b) one-on-one interactions, involving strong tidal in-teractions between companion galaxies. The relation of theprevalence of AGN with these two aspects can be differ- ⋆ E-mail: [email protected] (JS); [email protected] (PNB) ent or even opposite. For example, while the prevalenceof radiatively efficient AGN decreases toward higher localdensities of galaxies (Carter et al. 2001; Miller et al. 2003;Kauffmann et al. 2004), it is enhanced when one-on-one in-teractions are stronger (Petrosian 1982; Koulouridis et al.2006; Alonso et al. 2007; Rogers et al. 2009; Ellison et al.2011; Liu et al. 2012; Hwang et al. 2012). Hence, those dif-ferent aspects of the environment should be considered sep-arately (SBA13).One of the most (if not the most) important factors thataffects the triggering of an AGN is the galaxy mass. Theprevalence of AGN depend strongly on the stellar (or blackhole) mass of the host galaxy (e.g. Kauffmann et al. 2003;Best et al. 2005; Silverman et al. 2009; Tasse et al. 2011).The relation of the galaxy mass with the fraction of AGNis so strong that it will seriously bias any studies that donot take it into account properly. In SBA13, galaxy masswas accounted for using stratified statistical methods, andit was found that (at fixed mass) the prevalence of opticalAGN is a factor 2–3 lower in the densest environments, butincreases by a factor of ∼ c (cid:13) J. Sabater, P. N. Best and T. M. Heckman
Li et al. (2008b) found that, if the central star formationand the AGN prevalence are considered together, there isno enhancement of the nuclear activity in galaxies with closecompanions. Similarly, Reichard et al. (2009) found that thelopsidedness of a galaxy is related to an enhanced activitylevel of the central black hole but, if the age of the cen-tral stellar population is matched in the comparison, thisenhancement is no longer visible. These findings would berelated to the suggestion of Park & Choi (2009) that galaxymass and morphology are the main factors determining therest of the properties of galaxies. All of this may indicatethat the observed environmental dependence of both AGNand central star formation arises from the same underlyingmechanisms and that the environment affects the AGN ac-tivity only by affecting the gas supply. If this is correct, thetrends of AGN fraction with large scale environment shouldalso disappear when the central star formation rate is con-trolled together with galaxy mass. The goal of this study isto extend the analysis of SBA13 to test this hypothesis, andalso to look independently at interactions.AGN come in two flavours depending on their feed-ing mechanism (see Heckman & Best 2014, and referencestherein): (a) quasar or radiative mode, believed to be fu-elled by cold gas, observed as X-ray AGN, optical AGN andhigh excitation radio galaxies, and, (b) radiatively inefficientor jet-mode, probably fuelled by gas cooling from hot halos,and observed primarily as low excitation radio galaxies. Therelation of these two types of AGN with the environment andinteractions can be different or even opposite, as shown inSBA13. We will focus in this study on radiatively efficientAGN which are supposed to be fuelled by the cold gas thatcan also trigger the central star formation that we aim tocontrol for.We aim to study the effect of different aspects of theenvironment on radiatively efficient nuclear activity aftertaking into account the effect of both the mass and the starformation activity. We characterize the environment for asample of ≈ m = 0 .
3, Ω Λ = 0 . H = 70 km s − Mpc − . We use the sample presented in SBA13. This sample wasbased on the seventh data release of the Sloan Digital SkySurvey (SDSS DR7; Abazajian et al. 2009). It is composedof galaxies in the main spectroscopic sample with magni-tudes between 14.5 and 17.77 in r-band (Strauss et al. 2002)and with redshift between 0.03 and 0.1. The final number ofgalaxies in the sample is 267973. There were duplicated data for 4 galaxies in the catalogue pre-sented on SBA13. The indices of these galaxies are: 51930-285-80; 51999-286-559; 52173-644-540 and 52468-717-223 (mjd-plate-fiberid). The duplicate entries were removed from the catalogue.
