Astronomy
&
Astrophysics
A&A 545, A80 (2012)
DOI: 10.1051/0004-6361/201218769
c ESO 2012
A six-parameter space to describe galaxy diversification⋆
D. Fraix-Burnet1 , T. Chattopadhyay2 , A. K. Chattopadhyay3 , E. Davoust4 , and M. Thuillard5
1
2
3
4
5
Université Joseph Fourier – Grenoble 1/CNRS, Institut de Planétologie et d’Astrophysique de Grenoble, BP 53,
38041 Grenoble Cedex 9, France
e-mail: fraix@obs.ujf-grenoble.fr
Department of Applied Mathematics, Calcutta University, 92 A.P.C. Road, 700009 Kolkata, India
Department of Statistics, Calcutta University, 35 Ballygunge Circular Road, 700019 Kolkata, India
Université de Toulouse, CNRS, Institut de Recherches en Astrophysique et Planétologie, 14 Av. E. Belin, 31400 Toulouse, France
La Colline, 2072 St-Blaise, Switzerland
Received 3 January 2012 / Accepted 9 June 2012
ABSTRACT
Context. The diversification of galaxies is caused by transforming events such as accretion, interaction, or mergers. These explain
the formation and evolution of galaxies, which can now be described by many observables. Multivariate analyses are the obvious
tools to tackle the available datasets and understand the differences between different kinds of objects. However, depending on the
method used, redundancies, incompatibilities, or subjective choices of the parameters can diminish the usefulness of these analyses.
The behaviour of the available parameters should be analysed before any objective reduction in the dimensionality and any subsequent
clustering analyses can be undertaken, especially in an evolutionary context.
Aims. We study a sample of 424 early-type galaxies described by 25 parameters, 10 of which are Lick indices, to identify the most
discriminant parameters and construct an evolutionary classification of these objects.
Methods. Four independent statistical methods are used to investigate the discriminant properties of the observables and the partitioning of the 424 galaxies: principal component analysis, K-means cluster analysis, minimum contradiction analysis, and Cladistics.
Results. The methods agree in terms of six parameters: central velocity dispersion, disc-to-bulge ratio, effective surface brightness,
metallicity, and the line indices NaD and OIII. The partitioning found using these six parameters, when projected onto the fundamental plane, looks very similar to the partitioning obtained previously for a totally different sample and based only on the parameters of
the fundamental plane. Two additional groups are identified here, and we are able to provide some more constraints on the assembly
history of galaxies within each group thanks to the larger number of parameters. We also identify another “fundamental plane” with
the absolute K magnitude, the linear diameter, and the Lick index Hβ. We confirm that the Mg b vs. velocity dispersion correlation
is very probably an evolutionary correlation, in addition to several other scaling relations. Finally, combining the results of our two
papers, we obtain a classification of galaxies that is based on the transforming processes that are at the origin of the different groups.
Conclusions. By taking into account that galaxies are evolving complex objects and using appropriate tools, we are able to derive an
explanatory classification of galaxies, based on the physical causes of the diverse properties of galaxies, as opposed to the descriptive
classifications that are quite common in astrophysics.
Key words. galaxies: elliptical and lenticular, cD – galaxies: evolution – galaxies: formation – galaxies: fundamental parameters –
methods: statistical
1. Introduction
Galaxies are complex and evolving objects. Their diversity appears to increase rapidly with the instrumental improvements
that produce huge databases. A good understanding of the
physics governing the processes at work within and between the
different components of galaxies requires numerical simulations
that produce synthetic populations of hopefully realistic objects.
The number of physical processes that may operate together with
their infinite possible configurations, render the morphological
Hubble classification and its equivalents obviously too simple.
Morphology, as detailed as it can be determined in the visible, is
only one component of the physics of galaxies, and ignores many
ingredients of galaxy evolution, such as kinematics and chemical composition (e.g. Cappellari et al. 2011). In addition, these
classifications do not make full use of the wealth of information
that observations and numerical simulations provide.
⋆
Appendices A–C are available in electronic form at
http://www.aanda.org
Multivariate partitioning analyses appear to be the most
appropriate tools. One basic tool, the Principal component
analysis, is relatively well-known (e.g. Cabanac et al. 2002;
Recio-Blanco et al. 2006), but this is not a clustering tool in
itself. Many attempts to apply multivariate clustering methods have been made (e.g. Ellis et al. 2005; Chattopadhyay &
Chattopadhyay 2006, 2007; Chattopadhyay et al. 2007, 2008,
2009a,b; Fraix-Burnet et al. 2009; Sánchez Almeida et al. 2010;
Fraix-Burnet et al. 2010). Sophisticated statistical tools are used
in some areas of astrophysics and are being developed steadily,
but multivariate analysis and clustering techniques have not yet
been widely applied across the community. It is true that the interpretation of the results is not always easy. Some of the reasons
are given below.
Before using the available parameters to derive and compare the physical properties of galaxies, it is important to check
whether they can discriminate between different kinds of galaxies. A partitioning of objects into robust groups can only be obtained with discriminant parameters. This does not necessarily
Article published by EDP Sciences
A80, page 1 of 24
A&A 545, A80 (2012)
preclude using other information to help the physical and evolutionary interpretation of the properties of the groups and the relationships between groups. Among the descriptors of galaxies,
many come directly from the observations independently of any
model. In principle, all the information is contained within the
spectrum. However, since it is a huge amount of information, it is
usually summarized by broad-band fluxes (magnitudes), slopes
(colours), medium-band and line fluxes (e.g. the Lick indices).
This dimensionality reduction is often guided by observational
constraints or some physical a priori, but not by discriminant (i.e.
statistical) properties.
Multivariate partitionings group objects according to global
similarities. They yield a descriptive classification of the diversity, but do not provide any explanation of the differences in
properties between groups. Modelling or numerical simulations
must be used to understand physically the partitioning and the
relationships between the groups.
However, galaxy properties are indeed explained by evolution. Mass, metallicity, morphology, colours, etc., are all the
result of galaxy evolution. It can thus be expected that the
relationships between the groups are driven by evolution. In
addition, galaxy (formation and) evolution proceeds through a
limited number of transforming processes (monolithic collapse,
secular evolution, gravitational interaction, accretion/merger,
and sweeping/ejection, see Fraix-Burnet et al. 2006b,c). Since
they depend on so many parameters (initial conditions, nature
of the objects involved, impact parameters, ...), the outcomes
of each of them vary a lot, so that diversity is naturally created
through evolution of the galaxy populations. This is what we call
diversification.
It is easy to see that each of these transforming processes
follows a “transmission with modification” scheme, because a
galaxy is made of stars, gas, and dust, which are both transmitted and modified during the transforming event (Fraix-Burnet
et al. 2006b,c,a). This is why one might expect a hierarchical organization of galaxy diversity with evolutionary relationships between groups. Cladistics has been shown to be an adequate (Fraix-Burnet et al. 2006b,c,a) and quite effective (e.g.
Fraix-Burnet et al. 2009, 2010) tool for identifying this hierarchical organization. Instead of a descriptive classification of
galaxy diversity, we hope to be able to build an explanatory classification that is physically more informative.
In a previous work (Fraix-Burnet et al. 2010), we found that
the fundamental plane of early-type galaxies is probably generated by diversification. We believe that this result can be of such
importance as to deserve dedicated studies to assess its robustness. The present work is novel in several fundamentall ways:
– a distinct data set is used, for which more parameters are
available, useful for the analyses themselves and also for the
subsequent interpretation, once the groups are determined
(Sect. 2.1);
– two additional methods, PCA and MCA, are used (Sect. 2.2);
– a new set of parameters is used for the various partitioning
methods (Sect. 4), which are selected in a rather objective
way (Sect. 3), unlike the ad hoc selection of parameters from
longtime conventional wisdom (as followed in Fraix-Burnet
et al. 2010);
– measurement errors are used in the classification in the case
of the cladistic analysis (Sect. 2.1 and Appendix B.2);
– the combination of the partitionings in the present paper and
those in Fraix-Burnet et al. (2010) help us to devise a new
scheme for galaxy classification.
A80, page 2 of 24
We present the data in Sect. 2.1 before describing the philosophy of our approach with the different methods used to analyse
the discriminant properties of the parameters (i.e. their ability
to discriminate between different groups) and the partitioning of
the sample in Sect. 2.2. We then give the results of these analyses and the “winning” set of parameters (Sect. 3) that is used for
the partitioning (Sect. 4). We then comment on the discriminant
parameters (Sect. 5) and detail the group properties (Sect. 6).
Scaling relations, correlations, and scatter plots are presented in
Sect. 7, and we discuss the well-known fundamental plane of
early-type galaxies as well as another fundamental plane, which
we discover in this paper, in Sect. 7.4. Finally, we combine our
present result with the one in Fraix-Burnet et al. (2010) by plotting a cladogram that summarizes the inferred assembly histories
of the galaxies and thus is a tentative new scheme for classifying
galaxies (Sect. 8). The conclusion of this study closes this paper
(Sect. 9).
2. Data and methods
2.1. Data
We selected 424 fully documented galaxies from the sample
of 509 early-type galaxies in the local Universe of Ogando
et al. (2008). As these authors point out, this sample appears
to be relatively small compared to those at intermediate redshifts that have been obtained with large surveys (such as the
Sloan Digital Sky Survey). However, they do have the advantage of higher quality spectroscopic data and more reliable structural information such as the effective radius. To describe the galaxies, we took from Ogando et al. (2008) the
10 parameters that belong to the set of 25 Lick indices defined by Worthey & Ottaviani (1997): Hβ, Fe5015, Mg1, Mg2 ,
Mg b, Fe5270, Fe5335, Fe5406, Fe5709, and NaD. From these
Lick indices, we computed two other parameters defined as:
[MgbFe]′ = Mgb ∗ (0.72 ∗ Fe5270 + 0.28 ∗ Fe5335) (Thomas
et al. 2003) and Mg b/Fe = Mg b/( 21 (Fe5270 + Fe5335))
(Gonzáles 1997), which are indicators of metallicity and lightelement abundance, respectively. The other parameters taken
from Ogando et al. (2008) were the number of companions nc ,
the morphological type T , the line index OIII, the velocity dispersion (log σ), and the linear effective radius (log re ).
The surface brightness within the effective radius (Brie ) and
the disc-to-bulge ratio (D/B) were taken from Alonso et al.
(2003). The absolute magnitude in B (Mabs ) and the distance
of the galaxies were taken from Hyperleda1, which adopts a
Hubble constant of 70 km s−1 Mpc−1 . The distances to three
galaxies not available in Hyperleda were taken from the literature: NGC 1400 (27.7 Mpc, from Perrett et al. 1997), NGC 4550
(15.49 Mpc from Mei et al. 2007), and NGC 5206 (3.6 Mpc from
Karachentsev et al. 2002). The colour B-R was calculated from
the corrected apparent B magnitude in Hyperleda and the total
R magnitude given by Alonso et al. (2003). The linear diameter
(log(diam)) was computed from logdc given in Hyperleda. The
infrared magnitudes and colours were taken or calculated from
NED2 .
Altogether, we have 25 parameters to describe the 424 galaxies. However, two parameters were removed from our analyses,
namely the number of companions nc and the morphological
type T , which are both discrete parameters. More importantly,
nc is not a property of the galaxies, but their local environment,
1
2
http://leda.univ-lyon1.fr/
http://nedwww.ipac.caltech.edu/
2.2. Methods
The philosophy of our approach is to use multivariate tools in a
first step to select the parameters that can discriminate different
groups within the whole sample. These parameters, called discriminant parameters, are then used in a second step to partition
the data into several groups.
In this paper, we use four methods, which are described in
more details in Appendix A. Three of them are used to analyse
the parameters: principal component analysis (PCA, Sect. A.1),
minimum contradiction analysis (MCA, Sect. A.2), and cladistics (Sect. A.3), while the groupings are performed with the two
latter (MCA and cladistics) together with a cluster analysis (CA,
Sect. A.4),
The four approaches are all very different in philosophy and
technique. Since there is no ideal statistical method, it is useful to compare results obtained with these independent methods.
Convergence improves confidence, but since the assumptions behind the different techniques are different, exact agreement cannot be expected. In the end, it is the physics that decides whether
a partitioning is informative.
3. Analyses of the parameters
In this section, we investigate the behaviour of the observables
using three of the methods presented above: PCA, MCA, and
cladistics. These three multivariate techniques use the parameters directly instead of distance measures, as in the cluster analysis also considered in this paper. Hence, much information can
be gained about the parameters themselves, such as their correlations (PCA) or their respective behaviour in the partitioning
process (MCA and cladistics). From this information, one can
infer the discriminant power of all parameters for the studied
sample.
6
4
2
0
while T is qualitative and subjective. The full set of 25 parameters is naturally used to interpret our results.
We thus used 23 parameters for the analyses in this paper:
three are geometrical (D/B, log re , and log(diam)), two come
from medium-resolution spectra (log σ and OIII), in addition
to the ten Lick indices (Hβ, Fe5015, Mg1, Mg2 , Mg b, Fe5270,
Fe5335, Fe5406, Fe5709, and NaD) and [MgbFe]′, Mg b/Fe, the
six others are broad-band observables (Brie, the absolute magnitudes in B (Mabs ) and K (Kabs ), the total colours B-R, J-H,
and H-K).
Ogando et al. (2008) provides error bars for each of these
parameters. However, evaluating the influence of measurement
errors on the partitioning is difficult because their multivariate
distribution function is unknown. This appears to be a big statistical problem. Fuzzy cluster analyses could perhaps be useful
but they are quite complicated to implement, and the very good
agreement between all our results indicate that such an investment is unnecessary at this point. In addition, measurement uncertainties can easily be integrated into the cladistic analysis. We
thus limit ourselves to two restricted assessments: the influence
of two determinations of the distances of galaxies (needed to determine re ) on the result, and the cladistic analysis with errors.
These are described in Appendix B. In any case, one should consider that the physical nature of galaxies implies that there are
continuous variations in the parameters, hence the partitions are
necessarily fuzzy with no rigid boundaries between groups. This
implies that there is some uncertainty in the placement of the
individual objects in the multivariate parameter space.
8
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
Fig. 1. PCA eigenvectors for the sample of 424 objects with 23 parameters. The eigenvalue 1 is indicated by the horizontal line and the eight
eigenvectors higher than 1 or so are darkened.
3.1. Principal component analysis
We performed a PCA analysis (Sect. A.1) on the set of 23 parameters. Six principal components (PC or eigenvectors) have
eigenvalues greater than 1 and two others are very close to 1
(Fig. 1), so that eight components describe most (82%) of the
variance of the sample while the first five account for 69% of the
variance.
The loadings (i.e. the coefficients of the parameters composing the eigenvectors, Table A.1) give some indications as to
which parameters are correlated, redundant, or discriminant. The
most important parameters (the first few with the highest loadings in the first eigenvector, and the first parameter for the other
eigenvectors) in each PC are:
1. Mg b, log σ, Mg2 , [MgbFe]′ , NaD, Mg1
2. Brie
3. OIII
4. Brie
5. H-K
6. J-H
7. D/B
8. Fe5709.
Since the parameters Mg b, Mg1, and Mg2 are so closely related
(Burstein et al. 1984), these three quantities are undoubtedly redundant. Moreover, [MgbFe]′ depends very much on Mg b, and
is considered a more accurate estimate of the metallicity of a
galaxy. Hence, the most important a priori non-redundant parameters are: log σ, [MgbFe]′, NaD, Brie , OIII, H-K, J-H, D/B,
and Fe5709.
