A New Local Temperature Distribution Function for X-ray Clusters: Cosmological Applications
Extracted source text ยท arXiv astro-ph/9908037
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% corrections du 23/07/99
% corrections du 26/07/99 J.G. Bartlett
% corrections du 10/10/99 Revisions
% corrections du 1/01/00 last revisions ? oh, no...
% corrections du 03/01/00
% corrections du sam 08/01/00
% corrections du 02/06/00
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\documentclass[referee]{aa}
%\usepackage{aabib99}
%\bibliographystyle{aabib99}
%\documentclass{aa}
\usepackage{graphics}
\begin{document}
\thesaurus{}%12.03.4;12.03.3;12.04.01;12.12.1;11.03.1}
\title{A New Local Temperature Distribution Function for X--ray Clusters:
Cosmological Applications}
\author{A.~Blanchard$^{1,2}$ \and R.~Sadat$^1$ \and J.G.~Bartlett$^1$ \and
M.~Le~Dour$^1$}
\institute{ $^1$ Observatoire Midi-Pyr\'en\'ees, 14, Av. Edouard Belin, 31400
Toulouse, France\\
$^2$ Observatoire astronomique de Strasbourg, ULP,
11, rue de l'Universit\'e, 67000 Strasbourg, France }
\offprints{A.Blanchard}
\mail{blanchard@astro.u-strasbg.fr}
\date{Received \rule{2.0cm}{0.01cm} ; accepted \rule{2.0cm}{0.01cm} }
\titlerunning{Cluster Temperature Function}
\authorrunning{Blanchard, Sadat, Bartlett \& Le Dour }
\maketitle
\begin{abstract}
We present a new determination of the local temperature
function of X-ray clusters using a sample of X-ray clusters with fluxes above 2.2
$10^{-11}$erg/s/cm$^2$ in the $[0.1-2.4]$ keV band, most of these clusters come from the Abell XBAC's
sample to which a handfull of known non-Abell clusters
has been added. We estimate this sample to be 85\% complete, and should
therefore provide an useful estimation of the present-day
number density of clusters.
Comprising fifty clusters for which the temperature information is available, it is the largest complete sample of this kind. It is
therefore expected to significantly improve the
estimation of the temperature distribution function of clusters. We find that the resulting temperature function is higher than
previous estimations, but it agrees with the temperature distribution
function inferred from the BCS and RASS luminosity function
(Ebeling et al., 1997; De Grandi et al. 1999a). We have used this sample to constrain the amplitude of the matter
fluctuations $\sigma_c $ on cluster's scale of $8\sqrt[3]{\Omega_0}^{-1}h^{-1}$Mpc,
assuming a mass-temperature relation based on recent numerical simulations.
We find $\sigma_c \sim 0.6\pm 0.02$ for an $\Omega_0 = 1$ model (for which
$\sigma_c = \sigma_8$). Our sample
provides an useful reference at $z \sim 0$ to use in the application of the
cosmological test based on the evolution of X-ray clusters abundance
(Oukbir \& Blanchard 1992, 1997). We have therefore
estimated the temperature distribution function at $z = 0.33$ using
Henry's sample of
high-z X-ray clusters (Henry, 1997; hereafter H97) and performed a preliminary
estimate of $\Omega_0$. We find that the abundance of clusters at $z = 0.33$
is significantly smaller, by a factor larger than 2, which shows that the
EMSS sample provides strong evidence for evolution of the cluster
abundance. A likelihood analysis leads to a rather high value of the mean
density parameter of the universe: $\Omega_0 = 0.92^{+0.255}_{-0.215}$ ($1 \sigma$) for open universes and
$\Omega_0 = 0.865^{+0.35}_{-0.245}$ for flat
universes, which is consistent with a
previous independent estimation based on the full EMSS sample by
Sadat et al.(1998). Some systematic uncertainties which could alter
this result are briefly discussed.
\keywords{Cosmology: observations --
Cosmology: large--scale structure of the Universe --
Galaxies: clusters: general}
\end{abstract}
\section {Introduction}
Clusters are believed to be the largest virialized
concentrations of dark matter. Therefore, they offer privileged regions for
studying dark matter distribution on large scales in the universe.
X-ray and weak lensing mass measurements have been added to
the traditional mass estimates based on optical velocity dispersions.
Weak lensing analyses are still in their infancy and at present
there exists no sample of weak lensing observations large enough to establish
a mass function. The velocity distribution function of clusters
could be used to derive the mass function. However, as
Evrard pointed out (Evrard, 1989) the error on individual measurements
can introduce a significant overestimate. Furthermore, velocity measurements
at least in the case of distant clusters, can be corrupted by
projection effects that might be difficult to remove in practice
(Frenk et al., 1990). Moreover, Sadat et al. (1998) have shown that
X-ray temperatures of some of the CNOC clusters show a significant
difference with what is expected from their velocity dispersion measurements.
For these reasons, it has been argued that the X-ray temperature
is a better indicator of cluster mass. Numerical simulations have
greatly helped to understand the relation between X--ray temperature and
the mass, and useful constraints have been placed on the amplitude and
the shape of the spectrum of mass density
fluctuations. Still, the size of the samples of X-ray clusters
homogeneously selected is limited: 25 clusters in the
Henry and Arnaud (1991, hereafter HA91) sample and 30 clusters in Markevitch's
(1998). ROSAT selected clusters samples have significantly improved the
situation in this domain. In order to achieve stronger constraints on
theoretical models, it will be necessary to obtain more temperature
measurements, a vast program that will probably be possible
with the next generation of X-ray satellites such as AXAF and XMM.
The cluster X--ray temperature function is a powerful
tool for cosmology. Provided that the mass-temperature relation is reasonably
well known, the Press and Schechter (1974) formalism allows one to constrain
the amplitude and the shape of the power spectrum for a given cosmological
background density (see Bartlett, 1997, for a recent review on the subject).
Since there is a nearly complete degeneracy between the amplitude of the
fluctuations and the mean density of the universe, the cosmological parameter
$\Omega_0$ cannot be determined solely from the local temperature function.
However, the evolution of this temperature distribution function, once
normalized to the present day cluster abundance, varies significantly
with $\Omega_0$ offering an interesting new cosmological probe (Oukbir \& Blanchard 1992, hereafter OB92;
Hattori \& Matsuzawa, 1995). This test has received considerable
attention in recent years (Donahue, 1996; Carlberg et al., 1997;
H97; Oukbir \& Blanchard 1997, OB97, hereafter; Blanchard \& Bartlett, 1998; Eke et al., 1996, 1998;
Sadat et al., 1998; Viana \& Liddle, 1996, 1999a; Blanchard et al., 1999;
Donahue et al., 1999; Donahue and Voit, 1999; Reichart et al., 1999). Variants have been proposed using the
Sunyaev--Zeldovich (Barbosa et al, 1996) and weak lensing effects
(Kruse \& Schneider, 1999). Modeling the redshift distribution of the
EMSS X-ray selected sample (OB97) given the absence of
any significant evolution in the
$L_x-T_x$ relation (Mushotsky \& Sharf, 1997; Sadat et al., 1998),
seems to favor high value of the density parameter (Sadat et al, 1998;
Reichart et al, 1999). Modeling the RDCS redshift distribution
leads to consistent results (Borgani et al, 1998).
A more direct estimate, free from any consideration on the possible
evolution in the $L_x-T_x$ relation, could be obtained from the measurement
of the evolution of the temperature distribution function, which in turn requires a good knowledge of the selection function of the sample of clusters.
The first sample of X-ray selected clusters at non--zero redshift
with measured temperatures has recently become available (H97) and has
led to an apparent median value of $\Omega_0$
in the range $0.2-0.6$ (H97; Eke et al, 1998), although higher values
were found by Viana and Liddle (1999a) and Blanchard et al. (1999). It has been
argued that current data are not good enough to allow a reliable estimate
of the mean density of the universe from such techniques
(Colafrancesco et al., 1997), a conclusion that
appears wise given that the local sample of X-ray clusters used up to now
is that of HA91 which contains only 25 clusters while the high--redshift
sample comprises only 10 clusters with moderately accurate temperature
measurements (recently, the redshift of one cluster in the sample has been
revised, reducing the sample to 9 clusters, Donahue et al., 1999). However,
the fact that some conclusions have already been drawn, even if too optimistic
regarding possible systematics, demonstrates the power of this test:
clearly ten more clusters or so at high
redshift and an accurate determination of the local temperature distribution
function would provide a very robust determination of the
mean density of the universe. In fact, going to high redshift makes a dramatic
difference in the abundance of hot clusters (OB92), and
the existence of few EMSS clusters with a high temperature has been argued to
already provide a strong evidence for a low density universe (Donahue 1996;
Bahcall \& Fan, 1998; Donahue et al., 1998; Eke et al., 1998).\\
Until recently, the X-ray cluster temperature distribution function was
inferred from catalogs built from the HEAO1 survey in the 2-10 keV band.
The need for a new estimation of the temperature distribution function
has been recognized. Markevitch (1998, hereafter M98) has provided such a new estimate based on a sample
of X-ray clusters for which ROSAT fluxes were available.
Most of his clusters come from the XBACS sample of bright Abell
clusters (Ebeling et al., 1996). Temperatures are
derived from ASCA observations, and for this reason clusters at redshift
smaller than 0.04 were not considered. This means that the sample is
not a strictly flux limited sample. M98 corrected both the fluxes
and temperatures of his sample for the presence of cooling flows in the
central regions. Cooling flows are believed to be an important source of
dispersion in the $L_x-T_x$ relation (Fabian et al., 1994; M98;
Arnaud \& Evrard, 1999), but have negligible effect on the temperature
distribution function. \\
In the following, we present the analysis of a new sample containing
fifty clusters essentially based on the XBACS sample (Ebeling et al., 1996),
using a flux limit similar to that of M98. Note that M98 considered his
sample to be reasonably complete. In the present analysis, the effect of
temperature error measurements is explicitly taken into account using a
Bayesian correction. Our sample does not require any correction for
redshift incompleteness. We also do not correct for cooling flows in order
to allow a direct comparison between our sample and those at high redshift.
The resulting temperature distribution function we obtain is smooth
and can be used for useful comparison with theoretical models.
