The operational definition of a group is
a number density enhancement in (redshift) space
(HG82; NW87 have a slightly different point of view).
The algorithm can be thought of as a percolation technique,
but truncated to a specified value R0 of the connecting link :galaxies closer than R0 are "friends'' of each other,
friendship is transitive,
and an isolated set of (at least 3) friends is what we call a galaxy group.
In the ideal case of a luminosity-complete and volume-limited sample with
N galaxies, volume V, average density
,
with no redshift distortions,
groups would then be selected above a fixed number density threshold
, given by:
![]() |
(1) |
![]() |
(2) |
To overcome effect (ii),
the strategy is to "compensate'' the decrease in by allowing the links to be "more generous'' at larger r or fainter
.To deal with effect (i),
the spatial link
is replaced by
a transverse "sky-link''
and a radial "redshift-link''
.Different authors proposed qualitatively different solutions
to implement such a strategy
(HG82; NW87; see also Gourgoulhon et al. 1992, N93; Garcia 1993;
we refer to the original papers for details),
involving also dynamical considerations about
because of effect (i).
Advantages and shortcomings of one recipe over another are discussed in
NW87, Garcia (1993) and F95a, b.
We adopt the scaling recipe of HG82, though with stricter normalizations.
Both and
are normalized by
and
at some fiducial redshift cz0=H0 r0.
They are then scaled with cz as
,where czij=(czi+czj)/2 is the median redshift of the ij-th pair of galaxies.
Galaxies are linked if their transverse and radial separations
and
satisfy
and
, respectively.
The number density within groups at distance r is
![]() |
(3) |
![]() |
(4) |
According to HG82, groups are selected above a density contrast
given a priori.
The physical justification is the hypothesis-requirement that
galaxy groups correspond to dynamically bound matter overdensities,
whose dynamical state is dictated by their density contrast
,in turn related to the number density contrast
,maybe through some mechanism of biased galaxy formation (e.g., Kaiser 1984).
Unfortunately, one does not know exactly the value
of the normalization of
in Eq. (1).
Due to the presence of LSS, this may vary by a factor of 2 from survey to survey,
and fluctuates strongly even within the same galaxy survey
(e.g., de Lapparent et al. 1988, 1989; see also Sect. 3.4).
Fluctuations and uncertainties about the shape of the LF
also have a (slight) effect on the D0-
relation (4)
(TB96; Sect. 4.3).
Moreover, the relation among
the (physically well-motivated) mass density-contrast
and the observatio galaxy number density-contrast
,is still very uncertain (e.g., Bower et al. 1993).
Actually, the very existence of a universal value of
, valid at any
sufficiently large spatial scale, has been repeatedly questioned
(e.g., Coleman & Pietronero 1992; Baryshev et al. 1994)
and the debate on this point is still alive
(Davis 1996; Pietronero et al. 1997).
What normalizations should we adopt?
For the sake of clarity, let us first consider only
data samples of the same depth, but different galaxy LF.
(We will discuss the effect of different depth,
and of further sample-to-sample variations
due to local LSS features, later on.)
The RGH89 group catalog in the CfA2 Slices is normalized by
and
at
.With the galaxy LF adopted by RGH89, D0 translates into
.This very value may be regarded as an update of
used by HG82 and based on theoretical considerations,
enhanced by a factor 4 to account for the locally high density of the
the Great Wall region within CfA2 Slice where most of the groups reside
(Ramella et al. 1992; Ramella, private communication).
Subsequent observational (Ramella et al. 1995a-b, 1996) and theoretical (F95a,b) studies
confirmed such values of V0 and D0
as "optimal'' for the compilation of a group catalog,
though different normalizations may be more appropriate in different contexts
(NW87; MFW93; N93; F95a, b; NKP94, NKP97).
We also note that the specific conclusions of most such works should be taken with caution,
as (i) they do not always use the precise HG82 scaling for
and
,and (ii) their observational sample is often the CfA1 survey,
brighter and shallower than CfA2 and PPS2. (For a discussion of the effect of depth, see below.)
