0803.1161/ms.tex
1: %\documentclass[aps,prd,amsfonts,amsmath,preprint]{revtex4}
2: \documentclass[aps,prd,amsfonts,amsmath,showpacs]{revtex4}
3: \usepackage[dvips]{graphicx}
4: \usepackage{bm}
5: 
6: \begin{document}
7: 
8: \title{Point Source Contamination in CMB Non-Gaussianity Analyses}
9: 
10: \author{Daniel Babich}
11: 
12: \email{babich@tapir.caltech.edu}
13: 
14: \affiliation{Theoretical Astrophysics, California Institute of Technology, Pasadena, CA 91125 U.S.A.}
15: 
16: \author{Elena Pierpaoli}
17: 
18: \email{pierpaol@usc.edu}
19: 
20: \affiliation{University of Southern California, Los Angeles, CA 90089 U.S.A.}
21: 
22: \date{\today. To be submitted to Phys. Rev. D.}
23: 
24: \pacs{98.70.Dk, 98.80.-k}
25: 
26: \begin{abstract}  	
27: In this paper we analyze the biasing effect of point sources, either thermal Sunyaev-Zeldovich clusters 
28: or standard radio sources, on the estimated strength of the non-Gaussianity in the Cosmic Microwave Background (CMB). 
29: We show that the biggest contribution comes from the cross--correlation of the CMB with the matter density rather 
30: than from the poisson term which is conventionally assumed in these calculations. For the three year WMAP data, 
31: we estimate that point sources could produce a non--Gaussian signature equivalent to a bias in $f_{NL}$ of 
32: $0.35, 0.24, -0.097, -0.13$ in the Ka, Q, V and W bands respectively. The level of bias we find is largely insufficient 
33: to explain the very high $f_{NL}$ values recently detected by Yadav and Wandelt. 
34: For Planck, we estimate the point source bispectra to contaminate the $f_{NL}$ estimator with a bias of $1.3, 0.34, -0.25, -0.48$
35: at $30, 44, 70, 100~{\rm GHz}$ respectively. These results depend on the assumed redshift distribution of the point sources. 
36: However, given the projected Planck sensitivity of $\Delta f_{NL} \simeq 5$ (95 \% C.L.), a good estimate of point sources' properties 
37: including their number density and redshift distribution is essential before deriving strong conclusions on primordial 
38: non--Gaussianity.
39: 
40: \end{abstract}
41: 
42: \maketitle
43: 
44: \section{Introduction}
45: 
46: Recent claims by Yadav and Wandelt \cite{Yadav07} of the detection of strong primordial non-Gaussianity in the three-year 
47: Wilkinson Microwave Anisotropy Probe (WMAP) data \cite{Spergel07} have the potential to revolutionize our understanding of 
48: the early universe. These results were also found in the WMAP five-year analysis, although with less statistical 
49: significance \cite{Komatsu08}. The strength of the non-Gaussianity detected in their analysis is more than two orders of 
50: magnitude larger than the non-Gaussianity expected in the simplest model of single field, slow roll model of inflation 
51: \cite{Maldacena03}. This detection, if it stands up to scrutiny, will be the first definitive indication that the simplest 
52: model of inflation cannot adequately explain all of the current cosmological observations and must be modified in some way. 
53: 
54: In Yadav and Wandelt's analysis \cite{Yadav07}, the detection of the non-Gaussian signal is due to the simultaneous 
55: reduction in the estimator's error bars and the shift in the central value as smaller angular scale information
56: was included in the analysis. One obvious concern is that the estimator is contaminated by foreground emission, 
57: in particular radio points sources and the thermal Sunyaev-Zeldovich effect which become increasingly more important 
58: on small angular scales. 
59: 
60: In this paper we will analyze the influence of point sources, for the experiment resolution of WMAP both the radio sources 
61: and SZ clusters effectively act as point sources, on the standard non-Gaussianity estimator.
62: While it has been claimed that the non-Gaussianity caused by Poisson fluctuations in the number density of radio point
63: sources can be safely separated from the primordial non-Gaussian signal \cite{Komatsu02}, we will demonstrate
64: the other forms of point source non-Gaussianity cannot be safely ignored. In addition to the standard forms 
65: of non-Gaussianity produced by point sources, we will show that cross-correlation between the point source 
66: power spectrum and the CMB temperature anisotropies or instrument noise will produce non-Gaussianity of a very
67: similar form as the local model. The local model is the form typically assumed in analyses of primordial 
68: non-Gaussianity, so these new non-Gaussian contributions will bias the estimator in a fashion that cannot easily 
69: be corrected.