We have performed such a comparison in order to constrain
the density fluctuation power spectrum with different values of the
cosmological background density $\Omega_0$.
This paper is organized as follows: section 2 summarizes the results from
previous estimates of the local temperature function. In section 3
we give a brief description of our sample. In section 4, we present our
method to calculate the local temperature function as well as
a determination of the distribution function of the cluster abundance
estimator. We also discuss how our temperature function compares with previous
results. Section 5 presents the first application of this new temperature
distribution function to constrain cosmological parameters.
\section{Previous estimates}
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{histo1.ps}\includegraphics{histo2.ps}}
\hspace*{1.8cm}a) \hspace*{8.4cm} b)
\caption{\label{fig:histoT1} The distribution of clusters with revised
temperatures in the Henry \& Arnaud sample, HA91 (left panel). The right
panel shows the histogram of the 21 clusters in the Edge et al. (1990) sample,
not included in HA91.
%Because both samples are considered as X-ray selected, one should
%expect the same distribution. Clearly this is not the case. The reason for
%this difference is unclear, but translates in a significant difference
%in the inferred temperature function.
}
\end{figure*}
In order to determine the temperature function, it is important to have a
well controlled sample of clusters. The knowledge of the selection
function is therefore critical. For instance, the completeness of the Edge et al.
(1990) sample (hereafter E90) was
questioned by HA91, and in fact the difference between the
E90 and HA91 temperature functions is striking. We therefore re--examine
these two samples, although the revised version of HA91 appears to be closer
to E90. Rather than plotting the temperature function, we directly
examine the histogram of the number of clusters in temperature bins
for each sample. In Figure \ref{fig:histoT1}a, we show the histogram
of the revised temperatures in the HA91 sample, while in Figure
\ref{fig:histoT1}b, we give the corresponding histogram for the E90
clusters {\em not present} in the HA91 sample; this latter will be referred to
as the E-HA sample. Comparison of the two histograms clearly reveals
a striking difference: there is an apparent deficit of clusters
with $T \sim 4-6$ keV in the HA sample as compared to E-HA.
Obviously, the smaller number of 4 keV clusters present in the HA sample
leads to a lower number density. This is of critical importance in the
theoretical modeling, because the abundance of 4 keV clusters is
crucial in determining the amplitude of the fluctuations on a 8$h^{-1}$Mpc
scale (for $\Omega_0 = 1$). The E-HA set is astonishingly different:
the number of clusters reaches a maximum at $T = 4$ keV, and then tends to
decrease monotonically for higher $T$, meaning that there are relatively
fewer hot clusters in the E-HA data set (4 clusters from the E90 sample
with poor temperature measurements are not taken into account in E-HA,
which adds 3 clusters to the region $T > 8$ keV, but this does not change
the shape of the histogram). Although
statistical analysis of such a small sample is a
hazardous exercise, this difference is highly significant: assuming that the number of clusters found in the HA sample between 4 and 6 keV (4 clusters), the probability to have ten clusters or more in the E-HA sample in the same range is less than $10^{-3}$. When dealing with small samples, one must of course be careful of
spurious noise introduced by Poisson statistics. However, using different tests
we always find this difference significant at a confidence level greater than 95\%. It is difficult to understand the
origin of such a difference: both samples are X-ray selected
and incompleteness is expected to be nearly independent of temperature.
It is possible that the existence of
large--scale fluctuations, as can be seen from
the visual appearance of the clusters
distribution or from the long--range correlation function, indicates that
noise due to a finite number of objects may be larger than that
expected from Poisson noise.
The comparison we have presented in Fig. \ref{fig:histoT1} clearly
calls for a larger, complete flux limited sample strictly X-ray selected.
However, only the Rosat All Sky Survey (RASS) would allow the construction of such a sample. The Bright Cluster Survey BCS (Ebeling et al., 1998) and the RASS South (De Grandi et al., 1999b) have just become available at the time
of writing, but the lack of temperature information for a significant
fraction of these clusters prevents us from using these samples
to usefully estimate the temperature distribution function.
Recently, M98 provided a new local temperature function based on the
XBACS sample (Ebeling et al, 1996).
XBACS is an essentially complete, flux-limited X-ray sample of Abell clusters.
In principle, the restriction to Abell clusters implies that the sample
is possibly incomplete. There are three possible sources of incompleteness:
i) low mass (i.e., low temperature) clusters may be
missing because they are not rich enough to be selected
by the Abell criteria; ii) distant clusters could be missed because at
large redshifts clusters hardly meet the Abell criteria;
iii) there is a significant difference between the optical
based Abell criteria and those based on X-ray fluxes.
M98 restricted his sample to the redshift range
$0.04\leq z \leq 0.09$, the lower limit being imposed by ASCA.
This redshift restriction eliminates intrinsically faint clusters from
his sample. M98 corrected for this incompleteness
by using a weight to compensate for the selection function.
The resulting temperature function is noticeably higher than previous
estimates, although it is claimed to be consistent with them.
However, the correction procedure could introduce a systematic error
in the derived abundance of cool clusters as we will see in the next
section.
\section{The present sample}
\begin{center}
\begin{table*}[ht]
\begin{tabular}{llllll}
\hline
\hline
Cluster & z &$T_{X}$$^{1}$ & $L_{X}$$^{2}$ & $s_{X}$$^{3}$ & Ref$^{4}$.\\
\hline
Virgo& 0.0036 & 2.2$^{ +0.1 }_{ -0.1 }$ &
1 & 82.0 &{\sl 6,10 } \\
A1060& 0.0114& 3.25$^{ +0.1 }_{ -0.1 }$ &
0.47& 7.&{\sl 4,1 } \\
A3526& 0.0109& 3.54$^{ +0.1}_{ -0.1}$ &
0.83& 19.&{\sl 8,1} \\
A262 & 0.0161 & 2.15$^{ + 0.1}_{ -0.1}$ &
0.55& 4.8&{\sl 4,1 } \\
A3581& 0.02& 2.0$^{ + 0.1}_{ -0.1}$ &
0.57& 2.9&{\sl 5,1 } \\
A1367& 0.0215 & 3.55$^{ +0.1}_{ -0.1}$ &
1.63 & 8.3&{\sl 4,1 } \\
A1656& 0.0232 & 8.2$^{ +0.1}_{ -0.1}$ &
7.21 & 31.6 &{\sl 9,1 } \\
A4038& 0.029& 3.3$^{ + 2.6}_{ -1.3}$ &
1.9& 5.27 &{\sl 2,1 } \\
A2199& 0.0309 & 4.5$^{ +0.1}_{ -0.1}$ &
3.65 & 9.5&{\sl 7,1 } \\
A496& 0.032& 4.1$^{ +0.1}_{ -0.1}$ &
3.54 & 7.5&{\sl 4,1 } \\
A2634& 0.0321 & 3.7$^{ + 0.3}_{ -0.3}$ &
0.94& 2.3&{\sl 4,1 } \\
A2063& 0.0337 & 3.7$^{ + 0.1}_{ -0.1}$ &
2.03 & 3.7&{\sl 4,1} \\
A2052& 0.0348 & 3.1$^{ + 0.3}_{ -0.3}$ &
2.52 & 4.7&{\sl 2,1 } \\
Cl0336& 0.0349 & 3.$^{ +0.1}_{ -0.1}$ &
5.87 & 8.76 &{\sl 4,1 } \\
A2147& 0.0356 & 4.9$^{ +0.3}_{ -0.3}$ &
2.84 & 5.3&{\sl 4,1 } \\
A576& 0.0381 & 4.3$^{ + 0.5}_{ -0.4}$ &
1.39 & 2.24 &{\sl 2,1 } \\
0422--09& 0.039 & 2.9$^{ + 0.5}_{ -0.4}$ &
{\em 2.$^*$} & {\em 3. $^*$} & {\sl 2,2 } \\
A3571& 0.0391 & 6.9$^{ + 0.3}_{ -0.3}$ &
7.36 & 10.9 &{\sl 3,1 } \\
A2657& 0.04& 3.7$^{ + 0.3}_{ -0.3}$ &
1.6& 2.33 &{\sl 3,1 } \\
A2589& 0.0416 & 3.7$^{ + 0.6}_{ -0.6}$ &
1.86 & 2.5&{\sl 6,1 } \\
A119& 0.044& 5.8$^{ + 0.4}_{ -0.4}$ &
3.23 & 3.8&{\sl 3,1 } \\
MKW3s& 0.045& 3.5$^{ + 0.2}_{ -0.2}$ &
3.& 3.43 &{\sl 3,3 } \\
A1736& 0.0461 & 3.5$^{ + 0.4}_{ -0.4}$ &
2.37 & 2.6&{\sl 3,1 } \\
A3376& 0.0464 & 4.3$^{ + 0.6}_{ -0.6}$ &
2.48 & 2.69 &{\sl 3,1 } \\
A3558& 0.0478 & 5.5$^{ + 0.3}_{ -0.3}$ &
6.27 & 6.2&{\sl 3,1 } \\
\hline
\hline
\end{tabular}
\hspace*{2mm}
\begin{tabular}{llllll}
\hline
\hline
Cluster & z &$T_{X}$$^{1}$ & $L_{X}$$^{2}$ & $s_{X}$$^{3}$ & Ref$^{4}$.\\
\hline
A1644& 0.048& 4.7$^{ +0.8}_{ -0.8}$ &
3.52 & 3.64 &{\sl 2,1 } \\
A4059& 0.048& 4.0$^{ + 0.1}_{ -0.1}$ &
3.09 & 3.12 &{\sl 4,1 } \\
A3562& 0.0499 & 3.8$^{ + 0.8}_{ -0.8}$ &
3.33 & 3.08 &{\sl 2,1 } \\
A3395& 0.05& 4.8$^{ +0.4}_{ -0.4}$ &
2.8& 2.65 &{\sl 3,1 } \\
A85& 0.0518 & 6.1$^{ +0.2}_{ -0.2}$ &
8.38 & 7.2&{\sl 3,1 } \\
A3667& 0.053& 7.$^{ + 0.6}_{ -0.6}$ &
8.76 & 7.3&{\sl 3,1 } \\
A754& 0.0534 & 7.6$^{ +0.3}_{ -0.3}$ &
8.01 & 12.&{\sl 6,1 } \\
A780& 0.05384 & 3.8$^{ +0.2}_{ -0.2}$ &
6.63 & 4.8&{\sl 3,1 } \\
A3391& 0.054& 5.7$^{ + 0.7}_{ -0.7}$ &
2.32 & 2.39 &{\sl 3,3 } \\
A3158& 0.059& 5.5$^{ +0.6}_{ -0.6}$ &