We then performed a second PCA analysis after removing
the supposedly redundant parameters (Mg b, Mg1, Mg2 , Mabs )
and log re and log(diam) that are affected by the uncertainties in
the distance determination (see Appendix B.2). We also disregarded J-H, which has an outlier and otherwise quite constant
values. We note that somewhat paradoxically this behaviour
could explain why J-H appears in the sixth principal component above. We are now left with the 16 parameters log σ, Brie ,
D/B, Hβ, Fe5015, Fe5270, Fe5335, Fe5406, Fe5709, NaD, OIII,
H-K, B-R, Kabs , [MgbFe]′, and Mg b/Fe. Five eigenvectors have
eigenvalues higher than 1 and account for 68% of the variance.
The most important parameters are now:
1. [MgbFe]′ , log σ, NaD
2. Fe5015
A80, page 3 of 24
A&A 545, A80 (2012)
3. Brie
4. Fe5709
5. H-K.
The agreement is very good with [MgbFe]′, log σ, NaD, Brie ,
Fe5709, and H-K still present, while OIII has disappeared but
has loadings very close to those of Fe5015 and Fe5709 in components 2 and 4, respectively. The disc-to-bulge ratio D/B does
not appear to be as important in this analysis.
3.2. Minimum contradiction analysis
The MCA analysis (Sect. A.2) uses all parameters and explores
them to determine the best order (partitioning) that can be obtained. It is possible to derive the discriminant capacity of the
parameters according to their respective behaviour as formalised
in Thuillard & Fraix-Burnet (2009). We find that:
– log σ, Fe5270, NaD, [MgbFe]′ , Brie , B-R, OIII, and D/B appear as discriminant parameters;
– log σ, Kabs , log(diam) are strongly correlated.
In contrast to PCA, the correlations are not automatically removed, some or all of them may remain. In the present case,
the three strongly correlated parameters are not obviously redundant since they are not related by a direct causal relation
(see Fraix-Burnet 2011). However, keeping all of them for the
MCA analysis does not bring any more discriminating information than keeping only one (Thuillard & Fraix-Burnet 2009). As
a consequence, since log σ is listed as one of the discriminant parameters, Kabs and log(diam) can be disregarded in the analysis.
Consequently, the MCA analysis finds eight discriminant
parameters.
3.3. Cladistic analysis
Each cladistic analysis (Sect. A.3) uses and investigates a given
set of parameters. To improve our understanding of the 23 parameters, it would be necessary to analyse all possible subsets. Since this require too much computing time, we decided
to eliminate obvious redundancies (Mg b, Mg1, Mg2 , Mabs , and
log(diam)). The index Hβ, which is an age indicator for stellar
populations older than a few hundred Myr, is problematic for
cladistics because age is a property of all groups. This parameter
might be able to trace recent transformative events accompanied
by starbursts, if it were not for the degeneracy between the age of
a younger stellar component and its relative contribution to the
total stellar mass or luminosity. Guided by the PCA and MCA
analyses, we also disregarded Fe5335, Fe5406, H-K, Mg b/Fe,
and J-H. It is remarkable that log re is not found in the PCA and
MCA analyses as a discriminant parameter. We thus also disregarded it here, but kept it for a specific analysis of the fundamental plane together with log σ, Brie , and Mg2 (see Sect. 7.4.1 and
Fraix-Burnet et al. 2010).
Finally, we studied in more detail the remaining eleven parameters: log σ, D/B, NaD, [MgbFe]′ , Brie , OIII, Mg b, Fe5015,
Fe5270, Fe5709, and B-R. To find the most discriminant ones in
this list, we examined the relative robustness of the trees obtained by cladistic analyses using the eight subsets of these parameters listed in Table 1. Analyses of each parameter subset
were performed with the full sample and several subsamples,
and all the results were compared. The details of our procedure
are presented in Sect. A.3.
Five or six discriminant parameters are favored by the cladistic analysis because the results are then more stable. The trees
A80, page 4 of 24
Table 1. Subsets of parameters used in cladistic analyses to determine
the most discriminant parameters.
Subset
4cA
5c
5cA
6c
6cA
7c
8c
10c
Parameters
log σ D/B NaD [MgbFe]′
log σ D/B NaD [MgbFe]′ Brie
log σ D/B NaD [MgbFe]′ Mg b
log σ D/B NaD [MgbFe]′ Brie OIII
log σ D/B NaD [MgbFe]′ Mg b Brie
log σ D/B NaD [MgbFe]′ Brie OIII Fe5015
log σ D/B NaD [MgbFe]′ Brie OIII Fe5015 Mg b
log σ D/B NaD [MgbFe]′ Brie OIII Mg b Fe5270
Fe5709 B-R
Notes. The names of the subsets include the number of parameters.
from subsets 5c and 6c are in very good agreement, that of 6c
being almost entirely structured, and the result for subset 6c is in
very good agreement with the cluster analysis (Sect. 4).
We conclude that the most discriminant set of parameters
from the cladistic analyses is that of 6c, namely log σ, D/B,
NaD, [MgbFe]′ , Brie , and OIII.
3.4. Final set of discriminant parameters
The PCA, MCA, and cladistic analyses agree in terms of the
five parameters log σ, [MgbFe]′, NaD, Brie , and OIII, while
they globally identify five to eight discriminant parameters. The
cladistic and MCA analyses find that D/B is an important parameter, which appears only weakly in PCA. B-R is discriminant
in MCA only, while the iron indices appear with Fe5015 and
Fe5709 on one side (PCA), and Fe5270 on the other (MCA).
None of these four parameters are preferred in the cladistic
analyses.
Hence, we select the consensual six parameters log σ, D/B,
NaD, [MgbFe]′, Brie , and OIII for partitioning analyses of our
sample.
4. Partitioning the sample galaxies
We now compare the partitioning obtained with four methods:
a cluster analysis using eight principal components (Sect. 4.1),
a cluster analysis (Sect. 4.2), a MCA optimisation (Sect. 4.3),
and a cladistic analysis (Sect. 4.4), the latter three using the six
parameters listed in Sect. 3.4. The partitionings are compared at
the end of this section and in Fig. 2. The order of the groups for
cladistics is essentially dictated by the tree (and its rooting, see
Sect. 4.4), while for the other methods, the order was arbitrarily
chosen to correspond as much as possible to the cladistic order.
4.1. PCA plus cluster analysis
A cluster analysis (Sect. A.4) was performed using the eight PCs
obtained by PCA (Sects. 3.1 and A.1). In this paper, this analysis is denoted PCA+CA. Three groups are found and labelled
PCACA1, PCACA2, and PCACA3. The first two contain about
100 objects, while the third is about twice as big (Fig. 2).
As noted in Sect. A.1, the use of principal components in
multivariate clustering very likely obscures a significant part
of the underlying physics since it suppresses all correlations,
even those that are due to hidden parameters or independent
evolutions (see Sect. 6). We present this result here mainly as
an illustration of this point.
Cluster Analysis
100
NGC5583
NGC1412
NGC4515
NGC1419
NGC5770
NGC4318
NGC0770
IC1317
IC4180
NGC3522
ESO286G050
NGC3056
ESO358G059
NGC0516
NGC1331
NGC7404
ESO235G051
ESO440G038
IC5267B
ESO447G031
ESO489G037
NGC1366
IC4796
ESO384G019
NGC4685
NGC2865
ESO378G020
NGC6172
IC2200A
NGC4997
ESO231G017
NGC4215
IC2552
NGC1439
NGC3142
NGC6548
ESO445G049
NGC4553
NGC7302
ESO499G013
ESO443G039
ESO507G014
NGC1427
UGC04228
NGC3300
NGC3617
NGC0277
ESO501G025
IC1445
ESO498G004
NGC2888
ESO545G042
NGC0442
NGC5516
NGC2679
ESO384G021
ESO384G023
NGC4612
NGC4489
NGC3412
IC4913
NGC7280
NGC1393
NGC7079
ESO490G006
NGC4739
NGC6909
ESO317G021
NGC1379
ESO507G024
ESO548G033
NGC7365
NGC7185
NGC1336
IC2764
NGC1315
NGC1172
UGC12840
ESO579G017
NGC0830
NGC4239
NGC0349
NGC7351
NGC4694
NGC2106
UGC09288
NGC7600
NGC2716
ESO283G020
IC1729
ESO567G052
ESO386G038
NGC0125
NGC5726
NGC0790
ESO437G021
ESO437G045
NGC3892
ESO533G025
NGC6725
NGC1979
UGC05226
ESO563G024
ESO058G019
IC4943
ESO510G054
NGC2984
NGC5330
NGC0448
NGC4603C
NGC4379
NGC3605
NGC4474
ESO242G014
NGC3325
ESO323G092
UGC04670
ESO501G056
ESO182G007
UGC08872
IC1858
ESO341G013
NGC4993
NGC4373A
NGC4645A
NGC5304
NGC0599
NGC0875
ESO282G028
ESO436G027
NGC2729
ESO139G005
NGC1993
NGC4191
NGC7145
NGC0155
NGC0968
NGC6706
IC1609
NGC1403
NGC4855
NGC1567
NGC5761
NGC3483
ESO263G033
IC4310
NGC2073
NGC2272
NGC3282
MCG0135007
IC0555
NGC0502
NGC2723
ESO507G027
NGC3954
NGC1374
ESO423G024
NGC0809
ESO316G046
ESO437G038
NGC3082
NGC7029
NGC0641
NGC0273
NGC3546
NGC0430
ESO491G006
NGC3904
NGC0682
MCG0207002
NGC1339
NGC3564
NGC5831
ESO387G016
IC2526
NGC2502
NGC4958
NGC5858
NGC1383
NGC5017
NGC4551
NGC7805
NGC4339
NGC4270
MCG0532074
NGC4417
NGC3085
ESO428G011
ESO221G037
NGC2191
ESO486G019
NGC5493
ESO437G015
NGC0862
ESO569G012
NGC0113
IC0999
NGC5343
NGC3591
NGC0596
NGC4033
NGC1381
NGC5049
NGC3224
NGC1389
ESO447G030
ESO376G009
NGC7576
NGC2891
UGC10864
NGC7359
NGC4550
NGC6017
NGC4240
ESO382G034
NGC3599
ESO576G030
NGC3335
NGC1297
IC0494
IC2587
NGC1298
NGC2267
NGC1426
ESO501G003
ESO084G028
NGC7144
NGC6924
NGC0774
NGC3096
NGC4756
MCG0127015
NGC3311
ESO512G019
MCG0157004
NGC0474
NGC5928
NGC1132
NGC3768
NGC6364
NGC6442
NGC3308
IC1492
NGC6587
UGC07813
NGC3022
NGC4714
ESO466G026
NGC1713
NGC5626
NGC0163
NGC6375
NGC5903
ESO367G008
NGC3302
NGC0969
NGC5424
ESO500G018
IC1864
MCG0103049
UGC01169
NGC2211
NGC1394
NGC3332
MCG0213009
NGC0193
IC0642
NGC3051
NGC2945
NGC1653
NGC4404
IC0513
IC0164
NGC5628
ESO126G014
IC0395
ESO507G021
NGC2911
NGC6614
NGC3100
NGC4825
NGC2974
NGC7097
NGC4546
IC1860
NGC0050
IC5157
NGC7626
IC3896
IC3289
NGC1549
NGC5813
NGC6903
IC4329
NGC1453
NGC4730
IC4350
NGC0541
NGC1200
ESO137G024
NGC5111
IC2437
NGC7832
NGC4233
ESO282G010
NGC2325
ESO142G049
NGC3805
NGC2749
NGC5898
NGC1553
ESO445G028
UGC12515
ESO208G021
NGC4374
NGC2822
NGC5507
IC0090
NGC2983
NGC3316
IC0232
NGC1162
NGC5397
NGC1380
IC5328
NGC4683
NGC0842
NGC4830
NGC2918
NGC3598
NGC0686
NGC1199
NGC3615
NGC5869
NGC5193
NGC5203
NGC3886
NGC1521
ESO263G048
NGC7778
NGC6495
NGC7458
NGC2765
MCG0127018
ESO316G034
MCG0103018
NGC7075
ESO322G051
IC4421
ESO509G108
IC4731
NGC3305
NGC3778
NGC0636
NGC1461
NGC4697
IC0552
ESO124G014
NGC1201
NGC4645B
ESO377G029
NGC7173
ESO425G014
ESO417G021
NGC1574
ESO498G006
ESO489G035
NGC0584
NGC1209
UGC11082
NGC0936
NGC4078
ESO443G053
NGC3042
ESO221G020
ESO425G019
NGC1700
ESO510G071
ESO508G008
NGC7785
NGC6851
NGC7562
ESO141G003
NGC3585
NGC3273
NGC7176
NGC2663
ESO467G037
NGC7196
IC2586
NGC5846
NGC0547
ESO286G049
ESO235G049
NGC6868
NGC7049
ESO510G009
NGC4472
NGC1407
NGC5062
NGC1399
NGC1332
NGC6861
IC1459
NGC7507
NGC5087
NGC1404
NGC3923
NGC1395
NGC0883
NGC1400
NGC3919
NGC0720
NGC6841
ESO282G024
NGC5028
NGC3209
ESO494G031
NGC1550
NGC2986
NGC3091
NGC5419
NGC7014
IC4296
80
200
3
1
2
3
4
5
6
7
100
0 1 2 3 4 5 6 7 8
3
60
20
1
2
3
4
5
6
7
100
0 1 2 3 4 5 6 7 8
3
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
20
40
0
20
0
2
Clad2
40
60
60
80
150
100
50
0
1
Clad1
Fig. 3. Most parsimonious tree found with cladistics with the identification of the eight groups and their corresponding colours.
80
200
120
2
0
0
20
50
40
40
60
100
80
150
80
200
120
2
0
0
20
20
40
40
60
60
80
150
100
50
0
1
1
Colours for clades
Cladistic analysis
120
PCA+CA
0
Colours for clusters
Colours for PCA+CA groups
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
1
2
3
4
5
6
7
0 1 2 3 4 5 6 7 8
Fig. 2. Comparison of analyses with three different methods: PCA+CA
(left panels), cluster analysis (middle panels), and cladistics (right panels). The MCA result is not shown since it can be easily compared to
the cladistic result (see Sect. 4.5). In the first row, the colours identify the eight groups found in the cladistic analysis; in the second row,
they identify the seven groups of the cluster analysis and, in the third
row, the three groups found in PCA+CA. The colours for the cluster
and PCA+CA groups are chosen to more easily visualize the agreement
with the cladistic partitioning.
4.2. Cluster analysis
A cluster analysis (Sect. A.4) was performed with the six parameters listed in Sect. 3.4. Seven groups were found and named
Clus1 to Clus7. There are three large groups, with about 80
to 120 objects. The other contain from 20 to 40 objects (Fig. 2).
4.3. Minimum contradiction analysis
With the six parameters listed in Sect. 3.4, the MCA analysis
performs an optimisation of the order to minimise the contradiction (Sect. A.2). The result is four groups, and maybe two others.
The groups are globally very fuzzy, i.e. they have no sharp limits.
This is expected because of the continuous nature of the parameters, and because of both uncertainties and measurement errors.
This is an important point that is essentially overlooked by the
other methods, and that should be kept in mind.
As we see below, these four groups are easily identified with
the groups obtained by cladistics, and for this reason they are not
given labels in this paper.
4.4. Cladistic analysis
The cladistic analysis, performed with the six parameters selected in Sect. 3.4, produces a most parsimonious tree (shown
in Fig. 3), on which we can identify groups. There is no absolute
rule for defining groups on a cladogram. However, substructures
in the tree are a good guide.
We identified eight groups in this tree, three large ones with
more than 80 objects, an intermediate group with about 50 objects, and four smaller ones with fewer than 30 members (Fig. 3).