We will use two of the environmental parameters de-rived in SBA13. In that study three environmental param-eters where considered, to trace different aspects of the en-vironment and interaction: (a) a local galaxy density pa-rameter (hereafter “density”), (b) a tidal forces estima-tor, and (c) a cluster richness estimator (from Tago et al.2010). The density parameter is derived from the densityof galaxies up to the 10 th nearest neighbour; defined aslog(10 / Vol( r th )), where r th is the projected distance inMpc to the 10 th companion. The tidal estimator tracesthe relation between the tidal forces exerted by compan-ions and the internal binding forces of the galaxy; definedas log( P i [( L r /L r i ) × (2 R/d i ) ]), where L r is the luminos-ity in r-band of the galaxy, L r i the luminosity in r-band ofthe companion, R the radius of the galaxy and d i the dis-tance between the galaxy and the companion. A PrincipalComponent Analysis (PCA) was applied to consider and re-move the possible correlations between the environmentalparameters. We found that the local density of companionsaround the target galaxy (a measure of the larger-scale en-vironment in which the galaxy lives) is one of the main en-vironmental driving factors for AGN activity and was welltraced by the density parameter (also largely equivalent tothe PCA1 component). We found that one-on-one interac-tions are also important driving factors, and are best tracedby the the PCA component denominated PCA2 (defined as0 . × tidal − . × density). PCA2 seems to take intoaccount and corrects, at least partially, the possible 2D pro-jection effects that would affect a pure tidal estimator indense environments. Hence, we selected density and PCA2for this study.We also used the total stellar mass of the galaxy andthe specific star formation rate measured in the central areacovered by the spectrograph fibre (Kauffmann et al. 2003;Brinchmann et al. 2004; hereafter we will use sSFR to re-fer to this central measurement of the specific star forma-tion rate: sSFR = nuclear star formation divided by totalstellar mass). We will also use the optical AGN classifica-tion. A galaxy is considered to harbour an optical AGN if itis classified as a Seyfert, LINER or transition object usingthe standard emission line ratio diagnostic diagrams (e.g.Kewley et al. 2006) and the luminosity of its [O iii ] emissionline is higher than 10 . L ⊙ . This limit selects only brightAGN but avoids the possible classification biases arisingfrom the different redshifts of the galaxies of the sample(see SBA13). The number of AGN galaxies classified as eachtype using the former criteria are the following: 3347 Seyfert,644 LINER, 1704 transition objects. The mean L [O iii ] of theAGN is 6 . ± .
34. Note that because of the high L [O iii ] limit, the vast majority of the selected AGN are Seyferts,so this will be a study of radiatively efficient AGN, even ifLINERs are classified as jet-mode AGN (Heckman & Best2014).There is a chance for relatively faint AGN to be mis-classified as SF nuclei if their emission is concealed by thestrong emission of a powerful star forming host. The high L [O iii ] limit helps to minimise the misclassification of AGNwithin galaxies with high star formation rates. Furthermore,for galaxies of a given mass and SFR, there is no reasonto expect that any misclassification should be a functionof environment. To check this, we examined AGN-tracers(e.g. L [O iii ] ) in the SF-classified galaxies in both high and c (cid:13) , 1–8 riggering optical AGN Figure 1.
