Planck 2013 results. XXI. Cosmology with the all-sky Planck Compton parameter y-map

Transcription

1 Astronomy & Astrophysics manuscript no. sz powerspectrum ESO 2013 March 20, 2013 Panck 2013 resuts. XXI. Cosmoogy with the a-sky Panck Compton parameter y-map Panck Coaboration: P. A. R. Ade 87, N. Aghanim 61, C. Armitage-Capan 93, M. Arnaud 74, M. Ashdown 71,6, F. Atrio-Barandea 20, J. Aumont 61, C. Baccigaupi 86, A. J. Banday 96,10, R. B. Barreiro 68, J. G. Bartett 1,69, E. Battaner 98, K. Benabed 62,95, A. Benoît 59, A. Benoit-Lévy 27,62,95, J.-P. Bernard 10, M. Bersanei 36,52, P. Bieewicz 96,10,86, J. Bobin 74, J. J. Bock 69,11, A. Bonadi 70, J. R. Bond 9, J. Borri 15,90, F. R. Bouchet 62,95, M. Bridges 71,6,65, M. Bucher 1, C. Burigana 51,34, R. C. Buter 51, J.-F. Cardoso 75,1,62, P. Carvaho 6, A. Cataano 76,73, A. Chainor 65,71,12, A. Chambau 74,17,61, L.-Y Chiang 64, H. C. Chiang 29,7, P. R. Christensen 82,39, S. Church 92, D. L. Cements 57, S. Coombi 62,95, L. P. L. Coombo 26,69, B. Comis 76, F. Couchot 72, A. Couais 73, B. P. Cri 69,83, A. Curto 6,68, F. Cuttaia 51, A. Da Siva 13, L. Danese 86, R. D. Davies 70, R. J. Davis 70, P. de Bernardis 35, A. de Rosa 51, G. de Zotti 48,86, J. Deabrouie 1, J.-M. Deouis 62,95, F.-X. Désert 55, C. Dickinson 70, J. M. Diego 68, K. Doag 97,79, H. Doe 61,60, S. Donzei 52, O. Doré 69,11, M. Douspis 61, X. Dupac 42, G. Efstathiou 65, T. A. Enßin 79, H. K. Eriksen 66, F. Finei 51,53, I. Fores-Cacho 10,96, O. Forni 96,10, M. Fraiis 50, E. Franceschi 51, S. Gaeotta 50, K. Ganga 1, R. T. Génova-Santos 67, M. Giard 96,10, G. Giardino 43, Y. Giraud-Héraud 1, J. Gonzáez-Nuevo 68,86, K. M. Górski 69,100, S. Gratton 71,65, A. Gregorio 37,50, A. Gruppuso 51, F. K. Hansen 66, D. Hanson 80,69,9, D. Harrison 65,71, S. Henrot-Versié 72, C. Hernández-Monteagudo 14,79, D. Herranz 68, S. R. Hidebrandt 11, E. Hivon 62,95, M. Hobson 6, W. A. Homes 69, A. Hornstrup 18, W. Hovest 79, K. M. Huffenberger 99, G. Hurier 61,76, T. R. Jaffe 96,10, A. H. Jaffe 57, W. C. Jones 29, M. Juvea 28, E. Keihänen 28, R. Keskitao 24,15, T. S. Kisner 78, R. Kneiss 41,8, J. Knoche 79, L. Knox 30, M. Kunz 19,61,3, H. Kurki-Suonio 28,46, F. Lacasa 61, G. Lagache 61, A. Lähteenmäki 2,46, J.-M. Lamarre 73, A. Lasenby 6,71, R. J. Laureijs 43, C. R. Lawrence 69, J. P. Leahy 70, R. Leonardi 42, J. León-Tavares 44,2, J. Lesgourgues 94,85, M. Liguori 33, P. B. Lije 66, M. Linden-Vørne 18, M. López-Caniego 68, P. M. Lubin 31, J. F. Macías-Pérez 76, B. Maffei 70, D. Maino 36,52, N. Mandoesi 51,5,34, A. Marcos-Cabaero 68, M. Maris 50, D. J. Marsha 74, P. G. Martin 9, E. Martínez-Gonzáez 68, S. Masi 35, S. Matarrese 33, F. Matthai 79, P. Mazzotta 38, A. Mechiorri 35,54, J.-B. Mein 17, L. Mendes 42, A. Mennea 36,52, M. Migiaccio 65,71, S. Mitra 56,69, M.-A. Mivie-Deschênes 61,9, A. Moneti 62, L. Montier 96,10, G. Morgante 51, D. Mortock 57, A. Moss 88, D. Munshi 87, P. Nasesky 82,39, F. Nati 35, P. Natoi 34,4,51, C. B. Netterfied 22, H. U. Nørgaard-Niesen 18, F. Novieo 70, D. Novikov 57, I. Novikov 82, S. Osborne 92, C. A. Oxborrow 18, F. Paci 86, L. Pagano 35,54, F. Pajot 61, D. Paoetti 51,53, B. Partridge 45, F. Pasian 50, G. Patanchon 1, O. Perdereau 72, L. Perotto 76, F. Perrotta 86, F. Piacentini 35, M. Piat 1, E. Pierpaoi 26, D. Pietrobon 69, S. Paszczynski 72, E. Pointecouteau 96,10, G. Poenta 4,49, N. Ponthieu 61,55, L. Popa 63, T. Poutanen 46,28,2, G. W. Pratt 74, G. Prézeau 11,69, S. Prunet 62,95, J.-L. Puget 61, J. P. Rachen 23,79, R. Reboo 67,16,40, M. Reinecke 79, M. Remazeies 61,1, C. Renaut 76, S. Ricciardi 51, T. Rier 79, I. Ristorcei 96,10, G. Rocha 69,11, C. Rosset 1, M. Rossetti 36,52, G. Roudier 1,73,69, J. A. Rubiño-Martín 67,40, B. Rushome 58, M. Sandri 51, D. Santos 76, G. Savini 84, D. Scott 25, M. D. Seiffert 69,11, E. P. S. Sheard 12, L. D. Spencer 87, J.-L. Starck 74, V. Stoyarov 6,71,91, R. Stompor 1, R. Sudiwaa 87, R. Sunyaev 79,89, F. Sureau 74, D. Sutton 65,71, A.-S. Suur-Uski 28,46, J.-F. Sygnet 62, J. A. Tauber 43, D. Tavagnacco 50,37, L. Terenzi 51, L. Toffoatti 21,68, M. Tomasi 52, M. Tristram 72, M. Tucci 19,72, J. Tuovinen 81, G. Umana 47, L. Vaenziano 51, J. Vaiviita 46,28,66, B. Van Tent 77, J. Varis 81, P. Vieva 68, F. Via 51, N. Vittorio 38, L. A. Wade 69, B. D. Wandet 62,95,32, S. D. M. White 79, D. Yvon 17, A. Zacchei 50, and A. Zonca 31 (Affiiations can be found after the references) Preprint onine version: March 20, 2013 ABSTRACT We have constructed the first a-sky map of the therma Sunyaev-Zedovich (tsz) effect by appying specificay taiored component separation agorithms to the 100 to 857 GHz frequency channe maps from the Panck survey. These maps show an obvious gaaxy custer tsz signa that is we matched with bindy detected custers in the Panck SZ cataogue. To characterize the signa in the tsz map we have computed its anguar power spectrum. At arge anguar scaes ( < 60), the major foreground contaminant is the diffuse therma dust emission. At sma anguar scaes ( > 500) the custered Cosmic Infrared Background (CIB) and residua point sources are the major contaminants. These foregrounds are carefuy modeed and subtracted. We measure the tsz power spectrum in anguar scaes, 0.17 θ 3.0, that were previousy unexpored. The measured tsz power spectrum is consistent with that expected from the Panck cataogue of SZ sources, with additiona cear evidence of signa from unresoved custers and, potentiay, diffuse warm baryons. We use the tsz power spectrum to obtain the foowing cosmoogica constraints: σ 8 (Ω m /0.28) 3.2/8.1 = 0.784±0.016 (68 % C.L.). Marginaized band-powers of the Panck tsz power spectrum and the best-fit mode are given. The non-gaussianity of the Compton parameter map is further characterized by computing its 1D probabiity distribution function and its bispectrum. These are used to pace additiona independent constraints on σ 8. Key words. cosmoogica parameters arge-scae structure of Universe Gaaxies: custers: genera Corresponding author: J. F. Macías-Pérez, macias@psc. in2p3.fr 1

