The nature of the SDSS galaxies in various classes based on morphology, colour and spectral


Mon. Not. R. Astron. Soc. 000, 1–17 (2008)

Printed 1 July 2008

A (MN L TEX style ?le v2.2)

The nature of the SDSS galaxies in various classes based on morphology, colour and spectral features – I. Optical properties

arXiv:0807.0110v1 [astro-ph] 1 Jul 2008

Joon Hyeop Lee1,2?, Myung Gyoon Lee1? , Changbom Park3?, Yun-Young Choi4? 1
Astronomy Program, Department of Physics and Astronomy, Seoul National University, Seoul 151-742, Korea Astronomy and Space Science Institute, Daejeon 305-348, Korea 3 Korea Institute for Advanced Study, Dongdaemun-gu, Seoul 103-722, Korea 4 Astrophysical Research Centre for the Structure and Evolution of the Cosmos, Sejong University, Seoul 143-747, Korea
2 Korea

Accepted 2008 June 28. Received 2008 June 04; in original form 2008 April 14


We present a comprehensive study of the nature of the SDSS galaxies divided into various classes based on their morphology, colour, and spectral features. The SDSS galaxies are classi?ed into early-type and late-type; red and blue; passive, HII, Seyfert, and LINER, returning a total of 16 ?ne classes of galaxies. We examine the luminosity dependence of seven physical parameters of galaxies in each class. We ?nd that more than half of red early-type galaxies (REGs) have star formation or AGN activity, and that these active REGs have smaller axis ratio and bluer outside compared to the passive REGs. Blue early-type galaxies (BEGs) show structural features similar to those of REGs, but their centres are bluer than REGs. HII BEGs are found to have bluer centres than passive BEGs, but HII REGs have bluer outside than passive REGs. Bulge-dominated late-type galaxies have red colours. Passive red late-types are similar to REGs in several aspects. Most blue late-type galaxies (BLGs) have forming stars, but a small fraction of BLGs do not show evidence for current star formation activity. Di?erences of other physical parameters among di?erent classes are inspected, and their implication on galaxy evolution is discussed. Key words: galaxies: general – galaxies: evolution – galaxies: statistics – galaxies: elliptical – galaxies: spiral – galaxies: active



One of the fundamental issues of the observational cosmology is the evolutionary connection between various classes of galaxies. Today, galaxies are classi?ed with various criteria. The most classical classi?cation of galaxies is the Hubble sequence: elliptical galaxies, lenticular galaxies, spiral galaxies, barred spiral galaxies, and irregular galaxies (Hubble 1936; Sandage 1961; De Vaucouleurs 1974). The main criterion of this classi?cation is the morphology of galaxies; that is, the existence, size ratio, and appearance of spiral arms, disc, bulge, and bar. The Hubble sequence was established mainly based on the galaxies without nuclear activity, and such galaxies are often called normal galaxies. On the other hand, more and more galaxies showing active nuclei with broad line emission are being found; they are often called abnormal galaxies. According to the features of


E-mail: jhl@kasi.re.kr (JHL); mglee@astrog.snu.ac.kr (MGL); cbp@kias.re.kr (CBP); yychoi@kias.re.kr (YYC)

the line emission, active galactic nuclei (AGN) host galaxies are classi?ed into several sub-classes: Seyfert 1 galaxies, Seyfert 2 galaxies, broad-line radio galaxies, narrow-line radio galaxies, low ionisation nuclear emission regions (LINERs), and so on. Luminosity is another criterion to classify galaxies. Sandage & Binggeli (1984) de?ned the galaxies with MB > ?18 as dwarf galaxies. In addition to those general classes of galaxies, there are some unusual galaxy classes with interesting properties: E+A galaxies with the spectral features of both very old stellar populations and very young stellar populations (Dressler & Gunn 1992; Yang et al. 2006, 2008); ultra-luminous infrared galaxies (ULIRGs) that are very bright in the mid- and far-infrared bands (Sanders & Mirabel 1996; Hwang et al. 2007); blue compact galaxies that show very compact morphology and high surface brightness with blue colour (Thuan et al. 1997); extremely red objects (EROs) whose optical ? infrared colour is extremely red (Elston et al. 1988); and so on. Recently, to deal with a large amount of survey data, many astronomers have classi?ed galaxies simply into red se-


