MNRAS 000, 000–000 (0000) Preprint 18 January 2016 Compiled using MNRAS LATEX style file v3.

0

Probing dark energy via galaxy cluster outskirts

Andrea Morandi1⋆ , Ming Sun1
1 Physics Department, University of Alabama in Huntsville, Huntsville, AL 35899, USA
arXiv:1601.03741v1 [astro-ph.CO] 14 Jan 2016

ABSTRACT
We present a Bayesian approach to combine Planck data and the X-ray physical properties
of the intracluster medium in the virialization region of a sample of 320 galaxy clusters
(0.056 < z < 1.24, kT > ∼ 3 keV) observed with Chandra. We exploited the high-level
of similarity of the emission measure in the cluster outskirts as cosmology proxy. The cos-
mological parameters are thus constrained assuming that the emission measure profiles at
different redshift are weakly self-similar, that is their shape is universal, explicitly allowing
for temperature and redshift dependency of the gas fraction. This cosmological test, in combi-
nation with Planck+SNIa data, allows us to put a tight constraint on the dark energy models.
For a constant-w model, we have w = −1.010 ± 0.030 and Ωm = 0.311 ± 0.014, while
for a time-evolving equation of state of dark energy w(z) we have Ωm = 0.308 ± 0.017,
w0 = −0.993 ± 0.046 and wa = −0.123 ± 0.400. Constraints on the cosmology are fur-
ther improved by adding priors on the gas fraction evolution from hydrodynamic simulations.
Current data favor the cosmological constant with w ≡ −1, with no evidence for dynamic
dark energy. We checked that our method is robust towards different sources of systematics,
including background modelling, outlier measurements, selection effects, inhomogeneities of
the gas distribution and cosmic filaments. We also provided for the first time constraints on
which definition of cluster boundary radius is more tenable, namely based on a fixed overden-
sity with respect to the critical density of the Universe. This novel cosmological test has the
capacity to provide a generational leap forward in our understanding of the equation of state
of dark energy.
Key words: cosmology: observations – cosmology: large-scale structure of Universe – galax-
ies: clusters: general – X-rays: galaxies: clusters – methods: data analysis – methods: statisti-
cal.

1 INTRODUCTION (e.g. the evolution of the cluster abundance) can be therefore used
to constrain the cosmology, since a consistency relation between all
The hierarchical large-scale structure scenario provides a picture
the related observables must occur in any theory invoked to explain
where gravity is constantly pulling lumps of matter together to form
cosmic acceleration of the Universe.
increasingly larger structures, with galaxy clusters sitting atop this
hierarchy, i.e. at the intersection of filaments. Clusters, by virtue of Recently, the improvements of the data quality of the obser-
their position at the high end of the cosmic mass power spectrum, vations, with the accumulation of observational data from CMB
provide a number of signatures that are sensitive tracers of cosmic measurements, large scale structure surveys (LSS), and SNIa ob-
evolution and a unique probe of mass distribution on large scales. servations, has allowed to constrain the cosmological parameters
In particular, measurements of cluster growth via their with a greatly improved accuracy, at few percent level. In particular,
mass function (Vikhlinin et al. 2009) and relative baryon budget the Planck collaboration has released the first cosmological papers
(Mantz et al. 2014) are highly sensitive tests of the geometry and providing the highest resolution, full sky maps of the CMB tem-
matter content of the Universe. These constraints are complemen- perature anisotropies. These results suggest that a model in which
tary to measurements of the expansion history from type Ia su- the dark energy is a cosmological constant (w = −1) is preferred
pernovae (SNIa), the cosmic microwave background (CMB) tem- (Planck Collaboration 2015). These findings have been indepen-
perature and polarization spectra, and baryon acoustic oscillations dently corroborated by SNIa (Betoule et al. 2014) and galaxy clus-
(BAO). The combination of observables pertaining to the expan- ter growth analyses (Vikhlinin et al. 2009). However, observational
sion history and geometry of the Universe (e.g. baryon fraction in signatures of such deviations of w from -1 are very small, and
galaxy cluster, SNe, BAO), and those pertaining to structure growth hence the measurements are prone to statistical and systematic er-
rors. Therefore, current cosmological constraints are still limited by
the quality of the data and systematics in complementary proxies of
⋆ E-mail: andrea.morandi@uah.edu the expansion history of the Universe.

c 0000 The Authors

2 Morandi & Sun
For example, the use of clusters as accurate cosmological similar model, cosmological framework and we describe the origin
probes requires to unearth and disentangle the cosmological and of the constraints on the cosmological parameters from the observ-
astrophysical signatures since, in practice, the cluster observables ables. In §3 we present our data set, while in §4 we present the sta-
are customarily locked to the baryonic component. With respect to tistical method used to infer both model and physical parameters.
the DM, cluster baryons are indeed subject to more complex (and In §5 we present results on the EoS of dark energy, a model com-
currently still uncertain) physical processes (e.g. galaxy formation parison (§6), the analysis of the systematics (§7) and the potential
and feedback) besides the gravity. This can clearly limit the use of for improvements in cosmological constraints from future surveys
clusters as cosmological proxies. While it is paramount to improve (§8). Throughout this work we assume the flat CDM model, with
our modelling of clusters, it is also crucially important that the dark matter density parameter Ωm , EoS parameter w, and Hubble con-
energy constraints at this level of accuracy are obtained from com- stant H0 = 100h km s−1 Mpc−1 . The factor Ez = H(z)/H0
bination of several independent techniques and data sets. This not is the ratio of the Hubble constant at redshift z to its present day
only reduces systematics but also improves statistical accuracy by value. Unless otherwise stated, we report the errors at the 68.3%
breaking degeneracies in the cosmological parameter constraints. confidence level.
One cosmological test via galaxy clusters which has been
previously proposed (e.g. Arnaud et al. 2002) hinges on the self-
similarity. Indeed, for an assumed cosmology, similarity of shape
in cluster radial quantities has been recently shown in X-ray obser- 2 COSMOLOGY VIA OUTSKIRTS OF GALAXY
vations (Arnaud et al. 2002; Arnaud et al. 2010; Eckert et al. 2012; CLUSTERS: METHODOLOGY
Morandi et al. 2015), with weak-lensing (Postman et al. 2012) and
SZ (Plagge et al. 2010; Planck Collaboration et al. 2013). These In this Section we will review the basic idea of the self-similar
homogeneous physical properties of the ICM are theoretically ex- model for cluster outskirts (§2.1), emphasizing how clusters are
pected since the scale-free dark matter (DM) collapse drives the identical objects with respect to a critical overdensity and when
evolution of halo concentration and the physical properties of the scaled by their temperature. We then outline our current cosmo-
baryons across cosmic time (Limousin et al. 2013). The ensuing logical framework, in particular with respect to the dependency of
similarity yields both a universal dark matter distribution and sim- the cluster observables on the underlying cosmology (§2.2). Next,
ple global scaling relations between the observables (Arnaud et al. we discuss how to pin down the cosmological parameters through
2010; Voit 2005; Kaiser 1986), retaining the fingerprints of the the observables (§2.3), comparing our constraints to those from
underlying cosmology, gravity but also astrophysics that govern other independent measurements (e.g. CMB, LSS, SNIa). Finally,
their formation and evolution. Given the current large samples we present a generalized self-similar model which accounts for the
of clusters e.g. from Planck or Chandra, the most striking out- impact of the baryon component at the top of the evolution of the
growth of self-similarity is the ability to greatly exploit this large cosmic expansion history (§2.4).
amount of data, by stacking the X-ray/SZ signal to study the viri-
alization region of clusters (Plagge et al. 2010; Eckert et al. 2012;
Planck Collaboration et al. 2013; Morandi et al. 2015).
2.1 The self-similar model
In this respect, in the present paper we investigated cosmolog-
ical constraints on the expansion history and geometry of the Uni- In the current hierarchical structure formation model, galaxy clus-
verse to study dark energy models, by exploiting the high-level of ters form from gravitational collapse of the primordial cosmolog-
similarity of X-ray emission measure profiles in the outskirts of the ical density perturbations, when their linear density fluctuations
hot cluster population. In a companion paper (Morandi R et al. 2015), reach the collapse threshold δc = 1.686. On large scales, the ma-
we stacked the emission measure profiles EM ∝ n2e dl of a X- jority of the baryonic component is in the form of X-ray emitting
ray Chandra sample of 320 clusters to detect a signal out to and ICM and is expected to follow the distribution of the gravitation-
beyond R200 . In that work we proved that the physical properties ally dominant DM. The self-similar model (see, e.g., Kaiser 1986)
of clusters are indeed universal outside of the core (R > ∼ 0.2R200 ) predicts that cluster gas profiles for a given mass (or peak height)
once scaled according to a certain radius e.g. R200 , where R200 appear universal when they are scaled according to a certain radius
is the radius within which the mean total density is 200 times the (e.g. R200 ) with respect to the reference background density of the
critical density of the Universe. The idea of this cosmological test Universe (see, e.g., Voit 2005). The underlying idea is that grav-
is that the X-ray emission measure profiles, once rescaled by the ity is the only responsible for the observed values of the physical
self-similar predictions, can be used as ‘standard candle’ to con- properties of clusters and the ICM is heated only by the shocks
strain the cosmological parameters, being a sensitive proxy of the associated with the collapse. Hence clusters are identical objects
underlying cosmology via both the angular distance and the critical when scaled by their mass or temperature and power law scaling
density of the Universe ρc,z . The correct cosmology is thus con- relations are expected between the observables. Baryonic physics,
strained assuming weak self-similarity of EM for the outskirts of such as radiative cooling, star formation and feedback might break
hot clusters, that is the various EM profiles at different redshifts self-similarity, but their effects are mostly confined within the clus-
are homologous, explicitly allowing for temperature and redshift ter core.
dependency of the gas fraction fgas . In fact, at R > R1000 , even Assuming the spherical collapse model for the DM halo and
groups and poor clusters become more self-similar (e.g. Sun 2012). the equation of hydrostatic equilibrium to describe the distribution
Using the wealth of data in hand, both independently of and in of baryons into the DM potential well, in the self-similar model
combination with complementary data sets, we present constraints the cluster mass at an overdensity ∆ (e.g. ∆ = 500, 200) and
on the dark energy models within the ΛCDM, in particular to de- temperature are related by Ez Mtot ∝ T 3/2 . The overdensity ∆
termine whether there is evidence for dynamical dark energy with is calculated with respect to the reference background density of
equation of state (EoS) parameter w 6= −1. the Universe, customarily the critical (or mean) density (see §6.2
The paper is organized as follows. In §2 we describe the self- for further discussion). So we have R∆ ∝ (M/(ρc,z ∆))1/3 ∝

MNRAS 000, 000–000 (0000)