2 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map 1. Introduction This paper, one of a set associated with the 2013 reease of data from the Panck 1 mission (Panck Coaboration I 2013), describes the construction of a Compton parameter map and its anguar power spectrum and high order statistics. The therma Sunyaev-Zedovich (tsz) effect (Sunyaev & Zedovich 1972), produced by the inverse Compton scattering of CMB photons by hot eectrons aong the ine of sight, has proved to be a major too to the study of the physics of custers of gaaxies as we as structure formation in the Universe. In particuar, tsz-seected cataogues of custers of gaaxies have been provided by various experiments incuding the Panck sateite (Panck Coaboration VIII 2011; Panck Coaboration XXIX 2013), the Atacama Cosmoogy Teescope (ACT, Hassefied et a. 2013) and the South Poe Teescope (SPT, Reichardt et a. 2013). From these cataogues, and their associated sky surveys, a weath of studies have been performed both on the physics of custers of gaaxies (Panck Coaboration XII 2011; Panck Coaboration XI 2011; Panck Coaboration X 2011) and on their cosmoogica interpretation (Panck Coaboration XX 2013; Benson et a. 2013; Das et a. 2013; Wison et a. 2012; Mak & Pierpaoi 2012). The study of number counts and their evoution with redshift using tsz detected custers of gaaxies has been recognized as an important cosmoogica test (Carstrom et a. 2002; Dunkey et a. 2013; Benson et a. 2013; Panck Coaboration XX 2013). As a compement, the measurement of the tsz effect power spectrum has been proposed by Komatsu & Sejak (2002). One advantage of using the tsz anguar power spectrum over custer counts is that no expicit measurement of custer masses is required. However, drawbacks of using the tsz anguar power spectrum incude potentia contamination from point sources (Rubiño-Martín & Sunyaev 2003; Taburet et a. 2010) and other foregrounds. Aso, ower mass, and therefore fainter, custers, that may not be significanty detected as individua objects contribute to this statistica signa (Battagia et a. 2010; Shaw et a. 2010). To date, indirect measurements of the tsz power spectrum are ony avaiabe from high resoution CMB oriented experiments ike ACT (Sievers et a. 2013) and SPT (Reichardt et a. 2012). In these studies, constraints on the ampitude of tsz power spectrum at = 3000 are obtained by fitting a tsz tempate in addition to other components (i.e., CMB, radio and infrared point-source and custered Cosmic Infrared Background (CIB)) to the measured tota power spectrum. These constraints are obtained at anguar scaes where the tsz signa dominates over the CMB, but at these same scaes the contamination from point sources and custered CIB is important and may affect the measured tsz signa. Moreover, the scaes probed are particuary sensitive to the uncertainties in modeing the intra-custer medium (ICM) over a broad range of masses and redshifts, and at arge custer-centric radii (Battagia et a. 2010). Recent work, using hydrodynamica simuations (Battagia et a. 2010; Battagia et a. 2012) N-body simuations pus semi-anaytic gas modes (Trac et a. 2011) and anaytic modes (Shaw et a. 2010), have significanty reduced the tension between the observed and predicted vaues. However the distribution of ampitudes 1 Panck ( is a project of the European Space Agency (ESA) with instruments provided by two scientific consortia funded by ESA member states (in particuar the ead countries France and Itay), with contributions from NASA (USA) and teescope refectors provided by a coaboration between ESA and a scientific consortium ed and funded by Denmark. Tabe 1. tsz Compton parameter y conversion factors to CMB temperature units and the FWHM of the beam of the Panck channe maps. Frequency T CMB g(ν) FWHM [GHz] [K CMB ] [ ] between different modes and simuations is sti significanty arger than the measurement errors, degrading the constraints that can be paced on cosmoogica parameters with these methods (Dunkey et a. 2013; Reichardt et a. 2013). In addition to the power spectrum, and as pointed out in (Rubiño-Martín & Sunyaev 2003), the skewness (or equivaenty, the bispectrum) of the tsz signa is a powerfu and independent too to study and to isoate the signa of custers, separating it from the contribution of radio and IR sources. Recenty, Bhattacharya et a. (2012) showed that the bispectrum of the tsz effect signa is dominated by massive custers at intermediate redshift for which high-precision X-ray observations exist. This contrasts with the power spectrum, where the signa mainy comes from the ower mass and higher redshift groups and custers (e.g., Trac et a. 2011). The theoretica uncertainty in the tsz bispectrum is thus expected to be significanty smaer than that of the SZ power spectrum. Combined measurements of the power spectrum and the bispectrum can thus be used to distinguish the contribution to the power spectrum from different custer masses and redshift ranges. The bispectrum ampitude scaes as σ (Bhattacharya et a. 2012). Simiary, Wison et a. (2012) used the unnormaized skewness of the tsz fuctuations, T 3 (n), which scaes as σ , to obtain an independent determination of σ 8. Thanks to its a sky coverage and unprecedented wide frequency range, Panck has the unique capabiity to produce an a-sky Compton parameter (y) map and an accurate measurement of the tsz power spectrum at intermediate and arge anguar scaes, for which the tsz fuctuations are amost insensitive to the custer core physics. The Panck Compton parameter map aso offers the possibiity to study the properties of the non-gaussianity of the tsz signa using higher order statistica estimators as the skewness and the bispectrum. In this paper we derive such a tsz a-sky map from the individua Panck frequency maps and compute its power spectrum, its 1D probabiity density function (1D PDF), and the associated bispectrum. The paper is structured as foows. Sect. 2 describes the Panck data used to compute the tsz a-sky map and the simuations used to characterize it. We discuss detais of the modeing of the tsz effect power spectrum and bispectrum in Sect. 3. In Sect. 4 we present the Panck a-sky Compton parameter map. Sect. 5 describes the power spectrum anaysis. Cross-checks using high-order statistics are presented in Sect. 6. Cosmoogica interpretation of the resuts is discussed in Sect. 7 and we present our concusions in Sect. 8. 2