5.31 & 3.57 &{\sl 2,1 } \\
A3266& 0.0594 & 7.7$^{ + .8}_{ -0.8}$ &
6.15 & 4.8&{\sl 3,1 } \\
A2256& 0.0601 & 7.5$^{ +0.4}_{ -0.4}$ &
7.05 & 4.9&{\sl 3,1 } \\
A133& 0.0604 & 3.9$^{ + 1.6}_{ -0.7}$ &
3.57 & 2.29 &{\sl 2,1 } \\
A1795& 0.0616 & 6.$^{ + 0.3}_{ -0.3}$ &
11.1 & 6.7&{\sl 3,1 } \\
A3112& 0.07& 4.7$^{ + 0.4}_{ -0.4}$ &
7.7& 3.6&{\sl 3,1 } \\
A399& 0.0715 & 7.4$^{ + 0.7}_{ -0.7}$ &
6.45 & 2.9&{\sl 3,1 } \\
A2065& 0.072& 5.4$^{ + 0.3}_{ -0.3}$ &
4.95 & 2.2&{\sl 3,1 } \\
A401& 0.0748 & 8.3$^{ + 0.5}_{ -0.5}$ &
9.88 & 4.26 &{\sl 3,1 } \\
A2029& 0.0767 & 8.7$^{ + 0.3}_{ -0.3}$ &
15.35 & 6.16 &{\sl 3,1 } \\
A1651& 0.0825 & 6.3$^{ + 0.5}_{ -0.5}$ &
8.25 & 2.7&{\sl 3,1 } \\
A1650& 0.085& 5.6$^{ + 0.6}_{ -0.6}$ &
7.81 & 2.56 &{\sl 3,1 } \\
A2597& 0.085& 3.6$^{ + 0.2}_{ -0.2}$ &
7.97 & 2.59 &{\sl 3,1 } \\
A2142& 0.0899 & 9.$^{ + 0.2}_{ -0.2}$ &
20.74 & 6.14 &{\sl 11,1 } \\
A478& 0.09& 7.1$^{ +0.4}_{ -0.4}$ &
12.95 & 3.9&{\sl 3,1 } \\
A2244& 0.097& 7.1$^{ +1.2}_{ -1.2}$ &
9.09 & 2.28 &{\sl 2,1 } \\
\hline
\hline
\end{tabular}
\noindent $^{1}$ in {\sl keV}\\
$^{2}$ in 10$^{+44}$erg/s ($h = 0.5$)\\
$^{3}$ in 10$^{-11}$erg/s/cm$^2$\\
$^{4}$ the first reference is for temperature, the second for the luminosity\\
$^*$ inferred from a non-ROSAT measurement.\\
\caption{ \label{Table 1} X-ray temperatures, 90\% uncertainties, ROSAT 0.1--2.4 keV luminosities (h$ = 0.5$) and fluxes.}
${\bf Ref.}$
{\sl (1)} Ebeling et al. 1996;
{\sl (2)} David et al., 1993;
{\sl (3)} Markevitch, 1998;
{\sl (4)} Fukazawa et al., 1998 (temperature are given with the emission from the central $0.05h^{-1}$Mpc excluded);
{\sl (5)} Johnstone et al., 1998;
{\sl (6)} Arnaud \& Evrard, 1999;
{\sl (7)} Arnaud, 1994;
{\sl (8)} Yamashita et al., 1992;
{\sl (9)} Hughes et al., 1993;
{\sl (10)} Ebeling et al., 1998;
{\sl (11)} White et al., 1994;
\end{table*}
\end{center}
As it is clearly an advantage to have a survey which is not redshift
restricted, we have collected existing information on clusters with
$ f_x \geq 2.2 10^{-11}$ erg/s/cm$^{2}$ in the ROSAT
band 0.1--2.4 keV (and with $| b_{II}| > 20^\circ $). In practice, most
of the clusters were obtained from the XBACS sample complemented by a handful of non-Abell clusters which satisfy our flux requirement
(MKW3s, Virgo, Cl0336 and 0422--09). For all these clusters,
temperature measurements were found in the literature.
In the following analysis, the redshift range is still restricted to $z< 0.1$, because of
the possible substantial incompleteness of the sample at higher redshifts.
Our final sample contains fifty clusters. It is important to examine its completeness.
A classical test for checking the homogeneity of a survey is to use the
$V/V_m$ test (Schmidt, 1968). Applying this test, we find a mean value
of $V/V_m$ = 0.490, consistent with a homogenous sample.
The plot of $V/V_m$ versus temperature given in Figure \ref{fig:vvm}
does not reveal any unexpected trends.
The recent availability of the BCS and RASS samples allows us to check this
in a more quantitative way: at our chosen flux limit, four clusters
in the BCS and RASS1 are not present in our sample (RXJ0419, IIZw108, Z5029, NGC1550). None of these
(non-Abell) clusters has a measured temperature, but from their
luminosity, only two of them would have a temperature
greater than 4 keV. This is consistent with our sample being complete at a
level of 95\% for $T \geq 4$ keV, while the
completeness of the BCS and RASS is estimated to be better than 90\%.
Therefore, our estimated completeness is 85\% and we will treat it as
an essentially complete, X-ray selected ensemble of clusters, at least for clusters with temperature greater
than 4 keV. It is the largest homogenous sample of X-ray selected clusters
currently available for the estimation of the temperature
distribution function. Notice that the possible incompleteness of our sample
would imply that the inferred temperature function could be an
{\em under-estimate}.
At redshifts $z \geq 0.04$, our sample is similar to M98.
The temperature histograms of both M98 sample and ours are shown
in Figure \ref{fig:histoT2}. As one can see, the two histograms are
nearly identical on the hot end, but differ on the cool end
(for $T < 5$ keV). This difference is not surprising given the
restricted redshift range investigated by M98. However, it is important
to investigate the temperature function on the cool end since the number of
clusters in the two samples differs noticeably. Although M98 corrected
for the incompleteness of his sample, using a larger sample will result in a
statistically better estimate with reduced sensitivity to
systematics.
Such a sample has two main advantages: with a selection in the ROSAT band,
we are much less sensitive to the selection function of the 2.--10. keV band,
which obviously favors hot clusters (even if in principle the effect of
the selection function can be corrected for), and the number of clusters in
our sample is twice as large as the HA91 sample, leading to better
statistics especially on the cool end ($T \sim 4$ keV).
We have examined the temperature histograms for
clusters with fluxes in the range $ 2.2 -4.0 \; 10^{-11}$ erg/s/cm$^{2}$
and those with $f_{X} > 4.0 \;10^{-11}$ erg/s/cm$^{2}$.
We find that these two histograms differ noticeably in the same sense
as the HA91 and E-HA samples as discussed previously.
The reason of this difference is unclear, but it is clearly not due
to the 2.--10 keV selection only. Inspection of the redshift distribution does
not provide evidence that this could be due to large--scale structures
(although the BCS reveals significant fluctuations around $z \sim 0.05$).\\
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{vvm.ps}}
\caption{ \label{fig:vvm} $V/V_m$ distribution for clusters in our
sample as a function of temperature. No sign of
incompleteness can be seen from this plot for clusters with
$T \geq 3.5$keV.}
\end{figure}
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{histo4.ps}\includegraphics{histo3.ps}}
\hspace*{1.8cm}a) \hspace*{8.4cm} b)
\caption{ \label{fig:histoT2} Same as in Fig. \ref{fig:histoT1} for clusters
selected in the ROSAT band.
Figure a) represents the distribution of clusters in Markevitch's
sample. b) represents the
histogram
of the 50 clusters in our sample (XBACS clusters with flux
greater than $2.2 \; 10^{-11}$erg/s/cm$^2$ and redshift $z < 0.1$).
The shapes clearly differ for temperature smaller than 5 keV,
while they are nearly identical above. This difference is mainly due to the fact that
the Markevitch sample is restricted to redshift $> 0.04$.
}
\end{figure*}
\section{The Determination of the Temperature Function}
\subsection{The method}
Estimation of the luminosity distribution function $\phi(L)$ from a sample
for which the selection function is well known is relatively straightforward:
the observed number of clusters in the sample $\cal N$ in the luminosity range
$(L-L+\Delta L)$ (with $\Delta L \ll L$)
is the realization of a stochastic Poisson process with a mean value
$ \phi(L) V(L)\Delta L$, where $V(L)$ is the volume searched
for objects with luminosity $L$. When the selection function $s(z)$ is that
of a strictly flux limited sample, $V(L)$ is just the volume
$V_m$ corresponding to
the maximum distance at which the cluster would have been detected. If the
selection function $s(z)$ is more complex, then this volume can always be
computed as an integral:
\begin{displaymath}
V(L) = \int_0^{+\infty} s(z) dV(z)
\end{displaymath}
An unbiased estimation of $ \phi(L) \Delta L$ is therefore given by
${\cal N}/V(L) $. In the following, when $B$ is an estimator of some
quantity $A$, we use the notation below:
\begin{displaymath}
A \equiv B
\end{displaymath}
According to this notation:
\begin{displaymath}
\phi(L) \Delta L \equiv {\cal N}/V(L) = \sum 1/V
\end{displaymath}
where the summation is performed over all the relevant clusters in the sample.
This immediately provides an estimator of the integrated luminosity function:
\begin{displaymath}
\phi(>L) \equiv \sum_{L_i \geq L} 1/V(L_i)
\end{displaymath}
We can now recover the standard estimator of the temperature distribution
function:
\begin{equation}
N(T) \Delta T = \int_0^{+\infty}\phi(L,T) \Delta T dL \equiv \sum_{i} 1/V(T_i)
\label{eq:nesT}
\end{equation}
where the summation is performed over all clusters in the sample in the
relevant temperature range. It is important to notice that this estimator is
unbiased and fully accounts for the intrinsic dispersion in the
temperature-luminosity relation. This procedure may not always be used in
practice, if for some region of the parameter space in the sample $V(L) = 0$.