we can assume that the raw emission measure pro-  Z z  1 + w(z) file has a powerlaw shape in the external regions. dimming Sx ∝ Ez3 (1 + z)−4 .e. When from self-similarity. a time-evolving w(z) we choose the popular EoS parameterization This is referred as strong self-similarity (Kaiser 1986).. Therefore. and via the factor Ez = H(z)/H0 appearing both in the according to the standard self-similar model (Equation 3). we can then tory factor Ez = H(z)/H0 and the angular diameter distance Da write the angular diameter distance and the factor Ez as: (Figure 1). the angular distance Da used to convert angular radii θ to physi- The X-ray surface brightness profiles have been then scaled cal radii. thus introducing unphysical departures interact with other constituents other than through gravity. However. and sets the w(z) = w0 + wa z/(1 + z) given by Chevallier & Polarski (2001) powerlaw slopes of the scaling relations. In particular.3 for a more detailed discussion on the history factor Ez = H(z)/H0 and the angular diameter distance dependency of the (scaled) EM on the cosmological parameters. assuming considering a dynamical dark energy component. e.e. that Da = . with e. thus in- normalization of the profiles and in R∆ (i. ω = |Ωk |1/2 0 E(ζ) More in details.16 × dictions (see. In our baseline model we assume that lying cosmology produces a shift along the vertical axis (∝ Ez−3 ) the dark energy is a cosmological constant with current density pa.2 Cosmological framework parameters from cluster observables. θ being the an- ergy density. cosmological parameters (Equations 3 and 2). a nearby cluster like Coma sity. to ΩΛ as the dark en- brightness Sx (θ) (expressed in counts/s/arcmin2 . for dark energy models with redshift are identical objects when scaled according to Equation 1.e. g (x) profiles at different z (that is the shape similarity of the EM ize the equation of state either as a constant w or as a function g of EM (x) is universal) and for x > ∼ 0.g. From the previous equation it follows that: of the redshift.726 K)4 . ∆z = ∆ = 200) and cosmological model and its relative parameters. Hence the rescaled emission measure EM g (x) (see be- c S(ω) low for its definition) is self-similar in the correct cosmology. we will consider the case of a cosmological constant (standard ΛCDM scenario). (2) 0 1+z x1−6β . and assume that the dark energy does not emission measure profile. we parameter. Da are the proxies of the cosmology. The variation with redshift of Da and Ez depends on Ωm . Probing dark energy via galaxy cluster outskirts 3 T 1/2 Ez−1 ∆−1/2 . For example. EM (x) ∝ λz = exp 3 dz .. R∆ ∝ Ez−1 ).6 − 1 (a factor ∼ 2 − 3) with respect to the cosmological dimming factor (1 + z)4 ∼ 6 − 16. Equation 25). Without loss of EM Ez = Ωm (1 + z)3 + Ωk (1 + z)2 + ΩΛ λz generality. e.2 (excluding the inner vol- MNRAS 000. we can infer the cosmological with θ = x R∆ /Da . different clusters show signif- mation cluster outskirts are thought to be self-similar when they icant different ranges of X-ray surface brightness both in normal- are normalized to an overdensity ∆ with respect to the critical den- ization and angular extent. i. following a model selection as well: a convenient approach by means of their self-similar evolution.z = 3H02 Ez2 /(8πG). These raw Ωm + ΩΛ + Ωk = 1.g. where fgas is the gas fraction. ω. verse. We define the scaled emission measure as where S(ω) is sinh(ω). and to perform brighter (EM (x = 0) ∝ Ez3 ) with increasing redshift. assuming the self-similar evolution g (x) ∝ T −1/2 fgas EM −2 −3+(1−6β) Ez Da1−6β (3) X model and similarity (that is universal shape) of EM at various redshifts (in the correct cosmology). wich corresponds to an only modest attenuation at z ∼ 0. while high-z clusters have a typical EM profiles depend on the assumed cosmological parameters via size of the order a few arcminutes. we present a toy model showing the depen- tio between the pressure and the energy density in the equa. 000–000 (0000) . with β = β(x) increasing with the cluster-centric ra- dius x = R/R200 . clusters of given tempera. 1998. Arnaud et al. which requires w = −1. Thus. defined as the ra. we adopt the Einstein equation in the form gular distance) and global spectroscopic temperatures. finally. Ωk and w(z). by its equation-of-state parameter w = P/ρ. i. the problem is to recover simultaneously both the ture appear smaller (R∆ ∝ Ez−1 . We refer to §2. Our raw observables are represented by the X-ray surface We refer to Ωm as the total matter density. ρc. 2002). Note that in our analysis we neglect the energy associated with the cosmic radiation. Changing the under- Wang & Steinhardt 1998). Therefore. As can be seen from Equation 1. troducing a dependency upon the cosmology via the expansion his- From. equal to g (x) = EM (x)Ez−3 fgas EM −2 −1/2 TX exploiting the self-similar pre- and less than 0. dency the scaled emission measure EM g (x) (Equation 3) on the tion of state of the dark energy component (Caldwell et al. respectively. quantities can be regarded as cosmological independent. since they We remember that in an hierarchical scenario of structure for- are direct measurements. the model with a constant EoS by taking In this self-similar model. By playing the reverse game.g. H0 (1 + z) |Ωk |1/2 g (x) is universal for clusters with different tem- is the shape of EM Z z dζ peratures and redshifts. Moreover. Ωk 6= 0. (1992) (cf.e. This translates into a surface brightness Bayesian inference has been presented in §4 and Appendix A. fgas = Mgas /Mtot being the gas mass fraction. where Ωk accounts for the curvature of space. with x = R/R∆ . We shall also discuss the case of a non-flat Uni- evolve with redshift and are independent of the cluster mass. In Figure 1 (left). ters: ΩΛ . which are not predicted to (model wz CDM). Ωr ≃ 4. 2. We can thus highlight the dependency of EM g (x) on the cosmological parame- 10−5 (TCMB /2. Carroll et al. Hereafter we shall consider parameterizations of Z w encompassing a pressure-to-density ratio w constant and time- 1/2 EM (x) ∝ n2e dl ∝ Mgas 2 /R∆5 2 ∝ fgas Ez3 TX (1) evolving in a flat Universe (Ωk = 0). depending on Ωm and w = The properties of dark energy are commonly characterized w(z). from Equation 1 the (scaled) extends out to a few degrees. The expansion parameters. and the horizontal axis (∝ R/R200 ∝ Da /Ez−1 ) of the scaled rameter ΩΛ = 1 − Ωm . here we outline the dependency of  1/2 g (R/R200 ) on the cosmological parameters.3 From observables to cosmology In this Section we discuss how we can pin down the cosmological 2. galaxy clusters of all masses and w(z) = w0 (model wCDM). Kaiser 1986. sin(ω) for Ωk greater than. i.

3.g. depending on both the angular distance.4 Morandi & Sun Figure 1. The EM(x) from nearby clusters at z = 0. being not thought more like ’cosmic buoys’. following the idea i.g. galaxy clusters can be the value of H0 ) is a nuisance parameter in our analysis.7). anchored to the evolution of the of immediate cosmological interest. i. the absolute normalization of EM g (x) (which depends on distance-determination methods (e. model (§2. with the arrows in the Figure showing the magnitude and direction of this shift.05 in a reference g cosmology (Ωm = 0. Sx ∝ Ez3 (1 + z)−4 and R200 ∝ Ez−1 ). 000–000 (0000) . The scaled profiles are not used as simple distance indicators. we can recover the correct cosmology (e. in the normalization (via a temperature-dependency of the baryon content) malization. where β ∼ 1 (Morandi et al. arise due to the impact of non-gravitational processes (e. In other describes their evolution with z in the self-similar model. It relative hot gas fraction fgas (whose value depends upon the Hub- is important to underscore that the cosmological parameters that we ble constant) should be representative of the cosmic value Ωb /Ωm . We considered the measured gas fraction fgas = Mgas /Mtot ∝ Da (Ωm . and evolution with redshift (via a redshift-dependency of the baryon con- bination of Ez and Da ) on the cosmological parameters (Equation g tent) of EM(x). can actually constrain are those entering in the diameter distance We point out that the dependency of EM g (x) on the underly- and E(z). Changing the underlying cosmology introduces a shift upwards (∝ Ez−3 ) and towards smaller radii (∝ R/R200 ∝ Da /Ez−1 ) of the scaled emission measure profile.e. we also present the dependency of the mea. w)3/2 on and feedback). Right panel: sensitivity to Ωm and w of different cosmological proxies Obs. whose parameters must be determined a posteriori.3. w(z). It also important to point out that this cosmological test is in- in the strict sense. MNRAS 000. since we simply re- conversion of angular radius into physical radius. ΩΛ . (2002). in the nor. where self-similar objects (e.e. capturing possible departures from self-similarity the interiors (β <∼ 0.e. The magnitude of the scaling and renormalization has been increased of a factor of 2(4) for z = 0. we present the sensitivity of EM as we move towards the outer volumes (for increasing β). for example. ing cosmology is stronger than for fgas . parameters. w)3/2 and EM(x̃) g at a fixed radius x̃. i. and also upon quiring that EMg (x) is constant with redshift in the correct cosmol- the cosmology via the cluster-growth term Ez = H(z)/H0 .e. via the factor Ez−3 ) and radial dependency (via a com. via the sensible to the value of the Hubble constant.3. background cosmology in a very simple way provided that they are from the standard hot gas mass fraction cosmological test. we can recover the correct cosmology. 2015) with respect to g dependency of EM(x).2) describing the scaled emission measure EM(x).1).2(z = 1) for visualization purposes. assuming similarity of the EM(x) g profiles at different z (that is the shape of EM(x) is universal) and for x > ∼ 0.4). and it is more pronounced g (x) at a In Figure 1 (right). g By playing the reverse game. Left panel: toy model (a β-model profile with slopes β = 2/3 and 1. For comparison. for x̃ > ∼ 1 (Equation 3) as cosmological proxies Obs with respect to a reference cosmology (Ωm = 0. Ωm . This provides both a global (i. 2002). w = −1. thus introducing X 200 g a dependency upon the cosmology via Ez = H(z)/H0 and the angular diameter distance Da . This is different.2. g i. Unlike words. which ogy1 . renormalized by 2 T 1/2 and scaled by R Ez3 fgas .2 and z = 1 in a different dark energy model (Ωm = 0. The emission measure profiles have been scaled according to the standard self-similar model (Equation 3).g. it does not provide a radial dependency on the cosmological successfully applied by Arnaud et al. These departures from a strong self-similar model might 3). w = −1). Arnaud et al. umes). we will explicitly allow for temperature and redshift outer regions.e. SNIa). that is Ωm . galaxy formation sured gas fraction fgas = Mgas /Mtot ∝ Da (H0 . thus introducing unphysical departures from self-similarity. leading to a generalized scenario known as weak self-similar the cosmology: note that the gas fraction is a cumulative quality. We point both a global and radial dependency upon the cosmological param- out that the EMg (x) dependence upon the underlying cosmology is stronger for increasing redshifts and for increasing β. providing fixed radius x̃ to the cosmological parameters Ωm and w. in the 1 More in general.g. w = −1) is compared with the predictions of EM(x) at z = 0.

15 − 1 R500 (e. 2006). Note that the normalization of fgas is irrelevant for our purpose Ultimately.g. 7 keV ogy) is constant. ΩΛ . the comparison of cluster mass (or temperature) and redshift can be straightforwardly and emission measure at different redshifts can be thus used as ‘stan. 2009). Young et al. observations are very challenging did not find evidence of dependency of fgas (in the outskirts) with in cluster outskirts. self-consistently accounted for. the (rescaled) emission measure profiles. As for the redshift dependency. e. the question arises how much this assumption is tenable. i. on the since. Probing dark energy via galaxy cluster outskirts 5 eters. 3 Strictly speaking. However. 2009). and which might evolve with redshift of hydrostatic equilibrium and three-dimensional geometry (since (Ettori et al. customarily measurements of icance of the EM g (x) measurements and to reduce the observed the apparent evolution of the cluster X-ray gas mass fraction has scatter of the emission measure profiles. we address this issue in a less cosmology- self-similarity dependent way by incorporating a power-law with temperature and linear with redshift dependency into the model. This allow us to measure EM g (x) for the stacked for the first time a test of this assumption. the potential source of systematics in the use of cluster (Equation 3). Gas fractions and mass function EM relatively massive objects. gas fraction: However. perfectly self-similar) DM component. Cluster baryons are indeed rect. simulations of cluster formation support a pic- also relatively insensible to X-ray calibration systematics. 2011). our work (Morandi et al. Measurements by Vikhlinin et al. fgas . stellar baryon fraction or model of mass indicate the impact of the baryon component at the top of the evo- function (calibrated e. range 0. 000–000 (0000) . nor lution of the cosmic expansion history which affect the cluster ob- completeness of the cluster sample: we only require similarity of servables via the term Hz . 2015). providing an independent way (and hinging on and feedback) besides the gravity. since the X-ray surface brightness drops below global temperature. the underlying idea is that the 1). been also used to probe the acceleration of the Universe. This cosmological test is In this respect. since the global temperature is calculated in the radial ness’) will be discussed in §7. 2009). mass when considered over a range extending from group to clus- terize the dependency of fgas on mass and redshift (§2.g. on N-body simulations) are required. based on observed quantities (the emission measure) on subject to more complex physical processes (e. galaxy formation the plane of the sky. in practice. ter scales (e.g. 2014. being a sen. 2 The possible impact of inhomogeneities of the gas distribution (’clumpi. In other words. homo. with kT > 5 keV. with normalization EM0 ∝ Ez3 and mic expansion history due any possible dependency of fgas on boundary radius R200 ∝ Ez−1 . Un. which is probability due to the fact that most the background level at large radii. uncertainties in the cos- self-similar objects. w(z). greater than 5 keV. clusters of given temperature ize over the slopes of this relation simultaneously with the full set evolve with redshift in a very simple way provided that they are of cosmological parameters. A temperature and/or redshift dependency of fgas could geneous gas distribution2 .e.2 are (weakly) self-similar (see are somewhat cosmology-dependent.g. it is less clear whether such a trend persists for g (x) (∼ 30 − 60 percent).4 Temperature and redshift dependency of fgas : weak mological parameters. The correct cos. 2015) we Vikhlinin et al.3. Note that ture where the cumulative gas fraction is mildly increasing with the absolute normalization of EM g (x) can vary. via stacking (Morandi et al. we shall use the following parameterization of the sitive proxy of the underlying cosmology. (2006) methods can provide constraints also on other cosmological pa. the to the baryonic component rather than the invisible (and nearly- proposed method has also the major advantage that it is more di. Hence. We then marginal- As can be seen from Equation 1. Observationally. independent of the redshift.g. the dependence of fgas on the assumed distances to the clusters logical information ’buried’ in the outskirts of clusters (see Figure (Ettori et al. ing trend of the cumulative gas mass fraction with temperature trinsic scatter for individual clusters (∼ 10 − 20 percent) than (Sun et al.g. while the gas frac- However. mass function). tion in groups and intermediate mass clusters implies an increas- tion measurements are characterized by significantly smaller in. since we are interested in exploiting the large cosmo. These uncertainties might be enclosed in the we do not infer the three-dimensional gas and total mass). TX MNRAS 000. including the Hubble constant H0 consistent with being constant with cluster temperature. independent of the cluster temperature and redshift. for objects and the baryon fraction Ωb . in practice we do not use EMg (x) for individual clusters. 2015) provided bins of redshifts. While at the first order TX fgas ∝ (1 + αz) (4) of approximation we can assume that fgas (in the correct cosmol. a large cluster sample of the clusters (264 out of 320) in our sample have temperatures is required to robustly measure EM g (x) in the outer regions e. dard candle’ to constrain the cosmological parameters. through Finally. rather gas mass fraction is independent of the redshift in the ’correct’ cos- we stack the raw emission measure of several clusters grouped in mology. We also point out that in Equation 4 we refer to a de- as cosmological probe could be ascribed to the presence of baryons pendency of the gas fraction upon TX and z. In this way. We caution the reader that the aforementioned findings on fgas ent bins of redshift and for x > ∼ 0. i. which are currently uncertain different assumptions) with respect to the former approaches. Hereafter. In order to consistently investigate any possible temperature and redshift dependency of fgas and his ramification on the desired cos- 2. to improve the statistical signif. Naively.g. In our previous work (Morandi et al. indicating that fgas is profiles out to R100 and in several bins of redshift. no assumption (Kravtsov & Borgani 2012). since we parame. fgas term. the former two quantities being cosmology-independent3 .4 for a definition of weak self-similarity). it is also important to underscore that gas frac. mology is thus that for which the stacked profiles EMg (x) at differ. while the cosmological pa- §2. §3). rameters themselves are ultimately the quantity we are interested on. rather than e. respectively (Mantz et al.g. and σ8 . Moreover. (2008) (limited to the interiors of clusters) are both rameters besides Ωm . and Allen et al.4). Compared to other techniques (e. (Voit 2005) and difficult to model in hydrodynamic simulations like the gas fraction or the mass function methods. the cluster observables are customarily locked mass. equation 3 highlights also the dependency of the  δ scaled emission measure on the gas fraction.e. R500 being cosmology-dependent.