3 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map 2. Data and simuations 2.1. The Panck Data This paper is based on the first 15.5 month survey mission corresponding to two fu-sky surveys. We refer to Panck Coaboration II (2013) and Panck Coaboration VI (2013) for the generic scheme of Time Ordered Information (TOI) processing and map-making, as we as for the technica characteristics of the Panck frequency maps. The Panck channe maps are provided in HEALPix (Górski et a. 2005) N side = 2048 at fu resoution. An error map is associated with each channe map and is obtained from the difference of the first haf and second haf of the survey rings for a given pointing position of the sateite spin axis. The resuting maps, caed nu maps, are mainy free from astrophysica emission and they are a good representation of the statistica instrumenta noise. Nu maps have aso been used to estimate the noise in the fina Compton parameter maps. Here we approximate the Panck beams by effective circuar Gaussians (Panck Coaboration IV 2013; Panck Coaboration VII 2013) The FWHM for each frequency channe are given in Tabe 1. Athough tests have been performed using both LFI and HFI channe maps, the work presented here wi refer mainy to resuts using the HFI data ony FFP6 Simuations We aso use simuated Panck frequency maps obtained from the so-caed Fu Foca Pane (FFP6) simuations (described in Panck Coaboration ES 2013)). These simuations incude the most reevant sky components at microwave and miimeter frequencies: CMB; therma SZ effect; diffuse Gaactic emissions (synchrotron, free-free, therma and spinning dust); radio and infrared point sources, and the custered Cosmic Infrared Background (CIB) based on foreground modes from the Panck Sky Mode Deabrouie et a. (PSM, 2012). The simuated tsz signa was constructed using hydrodynamica simuations of custers of gaaxies up to redshift 0.3, competed with pressure profie-based simuations of individua custers of gaaxies randomy drawn on the sky. The noise in the maps was obtained from Gaussian random reaizations of noise in the time domain and therefore accounts for noise inhomogeneities in the maps. 3. Modeing the tsz effect The therma SZ Compton parameter in a given direction, n, is y(n) = n e K B T e m e c 2 σ T ds (1) where ds is the distance aong the ine-of-sight, n, and n e and T e are the eectron number density and temperature, respectivey. In units of CMB temperature the contribution of the tsz effect to the Panck maps for a given observation frequency ν is T T CMB = g(ν) y. (2) Negecting reativistic corrections we have g(ν) = ( x coth( x 2 ) 4), with x = hν/(k B T CMB ). Tabe 1 shows the Compton parameter to CMB temperature, K CMB, conversion factors for each frequency channe after integrating within the bandwidth tsz power spectrum Decomposing the map in spherica harmonics we obtain y(n) = y m Y m (n), (3) m Thus, the anguar power spectrum of the Compton parameter map is C tsz 1 = y m y m (4) m Note that C tsz is an a-dimensiona quantity as y. To mode the tsz power spectrum we consider a two hao mode to account for intra-hao and inter-hao correations C SZ = C 1hao + C 2haos. (5) The 1 hao term, aso known as the Poissonian contribution, can be computed by summing the square of the Fourier transform of the projected SZ profie, weighted by the number density of custers of a given mass and redshift (Komatsu & Sejak 2002): C 1hao = zmax 0 dz dv c dzdω Mmax M min dm dn(m, z) dm y (M, z) 2, (6) where dv c /(dzdω) is the comoving voume per unit redshift and soid ange and n(m, z)dm dv c /(dzdω) is the probabiity of having a gaaxy custer of mass M at a redshift z in the direction dω. The quantity ỹ = ỹ (M, z) is the 2D Fourier transform on the sphere of the 3D radia profie of the Compton y-parameter of individua custers, ỹ (M, z) = 4πr s 2 s ( σt m e c 2 ) 0 dx x 2 P e (M, z, x) sin( x/ s ) x / s (7) where x = r/r s, s = D A (z)/r s, r s is the scae radius of the 3D pressure profie, D A (z) is the anguar diameter distance to redshift z and P e is the eectron pressure profie. The two-hao term is obtained by computing the correation between two different haos (Komatsu & Kitayama 1999; Diego & Majumdar 2004; Taburet et a. 2011) C 2haos = zmax 0 ( Mmax dz dv c dzdω M min dm ) 2 dn(m, z) dm ỹ (M, z) B(M, z) P(k, z), (8) where P(k, z) is the 3D matter power spectrum at redshift z. Here B(M, z) is the time-dependent inear bias factor that reates the matter power spectrum, P(k, z), to the power spectrum of the custer correation function. Foowing Komatsu & Kitayama (1999, see aso Mo & White 1996) we adopt B(M, z) = 1+(ν 2 (M, z) 1)/δ c (z), where ν(m, z) = δ c (M)/D(z)σ(M), σ(m) is the present-day rms mass fuctuation, D(z) is the inear growth factor, and δ c (z) is the threshod over-density of spherica coapse. Finay, we compute the tsz power spectrum using the Tinker et a. (2008) mass function dn(m, z)/dm incuding an observed-to-true mass bias of 20%, as discussed in detai in Panck Coaboration XX (2013), and we mode the SZ Compton parameter using the pressure profie of Arnaud et a. (2010). This approach is consistent with the ingredients of the custer number count anaysis in Panck Coaboration XX (2013). 3

4 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map 3.2. Nth moment of the tsz fied To cacuate the Nth moment of the tsz fied, we assume to first order that the distribution of custers on the sky can be adequatey described by a Poisson distribution corresponding to the one hao term. We negect the contribution due to custering between custers and their overap (Komatsu & Kitayama 1999). The Nth moment is then given by (Wison et a. 2012) zmax 0 dz dv c dzdω Mmax M min dm dn(m, z) dm d 2 θ y(θ, M, z) N, (9) where y(θ, M, z) is the integrated Compton parameter aong the ine-of-sight for a custer of mass M at redshift z Bispectrum The anguar bispectrum, anaogous to the 3-point correation function in harmonic space, is the owest order indicator of the non-gaussianity of a fied. It is given by B m 1m 2 m = y 1 m 1 y 2 m 2 y 3 m 3, (10) where the ange-averaged quantity in the fu-sky imit can be written as ( ) 1 B( ) = 2 3 B m 1 m 2 m m 1m 2 m , (11) m 1 m 2 m 3 which has to satisfy the conditions: m 1 +m 2 +m 3 = 0; = even; and i j k i + j for the Wigner 3 j function in brackets. For iustration we compute the bispectrum assuming a Poissonian distribution, given by (Bhattacharya et a. 2012): zmax 0 B( ) (21 + 1)( )( ) 4π ( dz dv Mmax c dn(m, z) dm dzdω M min dm ỹ 1 (M, z)ỹ 2 (M, z)ỹ 3 (M, z). (12) 4. The reconstructed a-sky tsz map 4.1. Reconstruction methods The contribution of the tsz effect in the Panck frequency maps is subdominant with respect to the CMB and other foreground emissions. Furthermore, the tsz effect from gaaxy custers is spatiay ocaized and eads to a highy non-gaussian signa with respect to that from the CMB. CMB-oriented componentseparation methods (Panck Coaboration XII 2013) are