This is the case for instance, in a flux limited sample which is
further restricted to a redshift range. Because of this further restriction,
clusters fainter than some intrinsic luminosity may simply not be present in
the sample, consequently leading to $V(L) = 0$. In this case, the estimation
from Eq. (\ref{eq:nesT}) is clearly not appropriate anymore.
The contribution of the missing clusters can
however be estimated, provided the probability distribution
of clusters at a given temperature $p(L|T)$ is known. The observed number of
clusters is still a
Poisson realization of a process whose mean is given by:
\begin{eqnarray}
\nonumber \int_0^{+\infty}\phi(L,T) \Delta T V(L) dL = \int_0^{+\infty}N(T) \Delta T p(L|T)V(L) dL \\
= N(T)\int_0^{+\infty}p(L|T) \Delta T V(L) dL = N(T) \Delta T V_s(T)
\end{eqnarray}
The last equality defined $V_s(T)$, the effective average volume per cluster of given temperature in the survey. Therefore:
\begin{displaymath}
N(T) \Delta T \equiv \sum_{i}1/V_s(T_i)
\end{displaymath}
This estimator has been used by several authors (Eke et al., 1998; M98).
It is unbiased {\em provided that $V_s(T)$ is known exactly}.
In practice, this is of course not the case, since the statistical description
of the temperature-luminosity relation and the distribution $p(L|T)$ are not
perfectly known, which leads to a source of {\em systematic uncertainty}.
Consequently, a specific choice of $p(L|T)$ implies a bias in the above
estimator. Furthermore, as fewer clusters are
actually present in the sample, the Poisson noise increases.
It is clear therefore, that this approach should be avoided when
possible, unless one can be confident that the resulting uncertainties
remain small.
A second important issue concerns errors on
temperature measurements: when the true underlying cluster
distribution is very steep, positive
errors will move many more clusters upwards in temperature than
negative errors move hotter clusters downwards, just because of
the difference in the intrinsic abundances. This effect has been
pointed out by Evrard (1989) for velocity dispersion measurements.
A mean statistical correction was applied in the modeling
by Eke et al.(1998), while Viana and Liddle
(1999a) estimated the magnitude of this effect using a bootstrap
resampling technique on Henry's sample. It is
important to realize that when the errors on individual measurements are not
identical, a mean correction is not necessarily
sufficient, because the weight $1/V_i$
varies significantly from one cluster to another. A further problem is
that, in principle, the error on the temperature might correlate with
the apparent flux of a cluster, as fainter clusters will have lower
signal--to--noise and will be those having the larger $1/V_i$ as well.
In such cases, only a Monte-Carlo reproducing the exact
conditions of the observations including the different integration times
for the different sources, would in principle allow a complete separation
of the various effects of temperature measurement errors.
For the local temperature distribution function, the errors are small
enough that the above issue is not a major
problem. However, due to larger temperature errors this might be a more critical point for the high--redshift sample.
The bootstrap method used by Viana and Liddle
(1999a) is certainly well adapted. Another possible way to solve
this problem is to use a Bayesian approach: we can take as a prior
the distribution function of X-ray temperatures assuming
$N(T) \propto T^{-\alpha}$, with $\alpha \approx 5$.
Given that a cluster is observed with a measured
temperature $T_0$ and assuming that the errors are log-normal distributed
with a dispersion $\sigma$ (corresponding the observed $1 \sigma$ uncertainty), the a posteriori probability
that the actual temperature
of the cluster be $T$ is given by:
\begin{equation}
p(T|T_0) \propto N(T) \exp( -(\ln (T) - \ln (T_0))^2/2.\sigma^2)
\end{equation}
this distribution is also log-normal with the same dispersion but
with the most likely temperature $\overline{T}$ shifted compared to $T_0$:
\begin{equation}
\overline{T} = T_0 \exp( - \alpha \sigma^2)
\end{equation}
The above formula may appear strange, because it seems to suggest that a
systematic bias exists in the temperature measurements. This is actually
not the case, since the observed number of clusters within the sample
is rather constant among the different temperature bins. Only the estimated
density is biased.
Another effect to take into account for cosmological applications (see section 5.3) when one wishes to relate the mass function to the temperature distribution function, is the dispersion in the $M-T$
relation. The importance of this effect has been discussed by Eke et al (1998). They assumed an
intrinsic dispersion of 20\%, but neglected the effect
of the errors on the actual measurements considering them to be
smaller.
However, as the effect goes as the square of the error, one should be
cautious when dealing with these corrections: in practice, a
20\% error (or dispersion in the $M-T$ relation)
begins to make a significant difference
in the inferred abundance of clusters, while an amplitude of 10\% leads to a
change that is essentially negligible. In their numerical simulations,
Bryan and Norman (1998a) found an intrinsic dispersion of
the order of 10\% in the $M-T$ relation, twice smaller than the dispersion
adopted by Eke et al.(1998) ($\approx 20\%$). In order
to understand the bias introduced by these effects (errors on temperature
measurements and dispersion in the $M-T$ relation), the Bayesian correction
is illuminating: as we have seen, such a correction is equivalent to
a modification of the observed temperature, that is to say, the
mass-temperature relation in the modeling (this is only an
approximation because the Bayesian term depends on the actual slope
of the N(T), which might vary with T and redshift, but this is a second
order effect). The corrections to implement for observational errors
are slightly different in
nature though: errors vary from one cluster to another, and also may vary in a
systematic way, for instance accordingly to the apparent flux. Furthermore, since
the correcting factor varies in a nonlinear fashion, a
mean correction is inadequate.
Finally, it is worth noticing that errors for high redshift clusters are
larger than for the low redshift clusters, and therefore a larger
correction is necessary. The
practical implication of such a change in the modelling will be discussed
is section 5. \\
\subsection{Temperature Distribution Function}
We compute the integrated temperature distribution from
our sample following the method described in the previous section, i.e. using the estimator:
\begin{displaymath}
\nonumber
N(>T) \equiv \sum_{\overline{T}_i > T}1/V(\overline{T}_i)
\end{displaymath}
The resulting temperature distribution function is plotted as the
continuous thin line in Figure
\ref{fig:nt1a}. A smoothed version is also given. We have also checked that using
temperature when the central emission is removed makes negligible difference.
For comparison,
we have also estimated the temperature distribution
function from the HA91 sample, using available updated temperature
measurements and the standard estimator (Eq. \ref{eq:nesT}) without
the Bayesian correction factor. This temperature distribution is shown as
the dotted line in Fig. \ref{fig:nt1a}. As one can see, the difference is quite noticeable, in
particular around 4 keV, where it is roughly a factor of two. We also show the temperature distribution inferred from the BCS luminosity function
(Ebeling et al, 1997) assuming the $L_x-T_x$ relation
as given by OB97. A convenient fit of our temperature distribution function is:
\begin{equation}
N(>T) = 1.42 10^{-6}h^3 {\rm Mpc}^{-3} \left(T/4. {\rm keV}\right)^{-1.6}\exp \left( -(T-4. {\rm keV})/2.6 \right)
\label{eq:ourNT}
\end{equation}
which is plotted in Figure \ref{fig:nt1b}.
We also show the fit to the temperature distribution function derived by M98.
As one can see, we find
a larger abundance of cool clusters ($T \leq 4$ keV) than
M98 while we have a nearly
perfect agreement with the temperature distribution function
inferred from the BCS luminosity function.\\
The difference between our temperature distribution function and
previous estimations is striking. The
difference with M98 can be easily understood -- his sample being
restricted to the redshift range $z > 0.04$, a significant fraction of
faint (and cool) clusters are missing. He has corrected for this
incompleteness by a weighting scheme based on an estimated dispersion in the
$L_x-T_x$ relation and following the method discussed in section 4.1. However, in
order to properly evaluate such a correction, one needs an accurate estimate
of the dispersion in the $L_x-T_x$ relation, which can only be obtained
from a flux--limited sample. The difference with the original HA91 sample
is more difficult to understand. One may argue that the
difference comes essentially from the band selection ($2-10$ keV ), however as we have already mentioned in section 2, comparison with the
E-HA sample suggests another origin.
We have therefore divided our sample into two equal
sub-samples, corresponding to the brightest and the faintest
clusters (in apparent flux). Again, since both samples are X-ray
selected (in the same band), they should be statistically equivalent.
The corresponding temperature distribution functions are shown in Figure
\ref{fig:nt1b}. The bright sub-sample leads to a temperature distribution
function close to the one based on the HA91 sample. This is not surprising
on the high temperature end, since the clusters are almost the same in
these two samples. The fact that the temperature distribution functions
are almost the same for the cooler clusters as well, indicates that the
original HA91 sample does not suffer from any significant bias and that
the effects of the chosen band are relatively small. On the other hand, the
difference between the faint and the bright samples is quite significant,
and the relative abundances of hot and cool clusters is noticeably different.
As the difference is marginally 1 $\sigma$, these two
samples could be considered as representing two realizations which are
slightly in the tail of the distribution, without being really suspicious.
Nevertheless, such a difference have some implications in practice as we will see in the next section. The fact that our temperature distribution function is in good agreement
with that inferred from the BCS luminosity function is rather encouraging
and confirms the absence of any systematics that would undermine our analysis.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{nt1a.ps}}
\caption{ \label{fig:nt1a} The integrated temperature distribution function
inferred from our sample is given by the continuous (red) thin line. A
smoothed version is also given (thick line). The dotted line is
the same quantity for the original HA91 sample estimated by
means of Eq. \ref{eq:nesT}.
The dashed (green) line is derived from the BCS luminosity function.
The error bars represents the 68\% interval for the distribution of the estimator from the bootstrap.}
\end{figure}
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{nt1b.ps}}
\caption{ \label{fig:nt1b} A fit to our temperature distribution function
is given by the continuous (red) thick line. The two thick dashed lines
correspond to two sub--samples: the short dashed line corresponds to the bright one ($2.2 10^{-11} > s_x > 4. 10^{-11}$ erg/s/cm$^2$), the long dashed corresponds to the faint sub-sample ($s_x > 2.2 10^{-11}$ erg/s/cm$^2$). The long-dashed-3-dotted
(green) line is the temperature distribution function from Markevitch (1998).}
\end{figure}
Differences such those we have found have significant
consequences when constraining the spectrum of the
fluctuations from the temperature distribution function: a sample in which
the abundance of cool clusters is underestimated leads to a too flat
spectrum and the amplitude of fluctuations at cluster scales will be also underestimated. All these effects enter as a
source of systematic uncertainty in the estimation of $\Omega_0$.\\
\subsection{ Distribution function of the estimator and Confidence intervals }
It is useful to have a statistical description as comprehensive as
possible in order to derive strong constraints on models. The number
$\cal N$ of clusters in a given temperature range can be regarded as a
Poisson realization of mean $\int_0^{+\infty}$ $\phi(L)$ V(L) $\Delta$ L.