given the difficulties to infer X-ray signal in the out. Hence. A posteriori. We emphasize again gion by stacking the emission measure profiles EM ∝ n2e dl of the cluster sample to detect a signal out to and beyond R200 . given the high level of self-similarity of the clus- ter outskirts. and mation ’buried’ in the outskirts of clusters.g. Since the amount and nature of dark energy is poorly con- strained by local clusters (Equations 2). Cluster emission measure profiles distinguish these two evolutions which can be attributed to both are presented in Figure 2. we sort the clusters in redshift and priors on the astrophysical parameters (e. source-free regions of the cluster observations are always present. for the purpose of this work we did not require that the sam- previous works on studying cluster physical properties. Note the remarkable similarity of the profiles outside the cluster cores. Therefore. and iv) with the aid of complementary data sets (e. By combining Equations 3 and 4. ob- R serving time.6 Morandi & Sun We shall refer to α and δ as the astrophysical parameters. The curves are color-coded by redshift bins (§4).g. (the median redshift is ∼ 0. non-gravitational pro.3 and H0 = 70 km s−1 Mpc−1 in the self-similar scal- ray analysis. The expected evolution in the normalization of the scaling relations. and the slopes of scaling relations should remain close to the theo- might slightly depend on the assumed cosmology. which is solely due to the changing density of the Universe with redshift in the strong self- similar model. the ACIS-I imaging array and they have been selected because the In the following Section we will present our X-ray sample. are discussed in §2. we emphasize the im- portance of external data sets in order to have a bound on both the parameters of the underlying cosmology and the evolution of fgas from theory or simulations. (2002) (whose results hinge on the assumption of strong self- similarity). Although the sample was selected based on the quality of the 3. ter Ωm = 0. re-scaled according the self-similar model (§2. with high-z cluster showing similarity in shape and normalization with skirts. we underscore that we need a Figure 2. prove the readability of the figure. e. ing. ii) sample. iii) where we assumed a reference flat ΛCDM model. Indeed. but at the same time it could be related to the evolution of the archive (Morandi et al. to distin- guish them from the cosmological parameters. Thus. relaxation state. retical scaling laws.4. Departures from these theoretical expectations this dependency is substantially negligible. 2015). and repeating our analysis (§7). and stacked the profiles in bins of 4 sources to reduce the scatter and im- in order to break the degeneracy among the parameters.1 X-ray sample and analysis existing observations and hence might be subject to selection ef- This X-ray data analysis project represents a follow-up of our fects. different from Arnaud et al. For each redshift interval.4) and temperatures kT > ∼ 3 keV (the cesses (Equation 4). tion of the profiles. and gas fraction in the outskirts (Morandi et al. that the present study concerns primarily relatively hot clusters. and details on the data analysis. sources are ideal for observations of cluster outskirts.1). the real challenge is to measure and median temperature is ∼ 7 keV). that is the shape of EM g (x) is univer- sal. see Figure 1). However.g. These sources have been observed with cosmological and astrophysical processes. we verified that our working hypothesis emission measure. We have shown that Chandra can accurately (self-similarity) is robust towards systematics by considering clus- measure the physical properties of clusters in the virialization re. capturing effects of non-gravitational processes at play in clusters (Morandi & Ettori 2007). with redshift range z = 0. and in light of the depen- dency of the observables quantity EM g (x) on the underlying cos- mological parameters (Equation 3). temperature. explicitly allowing for temperature and redshift dependency of fgas . In particular.056−1. in practice. we can outline the depen- g (x) on both the astrophysical and cosmological pa- dency of EM rameters:  −1/2−2δ g TX EM (x) ∝ (1 + αz) −2 Ez−3+(1−6β) Da1−6β (5) 7 keV with x = R/R200 and β = β(x). allowing measurements of the local background which is funda- mental for robust measurements of the physical parameters in the 3 DATA SETS AND ANALYSIS outskirts.g. these including particulars on the temperature and redshift of the clusters. 2015). ter sub-samples based on redshift. The Chandra emission measure profiles of all clusters in our large sample of clusters: i) spanning a wide range of redshift. we only require weak self-similarity of the (rescaled) emission measure profiles. for which non-gravitational effects like pre-heating should be minimal. gas density. X-ray observations encompass R200 of each cluster. 000–000 (0000) . it thus described in this weak self-similar model. A straight line where EM g (x) can be inferred out to large radii via e. including ple be complete. stacking is drawn between the data points for each cluster for a better visualiza- (since we are interested in exploiting the large cosmological infor. MNRAS 000. from simulations). Moreover. an apparent evolution (or lack of) in EM g as measured from X-ray data could be ascribed to the cosmic expansion history and EoS of dark energy (Equation The sample is composed of 320 clusters from the Chandra 3). CMB) nearby sources.g.24 overall hot fraction budget fgas due to e. with matter density parame- we have an exquisite acknowledge of the systematics due to the X. concerning the last point.

2006). we considered the re.96 ± 0. one unabsorbed component of feedback from AGNs. 2011. the gas mass.15 − 1 R500 via mock simulations. 2012) in combination with Planck CMB We have performed a spectral analysis in order to infer the data (Planck Collaboration 2015) and WMAP polarization mea- global spectroscopic temperature TX via the XSPEC package surements (Bennett et al. wCDM and wz CDM and definition of ’virial’ radius are more likely given 3. which has been then depro. towards other definition of virial radius. For these data sets. while we fix the redshift to the value obtained from SNIa constraints. version 12. drogen column density NH to the Galactic value by using the Lei.g. For further details we refer to Morandi et al. This correction factor is Hydrodynamic simulations of cluster formation indicate that the 0. which can be determined from the flattened portion at the outer Specifically. 4 MODEL FITTING & SELECTION Note that in our work we have considered all the physical quantities at fixed overdensity (e. desired cosmological parameters. star formation and feed- CXB component by an absorbed power law with index 1. they have become background. 2013). gas mass fraction at intermediate to large cluster radii should have R a small cluster-to-cluster scatter and evolve little or not at all with We then measure the emission measure profile EM ∝ n2e dl from the X-ray images. For our sample we have R500 = (0. We also used the Tuebingen-Boulder absorption model ΩΛ and w(z)) from these likelihood functions for the given cos- (tbabs) for X-ray absorption by the ISM. Thanks to the improvements in the models of of the cosmic X-ray (CXB)+particle+readout artifact backgrounds. e. for which we always have some areas (> ∼ R100 ) free from source 3. which car- the readout background using the make readout bg routine.36 ± 0.8. we modeled the soft of simulations including radiative cooling.2). optical spectroscopy. They evaluated the gas fraction and de- with a fixed temperature of 0. 2009). The background spectra have been extracted from regions of the same exposure for the ACIS–I observations. Sun et al.02.0. 100) from the known value of R500 by assuming a NFW distribution with pa. distributions of the subset of cosmological parameters (e. (2014).25 keV. rived with the exposure correction. and two sets nally. cosmological parameters from the Markov Chain simulations. we shall determine which cosmological model ΛCDM . We will discuss how this definition is robust parameters (Ωm . Equation 4).g. we are primarily interested in con- straining the EoS of dark energy. The only free parameters this way. Fi. we infer R500 from the observables quantities via ter describing the evolution with z of fgas . we exploit the ’Union2. In order to measure the CXB. pletion from these simulations at different overdensities. (2009) sample ever. For each cosmological set.z .01)R200 . This provides an es- timate of R∆ for given value of R500 . Finally. MNRAS 000. with the concentration parameter c from the impact of a different set of simulations (Battaglia et al. Note that the likelihood for the Planck data is not computed den/Argentine/Bonn (LAB) HI-survey (Kalberla et al. (2014) (see their Table 3). Ωm . pilation (Suzuki et al. The radial emission measure profile is de. 2005). we recover the entire information from the location and in the spectral fit are the temperature. and AGN-simulation results of Planelles et al. using the covariance matrix for the cosmological parameters of in- tra have been thus extracted from a circular region in the radial terests. 2013.1’ SNIa com- analysis. (2009). Thus. Ωk and w(z)) inference and model selection. Young et al. studying nent with a temperature derived from spectral fits (∼ 0. c − M relation of Dutton & Macciò (2014). Battaglia et al.2) database (Planck Collaboration 2015). baryonic physics in clusters. like- (Arnaud 1996. incorporating their results as Gaussian priors on α (the parame- the X-ray Compton parameter YX = Mgas TX (Kravtsov et al. in order to account for the vignetting.2 External Cosmological Data Sets the data. evolution with redshift. Concerning the model selection. In particular. as well as from constraints on the cosmology model (Equation 3). we consider the recent smoothed particle hydro- region of the surface brightness profile. greatly reducing their degeneracy. the normalization and the relative amplitudes of the peaks in the CMB power spectrum and metallicity. framework used to infer both model and physical parameters. Spec. provides a robust method to constrain both cosmological and astrophysical parameters at play in our analysis. 000–000 (0000) . We adopted the APEC emissivity lihood functions are provided by the Planck collaboration model (Foster et al. We then corrected TX to infer a value in the radial range 0. Concerning the former point. we will also gauge the rameters (c. Probing dark energy via galaxy cluster outskirts 7 Here we summarize the most relevant aspects of the X-ray rameters. Our analysis will make use of complementary data sets as We refer to Appendix A for further details on the statistical well in constraining both astrophysical and cosmological pa. ΩΛ . we calculated the median distribu- tion of the stacked profile EMg (x). we will also investigate the impact of their different set of simulation (NR and CSF) on the evolution of fgas and hence the of clusters. ried out one set of non-radiative (NR) simulations. R200 ).3 Priors on fgas evolution from hydrodynamic simulations emission. their evolution (or lack of) with redshifts according to Equation 4. in particular modelling of the effects For the particle background modelling. one of which also accounting for the effect thermal components at zero redshift.65 ± 0.1 keV and another absorbed compo.g. Combining such predictions with measurements of the X-ray EM (x) has been then rescaled according to the self-similar emission measure. 2013). able to better recover the gas fraction content in clusters and its gion where the CXB is more dominant than the cluster emission. We also accounted for dynamics (SPH) simulations of Planelles et al.4 and two back from SNe (CSF). We then calculated R∆ (∆ = 200. mological model. we use the scaled stowed of feedback from active galactic nuclei (AGN). rather we explore the entire multidimensional grid of the range 0. redshift (Nagai et al.15 − 0. We inferred marginal probabilities of atomic data and the solar abundance ratios from Asplund et al. We applied a direct subtraction Planelles et al. from external data sets. We will primarily focus on the jected in order to infer the gas density profiles. the AtomDB (version 2. ∆ = 200) with respect to the A Bayesian statistics can be used to constrain both cosmological critical density ρc. 2014). How- the YX − M relation measured for the Sun et al.01)R200 and R100 = (1. in order to infer a renormalized value EMg (x). and we fixed the hy. 2007.75 R500 around the cluster centroid. 2012). once the con. (2015). In taminating point sources are masked.