not optimized to recover the tsz signa. We therefore need to use specificay taiored component separation agorithms that are abe to reconstruct the tsz signa from the Panck frequency channe maps. These optimized a-sky component separation techniques rey on the spatia ocaization of the different astrophysica components and on their spectra diversity to separate them. We present,in the foowing, the resuts of two agorithms, MILCA (Modified Interna Linear Combination Agorithm, Hurier et a. 2010) and NILC (Needet Iindependent Linear Combination, Remazeies et a. 2011). Both are based on the we known Interna Linear Combination (ILC) approach that searches for the inear combination of the input maps that minimizes the variance of the fina reconstructed map under the constraint of offering unit gain to the component of interest (here the tsz effect, whose frequency dependence is known). Both have been extensivey tested on simuated Panck data. ) MILCA MILCA (Hurier et a. 2010) uses two constraints: preservation of the tsz signa, assuming the tsz spectra signature; and remova of the CMB contamination in the fina SZ map, making use of the we known spectrum of the CMB. In addition, to compute the weights of the inear combination, we have used the extra degrees of freedom in the inear system to minimize residuas from other components (two degrees) and from the noise (two degrees). The noise covariance matrix was estimated from the nu maps described in Section 2.1. To improve the efficiency of the MILCA agorithm, weights are aowed to vary as a function of mutipoe, and are computed independenty on different sky regions. We have used 11 fiters in space with an overa transmission of one, except for < 8. For these arge anguar scaes we have used a Gaussian fiter to reduce foreground contamination. The size of the independent sky regions was adapted to the mutipoe range to ensure sufficient spatia ocaization at the required resoution. We used a minimum of 12 regions at ow resoution and a maximum of 3072 regions at high resoution NILC In the muti-component extensions of NILC Deabrouie et a. (2009); Remazeies et a. (2011), initiay deveoped to extract the CMB, the weights for component separation (i.e., covariances) are computed independenty in domains of a needet decomposition (in the spherica waveet frame). The needet decomposition provides ocaization of the ILC fiters both in pixe and in mutipoe space, aowing us to dea with oca contamination conditions varying both in space and in scae. We imposed constraints to remove the CMB contamination and preserve the tsz effect. To avoid strong foreground contamination, the Gaactic pane was masked before appying NILC to the Panck frequency maps. In both methods, we mask the brightest regions in the Panck 857 GHz channe map, corresponding to about 33 % of the sky. We use the HFI channe maps from 100 to 857 GHz that are convoved to a common resoution of 10. The 857GHz channe map is mainy expoited in the interna inear combination as a tempate to remove the therma dust emission on arge anguar scaes. However it induces significant CIB residuas in the tsz map on sma scaes. To avoid this contamination, whie enabing efficient remova of the diffuse therma dust emission at arge anguar scaes, we use the 857GHz channe ony for < Reconstructed Compton parameter y map Figure 1 shows the reconstructed Panck a-sky Compton parameter map for NILC (top pane) and MILCA (bottom pane). For dispay purposes, the maps are fitered using the procedure described in Sect. 6. Custers appear as positive sources: the Coma custer and Virgo supercuster are ceary visibe near the north Gaactic poe. As mentioned above, the Gaactic pane is masked in both maps, eaving 67% of sky. Other weaker and more compact custers are visibe in the zoomed region of the Southern cap shown in the bottom pane of Fig. 2. Strong Gaactic and extragaactic radio sources show up as negative bright spots on the maps and were masked prior to any scientific anaysis, as discussed beow in Sect Residua Gaactic contamination is aso visibe around the edges of the masked area; extra masking was performed to avoid this highy contaminated area. The difference of contrast observed between the NILC and MILCA maps comes both from differences in the noise and in- 4

5 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map Fig. 1. Reconstructed Panck a-sky Compton parameter maps for NILC (top) and MILCA (bottom) in orthographic projections. The difference of contrast observed between the NILC and MILCA maps comes both from differences in the noise and instrumenta systematic contribution and from the differences in the fitering appied for dispay purpose to the origina Compton parameter maps. 5

6 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map Fig. 2. Zooms into the reconstructed Panck a-sky Compton parameter maps for NILC (eft) and MILCA (right) at intermediate Gaactic atitudes in the southern sky. strumenta systematic contribution (the NILC map is sighty noisier but ess affected by systematic effect and foreground emissions than the MILCA map, as discussed in Sect. 5.2) and from the differences in the fitering appied for dispay purpose to the origina Compton parameter maps, as discussed in Sect In addition to the fu Compton parameter maps, we aso produce the so-caed FIRST and LAST Compton parameter maps from the first and second haves of the survey rings (i.e., pointing periods). These maps are used for the power spectrum anaysis in Sect Point source contamination and masking Point source contamination is an important issue for the cosmoogica interpretation of the Panck Compton parameter map. Radio sources wi show up in the reconstructed tsz maps as negative peaks, whie infrared sources wi show up as positive peaks, mimicking the custer signa. To avoid contamination from these sources we introduce a point source mask (PSMASK, hereafter). This mask is the union of the individua frequency point-source masks discussed in Panck Coaboration XXVIII (2013). To test the reiabiity of this mask we have performed a search for negative sources in the Compton parameter maps using the MHW2 agorithm (López-Caniego et a. 2006). We have found that a resoved radio sources in the Compton parameter maps are masked by the PSMASK. For infrared sources, estimating the efficiency of the masking is hampered by the tsz signa itsef. The residua contamination from point sources is discussed in Sect. 5.2 and Sect. 6. It