J. H. Lee et al.
the distribution of photometric, spectroscopic and structural parameters for 350 nearby galaxies using the Millennium Galaxy Catalogue (Liske et al. 2003), arguing that most properties show a clear distinction between early-type galaxies and late-type galaxies. Later, Conselice (2006) carried out principal-component analyses of the properties of 22000 galaxies at z < 0.05 using the Third Reference Catalogue of Bright Galaxies (RC3; De Vaucouleurs et al. 1991), ?nding that the three parameters determining a galaxy’s physical state may be mass, star formation and interactions/mergers. These studies presented how useful to consider various parameters in galaxy classi?cation, although some important components like AGNs were not considered in their analyses. Conselice (2006) showed that multiple independent components, rather than any single component, may determine the various properties of galaxies. Therefore, to understand the nature of galaxies more comprehensively and to give stronger constraints on galaxy evolution models to explain all kinds of galaxies, it is necessary to investigate the nature of galaxies in various and ?nely-divided classes, and to ?nd out the evolutionary connections between the classes. We have been doing a comprehensive study on a set of ?ne galaxy classes in the Sloan Digital Sky Survey (SDSS; York et al. 2000), based on their morphology, colour and spectral features. In this paper, the ?rst in the series, we present the optical properties of galaxies in various ?ne classes. The outline of this paper is as follows. Section 2 shows the data set we used, and §3 describes the methods to classify the SDSS galaxies and to select volume-limited samples. We present the statistics of selected optical parameters and their luminosity dependence in §4. Based on the optical properties, we discuss the nature of galaxies in the ?ne classes in §5. Finally, the conclusions in this paper are given in §6. Throughout this paper, we adopt the cosmological parameters h = 0.7, ?Λ = 0.7, and ?M = 0.3.

quence galaxies and blue sequence galaxies according to their colour-magnitude relation (e.g. Martin et al. 2007). For a long time, the individual properties of galaxies in those various classes have been investigated, and many e?orts have been made to answer fundamental questions about galaxy evolution and connections between galaxy classes. Some examples of such fundamental questions are as follows: Why is there a conspicuous bimodality in the colour distribution of galaxies? Why does the colour bimodality not exactly agree with the morphological segregation? How do the environments a?ect the properties of galaxies? How did active galactic nuclei (AGNs) form? Is there any transition among the di?erent classes of galaxies? Recent studies using large survey data discovered several interesting aspects of galaxy evolution, providing some answers to those fundamental questions. For example, Ball et al. (2006) studied the bivariate luminosity functions with galaxy type classi?cation, and found that there is a clear morphological bimodality supporting the idea that merger and accretion are associated with bulges and discs, respectively. Bernardi et al. (2005) showed that both luminosity and colour of early-type galaxies are correlated with stellar velocity dispersion, and that velocity dispersion may be also closely correlated with the age and metal abundance of early-type galaxies. Choi et al. (2007) found that late-type galaxies show wider dispersion in several physical quantities than early-type galaxies, and that those physical quantities manifest di?erent behaviours across M? ± 1 mag. Park et al. (2007) investigated the environmental e?ects on various physical properties of galaxies, ?nding that a key constraint on galaxy formation models is the morphology-density-luminosity relation. Mateus et al. (2006) analysed the colour, 4000? break, and age in various A spectral classes of galaxies, suggesting that the median lightweighted stellar age of galaxies is directly responsible for the colour bimodality in the galaxy population. Those previous studies, however, are not yet enough to explain the origins of the various galaxy classes and the evolutionary connections between the classes. One of the major limitations is that the galaxy classi?cations in most previous studies were often limited to only one or two properties. For instance, automatically-classi?ed morphology (e.g. Park & Choi 2005; Ball et al. 2006), galaxy colour (e.g. Martin et al. 2007), or spectral line features (e.g. Mateus et al. 2006) are used frequently in recent studies based on large survey data, but a multilateral classi?cation study using all these criteria at the same time has not yet been seen. Such simpli?cations may be often very useful to understand the various and complicated phenomena in galaxy evolution. However, a simple classi?cation can not distinguish detailed aspects of galaxy evolution. For example, early-type galaxies are often regarded as red and passive galaxies in most studies, but that is not always true. Some kinds of galaxies with early-type morphology were found to have blue colours possibly originating from young stellar populations or active nuclei (Abraham et al. 1999; Im et al. 2001; Ferreras et al. 2005; Lee et al. 2006). It is di?cult to understand these kinds of unusual classes in studies with simple classi?cations. Closely related to galaxy classi?cation, a couple of studies searching for principal components of galaxy properties have been conducted recently. Ellis et al. (2005) examined