zi ) K = P (Hi |D)/P (Hj |D) = P (D|Hi )/P (D|Hj ) (7) on the cosmological and astrophysical parameters is outlined in Equation 5. zi ) profiles which We will then tackle a progressively more complex cosmolog- reflect the proposed parameters θk (Equation 5). CMB.2<z<0.4) from hydrodynamic simulations.g. we For each cosmological model our merit function is repre. complementary data sets D (e. z5 :z>0. We then generated proposed parameters with θ of the cosmological model Hi (e. Hi ) for tion of C. we use the evidence-based criterion (Equation z< A3).2 parameters θk .i and for an detection of the X-ray signal in the outskirts at each redshift zi . we have a feeling on the quality and degeneracies of these constraints. This will allow us to profiles in each bin of redshift. θ = Hi is more strongly supported by data under consideration than {Ωm . which can be regarded as cos.6. which asks the we will consider separately combinations which include clusters. where K is called Bayesian factor. i = 1. .i (total scatter) on the interpolated points were ical probes and contraints on the evolution of fgas from hydrody- propagated to the stacked profiles according to the MC approach namic simulations used as priors in our Bayesian analysis. we perform an adaptive MCMC. trophysical+cosmological parameters.. consists in splitting the chains into 2. This level of inference is repeated for different discussion about this methodology (MacKay 1997). H = wz CDM ). This will described in Morandi et al. zi ) MCMC iterations is ∼ 106 for the post-convergence.g. by capturing the covariance matrix C of the parameters while distribution. with expectation updating each from conditional distributions. In this way. z4 :0. wz CDM etc. Concerning the model comparison. We then repeated the cosmological analysis. the at the lowest redshifts. with uncertainty σj. following a predefined binning in units of by presenting the constraints arising in a flat (and non-flat) ΛCDM R200 (calculated in a reference cosmology) common to all clusters. (2015).8). code to automatically ’learn’ better parameter values ’on the fly’. and a few for the stacked profiles out to R100 . zi ). We obtain parameter constraints using the likeli.1 to our assumption of (weak) self-similarity of the stacked emission measure EM g (x) (the ’likelihood’ P (D|θ.8. and RESULTS we describe our approach below.6<z<0. allow us to constraints simultaneously cosmological models and as- Moreover.4<z<0. For each proposed ical model using the cluster data. α.2).i is the uncertainty (total scatter) of EM g (xj .). 3 or 4 parts and comparing ters (see Figure 1). We thus verified an agreement between input and ∼ are (weakly) self-similar (see §2. we point out that a Hence the likelihood (the proposal distribution) L ∝ Bayesian model selection will be possible since our priors on the exp(−χ2 /2). This rebinning ensures a statistically-significant profiles for each redshift bin zi . C). δ}) of cosmological+astrophysical parame. biased χ2 via the following test. 5. z1 ))2 g (xj . zi ) profiles which reflect the proposed stacked profiles EMg (xj . w0 . cosmology). in order to achieve value θ and covariance C. performed a series of convergence tests: the Gelman & Rubin R sented by EM g . ΩΛ . Assuming that we choose to assign equal priors P (Hi ) to ∼ 0. defin- ing statistical information about fitting variables. In addition to a complete statistical analysis of the chains.e.j σj. SNe) and priors on the as- ison of the scaled emission measure profiles of distant clusters trophysical parameter α (§2. zi ) − EM deed hinges on the constraints on the cosmological parameters due χ2 = 2 2 (6) i>1.8 Morandi & Sun Concerning the model fitting. in order to have > g (xj . as well as via external cosmolog- state θk the errors σj. We calculated the raw X-ray sur- face brightness Sx (θ) profile (expressed in counts/s/arcmin2 . multinormal probability distribution L(θ. 000–000 (0000) .2. (xj = Rj /R200 ) and in the ith bin of redshift.4.g. We divided our Finally. Note that the amount and nature of dark energy and mat- ter density have a very small effect on the scaled emission measure the alternative cosmological models Hi (e. we empirically verified that Equation 6 gives an un- 320 X-ray clusters in five redshift bins (z1 :z<0. zi ) cosmological model Hi is known a priori. cosmological models Hi .e. we used the Metropolis.. models are ranked by evaluating the evidence ratios: The dependency of the stacked emission measure EM g (xj . we cycle through all of the parameters the algorithm runs. and compute new stacked EM g (xj . we created mock EM clusters per bin. MNRAS 000. since we are interested in exploiting statistics (Gelman & Rubin 1992) and a split-test (which essentially the large cosmological information ’buried’ in the outskirts of clus. with This Section presents the cosmological constraints obtained from θ = x R200 /Da ) of each cluster. our analysis of the cluster stacked emission measures. where the We then performed the median to compute stacked EM g (xj . ternal data sets and priors on α from hydrodynamic simulations) in the MCMC analysis. we stack the emission measure of several clus. similarity). the difference in the parameter quartiles). with a local reference profile derived from hot nearby clusters (at In particular..g. For each proposed state θk . This is computationally demanding since we need to re-calculate the stacked emission measure EM g (xj . compared to other constraints from observables pertaining to the lowing a new binning in units of R200 (calculated in the proposed expansion history and structure growth. This but also on priors P (θ|Hi ) on the cosmological parameters from posterior probability allows us to perform a quantitative compar. in a classical Bayesian framework. in order to seek for an improved proposal distri. The typical number of ters grouped in bins of redshifts in order to measure EM g (xj . z2 :0.4 for a definition of weak self. by performing a Cholesky decomposi- convergence to a stationary posterior distribution P (θ|D. The correct cosmology is thus that for which the by inferring new EM g (xj . C). Our analysis in- X (EM g (xj . bution L(θ. where σj. in the jth radial annulus ∼ 104 for the burn-in period. i. Hastings MCMC algorithm for sampling from multi-dimensional i. scale of Kass & Raftery (1995).i + σj. wa . In practice. When combining data sets. and wCDM models. and χ2 reads: cosmological and astrophysical parameters are informative. zi ) ∼ 50 sure averaged on the whole sample. We will start mological independent. zi ) 5 CONSTRAINTS ON THE EOS OF DARK ENERGY: for each proposed state θk in all the radial and redshift bin. zi ) at redshifts zi and for x > 0. We refer to the classical statistical books for a detailed a given data set D. ΛCDM . interpolated the individual surface brightness profiles Sx (θ) fol. From the rescaled emission mea- z3 :0. assumed cosmology. A value of K > 1 means that hood function computed for each proposed state θk (e. Hj . output cosmological parameters. For the interpretation of the Bayesian factor K we refer to the ters affecting cluster observables (and with/without priors from ex. Hi ) in Equation A2).

1 data sets provide Ωm = 0. 2013). α. our analysis provides prior/external data set). from CMB-only Planck temperature while Ωk < 0 to positively-curved 3D-geometries (or closed Uni. in agreement with the uncertainty found by flat ΛCDM model (Ωk CDM ). generacy mostly along the Ωm axis. In this respect. will be discussed in §7. the apparent evolution with z (or lack of) of the ments (Bennett et al.690 ± 0. Bennett et al.005).009 ± 0. with a small (pos- itive) correlation between the two parameters. we have one cosmological and Notes on the complementary data sets: ’Planck’ refers to the CMB analysis two astrophysical parameters. combined with Planck lensing. no evolution of fgas . The base ΛCDM cosmology assumes an FRW metric ter 0. For comparison. We have ysis of EM ence of outlier measurements.1 Constraints on non-flat ΛCDM models straining the cosmological parameters.350 ± 0. (2014). tical errors. As we previously of Suzuki et al. with Ωk = −0.1’ refers to the SNIa compilation size a strong anti-correlation between Ωm and α. ΩΛ . In Figure 4 we can easily identify the role of using external data sets and/or priors on the hot gas fraction evolution in con- 5.305 ± 0. While the two independent data sets are in good agreement. which limits our The impact of several sources of systematics in our analysis. Planck Collaboration (2015) (∼ 0. α.012. For this model. external data sets. Suzuki et al.690 ± 0. The simplifying assumption of large-scale homogene. with Ωk = −0. We obtain Ωm = 0.e. with the external data sets providing a more stringent constraint on the curvature Ωk of the Universe.308 ± 0. Combining all the data sets (without additional priors). outlining how the CMB constraints are more with a flat curvature (Ωk = 0). The cluster constraints on Ωm combined with ity and isotropy on large scales leads to the Friedman-Robertson.310 ± 0. the pres. Clearly.144 for the case without external data sets and/or priors. stringent for ΛCDM in constraining the matter density parameter scription of our Universe as constrained by CMB measurements with respect to those from X-ray-only data (without any additional (Planck Collaboration 2015).e.1 (Planck Collaboration 2015. we have a matter density parame- verses). inhomogeneities of Ωm = 0.022 and α = −0. ’WP’ refers to the WMAP polarization measure- pointed out (§2.1 data sets (dashed line) and constraints on the evolution of fgas from hydrodynamic simulations (dotted line) as pri.001 ± 0. g ray-only analysis of the stacked emission measure EM(x) in the whole Chandra sample (solid line). g (x) without external data sets and/or priors. θ = {Ωm . 2012. ability to infer the desired parameters from our X-ray-only anal- including biases due to the modelling of the background. we ob- tain tight constraints strongly preferring a flat Universe: Ωm = 0. an (apparent) positive evolution with z Planck CMB or complementary data sets. stacked emission measure EM be ascribed to both the cosmic expansion history of the Universe (via the parameter Ωm ) and to the evolution parameter α of fgas due to e. The spatial curvature of the Universe is thus found to be close to zero with an uncertainty Figure 3. the constraints obtained from confidence level. ’Union2. The best-fit values and statis- where for non-flat ΛCDM models (hereafter Ωk CDM ) Ωk = 1− tical uncertainties for Ωm are very close to those derived from the Ωm − ΩΛ . right panel) or with negative evolution of the hot gas tions and systematics. We show constraints arising from our X. For this model.062. The thick long-dashed line shows the marginalized posterior distribu- tion only from the Planck+WP+Union2. non-gravitational processes (Equation 5). as well as priors on the g could be qualitatively achieved both by a smaller value of of EM fgas evolution. θ = {Ωm . Also shown are independent constraints from Planck+WP+Union2. If we add priors on the gas fraction evolution from hydro- dynamic simulations we have Ωm = 0. δ}.116. confidence level.042 and ΩΛ = 0. it is interesting to point out that the degeneracy between Ωm and ΩΛ in these data sets is nearly orthogonal.012. This translates into an intrinsic degeneracy be- parameters. and remove the degeneracy among the fitting fraction (α < 0).4). and by cumulatively adding the likelihood from the Planck+WP+Union2.01. the constraints on the evolution of fgas from g as measured from X-ray data could hydrodynamic simulations refer to the results of Planelles et al. with a significant improvement in the statis- respond to negatively-curved 3D-geometries (or open-Universes). For FRW models Ωk > 0 would cor. it breaks the de- In this subsection. Probing dark energy via galaxy cluster outskirts 9 (with/without standard priors and external data sets) are shown as contours at 1-σ confidence level in Figure 3. The contours are at 1-σ For a flat ΛCDM models (ΛCDM ). 2D marginalized posterior distributions for Ωm and ΩΛ in a non. in order to gauge the impact of different assump- Ωm (Figure 1.037 and Ωk = −0. 2013).2 Constraints on flat ΛCDM models ors.011.021 ± 0.037 ± 0. We empha- of Planck Collaboration (2015). consistent with the gas distribution and cosmic filaments.015 and ΩΛ = 0. Planck+WP+Union2. With respect to the correct cosmology. an independent method to test empirically this assumption. ΩΛ = 0.058 ± 0. while the expectation value is represented by a point and the full X-ray data set are shown in Figure 4 as contours at 1-σ refers to the combination of all the data sets. tween astrophysical and cosmological parameter. i. we have two cosmological and two astrophysical parameters.1 data sets. δ}. selection effects. of the order of 0. α = −0. i.039. Adding the prior on α from hydrodynamic simulations leads Our constraints obtained from the full X-ray data set to the CMB-free constraints on the matter density parameter MNRAS 000. (2012).708 ± 0. 5.046. we investigate constraints on the parameter Ωk . that appears to be an accurate de.019.g.014 and Walker (FRW) metric. as can be seen in the figure. 000–000 (0000) .311 ± 0.318 ± 0.

In order to gauge the impact of these value of δ (the parameter describing the temperature-dependency systematic uncertainties in our recovered parameters.g. and by cumulatively adding the likelihood from Planck+WP+Union2. since non-radiative hydrodynamic simulations have incorporated more accurate models of baryonic simulations likely do not capture the correct physics in the ICM. This gas fraction evolution significantly improves the parameters con. see Rasia et al.g.g. (2015) we also found a lack of dependency of the use of the other predictions on α for CSF and NR simulations. e. such that the clus- biases in our analysis (when these priors are included). ’Union2. the In Morandi et al. 000–000 (0000) . For all the panels. we have Ωm = and the impact of the particular baryonic physics implemented 0.1 data sets (dashed line) and constraints on the evolution of fgas from hydrodynamic simulations (dotted line) as priors. as estimates the true systematics. systematic differences between different sets of hydrody. while the expectation values are represented by a point and refers to the combination of all the data sets. a trend with cluster temperature (or mass) of fgas .006 ± 0. in particular modelling the effects of feed. by combining all the data sets and priors. we used the of fgas . e. Left panel: 2D marginalized posterior distributions for Ωm and α (the parameter describing the evolution with z of fgas . These are based on one set of non. MCMC analysis performed on mock data sets (see §7. we show constraints arising from our X-ray-only analysis of the stacked emission measure EM(x) g in the whole Chandra sample (solid line).034±0. as we verified by fixing the temperature-dependence to zero. However. By incor. Similar conclusions hold by consid- ter recover the baryonic content of clusters (Battaglia et al. (2014)) Thus. Notes on the complementary data sets: ’Planck’ refers to the CMB analysis of Planck Collaboration (2015). 2014) suggest that the use of these theoretical pri. in their er- one of which also accounting for the effect of feedback from AGNs. ters in our sample should show little dependence of the physical porating these different predictions on α as priors in our Bayesian properties on the global temperature. (2014). ering the predictions on the gas fraction evolution of Battaglia et al. must be compared to statistical uncertainties ∼ 5% (with external straints.g. We also quote the ors on α merits some caution. (2014).1 for fur- tainties and in agreement with the CMB-only constraints.e. a negligible effect on our cosmological constraints (e.1’ refers to the SNIa compilation of Suzuki et al. 0. Indeed most of the clusters MNRAS 000. 2013). which is competitive in term of statistical uncer.10 Morandi & Sun Figure 4. (2012). Equation 4): δ = 0.307 ± 0.5% and < ∼ 1% While the advantage of introducing theoretical priors for the for the CSF and NR hydrodynamic simulations.011. the constraints on the evolution of fgas from hydrodynamic simulations refer to the results of Planelles et al. Right panel: 2D marginalized posterior distributions for Ωm and δ (the parameter describing the temperature-dependency of fgas . to estimate these systematics. Equation 4) in a flat ΛCDM model.009 and α = −0. (2013). can give us a sense of the impact of these ature range spanned in our sample (> ∼ 3 keV). ’WP’ refers to the WMAP polarization measurements (Bennett et al.309±0. Equation 4) in a flat ΛCDM model. The level of bias from NR simulations probably over- ical predictions might deserve some prudence. 2013).072. rors). with no evidence for results Planelles et al. marginalizing over an fgas − kT slope has radiative cooling. e. Increasingly. This is probably explained by the relatively narrow temper- the physics prescriptions. While we usually refer to their AGN results as baseline prior. namic simulations (e. physics in clusters. Note that. We shall therefore focus on the two cases AGN and CSF in order back from AGNs in cluster cores. The contours are at 1-σ confidence level. data set). ther details on the mocks).024.g. no evolution of fgas ). with the latter quantity still (Planelles et al. respectively. star formation and feedback from SNe (CSF). unlike radiative (NR) simulations. they have become able to bet. smoothed-particle hydrodynamics (SPH) or adaptive-mesh refinement (AMR) codes. consistent with zero (i. gas fraction on the cluster temperatures (in a cosmology-dependent which encompass the aforementioned systematic uncertainties in way). and two sets of simulations including the fgas − z relation. we conservatively estimate that the bias (downwards) on the matter density parameter is < ∼ 0. we caution the reader that the reliability of the theoret.