is aso important to note that the PSMASK may aso excude some custers of gaaxies. This is particuary true in the case of custers with strong centra radio sources, such as the Perseus custer (see Panck Coaboration XXIX 2013, for detaied discussion) tsz signa from resoved sources As a very first vaidation step of the Compton parameter maps we perform a bind search of the SZ signa from resoved sources and compare it to the Panck cataogue of SZ sources (Panck Coaboration XXIX 2013). The atter comprises 861 confirmed custers out of 1227 custer candidates and 54 CLASS1 highy reiabe candidate custers Yieds Two ists of SZ sources above a signa-to-noise threshod of 4.5 are constructed from both MILCA and NILC a-sky Compton parameter maps outside a 33% Gaactic mask. The point source detections are undertaken using two methods: SMATCH, in which sources are detected using the SEXtractor agorithm (Bertin & Arnouts 1996) over the whoe sky divided into 504 patches. A singe frequency matched fiter (Mein et a. 2006a) is then appied to measure the SZ fux density and signa-to-noise using the Arnaud et a. (2010) pressure profie. Using this method, we detect 843 and 872 sources in MILCA and NILC, respectivey. MHWS, in which SZ sources are detected in the maps using IFCAMEX (MHW2, Gonzáez-Nuevo et a. 2006; López- Caniego et a. 2006). The fux density and signa-to-noise are then estimated using SEXtractor on patches. We detect 1036 and 1740 sources in MILCA and NILC, respectivey, with this method. The difference between the yieds of the two methods is understandabe, as SMATCH is by construction dedicated to the search for SZ sources and the precise measurement of their fux (incuding assumptions on the spatia distribution of the SZ signa), whereas MHWS targets a types of compact source (incuding IR and radio sources) and uses a more generic fux estimation procedure. 6

7 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map Y5R500,y MAP [arcmin 2 ] Y5R500,NILC [arcmin 2 ] Y 5R500,PSZ [arcmin 2 ] Y 5R500,MILCA [arcmin 2 ] Fig. 3. Comparison of the measurement of Y 5R500. Left: between the vaues derived from the detection methods used to buid the Panck cataogue of custers (Y 5R500,PSZ ), and those from the a-sy reconstructed MILCA tsz map (Y 5R500,y MAP ). Right: between the MILCA (Y 5R500,MILCA ) and NILC (Y 5R500,NILC ) a-sky tsz effect maps. The equaity reationship is marked as a dashed back ine. We have compared these two ists of sources with 790 confirmed custers and CLASS1 high reiabiity candidates from the Panck cataogue of SZ sources that fa outside the 33% Gaactic mask. The association is performed on the basis of the source positions within a search radius of 10 (the resoution of the SZ a-sky maps). We found 583 and 529 matches in the MILCA source ist with the SMATCH and MHWS methods, respectivey (614 and 414 from the NILC source ist). This match of about 52 to 77% per cent, respectivey, is satisfactory. Indeed, as shown in Mein et a. (2012), indirect detection methods based on reconstructed y-maps are ess efficient at extracting custers of gaaxies than dedicated direct methods such as those used to buid the Panck cataogue of SZ sources (i.e., MMF1, MMF3 and PwS Herranz et a. 2002; Mein et a. 2006b; Carvaho et a. 2011; Panck Coaboration XXIX 2013) Photometry Of more importance than a comparison of yieds is the comparison in terms of photometry. For a-sky map detections that are associated with custers in the Panck SZ cataogue, the SZ fux measurement from the a-sky maps correates very we with the maximum ikeihood vaue of the integrated Compton parameter, Y 5R500 2, provided by the dedicated SZ-detection methods in the Panck SZ cataogue. As shown in the top pane of Fig. 3, the correation is very tight, with itte dispersion (0.1 dex). We note that the few points at high Y 5R500 that ie significanty above the one-to-one ine are expected; they correspond to nearby and extended custers. On the one hand, the significance of SZ fux measurement increases with the fux. On the other hand, the cataogue detection methods are not optimized for the extraction of such extended sources (see Panck Coaboration XXIX 2013, for detais), therefore they tend to miss part of the SZ fux, which 2 R 500 refers to the radius inside which the mean density is 500 times the critica density at the custer redshift. is recovered together with a better estimate of the custer size from the Compton parameter map directy. As a sanity check, we have aso matched the ist of sources detected by a given method on both MILCA and NILC maps in order to compare the SZ photometry. The bottom pane of Fig. 3 shows very good agreement between the methods. There is ony 0.07 and 0.01 dex dispersion between them for the SMATCH and MHWS extraction methods, respectivey. Together, these resuts indicate that we can be confident in the fideity with which the tsz signa is reconstructed over the whoe sky by the MILCA and NILC methods. 5. Anguar power spectrum of the reconstructed y-map 5.1. Methodoogy To estimate the power spectrum of the tsz signa we use the XSPECT method (Tristram et a. 2005) initiay deveoped for the cross-correation of independent detector maps. XSPECT uses standard MASTER-ike techniques (Hivon et a. 2002) to correct for the beam convoution and the pixeization, as we as the mode-couping induced by masking foreground contaminated sky regions. We appy XSPECT to the FIRST and LAST y-maps obtained using NILC and MILCA. We consider the foowing map pairs: the MILCA FIRST and LAST (MILCA F/L); the NILC FIRST and LAST (NILC F/L); and the NILC FIRST and MILCA LAST (NILC-MILCA F/L), or equivaenty the MILCA FIRST and NILC LAST (MILCA-NILC F/L). As the noise is decorreated between the map pairs the resuting power spectrum is not biased and we preserve the variance. In the foowing, a the spectra wi use a common mutipoe binning scheme that was defined in order to minimize the correation between adjacent bins at ow mutipoes and to increase the signa-to-noise at high mutipoe vaues. Error bars 7