However, the understanding of this quantity requires good information
on the $L_x-T_x$ relation: its shape, its normalization and the distribution function around it. In order to determine the expected distribution of
our estimator for $N(T)$, we use a Bayesian bootstrap technique that
allows us to avoid the use of a specific $L_x-T_x$ relation, getting rid of some possible bias thereby introduced. We therefore
assume that the distribution of the estimator $ \tilde{N}(>T)$ for a given
density $ N(>T)$ is of the following form:
\begin{equation}
D(\tilde{N})d\tilde{N} = d(\lambda)d\lambda
\label{eq:distri}
\end{equation}
with $\lambda = \tilde{N} / N(>T)$. We then use a Bayesian bootstrap to
reconstruct the distribution function $d$: $10^{4}$ fake samples are built
from the original sample, each having the same mean number of clusters, but
with a given dispersion to account for Poisson noise.
This procedure was used to find the distribution function of the
estimator of $N(>T)$, as well as of the estimator for $N(>T_1)-N(>T_2)$.
The latter quantity is easier to handle with in a statistical analysis,
as different temperatures can be chosen in such a way that the
measurements are independent
quantities (typically $N(>T_1)-N(>T_2)$, $N(>T_2)-N(>T_3)$,...).
We found that a $\chi^2$ distribution function fits quite well the
distribution function of the
estimator, provided that the number of degrees of freedom is left as a free
parameter (and is not necessary an integer). An example is given in
Figure \ref{fig:distri}. As one
can see, the $\chi^2$ distribution provides an adequate fit to the overall
distribution. From this we can infer confidence intervals on the density.
The error bars given in Figure \ref{fig:nt1a} reflect the 68\%
range of the value of the estimator of the density of clusters.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{distri00.ps}}
\caption{ \label{fig:distri} The finite size of the sample
implies that our estimate of the abundance of clusters has some uncertainty.
The distribution of values that our estimator can take, given the sample we
have, is estimated by a Bayesian bootstrap.
Here we illustrate the histogram of the values taken
by our estimator of the
abundance of clusters at present epoch with temperature above 4 keV . A convenient analytical fit of
this distribution is also shown: a $\chi^2$ distribution with 6 degrees of freedom.}
\end{figure}
\section{ Cosmological Applications}
\subsection{Mass-temperature relation}
In order to relate the observable properties of clusters to their
mass, physical correlations are necessary. As discussed in the
introduction, the temperature of the X-ray gas is probably the most
convenient cluster mass estimator. Kaiser (1986) developed the scaling
arguments which are essential to this approach (they were already used
in 1980 in the pioneering by Perrenod, 1980).
Some of the scaling arguments are exact, and they are remarkably powerful.
In an Einstein--de Sitter universe with initial fluctuations described by a power--law spectrum, the only
process acting is gravity and therefore for
the physics of the gas, scaling arguments will hold as long as no other
mechanism plays an important role (like cooling or bulk
heating). In such a situation, the only possible scale of the problem is
associated with the non-linear scale, even in the presence of
shock heating. This guarantees that the
$T-M$ relation for $M_*$ clusters follows an exact scaling law (for a given
spectrum):
\begin{displaymath}
T_* \propto M_*/R* \propto M_*^{2/3}(1+z)
\end{displaymath}
It is reasonable to expect that such a relation will also hold between
different masses forming at a given redshift and that is nearly
independent of the power spectrum. However, this remains an approximation
as the geometrical aspect of the peaks corresponding to different masses at
a given time may differ (due to the difference in the height of
the peaks). The same remark applies for two different spectra.
Further checks are therefore needed.
Analytical modeling of clusters adopting hydrostatic
equilibrium is often completed by the isothermal assumption.
However, such a mass estimator
can be misleading, even when temperature profiles are taken into account
(Balland \& Blanchard, 1997). A further problem is the fact that a
significant part of the pressure support of the gas could come
from turbulent motions in the gas (Bryan and Norman, 1998b). Furthermore,
the scaling laws are not expected to hold exactly in open models, significant departures may actually exist in
this case (Voit \& Donahue, 1998). It seems therefore safer to rely on
numerical simulations in order to establish the mass-temperature relation.
There are several ways to define the mass of clusters, which are not
actually objects with well defined and sharp boundaries.
Considerable progress has been made in recent years thanks to Navarro, Frenk and White (1996), who showed that clusters are well fitted by
a universal profile involving few parameters (which is not what would be
expected based on the isothermal hydrostatic $\beta$-model). Still,
various definitions of the mass are used, resulting in some confusion
when one wishes to make detailed comparisons. The mass of a cluster could be defined as the mass within the so-called virial radius corresponding to a spherical region with a contrast
density $\delta=178$ (in the Einstein-de Sitter case). The mass can also be defined by $\delta=200$, $\delta=500$, or by the mass within a fixed physical radius such as the Abell radius or its equivalent in comoving coordinates. As
most of the numerical checks of the mass function have been performed at
the virial radius, it is certainly safer to estimate the mass at this
very radius. The validity of the scaling
models has been impressively demonstrated (Evrard 1989), and no noticeable difference
has been detected among different spectra, masses and cosmological backgrounds.
Furthermore, an intensive comparison of the results from various codes
has verified that different numerical techniques do not lead to significant
differences in inferred properties (Frenk et al., 1999). This is
especially true for the $M-T$ relation. Even the inclusion of a
significant energy injection does not introduce significant changes in
these relations (Metzler and Evrard, 1998). The recent numerical work of
Bryan and Norman (1998a) is certainly
the most advanced in this area, and the scaling relations they obtained are
impressive (see their figure 4) and convincingly independent of the
underlying cosmological model. Their result can be summarized by the
fact that
the following mass-temperature relation:
\begin{displaymath}
T_v = 3.8 {\textrm keV}
M_{15}^{2/3}(\Omega_0\Delta/178)^{1/3}(1+z)
\end{displaymath}
(given for $h =0.5$) reproduces well the results of their numerical simulations ($M_{15}
\equiv M/10^{15}$ solar masses). The statistical
dispersion they found around this relation is only 10\%. The slope
$2/3$ might not be guaranteed to better than 15\%. However,
the normalization of the above relation in the various numerical
simulations can differ (for instance, Evrard et al., 1996 found a value
20\% higher than Bryan \& Norman, 1998a, while the mean value obtained
by Frenk et al., 1999, is only 5\% higher with a 5\% dispersion among
the various simulations). The existence of turbulence on
small scales (Bryan and Norman; 1998b) indicates that one should be
cautious on this issue, as the modeling of turbulence is often
limited by the resolution of the simulation. Finally, it is
important to notice that such masses are 70\% higher than what
would be inferred from the isothermal hydrostatic equation (Roussel et al., 2000) and more than twice the mass inferred recently from
decreasing temperature
profiles as considered by M98, or more recently by
Nevalainen et al. (1999). In our modeling, we
have adopted the above normalization, corrected for the effect of
introducing a 10\% dispersion in the $M-T$ relation:
\begin{displaymath}
T_v = 4.0 {\textrm keV} M_{15}^{2/3}(\Omega_0\Delta/178)^{1/3}(1+Z)
\end{displaymath}
($h =0.5$). We will examine the consequences of any change in this relation.
\subsection{Constraining the models}
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{S-Om.ps}}
\caption{ \label{fig:sigom} Contours giving the allowed range of values for
the amplitude of matter density fluctuations on cluster scales. The
continuous line is the 68\% contour, the dotted line is the 90\% contour,
and the dashed line is the contour at 95\%.}
\end{figure}
The local temperature distribution function can be used to constrain the
properties of the matter density fluctuations. Although this
can also be applied to a non-Gaussian scenario (OB97; Robinson et al., 1999),
we will restrict ourselves to Gaussian fluctuation models only.
This consists basically in finding the
best parameters and confidence intervals from a fit to the
temperature distribution function. However, there is a degeneracy
between different cosmological models: equally satisfying fits can be
obtained in either low- or high--density
models (OB92). Nevertheless, the density fluctuation amplitude can
be usefully constrained for a given model. It is generally assumed
that cluster abundances lead to a determination of the fluctuation
amplitude at $8h^{-1}$Mpc and therefore the normalization is communly
expressed in term of $\sigma_8$. This is not completely correct.
In an $\Omega_0 = 1$ universe, the temperature
distribution function of clusters does actually provide
the normalization of the fluctuations on this scale. This comes from
the fact that the mass of a $8h^{-1}$Mpc sphere collapsing today will
lead to a 4.5 keV cluster, which is a typical temperature of a moderately
rich cluster for which X-ray
samples provide a reasonably well established abundance
(at least this what the authors now believe!). In contrast, in a
low--density universe ($\Omega_0 \sim 0.3$), a $8h^{-1}$Mpc sphere will
produce a 2 keV cluster, for which we actually do not yet have reliable
abundance estimates. This is not just a semantic difference,
since for a low--density universe ($\Omega_0 \sim 0.2$), the amplitude of
$\sigma_8$ could
differ by up 50\% (OB97). Following Blanchard \& Bartlett (1998),
we will use the following definition:
\begin{displaymath}
\sigma_c = \sigma(\Omega_0^{-1/3}8h^{-1}\textrm{Mpc})
\end{displaymath}
and give constraints in terms of $\sigma_c$. This has the additional
advantage
of being independent of the power spectrum: as $\sigma_8$ and the power index $n$ are
correlated, the uncertainty on $\sigma_8$ is larger than on $\sigma_c$. In order to illustrate this point we have
plotted both quantities in Fig. \ref{fig:norm}.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{S8n.ps}\includegraphics{Scn.ps}}
\caption{ \label{fig:norm} Contours giving the range of allowed values for
the index $n$ of the primordial matter fluctuation spectrum on cluster scales
for a Gaussian distribution function versus the normalization expressed in term of $\sigma_8$ (left side) and in term of $\sigma_c$ for a density parameter
$\Omega_0 = 0.4$. The continuous lines are
the 68\% contours, the dotted lines are the 90\% contours, and the
dashed lines
are the contours at 95\%. }
\end{figure}
To find parameter constraints, we find the best model according to
the following likelihood function:
\begin{equation}
{\cal L} = \prod p\left(N(>T_i,z_i)-N(>T_{i+1},z_i)| N_{\{\Omega_0,n,\sigma_c\}}(>T_i,z_i)-N_{\{\Omega_0,n,\sigma_c\}}(>T_{i+1},z_i)\right)
\end{equation}
in which $N$ is the measured abundance and $N_{\{\Omega_0,n,\sigma_c\}}$ is the value given
by the model, which depends on the density parameter $\Omega_0$,
the local slope of the power
spectrum $n$ and its amplitude specified by $\sigma_c$.