by combining only Planck+WP+Union2.123 ± 0. consider here the case of a Taylor expansion of w at first order ment with (or departing from) -1. usually based on a spatially flat Universe (Ωk = 0).40+0. We have w = −1.009 ± a local reference profiles derived from the sample of 15 hot nearby 0. can thus elucidate the nature of the dark energy (Chevallier & Polarski 2001).4 Contraints on flat wz CDM models the energy density in the equation of state of the dark energy com- ponent. w0 = −0. it is likely to change with time.026). Hence our choice is to present directly der of 0.3 Constraints on flat wCDM models Mantz et al.311 ± 0.2). hence we cannot constraint H0 . WMAP and BAO measurements. This trend is also con. however.g. in excellent agreement with ies of the past 20 years (Riess et al. and is therefore compatible with a eters.030.1 Adding priors on the cosmology from Planck+WP+Union2. only a moderate improvement is achieved by the scaled emission measure profiles of these distant clusters with adding priors from hydrodynamic simulations (i. containing matter and dark energy. The acceleration can be the present work. as discussed 0. see §5. by means of measurements attributed to the presence of a significant energy density component of the cluster gas fraction. An alternative model is Robertson-Walker metric. (2006) and Allen et al. w takes the value w0 driving the accelerated expansion of the Universe. While the sta- by Vikhlinin et al. we conser- vatively estimate that the bias on w is <∼ 0. which does not appear to depend on the selected tween ours and these external data sets (Figure 5). Figure 5 right panel). combined with the CMB and other cosmological data sets. see §5. they obtain w = −0. which stems from the degeneracy with the hot gas model.e. for objects with the external data sets. defined as the ratio between the pressure and 5.991 ± 0. we compare our results with Arnaud et al.039(sys). We adopt an evolving param- scalar field. and determine at the present day and w0 + wa in the high-redshift limit. when they are keV. remaining fixed to zero. due to the two-dimensional geometric de. observed with ROSAT. with negative pressure commonly referred as dark energy.036. in a dynamical dark energy (Caldwell et al. (2002). (2014). w = −1. by incorporating differ- gent constraints on the matter density of a ΛCDM model. Using (strong) self-similar evolution model. w of the order of 3%. Accurate measurements of w in agree. fitting cluster mass function Recent accelerated expansion of the Universe detected in the Hub.079 and wa = −0. 2014). w0 = −1 and wa = 0. thanks to a nearly orthogonal degeneracy between ours constraints of the CMB in combination with lower redshift distance and these independent data sets (Figure 6). as well as an possible evolving in the scale factor.020 ± 0. Measurements Ωm = 0. While the statistical uncertainty on (Planck Collaboration et al. The former. Curvature Ωk is not allowed Ωm = 0. Note that the CMB alone would the matter density parameter is driven by the external data sets. (2009) and 5. We remember that the absolute normal- Planck+WP+Union2. −0. parameterized by w(z) = w0 + wa z/(1 + z) value of w = w(z).051 from X-ray-only data sets. wa = 0.654. models (wCDM . we emphasize the stringent constraints on kT > 5 keV.319 ± 0. more accurate of those from Planck+SNe firmed by the level of the intrinsic scatter of fgas (∼ 15% at R500 . adding priors on the cosmology from Planck+WP+Union2.093. or eterization of the EoS of dark energy which allows w to depart alternative models which calls for modifications of general relativ. A cosmological constant in the context of general relativity In this Section we consider cosmological models with a Friedmann- corresponds to a non-evolving w = −1. In this model. The data are thus consistent with again the non-negligible statistical uncertainties on the matter den.017.1 data tion with temperature (Sun et al. The dark energy is w = P/ρ. We compare our results with Vikhlinin et al. we consider spatially flat models with a model (ΛCDM .09± data sets greatly reduces the degeneracy among the fitting param- 0. with our The resulting constraints on w0 and wa are shown in Figure constraint on the equation of state being w = −1. and the cumulative combination g (x) (which depends on the value of H0 ) is a nuisance ization of EM of priors from external data sets and hydrodynamic simulations.98 ± 0. wa = 0). If w differs from -1. making our prior-free results robust towards any theoreti- clusters.993 ± 0. A non-negligible improvement is achieved by adding priors MNRAS 000. (Planck Collaboration 2015). 1998).4. tioned degeneracy between ours and these independent data sets Finally. 000–000 (0000) . explains why dark energy constraints from EM g (x) data is com- They study the surface brightness profiles of a sample of 25 distant petitive with those e.305 ± 0. 1998).014 and α = −0. ters implies an increasing trend of the cumulative gas mass frac. obtain w = ble diagram for distant SNIa is one of the most significant discover. stress the stringent constraints on wa with an uncertainty of the or- generacy in these models. parameter in our analysis. Even for this general sity parameter. 2009). the ΛCDM model (w0 = −1.058 and 6 as 1-σ confidence level contours.036 ± 0. Probing dark energy via galaxy cluster outskirts 11 (264 out of 320) in our sample have temperatures greater than 5 The real strength of the cluster data is.3%. For comparison. in §5. and constant-w constant-w dark energy equation of state (wCDM model). tistical uncertainty on the matter density parameter is driven by tent with being constant with cluster temperature. (2014)). ent priors on α as priors in our Bayesian MCMC analysis (AGN. or X-ray-only data.26. We ity on cosmological scales. in agreement with our findings.045(stat) ± 0. thanks to a nearly orthogonal degeneracy be- ∼ 25% at R200 ).15 −0. The aforemen- subsamples.009 ± 0. Indeed.046 cosmological constant. While the gas fraction in groups and intermediate mass clus. We have Ωm = 0.3).2. as assumed in the base ΛCDM cosmology and wa = −0. from SNIa (Suzuki et al. jointly with the SNe. The different cosmological probes are in good agreement.12 at 90% confidence level. This pa- whether dark energy is truly the cosmological constant. which typically has a time varying w with w > −1. Our constraint is presented evant cosmological parameter for the analysis of cluster data is the in Figure 5 (left panel) along with independent constraints from present-day matter density.308 ± 0. 2012) and CMB and hot clusters.995 ± 0. w0 = −0.400. it is less clear whether sets greatly reduces this degeneracy. CSF and NR simulations from Planelles et al. rameterization contains as special cases the cosmological constant In this respect.1 data sets. the rel- ΛCDM model (where w = −1).19 (95% confidence level). our X-ray-only analysis provides a tight constraint Ωm = fraction evolution (parameter α. from -1.010±0. They performed a quantitative comparison of Moreover.1 the equation of state for dark energy is constrained to w = −1. we not strongly constrain w. As for the latter work. such a trend persists for relatively massive objects. measures in order to break the degeneracy among the parameters. tures from ASCA. Next. We underscore to vary. obtain. This represents the most basic phenomenological extension of a Beyond the dark energy equation of state w = w(z). (2008) are both consis. they put strin. ing Ωm = 0. cal assumption on the fgas evolution. with published tempera.

w0 = 6. can thus cast in the light of the data.g.03 ± 0. ∼ 0. we perform a second level of Bayesian inference. Ac- In this Section. by In this Section we wish to measure the relative quality of statistical incorporating different constraints on α from hydrodynamic simu. We finally compare our result with Mantz et al. Equation 4) in a flat wCDM model. The contours are at 1-σ confidence level. ’Union2. For all the panels. 2013) they obtain question arises whether the dark energy w = P/ρ (the ratio be- Ωm = 0. as well as a possible evolving value of w = w(z). Assuming that we choose to assign equal priors P (Hi ) to the on the cosmological and astrophysical parameters are informative. Here we wish to compare the models -1. An alternative model is dynamical dark energy. A cosmological constant in the context of general relativity corre- sponds to a constant EoS of the dark energy parameter w = −1.298 ± 0. Left panel: 2D marginalized posterior distributions for Ωm and w in spatially flat cosmological models with a constant-w dark energy equation of state (wCDM model). we used the evidence-based criterion (Equation that this Bayesian model selection will be possible since our priors A3).1+0. curate measurements of w in agreement with (or departing from) which is a model comparison.997±0. However.g. 2013). of the dark energy component) is truly the cosmological constant. (2014). model (ΛCDM ). 1998). While the recent accelerated expansion of the Universe de- tively. we show constraints arising from our X-ray-only analysis of the stacked emission measure EM g (x) in the whole Chandra sample (solid line).6 .4% and < ∼ 0. sion of the Universe.015.1 data sets. Given different cos- lations (e. Fit. 1998).034 and wa = −0. the WMAP polarization (Planck Collaboration et al.081±0. tected in the Hubble diagram for distant SNIa is one of the most ting cluster mass function jointly with the 1-year Planck data plus significant discoveries of the past 20 years (Riess et al. Note In this respect.2) are more likely given the data.12 Morandi & Sun Figure 5. cosmological models for a given set of data.1’ refers to the SNIa compilation of Suzuki et al. CSF and AGN simulations) as priors in our Bayesian mological models for the data. the constraints on the evolution of fgas from hydrodynamic simulations refer to the results of Planelles et al. while the expectation values are represented by a point and refers to the combination of all the data sets. light on the nature of the dark energy driving the accelerated expan- ing to our different assumptions.1 data sets (dashed line) and constraints on the evolution of fgas from hydrodynamic simulations (dotted line) as priors.1 Comparison of CDM models −0.267). with an improvement of a factor of 1. from hydrodynamic simulations (Ωm = 0. ’WP’ refers to the WMAP polarization measurements (Bennett et al.18 and wa = −0. Notes on the complementary data sets: ’Planck’ refers to the CMB analysis of Planck Collaboration (2015). e.309 ± 0. We will determine which cosmo.010. w0 = −1. we conservatively estimate that the bias on the model is more likely’ given the data. Only a value very close to -1 would determine logical model ΛCDM . in excellent agreement with the present work. (2012). providing the best trade-off matter density parameter and w0 is < between the goodness of fit of the model and its complexity. and by cumulatively adding the likelihood from the Planck+WP+Union2. alternative models (e.g. wCDM and wz CDM (§6. we would like to determine ’which MCMC analysis.6 tween the pressure and the energy density in the equation of state −0.3%. tion of boundary radius (§6. 000–000 (0000) .5 for the statistical uncertainties on wa . The thick long-dashed line shows the marginalized posterior distribution only from the Planck+WP+Union2. respec. (2014). wz CDM ) with respect to the reference defining statistical information about fitting variables. Right panel: 2D marginalized posterior distributions for Ωm and α (the parameter describing the evolution with z of fgas .1) and defini. or models which 6 CONSTRAINTS ON THE EOS OF DARK ENERGY: calls for modifications of general relativity on cosmological scales. and assign some sort of preference or rank. MODEL SELECTION which typically has a time varying w (Caldwell et al. we rank cosmological models by evaluating the MNRAS 000. that the dark energy is truly the cosmological constant.

e.g. For all the panels. The thick long-dashed line shows the marginalized posterior distribution only from the Planck+WP+Union2. ’Union2. while the expectation values are represented by a point and refers to the combination of all the data sets. (2014).g. Equation 4) in a flat wz CDM model. (2012). Notes on the complementary data sets: Planck’ refers to the CMB analysis of Planck Collaboration (2015). and by cumulatively adding the likelihood from Planck+WP+Union2.1’ refers to the SNIa compilation of Suzuki et al. MNRAS 000.1 data sets (dashed line) and constraints on the evolution of fgas from hydrodynamic simulations (dotted line) as priors. Equation 4) in a flat wz CDM model. the constraints on the evolution of fgas from hydrodynamic simulations refer to the results of Planelles et al. Top-right panel: 2D marginalized posterior distributions for Ωm and wa in a flat wz CDM model. 2013). 000–000 (0000) . Bottom-left panel: 2D marginalized posterior distributions for wa and δ (the parameter describing the temperature-dependency of fgas . Probing dark energy via galaxy cluster outskirts 13 Figure 6. Top-left panel: 2D marginalized posterior distributions for w0 and wa in spatially flat cosmological models with an evolving dark energy equation of state w(z) (model wz CDM. ’WP’ refers to the WMAP polarization measurements (Bennett et al. The contours are at 1-σ confidence level. we show constraints arising from our X-ray-only analysis of the stacked emission measure EM g (x) in the whole Chandra sample (solid line). e. Chevallier & Polarski (2001)).1 data sets. Bottom-right panel: 2D marginalized posterior distributions for wa and α (the parameter describing the evolution with z of fgas .