Distribution of mass and sSFR of the whole sample.The separation lines of sSFR = 10 − yr − and M = 10 . M ⊙ are shown as dashed lines. The grid marks the bins used for someof the stratified statistical studies of Section 3 (see text). low density (or PCA2) environments and found no evidenceof any differences. We also confirmed that our results wereunchanged (within the errors) if transition objects (a limitcase mix of SF and AGN) were excluded from the analysis.Therefore, this effect should not affect the results.The distribution of the mass and sSFR is shown inFig. 1. The mass of the galaxies of the sample ranges froma minimum of 10 . M ⊙ to a maximum of 10 . M ⊙ witha mean of 10 . M ⊙ . The sSFR follows the well-establisheddistribution (e.g. Strateva et al. 2001) with a star formingpopulation (sSFR ∼ > − yr − ) and a passive population(sSFR ∼ < − yr − ) and 99 per cent of the galaxies withvalues between 10 − . yr − and 10 − . yr − . We will usea value of sSFR= 10 − yr − to separate the high and lowsSFR populations when needed. An additional separation inmass of M = 10 . M ⊙ will be used as well when required.The relation of the environmental parameters, densityand PCA2, with respect to the mass and the sSFR is shownin Fig. 2. The correlations are subtle but may be strongenough to bias the study if not taken into account at a laterstage. The two sSFR populations are clearly visible on theupper panels. The high density end is populated by massivegalaxies (visible on the lower-left panel) and these galaxiesare mainly low sSFR galaxies (upper-left panel). There isa weak trend for low sSFR galaxies to be located at lowerlevels of PCA2 than high sSFR galaxies (upper-right panel).Finally, for some computations we used the mass ofthe black hole, which was derived using the relation ofMcConnell & Ma (2013) as explained in Heckman & Best(2014). We also considered the bolometric luminosity of theAGN to be L bol = 3 . × L [O iii ] , where L [O iii ] is not cor-rected for dust extinction. We investigate the relation of the prevalence of optical AGNwith the environmental parameters, density and PCA2, tak- ing into account the effect of the mass and the sSFR. To dothat, we compute the odds ratio of a galaxy harbouring anAGN at values of the environmental parameter above or be-low its median value ( − .
603 for the density and − .
043 forPCA2; see Fig. 2). The masses are divided in strata of 0.25[M ⊙ ] from log( M ) = 8 .
50 [M ⊙ ] to log( M ) = 11 .
75 [M ⊙ ],and the sSFR are divided in strata of 0.5 [yr − ] fromlog(sSFR) = − . − ] to log(sSFR) = − . − ]. Thegrid composed by the different strata is shown in Fig. 1. Ineach of these bins, a Fisher’s exact test is performed. Weobtain the odds ratio (the strength of the relation betweenharbouring an AGN and a higher value of the environmentalparameter; if this is ≈ p -value that indicates whether thetrend found is significant.The results of the Fisher’s test are shown in Fig. 3. Forthe density parameter, the odds ratios are usually close to 1and the p -values are, in general, higher than 0.05 indicatingthat the possible trends are not significant. There is only onebin in the high sSFR and high mass region where the trendmay be significant ( p .
01) but the value of the odds ratiois not far from 1 in this case. We therefore find no significanttrend of the prevalence of AGN with respect to the density.In the case of PCA2, we find a significant positive trend forone bin in the low sSFR and high mass region and a clearlysignificant negative trend for 4 bins in the high sSFR andlow mass region.It should be noted that the size of the bins may af-fect the statistics of the test. The bins need to be small toavoid biases due to strong trends of the fraction of AGNwith mass, but the smaller the bins the lower the statis-tical significance of the test. Therefore, we will use a sta-tistical method that aggregates the information of the binsto gain in statistical significance but that still accounts forthe different strata used: the Cochran-Mantel-Haenzsel test(CMH; Cochran 1954; Mantel & Haenszel 1959). It givesthe strength of association between two bi-valued variables(odds ratio) and its statistical significance after taking intoaccount the effects of the possible confounding factors de-fined as the strata.It is also clear that the whole sample may not homoge-neously follow a trend; that appears to be the case for PCA2(right panel of Fig. 3). A Woolf’s test of homogeneity ofodds ratios among strata (Woolf 1955) was applied to checkwhether the trends were homogeneous enough within thesample. A low p -value for this test means that the trends aredifferent for different parts of the parameter space. Hence,we need to separate the analysis in different regions to obtainmeaningful values of the CMH test. For density, we found noevidence of any heterogeneity. In the case of PCA2, we foundsigns of statistically significant heterogeneity using this testfor the whole sample and also when considering only thehigh sSFR sub-sample. However, when splitting into foursub-samples, by mass and sSFR, no evidence of heterogene-ity was found in PCA2 for any sub-sample.We applied the CMH test to check for the significanceof the relation between the presence of an optical AGN andvalues of the environmental parameters above or below theirmedian. The confounding factors considered are again sSFRand mass. The test was applied to the complete sample, tothe high and low sSFR sub-samples and to the four sub- c (cid:13) , 1–8 J. Sabater, P. N. Best and T. M. Heckman
Figure 2.