These groups are named “Clad1” (the most ancestral one, at the
top) to “Clad8” (at the bottom) (Fig. 3). This numbering and
presentation of the tree should not a priori be seen as a diversification arrow since branches can be switched graphically. It is the
physical interpretation that both confirms the possible ancestrality of group Clad1 and gives the right order of diversification.
The rooting of the tree (i.e. the choice of objects that appear
graphically at the top of the tree and are supposed to be the closest to the common “ancestor species” of all the objects of the
sample) is necessary to define the direction of diversification,
and in general affects the contours of the groups. The tree is here
rooted with the group of the lowest average metallicity as measured by [MgbFe]′ according to our assumption that the lowest
metallicity corresponds to the most ancestral objects. This guess
is monovariate and might not represent the best choice in a multivariate study like the present one. However, we do not yet have
a better multivariate criterion for primitiveness. The rooting of
the tree can easily be changed.
In contrast to the other partitioning methods, some objects
appear to be isolated on the tree and consequently cannot be easily grouped with others. Each of them could indeed represent a
class, but, for the sake of simplicity, we decided not to identify
them with specific colours. We simply gather all such objects
as Clad0, give them a grey colour in plots or simply disregard
them in the discussions that deal with the statistical properties of
groups.
4.5. Comparison of the four partitionings
The four methods produce three (PCA+CA), four (MCA), seven
(cluster analysis), and eight (cladistics) groups. They thus all
agree for a relatively small number of groups.
The agreement between cladistics and PCA+CA is quite
good (see Fig. 2), if we identify the three following groups:
(Clad1, Clad2, Clad3, Clad4, Clad5, and a part of Clad6),
(Clad7), and (Clad6, Clad8) with PCACA1, PCACA2, and
PCACA3 respectively.
A80, page 5 of 24
A&A 545, A80 (2012)
The agreement between the PCA+CA and cluster analyses is
also quite good with PCACA3 being composed of Clus4, Clus7,
and a part of Clus6, PCACA2 being composed essentially of
Clus5 and partly Clus6, and PCACA1 being mainly composed
of Clus1, Clus3, and part of Clus4.
Cluster and cladistic partitionings agree very well for
the three large groups (Clus4 ≃ Clad6, Clus5 + Clus6 ≃ Clad7,
and Clus7 ≃ Clad8). The situation is slightly more complicated for the other groups, but still convergent with
Clus1 ≃ Clad1 + Clad3, and Clad2 being split mainly into Clus2,
Clus3, and also Clus4. Conversely, Clus2 is contained mainly in
Clad2 and also in Clad7.
The four groups from MCA are in very good correspondence
with groups Clad6, Clad7, Clad8, and Clad1 + Clad3. On the
other hand, Clad2 + Clad5 does not seem well-justified based on
the MCA result. Interestingly, Clad6 and Clad8 are not quite independent, in agreement with PCACA3 being mainly composed
of Clad6 and Clad8 as seen above.
The PCA+CA identifies a smaller number of groups than
the other partitionings do. This was expected, because of the
effect of the PCA analysis, which eliminates too many of the
correlations (Sect. A.1). The MCA result reinforces the significance of groups Clad6, Clad7, Clad8, and Clad1 + Clad3, which
are all also identified in the cluster analysis. The other groups
from cladistics and cluster analyses are either less robust or more
fuzzy.
We conclude that the number of groups is at least four, and
probably seven or eight. In the following, we consider the cladistic result with eight groups because it provides the very important evolutionary relationships between them.
Supplementary figures are given in Appendix C for the cluster partitioning and can be used to check that our interpretation
does not depend on the detailed boundaries of the groups. In addition, two complementary cladistic analyses were performed in
order to check the influences of two different determinations of
the distance of galaxies (needed to determine re ), and of measurement errors are presented in Appendix B.
5. Discriminant descriptors of galaxies
Among the initial 23 quantitative parameters (Sect. 2.1), only 6
are discriminant and actually yield a relatively robust partitioning (Sect. 3.4). The 17 remaining parameters do not yield enough
information to distinguish different classes of objects, because
either intrinsically they are not informative, they bear the same
redundant information as the discriminant ones, or they are not
discriminant for the sample under study.
It is remarkable that the global luminosity of the galaxies
(Mabs Kabs) is not discriminant. It is usually used as an indicator of mass and chosen as a main criterion of a priori classification. Luminosity is also often assumed to characterize the
level of evolution (for instance in the so-called “downsizing effect”). However, from a diversification point of view, the absence of the global luminosity is expected, since mass is a global
property that can be acquired by different processes, i.e. accretion or merging, which have different timescales and perturbing
powers. Such parameters, which show too much convergence,
are not well-suited to establish phylogenies, that is, they are
not good tracers of the assembly history of galaxies. Mass is
bound to increase, it is thus not specific to any particular assembly history, which could distinguish different kinds of galaxies. Nevertheless, mass is not entirely absent and is represented
somehow in log σ and Brie , which are certainly better tracers
than mass itself of the way in which mass has been assembled.
A80, page 6 of 24
The index OIII, which tends to decrease in more metallic
galaxies, is a discriminant parameter, but Hβ, which is often used
as an age indicator, is not. This is not so surprising since age is
not an indicator of diversity, as it is shared by all objects (see a
discussion in Fraix-Burnet et al. 2009). Age, even more so than
either mass or size, is bound to increase independently of the
assembly history. Anyhow, defining an age for a galaxy is tricky
and is often taken as the average age of the stellar populations,
which is a poor tracer of the assembly history.
The size parameters log re and log(diam) are not discriminant. They are probably merely scaling factors that are somewhat
similar to mass, and bound to increase regardless of the sequence
of transforming events that occur during the assembly history of
galaxies. However size does not seem to be represented at all,
and if so probably weakly in log σ and Brie , or even in some
hidden correlation, which we study later on in this paper.
However, one may wonder why Fraix-Burnet et al. (2010)
found a robust partitioning using only four parameters. Two of
them are in our list of six (log σ and Brie ), one of them (Mg2 )
is very similar to [MgbFe]′ , but the fourth one is log re , which is
not a discriminant parameter in the present analysis. There are
several reasons for this.
First, in Fraix-Burnet et al. (2010), the four parameters were
not the result of a multivariate and objective selection, but were
chosen because common wisdom suggests that they may be important for characterizing the physics of galaxies. The very positive result obtained with these four parameters strongly supports
this a priori, but the present paper demonstrates that only three
of them are really discriminant parameters.
Second, three parameters out of four are discriminant, so that
the partitioning signal is borne by these three. Unless the fourth
parameter (log re ) is strongly erratic or contradictory, this signal
is not expected to be entirely destroyed (see Sect. 7.4.1).
Third, the discriminant parameters may in principle be different from one sample to another, if the diversity of objects is
not equally covered. They may also depend on the initial set of
parameters, if more discriminant ones are present in a larger list.
This is probably the case for log re , which has been replaced by
better observables.
It is thus unsurprising that the four parameters used in
Fraix-Burnet et al. (2010) yielded a robust partitioning and that
we find more discriminant parameters in the present study. The
six parameters selected in the present analyses are not necessarily the most suitable ones for other samples, for which new
partitioning analyses should ideally be conducted.
6. Group properties
The groups identified by the partitioning methods must be understood in the light of their statistical and comparative properties.
In this section, we first identify the main trends along diversification and then describe the distinctive properties of the groups.
For this purpose, we use boxplots, which give the four quantiles of each parameter for the eight groups. We consider two
additional parameters. The dynamical mass is defined as Mdyn ≃
Aσ2 Re /G with A = 3.8 (with Mdyn in solar mass, σ in km s−1
and Re in kpc) according to Hopkins et al. (2008), as done in
Fraix-Burnet et al. (2010). This makes Mdyn ≃ 5.95 ∗ σ2 Re .
Using this mass, we compute the mass-to-light ratio M/Lr using Brie = −2.5 log(Lr /πre2 ) + 4.29.
We show the most informative boxplots in Fig. 4; the others do not indicate significant differences between groups. We
show the boxplots for the cluster partitioning in Fig. C.1 of
Appendix C.
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
logre
Mabs
−18
−20
−22
1.8
0.0
2.0
0.5
2.2
2.4
1.0
2.6
logs
1
2
3
4
5
6
7
8
1
2
3
5
6
7
8
1
2
3
D/B
5
6
7
8
6
7
8
6
7
8
6
7
8
7
8
7
8
7
5
3
0.0
2
18
19
1.0
4
20
21
2.0
6
22
23
4
NaD
3.0
Brie
4
1
2
3
4
5
6
7
8
1
2
3
5
6
7
8
1
2
Kabs
3
4
5
[MgbFe]’
1
2
3
4
5
6
7
4.0
3.0
2.0
−26
−1.0
−24
0.0
−22
1.0
2.0
−20
OIII
4
8
1
2
3
4
6
7
8
1
−2.0
−3.0
80 100
60
−4.0
40
−5.0
3
4
5
6
7
4
5
Hbeta
20
2
3
T
0
1
2
0.5 1.0 1.5 2.0 2.5 3.0
nc
5
8
1
2
3
5
6
7
8
1
2
Mgb
3
4
5
Fe5270
2
3
2.0
3
4
2.5
4
5
3.0
5
6
6
3.5
7
Fe5015
4
Figure 4 shows that log σ, log re , NaD, [MgbFe]′, Mg b,
log(diam), Fe5270, Fe5335, Mg b/Fe, Mdyn , and M/Lr essentially increase along the diversification rank defined on the tree
of Fig. 3, while Hβ, Mabs , Kabs , and possibly D/B decrease. As
already mentioned (Sect. 4.4), this rank is not necessarily as linear as it seems. Anyhow, the adopted rooting of the tree gives
a very sensible result: globally, galaxies tend to become more
metallic, more luminous, more massive, and larger with increasing diversification. At the same time, they acquire a larger central velocity dispersion, which is often related to the higher mass,
and NaD is also known to increase with both mass and velocity
dispersion. In addition, the decrease in Hβ indicates that the average age of the stellar populations increases with diversification.
Mg b/Fe increases with diversification. The index [α/Fe] is
known to increase with galaxy mass and age, because successive mergers and accretions trigger more intense star formation
on shorter timescales. These events clearly participate in the diversification of galaxies, confirming our observed increase in
Mg b/Fe.
The OIII index does not show any trend with diversification,
but has a lower median value for the three groups Clad3, Clad4,
and Clad7.
There is no systematic trend in environmental properties
with diversification, nc having a wide range in all groups, except in Clad4 where it is small. Since an observed galaxy is the
result of a long and multiple sequence of transforming events, it
is probably the past environment, rather than the observed one,
that plays a role in the diversification process.
The most diversified groups (Clad5 to Clad8) have on average a lower D/B ratio, suggesting that transforming events, such
as accretion, interaction, and mergers, tend to destroy discs and
build larger bulges, presumably by randomizing stellar orbits.
The morphological type is unevenly distributed among
groups, Clad2 and Clad4 having nearly only galaxies with T =
−2, whereas T = −5 galaxies are found mainly in Clad1 and
Clad8 (respectively, the most ancestral and one of the most diversified groups).
Apart from the general trends with diversification, the groups
have distinctive properties, otherwise there would be no reason for finding separate groups. Their distinctive properties
are the following, where the number of members is given in
parentheses:
1
2
3
4
5
6
7
8
1
2
3
5
6
7
8
1
2
log(diam)
3
4
5
6
Mgb/Fe
1.4
1.0
0.6
1.5
2.0
1.0
2.5
1.4
3.0
1.8
1.8
3.5
Fe5335
4
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
M/Lr
0
9.5 10.0
100
200
11.0
300
12.0
400
Mdyn
1
1
2
3
4
5
6
7
8
1
2
3
4
5
6
Fig. 4. Selection of the most interesting boxplots.
7
8
2
3
4
5
6
– Clad1 (20 objects): these galaxies have the properties expected from an ancestral group in being small, faint, discy,
and of low metallicity. They are young (although have a large
spread in Hβ) and have a large spread in morphological type.
They have very low Mdyn and log σ, and a low M/Lr .
– Clad2 (45): these galaxies have the same average Hβ, Brie ,
and OIII as Clad1. They are larger, brighter, more massive,
more metallic, and have a much higher log σ than both Clad1
and Clad3. They are all of morphological type T = −2, and
have the highest D/B of all groups by far.
– Clad3 (16): these galaxies have a large range in many parameters (Hβ, Mabs and Kabs , log(diam), nc , OIII, Mg b/Fe), but
not in [MgbFe]′, log re , log σ, Brie , NaD, or Mdyn . They have
a small log(diam) similar to Clad1, but a much higher log re .
They are relatively faint with a relatively low log σ and OIII,
and a high Brie .
– Clad4 (13): these galaxies resemble those of Clad2 in most
respects (see discussion below on the respective placements of Clad2 and Clad3). In particular, they are all of
morphological type T = −2. The main differences are that
Clad4 objects are large (the highest log re after Clad7), of low
A80, page 7 of 24
A&A 545, A80 (2012)
–
–
–
–
surface brightness (high Brie ), have higher Mdyn and M/Lr ,
and are slightly less discy.
Clad5 (30): these galaxies are very similar to the Clad4 ones,
except that they have a low Brie and a very low D/B, the
lowest of all groups with Clad8.
Clad6 (85): this is one of the three largest groups, which are
also the most diversified. Its galaxies have unexpectedly low
values of Mdyn , log re , and Brie . Interestingly, they have very
similar properties to those of Clad2 galaxies, except for a
much lower D/B, slightly lower Brie , and Hβ, and a slighty
higher Mg b/Fe.
Clad7 (94): these galaxies are the largest in this sample. They
are the most luminous and have the highest log σ, Mdyn ,
and M/Lr together with Clad8. Clad7 galaxies have a higher
Brie , slightly lower Hβ and OIII, and a slightly higher D/B
than Clad6 and Clad8.
Clad8 (106): their distinctive properties are a very low D/B
and a very large spread in morphological type, in a similar
way to Clad1. They have a higher log σ and a lower Brie
than galaxies of Clad7. They have values of Mdyn and M/Lr
that are as high as Clad7.
The Clad2 group often departs from the general trend along diversification (Fig. 4), which would seem smoother if Clad2 and
Clad3 were inverted. We have noted (Sect. 4.5) that Clad2 is split
between Clus2, Clus3, and also Clus4. It is significantly higher
than expected in log σ, D/B, NaD, [MgbFe]′, Mg b, log(diam),
Fe5270, Fe5335, and lower in Mabs and Kabs . This means that,
because of some parameters, it seems misplaced in the diversification scenario for other parameters. This behaviour is visible in
the partitioning from the cluster analysis (Fig. C.1) since Clus2
and Clad2 are partially similar. Hence, why was Clad2 placed so
early by the cladistic analysis, while it is more diversified in four
of the six parameters used for the analysis?
The diversification scenario given in the tree of Fig. 3 is obtained from the parsimony criterion, which chooses the simplest
combined evolution of all parameters. Taken individually, the
simplest evolutionary curve of each variable is monotonic with
as few reversals as possible. For instance, in the log σ boxplot of
Fig. 4, one would expect Clad2 and Clad3 to be inverted to avoid
the Clad2 box to “peak”. However, this is a multivariate compromise, and since Clad2 would be better placed in the very first
position on the D/B plot, to “smooth” the evolutionary curve, it
is understandable that this is the most parsimonious placement
on the tree. In addition, while the two discriminant parameters
Brie and OIII show a variable behaviour, they would nevertheless induce us to place Clad2 before Clad3.