we wish to determine by means of observables dark energy given the data. By inferring EM 4 Only in this Section we will use the notation ∆c rather than ∆ to explic.g.z 4 . in our Bayes factor calculation we Eckert et al. other definitions of ’virial’ radii hinge on the aver- age matter density of the Universe. non-flat geometry etc.z = 18π 2 + 82(Ωz − 1) − 39(Ωz − 1)2 (8) to consider the evolution with z of the physical properties. with a redshift-dependent or constant overdensity.z Ωz (Lau et al. N-body simulations indicate that the dependency of the over- caution. where the X-ray signal can be measured reliably. Finally. 2011. boundaries. the standard of the large scale structure formation (Kravtsov & Borgani 2012. verse ρc.g.g. This cluster boundary definition is conventional and 3) a constant overdensity ∆m with respect to the average matter it is borne out of the spherical top-hat model for non-linear col. of the ICM of our population of clus- Thus. physical boundaries. This generalization of the parameter ∆c. since it is close to the value 18π 2 for the critical Einstein-de Sitter (EdS) cosmological model (with In general.g. We arbitrarily rescaled MNRAS 000. since we do not particular. this overdensity has also a dependence on the cosmology and redshift through the parameter different definitions of cluster boundary: Ωz = Ωm (1 + z)3 /Ez2 . The fiducial radius of a cluster is thus defined as the radius of a sphere enclos- that clusters are identical objects when renormalized by their mass or temperature.. virialization region (Vikhlinin et al.g. strictly speaking. or even distinct. However. while one would like to define that boundary so as to Here we are interested to compare three definitions of bound- maximize the simplicity of the relationships between cluster mass ary radius. Planck Collaboration et al. 2013). factor K ∼ 288 with respect to a non-flat ΛCDM model.z .2 Boundary radius definition ρm. namely the scaled emission measure. 2006. the relations presented in chooses to define a clusters outer boundary. (∆ = {∆c.z for ies of the cosmology.z and ∆m . 2015).z /∆c ) . However. the question arises whether the general- model including e. ΛCDM models: Here we propose to use self-similarity to discriminate which of the definition of cluster boundary is more tenable.14 Morandi & Sun evidence ratios. dropping the subscript ’c’. and the consistent with zero and poorly correlated with the other parame. recover the definition of ’virial’ radius R∆ (which enters in Equa- itly indicate an overdensity with respect to the critical density. Thus.z g (x) ∝ T −1/2 fgas EM X −2 −3+(1−6β) fz Da1−6β (9) can only be calculated by numerically integrating the equations of 1/2 motion for the spherical top-hat collapse in a ΛCDM cosmology. 000–000 (0000) . the ΛCDM cosmology is more supported. which are not captured by the spherical top- spectrum of adiabatic scalar perturbations. z)3/2 } for the aforementioned overdensities ∆c . ∆c = 500 (fixed overdensity) has been used in the anal- require any bound for this parameter.1 we described the self-similar model. we fixed δ = 0 and we combined all the data sets and priors. 2012). for the sake of simplicity a toy model depicting the evolution of the boundary radius R∆ in the notation. Since cluster’s mass and all the relations link- ing the mass to other observable quantities depend on how one lations (Jenkins et al. a robust definition of §2. and rescaled according to a certain radius (e. The idea is ∆c. ing an average matter density equal to a reference overdensity ∆m R200 ) with respect to the reference background density of the Uni. density of the Universe. and thus In the literature ρc. ization of the parameter ∆c. we verified that this parameter is substantially a nuisance. (1 + with the assumption that the baryonic matter follows the dark mat. being volumes. (1998) provided fitting formulae for ∆c.g. In Figure 7 we present use ∆ (e.∆c. Using ρm. Thus.z has been widely used to define cluster the latter is strongly disproved. what is the best definition of cluster boundary. 6. For exam- Therefore.g. times an average background density ρm. depending on Ωm and w = w(z). can expect that the ICM in galaxy clusters features some hydro- Finally. e. which is independent of other In §2. The ΛCDM model shows a Bayes responds to fz ≡ Ez (see below Equation 9).z . Elsewhere we tion 9) which better preserves self-similarity. The expansion history factor fz = {Ez . our quantitative Bayes factor calculation merits some ple. the relations presented in §2. rather than that critical. However. re- ter and neglecting effects of hydrodynamics in structure formation. 2008). 2004). Ultimately. Thus.z = ρc. Simionescu et al.z with respect to the crit- total density is ∆c = 200 times the critical density of the Uni. dynamic dark energy. 2001). 2012.z /∆)1/2 (see also showing a Bayes factor K ∼ 7(13) with respect to the alternative Ettori et al. Ez (∆c. ters (see Figures 4 and 6). the Hubble parameter. clusters formed from hierarchical clustering 1) a constant overdensity ∆c with respect to the critical density do not possess unique. While the exact Bayes factor yses of Chandra and XMM-Newton X-ray observations of galaxy calculation may slightly depend on the allowed range of values of clusters. namely based on: and other observables. ’virial’ radius is clearly important. as well as an highly non-spherical topology of the dark energy parameter w) with respect to e.z in Equation 8 is a nuisance. which captures a redshift-dependent overden- model wCDM (wz CDM ) (for the scale for the interpretation of sity parameter ∆c. and we underscored cosmological parameters.g. we can rewrite Equation 3 and highlight the de- pendency of EM g (x) on the cosmological parameters with these Ωm = 1). and the angular diameter distance Da are thus the prox- Eke et al. ical density of the Universe. In this case. evidence of non-constant (w 6= −1) or time-evolving (w = w(z)) 2008).z to define DM halos also leads to universal mass function in N-body simu- verse (Kaiser 1986). or any extension of this hat collapse model. The special case of constant overdensity cor- K see Kass & Raftery 1995). replacing Ez with the term fz ≡ Ez (∆c.z . baseline model defines R200 as the radius within which the mean 2) a redshift-dependent overdensity ∆c. we caution the reader that our analysis can constrain dynamic modifications compared to the gravity-dominated scale- only a small subset of parameters (e. In Note that the prior on δ is non-informative.1 can be generalized by ters by rescaling with respect to various definitions of boundary g (x) (x = R/R∆ ) at different z. Evrard et al. spatially-flat six-parameter ΛCDM cosmology with a power-law Böhringer et al. although semi-quantitatively we emphasize that there is no density parameter on the redshift can be neglected (Evrard et al. we can radius. e. spectively. matter density and EoS free DM distribution. lapse (e. since it corresponds to the transition between the inner δ.1 can be generalized by setting fz ≡ (1 + z)3/2 and replacing Ez with fz . ∆m }) renormalized by R200c . ∆ = 200). Our of the Universe.

simplified toy-method to validate our cosmological analysis (§7. thus introducing unphysical substructures. As for ∆ = ∆m . we caution the reader that those findings are cosmological-dependent given the assumed cosmological model. The effect is larger for ∆ = ∆m and moving towards the outer vol. In other words. files holds out to 3 × R500c ∼ 1. (2012). we are able to constrain ICM profiles out to R100c . Planck Collaboration et al. given the low signal-to-noise in the high-z sources departures from self-similarity. outlier measurements and selection effects. lowering the value of the Bayesian evidence. However.25). As we can see. Lau et al. following the In this respect. we caution the reader that R∆ in such a way R∆ /R200c = 1 at z = 0.2). while it breaks if we introduce the alternative definition of boundary radius (R∆m ). Finally. Moreover. Probing dark energy via galaxy cluster outskirts 15 by combining all the data sets and priors (for the scale for the in- terpretation of K see Kass & Raftery 1995). By rescaling the profile by may lead to higher emission measure. We start by presenting a as previously discussed (§2. we observe that EM g at high-z is ∼ 3 − 4 Figure 7. for an as- ary radius R∆ shows larger values for the high-z cluster popula. 2015) in- evolution of this ratio with redshift. which stems from the lower stacked signal. our previous results (Morandi et al. and study the evolution of EMg (R/R∆m ) vs EM g (R/R200c ) with z in the outer volumes (R > ∼ R200c ). We arbitrarily rescaled R∆ in such at ∼ 6σ statistical significance level. 000–000 (0000) . (2015) found that the self- similarity of the outer ICM profiles is better preserved if they are normalized with respect to the mean density of the Universe. in agreement with the present umes due to the steeper emission measure profiles. while the inner profiles are more self-similar when normalized us- ing the critical density. with K ∼ 2×102 of the gas distribution and cosmic filaments (§7. which is an alter- by considering the two cases ∆ = ∆c.2) describing the scaled emission mea. 2015).3). which is statistically disproved {∆c. Böhringer et al. In order to determine which definition of boundary radius bet. thus it is still tion of this ratio with redshift. We point out that we cannot apply the previous Bayesian inference to this case. a way that R∆ /R200c = 1 at z = 0 and for Ωm = 0. we discuss the impact of inhomogeneities by our data with respect to the alternative model. (2008). we con- firmed our previous findings that self-similarity is very well pre- served with respect to a fixed critical overdensity. (Lau & Nagai. We then discuss the impact of outlier measurements by discussing ter preserves similarity of the profiles we calculate the Bayes factor the results via a Bagging-MCMC analysis (§7.1). Moreover. redshifts. By propagating all the uncertainties related to statistical+systematic measurement errors (Morandi et al. due to biases in the recovered physical parameters.z ) would bias downwards the estimate of Ωm (∼ 0. personal communication).4). Clearly this value of Ωm is also in tension with the Planck cosmology. we should observe a break of it if we intro- duce other definitions of ’virial’ radii. similarity of the shapes of the pressure pro- tion. 2015) and to the cosmological parameters. dicate that filaments can contribute significantly to the emission at g ∆ /R200c ) (renormalized in such a way that it is equal to unity sure EM(R large radii.z and ∆ = ∆c in deriving native method with respect to the standard MCMC analysis previ- g (x) at various redshifts. In this Section we tackle the impact of systematics in our analysis Ultimately. MNRAS 000. The emission measure profiles have been scaled according to z via the method we employed (see Morandi et al. thus introducing unphysical departures from self-similarity. independent SZ analyses suggest that.z . In the latter case. that is we are probing progressively more better preserved if cluster observables are normalized with respect internal regions (with respect to R200c ) as we are moving to higher to the mean density of the Universe. Filaments could be more difficult to remove at high- at z = 0). analysis. the models (∆ = ∆m and ∆ = ∆200c ) being not independent and overlapping out to large radii (out to ∼ R500c − R200c ). Thus.3. The ensuing (rescaled) emission measure at the bound. Upper panel: evolution of the boundary radius R∆ (∆ = larger than the local cluster values. this ratio is smaller possible that beyond our boundary radius the self-similarity might than the unity at high-z. we would like to disentangle the evolution of the cos. A simpler approach is to re- peat the exercise presented in the toy model of Figure 7 for real data. and we confirmed the aforementioned results but at smaller statistical significance (∼ 3σ). larger EM While we proved in our previous work (Morandi et al. following the evolu. We thus repeat the analy- sis by masking the filaments. which the standard self-similar model (Equation 3). the presence of mic expansion history from the evolution of the baryon properties. ∆m }) renormalized by R200c . the ensuing emission measure at the boundary radius R∆ shows tribute to observed biases if we do not properly remove undetected larger values for the high-z cluster population. Lower panel: toy model (a β-model profile with slopes β = 2/3 and 1. 2015) that self-similarity is very well preserved in a framework where 7 ANALYSIS OF THE SYSTEMATICS ∆ = ∆c . the use of this alternative definition of boundary radius (∆ = ∆c. Similar conclusions have been drawn by Evrard et al. if self-similarity of the X-ray observables holds by rescaling them with respect to R200c . 2013). 2010.5R100c (Plagge et al. ∆ = ∆c is strongly supported EM ously discussed. with high-z clusters having g (x) with respect to the low-z population. Subclumps could also con- R200c . sumed cosmology.