Distribution of mass, sSFR, density and PCA2 for the whole sample. The medians of the density and PCA2 are marked withdotted lines. The separation lines of sSFR = 10 − yr − and M = 10 . M ⊙ are shown as dashed lines. Figure 3.
Odds ratios of the prevalence of AGN with respect to the level of density and one-on-one interactions (PCA2). The binscorrespond to those shown in Fig. 1. An odds ratio lower than one indicates a negative relation between the probability of harbouring anAGN and a higher value of the environmental parameter; a value higher than one indicates a positive relation. Note that the statisticalsignificance must be taken into account in order to interpret these relations. The bins for which the p -value is higher than 0.05 are markedwith a cross and those in which the p -value is between 0.05 and 0.01 are marked with a plus sign. Only in the one (left panel) and six(right panel) bins with no symbol is the odds ratio significantly ( p .
01) at variance with unity. samples obtained with the sSFR = 10 − yr − and M =10 . M ⊙ separation lines. We checked that the results werenot affected by the sizes of the bins. The results are the sameusing from 4 to 128 bins per sample or sub-sample.The results of the CMH and the Woolf’s tests are shownin Table 1 and Fig. 4. The values of the ratios for the densityare in general compatible with the absence of a clear trend.Even in the two cases where the deviation of the ratios fromunity can be considered statistically significant, their valuesare very close to 1 implying changes in the probability oftriggering an AGN of less than 10 per cent. In the case of PCA2 the result for the whole sample is compatible with1 but with heterogeneity. When the sample is separated bysSFR an opposite significant trend appears in the two sSFRregimes; galaxies with low sSFR and higher values of PCA2harbour more AGN (about 20 per cent more when the valueof PCA2 is above the mean). On the other hand, galaxieswith high sSFR, especially galaxies with low mass and highsSFR seem to harbour fewer AGN in this case. The AGNprevalence is about 30 per cent higher at low than at highPCA2 values for high sSFR, low mass galaxies.We checked the activity level of the AGN with respect c (cid:13) , 1–8 riggering optical AGN A ll H i g h s S F R L o w s S F R H i g h s S F R L o w M a ss H i g h s S F R H i g h M a ss L o w s S F R L o w M a ss L o w s S F R H i g h M a ss C M H r a t i o Density A ll H i g h s S F R L o w s S F R H i g h s S F R L o w M a ss H i g h s S F R H i g h M a ss L o w s S F R L o w M a ss L o w s S F R H i g h M a ss C M H r a t i o PCA2 (one-on-one interactions)
Figure 4.
CMH ratios for the prevalence of optical AGN with respect to the density and one-on-one interactions (PCA2). The error barsmark the 95 per cent confidence interval. The shape of the symbol indicates if the trend found is statistically significant and if the dataof the sample tested is homogeneous enough for the CMH ratio to be relevant (see text). Statistically significant trends ( p .
05) aremarked with solid symbols (triangles or circles). A value of the Woolf test below 0.05, indicating that the odds ratios are heterogeneousamong the strata, is marked as a plus sign (instead of a cross) or a triangle (instead of a circle).
Table 1.
Statistical study results from the CMH test. For eachcell of the table the first row shows the CMH common odds ra-tio and its 95 per cent confidence interval and the second rowshows the statistical significance or p -value (i.e., the probabilityof the trend occurring by chance) measured by the CMH test. Thehypothesis tested was that the observed nuclear activity type isindependent of the density or PCA2 parameter. The typeface ofthe p -value depends on its value: bold if p < .