The probability distribution function $p$ cannot be inferred from
first principles and is therefore estimated with our bootstrap procedure.
The maximum value of the likelihood has been normalized to unity, and
confidence intervals on the parameters were estimated from the
contours at ${\cal
L }= 0.6, 0.26, 0.13$ corresponding respectively to the 68\%,
90\% and 95\% contours (strictly speaking this holds only for normal distributions). The constraint on the shape of the spectrum
$n$ is rather poor. We find a broad range of possible values.
The only obvious result is that models with $\Omega_0 = 1$ and
$n\sim -1$, representing the standard Cold Dark Matter spectrum,
are excluded, as previously found (Blanchard \& Silk, 1991;
Bartlett \& Silk, 1993). The
effect of statistical uncertainties on the amplitude of matter
fluctuations on cluster scales is quite small, and therefore only
a restricted range of values is allowed for a given $\Omega_0$.
The dependence on $\Omega_0$ is weak :
\begin{eqnarray}
\nonumber \sigma_c \approx & \: 0.6~^{+0.035}_{-0.025} \:\Omega_0^{-0.14} \: (95\%)& \textrm{open case}\\
\sigma_c \approx & \:0.6~^{+0.035}_{-0.025} \:\Omega_0^{-0.18} \: (95\%) & \:\textrm{flat case}
\end{eqnarray}
(uncertainties are statistical and are given for $\Omega_0 = 1$).
Different values have been published for the amplitude obtained from the
cluster normalization. These differences are mostly due to different cluster abundances used. Our
normalization is slightly higher than what is often derived from abundances
of X-ray clusters (but in better agreement with some normalizations based on
optical surveys, e.g., Girardi et al, 1998). This is due to
the fact that our number density of clusters at 4 keV is higher than previous estimations and that we used the Bryan and Norman (1998) normalization of
the $M-T$ relation which is slightly lower than that of Evrard et al. (1996). Finally, the differences with previous works concerning
the dependence on $\Omega_0$ come from the fact that we
refer to $\sigma_c$, which is the normalization on the true cluster scale rather than to $\sigma_8$, which depends on the assumed
spectrum (for $n =-1.5$ we find $\sigma_8 = 0.96$ in an open $\Omega_0 =0.3$
universe).
It is interesting to try to evaluate the magnitude of possible
systematic uncertainties on $\sigma_c$. As we have
argued, the local cluster abundance now seems well established,
so that we do not expect this to be a source of systematic error
larger than the statistical uncertainty already quoted. The remaining
dominant uncertainty comes from the use of the Press and Schechter mass
function. Different results exist in the literature concerning
its accuracy to describe the actual mass function
(Lacey \& Cole, 1994).
Numerical simulations mostly suggest that the mass function is well described
by the Press and Schechter formula: for instance,
Borgani et al. (1998) have shown that it can be safely used for cluster modeling in arbitrary cosmology.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{n-Om.ps}}
\caption{ \label{fig:nom} Contours giving the range of allowed values for
the index $n$ of the primordial matter fluctuation spectrum on cluster scales
for a Gaussian distribution function. The continuous line is
the 68\% contour, the dotted line is the 90\% contour, and the dashed line
is the contour at 95\%. }
\end{figure}
\subsection{Estimating $\Omega_0$}
The determination of the mean density of the universe is obviously
a very important goal of observational Cosmology. Although it is
fashionable to claim that current data favor a low value, it seems to
us that the observational situation has not improved that much since the
1980's. Indeed, bulk flows on large scales lead to somewhat contradictory
results and interestingly enough,
fluctuations in the CMB seem to favor flat models and to exclude open
ones (Lineweaver et al., 1997; Lineweaver and Barbosa, 1998; Le Dour et al., 2000).
Perhaps the most stringent constraint comes from the baryon fraction
observed in clusters (White et al, 1993). Still, a firm upper limit on
$\Omega_0$ is lacking, because of the uncertainties in primordial
nucleosynthesis and cluster mass estimates (notice that
virial masses estimated from the scaling models with the normalization we adopt are
nearly twice as large as those based on hydrostatic estimations!).\\
The evolution of clusters is known to be a powerful probe of the mean
density of the universe (Perrenod, 1980; Peebles et al. 1989). Evolution
of the cluster abundance with redshift offers a new pass to $\Omega_0$ (OB92, OB97) which is
insensitive to the cosmological constant, and above everything else
is really a global probe rather than a local one, as it is the case with the classical $M/L$ argument. It has been shown that the EMSS
distribution can be modeled equally well within low- and high--density
universes, provided that some evolution in the
$Lx-Tx$ relation is adopted (OB97). The required evolution depends on
$\Omega_0$ as follows:
\begin{equation}
L_x \propto T_x^3 (1+z)^B
\end{equation}
with
\begin{equation}
B \sim 4 \Omega_0 - 3.
\end{equation}
It is important to ask whether this modeling based on limited
observational data is robust or not. Since this work, the
observational situation has been considerably improved,
thanks to the ROSAT satellite.
The number counts for X-ray clusters are now available and fall well within the range of the predictions given by OB97.
The ROSAT X-ray cluster luminosity function was derived from the
BCS sample (Ebeling et al. 1997) and recently confirmed by De Grandi et al.
(1999a). This observed luminosity
function matches extremely well the one predicted by OB97. Finally, the amplitude of the fluctuations which was
normalized to the original HA91 temperature distribution function, is in good agreement with what has been obtained in the present work. This
considerably reinforces the strengt of OB97 modeling. Reichart at al. (1999) have recently re-analyzed the EMSS sample and have reached similar conclusions. Furthermore, similar value of $B$ has been obtained
by Borgani et al. (1998) from their analysis of the RDCS survey.
The determination of $\Omega_0$ from
the evolution of the $L_x-T_x$ relation seems therefore possible,
and we expect it to lead to reliable estimates. Thanks to the availability
of ASCA temperature measurements for a significant number of high--redshift
clusters, it has been shown that this relation undergoes at most only
a mild evolution (in the positive direction, i.e., high--redshift clusters
are more luminous than local ones with the same luminosity; Sadat
et al., 1998). Using the relation between B and $\Omega_0$ (eq. 9),
Sadat et al. (1998) concluded in favor of a high value of the density parameter ($\Omega_0 \sim 0.8$) with a quoted uncertainty of $0.2$
(this uncertainty only reflects the uncertainty due to the Poisson noise
in the EMSS distribution, and does not
take into account the uncertainty in the overall modeling). \\
Thanks to the availability of the first X-ray selected sample of
high--redshift clusters with measured temperatures (H97), it is now
possible to constrain the density of the universe using directly the evolution of the X-ray cluster abundance to an intermediate
redshift. A value of $\Omega_0 = 0.5$ was obtained (H97) (this value has been abusively referred
to as a low $\Omega_0$.). Eke et al. (1998) who studied in detail the
various systematic uncertainties have found a similar $\Omega_0$. A higher value of $\Omega_0$ has been derived by Viana and Liddle (1999a) who discussed the importance of a proper correction from the uncertainties introduced by temperature measurement errors. Since these authors used a
lower local abundance than the one we have inferred from the present work, it is interesting to re-address
this question. Our primary goal here is to analyze the consequences
of the higher cluster abundance at zero redshift that we derive.
We leave to a future work a more complete investigation of the
high--redshift sample.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{nt2.ps}}
\caption{ \label{fig:nt2} The integrated temperature distribution function
at $z = 0.33$ inferred from Henry's sample is given by the continuous thick (red)
line (corresponding to the case $\Omega_0 = 0.5$). The two triangles are
the estimation from Viana \& Liddle (1999a, 1999b) with the original H97
sample.
The line is the best fitting model. The continuous grey (green) line is the fit to our local temperature distribution function. The dashed grey (green) line is our best model at $z =0.33$.
}
\end{figure}
We have estimated the temperature distribution function
of the distant cluster sample as given by H97, and estimated the
suitable volume correction for the various cosmological models. Temperatures
were corrected by the Bayesian term. The resulting temperature distribution
function is given in Figure \ref{fig:nt2}.
For comparison, we have also plotted the temperature distribution function
inferred from the local luminosity function. One may worry that
our Bayesian correction is adequate. The bootstrap approach followed by
Viana and Liddle (1999a) is certainly well adapted to the treatment of
measurement errors. We have therefore compared their inferred abundances
to ours and found very good agreement (using the 10 original clusters).
This confirms that our analytic method provides a correction term which is
as good as the bootstrap resampling technique. Our inferred temperature
distribution is shown as the thick continuous line
in Figure \ref{fig:nt2}.