Later, PGHR94 and RPG97 used a slightly different
,
due to the higher density of CfA2 North and to their requirement
.The galaxy LF of PPS2 is still different,
yielding further combinations of D0 and
.With our choice of
(Sect. 3.4),
D0=0.231 and
yield
and 108, respectively,
instead of the desired
.The latter is only recovered if we adopt
,substantially larger than D0 used by RPG97.
This is consistent with the different normalizations
suitable for PPS2, CfA2 Slice, and CfA2 North
(
,
, and
, respectively)
and well within the LF normalization uncertainty.
We are then led to the following question.
In order to build group catalogs physically as similar as possible to each other,
should they be compiled using (the same link V0 and)
the same link D0, or
the same density threshold ?
We emphasize that the parameter
defined in Eq. (4)
is customarily used only in order to label a given FOF catalog.
In fact, no such assumption is needed to identify the groups,
and one could as well use as label the D0 parameter
effectively used by the FOF algorithm itself.
Moreover, the theoretically motivated mass overdensity
is ideally referred to the mean mass density
of the whole Universe,
and not
averaged only over the considered sample.
However, no matter whether the universal value of
coincides with
and whether it is known or not,
for a given
identical D0's will correspond to identical
.So, though strong theoretical motivations suggest
to be the relevant
physical quantity, and suggest
as its observational counterpart,
on practical grounds it is more justified to relate the link normalization directly to D0.
As a matter of fact,
the issue of D0 vs.
was considered also by Maia et al. (1989).
They compared FOF groups in SSRS1 and in CfA1, identified either with the same D0
or with the same
, and found that the median physical properties
of the groups were more similar in the former case than in the latter.
However, it is not clear how to interpret such result, since
the two galaxy samples SSRS1 and CfA1 differ in depth and selection criteria.
We can now discuss the case of data samples of different depth,
for which things are still slightly more complicated.
For simplicity, we assume them to have the same parent LF.
We adopt the Schechter (1976) functional form
.Then from Eq. (4) above, constant
requires
![]() |
||
(5) |
![]() |
(6) |
![]() |
Figure 3: Members of loose groups in PPS2: 4+1 diagram. Each dot is a group member galaxy. Everything else as in Fig. 1 |
All these complications, due to the nature of the FOF grouping algorithm,
would be avoided if group properties turned out
to change little for reasonable variations of FOF parameters.
Actually, RPG97 report that group properties are non-significantly sensitive
to the choice of linking parameters over a wide range of .On the other hand, the symmetry and the "orthogonality'' of
and
in the HG82 algorithm suggests
D0 to be directly related to the group size (
or
), and
V0 to be directly related to the group velocity dispersion
,possibly with some residual "non-orthogonal'' dependence.
Since
,
any dependence of a given group property X on D0 or
is equally recovered if one plots X against
or
.In summary, even a substantial dependence of group properties on the links V0 and D0,
though not removed, might be missed or hidden
by focusing attention only on the customary
parametrization.
In order to test such effects on group properties,
we built several arrays of group catalogs with different normalizations,
and the same LF adopted for PPS2.
The first array has given and variable
D0=
0.184, 0.192, 0.202, 0.215, 0.231, 0.254,
0.270, 0.290, 0.300, 0.330,
,
including the D0's used by RPG97 (Ramella, private communication) and yielding
342, 301, 259, 214, 173, 129,
108, 87, 79, 59, 45, respectively, with our LF.
The second array has given
and variable
V0= 150, 250, 350, 450, 600,
.The third array is like the second one, but for
.In Fig. 2a, we plot
the line-of-sight velocity dispersion
(medians) against the redshift link V0.
Analogously,
in Fig. 2b, we plot
the mean harmonic radius
and mean pairwise member separation
(medians)
against the spatial link D0.
In all cases, we find strong, approximately linear correlations:
,
,
.Similar results also hold for groups
in CfA2 North (Fig. 2, RPG97's data), in CfA1 (NKP94; NKP97),
and in SSRS1 (Maia et al. 1989, their Table 5).