The conclusion is that Clad2 is correctly placed in second
position, because this is a multivariate analysis, which seeks a
compromise among several parameters. This shows the importance of selecting the parameters objectively, with multivariate
tools. Otherwise, with too many redundant parameters, the peculiar properties of Clad2 could have easily been lost.
The relative and distinctive properties of the galaxies from
the different groups obviously cannot be summarized with only
one or two physical parameters. The relative properties of the
groups show that the evolution of galaxies is not linear. The
global trend in some properties (such as mass, metallicity, or Hβ)
may appear to be roughly linear globally, but a detailed analysis, and especially the distinctive properties within each group,
give many clues to understand the assembly history of the corresponding galaxies.
Having highlighted the group properties, we examine the
possible correlations between them in the next two sections.
A80, page 8 of 24
7. Scatter plots and correlations
Scatter plots must be examined using the partitioning to look for
different behaviours between groups or within groups.
In the first case, the distribution of the groups traces the
projected evolutionary track given by the tree. The fundamental plane is one example (Sect. 7.4.1 and Fraix-Burnet et al.
2010). We however focus on cases showing a roughly linear
track, where the groups are approximatively ordered along a linear correlation. We refer to these relations as evolutionary correlations (Sect. 7.1) since these groups are related in cladistics
by evolutionary relationships. They are important since they imply that the observed relation can be generated mainly by evolution, as found in Fraix-Burnet et al. (2010) and formalised in
Fraix-Burnet (2011).
In the second case (Sect. 7.3), some correlation may or may
not be present within a given group, independently of the global
behaviour between groups.
7.1. Evolutionary correlations
We confirm the evolutionary nature of the Mg2 – log σ correlation found by Fraix-Burnet et al. (2010) and identify several
other cases, the clearest ones being shown in Fig. 5. Such evolutionary correlations are revealed by the succession of groups
ordered along the correlation with the most ancestral group
(Clad1) at one end and the most diversified ones (Clad7 and
Clad8 here) at the other end.
Several of these evolutionary correlations involve the following set of parameters: log σ, log re , Mabs (and Kabs ), Hβ, Mg b
(and Mg1, Mg2 , [MgbFe]′, and Mg b/Fe), NaD, and log(diam).
Some relations are particularly tight (such as log(diam) vs. Kabs
or Mg b vs. [MgbFe]′ ). The iron Lick indices Fe5270, Fe5335,
and Fe5406, as well as H-K, also follow an evolutionary correlation with Mabs , although it is quite loose.
In all cases, except for log(diam) vs. Kabs and Mg b vs.
[MgbFe]′ (which are discussed in Sect. 7.3), the correlation is
not present within each group. This is a clear sign that there is no
direct causal physical link between the two variables, but simply
a change on average with galaxy diversification.
Thomas et al. (2005) discuss the origin of the Mg b vs. log σ
correlation. They find that metallicity, not age, is the main driving factor. This would again justify the use of metallicity as a
reasonable tracer of diversification. However, we find instead
that the Mg b vs. log σ correlation is an evolutionary correlation,
implying that diversification is indeed the real driver: metallicity,
like central velocity dispersion, is bound to change on average as
the galaxies evolve. This could explain why investigations find
that this correlation appears so sensitive to several parameters:
it has been proposed to be driven by metallicity, age, and relative abundance of different heavy elements (see Matković et al.
2009, for references). This sensitivity more probably points to
an underlying, hidden, and confounding factor, which creates the
apparent correlation (Fraix-Burnet 2011).
The correlations between Mg b and [MgbFe]′ and between
NaD and [MgbFe]′ are clearly evolutionary, with diversification
increasing from left to right in Fig. 5, while the correlations
found by Thomas et al. (2011) are driven by total metallicity,
which, for a given age and light-element ratio, increases from
left to right in their Figs. 6 and 8. The dispersion in the NaD vs.
[MgbFe]′ relation is larger than that in the Mg b vs. [MgbFe]′ relation for our data and theirs, and not well-accounted for by their
model, presumably because of the fixed age and light-element
ratio assumptions. Nevertheless, there is agreement between our
−18
Mabs
−22
−20
0.30
0.20
0.10
Mg2
0.40
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
1.8
2.0
2.2
2.4
2.6
1.8
2.0
2.2
2.4
2.6
2.4
2.6
logσ
2
4.0
2.0
3.0
[MgbFe]’
4
3
Mgb
5
6
logσ
1.8
2.0
2.2
2.4
2.6
1.8
2.0
logσ
2.2
logσ
result and their model since the average metallicity of galaxies
obviously increases with diversification.
The [MgbFe]′ vs. log σ, Mg b/Fe vs. log σ, and Mg b vs.
log σ correlations (Fig. 5) can be compared with the Z/H vs.
log σ and α/H vs. log σ in Fig. 16 in Kuntschner et al. (2010).
The same correlation is present, but we clearly show its evolutionary nature. Kuntschner et al. (2010) ask the question: “What
drives the [α/Fe] – log σe (or mass) relation?”. Our answer is
simply: diversification. They indeed arrive at the same conclusion because they find that, in their sample, “there is evidence
that the young stars with more solar-like [α/Fe] ratios, created
in fast-rotating disc-like components in low- and intermediatemass galaxies, reduce the global [α/Fe] and thus significantly
contribute to the apparent [α/Fe] – log σe relation”. These
galaxies belong to our Clad1 group as stated above, but they are
not the sole responsible cause of this “apparent” relation, since
all our groups are aligned along the same trend.
The Faber-Jackson relation, Mabs vs. log σ (Fig. 5), also appears to be a purely evolutionary correlation: the sequence of
evolutionary groups are aligned along this correlation. If mass
had been the hidden parameter, then a similar correlation should
exist within each group. This is not the case. This result was corroborated in an independent way by Nigoche-Netro et al. (2011).
6
5
2
1.0
3
4
NaD
1.8
1.4
Mgb/Fe
7
7.2. Diversification or ageing?
1.8
2.0
2.2
2.4
2.6
1.8
2.0
2.2
2.4
2.6
logσ
2
2
3
4
Mgb
5
4
3
NaD
5
6
6
7
logσ
2.0
3.0
4.0
2.0
3.0
3.0
M/L
1.5
2.0
2.5
−22
−24
−26
Kabs
4.0
[MgbFe]’
−20
[MgbFe]’
0.6
1.0
1.4
log(diam)
1.8
8.0
8.5
9.0
9.5
L
Fig. 5. Scatter plots showing evolutionary correlations. Colours are the
same as in Figs. 4 and 3.
There is a well-known degeneracy between the age (measured by
Hβ) and the metallicity (measured by [MgbFe]′ or (0.69 ∗ Mgb +
Fe5015)/2) of stellar populations, which models of stellar evolution have tried to circumvent (e.g. Tripicco & Bell 1995). In
particular, Thomas et al. (2011) and Kuntschner et al. (2010) superimposed evolutionary tracks from different models on galaxy
observations of Hβ as a function of a metallicity indicator defined by (0.69∗Mgb+Fe5015)/2. We reproduce the same figures
in Fig. 6 (left), showing with crosses the median for each group
and with arrows the principal direction of increases in age and
metallicity. We emphasize that the stellar evolution models used
by Thomas et al. (2011) and Kuntschner et al. (2010) to produce
their figures are single population models with fixed solar value
of [α/Fe], while the data used in this paper (Sect. 2.1) are integrated over the whole galaxy, mixing together the contributions
of possibly several different stellar populations.
Our groups are clearly arranged according to diversification
following an increase in both age and metallicity. The spread of
the correlation is large, and, within each individual group, the
range in age and metallicity is also large. This dispersion could
certainly be explained by many factors, such as an extended
horizontal branch, which can increase Hβ (e.g. Greggio 1997;
Matković et al. 2009). In all cases, the median for each group is
nearly perfectly aligned between the two axes for age and metallicity, ordered as in Fig. 3, except for Clad2 and Clad7, which
depart from the main alignment. This kind of plot indeed merely
tells us that the age, metallicity, and Hβ of galaxies evolve on
average with time.
However, we find no correlation between age and metallicity within individual groups. This is clearly shown in Fig. 6
(right), which plots Hβ vs. [MgbFe]′, which is a better indicator
of metallicity than Mg b. It is striking that the elongated inertial ellipses for Clad3, Clad5, and Clad7 are well-aligned with
the Hβ axis, with relatively little spread in metallicity, the one
for Clad1 is slightly inclined, and the one for Clad2 is aligned
along the global trend. Since the other ellipses are quite round,
Clad1 and Clad2 are the sole groups that might show a barely
significant correlation between Hβ and a metallicity indicator.
A80, page 9 of 24
2.2
3.0
3.0
A&A 545, A80 (2012)
2.5
1.4
1.6
2.0
3.8
4.0
4.2
4.4
1.0
Hbeta
1.0
3.6
1.5
3.4
0.5
0.5
1.0
Hbeta
2.0
1.2
Age
1.5
2.5
1.8
2.0
Metallicity
2.5
3.0
3.5
4.0
4.5
5.0
5.5
(0.69*Mgb+Fe5015)/2
2.0
2.5
3.0
3.5
4.0
4.5
[MgbFe]’
Fig. 6. Left: equivalent of Fig. 9 by Thomas et al. (2011) or Figs. 3 and 8 by Kuntschner et al. (2010). The inset shows the median for each group
and the arrows indicating the direction of increase of age and metallicity from single stellar population evolutionary models as shown by these
authors. Right: group inertia ellipses for the Hβ vs. [MgbFe]′ scatter plot.
The wider range in Hβ for the less diversified groups, especially Clad1 and Clad3, clearly appears in Fig. 6 and can also
be seen in Fig. 4. It probably corresponds to the well-known
wider range in age of the low-mass objects (e.g. Matković et al.
2009). Our interpretation is not that the low-mass galaxies have
had a longer star formation history: we propose instead that
these objects formed or appeared over a longer timescale in the
Universe’s history and the older ones have not changed much,
apart from the ageing of stars. Larger galaxies on average, necessarily took more time to assemble and complexify (diversify), so
that it is very unlikely to find young and very diversified galaxies. However, the notion of galaxy age must be questioned.
Figure 6 illustrates the fundamental difference between diversification and age. The Clad1 group, which is assumed to
be ancestral because of its low metallicity, appears to be the
youngest group on average according to stellar evolution models. If a galaxy is to resemble the most pristine objects, it must
not have been transformed too much even by secular evolution.
This is why the most pristine objects are necessarily relatively
young and metal-poor. Conversely, the most diversified objects
have a higher average stellar age, which provides no information on the epoch of the transforming events that gave them their
observed properties. As a consequence, old galaxies are not obvious ancestors. In addition, the spread in age within each group
is generally large and overlaps with the spread in age of the other
groups. Age is thus not a good landmark of evolution. From the
point of view of astrocladistics, the so-called downsizing effect
results from a confusion between age and the level of diversification (Fraix-Burnet et al. 2006b,c, 2009).
Hence, the age of a galaxy or a group of galaxies is probably
not so important and may even be meaningless (Serra & Trager
2007). We even find this term misleading, and suggest it be replaced by “average stellar age”. The diversification state should
be used instead, as it reflects the actual assembly history of a
galaxy.
7.3. Correlations within groups (“specific correlations”)
As previously seen (Sect. 7.1 and Fig. 5), the two scatter
plots of log(diam) vs. Kabs and Mg b vs. [MgbFe]′ show both
A80, page 10 of 24
a global evolutionary correlation and correlations within the
groups (which we call “specific correlations”). Four other scatter
plots only show specific correlations, the global correlation being less obvious and/or more dispersed : Brie and Kabs vs. log re ,
log(Mdyn ) vs. log re , and [MgbFe]′ vs. Fe5270. These six specific
correlations are shown in Fig. 7.
Diversification within each group is determined by the structure of the tree in Fig. 3. If we examine the evolution of the
parameters involved in the correlations along each branch (thus
each group) of the tree, we find that:
– Kabs increases slightly with diversification within Clad6 and
Clad8;
– [MgbFe]′ might possibly increase in Clad8;
– log re might possibly decrease in Clad6;
– Mdyn decreases slightly in Clad6 and might possibly increase
in Clad5 and Clad8.
This is clearly not enough to explain all the observed specific
correlations with evolution within the groups. However, the difference between objects of a same group is weaker than for the
whole sample and thus would require refined cladistic analyses
with possibly additional descriptors. We now examine in some
detail the scatter plots in Fig. 7.
The correlation, which is particularly tight and linear, between Mg b and [MgbFe]′ , also holds within each group. Since
the first parameter largely depends on the α-elements, while the
second is essentially independent of it (Thomas et al. 2011),
these specific correlations can probably be explained, in a
similar way to the global one, by evolution. The correlations
between [MgbFe]′ and either Fe5270 or Fe5335 have larger
scatters.
The correlation between log(diam) and Kabs seems to be
present within each group. This is not proof however that it is
a causal relation because the larger the galaxy the more luminous, it still be due to evolution within each group or to some
other confounding parameter (Fraix-Burnet 2011). In addition,
all correlations, either global or specific, are approximately similar, which suggests that they have the same explanation.
The Kormendy relation (Brie vs. log re ) clearly appears to
depend on the group, and has a far smaller scatter for the most
Kabs
2.0
2.5
3.0
3.5
4.0
−26 −25 −24 −23 −22 −21 −20
4
2
3
Mgb
5
6
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
4.5
0.6
0.8
Kabs
0.5
1.2
1.0
0.0
0.5
1.6
1.8
2.0
1.0
logre
9.5
3.5
3.0
2.0
2.5
10.0
10.5
[MgbFe]’
11.0
4.0
11.5
4.5
12.0
logre
log(Mdyn)
1.4
−26 −25 −24 −23 −22 −21 −20
22
21
Brie
20
19
18
0.0
1.0
log(diam)
23
[MgbFe]’
0.0
0.5
logre
1.0
2.0
2.5
3.0
3.5
Fe5270
here (log(diam) vs. Kabs and Mg b vs. [MgbFe]′), the global correlation is evolutionary (Sect. 7.1) and since all correlations appear to have approximately the same slope, then the specific
correlations should be evolutionary as well.
In the other cases where there is no obvious global correlation, the reason must be specific to the group, and probably
different from one group to another. The correlations might be
explained by a direct physical cause, or by a confounding parameter, which can still be evolution. Note that the confounding
factor may depend on the group.
Anyhow, the origin of the correlations and their properties
is quite complex. In the case of the Mdyn vs. log re relation, numerical simulations show that it is determined by several variables involved in the assembly history, such as the epoch of
the last merger, the level of dissipation, the number of accretion events, the impact parameters, and so forth (Robertson et al.
2006). Thanks to a good diversity of simulated galaxy populations, Fraix-Burnet et al. (2010) were able to derive the history
assembly of each group. The specific correlations can then be
explained by either several drivers from the physical point of
view, “cosmic variance” within each group from the observational point of view, or confounding factors from a statistical
point of view.
Consequently, the two scatter plots showing both global
and specific correlations are probably driven by a dominant
general evolutionary factor (such as perhaps dynamical evolution for log(diam) vs. Kabs and chemical evolution for Mg b
vs. [MgbFe]′) affecting all galaxies of the sample, while the
other ones have multiple and necessarily specific factors, as
in the Mdyn vs. log re relation. For instance, the importance of
merger events applies only to these galaxies that have experienced such a catastrophic transforming process during their assembly history.
Fig. 7. Scatter plots showing correlations within groups.
7.4. Fundamental planes
7.4.1. The fundamental plane of early-type galaxies
diversified groups. There is an “evolution” in the correlation
curve following diversification, the different relations appearing
stacked on each other. At first glance, galaxies are brighter when
more diversified, but this is not so simple if we look at Clad7
and Clad8: galaxies from the first group are globally larger and
fainter. The correlation also has a larger scatter for Clad1 and
Clad3.