We use the SDSS Data Release 4 (DR4; Adelman-McCarthy et al. 2006)1 in this study. The DR4 imaging data cover about 6670 deg2 in the ugriz bands, and the DR4 spectroscopic data cover 4783 deg2 . The photometric and spectroscopic observations were conducted with the 2.5-m SDSS telescope at the Apache Point Observatory in New Mexico, USA. The median width of the point-spread-function in the r band photometry is 1.4′′ , and the wavelength coverage in the spectroscopy is 3800 ? 9200?. We use the photometric and structural A parameters from the SDSS pipeline (Stoughton et al. 2002) data, and the spectroscopic parameters from the Max-Planck-Institute for Astronomy catalogue (MPA catalogue; Kau?mann et al. 2003a; Tremonti et al. 2004; Gallazzi et al. 2006). In addition, we use the velocity dispersion estimated using an automated spectroscopic pipeline called idlspec2d version 5 (D. Schlegel et al. 2008, in preparation), and colour gradient, inverse concentration, and axis ratio of the SDSS DR4plus (Choi et al. 2007)


See http://www.sdss.org/dr4/.

Optical properties of SDSS galaxy classes
sample, which is one of the products of the New York University Value-Added Galaxy Catalogue (Blanton et al. 2005). After foreground extinction correction (Schlegel et al. 1998), the magnitude of each galaxy was corrected in two more aspects: the redshift e?ect (k-correction) and galaxy luminosity evolution (evolutionary correction). We used the method of Blanton et al. (2003) to conduct k-correction, and the empirical formula of Tegmark et al. (2004) to conduct evolutionary correction. Using these methods, we corrected the observed magnitude of each galaxy into the magnitude at redshift z = 0.1, where the SDSS galaxies were observed most frequently. Since the corrections are applied optionally in this paper, we denote the magnitude with k-correction as 0.1K m and the magnitude with both k-correction and evolutionary correction as 0.1KE m, if the observed magnitude is m. We use Petrosian magnitudes to represent optical brightness, while we use model magnitudes to calculate galaxy colours. The colour gradient ?0.1K (g ? i) is de?ned as the di?erence in (g ? i) colour between the region at 0.5Rpet < R < Rpet and the region at R < 0.5Rpet (negative ?(g ? i) for blue outside; Choi et al. 2007). The inverse concentration (R50/R90) and axis ratio of galaxies were estimated using ellipsoidal ?tting (Choi et al. 2007).2 The colour gradient, concentration, and axis ratio are corrected for seeing e?ects. The equivalent width of Hα emission and the 4000? break A index, Dn (4000), de?ned in Bruzual (1983), were retrieved from the MPA catalogue.


Figure 1. Segregation between early-type galaxies (dark points) and late-type galaxies (light points) in the colour vs. colourgradient space. The lines represent di?erent segregation guidelines for di?erent magnitude ranges (solid line for 14.5 < rpet < 16.0, short-dashed line for 16.0 < rpet < 16.5, long-dashed line for 16.5 < rpet < 17.0, and dotted line for 17.0 < rpet < 17.5).