8 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map ( + 1)C/2π tsz dust CIB PS NILC F/L NILC-MILCA F/L 10 1 Fig. 4. Anguar power spectrum of the main foreground contributions as estimated using the FFP6 simuations. We pot the diffuse Gaactic emission (bue), custered CIB (green) and point source (cyan) contributions, as we as the tsz signa (red). The soid and dotted ines correspond the NILC F/L and to the NILC- MILCA F/L cross-power spectra, respectivey. For iustration we aso show the Panck instrumenta noise power spectrum (dashed back ine) in the MILCA Compton parameter map ( + 1)C/2π Fig. 5. Anguar cross-power spectra of the Panck NILC F/L reconstructed Compton parameter maps for different Gaactic masks, removing 30 (cyan), 40 (back points and error bars), 50 (red), 60 (green) and 70 (bue) % of the sky. in the spectrum are computed anayticay from the auto-power and cross-power spectra of the pairs of maps, as described in Tristram et a. (2005). A of our Compton parameter maps assume a circuar Gaussian beam of 10 FWHM. The additiona fitering at arge anguar scaes in the MILCA Compton parameter maps is aso accounted for and deconvoved Foreground contamination The chaenge in computing the tsz power spectrum is to estimate and minimize foreground contamination. We do not intend here to provide a detaied foreground anaysis, but rather to identify the main foreground contaminants at different mutipoes. We first identify the dominant foregrounds in the reconstructed Compton parameter maps. To do so, we appy to the FFP6 simuated maps the inear combination weights of NILC and MILCA derived from the rea data. In this way we have constructed maps of the expected foreground contamination in the fina Compton parameter maps. Figure 4 shows the anguar power spectra for these reconstructed foreground contamination maps. The PSMASK and a conservative common Gaactic mask that eaves 50% of the sky are used. The Gaactic mask is constructed by masking the 50% brightest regions of the sky in the 857 GHz intensity map, as detaied beow in Sect We show the diffuse Gaactic contamination (bue), the custered Cosmic Infrared Background contamination (green), and point source contamination (cyan). We consider here the foreground contamination in the cross-power spectra of the NILC F/L (dotted ines) and NILC-MILCA F/L maps (soid ines). The tsz power spectrum for the FFP6 simuations is potted in red. For iustration we aso show the Panck instrumenta noise power spectrum (dashed back ine) in the MILCA Compton parameter map. We ceary observe that, as expected, the diffuse Gaactic emission, mainy therma dust, dominates the foreground contribution at ow mutipoes. For arge mutipoes the custered CIB and point source contributions dominate the power spectrum. It is important to notice that the tsz signa dominates the anguar power spectrum in the mutipe range 100 < < 800. We aso note that foreground contamination differs depending on the reconstruction method. We observe that MILCA is more affected by foreground contamination. However, we find that at arge anguar scaes the diffuse Gaactic dust contamination is significanty ower in the NILC- MILCA F/L cross-power spectrum than in the NILC F/L crosspower spectrum. This indicates that the residua dust contamination is not 100 % correated between the reconstructed MILCA and NILC Compton parameter maps. In contrast, the custered CIB and point source contamination eves are simiar for the two cross-power spectra at high mutipoes, indicating that the residua contamination is 100% correated between the MILCA and NILC maps Low-mutipoe contribution The diffuse Gaactic foreground contribution can be significanty reduced by choosing a more aggressive Gaactic mask. Assuming that at arge anguar scaes the Compton parameter maps are mainy affected by diffuse Gaactic dust emission, we have tested severa Gaactic masks by imposing fux cuts on the Panck 857 GHz channe intensity map. In particuar we considered masking out 30, 40, 50, 60 and 70% of the sky. The edges of these masks have been apodized to imit ringing effects on the reconstruction of the anguar power spectrum. Figure 5 presents the anguar cross-power spectrum of the reconstructed NILC F/L Compton parameter maps for some of these Gaactic masks: 30 (cyan); 40 (back); 50 (red); 60 (green); 70 (bue)%; and the PSMASK. We find that when masking 40% or more of the sky the tsz anguar power spectrum does not change significanty. That is why, conservativey, we seect the 50% mask (GALMASK50 hereafter) that wi be used in the remainder of this anaysis. We checked if the foreground contribution in the reconstructed Panck Compton parameter maps aso depends on the reconstruction method. From the anaysis of the FFP6 simuations we have found that the contribution from foregrounds in the NILC and MILCA Compton parameter maps is not the same, and it is not fuy correated. Simiar resuts are found for the Panck data. Figure 6 shows the cross-power spectra between 8

9 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map ( + 1)C/2π Fig. 6. Anguar cross-power spectra between the reconstructed Panck MILCA F/L (back), the NILC F/L (red) and the NILC- MILCA F/L (bue) maps ( + 1)C/2π Fig. 7. NILC F/L cross power spectrum before (back points) and after (red points) foreground correction, compared to the power spectra of the physicay motivated foreground modes. Specificay we show: custered CIB (green ine); infrared sources (cyan ine); and radio sources (bue ine). The statistica (thick ine) and tota (statistica pus foreground, thin ine), uncertainties are aso shown. We aso show the best-fit tsz power spectrum mode presented in Sect.7.1 as a soid red ine. the MILCA F/L maps (back) 3, the NILC F/L maps (red) and the NILC- MILCA F/L maps (bue), as a function of. We observe that the MILCA F/L cross-power spectrum presents a arger ampitude than the NILC F/L cross-power spectrum. This is most probaby due to a arger foreground contamination in the MILCA Compton parameter map. In addition, we find that the NILC-MILCA F/L 4 cross-power spectrum shows the owest ampitude at ow mutipoes ( < 100). This is due to a reduction of the dust contamination in the cross-correation of the NILC and MILCA Compton parameter maps with respect to the dust contamination in the origina maps. We aso find that the NILC-MILCA F/L ies between the MILCA F/L and NILC F-L cross-power spectra at high mutipoes. This can be expained by the differences in the custered CIB contamination in the MILCA and NILC Compton parameter maps. An 3 The excess of power at ow observed in the MILCA F/L maps anguar cross-power spectrum is due to the deconvoution from the extra ow-mutipoe fitering in the MILCA maps, discussed in Sect And equivaenty MILCA-NILC F/L that is not shown in the Figure. accurate mode of the custered CIB power spectrum is avaiabe. However, this is not the case for