characterized by deep ob. which is (by definition) an outlier- In this Section we wish to determine whether our results are robust g (x) out to large radii via resistant estimator. (2014). are largely unrelaxed. the proposed get. and iv) dividing our sample into two laxation state etc. intrinsic scatter and measurement errors in the used for constraining the cosmology.1.g. good agreement for all these cases with respect to the whole sample lier emission measure profiles has some ramification on the stacked analysis. in particular with respect to the stacking in bins of redshift (since we are interested in exploiting the modelling of the background.g. We only require (weak)-similarity and astrophysical parameters deserves further investigation. A Bagging-MCMC analysis can be thus used as a ground spectrum. fitting the model to each sample. In this respect. hence effect of clumpiness (Morandi & Cui liably measured and reliably compared to simulations. of the or.g. sumptions) with respect to the former approaches. z) and hence on the cosmological parameters. irrespective of the quality of the data. We finally applied the whole low signal-to-noise SX in the outskirts of individual clusters) on cosmological analysis described in §4 on the mock observations. fgas . The result is a In this way we can capture systematics due both possible spatial smoother prediction that is less affected by outliers or particularly variation of the CXB and temporal variations of the particle back. ibrated e. clusters with total net counts smaller and larger than our results are robust towards the modelling of the background. Mantz et al. sion measure.4%. in prep) and hence emission measure. temperature. Unlike the gas servations (∼ 2 Ms). Vikhlinin et static equilibrium and three-dimensional geometry (since we do not al. A g (x. we conservatively estimated an upper limit to the systematics on the cosmological pa. providing an independent way (and hinging on different as- other hand. We confirm EM comparison between the true and inferred cosmological parameters a very good agreement between the standard and Bagging-MCMC reveals the measurement errors fairly represent the error budget and analysis. 000–000 (0000) . ing systematics. This is probably due to the presence of non-thermal pressure method has also the major advantage that it is more direct. also. of the (rescaled) emission measure profiles. g (x) is a nuisance pa- the fact that the absolute normalization of EM tal and gas mass (and hence the measured gas fraction) can be re. Thus. indicating that our baseline self-similarity assumption is emission measure EM g (x. the cosmological parameters. Thus. e. asphericity and non-thermal pressure (Morandi & Limousin have an exquisite acknowledge of the systematics for individual 2012) do not have a major impact on the recovered cosmological measurements. Bagging hinges on bootstrap realistic backgrounds SX (x) for each observation. pared to other techniques (e. to. We validated our cosmological analysis via a simplified toy- see Figure 1) rather than using individual emission profiles. greater and smaller than 7 keV) but include all the clusters. From the rescaled emis- clusters and outliers in the stacked signal. §5. with systematics in the recovered cosmological parame- our recovered parameters are not significantly biased by uncertain. a cluster like A133. The entire procedure closely mimic the stacking tool to gauge the ramification of outlier measurements (which may analysis from which we infer the rescaled emission measure EM g reflect systematics. On the sky.5%. As a further test. on a coarser redshift rebinning.e. here we independently run bkg Bagging-MCMC simulations (§A3). While in the previous section we proved that subsamples. mass function).2.16 Morandi & Sun 7.g. aments. z) in each bin of redshift.g. Other clusters. and hence on tenable. according to the observation aspect information. where the geometry. based support and triaxiality. respectively. Us- method where we generated simulations of mock data sets includ- ing stacked profiles adds the benefit to infer SX out to large radii. ters < ∼ 0. (2015. which has been proved We emphasize that in our cosmological analysis we make use to be a tenable assumption outside the cluster core. The aforementioned validation tests suggest that our proposed servations (∼ 200 ks) shows a large gas fraction (and thus larger method is robust towards systematic due to background modelling. noisy data points. We thus added these backgrounds to the source brightness. has a low value of the gas fraction (< ∼ 0.s (R) for each observation. ii) masking the cosmic fil- by systematics due to e. with resampling.b). the median value of the whole sample.2 The impact of outlier measurements tively relaxed objects with centroid shift 6 2. in order to 2014). as described in Morandi et al. we created mock azimuthally-averaged source brightness images SX. In particular. no assumption of hydro- Vikhlinin et al. We find very the question arises whether the inclusion of some clusters with out. We then used blank-sky fields to produce in a standard MCMC framework (§4). the infer the three-dimensional gas and total mass). see. i. 2011a. and then aver- ter data. and ii) we infer EM towards our stacking methodology. re. iii) splitting the sam- ing problem. which boosts the X-ray surface brightness on observed quantities (the emission measure) on the plane of the and gas fraction measurements (Morandi et al. rameter in our analysis. a cluster like A1689. we repeated the cosmological analysis by considering: i) only rela- 7. observing time. our approach distribution. Morandi & Cui 2014. on N-body simulations) are required. This method closely follows the methodology dis- while minimizing the impact of measurement errors for individual cussed in Morandi et al. large cosmological information ’buried’ in the outskirts of clusters. (2015). relaxation state. is a challeng. selection effect and the presence of outlier measurements. which is non-negligibly larger than the cosmic bud. characterized by ultra-deep ob.1). MNRAS 000.g. e. For example. we believe that the ramification of stacking parameters.e. who only consider mas.1 Validation of the cosmological analysis primarily of: i) the median of the distribution of the (renormal- ized) emission measure profiles. which under- went a reduction procedure consistent with that applied to the clus. stellar baryon fraction or model of mass function (cal- is different from e. ties in the background modelling. This stems from sive and morphologically relaxed systems. averaged on the whole sample. which is the quantity we rely on). we verified that our results are robust towards rameters due to background modelling < ∼ 0. systematics by considering cluster sub-samples based on redshift. background modelling. emission measure. Com- der of 0. homogeneous gas Bullet clusters. a for While our cosmological results have been mostly carried out an assumed cosmology. 2006. fraction or the mass function methods. this cosmological test is also relatively insensible such heterogeneous sample in recovering the desired cosmological to X-ray calibration systematics. possibly affected proxy of the dynamical state of clusters. aging the predictions to get the bagged prediction. More in details. We remember that in our cosmological analysis we ple in two temperature bins (i. this quantity being a Making unbiased predictions from noisy data.

000–000 (0000) . We observed that the amount sion measure profiles of a sample of 320 hot (kT > 3 keV) galaxy and nature of dark energy and matter density have a very small ef. 2013. Morandi et al. Constraints on the cosmology are further cal uncertainties with respect to the lower-z (z < 0. similar model of cluster formation. since the absolute normaliza- nearly independent of the cosmological parameters of interest) to tion of the recovered stacked emission measure profiles is a nui- investigate the nature of dark energy.046 and with an improvement on w of a factor of 2(1. dency upon the system temperature). and ii) masking the cosmic filaments (where most of high-z targets. We thus repeat our cosmological analysis factor ∼ 2 − 3). high-z clus. as astrophysical parameters (gas fraction evolution and its depen- In particular. MNRAS 000.017. Clearly. 2013. Probing dark energy via galaxy cluster outskirts 17 7. The absolute value of the clumping in each cluster-centric as nearby objects (which provides a baseline emission measure. as well ble. stant with w ≡ −1. realizing this potential will also require Battaglia et al. and for which put constraints on the cosmological parameters (i. as we already observed (§2. in turn. radius is irrelevant for our purpose. to pendent of the cosmological parameters of interest.308 ± 0. with no evidence for dynamic dark energy. in order to mass-dependency of C). Morandi et al. The derived fect on the scaled emission measure at the lowest redshifts. we tion evolution from larger samples of hydrodynamic simulations. since the quality of in- A significant improvement on these constraints could be dividual observations does not allow to infer the clumping factor achieved. clusters. 9 CONCLUSIONS Measuring the EoS of dark energy is one of the largest efforts of 8 PROSPECTS FROM FUTURE SURVEYS observational cosmology. Current data favor the cosmological con- sets+priors on the gas fraction). of given temperature appear smaller (R200 ∝ Ez−1 ) and brighter This could translate into biases in the recovered cosmologi- (EM0 ∝ Ez3 ) with increasing redshift. In particular. allows us to put a tight constraint on the dark energy model.e. ters at z < 0. Ωm = 0. the comparison of the baseline emission mea.5) in term of statisti. since we are interested in the stacked X-ray signal the accretion happens.3 − 2 at R200 (Nagai & Lau 2011. as we discussed. and. clude that there is no significant bias in the recovered cosmological parameters due to inhomogeneities of the gas distribution and the presence of cosmic filaments (< ∼ 0. wards the emission measure (Morandi et al. data. Therefore. we have w = −1.1). only a moderate investment of time by: i) correcting the measured emission measure for the clump- by flagship X-ray telescopes will be required to observe these new ing factor. which are boosted by gas inhomogeneities z)−4 .5%. we can elu- Section 5 outlines the constraints on cosmological parameters ob. since e. profiles of distant clusters with a local reference profile derived In this work based on Chandra data we analyzed the emis- from hot nearby clusters (at z < ∼ 0. with weakly self-similar. the contribu- the discovery of new clusters at redshifts z > 0. (2015) (see their Equation 10 which outlines a mild significant improvements in the physics in simulations.6 for the wCDM model. constraining the EoS of dark energy.2). and thus they should account for most of in the self-similar outskirts rather than the emission from individual the gas clumping) via the method we previously employed (see sources. 2015).030 and to those from our whole sample analysis. and compare these results For a constant-w model. by leveraging on priors on the gas frac- C = C(R) (e. the EoS of the dark energy parameter w). covering a redshift range z = 0. to address the potential of future survey for dependency of fgas .24. density clumping might g analysis will also require dark energy constraints from the EM depend on cluster mass and/or redshift. This cosmological test. Observations and striking ramification of including high-z clusters in the cosmolog- simulations show that the gas clumping factor C =< n2e > / < ical analysis is to greatly improve the constraints on the EoS of ne >2 lies in the range ∼ 1. in combination with Planck+SNIa We repeat the cosmological analysis by considering only clus. used the predictions on C from the hydrodynamic simulations of However.993 ± 0. which corresponds to a modest dimming at z ∼ 0. The cosmological constraints sure from low-redshift cluster data with the cosmological-depend are borne out of the comparison of the scaled emission measure emission measure for high-z cluster (see Figure 1) produces a tight between nearby and higher-z clusters.010 ± 0. explicitly allowing for temperature and redshift the goal in mind e.g.6 from upcom- tion of cosmic filaments to clumping slightly increases with cluster ing surveys. Here we wish to investigate the ing the one for which the various profiles at different redshifts are impact of high-z clusters on the dark energy measurements. and determine whether dark energy is truly formed a quantitative comparison of the scaled emission measure the cosmological constant (w ≡ −1).014. w(z) we have Ωm = 0.311±0. since we assume (weak) self-similarity of the mea- evolution.123 ± 0. wa = −0. while for a time-evolving EoS of dark energy ters significantly increase the constraints on the EoS of dark energy. following their self-similar cal parameters.6) subsample.400. we have only a small improvement (∼ 15%) in term of statistical uncertainties on the matter density pa- In this Section we gauge the impact of inhomogeneities of the gas rameter with respect to the lower-z subsample. However. In this respect. Morandi & Cui 2014). we con- minimize any systematic effect from the theoretical predictions. which has been used. while only a slightly better accuracy is achieved on Morandi et al. at the same time. 2013. w0 = −0. when we include high-z clusters and if we con- and cosmic filaments sider a ΛCDM model. in particular. cidate the nature of the dark energy driving the accelerated expan- tained from the whole cluster sample. in §4 we per. The emission measure profiles are scaled according to the (weak) self- latter can be regarded as a baseline emission measure nearly inde. For the former test. the most distribution (’clumpiness’) and cosmic filaments.3% and < ∼ 0. improved by adding priors on the gas fraction evolution from hy- for the case where we include the external data sets (external data drodynamic simulations. This translates into a surface brightness Sx ∝ Ez3 (1 + sured EM profiles.056 − 1.g. 2015).g.3 The impact of inhomogeneities of the gas distribution Interestingly. By means of both these tests. As expected. Morandi & Cui Future large survey should target both high-z clusters as well 2014). the correct cosmology be- constraint on dark energy models. Significant improvement in sance parameter in our analysis. Morandi & Cui 2014). sion of the Universe. which will bias up- the matter density parameter. dark energy. respectively).6 − 1 (a (EM ∝ C < ne >2 ). clusters mass (Battaglia et al. matter density uncertainties related to the model of dark energy are thus negligi. By accurately measuring w.

the plane of the sky. This not only reduces systematics but also improves statistical model. which ence. Hence the value of the evidence naturally incorpo- model and physical parameters. given the data. the most powerful con- straints on current cosmological and astrophysical models of clus- ters (e. similarity of the stacked emission measure EMg (x). suppose we are interested in estimating inevitably depend on how one defines their boundary radius. and hence the measurements are prone to statistical and systematic er. model fitting. Moreover. P (Hi ). ing an era where complementary cosmological tests converge and Two levels of inference can often be distinguished in the pro- we are measuring the cosmological parameters more and more ac. Matthieu Roman.M. cess of data modelling. (A1) logical tests. We acknowledge support from Chandra grants GO2-13160A. is the subjective prior over our hypothesis acknowledges the hospitality of the Harvard-Smithsonian Center space. A. providing an independent way (and hinging on different assumptions) with respect to other cluster-based cosmo. clus. Thermodynamic integration ter. the qual- ity of the data being equal. gas fraction. This analysis is repeated for each rors.g. providing stringent constraints on the likelihood. which represents our state of knowledge (or prejudices) This makes this method extremely attractive. can be used to infer the Bayesian evidence (Gelman & Meng 1998). given the data. cluster growth (Vikhlinin et al. the larger number of parameters of the APPENDIX A: BAYESIAN INFERENCE model is significantly better at explaining the data (higher values of In this Section we review the statistical method used to infer both the likelihood). the prior and the evidence P (D|Hi ) (which is also often EoS of dark energy. mostly differ (see Figure 7). i. where different cluster boundaries definitions able for our analysis. priors on the namely when all the physical quantities are defined at a fixed over. GO2-13102A. constraints. accuracy by breaking degeneracies in the cosmological parameter The second level of inference is the task of model comparison. Jean-Baptiste P (D|Hi ) = P (θ|D. GO4-15115X and NASA grant NNX14AI29G. For any given model.4) from hydrody- especially crucial in analysis of sample of clusters spanning a wide namic simulations. rect. which improved have a small evidence (it will improbable) if it has large areas of the presentation of our work. cosmological parameters from complementary data sets (e. models were before the data arrived. based on observed quantities (the emission measure) on Bayes’ theorem then reads. called the marginalized likelihood). and are often summarized by the most probable parameter values. This From Equation A3. error bars on those parameters. rameters. scaling relations. our work provided for the first time constraints on which cosmological parameters due to our assumption of (weak) self- definition of cluster boundary radius is more tenable given the data. inhomogeneities of the gas distribution and cosmic filaments. selection ef- likelihood function. It is indeed crucially important that the dark energy con.e. So data of good quality and simpler models have larger evidence than data of bad quality and more complicated models. Hi )P (θ|Hi ) P (θ|D. it will more probable. Adam Mantz. Nabila Aghanim. 2012). Here we wish to compare the models in the light of the data. the presence of outlier measurements. the allowed parameter space with low values of the likelihood. mass function and scaled A1 The Bayesian theorem EM ) both hinge on the underlying assumption of self-similarity We start by reviewing briefly the basic principles of Bayesian infer- and arise from observations of how clusters evolve with time. Hi ) = . if more percomputer Authority. model will have a larger evidence. we infer what values its free parameters several independent techniques. To determine which model Hi is most plausible given the data. priors on the astrophysical parameter α (§2. i. Erwin Lau and Daisuke Nagai for valuable comments. we assume that a particular model is true. providing that they the data. we ACKNOWLEDGEMENTS rank Hi by evaluating the Bayesian evidence P (D|Hi ): Z Z We are indebted to Jimmy Irwin. the values of a set of parameters θ in some underlying model of We performed a validation of our findings. we see that the evidence is given by the work was made possible in part by a grant of high performance ’average’ of the likelihood function with respect to the prior.g. one may write down an expression are robust towards different assumptions and possible sources of for the likelihood P (D|θ. unless. This is SNe). P (D|θ. including biases due to the modelling of a particular set of values for the parameters θ. and assign some sort of preference or ranking to the alternatives. since observational signatures of should plausibly take. 2014). Hi )P (θ|Hi ) dθ (A3) Melin. Evidence 2009). P (D|Hi ) The cosmological constraints derived in this analysis are that is in remarkable agreement with the results obtained with SNIa (Suzuki et al. Thus. CMB data (Planck Collaboration 2015). or from the combined analysis of external data sets which gives the posterior distribution P (θ|D. For some data set D. Hi ) in terms of the (Planck Collaboration 2015). Hi ) dθ= P (D|θ. CMB. gratefully The second term. A computing resources and technical support from the Alabama Su. density with respect to the critical density of the Universe. Our analysis indeed hinges on the constraints on the have larger evidence than a more complicated theory. which expresses how plausible we thought the alternative for Astrophysics and of the NASA’s Goddard Space Flight Cen. Likelihood × Prior Posterior = (A2) ter gas fraction (Mantz et al. i. Maxim Markevich. At the first level of inference known as curately. while it will careful reading of the manuscript and suggestions. The results of this inference deviations in the EoS of dark energy from -1 are very small.18 Morandi & Sun Next. In addition to the the background. one may impose a prior P (θ|Hi ) on the pa- fects. in order to determine the EoS of rates the spirit of Ockham’s razor: a simpler theory will generally dark energy. 000–000 (0000) . We thank the anonymous referee for the of its allowed parameter space is likely.e. a Bayesian inference is particularly suit- range of redshift. Hi ) of obtaining the data vector D given systematics in our analysis. and we fit straints at this level of accuracy are obtained from combination of that model to the data. unless the MNRAS 000.e. since it is more di- regarding the values of the parameters before analyzing the data D. This is an additional sign that we are enter.