3 3.1

ANALYSIS Galaxy classi?cation

In this study, we classi?ed galaxies with three criteria: morphology, colour, and spectral line features. Morphology is one of the most fundamental criteria to classify galaxies. We selected early-type galaxies with the galaxy classi?cation method using the colour versus colourgradient space (Park & Choi 2005) as shown in Fig. 1. In this classi?cation method, colours and colour gradients are the main criteria for classi?cation, and the inverse concentration cut is also applied di?erentially for di?erent magnitude ranges: R50/R90 < 0.43 for 14.5 < rpet < 16.0, R50/R90 < 0.45 for 16.0 < rpet < 16.5, R50/R90 < 0.47 for 16.5 < rpet < 17.0, and R50/R90 < 0.48 for 17.0 < rpet < 17.5. The galaxy colour is another classi?cation criterion that is frequently used in large galaxy surveys. We classi?ed the galaxies into red galaxies and blue galaxies, based on the (g ? r) colour. Since the r band is the standard band in the SDSS photometry and the photometric uncertainty in the u band is relatively large compared to that in the g band, we selected the (g ? r) colour as the index for the galaxy colour segregation. We used the method of Lee et al. (2006) to segregate blue galaxies from red galaxies. First, we divided the redshift range of 0 z 0.4 into eight bins, and derived a 0.1K (g ? r) colour histogram of the early-type galaxies with

Figure 2. The 0.1K (g ? r) colour versus redshift of early-type galaxies. The solid line is the polynomial ?t of Gaussian peaks in the 0.1K (g ? r) distribution of early-type galaxies at 8 redshift bins between z = 0 and z = 0.4. The dashed line is the (peak ?3σ) line, which is used for red/blue galaxy separation.

Rpet is the Petrosian radius, and R50 (R90) is the semimajor axis length of an ellipse containing 50 per cent (90 per cent) of the Petrosian ?ux.


J. H. Lee et al.
Table 1. Abbreviations of the 16 ?ne galaxy classes Early-type Red Blue Passive HII Seyfert LINER pREG hREG sREG lREG pBEG hBEG sBEG lBEG Late-type Red Blue pRLG hRLG sRLG lRLG pBLG hBLG sBLG lBLG

Figure 3. (a) AGN selection in the BPT diagram (Baldwin et al. 1981). We used an empirical criterion (solid line; given by Kau?mann et al. 2003b) to distinguish AGNs from star-forming galaxies. (b) Seyfert-LINER segregation in the [OIII]/Hβ vs. [OI]/Hα diagram. The solid line is the criterion to distinguish between Seyferts and LINERs, given by Kewley et al. (2006).

0.1K Mpet (r) ?20 for each redshift bin. Then, we ?tted the colour distribution with a Gaussian function in each redshift bin, and derived a guideline for colour separation by ?tting a 5th-order polynomial function to the colours corresponding to the peaks in the eight bins, as shown in Fig. 2. We used a polynomial function instead of a linear function for more accurate ?tting, although the resulting peak line seems almost linear at z < 0.3. It is noted that the peak line shows a somewhat complex curve when it extends to higher redshift (z ? 1; Lee et al. 2006). We regard the early-type galaxies that are 3σ bluer than the Gaussian peak colour as blue early-type galaxies, and the other early-type galaxies as red early-type galaxies. The same colour guideline was used to separate the red late-type galaxies and the blue late-type galaxies. In spectroscopic studies, the line features of galaxies are often used to classify galaxies. Based on the spectral line features, we classi?ed the galaxies into passive galaxies, HII galaxies, Seyfert galaxies and LINER galaxies. First, We selected AGN host galaxies using the line ?ux ratio diagram of [OIII]/Hβ versus [NII]/Hα (BPT diagram; Baldwin et al. 1981) as shown in Fig. 3a. We used an empirical criterion to segregate AGN host galaxies from starforming galaxies: [OIII]/Hβ = 0.61/([NII]/Hα ?0.05) + 1.3, given by Kau?mann et al. (2003b). AGN selection was conducted for the sample of galaxies with a signal-to-noise ratio (S/N) of 3 for Hα, Hβ, [OIII] and [NII] lines (Kau?mann et al. 2003b), but we classi?ed some galaxies as AGNs even if their S/N ratios of one or two lines are smaller than 3, in some special cases. For example, a galaxy with log([NII]/Hα) > ?0.2 and its S/N ratios of [NII] and Hα larger than 3 but with its S/N ratios of [OIII] and Hβ