the dust contamination power spectrum, thus we restrict the power spectrum anaysis presented in Sect. 7.1 to > 60. Hereafter, we wi consider the NILC F/L cross-power spectrum as a baseine for cosmoogica anaysis and the NILC- MILCA F/L cross-power spectrum wi be used to cross-check the resuts High-mutipoe contribution The high- contamination from custered CIB and point sources affects the measurement of the tsz spectrum and its cosmoogica interpretation. Reaistic modes fitted to the Panck data are thus needed. We take advantage of the capabiity of the Panck survey to measure and constrain these foreground emissions and use the outputs of Panck Coaboration XVIII (2011) and Panck Coaboration (2013) for the custered CIB modeing. For the six Panck HFI frequencies considered in this paper, the custered CIB mode consists of six auto-power spectra and 24 crosspower spectra. For frequencies above 217 GHz, these spectra are fitted in Panck Coaboration (2013) to the measured CIB consistenty with Panck Coaboration XVIII (2011). The mode is extrapoated at 100 and 143 GHz foowing Béthermin et a. (2012) and Panck Coaboration XVIII (2011). The uncertainties in the custered-cib mode are mainy due to the crosscorreation coefficients that reate the cross-power spectra to the auto-power spectra. Foowing Panck Coaboration (2013) we consider 5% goba uncertainties on those coefficients. We use the Béthermin et a. (2012) mode to compute the star-forming dusty gaaxy contribution. Finay, we use the the Tucci et a. (2011) mode, fitted to the Panck ERCSC (Panck Coaboration Int. VII 2013), for extragaactic radio sources. Notice that these modes are aso used for the study of the custered CIB with Panck (Panck Coaboration 2013). We now estimate the residua power spectrum in the y-map after component separation. We appy the MILCA or NILC weights to Gaussian-reaization maps drawn using the crossand auto-spectra of each component at the six Panck HFI frequencies. The residua power spectrum in the y-map can be aso estimated in the spherica harmonic domain, as detaied in Appendix A. We have tested the consistency between the two approaches and we give here resuts for a map-based estimate using a tota of 50 a-sky simuations for each of the foreground components. Specific simuations, varying the foreground modes, were aso performed to propagate the 5% goba uncertainties of the mode-coefficients into the estimated residua power spectrum. We find a 50% uncertainty in the ampitude of each residua spectrum (custered CIB, star-forming dusty gaaxies and radio sources) in the y-map. Figure 7 shows the NILC F/L cross-power spectrum before (back points) and after (red points) foreground correction, using the refined foreground modes presented above. We aso show the custered CIB (green), infrared source (cyan) and radio source (bue) power spectrum contributions Contribution of resoved custers to the tsz power spectrum We simuate the expected Compton parameter map for the detected and confirmed custers of gaaxies in the Panck cataogue (Panck Coaboration XXIX 2013) from their measured 9

10 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map ( + 1)C/2π Fig. 8. Comparison of the measured tsz anguar power spectrum using the cross of the NILC F/L maps (back) with the expected anguar power spectrum of the confirmed custers in the Panck Custer Sampe (orange ine). In red we pot the NILC F/L crosspower-spectrum after masking these custers. The green points correspond to the difference of these two cross-power spectra. The cross-power spectrum between the NILC Compton parameter map and the simuated detected custer map is shown in bue. P(y) y x 10 6 Fig. 9. 1D PDF for the FFP6 simuation maps considering the MILCA inear combination weights obtained for the rea data. The tsz effect (red), diffuse Gaactic emission (cyan), custered CIB (bue) and radio source (back) contributions to the 1D PDF are shown. integrated Compton parameter, Y 5R500. The orange soid ine in Fig. 8 shows the power spectrum of this simuated map. Figure 8 aso shows the cross-power spectrum of the NILC F/L maps (back). In red we pot the cross-power spectrum of the NILC F/L maps after masking the confirmed custers from the PSZ cataogue. The green curve corresponds to the difference of the two cross power-spectra, with and without masking the custers. It is in good agreement with the modeed power spectrum of the confirmed custers of gaaxies. We aso compute the crosspower spectrum of the simuated custer map and the Panck reconstructed Compton parameter NILC map. This is shown in bue in the figure. Here again, the signa is consistent with the expected power spectrum of the confirmed Panck custers of gaaxies. These resuts show that a significant fraction of the signa in the reconstructed Panck Compton parameter maps is due to the tsz effect of detected and confirmed custers of gaaxies, verifying the SZ nature of the signa. In addition, by comparing the tsz power spectrum from the resoved custers with the marginaized tsz power spectrum presented in Sect. 7 we deduce that the measured tsz spectrum incudes an additiona tsz contribution from unresoved custers and diffuse hot gas. 6. Anaysis of High Order Statistics The power spectrum anaysis presented above ony provides information on the 2-point statistics of the Compton parameter distribution over the sky. A fu characterization of the fied can be performed by studying the higher order statistics in the 1D PDF of the map, or by measuring the bispectrum D PDF anaysis We performed an anaysis of the 1D PDF of the NILC and MILCA reconstructed Compton parameter maps. For the tsz effect we expect an asymmetric distribution with a significanty positive tai (Rubiño-Martín & Sunyaev 2003). We thus focus on the asymmetry of the distribution and its unnormaized skew- P(y) y x 10 6 Fig D PDF of the Panck y-map (back) and of the nu map (bue) for the MILCA method. ness. First, we fiter the maps in order to enhance the tsz signa with respect to foreground contamination and noise. To avoid residua point source ringing effects near the edges of the combined PSMASK and GALMASK50 masks we apodize them. We foow the approach of Wison et a. (2012) and use a fiter in harmonic space, constructed from the ratio between the anguar power spectrum of the expected tsz signa in the FFP6 simuations and the power spectrum of the nu y maps. We smooth this ratio using a 21-point square kerne and normaize it to one by dividing by its maximum vaue. Notice that this fiter ony seects the mutipoe range for which the tsz signa is arge with respect to the noise, and thus, it does not modify the non-gaussianity properties. Furthermore, we have found that the fiter used here behaves better than the more traditionay-used Wiener fiter, as it is ess affected by point-source ringing. Then, the 1D PDF of 10