01; bold italics if0 . p < .
05; and italics if p > .
05. An asterisk marks thetwo results where the data is deemed as heterogeneous using theWoolf test. Density PCA2All 0 . ± .
049 0 . ± .
051 *
High sSFR 0 . ± .
055 0 . ± .
054 *
Low sSFR 0 . ± .
10 1 . ± . High sSFR & Low Mass 1 . ± .
094 0 . ± . High sSFR & High Mass 0 . ± .
066 1 . ± . Low sSFR & Low Mass 0 . ± .
22 1 . ± . Low sSFR & High Mass 0 . ± .
11 1 . ± . to the environmental parameters. We computed the medianof the logarithm of the [O iii ] luminosity per unit black holemass, which is a measure of the Eddington-scaled accretionrate ( L bol /L Edd ∼ . L [O iii ] /M BH ), at different levels of theenvironmental parameters in the four different sub-samples.The results are shown in Fig. 5. The value of the medianlog( L [O iii ] /M BH ) does not show a significant trend with thedensity or PCA2 in any of the sub-samples and is, in general,compatible with the median value for each sub-sample. Onthe other hand, the median value can be seen to dependstrongly on both the mass and sSFR, which are used to define the sub-samples, emphasising the need to account forthese parameters.Finally, we checked if any possible contamination by jetmode AGN was affecting our results. We applied a cut-off onthe accretion rate of the AGN selecting only galaxies with L bol /L Edd > .
01 and repeated the study. More than 80 percent of the galaxies were selected and the results were thesame.
The test of the prevalence AGN with respect to the densityis fully compatible with the hypothesised scenario in whichthe galaxy density does not play any role after the galaxymass and the central sSFR are taken into account. The p -value is 2 per cent and the CMH ratio is very close to 1. Ifthe sample is separated by sSFR the p -values are above 5 percent. Given the size of the sample and the magnitude of theCMH ratio, we can consider that there is not any significanttrend with respect to the density of galaxies. We showed inSBA13 how the trend is significant if only the mass is consid-ered. Hence, the effect of the density and the central sSFRare strongly correlated. Additionally, we found the lack of aclear trend for the median of log( L [O iii ] /M BH ), which tracesthe activity level of the AGN, with respect to density. Themain variation of the activity level is driven by the galaxymass and the central sSFR. All of this is in agreement withthe picture arising from the results of Li et al. (2008b) andReichard et al. (2009) that the AGN activity depends on thecold gas supply to the nucleus and not on how it gets there.In higher galaxy density environments the gas supply to thegalaxy as a whole is reduced, and reduced gas supply to thecentre indirectly reduces AGN activity.Our own study on the effect of one-on-one interactionsis also broadly consistent with this picture. We find thatthe activity level of the AGN (as judged from the medianof log( L [O iii ] /M BH )) is independent of the PCA2 param-eter, once mass and sSFR are accounted for. This resultis in agreement with that of Reichard et al. (2009), who c (cid:13) , 1–8 J. Sabater, P. N. Best and T. M. Heckman -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0density1.51.00.50.00.5 M e d i a n l og ( L [ O III ] / M B H ) -1.75 -1.25 -0.75 -0.25 0.25 0.75 1.25 1.75PCA2 (one-on-one interactions)1.51.00.50.00.5 M e d i a n l og ( L [ O III ] / M B H ) High sSFRLow MassHigh sSFRHigh MassLow sSFRLow MassLow sSFRHigh Mass
Figure 5.