smaller than 3, was classi?ed as an AGN host galaxy. These S/N criteria are too generous if we intend to select a genuine sample of AGNs, but they may be useful to reduce the contamination of HII galaxies due to AGNs. Second, among the selected AGN host galaxies, we distinguished Seyferts from LINERs in the [OIII]/Hβ versus [OI]/Hα diagram. We used an empirical guideline: [OIII]/Hβ = 1.18 [OI]/Hα +1.3, given by Kewley et al. (2006). AGN host galaxies in the LINER domain were classi?ed as LINER galaxies, and the AGN host galaxies that are not LINER galaxies were classi?ed as Seyfert galaxies. In other words, we classi?ed AGN host galaxies without the signal of [OI] emission line as Seyfert galaxies in this paper. Third, we selected as HII galaxies, non-AGN galaxies with Hα emission with S/N 3. This criterion is more generous than those in previous studies, but useful to reduce the contamination of passive galaxies due to HII galaxies. Finally, passive galaxies were selected as galaxies with no or insu?cient (S/N< 3) signal of Hα emission. Based on the three independent classi?cations of galaxies, we classi?ed the galaxies into the ?nal 16 classes: [early-type, late-type] × [red, blue] × [passive, HII, Seyfert, LINER]. Hereafter, we use the following abbreviations for the 16 galaxy classes: REG (red early-type galaxy), BEG (blue early-type galaxy), RLG (red late-type galaxy), BLG (blue late-type galaxy), and p- (passive), h- (HII), s- (Seyfert), l- (LINER), as shown in Table 1. For example, ‘pREGs’ represents ‘passive red early-type galaxies’ and ‘sBLGs’ represents ‘Seyfert blue late-type galaxies’. Fig. 4 presents atlas images and spectra of sample galaxies in the 16 ?ne classes.


Sample selection

The selection of a galaxy sample is very important, because most properties of galaxies are known to be sensitive to their luminosity and redshift. We selected three sample volumes in the luminosity versus redshift space as shown in Fig. 5. Each selected volume is a rectangle because the SDSS spectroscopy has not only a lower-brightness limit but also an upper-brightness limit for completeness. Each volume has a small redshift range and a large luminosity range, which is adequate to investigate the luminosity dependence of galaxy properties. Since our sample covers only the nearby universe (z < 0.1), the e?ect of the redshift dependence is unlikely to be signi?cant. Possible small redshift variations of galaxy properties are checked by comparing the di?erence between the three volumes. The V1 and V2 volumes in Fig. 5 are within the redshift range with reliable spectral information (0.04 < z < 0.1) suggested by Kewley et al. (2006), but the V3 volume is not, implying that the spectroscopic parame-

Optical properties of SDSS galaxy classes


Figure 4. Atlas images and spectra of sample galaxies in the 16 ?ne classes. These images were retrieved from the SDSS.

Table 2. The number of galaxies in each class and in each volume (V1) Passive HII Seyfert LINER Total (V2) Passive HII Seyfert LINER Total (V3) Passive HII Seyfert LINER Figure 5. Three volumes (V1, V2, and V3) selected in this study. The dashed lines are the limits of the complete spectroscopic sample, and the solid lines are the completeness limits corrected for the redshift e?ect (Blanton et al. 2003) and the galaxy luminosity evolution (Tegmark et al. 2004). Total REG 5109 2479 1477 4829 13894 REG 925 705 302 824 2756 REG 116 189 45 72 422 BEG 26 148 98 159 431 BEG 23 141 65 32 261 BEG 40 253 26 14 333 RLG 446 1626 1770 3569 7411 RLG 196 815 547 846 2404 RLG 66 325 78 108 577 BLG 77 9632 1472 1281 12462 BLG 54 6348 396 233 7031 BLG 45 3997 76 41 4159 Total 5658 13885 4817 9838 34198 Total 1198 8009 1310 1935 12452 Total 267 4764 225 235 5491

ters of galaxies in the V3 volume may be less reliable than those in the V1 or V2 volumes.