The Kabs vs. log re relation is quite dispersed, but there are
slightly more convincing correlations for some groups, at least
for the most diversified ones. For Clad1, there is little variation in
log re , so that there is no real correlation. The difference between
this relation and the Kabs vs. log(diam) one is striking.
The Mdyn vs. log re relation is tight, with very clear correlations specific to each group and with relatively little overlap
between them. This plot can be usefully compared to numerical
simulations (e.g. Robertson et al. 2006) as done in Fraix-Burnet
et al. (2010, see Sects. 7.4.1 and 8).
The [MgbFe]′ vs. Fe5270 relation is quite dispersed, despite
the square-root relation linking both parameters. Correlations
can be easily seen within groups, and they appear to generally
differ (particularly in terms of the slope) from the global relation.
To summarize, there are several cases where the specific correlations are present in all or some of the groups, regardless of
whether the global correlation exists. Why is this so?
If the correlation is present both globally and within groups,
then we can guess that it is for the same reason. In the two cases
The well-known and intensively studied correlation between
Brie , log σ, and log re is called the fundamental plane. The first
multivariate analysis of this relation was performed recently
by Fraix-Burnet et al. (2010). The present sample and that of
Fraix-Burnet et al. (2010), which are both at low redshift, have
no galaxy in common and the parameters used to partition the
sample into groups are different.
The present partitioning is in excellent agreement with the
result of Fraix-Burnet et al. (2010) as illustrated in Fig. 8, which
shows the projection onto the fundamental plane (log σ vs. Brie )
of the partitionings obtained by cladistics and cluster analysis in
the present paper, and the partitioning obtained by Fraix-Burnet
et al. (2010). The structures within the fundamental plane found
in Fraix-Burnet et al. (2010) are thus confirmed.
There is a good correspondence between the groups in the
two studies, as can be seen in Fig. 8 and in more detail in
Fig. C.3: C1 includes Clad4 and a large part of Clad3, C4 ≃
Clad6, C3, C5, and C6 are essentially included into Clad7, and
C7 ≃ Clad8. The log re vs. Mdyn diagram (Figs. 9 and C.4) confirms these equivalences, pointing out that C1 overlaps both
Clad3 and Clad4 and is distributed in a way that is more similar to Clad3. There are however some differences.
Clad1 seems to occupy a region of the fundamental plane
(log σ vs. Brie ) that is not very well-covered by the sample used
by Fraix-Burnet et al. (2010).
A80, page 11 of 24
A&A 545, A80 (2012)
21
22
23
1.5
1.5
18
19
20
21
22
23
Brie
−0.5
1.0
2.2
2.0
17
11
12
logre
20
Brie
10
0.5
19
Clad2, Clad4
0.5
2.6
18
logσ
2.4
1.8
17
Clus1
Clus2
Clus3
Clus4
Clus5
Clus6
Clus7
1.8
2.4
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
2.0
logσ
2.0
Cluster Analysis
2.6
Cladistics
−0.5
this work
Fraix−Burnet et al 2010
2.2
C1
C2
C3
C4
C5
C6
C7
1.8
2.0
logσ
2.4
2.6
0.0
Fraix et al 2010
17
18
19
20
21
22
23
<mue>
Fig. 8. Projection on the fundamental plane. Comparison between the
partitionings obtained by cluster and cladistic analyses and that of
Fraix-Burnet et al. (2010). See also Fig. C.3.
Clad2 has no equivalent in Fraix-Burnet et al. (2010) when
projected onto the fundamental plane (Fig. 8). This group is also
plotted separately in Fig. 9 to show that it follows the same
correlation as the other groups, spanning nearly the full range
of both log re and Mdyn .
Clad5 is also absent in Fraix-Burnet et al. (2010). Since it appears in the very centre of both Figs. 8 and 9, we believe that it is
identified here because of the larger number of parameters used
here, which in effect provides a higher-resolution analysis. Its
properties were also found to be quite similar to Clad4 (Sect. 6),
so that Clad4 is quite different from C1 (see above).
Group C2 of Fraix-Burnet et al. (2010) is absent in the
present partitioning. This is perhaps because the corresponding
regions of the fundamental plane (Fig. 8) are not very populated
in our sample, or because of the different sets of parameters used
in the two studies.
To complete the comparison between the two studies, we
performed the same analysis as in Fraix-Burnet et al. (2010),
using the same four parameters log σ, log re , Brie , and Mg2
(Appendix B). Naturally, since these four parameters are not all
discriminant for the present sample, the resulting tree and the
corresponding partitioning are slightly less robust. However, the
agreement is still quite good. The sample used in the present paper (Ogando et al. 2008) is globally at a lower redshift than the
one used in Fraix-Burnet et al. (2010) (Hudson et al. 2001). This
renders the determination of the distance, and thus log re less accurate (see Ogando et al. 2008). This could partly explain why
log re has not been found to be discriminant or why the fourparameter result here is slightly less robust than in Fraix-Burnet
et al. (2010).
We thus confirm the result of Fraix-Burnet et al. (2010)
that there are structures within the fundamental plane. Since
the organisation of the groups on this plane defines very similar
A80, page 12 of 24
10
11
12
13
Mdyn
Fig. 9. Comparison between the partitioning obtained by cladistics and
that of Fraix-Burnet et al. (2010). In the inset, we plot the groups Clad2
and Clad4, which are difficult to see in the main graph. See also Fig. C.4.
evolutionary paths corresponding to a clear trend in diversification as defined by Fig. 3, we confirm that this global relation
in three-parameter space is mainly driven by diversification. For
the sake of clarity, we stress that this evolutionary interpretation
of the fundamental plane holds for the groups, not within the
groups, since neither the present paper nor that by Fraix-Burnet
et al. (2010) tackle this question. This would certainly merit further specific studies because Fraix-Burnet et al. (2010) found
that the tightness of the fundamental plane strongly depends on
the group considered, the least diversified ones showing a very
loose – if significant at all – correlation.
7.4.2. Other correlation planes
Concerning the fundamental plane, the groups follow one another along a path of diversification in the log re vs. log σ (edgeon projection) scatter plot, whereas they are clearly distinguishable and distributed in no obvious order in the Brie vs. log re
(face-on projection) scatter plot.
Hence, one can expect to find other “fundamental planes”
by looking at the behaviour of groups in two scatter plots made
with a set of three parameters. This is the case for Hβ, Kabs ,
and log(diam) (Fig. 10). The two latter parameters are obviously
reminiscent of Brie and log re , while Hβ is of a different nature
from log σ, so that this fundamental plane is not redundant with
the classical one. Replacing Hβ by D/B, one also obtains a nice
correlation, thus another fundamental plane.
It is interesting to note that the parameters of these fundamental planes might be easier to observe than for the classical one, but still allow the distance determination thanks
to log(diam).
There may be additional similar surfaces in higherdimension parameter spaces but that have more complex
projections. Would these be interesting to discover and would
they be more useful than the classical fundamental plane?
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
–
–
–
Fig. 10. Another “fundamental plane” in the parameter space defined
by Hβ, Kabs , and log(diam).
The answer to these questions is twofold. First, they are identical to the classical fundamental plane, in the sense that they
are essentially evolutionary correlations driven by diversification
(Fraix-Burnet 2011). All are simply projections of the tree in a
sub-parameter space. They are thus not more and not less informative. Second, they can be useful once the confounding factor
(here evolution) has been taken into account. This is however
beyond the scope of the present paper.
–
8. The assembly history of early-type galaxies
Fraix-Burnet et al. (2010) uncovered the assembly history of a
completely different sample of early-type galaxies by analysing
its properties in both the fundamental plane and a mass-radius
diagram with the help of numerical simulations from the literature. This sample was composed of galaxies in clusters, while
the present one is composed of galaxies in the field, groups, or
clusters, and has the advantage of more properties being documented. Emission-line galaxies were excluded from both samples. We repeat here the exercise for both sets of groups merged
together following the correspondence detailed in Sect. 7.4.1.
– Clad1: these galaxies have low metallicity and in many respects look quite primitive, having a very low dynamical
mass. They are less metallic and more primitive than the
Clad3/C1 ones, which were chosen as the most primeval
group in Fraix-Burnet et al. (2010). Clad1 galaxies could be
the remains of a simple assembly through a monolithic collapse with little dissipation and their somewhat discy nature
probably requires significant feedback and a few perturbations (Benson 2010).
– Clad2: this group shows a steep correlation between log re
and Mdyn . This could be indicative of some merger processes
(e.g. Ciotti et al. 2007) but the galaxies are discy. However,
they also have a high log σ. This suggests that these galaxies
are quite primitive objects similar to those of Clad1 but since
they are more massive, they underwent a more significant
secular evolution, perhaps as in the case of “pseudo-bulges”
(Kormendy & Kennicutt 2004; Benson 2010).
– Clad3/C1: chosen as the most primeval group because of its
low average Mg2 in Fraix-Burnet et al. (2010), this group is
here surpassed by the still lower Mg2 of Clad1. They found
that galaxies of C1 are possibly the remains of a simple assembly through a monolithic collapse with little dissipation,
–
–
and were probably perturbed by interactions. We propose instead that accretion is the main perturbations because the
Clad3 galaxies are small, not very much concentrated, and
have a low log σ.
C2: they were found in Fraix-Burnet et al. (2010) to be less
massive and smaller than the ones of Clad3/C1, and have a
slightly higher Mg2 . They are also somewhat brighter. They
could be the remains of wind stripping of some kind of more
diversified objects because of a strong interaction.
Clad4: this group is very similar to Clad2, but since the
Clad4 galaxies are larger and have a higher Mdyn , they might
have been perturbed by a strong interaction that yielded a
more massive central black hole.
Clad5: being very similar to Clad4 objects, they could be
these discy galaxies seen edge-on but this is statistically untenable because there are three times more Clad5 objects
than Clad4 ones. Since they have a very low D/B, a more
likely explanation is that Clad5 galaxies could be perturbed
versions of Clad4 members. These perturbations are probably mergers since these objects have lost the disciness of
Clad4 objects, but to preserve similar properties, little gas
should be involved, which implies that dry mergers are the
most probable transforming events.
Clad6/C4: three scenarios were proposed for C4 in
Fraix-Burnet et al. (2010): these objects could simply be either galaxies in which star formation has been continuous,
C1 galaxies in which the initially richer gas has not been
swept out, or the remnants of several minor mergers and accretion. The Clad6/C4 galaxies have unexpectedly low values of Mdyn , Brie , and log re , as well as to a lesser extent
log(diam). Otherwise, they do not look odd, so that a continuous star formation with few external perturbations could
also be a reasonable explanation. However, we find that they
are very similar to Clad2, which we proposed to have undergone significant secular evolution, but have a much lower
D/B. This suggests that many interactions, such as harassment (Moore et al. 1996), could be the culprit.
Clad7: because Clad7 includes the groups C3, C5, and C6,
their history might be complex according to Fraix-Burnet
et al. (2010), involving many transforming events (accretions, minor mergers, together with more or less dissipational major mergers). They are probably the remains of both
wet and dry mergers, the most recent ones being of the latter kind. The low value of Hβ, that would indicate that the
last star formation event is relatively ancient, reinforces this
interpretation. They could represent a kind of end state of
galaxy diversification.
Clad8/C7: C7 galaxies were found to be small and very
metallic with a high surface brightness, and to define a tight
FP. They seemed to be associated with the remains of a dissipative (wet) merger, with very little or no dry mergers. They
might also have formed through minor mergers and accretions but the tight FP favors the dissipative wet-merger scenario. The low D/B found here for Clad8/C7 tends to confirm this conclusion. We believe that they may well define
another possible end state for galaxy diversification.
We summarize the above histories in a single cladogram
(Fig. 11), combining the trees obtained in the two studies.
The best way to interpret the evolutionary scenario depicted
in this cladogram is to identify, for each node of the tree, a
particular transforming event that could characterize all groups
related by branches and sub-branches starting from this node
(Fraix-Burnet et al. 2006c). The sequence of nodes downwards
A80, page 13 of 24
A&A 545, A80 (2012)
Clad1
Monolithic collapse
Little dissipation
No perturbation
Clad2
Discy, old stellar population
Secular evolution
Clad3
Monolithic collapse
Little dissipation
Accretion
C2
Windstripped from more
diversified objects
Clad4
BH and size increase
Like Clad2 but with more massive BH
Clad5
Perturbed Clad4
Dry mergers
Clad6
Continuous stellar formation
Harassment
1
2
3
5
4
6
7
8
Clad7
Complex history
Ancient wet merger
More recent dry merger
Clad8
Wet mergers
Fig. 11. Combination of the tree from this paper (Fig. 3) and the one in Fraix-Burnet et al. (2010), showing the proposed assembly history. Numbers
indicated at nodes are referred to in the text.
starting at the upper left of the tree thus defines a sequence of “innovations” that occurred in a common ancestor and were transmitted to all its descendant species. In principle, these innovations are the properties of the galaxies that remain as imprints
transmitted through subsequent transforming processes. Here,
we consider the transforming events as innovations since they
are the origin of the modifications of the properties of galaxies. However, it should be kept in mind that parallel evolutions
(events occurring independently on two different lineages), convergences (different pathways leading to the same parameter
value), and reversals (backward parameter evolution) are probably present, making this exercise currently quite tentative. These
behaviours (called homoplasies) are supposedly not too numerous here since the parcimony optimisation of the cladistic analysis minimizes the occurrence of these types of parameter evolutions. We do not discuss them in this very first attempt, especially
because the properties at hand are few.
We attempt to identify these innovations in Fig. 11 using
the possible histories of each group previously identified. They
appear on the tree when they first occur in the history of the
Universe. It is thus expected that the most basic transforming
events occur very early and the more complex ones later, making the sequence of “innovations” along the tree a representation
of the so-called cosmic evolution.
Finally, we remind the reader that if the level of diversity
goes along the vertical axis in Fig. 11, the horizontal axis, especially the branch lengths, has no particular meaning in this representation. It is important to keep in mind that we are dealing
with present-day galaxies and properties, not those that prevailed
at the time of the transforming event indicated at the node from
which a given branch emerges. Simply speaking, this means that
the galaxies of, say, Clad1, are present-day galaxies that have
A80, page 14 of 24
passively evolved from a less diversified initial state than those
of either Clad6 or Clad7.
The nodes are identified by the numbers in Fig. 11. The corresponding proposed events are:
1.
2.
3.
4.
5.
6.
7.
8.
collapse;
secular evolution;
accretion;
interaction;
“gaseous” interaction;
dry merger;
harassment;
wet mergers.
The first transforming event (node 1), which marked the history of all the galaxies, is most likely monolithic collapse. This
is probably the simplest process to form a self-gravitating ensemble of gas, stars, and dust that we can call a “galaxy”. The
galaxies of Clad1 evolved passively after exhausting their gas
reservoir.
The next three events (nodes 2−4) must have been gentle,
since the discy morphology is well-preserved until Clad4 and
it is now well-established that minor mergers generally preserve
the structure of discy galaxies, while mergers of galaxies of comparable masses generally do not. We first have (node 2) the secular evolution, which is defined as the evolution of a galaxy in
isolation, which is expected to be far more frequent and capable
of significantly modifying the structure and properties of galaxies. At node 3, we must then invoke accretion must be invoked to
increase the masses of galaxies. A more complex event follows,
namely interaction, which is an external perturbation that, with
the wealth of possible impact parameters and galaxy properties,
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
is probably the main driver of galaxy diversity, especially during
the first Gyr of the Universe (Benson 2010).