11 Panck Coaboration: Cosmoogy with the a-sky Panck Compton parameter y-map the fitered Compton parameter map, P(y), is computed from the histogram of the pixes. Figure 9 shows the 1D PDF for the FFP6 simuation maps combined using the weights of the MILCA inear combination of the rea data. We present in red the 1D PDF of the tsz effect, which is ceary asymmetric, with a positive tai as expected. Moreover, the asymptotic sope of this red curve at high vaues of y scaes amost as P(y) y 2.5, impying that the underying source counts shoud scae in the same way (i.e., dn/dy Y 2.5 ). This is the predicted scaing behavior for custers (e.g., de Luca et a. 1995; Rubiño-Martín & Sunyaev 2003), and indeed, it is the scaing that we find in the actua number counts of custers in the simuation used. Simiary, the 1D PDF for radio sources (back) is aso asymmetric, but with a negative tai. By contrast, the custered CIB (bue) and diffuse Gaactic emission (cyan) distributions are to first approximation symmetric. From this anaysis we see that the fitering enhances, as expected, the tsz effect with respect to foregrounds and therefore heps their discrimination. For iustration, Fig. 10 shows the 1D PDF for the MILCA Compton parameter map in back. This is the convoution of the 1D PDF of the different components in the map: tsz effect; foregrounds; and noise. Indeed, it ceary shows three distinctive contributions: a Gaussian centra part that exceeds sighty the contribution from noise, as expected from the nu map 1D PDF (cyan curve); a sma negative tai, corresponding most ikey to residua radio sources; and a positive tai corresponding mainy to the tsz signa. A direct computation of the sope of the fu P(y) function in Fig. 10 shows that it converges to 2.5 for y > 10 5, as predicted from the custer counts. A simpe anaysis of the measured 1D PDF can be done by considering the asymmetry of the distribution: A + y p P(y)dy yp P(y)dy, (13) where y p is the peak vaue of the normaized distribution ( P(y)dy = 1). In addition, the non-gaussianity of the positive tai can be quantified by = + y p [ P(y) G(y) ] dy, (14) with G(y) the expected distribution if fuctuations were ony due to noise. For the NILC Compton parameter map we find A = and = Equivaenty, for the MILCA Compton parameter map we find A = 0.26 and = These resuts are consistent with a positive tai in the 1D PDF as expected for the tsz effect. The differences between the NILC and MILCA resuts come mainy from the difference in fitering. Aternativey, we can aso compute the skewness of the obtained distribution, y 3 P(y)dy/ ( y 2 P(y)dy ) 3/2. Foowing Wison et a. (2012) we have chosen here a hybrid approach, by computing the unnormaized skewness of the fitered Compton parameter maps outside the 50% sky mask. In particuar we have computed the skewness of the Panck data Compton parameter maps y 3, of the nu maps ynull 3. For the FFP6 simuations, we computed these for the tsz component y 3 FFP6,SZ and for the sum of a astrophysica components y 3 FFP6,ALL. Tabe 2 shows the resuts for the NILC and MILCA maps. The different fitering function derived for the NILC and MILCA y-maps prevents Tabe 2. Unnormaized skewness mutipied by Method y 3 y 3 NULL y 3 FFP6,S Z y 3 FFP6,ALL NILC MILCA a direct one-to-one comparison of the skewness. However the comparison for each map with the FFP6 simuations for the tsz component and for the sum of a components ceary shows the minor contribution of the foregrounds in both maps. This aow us to argue that the measured skewness vaue is mainy dominated by the tsz signa, as one woud expect from Figs. 9 and 10. By comparing the measured and mode skewness we present in the Sect. 7.2 iustrative constraints on σ Bispectrum Since the SZ signa is non-gaussian, significant statistica information is contained in the bispectrum, compementary to the power spectrum (Rubiño-Martín & Sunyaev 2003; Bhattacharya et a. 2012). We therefore compute the bispectrum of the NILC and MILCA reconstructed Compton parameter maps. The resuts presented here use the binned bispectrum estimator described in Bucher et a. (2010) and Lacasa et a. (2012), which is aso used for the Panck primordia non-gaussianity anaysis (Panck Coaboration XXIV 2013). We mask the maps with the combined PSMASK and GALMASK50, remove the best-fit monopoe and dipoe outside the mask, and degrade the resoution to N side = 1024 to reduce computing time. We use a mutipoe bin size = 64 and a maximum mutipoe max = 2048 for the anaysis. To correct for the bias introduced by masking we have produced non-gaussian simuations with a tsz-ike bispectrum and we have convoved the simuated maps with a Gaussian beam of 10 FWHM. We compute the bispectrum of the simuated fu-sky and masked maps and measure the average ratio between the two. This ratio is used to correct the measured bispectra and fag unreiabe ( 1, 2, 3 ) configurations for which mask effects are too arge to be corrected for. We checked that foreground residuas do not significanty affect the recovered tsz bispectrum using the FFP6 simuations described previousy. In the case of the MILCA reconstructed map (more affected by foregrounds), for exampe, Fig. 11 shows the tsz bispectrum as we as the (absoute vaue of the) bispectra of the different foreground residuas, in some specia configurations, namey equiatera (,, ), orthogona isoscees (,, 2), fat isoscees (,, 2) and squeezed ( min,, ). The foreground residuas yied negigibe bispectra, at east one order of magnitude smaer than the tsz bispectrum over the mutipoes of interest. In Fig. 12 we compare the tsz bispectrum measured on Panck data, with the tsz bispectrum of the FFP6 simuation and with the bispectrum of the maps of detected custers in the Panck cataogue presented above. Custers from the Panck cataogue contribute to an important fraction of the measured bispectrum, at east 30% on arge anguar scaes and more on smaer anguar scaes; the bispectrum hence aso probes the unresoved tsz signa, as was the case for the power spectrum. On arge anguar scaes this may be the signature of the cus- 11