4 4.1

OPTICAL PROPERTIES Class composition

Table 2 lists the number of galaxies in each class and in each volume. The class composition varies with respect to the volume. The three most dominant classes are hBLGs (28.2 per cent), pREGs (14.9 per cent) and lREGs (14.1 per cent) in V1, while they are hBLGs (51.0 per cent), pREGs

(7.4 per cent) and lRLGs (6.8 per cent) in V2, and hBLGs (72.8 per cent), hRLGs (5.9 per cent) and hBEGs (4.6 per cent) in V3. Since most properties of galaxies are known to depend on their luminosity, it is necessary to study the variation of class properties with respect to their luminosity. Fig. 6 presents the class fraction versus luminosity in each volume, showing that the variation of the class fraction is continuous even between di?erent volumes. Small discontinuities are found in several classes, but those discontinuities are not signi?cant, considering the fractional uncertainty based on the Poisson error. These results indicate that the e?ect of luminosity is more important than that of redshift on the difference in the class composition between di?erent volumes. Choi et al. (2007) showed that the fraction of early-type


J. H. Lee et al.

Figure 6. Class fraction as a function of luminosity. Percentage variation of each class with respect to luminosity was derived for three volumes: V1 (open circle), V2 (open rectangle), and V3 (open triangle). Each errorbar represents the Poisson error.

galaxies decreases as luminosity decreases, which is consistent with our result. In Fig. 6, however, we found that such a trend appears to be better established between the colour classes, than between the morphology classes, among passive galaxies. In other words, passive red galaxies are dominant at the bright end, while passive blue galaxies are dominant at the faint end, on average. The class fraction distribution of HII blue galaxies is similar to that of passive blue galaxies, but the class fraction of HII red galaxies is highest 0.1K at ?20.5 Mpet (r) ?19.5, unlike that of passive red galaxies. It is interesting that the fraction of AGN host BLGs shows decrease at the faint end, unlike that of nonAGN BLGs.

Figure 7. 0.1K (u?r) colour variation with respect to 0.1KE Mpet , for each class. Each open circle shows the median value at give magnitude bin in V1, and open rectangle and open triangle do in V2 and V3, respectively. Each errorbar represents the sample inter-quartile range (SIQR) of 0.1K (u ? r) colour at given magnitude bin. The line across symbols in each panel is the linear least-square ?t.


Optical colour


Luminosity dependence of optical properties

To understand the individual characteristics of each class, we investigate their photometric, structural, and spectroscopic properties, using seven selected physical quantities. Figs. 7 – 13 show the variation of each quantity with respect to luminosity for the 16 classes, respectively. To reduce the biases in those quantities due to internal extinction in latetype galaxies (Choi et al. 2007), late-type galaxies with axis ratio smaller than 0.6 are not used in the analysis of each physical quantity, except for axis ratio itself. To reduce the e?ect of abnormal data points, we use the median statistics.

Fig. 7 shows the luminosity dependence of 0.1K (u ? r) colour for each class. Nine of 16 classes have a clear trend that the fainter galaxies are bluer. Those variations are steady and continuous even in the transition ranges between volumes, implying that there is little e?ect from the redshift di?erence between volumes. pBLGs are the only class that are clearly redder when fainter, but the deviation in such a trend is somewhat large. Table 3 lists the results of linear least-squares ?tting for Fig. 7. We note several interesting features in Fig. 7 and Table 3. First, red galaxies (except for hRLGs) and HII blue galaxies show bluer colour at fainter luminosity. The colour variation in pREGs, called often the colour-magnitude relation (CMR), was explained in many previous studies. That is, the CMR of pREGs may re?ect mainly the di?erence in the metal abundance with respect to the mass of galaxies (Kodama & Arimoto 1997; Kau?mann & Charlot 1998). The colour variation in other classes shows somewhat complicated trends, possibly a?ected by both age and metallicity e?ects. In star-forming galaxies, the fraction of young stellar population may be an important factor making the colour variation, because it is known that galaxies with small mass