Although RPG97 produce a figure with a decreasing trend of
versus
, they do not mention that this trend
simply arises from the proportionality between median
and D0
(which would have been obvious had they made a log-log plot).
For all the above mentioned reasons, several of which are mainly practical,
we prefer to parametrize a given FOF algorithm by its D0 and V0.
In particular, we specify the threshold spatial separation used by the FOF algorithm
directly in terms of D0, instead of through the density contrast (HG82)
or as a fraction of the mean inter galaxy separation (NW87).
We then compute a posteriori through Eq. (4),
and interpret its value mainly as an approximate measure,
rather than a precise parametrization, of density contrast.
![]() |
(7) |
Recently, Marzke et al. (1994; MHG94 hereafter) estimated the LF of the
whole CfA2 Survey, and several subsamples therein.
For their CfA2 South subsample, very similar to PPS2, they get:
,
,
.Note that they explicitly corrected for (a form of) Malmquist bias,
the Eddington (1913) bias:
Zwicky magnitudes have an uncertainty
-0.4 mag,
which causes a "random diffusion'' of the more numerous fainter galaxies
towards brighter magnitudes, and in turn modifies the overall shape of the LF.
This effect induces
an artificially bright M*, and a correspondingly too negative
.
We use an estimate of based directly on our data (TB96).
First, though PPS2 and CfA2 South are similar samples,
they are still slightly different.
Their galaxy LF's might be different, and this might effect
group identification through the
ingredient of the FOF algorithm.
Second, group identification requires a Malmquist-uncorrected
.
Malmquist corrections to
are global,
and they do not apply to each single galaxy.
In fact,
only the study of intrinsic physical properties of the galaxies themselves
would require such (unknown) corrections
(and others, e.g. internal extinction).
Our sample PPS2 is spatially inhomogeneous (Fig. 1).
This requires a density-inhomogeneity-independent technique
(e.g., Efstathiou et al. 1988; de Lapparent et al. 1989;
see also the review of Binggeli et al. 1988).
Using the inhomogeneity-independent STY method (Sandage et al. 1979)
and Zwicky magnitudes (corrected for Milky-Way extinction),
TB96 found and
.By construction, the STY technique does not provide an estimate of the
density normalization
.Simple tests with the non-parametric inhomogeneity-independent C-method
(Lynden-Bell 1971; Choloniewsky 1986, 1987)
or, on the other hand, a countercheck with
the more naïve inhomogeneity dependent
technique
(e.g., Felten 1977),
all seem to suggest
for PPS2 (TB96).
However, matching the absolute magnitude counts
or the radial counts
is better accomplished using
(TB96).
Both values are consistent with the typical uncertainty
,and with the CfA2 South value
given by MHG94.
Since their analysis is superior TB96's on this point,
here we also adopt
.(Note that this only matters when we insist on translating
a given D0 in the grouping algorithm into
an approximate density threshold
.)
Regarding the other two Schechter parameters,
estimating errorbars as in Marshall (1985) the same tests also suggested an uncertainty
and
(TB96).
This is close to the typical observational uncertainties
(and/or the scatter among different techniques and/or data sets)
and
reported by MHG94, and the similar earlier studies quoted above.
The results of TB96 (not Malmquist-corrected) and of MHG94
(Malmquist-corrected
by
,
, both positive)
are in good agreement, once we take into account
the different details between the two analyses.
In fact, by using inhomogeneity-independent techniques,
several authors estimated uncorrected Schechter parameters and
their corresponding additive corrections
and
.In the first two Northern CfA2 Slices, de Lapparent et al. (1989) found
and
.For the whole CfA1, similar values were found
by Efstathiou et al.
(1988;
and
), and
by N93 (
and
).
Our LF and those of de Lapparent et al. (1988, 1989) for the CfA2 Slice(s)
(
-
,
,
not Malmquist-corrected),
used in previous group catalogs, are also rather similar.
Let us outline that, concerning group properties,
such residual small differences
have however a rather small effect (Sect. 4.3; TB96).
Copyright The European Southern Observatory (ESO)