Activity level of the AGN traced by the median of the log( L [O iii ] /M BH ) with respect to the density and one-on-one interactions(PCA2) for each of the sub-samples. The error bars mark the 95 per cent confidence interval. The dashed lines mark the median for eachsub-sample. found that the activity level depends clearly on the stel-lar age (traced by D4000 and correlated with the sSFR;Brinchmann et al. 2004) but with little dependence on thelopsidedness of the galaxy (their Fig. 14). Our CMH testfor the sample as a whole is also consistent with the lack ofany trend between AGN prevalence and PCA2, in line withthis picture. However, if the sample is separated into sub-samples some significant trends arise at a 20 to 30 per centlevel. Although significant, they are small in comparison tosome of the high ratio values found in SBA13 when cen-tral sSFR was not accounted for. The trends found do notdirectly fit into the previous model and suggest additionalfactors may be at play.The increase found for low sSFR galaxies could be ex-plained by the expected positive relation between the trig-gering of AGN and strong interactions present in theoreti-cal models (e.g. Hopkins & Hernquist 2006). However, un-der this assumption a similar trend would also be expectedfor high sSFR galaxies and this is not the seen. A morelikely explanation is the different time-scale between thetriggering of the central star formation burst and the AGN(e.g. Li et al. 2008b; Darg et al. 2010; Wild et al. 2010).Wild et al. (2010) found that the growth of the black holeis delayed by about 250 Myr with respect to the start of thestarburst. This time delay may be sufficient to weaken thecorrelation between central SF and AGN activity in inter-acting systems (high PCA2) and cause the observed trends:in interacting galaxies, AGN activity is less likely to occurwhen the sSFR is high (when the starburst is triggered)and more likely to be observed later, when the sSFR hasdropped.In general, we find that central sSFR and galaxy densityare strongly correlated and one-on-one interactions play asecondary role on the triggering of AGN. This results arecompatible with a scenario in which the presence of coldgas in the centre of a galaxy is the principal factor requiredfor the triggering of radiatively efficient AGN. The cold gassupply to the central regions depends upon both the galaxy’slarge-scale environment and any interactions. The AGN doesnot need one-on-one interactions to be triggered and can befed via secular processes. We presented a study of the relation between the prevalenceof optical nuclear activity with respect to the environmentin a sample of SDSS galaxies. We aimed to quantify the ef-fect of different aspects of the environment on radiativelyefficient nuclear activity. The study is based on a sample of ∼ c (cid:13) , 1–8 riggering optical AGN teractions only affect AGN activity in an indirect manner,by influencing the central gas supply. ACKNOWLEDGMENTS
JS and PNB are grateful for financial support from STFC.We thank the referee for quick and helpful comments.This research made use of
Astropy , a community-developed core Python package for Astronomy(Astropy Collaboration 2013);
Ipython (P´erez & Granger2007); matplotlib (Hunter 2007); numpy (Walt et al.2011); pandas (McKinney 2010); scipy (Jones et al. 2001–)and
TOPCAT
REFERENCES