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{distri12a.ps}
\includegraphics{distri12b.ps}}
\hspace*{2.8cm}a) \hspace*{6.4cm} b)
\caption{ \label{fig:distriz03} Figure a) represents the histogram
of the values taken by our estimator from $10^4$ bootstrap resampled samples (as in figure \ref{fig:distri}). The spikes are real
and reflect the fact that the abundance estimation is dominated by a single
cluster. In this case, our analytic expression for the distribution function
does not work very well (continuous line :
$\chi^2$ distribution with 1.5 dof). b) represents the distribution function of
the estimator with larger bins; in this case the fit is acceptable.}
\end{figure*}
We use the following likelihood function to infer the best-fit
value of $\Omega_0$:
\begin{eqnarray}
\nonumber \lefteqn{
{\cal L} = \prod_{i} p\left(N(>T_i,z_i)-N(>T_{i+1},z_i)| N_{\{\Omega_0,n,\sigma_c\}}(>T_i,z_i)-N_{\{\Omega_0,n,\sigma_c\}}(>T_{i+1},z_i)\right)
}\\
& & \prod_{j} p\left(N(>T_j,z_j)-N(>T_{j+1},z_j)| N_{\{\Omega_0,n,\sigma_c\}}(>T_j,z_j)-N_{\{\Omega_0,n,\sigma_c\}}(>T_{j+1},z_j)\right)
\end{eqnarray}
with $T_i=\{3., 4., 5.4, 6.6, 8., +\infty, z_i = 0.05 \}$ and
$T_j=\{3.6, 4.4, 5.5, 6.7, 8., +\infty, z_j = 0.33\}$.
The distribution function of
the measured abundances was estimated by a Bayesian bootstrap, as
discussed in section 4.3. Because of the small number of clusters in
the sample, the inferred abundance looks very spiky (see Figure
\ref{fig:distriz03}a), which simply means that one cluster
essentially dominates the statistic. In such a case, our fits are of course
never as good as in the case where the number of clusters is larger,
but they are reasonably acceptable when the binning in abundance is
enlarged (see Figure \ref{fig:distriz03}b).
The
likelihood function normalized to unity, is given in Figure \ref{fig:Lom}.
The most likely values of $\Omega_0$ are 0.92 (open case) and 0.865 (flat case).
The shape of the likelihood function
is very well fitted by a Gaussian, even if the probability functions we have
used are significantly non-Gaussian. We can therefore make direct
use of this function to give the confidence intervals at the $1 \sigma$ level.
Our best fitting values are :
\begin{eqnarray}
\nonumber \Omega_0 = & 0.92^{+0.255}_{-0.215} & \mbox{ {\textrm (open case)}}\\
\Omega_0 = & 0.865^{+0.35}_{-0.245} & \mbox{ {\textrm (flat case)}}
\end{eqnarray}
This is very consistent with Sadat et al. (1998) value
from a completely independent analysis. The constraint we have
obtained is quite severe: an open model with $\Omega_0 < 0.49$ is ruled
out at the 95\% confidence level, a flat model with $\Omega_0 < 0.37$ is ruled out at the same level, conclusions which are in clear
disagreement with previous analyses based
on the same high redshift sample (H97; Eke et al, 1998). The main differences come
from our higher abundance at low redshift and from the fact that we explicitly take into account the effect of temperature measurement errors
in the high--redshift sample. Note that Viana and Liddle (1999a) reached values which are consistent with ours. There are a number of
other issues that differ in these various analyses and it is important
to check whether these differences can result in significantly different
values of $\Omega_0$. Eke et al. (1998) have analyzed various sources of systematics
and concluded that they are not of critical importance.
We reach similar conclusions for the effects they have investigated.
For instance, we have modified the $M-T$ relation
by $\pm$ 20\%, which changes $\Omega_0$ by $\pm 0.1$. The uncertainty in
the validity of the Press and Schechter formalism is also significant
but small. Using the revised version by Governato et al. (1998), we found values arround 15\% higher. We have checked that our likelihood approach
is not biased by applying it to the predicted mean abundance in one model
($\Omega_0 = 0.85$, $n = -1.85$,
$\sigma_c = 0.606$) and found that the best fitting model is ($\Omega_0 = 0.85$, $n = -1.75$,
$\sigma_c = 0.61$). We also have checked that varying the shape of the fitting expression does not change the final likelihood at any appreciable level.
After this work was
finished, we learned about Donahue \& Voit (1999) and Henry (2000) works. The first authors found a significant evolution in the abundance of clusters over the redshift range [0.-0.8], in agreement
with Blanchard \& Bartlett (1998) and the present work. Nevertheless, they conclude it to be consistent with a lowdensity universe, but simultaneously they found a very flat spectrum $n \sim -2.2$, which is
excluded in our analysis based on the
local temperature distribution function. Similarly, Henry (2000) found a moderate value of $\Omega_0 \sim 0.5 $ using the HA sample. Indeed, the
strongest source of
systematic error we found comes from the reference sample we used at
zero redshift: using the abundances inferred from the sample limited at
the bright end ($f_x > 4. 10^{-11}$ erg/s/cm$^2$) leads to $\Omega_0 = 0.65$ consistent with Eke et al.(1998) and Henry (2000). With the fainter part of our sample
($2.2 10^{-11}$ erg/s/cm$^2$ $<f_x < 4. 10^{-11}$ erg/s/cm$^2$) we obtained
$\Omega_0 = 0.98$. The difference obtained just by dividing the sample
into two statistically equivalent sub-samples is surprisingly large (althouhg one may not worry of a difference at 1.5 $\sigma$ level):
one would expect the uncertainty to come primarily from the
high--redshift sample (which comprises only nine clusters) and not from
the low--redshift sample.\\
Our best fit value of $\Omega_0$ inferred from the H97 sample is significantly
higher than previous estimates from the same sample but it is consistent with the latest optical result by Borgani et al. (1999).
It is therefore important to examine the
robustness of our analysis. This disagreement with previous results can be due to a higher local abundances of X-ray clusters, and a somewhat different treatment of
the bias introduced by the errors at high redshift. As we have already argued, it seems
unlikely that our local sample leads to a significant overestimation of
the local temperature distribution function. In fact, one could argue that
we may underestimate the actual $N(T)$, because of the possible
incompleteness of our sample. However, regarding the agreement between
our local temperature distribution function and what can be inferred from
the luminosity function this is rather unlikely.
The high redshift sample might be more worrisome. As we have seen, a
systematic difference may exist between two samples, leading to larger
differences in $\Omega_0$ than expected from Poisson noise. It is therefore
conceivable that the high redshift sample is a statistical fluke. For
instance, Eke et al. noticed that the temperature distribution within the
redshift bin [0.3--0.4] in the original Henry's sample is statistically
surprising as 9 of the clusters lie in the range [0.3--0.35] (now 8 are left). Another
possibility is that high--redshift
clusters are more massive than what one would infer from their apparent
temperature (averaged over the luminosity which comes essentially from
the core region). This could be possible for instance, if high--redshift
clusters are more often dominated by cooling flows (or if the selection
procedure favors cooling flow clusters), making them appear
cooler than they actually are and producing an apparent evolution in
the $M-T$ relation.
%The discrepancy (if real) between X-ray and lensing masses observed for some %clusters seem to confirm such effect.
At the same time, in order not to produce an evolution in the $L -T$
relation (not seen in the data), these high-z clusters would have to
be fainter. This is not very compelling, since cooling flows are expected
to increase the X-ray luminosity.
Another potential problem in the above determination of the density
parameter is the quality of the EMSS sample: if the selection function is
not well understood, it could be that the sample is missing a significantly
larger fraction of the cluster population than expected. Indeed, clusters
selection in EMSS is rather problematic, as the detection algorithm
was designed to detect point sources. A mean correction for extended
sources was applied (see Gioia and Luppino, 1994 and OB97 for more details),
but one may nonetheless worry that this procedure is not well controlled.
However, the deficit of high-z clusters observed in Figure \ref{fig:nt2}
is of the order of 2 to 3. The possibility that the EMSS selection procedure
could have missed clusters to such an extent seems unlikely. Furthermore,
the modelling by OB97 {\em predicted } number counts which are
subsequently seen to be in good agreement with available ROSAT
counts argues against significant incompleteness of the EMSS sample.
An other potential problem
could be a systematical bias in the EMSS fluxes. No evidence has been found
by Nichol et al. (1997). However, Ebeling et al. (1999) claim that a
40\% offset in flux exists, which could explain half of the observed dimming.
This possibility does not seem very appealing, because it would imply a
very significant evolution in the $L_x-T_x$ relation. In order to get rid of
possible limitations of the EMSS sample, it will clearly be very
important to see whether consistent results could be obtained from
ROSAT selected samples of X-ray clusters.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{L-Om.ps}}
\caption{ \label{fig:Lom} The likelihood function provided by the comparison
of the abundances of clusters at $z = 0.05$ and $z = 0.33$. The continuous
line corresponds to the open case, the dashed line corresponds to the flat
case. The two horizontal line gives the 1$\sigma$ and 2$\sigma$ confidence
interval on one parameter for a gaussian distribution. The dotted-dashed
line is a fit by a two-tailled gaussian. }
\end{figure}
\section{Conclusion}
The local temperature distribution function of X-ray clusters is an important tool
for cosmology and can provide direct information on the dark matter
distribution. It is therefore essential to have a good estimation of the
local temperature distribution function. Furthermore, such a sample is
of crucial importance as a reference sample in order to properly evaluate
the evolutionary properties of the
cluster population and for other cosmological applications.
We have provided a new estimate of the temperature
distribution function for local clusters, based on a large sample of X-ray
clusters with measured temperatures. This sample is essentially flux limited and we have argued that it is likely to be reasonably representative with a completeness estimated to 85\%. We found an appreciably higher
abundance of clusters than previous estimates, but in good agreement with
the abundance inferred from the luminosity function and optical data. We have used the
sample to study the statistical properties of the matter density
fluctuations and obtained results consistent with previous works, although
we obtained a normalization slightly higher, $\sigma_8\sim 0.6$ for $\Omega_0 = 1.$ than previously found from X-ray clusters due to our
higher abundance of 4 keV clusters. Probably the most important application
of this new temperature function concerns the determination of the density
parameter via the test of the evolution of the cluster abundance with redshift.
In order to apply this test, we have used the Henry's sample
which provides for the first time a direct estimation of the temperature
distribution function at a non-zero redshift. We have found a clear
indication that the abundance of clusters was smaller at the epoch
corresponding to $z \sim 0.33$ consistent with Donahue and Voit (1999).
Using a likelihood approach, we inferred a high value of $\Omega_0 \sim 0.9
\pm 0.2$. Low--density open universes ($\Omega_0 \leq 0.49$) are excluded at
the 2 $\sigma$ level. The exclusion region is nearly as severe in
the flat case: $\Omega_0 \leq 0.37$ models are excluded at the 2 $\sigma$ level.