For node 5 between C2 and Clad4, interaction involving gas
must be invoked to either strip the gas in C2 galaxies by ram
pressure or feed the central black hole in Clad4 objects.
Mergers must be advocated at node 6 since the more diversified groups have lost their disciness. For Clad5 galaxies, the most
probable transforming event is a major merger without much gas
(dry mergers).
A substantial star formation must have occurred in Clad6
galaxies, and several properties indicate repeated perturbations,
implying that harassment is a good candidate for node 7 (Moore
et al. 1996). Harassment is the cumulative effect of high-speed
galaxy encounters, that heats the disc (logσ increases) and favors gas inflow to the galaxy center. This kind of transforming
event acts on a longer timescale than the dry mergers at node 6,
which explains why it appears “later” in the cladogram.
The two most diversified groups, Clad7 and Clad8, are found
to have had complex histories, certainly including wet mergers
(node 8).
Many associated processes, such as feedback and the
quenching of star formation (e.g. Bundy et al. 2006; Benson
2010) are not proposed here because we have concentrated on
more generic events. A significant difficulty of such an exercise
is to identify some properties with transforming events that are
very complex, involving diverse impact parameters and various
chemical, physical, and dynamical processes. We believe that it
is somewhat illusory to associate a particular feature with any
each of these events, and only statistical analyses of simulated
cases could provide average properties that could be compared
to statistical analyses of real objects similar to the one we have
performed here.
9. Conclusion
We have used several multivariate tools, first to select the most
discriminant parameters from the 25 initially available for the
sample of 424 fully documented galaxies of Ogando et al.
(2008), and second to partition the sample into groups. The three
partitioning methods yield similar numbers of groups and similar
composition for each of them, considering that some fuzziness
is expected.
Our first result is that among the initial 23 quantitative parameters available in this study, only 6 are discriminant and actually yield a relatively robust partitioning for this sample. Among
the 10 Lick indices, only 2 ([MgbFe]′ and NaD) are discriminant, together with log σ, Brie , OIII, and D/B.
The global evolutionary scenario found by astrocladistics
gives a very sensible result: galaxies tend to globally become
more metallic, more luminous (more massive), and larger with
increasing diversification. At the same time, they acquire a larger
central velocity dispersion, which is often related to the mass
increase, and NaD also increases along with both mass and velocity dispersion, as expected. These are global statistical trends
that are explained by general basic physical and chemical processes as a function of time since the Big Bang.
As a consequence, the many properties of galaxies that are
bound to evolve on average with galaxy diversification explain
the several evolutionary correlations found in this paper. In particular, we have confirmed the evolutionary nature of the Mg2
vs. log σ correlation found by Fraix-Burnet et al. (2010) for a
different sample and using a different set of parameters. Rather
interestingly, we have also found some correlations that are specific only to some groups. These can be attributed to either a
direct physical cause or a confounding factor specific to some
groups (such as the epoch of the last merger, the level of dissipation, the number of accretion events, the impact parameters, or
the number of mergers).
One of the most important results of our work is that the
structures defined by the partitioning, when projected onto either the fundamental plane (log σ vs. Brie ) or the log re vs. Mdyn
diagram, are very similar to those found by Fraix-Burnet et al.
(2010) for a totally distinct sample with different parameters.
The fundamental plane of early-type galaxies appears to
be very probably generated by diversification. In support of
this, we have also found another “fundamental plane”, a
three-dimensional correlation between Hβ, Kabs , and log(diam).
All scatter plots are basically simple projections onto a subparameter space of the partitioning established in the sixparameter space. Thus, there is less information in either these
scatter plots or “fundamental planes” than in the multivariate
partitioning and the evolutionary tree obtained with cladistics.
Another important result is that six parameters – no fewer
– are needed to describe the diversity of this sample. The three
parameters of the fundamental plane (log σ, Brie , and log re ),
plus the index Mg2 , have not yielded as robust a partitioning
here, although they did in a previous study for a distinct sample
(Fraix-Burnet et al. 2010). We argue that there is no contradiction here, because three discriminant parameters extracted in the
present paper are present in our previous study (log σ, Brie , Mg2
being replaced by [MgbFe]′). The multivariate analyses generally depend on both the objects in the sample and their initial
set of descriptors, both of which are different in the two studies. Nevertheless, as a consequence, similar analyses will have
to be conducted on other samples with other descriptors, since
our fairly small sample of nearby galaxies cannot represent the
diversity of galaxies throughout the Universe, and the available
parameters are here restricted to the visible domain.
We have combined the results of Fraix-Burnet et al. (2010)
and those in the present paper on a single cladogram, showing
the possible assembly history for each group. From this cladogram, we have attempted to identify the transforming events
that are at the origin of galaxy diversification. The transforming events that we have indicated as “innovations” are tentative,
because the information at hand is insufficient to identify them
with certainty. These proposed events show that the use of sophisticated statistical tools yields a very sensible classification.
Figure 11 is the basis of an explanatory classification linking
the objects to the fundamental transforming processes, i.e. to the
physics, rather than a descriptive classification adopted in most
current classifications of galaxies. In this respect, we note that
the Edwin Hubble classification is of the latter type, being based
on morphology, while his tuning fork diagram (often called the
Hubble sequence nowadays) is explanatory since it indicates the
links between the classes. Nearly one century later, we know that
galaxies are characterized by much more than only their morphology, so that we need to generalize the Hubble diagram to a
multivariate picture of galaxy diversification. Figure 11, which
we have produced using cladistics, is one step in this direction.
Hence, increasing both the sample size and the number of
descriptors is an absolute requirement. The six-parameter space
needed to describe the diversity of the sample used in the present
paper is probably a minimum space because of the complexity
of galaxies and their assembly history. The nature of the discriminant parameters might also change with the input of more observables. In addition, the number of groups and their boundaries
will certainly change. This is a double quest: classifying galaxies
into objectively established evolutionary and intelligible groups,
A80, page 15 of 24
A&A 545, A80 (2012)
and finding the parameter space in which these groups can be
identified. This quest is necessarily progressive, and will probably never end. However, one can hope that some convergence
will be reached.
A limitation of the present work is that cladistics cannot be
applied directly to very large samples as the necessary computer time would be prohibitively excessive. However, once the
most discriminant parameters are identified, it will be possible
to repeat the cladistic analysis for many subsamples, and subsequently combine the trees to define classes of galaxies. The
ultimate goal is to gather the huge number of galaxies in the
Universe into a tractable number of groups and establish the corresponding evolutionary relationships.
Acknowledgements. This research has made use of the HyperLeda database
(http://leda.univ-lyon1.fr) and the NASA/IPAC Extragalactic Database
(NED) which is operated by the Jet Propulsion Laboratory, California Institute
of Technology, under contract with the National Aeronautics and Space
Administration.
References
Alonso, M. V., Bernardi, M., da Costa, L. N., et al. 2003, AJ, 125, 2307
Babu, G. J., Chattopadhyay, T., Chattopadhyay, A. K., & Mondal, S. 2009, ApJ,
700, 1768
Benson, A. J. 2010, Phys. Rep., 495, 33
Bernardi, M., Alonso, M. V., da Costa, L. N., et al. 2002, AJ, 123, 2159
Bundy, K., Ellis, R. S., Conselice, C. J., et al. 2006, ApJ, 651, 120
Burstein, D., Faber, S. M., Gaskell, C. M., & Krumm, N. 1984, ApJ, 287, 586
Cabanac, R. A., de Lapparent, A., & Hickson, P. 2002, A&A, 389, 1090
Cappellari, M., Emsellem, E., Krajnović, D., et al. 2011, MNRAS, 416, 1680
Chattopadhyay, T., & Chattopadhyay, A. 2006, AJ, 131, 2452
Chattopadhyay, T., & Chattopadhyay, A. 2007, A&A, 472, 131
Chattopadhyay, T., Misra, R., Naskar, M., & Chattopadhyay, A. 2007, ApJ, 667,
1017
Chattopadhyay, T., Mondal, S., & Chattopadhyay, A. 2008, ApJ, 683, 172
Chattopadhyay, A., Chattopadhyay, T., Davoust, E., Mondal, S., & Sharina, M.
2009a, ApJ, 705, 1533
Chattopadhyay, T., Babu, J., Chattopadhyay, A., & Mondal, S. 2009b, ApJ, 700,
1768
Chattopadhyay, T., Sharina, M., & Karmakar, P. 2010, ApJ, 724, 678
Ciotti, L., Lanzoni, B., & Volonteri, M. 2007, ApJ, 658, 65
Ellis, S. C., Driver, S. P., Allen, P. D., et al. 2005, MNRAS, 363, 1257
Fraix-Burnet, D. 2011, MNRAS, 416, L36
Fraix-Burnet, D., Choler, P., & Douzery, E. 2006a, A&A, 455, 845
Fraix-Burnet, D., Choler, P., Douzery, E., & Verhamme, A. 2006b, J. Class., 23,
31
Fraix-Burnet, D., Douzery, E., Choler, P., & Verhamme, A. 2006c, J. Class., 23,
57
Fraix-Burnet, D., Davoust, E., & Charbonnel, C. 2009, MNRAS, 398, 1706
Fraix-Burnet, D., Dugué, M., Chattopadhyay, T., Chattopadhyay, A. K., &
Davoust, E. 2010, MNRAS, 407, 2207
Gonzáles, J. 1993, Ph.D. Thesis, University of California
Greggio, L. 1997, MNRAS, 285, 151
Hopkins, P. F., Cox, T. J., & Hernquist, L. 2008, ApJ, 689, 17
Hudson, M. J., Lucey, J. R., Smith, R. J., Schlegel, D. J., & Davies, R. L. 2001,
MNRAS, 327, 265
Karachentsev, I. D., Sharina, M. E., Dolphin, A. E., et al. 2002, A&A, 385, 21
Kormendy, J., & Kennicutt, Jr., R. C. 2004, ARA&A, 42, 603
Kuntschner, H., Emsellem, E., Bacon, R., et al. 2010, MNRAS, 408, 97
MacQueen, J. 1967, Fifth Berkeley Symp. Math. Statist. Prob., 1, 281
Maddison, W. P., & Maddison, D. R. 2004, Mesquite: a Modular System for
Evolutionary Analysis
Matković, A., Guzmán, R., Sánchez-Blázquez, P., et al. 2009, ApJ, 691, 1862
Mei, S., Blakeslee, J. P., Côté, P., et al. 2007, ApJ, 655, 144
Milligan, G. W. 1980, Psychometrika, 45, 325
Moore, B., Katz, N., Lake, G., Dressler, A., & Oemler, A. 1996, Nature, 379,
613
Murtagh, F., & Heck, A. 1987, Multivariate Data Analysis (Dordrecht: Reidel)
Nigoche-Netro, A., Aguerri, J. A. L., Lagos, P., et al. 2011, A&A, 534, A61
Ogando, R. L. C., Maia, M. A. G., Pellegrini, P. S., & da Costa, L. N. 2008, AJ,
135, 2424
Perrett, K. M., Hanes, D. A., Butterworth, S. T., et al. 1997 AJ, 113, 895
Recio-Blanco, A., Aparicio, A., Piotto, G., De Angeli, F., & Djorgovski, S. 2006,
A&A, 452
Robertson, B., Cox, T. J., Hernquist, L., et al. 2006, ApJ, 641, 21
Sánchez Almeida, J., Aguerri, J. A. L., Muñoz-Tuñón, C., & de Vicente, A. 2010,
ApJ, 714, 487
Serra, P., & Trager, S. C. MNRAS, 374, 769
Sugar, A., & James, G. 2003, JASA, 98, 750
Swofford, D. L. 2003, PAUP*: Phylogenetic Analysis Using Parsimony (*and
Other Methods)
Thomas, D., Maraston, C., & Bender, R. 2003, MNRAS, 339, 897
Thomas, D., Maraston, C., Bender, R., & Mendes de Oliveira, C. 2005, ApJ,
621, 673
Thomas, D., Maraston, C., & Johansson, J. 2011, MNRAS, 412, 2183
Thuillard, M. 2007, Evolutionary Bioinformatics, 3, 267
Thuillard, M. 2008, Evolutionary Bioinformatics, 4, 237
Thuillard, M., & Fraix-Burnet, D. 2009, Evolutionary Bioinformatics, 5, 33
Thuillard, M., & Moulton, V. 2011, Bioinformatics and Computational Biology,
9, 453
Tripicco, M. J., & Bell, R. A. 1995, AJ, 110, 3035
Whitmore, B. C. 1984, ApJ, 278, 61
Worthey, G., & Ottaviani, D. L. 1997, ApJS, 111, 377
Worthey, G., Faber, S. M., González, J. J., & Burstein, D. 1994, ApJS, 94, 687
Pages 17 to 24 are available in the electronic edition of the journal at http://www.aanda.org
A80, page 16 of 24
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
Appendix A: Methods
A.1. Principal component analysis
Principal component analysis (PCA) is well-known to astronomers. It is not a partitioning method: its aim is instead to
reduce the dimensionality of the parameter space. From the correlation matrix, PCA builds eigenvectors (the principal components) that are orthogonal and linear combinations of the physical parameters. These eigenvectors usually have no physical
meaning. In general, most of the variance of the sample can be
represented with only a few principal components (those having an eigenvalue greater than 1). They thus give a simpler representation of the data by eliminating the correlations between
physical parameters. Strongly correlated parameters are gathered in the same eigenvector, and the most important parameters
(with respect to variance) are the ones with the highest coefficient (loading) in each eigenvector. The physical interpretation
must be made back in the real parameter space.
PCA is thus very efficient at reducing the parameter space to
supposedly uncorrelated components and helps in detecting the
most discriminant or discriminating parameters. The number of
significant eigenvectors gives an idea of the number of parameters necessary to describe the sample. Principal components can
also be used for subsequent cluster or cladistic analyses.
There is however a caveat to be kept in mind. PCA eliminates all correlations, regardless of whether they are causal. It is
extremely useful to remove any redundancies, as well as physical
correlations between two parameters indicating the same underlying process. However, PCA also removes evolutionary correlations (which are called “spurious” or confounding in statistics,
Fraix-Burnet 2011), for instance between two parameters that
are independent but vary with time. The log σ-Mg2 correlation
for early-type galaxies (see Fraix-Burnet et al. 2010) is a good
example. Such independent evolutions are lost through the PCA
reduction of dimensionality.
A.2. Minimum contradiction analysis
Partitioning objects consists in producing some order. In some
cases, i.e. in either hierarchical clustering or cladistics, the arrangement of the objects can be represented on a tree. A tree is
a graph representing the objects as the leaves with a unique path
between any two vertices. A bifurcating tree has internal vertices
that all have a degree of at most 3 (at most 3 branches connect to
any such vertex).
By indexing circularly all the leaves of a planar representation of a weighted binary tree, one obtains a perfect order,
meaning that the corresponding ordered distance-matrix fulfills
all Kalmanson inequalities. Generally speaking, the Kalmanson
inequalities are fulfilled if the ordered distance matrix corresponds to a weighted binary tree or a superposition of binary
trees (Thuillard & Moulton 2011). The difference between the
perfect order and the order one obtains with a given dataset
is called the contradiction. The minimum contradiction corresponds to the best order one can get.
The minimum contradiction analysis (Thuillard 2007, 2008,
MCA) finds this best order. It is a powerful tool for ascertaining
whether the parameters can lead to a tree-like arrangement of the
objects (Thuillard & Fraix-Burnet 2009). Using the parameters
that fulfil this property, the method then performs an optimisation of the order and provides groupings with an assessment of
their robustness.