Optical properties of SDSS galaxy classes
Table 3. Slopes of the linear ?ts in the 0.1KE M pet plots REG Passive HII Seyfert LINER ?0.087 ± 0.005(2.792) ?0.092 ± 0.004(2.776) ?0.104 ± 0.006(2.771) ?0.104 ± 0.006(2.792) RLG Passive HII Seyfert LINER ?0.092 ± 0.007(2.792) ?0.020 ± 0.014(2.559) ?0.043 ± 0.010(2.617) ?0.046 ± 0.010(2.697)
0.1K (u


? r) versus

BEG ?0.023 ± 0.018(2.426) ?0.218 ± 0.028(2.139) ?0.029 ± 0.023(2.275) ?0.002 ± 0.016(2.407) BLG 0.115 ± 0.067(1.967) ?0.129 ± 0.007(1.810) 0.022 ± 0.016(2.110) 0.020 ± 0.017(2.222)

The slopes of the median 0.1K (u ? r) with respect to 0.1KE Mpet , and 0.1K (u ? r) at 0.1KE Mpet (r) = ?21 within parentheses.

are generally young in the local universe (Bernardi et al. 2005; Treu et al. 2005). Second, we found that pREGs and hREGs have a similar slope in their CMR within 1σ, and that the CMR slope of AGN host REGs is marginally steeper than that of pREGs. The di?erence in the CMR slope between di?erent spectral classes of REGs is relatively small, compared to those of BEGs, RLGs and BLGs, showing that the metallicity e?ect may be dominant in REGs, irrespective of their spectral class. In other words, the e?ects of star formation or AGN activities may be small in REGs. The CMR slope in pREGs is about ?0.09, which is signi?cantly smaller than those in hBLGs (about ?0.13). This indicates that the CMR slope of age-e?ect-dominated galaxies is steeper (larger colour variation) than the CMR slope of metallicity-e?ect-dominated galaxies, con?rming the result of Choi et al. (2007). Third, we found that AGN host galaxies in non-REG classes show very small variation in their colour. Particularly, AGN host blue galaxies do not show signi?cant colour variation with respect to luminosity. Since galaxies with small mass are generally expected to have bluer colours than massive galaxies (Kodama & Arimoto 1997; Bernardi et al. 2005; Choi et al. 2007), such constant colours of AGN host blue galaxies are somewhat unusual. It is inferred that AGN activity may be responsible for the colour constancy, and that such AGN e?ects may be most prominent in blue galaxies. Fourth, hBEGs show the largest variation in colour due to the very blue faint members of the class. The very blue faint hBEGs were also identi?ed by Choi et al. (2007), which may have much younger mean stellar ages or much poorer metal abundance than bright hBEGs. Finally, we found that pBEGs and hRLGs do not show signi?cant variation in their colour. Since both the metallicity-magnitude relation and the age-magnitude relation are known to cause the slope in CMR, the CMR of pBEGs and hRLGs may not be the result of any single mechanism. In other words, the combined e?ects of age and metallicity working di?erentially with respect to luminosity are one possible origin of the CMR of pBEGs and hRLGs. Another possibility is di?erential dust extinction with respect to luminosity.

Figure 8. The same as Fig. 7, but ?(g ? i) as Y-axis.