L. 2009. ARA&A. Compared to the classi. 389. G. To construct the Markov chain. atics. Sievers J. ApJ. Bond J. answers to question ’which model is more likely’ given the data. i. Pointe- couteau E. 2002... in Jacoby G. wz CDM etc.. Morris R. REFERENCES Bagging hinges on bootstrap with resampling. 539. order to obtain a new pseudo-data set D′ ... with severe ramification of outlier measurements (which may reflect system- outliers which might cause predictions to become inaccurate (i. ING) to reduce the prediction bias (or inaccuracy) due to outliers and/or departures from gaussianity in the measurements. H... 879 the bagged prediction. N. eds. Rapetti D. orginal data set D. 20 Going one step further.. Fabian model to each sample.g. . Piffaretti R. intrinsic scatter and measurement errors in the low signal-to- biased). parameters (§7. . and then averaging the predictions to get A. 383. then the full cycle is repeated in each iteration. When we have a sufficiently high number of time to seek convergence. Sievers J. Dn ) to produce M pseudo-data sets Astronomical Data Analysis Software and Systems V.... we choose candidate pa. 2008. we synthesize the idea underlying the resampling in CPU hours to infer the cluster likelihood functions on full parame- MCMC algorithms. MacKay 1997). 2012. in a data. with replacement. A22 technique.. Dolag K. 123 the M pseudo-data sets D′ .. ApJ. Smith 2013. Böhringer H. with initial values and a model H. A&A.. repeats steps 3 − 4 until the desired number of Monte Carlo rameter values θ = (θ1 . 1993). allowed parameter space by assuming a convenient distribution of Clearly. Grevesse N.II) we indicate the above two levels of in. L. unbi- ased) predictions (Clyde & Lee 2001). in order to achieve conver- A2 Model fitting gence to a stationary posterior distribution P (θ|D. Chon G.. The theoretical motivation is that the presence ter grids for a given cosmological model..... Ebeling H. 517. MNRAS. 2013. 1 with replacement. start the MCMC. A92+ or omitted for a particular pseudo data set. drawing samples from all the (MC) samples are obtained for each model H. regardless of the complexity of the same models). A&A. to evaluate Asplund M. 1. A&A. W. resample the orginal data set D using a bootstrap sample. 481 the best-fit model parameters one simply fit that model on each of Battaglia N.. 1996. while a Bayesian inference via a MCMC cycle. A&A. This required us ∼20. fitting the Allen S. H. R.. and then averages the results to get the Battaglia N. updating distribution. Barnes J. 777.g. The idea is to create a Markov chain with cal likelihood test.. A. 3. with some elements which may be repeated Arnaud M. Bagging can be implemented during MCMC to achieve smoother (i.... after Bayesian inference we have a tension between the need to improve each MCMC iteration (each full cycle through the parameters) in the quality of the fit and the need to have the simplest model to order to obtain M pseudo-data sets D′ . since it is basically an ’MCMC within iterations.. 568. structing a chain that has the desired distribution as its equilibrium 4..g.. J. A120 MNRAS 000... C. 2. Scott P. 2014. Croston J. the cluster emission mea. 2010. Hinshaw G. noise SX in the outskirts of individual clusters) in the cosmological Breiman (1996) introduced Bagging (Bootstrap AGGregat. via e. Rather. MC sam- ples are thus obtained for MCMC simulations and then merged to A3 Bagging in MCMC infer the desired (posterior) likelihood. 2003) after some ini- tioned step 4.. We refer to the literature for a spect to a MCMC framework. 47. A.g. Bagging naturally fits into the MCMC Betoule M. 2015.. on the Fisher-matrix approxi- detailed discussion about the acceptance probability of a new state mation.. 000–000 (0000) . where the better model which fits the data is the stationary distribution equal to the posterior distribution. updat- ing each from conditional distributions. of one or more observations (in our case... 101: original data D = (D1 . Thus at the end. which tial burn-in period required for the chain to reach equilibrium. data sets D′ . Sauval A.. Larson D. L. across many samples.... cycle through all of the parameters of the model H. W. estimates are taken as an average simultaneously So we can rewrite the Bayes theorem as: over the resampled data D′ and over the samples from the posterior distribution P (θ|D′ ). θ2 . the set of model parameters is updated sequentially tion ’which model fits best’ the data. ApJS.2). by re- one with larger likelihood (taking into account just the need to have sampling from the empirical distribution function of the observed better fits. 806. Pratt G. Ser. L. In Bagging. Weiland J. Bond J. H). we use the standard Bagging approach by means of to be updated (e. the Levenberg-Marquardt algorithm..).. H). in nique for sampling from a probability distribution based on con. The downside of the latter can thus determine expectation values and errors by considering approach is that it hinges on more restrictive assumptions with re- the percentiles of the distribution. the frequency of their occurrence tends to be the true a MC’. For each of the M pseudo- explain the data. cycle through all of the parameters of the model H. Neal tionary posterior distribution P (θ|D′ . Jarosik N.. The state of the chain after a number of steps is then each from conditional distributions.e. Odegard by the radial surface brightness Sx (θ) of the ith cluster. Bagging-MCMC simulations represent a tool to gauge the sure) farther from the true trend can influence the fit. the ith data set Di is represented Bennett C. ASP Conf. on the ference.e. Schmidt R.. 208. ΛCDM .. The bootstrap works by resampling the Arnaud K.II) (A4) MCMC algorithm incorporating Bagging can be summarized as: where with inference(I.e. Probing dark energy via galaxy cluster outskirts 19 latter significantly better describe the data. Pfrommer C. a likelihood test answers to ques. et al. seeking convergence to a sta- used as a sample of the desired distribution (Skilling 1998. by drawing n samples with equal probability and Arnaud et al.. The Markov Chain Monte Carlo (MCMC) method is a robust tech.. Prior × Likelihood = Posterior × Evidence For each model H (e. pp 17–+ (D′1 . this method may require a prohibitive computational this proposal density. several independent MCMC simulations (above steps 3 − 4) each performed on a different core of a large computing node. because MCMC predictions are themselves an average Böhringer H. A shortcut could be to perform a fit in the aforemen- probability distribution function (Gilks et al. θn ). R. 5. D′M ). 43 bootstrap estimate. We resample the original data set D.. We is computationally significantly faster. In other words. e.000 In this section. the final assumptions & measurements ⇒ inference(I. Pfrommer C. In our case. W.

716. et al.. L. Science... ARA&A..... B. 77. MNRAS. J.. V. ApJ. 122 Foster A. 2012. Couchman H. Bartelmann M. Verdugo T. Borgani S.. Nagai D. Dave R.. Voit G.. Jones C. Dolag K. MNRAS. S. Young O. A. 000–000 (0000) . Fabjan D.. Dolag K. 2012. 222. 2077 Morandi A. 508... 2012.. Rephaeli Y. Kravtsov A.... M. Donahue M. Killedar M. et al. 441.... P. 672... 729. et al. 2013. Limousin M.. Ford H... 440. Borgani S. 1118 Planck Collaboration 2015.. 1576 Skilling J.. 365.. ApJ.. 50. E. 450. Barkana R. 2014. et al. 2004.. Vikhlinin A. Lee H.. ApJ.. Granato G. L10+ Nagai D. Ettori S. 155 MacKay D. Jones C.. 14. Statistical Science. 116. 163 Gelman A.. Nagai D. A16 Planelles S... M. 2009. 2011. Dahle H. 2013.. M. Turner E.. 213 Clyde M.. 30. Balestra I. 1021 Ettori S. 2001. International Journal of Modern Physics D..... 650.. Van Speybroeck L... 692. Macciò A.. 2261 Nagai D. 791. K. 2011. 1521 Morandi A... J. 7.. 24.... Broadhurst T. ApJ. MNRAS. Hartmann D.... 693... 2009... 2006. 1998. 37 Morandi A. 2015. Lovisari L... Brickhouse N. T. 2014. ApJ. Markevitch M. Donahue M.. V. 2011b. Markov Chain Monte Carlo in Practice.. MNRAS. 2005. 128 Gelman A.... Probabilistic Data Analysis: An Introductory Guide Sun M. A&A. 2005. 773 Kravtsov A.. S. AJ. 438.. Broadhurst T. 1998.. R. 2003.. Pedersen K. H. 123 Wang L.. J.. Yoshida N. MNRAS.. 437.... Meng X. MNRAS. MNRAS. E. Statist. 372 Kaiser N. 691 Vikhlinin A. Physical Review Letters.. Terenziani F. 25 Rasia E. ApJ. 416. 2006. 1992. 177. Thomas P. 731.. 413. 1142 Suzuki N. Rubin D. Limousin M.. V. 195 Postman M. 2011a. A57 Eke V. 1 Neal R. Borgani S. Bradley L... Technical Report CRG-TR-93-1. Machine learning. 1986.. 2013. Steinhardt P. Murray S.. Rev. 321.. Pöppel W. 90.. 499 Chevallier M.... 2012.. Frenk C... Limousin M. Steinhardt P. 2001. ArXiv 1502.. 1123 Morandi A. A. MNRAS. 2012. Rosati P. 331...... 85 Vikhlinin A. 1995. Forman W. Artificial Intelligence and Statistics. Coe D. 045004 Sun M. 457 Gilks W.. Probabilistic inference using Markov chain Monte Carlo methods. 483 Caldwell R. Polarski D. 2567 Morandi A. Borgani S. ApJ. 1060 Voit G... Bajaja E.. 2001.. 806. E. New Journal of Physics. 2007. 61 Ettori S. 1993. Evrard A. 2007. 2001. 380.. Avestruz C. 1997. 775 Kass R. MNRAS. ApJ. et al. ApJ. L. 2011. D. 2014..... Chapman & Hall... Vikhlinin A... M. 440. Frenk C. Journal of the american statistical associa- tion.. R. 2009. 354. 668.. Pearce F. 501.. ApJ. K. Vikhlinin A. 746. et al. Ji L. 2008. Morras R. 569 Ettori S. 1998. Sun M. MNRAS.. 353 Kravtsov A. 2012.. et al. Meneghetti M... ARA&A. 68 Limousin M. Cui W... 1582 Carroll S.. V. L. B. Tozzi P. G.... ApJ. Sereno M. Richardson S.. G. 1992. 323 Kalberla P. 2014. Murante G. Space Sci. Ensemble learning for hidden Markov models Mantz et al. M. Colberg J. ApJ. 199. Arnal E. 2015. ApJ.. 503.. F. Vikhlinin A. ApJ... M. University of Toronto Plagge T. Ettori S...20 Morandi & Sun Breiman L. 2010. M. Burton W. Reviews of Modern Physics. Murante G. R.. ApJ. S. Short C.. Nagai D.-L. . 541. Cole S. Cui W. et al.. M. W. 1909 Morandi A.. MNRAS. 2006.. Press W. S. Morandi A. Sci. 1998. 691 80. A&A.. M. MNRAS... 3147 Morandi A.. E. ApJS. Ragone-Figueroa C. Navarro J. A&A. 1996. 571. 96 Riess A. MNRAS. Forman W.. 1998.01589.... A131 Planck Collaboration et al. 1998. T.... 207 MNRAS 000. 2014. A&A. MNRAS. Benitez N... 128 Lau E. et al. 13. 2014. A. Lau E.. Kravtsov A.. London Jenkins A.. 436. Nelson K. pp. 10... Forman W. Spiegelhalter D.. 640. A&A. Umetsu K. 1009 Simionescu A. Morandi A. 756. 421. 169 Dutton A.. 3359 Eckert D. E.. Raftery A. Smith R. Planck Collaboration et al. Jones C.. 111 Evrard A.. 550.. White S. 2012.