For taxa indexed according to a circular order, the distance
matrix, which is defined to be
Yi,n j =
1
(di,n + d j,n − di, j )
2
fulfils the so-called Kalmanson inequalities (Kalmanson 1975):
n
n
, Yk, j n ≥ Yk,i
(i ≤ j ≤ k)
Yi,n j ≥ Yi,k
(A.1)
where di, j is the pairwise distance between taxon i and j. The
matrix element Yi,n j is the distance between a reference node n
and the path i- j. The diagonal elements Yi,in = di,n correspond
to the pairwise distance between the reference node n and the
taxon i.
The contradiction on the order of the taxa can be defined as
2
2
n
n
C=
− Yi,n j , 0 +
− Y nj,k , 0
max Yi,k
max Yi,k
k> j≥i
k≥ j>i
(A.2)
for any i, j, k n. The best order of a distance matrix is, by definition, the order minimizing the contradiction (Thuillard 2007,
2008). Thuillard & Fraix-Burnet (2009) showed that the perfect
order is linked to the convexity of the variables in the parameter space, and is obtained for specific properties of the variables along the order. It is then possible to detect the discriminant potentiality of the variables. This is exactly what is done in
Sect. 3.2.
A.3. Cladistic analysis
Cladistics seeks to establish evolutionary relationships between
objects. It is a non-parametric character-based phylogenetic
method, also called a maximum parsimony method. It does not
use distances, because there is no assumption about the metrics of the parameter space. The “characters” are instead traits,
descriptors, observables, or properties, which can be assigned
at least two states characterizing the evolutionary stage of the
objects for that character. The use of this approach in astrophysics is known as astrocladistics (for details and applications,
see Fraix-Burnet et al. 2006b,c, 2009, 2010). Simply speaking,
the characters here are the parameters, the (continuous) values
of which supposedly evolve with the level of diversification of
the objects. The maximum parsimony algorithm looks for the
simplest arrangement of objects on a bifurcating tree. The complexity of the arrangement is measured by the total number of
“steps” (i.e. changes in all parameter values) along the tree.
The success of a cladistic analysis much depends on the behaviour of the parameters. In particular, it is sensitive to redundancies, incompatibilities, too much variability (reversals), and
parallel and convergent evolutions. It is thus a very good tool
for investigating whether a given set of parameters can lead to a
robust and pertinent diversification scenario.
In the present study, we used the same kind of analysis as
in our previous papers on astrocladistics. We discretized the parameters into 30 equal-width bins, which play the role of discrete
evolutionary states. This choice of 30 bins is justified by a fair
representation of diversity, a stability of the analysis in the sense
that the result does not depend on the number of bins, and a bin
width roughly corresponding to the typical order of magnitude of
the uncertainties (i.e. 7%, see Fraix-Burnet et al. 2009). We also
adopted the parsimony criterion, which consists in finding the
simplest evolutionary scenario that can be represented on a tree.
Our maximum parsimony searches were performed using the
A80, page 17 of 24
A&A 545, A80 (2012)
Table A.1. Loadings on the eight principal components of the PCA analysis made on the set of 23 parameters (see Sect. 3.1).
Mgb
logs
Mg2
mgbfe
NaD
Mg1
Kabs
mabs
ldiam
mgb.fe
logre
Fe5335
hbeta
Fe5406
Fe5270
H.K
Fe5015
D.B
B.R
Brie
Fe5709
OIII
J.H
Comp1
–0.9143
–0.9067
–0.8983
–0.8879
–0.8659
–0.8470
0.7780
0.7255
–0.7248
–0.6742
–0.6512
–0.5455
0.5389
–0.4906
–0.4900
–0.3065
–0.2763
0.2024
–0.1338
–0.0673
0.0602
–0.0311
–0.0175
Comp2
–0.122395
–0.012224
–0.158860
–0.303358
–0.099002
–0.132117
–0.357711
–0.359334
0.409603
0.245147
0.570850
–0.406391
–0.223947
–0.465277
–0.522075
0.117644
–0.550090
0.134962
0.363115
0.614240
–0.172935
–0.395042
–0.000856
Comp3
0.1594
0.0674
0.1190
0.0656
0.0527
0.1739
0.1878
0.2095
–0.2557
0.2938
–0.3297
–0.1312
–0.4974
–0.1337
–0.1240
0.0945
–0.5230
–0.3041
–0.2925
–0.4327
–0.1399
–0.6303
–0.0977
Comp4
–0.1593
0.0950
–0.1121
–0.1870
0.0657
–0.0977
–0.3577
–0.4395
0.3327
–0.0296
–0.2541
–0.2574
0.2296
–0.0838
–0.1365
0.2266
0.1094
–0.1834
–0.4631
–0.4875
–0.3042
0.1494
0.3589
heuristic algorithm implemented in the PAUP*4.0b10 (Swofford
2003) package, with the Multi-Batch Paup Ratchet method3 .
The results were interpreted with the help of the Mesquite software (Maddison & Maddison 2004) and the R-package (used for
graphics and statistical analyses).
Making cladistic analyses with different sets of parameters
both helps to find the most robust result and gives interesting
information on the behaviour of the parameters themselves. The
robustness of cladograms is always difficult to assess objectively,
so we use a criterion similar to that of other statistical distance
analyses: if a similar result is found by using different conditions
or methods, then it can be considered as reasonably robust. We
applied four possible tests here:
1. The occurrence of a branching pattern among most parsimonious trees: with so few parameters, many equally parsimonious trees are found, often arbitrarily limited to 1000. The
majority-rule consensus of all of them yields a percentage
of occurrence for each node. The higher this percentage, the
higher the probability that this node is “robust”.
2. The agreement of branching patterns between sub-sample
analyses, which can be called “internal consistency”: by
making analyses of several sets of arbitrarily selected subsamples, we can check whether a given pattern is present on
trees found with larger samples, including the full tree.
3. The comparison between different sets of parameters: any
result should preferably not depend too much on a single
parameter. Adding or removing a parameter should not drastically change the tree.
4. A comparison with the results of a cluster analysis: distancebased methods are totally independent, so any agreement can
instill us a fair confidence in the result.
Since we have many more objects than parameters, a lot of
“flying” objects are expected between different analyses, and
the above tests should be done with statistics in mind. The
first test is always positive in this study: percentages are higher
3
http://mathbio.sas.upenn.edu/mbpr
A80, page 18 of 24
Comp5
–0.1215
–0.0443
–0.1455
0.0543
–0.0496
–0.1787
–0.1884
–0.2075
0.1887
–0.3977
–0.0686
0.3127
0.0104
0.1044
0.2660
0.4477
–0.2198
0.2342
–0.1578
–0.0751
0.4027
–0.3828
–0.3548
Comp6
0.04717
0.02837
0.05519
–0.04251
–0.00899
0.01250
0.00503
0.06904
–0.09360
0.17776
–0.01075
–0.11438
0.18101
–0.13418
–0.18686
0.48040
0.17362
–0.15225
0.13444
–0.08047
–0.14195
0.25483
–0.69952
Comp7
0.11541
–0.05006
0.07249
0.00546
–0.00360
0.06330
0.04219
–0.06490
0.02929
0.26559
–0.01336
–0.13735
–0.01882
0.01741
–0.15191
–0.20626
0.05076
0.66215
–0.51784
0.03262
0.06450
0.08224
–0.23475
Comp8
0.03549
0.04723
–0.00744
–0.04366
–0.03957
–0.05674
–0.10201
–0.11826
0.06032
0.15522
–0.01325
–0.09773
–0.00539
–0.11408
–0.15975
–0.28296
–0.00969
–0.38279
–0.06306
–0.07633
0.74089
0.13941
–0.11213
than 70−75%, and most often they are above 95%. This is already an indication that some structure is present in the data.
The other three tests are described below.
The full sample of 424 galaxies was divided into three subsamples with 105 objects each and a fourth one with 109 objects. The first and fourth subsamples were found to belong exclusively to clusters 1 and 3, respectively, of the cluster analysis.
The diversity in the first subsample is less than for the others,
so that the resulting tree is generally less well-resolved. The two
first subsamples were also gathered to form a 210-object subsample, as well as the two last ones that form a 214-object subsample. Analyses were performed with these six subsamples, as
well as the full sample. We then estimated the internal consistency by comparing the seven trees two by two and by eye (with
the help of the program cophyloplot in the R-package, which
connects a given object between the two trees).
This procedure was applied to each of the eight-parameter
subsets given in Table 1. Subsets 5c, 5cA, 6cA, and 3c show
a rather good internal consistency, 4c, 7c, and 6c that which is
fairly good, and finally 8c and 10c that which is not so good.
This already shows that the optimal number of parameters is
around 5, 6, or at most 7. This is in excellent agreement with the
PCA analysis (Sect. A.1).
If we compare the trees obtained with the full sample for the
eight-parameter subsets, we find that subset 5c is very consistent
with 6c, 7c, and 8c. In addition, 5c, 6c, 5cA, and 6cA are in good
mutual agreement, while this is not the case for 6c, 7c, and 8c.
In Table A.2, we show for each tree the Rescaled
Consistency Index (RCI), which measures the fitness of a parameter on the phylogeny depicted by the tree. The higher the
RCI (indeed the closer it is to 1), the more discriminant the parameter. In other words, parameters with higher RCI are the most
responsible for the structure of the tree. The absolute value depends on the number of objects and parameters, so it cannot be
used to compare trees obtained with different data. Here, we can
only use it to compare parameters for a given tree. In Table A.2,
the parameters are ordered according to RCI.
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
Table A.2. Fitness of parameters on the cladograms obtained for each subset as represented by the Rescaled Consistency Index (RCI).
Subset
4cA
5c
5cA
6c
6cA
7c
8c
Order from RCI
D/B log σ [MgbFe]′ NaD
D/B log σ NaD [MgbFe]′ Brie
[MgbFe]′ Mg b log σ D/B NaD
log σ D/B NaD [MgbFe]′ Brie OIII
[MgbFe]′ Mg b log σ D/B NaD Brie
D/B log σ OIII NaD [MgbFe]′ Fe5015 Brie
Mg b [MgbFe]′ log σ NaD D/B OIII Fe5015
Brie
[MgbFe]′ Mg b NaD log σ D/B Fe5270 Brie
B-R OIII Fe5709
10c
RCI
0.102 0.086 0.086 0.075
0.077 0.063 0.063 0.059 0.051
0.098 0.090 0.080 0.075 0.066
0.059 0.055 0.055 0.050 0.040 0.039
0.076 0.073 0.061 0.060 0.054 0.043
0.053 0.051 0.044 0.041 0.040 0.037 0.031
0.055 0.052 0.050 0.044 0.041 0.038 0.033
0.030
0.075 0.055 0.044 0.042 0.034 0.033 0.025
0.025 0.023 0.020
A.4. Cluster analysis
5.5
Jk
4.5
3.5
2.5
2
4
6
8
10
K
3.0
Jk
2.5
2.0
1.5
1.0
0
2
4
6
8
10
K
Fig. A.1. Plots showing the jumps as defined in Sect. A.4. Top: jumps
for the PCA+CA analysis (Sect. 4.1). Bottom: jumps for the cluster
analysis with the six parameters (Sect. 4.2).
When Mg b and [MgbFe]′ are present together in a subset,
they dominate the shape of the tree (sets 5cA, 6cA, 8c, and 10c),
log σ and D/B being right after them. Mg b and [MgbFe]′ are
obviously redundant because they are very well-correlated and
are more or less the same measure. Hence, they cannot be used
simultaneously in the cladistic analysis, and the trees that we find
are more linear than the others. In contrast, log σ and D/B are
not at all correlated, but are always together, and dominate the
tree shape when Mg b is not present together with [MgbFe]′. In
addition, NaD is very discriminant, and only roughly correlated
with log σ and [MgbFe]′.
If we compare the clusters obtained with the clustering analysis, the agreement decreases roughly for 6c, 7c, 5A, 3c, 5c, 4cA,
8c, and 10c, the winner being undoubtedly 6c. The corresponding tree with the groups is shown in Fig. 3.
In the present study, we adopted K-means partitioning algorithm
of clustering following MacQueen (1967). This method constructs K clusters using a distance measure (here Euclidean).
The data are classified into K groups around K centres, such
that the distance of a member object of any particular cluster
(group) from its centre is minimal compared to its distance from
the centres of the remaining groups. The requirement for the algorithm is that each group must contain at least one object and
each object must belong to exactly one group, so there are at
most as many groups as there are objects. Partitioning methods are applied (Whitmore 1984; Murtagh 1987; Chattopadhyay
& Chattopadhyay 2006, 2007; Babu et al. 2009; Chattopadhyay
et al. 2009a; Chattopadhyay et al. 2010), if one wishes to classify the objects into K clusters where K is fixed. Cluster centres
were chosen based on a group average method, which ensures
that the process is almost robust (Milligan 1980).
To achieve an optimum choice of K, the algorithm is run
for K = 2, 3, 4, etc. For each value of K, the value of a distance measure dK (called the distortion) is computed as dK =
(1/p) minx E[(xK − cK )′ (xK − cK )], which is defined as the distance of the xK vector (values of the parameters) from the centre
cK where p is the order of the xK vector. If dK′ is the estimate
of dK at the Kth point, then the optimum number of clusters is
′−p/2
determined by the sharp jump in the curve JK = (dK′−p/2 − dK−1
)
vs. K (Sugar & James 2003). The jumps as a function of K for
our PCA+CA and CA analyses are shown in Fig. A.1.
Appendix B: Analysis with log σ, log r e , Bri e
and Mg2 , and error bars
B.1. Analysis with logσ, log re , Brie and Mg2
We complemented the study presented in this paper with
the analysis of our sample with the four parameters (log re ,
log σ, Brie , and Mg2 ) as in Fraix-Burnet et al. (2010). We
used the same three multivariate techniques (cluster analysis,
Miminimum Contradiction Analysis, and cladistics) as presented in Sect. 2.2 and Appendix A.
The resulting tree is less structured (more galaxies lie on individual branches) than the one obtained in the present paper using six parameters. This can be explained by log re and Mg2 not
having been found to be discriminant parameters for the considered sample. It is also less structured than in Fraix-Burnet et al.
(2010) which uses the same four parameters, which is probably
due to the problems in determining of log re .
To summarize the results, we show the projection of the three
trees – the one obtained in this paper with six parameters, the one
obtained here with four parameters, and the one of Fraix-Burnet
et al. (2010) – onto the fundamental plane (log σ vs. Brie )
A80, page 19 of 24
2.4
2.2
2.0
logσ
2.0
logσ
2.2
2.4
A&A 545, A80 (2012)
6 parameters
1.8
4 parameters
1.8
4 parameters
18.0
19.0
20.0
21.0
Fraix−Burnet et al 2010
18.0
19.0
21.0
20.0
21.0
2.2
2.0
logσ
2.0
logσ
2.2
2.4
Brie
2.4
Brie
20.0
6 parameters
6 parameters
1.8
1.8
4 parameters
Fraix−Burnet et al 2010
18.0
19.0
20.0
21.0
Brie
Fraix−Burnet et al 2010
18.0
19.0
Brie
Fig. B.1. Projection of the trees onto the fundamental plane for three cases: the analysis of this Appendix B and the one by Fraix-Burnet et al.
(2010) both using the four parameters of the fundamental plane, and the principal study of the present paper with six parameters. Thick lines
represent the “trunk” of the trees, while the small branches relate the trunks to the mean of each group. For clarity, results are compared two by
two, and only the trunks are shown for the three studies on the lower right diagram. These are evolutionary tracks in the sense of diversification,
and not the path of evolution for a single galaxy.
without the data points (Fig. B.1). Globally, there is good agreement and the groupings are consistent. However, the projected
tree from the present Appendix departs from the other two in the
top half of the figure. This is because this tree is less structured
than the others, so that instead of having one or two groups at
this level, there is a sequence of single branches that makes the
trunk of the tree to “follow” more closely individual objects.