Table 4. As Table 3 but for ?(g ? i) REG Passive HII Seyfert LINER 0.004 ± 0.002(?0.028) 0.005 ± 0.002(?0.036) 0.015 ± 0.009(?0.056) 0.012 ± 0.008(?0.060) RLG Passive HII Seyfert LINER 0.018 ± 0.008(?0.063) 0.037 ± 0.006(?0.118) 0.033 ± 0.007(?0.152) 0.037 ± 0.005(?0.158) BEG 0.000 ± 0.007(0.011) 0.050 ± 0.005(0.071) 0.020 ± 0.011(0.036) 0.025 ± 0.014(0.026) BLG 0.056 ± 0.021(?0.112) 0.044 ± 0.002(?0.163) 0.057 ± 0.004(?0.180) 0.055 ± 0.006(?0.184)


Colour gradient

In Fig. 8, the luminosity dependence of (g ? i) colour gradient for each class is shown, and the linear ?ts in Fig. 8 are summarised in Table 4. As found by Choi et al. (2007), REGs have negative colour gradients (blue outside) on average, and show little variation of their colour gradients with respect to luminosity. Such a colour gradient in REGs is consistent with the previous studies, and may originate from the internal metallicity gradient (Tamura & Ohta 2003; La Barera et al. 2005). The centres of BEGs are bluer than their outer parts, which results from the shape of the domain in which early-type galaxies were selected (Fig. 1). Choi et al. (2007) reported that the colour gradient of BEGs increases as luminosity decreases, and we additionally found


J. H. Lee et al.
Table 5. As Table 3 but for inverse concentration REG Passive HII Seyfert LINER 0.014 ± 0.003(0.325) 0.012 ± 0.002(0.330) 0.015 ± 0.002(0.335) 0.007 ± 0.002(0.329) RLG Passive HII Seyfert LINER 0.009 ± 0.004(0.372) 0.010 ± 0.002(0.389) 0.000 ± 0.004(0.393) 0.003 ± 0.003(0.378) BEG 0.017 ± 0.003(0.331) 0.018 ± 0.004(0.332) 0.008 ± 0.003(0.334) 0.022 ± 0.002(0.335) BLG ?0.015 ± 0.009(0.472) 0.005 ± 0.002(0.462) ?0.004 ± 0.007(0.435) ?0.005 ± 0.003(0.431)

Figure 9. The same as Fig. 7, but inverse concentration as Yaxis.

that such a trend is most conspicuous in hBEGs. This implies that star formation activity in BEGs may be centrally concentrated, and that the central star formation activity in faint BEGs may be more vigorous than those in bright BEGs. All sub-classes of late-type galaxies have negative colour gradients (i.e. bluer outside than centre) and show signi?cantly increasing colour gradients as luminosity decreases. The increase in BLGs is slightly larger than that in RLGs. Faint late-type galaxies show very small colour di?erence between their centre and outside, which implies that the fraction of Scd- and Im-type galaxies in late-type galaxies may increase as luminosity decreases, as pointed out by Choi et al. (2007). It is noted that AGN host BLGs show variation in their colour gradient larger than hBLGs, which is mainly due to the (negatively) large colour gradients of bright AGN host BLGs. This shows that bright AGN host BLGs may have vigorous star formation activity in their discs, or that the gas cooling in their galactic centre may be suppressed by AGN feedback (Cattaneo et al. 2007). The colours of AGNs cannot be responsible for those trends in colour gradients, because the colours of AGNs are known to be typically blue (Im et al. 2007), which would cause the opposite trend.

ation of inverse concentration are found between di?erent volumes. This may be mainly because we did not conduct morphological k-correction in estimating the inverse concentration. Table 5 lists the linear ?ts in Fig. 9. In earlytype galaxies, there is a clear trend that fainter galaxies are less concentrated, as expected from the early-type selection criteria and consistent with the well-known nonhomology in early-type galaxies (Kormendy & Djorgovski 1989; Michard & Marchal 1994). It is notable that no clear di?erence is found between pREGs, hREGs, pBEGs and hBEGs, which may indicate that the structures of these galaxies are similar. However, this does not necessarily mean that the mass pro?les of REGs and BEGs are similar, because BEGs may have young and bright stellar populations with relatively low mass-to-light ratio in their centre. Choi et al. (2007) showed that the brightest early-type galaxies are less concentrated, which may be due to recent mergers. We found that such a trend is most conspicuous in hBEGs