This result is entirely consistent with other independent analyses of the
EMSS sample (Sadat et al., 1998; Reichart et al., 1999). We therefore
confirm that the abundance of X-ray clusters as inferred from the EMSS
favors a high--density universe. We have pointed out that this result is
also consistent with what is known from existing ROSAT samples, although
the situation is not as clear as in the case of the EMSS sample. It is
important to keep in mind that the present method is one of the very few
cosmological probes of $\Omega_0$ that is
not based on local estimates, but is rather global in nature.
However, given the importance of the conclusion, we believe that it should
still be considered with caution. Our study of the local temperature
distribution function demonstrated that systematic uncertainties could
be more important than expected. It is therefore essential to perform this
test using an entirely different and independent sample. Temperature measurements with XMM of ROSAT
selected clusters will allow to obtain such a sample. The application of the cosmological test will then probably lead to a more definitive conclusion concerning
the mean density of the universe.
\begin{thebibliography}{}
\bibitem {} Arnaud, M. 1994, Cosmological Aspects of X-ray Clusters of Galaxies, W.C.Seitter ed., NATO ASI Series, Vol. 441, 197
\bibitem {} Arnaud, M. \& Evrard, A.E. 1999, MNRAS, 305, 661
\bibitem{} Bahcall, N.A. \& Fan, X., 1998, ApJ, 504, 1
\bibitem {} Balland, C. \& Blanchard, A. 1997, ApJ, 497, 541
\bibitem {} Barbosa D., Bartlett J.G., Blanchard A. \& Oukbir, J. 1996, A\&A,
314, 13
\bibitem {} Bartlett, J.G. 1997, Proceedings of the
1st Moroccan School of Astrophysics,
ed. D.~Valls-Gabaud et al., A.S.P. Conf. Ser.,
vol. 126, p. 365
\bibitem {} Bartlett, J.G. \& Silk, J. 1993, ApJL, 407, L45
\bibitem {} Blanchard, A. \& Bartlett, J. 1998, A\&A, 314, 13
\bibitem {} Blanchard, A., Bartlett, J. \& Sadat, R. 1999, CRAS, 327, 318.
\bibitem {} Blanchard, A. \& Silk, J. : 1991,
proceedings of the Moriond Conference, 1991, Editions Fronti\`eres, p93.
\bibitem {} Bryan, G.L. \& Norman, M.L. 1998a, ApJ, 495, 80
\bibitem {} Bryan, G.L. \& Norman, M.L. 1998b, astro-ph/9802335
\bibitem {} Borgani, S., Rosati, P., Tozzi, P. \& Norman, C. 1999, ApJ, 517, 40.
\bibitem {} Borgani, S., Girardi, M., Carlberg, R.G., Yee, H.K.C. \& Ellingson, E. 1999, ApJ, 527, 561
\bibitem {} Carlberg, R.G., Morris, S.L., Yee, H.K.C. \& Ellingson, E. 1997,
ApJ, 479, L19
\bibitem {} Colafrancesco, S., Mazzotta, P. \& Vittorio, N. 1997, ApJ, 488, 566
\bibitem {} David et al 1993
\bibitem {} De Grandi, S. et al. 1999a, ApJL, 513, L17
\bibitem{} De Grandi, S. et al. 1999b, ApJ, 514, 148
\bibitem {} Donahue, M. 1996, ApJ, 468, 79
\bibitem {} Donahue, M., Voit, G. M., Gioia, I., Lupino, G., Hughes, J. P. \&
Stocke, J. T. 1998, ApJ, 502, 550
\bibitem {} Donahue, M., Voit, G. M., Scharf, C. A., Gioia, I.,
Mullis, C. P., Hughes, J. P. \& Stocke, J. T. 1999, ApJ, 527, 525
\bibitem {} Donahue, M. \& Voit, G. M. 1999, ApJL, 523, L137
\bibitem {} Ebeling, H., Voges, W., B\"ohringer, H., Edge, A.C., Huchra, J.P. \& Briel, U.G. 1996, MNRAS, 281, 799
\bibitem {} Ebeling, H., Edge, A.C., Fabian, A.C., Allen, S.W.,
Crawford, C.S. \& B\"ohringer, H. 1997, ApJL, 479, L101
\bibitem {} Ebeling, H., Edge, A.C., B\"ohringer, H., Allen, S.W.,
Crawford, C.S. Fabian, A.C., Voges, W. \& Huchra, J.P. 1998, MNRAS, 301, 881
\bibitem {} Ebeling, H., Jones, L. R., Perlman, E., Scharf, C., Horner, D.,
Wegner, G., Malkan, M., Fairley, B.C. \& Mullis R. 2000, Ap.J., 534, 133
\bibitem {} Edge, A.C., Stewart, G.C., Fabian, A.C., \& Arnaud, K.A. 1990,
MNRAS, 245, 559
\bibitem {} Eke, V.R., Cole, S., Frenk, C.S., Henry, P.J. 1996,
MNRAS, 298, 1145
\bibitem {} Eke, V.R., Cole, S., Frenk, C.S., Henry, P.J. 1998,
MNRAS, 298, 1145
\bibitem {} Evrard, A.E. 1989, ApJ, 341, L71
\bibitem {} Evrard, A.E., Metzler, C.A., Navarro, J.F. 1996, ApJ, 469, 494
\bibitem {} Fabian, A.C. Crawford, C. S., Edge, A. C., Mushotzky, R. F.
1994, MNRAS, 267, 779
\bibitem {} Frenk, C.S., White, S.D.M., Efstathiou, G., \& Davis, M. 1990,
ApJ, 351, 10
\bibitem {} Frenk, C.S. et al. 1999, ApJ, 525, 554
\bibitem {} Fukazawa, Y., Makishima, K., Tamura, T., Ezawa, H., Xu, H,Ikebe, Y., Kikushi,K., Ohashi, T. 1998, PASJ, 50, 187.
\bibitem {} Gioia, I.M., Luppino, G.A. 1994, ApJS, 94, 583
\bibitem{} Girardi M., Borgani S., Giuricin G., Mardirossian F. Mezzetti, M. 1998, ApJ, 506, 45
\bibitem{} Governato F., Babul, A., Quinn, T., Tozzi, P., Baugh, C. M.,
Katz, N.\& Lake, G. 1999, MNRAS, 307, 949
\bibitem {} Hattori, M. \& Matsuzawa, H. 1995, A\&A, 300, 637
\bibitem {} Henry, J.P. \& Arnaud, K.A. 1991, ApJ, 372, 410
\bibitem {} Henry, J.P. 1997, ApJ, 489, L1
\bibitem {} Henry, J.P. 2000, ApJ, 535, 350
\bibitem {} Hughes, J.P., Butchler, J.A., Stewart, G.C. \& Tanaka, Y. 1993, ApJ, 404, 611
\bibitem {} Johnstone, R.M., Fabian, A.C. \& Taylor, G.B. 1998, MNRAS, 298, 854
\bibitem {} Kaiser, N, 1986, MNRAS, 222, 323
\bibitem {} Kruse, G. \& Schneider P. 1999, MNRAS, 302, 821.
\bibitem {} Lacey, C. \& Cole, S. 1994, MNRAS 271, 676
\bibitem {} Le Dour, M., Bartlett, J.G., Douspis, M. \& Blanchard, A. 2000, astro-ph/0004283, submitted to A\&A
\bibitem {} Lineweaver, C., Barbosa, D., Blanchard, A. \& Bartlett, J. 1997,
A\&A, 322, 365
\bibitem {} Lineweaver, C. \& Barbosa, D. 1998, ApJ, 496, 624
\bibitem {} Markevitch, M. 1998, ApJ, 504, 27
\bibitem {} Metzler, C.A. \& Evrard, A.E. 1997, astro-ph/9710324
\bibitem {} Mushotzky R. F. \& Scharf C. A. 1997, ApJ, 482, L13
\bibitem {} Navarro, J. F., Frenk, C.S. \& White, S.D.M. 1995,
MNRAS, 275, 720
\bibitem {} Nevalainen, J., Markevitch, M., Forman, W.R. 2000, Ap.J., 532, 694
\bibitem {} Nichol, R.C., Holden, B.P., Romer, A.K., Ulmer, M.P.,
Burke, D.J. \& Collins, C.A. 1997, ApJ., 481, 644
\bibitem {} Perrenod, S.C. 1980, ApJ, 236, 373
\bibitem {} Oukbir, J. \& Blanchard A. 1992, A\&A, 262, L21
\bibitem {} Oukbir, J. \& Blanchard A. 1997, A\&A, 317, 10
\bibitem {} Oukbir, J., Bartlett, J.G. \& Blanchard, A. 1997, A\&A, 320, 365
\bibitem {} Peebles, P. J. E., Daly, R. A. \& Juszkiewicz, R. 1989, ApJ, 347, 563
\bibitem {} Press, W.H. \& Schechter, P. 1974, ApJ, 187, 425
\bibitem {} Reichart, D.E. et al, 1999, ApJ, 518, 521
\bibitem {} Robinson, J., Gawiser, E. \& Silk, J. 2000, Ap.J., 532, 1
\bibitem {} Roussel, H., Sadat, R., Blanchard, A. 2000, A\&A, submitted
\bibitem {} Sadat, R., Blanchard, A. \& Oukbir, J. 1998, A\&A, 329, 21
\bibitem {} Schmidt, M. 1968, ApJ, 151, 393
\bibitem {} Viana, P.T.R. \& Liddle, A.R. 1996, MNRAS, 281, 323
\bibitem {} Viana, P.T.R. \& Liddle, A.R. 1999a, MNRAS, 303, 535
\bibitem {} Viana, P.T.R. \& Liddle, A.R. 1999b, electronic proceedings of the conference ``Cosmological Constraints from X-ray Clusters'', Strasbourg,
France, Dec. 9-11, 1998, http://astro.u-strasbg.fr/amas/proc/proceed.html
\bibitem{} Voit, G.M. \& Donahue, M. 1998, ApJ, 500, L111
\bibitem{} White, S.D.M., Navarro, J.F., Evrard, A.E. \& Frenk, C.S. 1993,
Nature, 366, 429
\bibitem{} Yamashita, K., in Frontiers of X-ray astronomy, Ed. Y. Tanaka, K., Koyama (Universal Academy Press, Tokyo), p.475
\end{thebibliography}
\end{document}