B.2. Influence of re and error bars on the partitioning
The effective radius log re in our sample is recomputed through a
statistical relation between the linear diameter of the galaxy (Dn )
and its velocity dispersion (σ), which was determined in another
paper (Bernardi et al. 2002). The reason given by Ogando et al.
(2008) is that, due to the very low redshift of the galaxies in
the sample, “the conversion of re in arcseconds to kpc needs a
reliable determination of the galaxy distance (D). Considering
just the redshift to calculate D, we may incur in error due to
the peculiar motion of galaxies. Thus, we adopted D given by
A80, page 20 of 24
the Dn vs. σ relation (Bernardi et al. 2002) to calculate re in
kpc.” However, this relation was obtained with some assumptions (such as the identical properties of galaxies in several clusters) and introduces a dependence of log re (through D) on log σ.
The two radii (Fig. B.2) are quite well-correlated with each
other, but the dispersion is relatively large. We performed two
cladistic analyses with the four parameters of the fundamental
plane (log re , log σ, Brie , and Mg2 ) as above using the two determinations of the effective radius. The agreement between the two
results is only fair. This can be explained by the relatively important discrepancy between the two different values of re (median
difference of 10%). This however is similar to the uncertainty in
log re , but much larger thn for the other parameters. In addition,
the radius or dimension of galaxies does not appear as a discriminant parameter in the study presented in this paper. Hence, it is
not so surprising that analyses using this parameter are not very
stable.
We now consider the robustness of our clustering result for
the six-parameter analysis when taking error measurements into
0.25
error
0.05
0.15
0.20
error
0.00
1.0
1.8
2.0
2.2
2.4
2.6
0.0
logσ
0.5
1.0
logre
Fig. B.2. Correspondence between the effective radius computed in two
separate ways.
account. It is statistically a very challenging task to assess the
influence of the errors. However, cladistics can easily take into
account the error bars since the optimisation criterion in all analyses performed so far in astrocladistics use the parsimony criterion: among all the possible arrangements of the objects on trees,
the simplest evolutionary scenario is retained. The parcimony is
measured by using the number of “steps”, that is the total number of changes in parameter values along all the branches of the
tree. If a missing value or an uncertain one (given by a range
of values) is included in the data matrix, all possible values are
considered and the ones corresponding to the simplest tree is
favored. This simply increases the number of possible cases to
consider. We note that all possible values within the range alllowed by measurement uncertainties are given the same weight,
whereas the probability distribution is generally expected to be
higher at the central value (ideally Gaussian).
We performed a cladistic analysis similar to that in Fig. 3
using the error bars given in Ogando et al. (2008) and Alonso
et al. (2003) for log σ and Brie , and for D/B we considered the
error given for log re in Alonso et al. (2003), There errors are
shown in Fig. B.3. For NaD, [MgbFe]′ , and OIII, we assumed a
face value of 10%, which is the upper limit estimated by Ogando
et al. (2008) for all the Lick index values.
The resulting tree shown in Fig. B.4 is slightly less structured
than the one in Fig. 3 but most groups are grossly preserved.
Clad3 appears to be mixed with Clad1 and Clad5 to be mixed
with Clad6. In addition, Clad7 and Clad8 are somewhat mixed
with each other. Interestingly, these behaviours are similar
to those inferred from the comparison with the partitioning
derived from the cluster analysis. In addition, the agreement is
quite satisfactory given the large uncertainties for half of the
0.04
error
1.0
logre
0.02
0.00
0.5
0.1 0.2 0.3 0.4
error
0.0
0.06
0.5
0.0
logre (H0)
0.10
1.5
0.30
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
18
19
20
µe
21
22
23
0.10
0.20
0.30
0.40
Mg2
Fig. B.3. Errors in log σ, Brie , and log re (taken for the errors in D/B).
ESO235G051
ESO440G038
ESO447G031
NGC1172
NGC1315
IC5267B
ESO545G042
IC2764
ESO384G023
NGC7404
NGC3056
NGC0516
NGC1331
NGC3522
ESO286G050
ESO358G059
NGC7351
NGC4694
NGC4489
NGC3412
NGC6172
NGC3599
NGC4612
IC2200A
ESO443G039
NGC4685
NGC5726
ESO386G038
NGC0125
ESO576G030
NGC2211
ESO378G020
ESO507G014
NGC4997
NGC1427
ESO500G018
NGC0790
ESO437G021
NGC3051
NGC3300
NGC0277
NGC3617
IC1445
ESO501G025
ESO498G004
NGC2888
NGC0442
NGC7302
ESO499G013
NGC5516
NGC2679
IC4913
ESO384G021
NGC7280
ESO490G006
NGC1393
NGC7079
IC2552
NGC4215
NGC1394
NGC2865
IC1864
ESO437G038
ESO231G017
NGC3564
NGC5424
NGC3892
ESO437G045
ESO533G025
NGC3335
IC4796
UGC04228
ESO316G046
NGC0502
ESO384G019
MCG0103049
NGC7029
NGC0273
NGC1339
NGC1374
ESO423G024
NGC4855
NGC1403
NGC2729
NGC6706
NGC2723
NGC3332
NGC5928
MCG0213009
NGC0193
NGC0968
IC0642
IC0555
NGC5017
NGC5831
NGC0842
NGC1567
NGC3483
NGC1389
ESO507G027
ESO387G016
ESO428G011
NGC2502
MCG0532074
IC2526
NGC4958
NGC3085
NGC5858
NGC5049
NGC5761
NGC2984
NGC1383
NGC3954
NGC5330
ESO263G033
NGC3224
NGC4339
NGC4270
NGC4551
NGC0448
ESO510G054
NGC2267
NGC7600
ESO283G020
ESO139G005
IC1609
NGC0155
NGC2945
UGC04670
NGC3325
NGC3591
NGC0862
IC2587
IC1858
NGC5304
MCG0157004
NGC0474
ESO341G013
IC4943
NGC4379
NGC4191
NGC4993
NGC4683
NGC3605
UGC12840
NGC4603C
ESO242G014
NGC1993
ESO323G092
NGC0113
NGC2716
NGC0830
ESO579G017
NGC6017
NGC7576
UGC10864
UGC09288
ESO058G019
NGC1979
NGC7145
ESO182G007
NGC6725
NGC2106
IC1729
ESO563G024
NGC4550
NGC2891
ESO376G009
ESO447G030
ESO221G037
NGC2191
NGC4417
NGC7359
NGC0596
IC0999
NGC1381
NGC5343
NGC4033
NGC4240
NGC4515
NGC1419
NGC1379
ESO382G034
NGC0349
NGC4239
NGC4474
NGC1412
NGC5583
ESO317G021
ESO548G033
NGC6909
ESO507G024
IC4180
NGC1366
NGC0770
IC1317
ESO569G012
ESO437G015
NGC5770
NGC4739
NGC4318
ESO489G037
UGC01169
NGC7365
NGC7185
ESO567G052
NGC1336
NGC6924
NGC0774
NGC3096
NGC0599
UGC05226
NGC1298
NGC1426
ESO084G028
NGC4830
UGC08872
ESO501G056
NGC4645A
ESO436G027
ESO282G028
ESO501G003
IC0232
NGC0686
NGC1162
IC5328
NGC0875
NGC2983
NGC3598
NGC3316
NGC4373A
NGC1199
NGC5397
NGC2918
NGC0936
NGC5493
NGC7805
ESO221G020
ESO486G019
ESO417G021
ESO425G014
NGC1574
ESO498G006
ESO489G035
NGC5507
ESO377G029
ESO491G006
NGC5193
IC0552
IC0090
NGC5628
IC0513
IC0164
IC4421
NGC5203
NGC7075
NGC3904
NGC7097
NGC3273
NGC0430
IC3896
NGC1653
NGC3585
NGC3546
NGC3615
NGC0050
ESO316G034
NGC2749
ESO494G031
ESO466G026
ESO282G024
IC5157
NGC4714
NGC6375
ESO322G051
NGC4404
NGC3082
MCG0207002
IC4310
MCG0135007
NGC2272
NGC3282
NGC0809
NGC0682
NGC2073
ESO367G008
NGC3302
NGC7144
NGC3768
NGC0641
NGC6364
NGC6442
NGC0969
NGC3308
NGC5903
IC1860
NGC1713
NGC1132
IC1492
NGC5626
NGC7626
NGC0163
NGC3022
NGC6587
UGC07813
NGC0541
IC3289
NGC4730
NGC1439
NGC1297
IC0494
NGC3100
NGC3142
NGC6548
ESO445G049
NGC1200
ESO137G024
NGC5111
IC4350
NGC5813
NGC1549
IC4329
NGC4553
NGC6903
NGC1453
NGC0584
UGC11082
NGC3778
NGC0636
ESO510G071
UGC12515
ESO124G014
NGC1201
NGC1380
NGC1461
NGC7176
ESO509G108
ESO508G008
ESO445G028
NGC4645B
NGC7173
ESO443G053
NGC1521
NGC4697
ESO141G003
ESO208G021
NGC4078
NGC3305
NGC7785
NGC6851
IC0395
NGC7562
ESO425G019
NGC7778
NGC6495
NGC5898
NGC1553
ESO507G021
NGC4233
NGC4756
ESO282G010
NGC2325
ESO512G019
MCG0127015
NGC3311
ESO142G049
NGC7832
MCG0127018
NGC4825
NGC5869
MCG0103018
IC2437
NGC7458
NGC2765
IC4731
NGC3042
NGC1209
NGC2822
NGC2974
NGC1700
NGC3886
NGC7196
NGC2663
ESO467G037
ESO235G049
NGC4374
NGC5087
ESO510G009
NGC1399
NGC6861
NGC1332
NGC7507
NGC1404
ESO286G049
NGC7049
ESO263G048
NGC0547
NGC3805
NGC3923
NGC5846
NGC3209
NGC2911
NGC6614
NGC1395
ESO126G014
NGC6868
NGC3919
NGC0883
NGC0720
NGC4546
NGC5062
NGC1407
NGC6841
NGC5028
NGC4472
IC2586
NGC1400
IC1459
NGC3091
NGC5419
IC4296
NGC1550
NGC2986
NGC7014
Fig. B.4. The most parsimonious tree found with cladistics taking uncertainties in the parameters into account. The colours correspond to the
groups defined in Fig. 3.
parameters (the Lick indices), a face value given to these uncertainties, and the equal probability given to all values within the
range of uncertainty.
These results shows that the cladistic analysis is relatively
robust to measurement errors, as found through the comparison
with different clustering methods.
A80, page 21 of 24
A&A 545, A80 (2012)
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
Mabs
−22
−20
0.30
Mg2
−22
0.10
0.0
−20
0.5
−18
1.0
2.4
2.2
2.0
1.8
1
−18
Mabs
0.20
logre
2.6
logs
0.40
Appendix C: Supplementary figures
1.8
2.0
2.2
2.4
2.6
1.8
2.0
logσ
D/B
2.6
2.4
2.6
2.4
2.6
7
6
7
1
2
3
4
5
6
7
5
2
3
4
5
6
4
Mgb
1
7
2.0
2
OIII
Kabs
[MgbFe]’
1.8
−20
4
5
6
7
2
6
4
5
6
7
6
7
6
7
7
1
2
3
5
6
7
1.8
1
2
Mgb
3
4
5
2.0
2.2
2.4
2.6
1.8
logσ
2.0
2.2
logσ
Fe5270
6
6
2
3
4
5
6
7
1
2
3
4
5
6
7
5
1
2
3
4
5
3
2
2
1
4
Mgb
5
4
3
2
3
2.0
3
4
2.5
NaD
4
5
3.0
5
6
6
3.5
7
7
Fe5015
4
3
0.5 1.0 1.5 2.0 2.5 3.0
−2.0
−4.0
−5.0
5
3
Hbeta
−3.0
80 100
60
4
7
1
6
3
NaD
2
T
40
3
2.2
2
1
20
2
2.0
logσ
2.0
7
0
1
1.8
1.8
6
2.6
1.4
5
nc
Mgb/Fe
4
2.4
1.0
3
2.2
logσ
3.0
−22
−24
−26
2
2.0
4.0
2.0
1.0
0.0
−1.0
1
3.0
5
[MgbFe]’
4
5
3
4
2
3
1
4.0
6
5
3
0.0
2
18
19
1.0
4
20
21
2.0
6
22
23
2.4
NaD
3.0
Brie
2.2
logσ
Fe5335
log(diam)
2.0
Mgb/Fe
3.0
4.0
2.0
3.0
[MgbFe]’
1
2
3
4
5
6
7
1
2
3
4
5
6
7
M/Lr
1
2
3
4
5
6
7
1
2
3
4
5
6
7
Fig. C.1. Same boxplots as in Fig. 4 but for the cluster partitioning.
Colours are the one given in Fig. 2.
A80, page 22 of 24
3.0
0.6
0
9.5 10.0
100
−26
200
11.0
300
12.0
400
Mdyn
2.5
7
M/L
6
2.0
5
1.5
4
−22
3
Kabs
2
−24
1
−20
1.4
1.0
0.6
1.5
2.0
1.0
2.5
1.4
3.0
1.8
1.8
3.5
[MgbFe]’
4.0
1.0
1.4
log(diam)
1.8
8.0
8.5
9.0
9.5
L
Fig. C.2. Scatter plots showing evolutionary correlations, like Fig. 5,
but for the cluster partitioning. Colours are the same as in Fig. C.1.
D. Fraix-Burnet et al.: A six-parameter space to describe galaxy diversification
2.6
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
1.8
logσ
2.6
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
1.8
logσ
2.2
C3
Brie
Brie
Brie
C4
C5
C6
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
1.8
logσ
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
1.8
logσ
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
2.6
17 18 19 20 21 22 23
2.6
17 18 19 20 21 22 23
2.6
17 18 19 20 21 22 23
1.8
logσ
C2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
1.8
logσ
2.6
C1
17 18 19 20 21 22 23
17 18 19 20 21 22 23
17 18 19 20 21 22 23
Brie
Brie
Brie
2.2
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
1.8
logσ
2.6
C7
17 18 19 20 21 22 23
Brie
Fig. C.3. Comparison of the positions of the groups found in Fraix-Burnet et al. (2010) and those of the present paper, as projected onto the
fundamental plane. The colour-coded ellipses are the inertia ellipses for each group from the present paper, and the black ellipse is the one for the
group from Fraix-Burnet et al. (2010) indicated on top each graph. See also Fig. 8.
A80, page 23 of 24
A&A 545, A80 (2012)
C2
11
12
11
12
1.5
0.5
−0.5
1.5
0.5
−0.5
10
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
13
10
11
Mdyn
Mdyn
C4
C5
C6
11
12
13
Mdyn
10
11
Mdyn
12
13
12
13
0.5
1.5
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
−0.5
logre
0.5
1.5
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
−0.5
0.5
1.5
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
10
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
Mdyn
−0.5
logre
13
logre
10
logre
0.5
−0.5
logre
1.5
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
C3
logre
C1
10
11
12
13
Mdyn
C7
0.5
−0.5
logre
1.5
Clad1
Clad2
Clad3
Clad4
Clad5
Clad6
Clad7
Clad8
10
11
12
13
Mdyn
Fig. C.4. Comparison of the positions of the groups found in Fraix-Burnet et al. (2010) and those of the present paper, as projected onto the log re
vs. Mdyn diagram. The colour-coded ellipses are the inertia ellipses for each group from the present paper, and the black ellipse is the one for the
group from Fraix-Burnet et al. (2010) indicated on top each graph. See also Fig. 9.
A80, page 24 of 24