Abazajian K. N. et al., 2009, ApJS, 182, 543Alonso M. S., Lambas D. G., Tissera P., Coldwell G., 2007,MNRAS, 375, 1017Astropy Collaboration et al., 2013, A&A, 558, A33Best P. N., Kauffmann G., Heckman T. M., BrinchmannJ., Charlot S., Ivezi´c ˇZ., White S. D. M., 2005, MNRAS,362, 25Brinchmann J., Charlot S., White S. D. M., Tremonti C.,Kauffmann G., Heckman T., Brinkmann J., 2004, MN-RAS, 351, 1151Carter B. J., Fabricant D. G., Geller M. J., Kurtz M. J.,McLean B., 2001, ApJ, 559, 606Cattaneo A. et al., 2009, Nature, 460, 213Cochran W. G., 1954, Biometrics, 10, 417Darg D. W. et al., 2010, MNRAS, 401, 1552Ellison S. L., Patton D. R., Mendel J. T., Scudder J. M.,2011, MNRAS, 418, 2043Ferrarese L., Merritt D., 2000, ApJ, 539, L9Gebhardt K. et al., 2000, ApJ, 539, L13Heckman T. M., Best P. N., 2014, ARA&A, 58 Hopkins P. F., Hernquist L., 2006, ApJS, 166, 1Hunter J. D., 2007, Computing in Science & Engineering,9, 90Hwang H. S., Park C., Elbaz D., Choi Y. Y., 2012, A&A,538, A15Jones E., Oliphant T., Peterson P. et al., 2001–, SciPy:Open source scientific tools for Python. [Online; accessed2014-08-26]Kauffmann G. et al., 2003, MNRAS, 346, 1055Kauffmann G., White S. D. M., Heckman T. M., M´enardB., Brinchmann J., Charlot S., Tremonti C., BrinkmannJ., 2004, MNRAS, 353, 713Kauffmann G. et al., 2007, ApJS, 173, 357Kewley L. J., Groves B., Kauffmann G., Heckman T., 2006,MNRAS, 372, 961Koulouridis E., Plionis M., Chavushyan V., Dultzin-Hacyan D., Krongold Y., Goudis C., 2006, ApJ, 639, 37LaMassa S. M., Heckman T. M., Ptak A., Urry C. M., 2013,ApJ, 765, L33Li C., Kauffmann G., Heckman T. M., Jing Y. P., WhiteS. D. M., 2008a, MNRAS, 385, 1903Li C., Kauffmann G., Heckman T. M., White S. D. M.,Jing Y. P., 2008b, MNRAS, 385, 1915Lintott C. J. et al., 2008, MNRAS, 389, 1179Liu X., Shen Y., Strauss M. A., 2012, ApJ, 745, 94Mantel N., Haenszel W., 1959, Journal of the National Can-cer Institute, 22, 719Marconi A., Hunt L. K., 2003, ApJ, 589, L21McConnell N. J., Ma C. P., 2013, ApJ, 764, 184McKinney W., 2010, in S. van der Walt, J. Millman, eds,Proceedings of the 9th Python in Science Conference. pp.51 – 56Miller C. J., Nichol R. C., G´omez P. L., Hopkins A. M.,Bernardi M., 2003, ApJ, 597, 142Moles M., Marquez I., Perez E., 1995, ApJ, 438, 604Park C., Choi Y. Y., 2009, ApJ, 691, 1828P´erez F., Granger B. E., 2007, Computing in Science andEngineering, 9, 21Petrosian A. R., 1982, Astrofizika, 18, 548Reichard T. A., Heckman T. M., Rudnick G., BrinchmannJ., Kauffmann G., Wild V., 2009, ApJ, 691, 1005Rogers B., Ferreras I., Kaviraj S., Pasquali A., Sarzi M.,2009, MNRAS, 399, 2172Sabater J., Verdes-Montenegro L., Leon S., Best P., Sulen-tic J., 2012, A&A, 545, A15Sabater J., Best P. N., Argudo-Fern´andez M., 2013, MN-RAS, 430, 638Schawinski K., Dowlin N., Thomas D., Urry C. M., Ed-mondson E., 2010, ApJ, 714, L108Silverman J. D. et al., 2009, ApJ, 695, 171Strateva I. et al., 2001, AJ, 122, 1861Strauss M. A. et al., 2002, AJ, 124, 1810Tago E., Saar E., Tempel E., Einasto J., Einasto M., NurmiP., Hein¨am¨aki P., 2010, A&A, 514, A102Tasse C., R¨ottgering H., Best P. N., 2011, A&A, 525,A127+Taylor M. B., 2005, in P. Shopbell, M. Britton, R. Ebert,eds, Astronomical Data Analysis Software and SystemsXIV. Astronomical Society of the Pacific Conference Se-ries, Vol. 347, p. 29Walt S. v. d., Colbert S. C., Varoquaux G., 2011, Comput-ing in Science & Engineering, 13, 22 c (cid:13) , 1–8 J. Sabater, P. N. Best and T. M. Heckman
Wild V., Heckman T., Charlot S., 2010, MNRAS, 405, 933Woolf B., 1955, Annals of Human Genetics, 19, 251 c (cid:13)000