70: 
71: This paper is organized as follows. In \S \ref{sec:ps} we demonstrate that point sources can produce a 
72: bispectrum that has the same form as the local model. In \S \ref{sec:bias} we derive the bias induced by 
73: the various point source bispectra. In \S \ref{sec:conc} we conclude. We use the WMAP three-year cosmological
74: model \cite{Spergel07} for numerical calculations.
75: 
76: \section{Estimator Bias}\label{sec:bias}
77:  
78: While non-Gaussianity generically implies that any higher-order connected correlation function is non-zero,
79: it is typical to focus on the three-point correlation function, or equivalently the bispectrum, because 
80: it has the simplest form of all non-Gaussian correlation functions and for weak non-Gaussianity it 
81: contains the nearly all of the information \cite{Babich05}. The three-point correlation function can be
82: factored into a component fixed by rotational invariance, which is assumed {\it a prioiri}, and a piece
83: determined by the underlying mechanism that produced the non-Gaussianity \cite{Komatsu02}. Rotational 
84: invariance forces the three-point correlation function to be proportional to the Gaunt integral, 
85: \begin{equation}
86: \mathcal{G}^{\ell_1 \ell_2 \ell_3}_{m_1 m_2 m_3} = \sqrt{\frac{(2 \ell_1 + 1)(2 \ell_2 + 1)(2 \ell_3 + 1)}{4\pi}}
87: \left(\begin{array}{ccc} \ell_1 & \ell_2 & \ell_3 \\ m_1 & m_2 & m_3 \end{array}\right)
88: \left(\begin{array}{ccc} \ell_1 & \ell_2 & \ell_3 \\ 0 & 0 & 0 \end{array}\right),
89: \end{equation}
90: and the CMB three-point correlation function can be written as
91: \begin{equation}
92: \langle a_{\ell_1 m_1} a_{\ell_2 m_2} a_{\ell_3 m_3} \rangle = \mathcal{G}^{\ell_1 \ell_2 \ell_3}_{m_1 m_2 m_3} 
93: b_{\ell_1, \ell_2, \ell_3},
94: \end{equation}
95: where the reduced bispectrum, $b_{\ell_1, \ell_2, \ell_3}$, contains information about the form of non-Gaussianity. 
96: 
97: The estimators used in CMB non-Gaussianity analyses are optimized for the detection of a signal with a very particular 
98: form, namely the local model \cite{Babich05}. The local model assumes that the initial curvature perturbations can be written as
99: \begin{equation}\label{eq:local}
100: \Phi({\bm x}) = \phi_g({\bm x}) + f_{NL}[\phi^2_g({\bm x}) - \langle \phi^2_g({\bm x}) \rangle ],
101: \end{equation}
102: where $\phi_g$ is Gaussian. The non-linear terms in this model lead to the following bispectrum for the initial
103: curvature perturbations
104: \begin{equation}
105: B(k_1,k_2,k_3) = 2f_{NL}[P(k_1)P(k_2) + {\rm cyc.}],
106: \end{equation}
107: where $P(k)$ is the power spectrum.
108: The ordinary linear radiative transfer function are subsequently used to calculate the CMB temperature anisotropies 
109: from these initial curvature perturbations. The statistical properties of the CMB temperature anisotropies will mirror
110: the statistical properties of underlying curvature perturbations since we are consider linear radiative transfer. The
111: levels of non-Gaussianity claimed by Yadav \& Wandelt are significantly larger than the expected non-Gaussianity produced
112: by non-linear radiative transfer.
113: 
114: However the estimators are sensitive to any bispectrum 
115: form that might be present in the data, regardless of its origin. In this section we will determine the induced bias 
116: produced by the various cross-correlation bispectra described in the next section.
117: 
118: The non-Gaussianity estimator can be expressed as \cite{Creminelli06}
119: \begin{equation}\label{eq:est}
120: \hat{f}_{NL} = \frac{1}{A} \left[ \sum \mathcal{G}^{\ell_1 \ell_2 \ell_3}_{m_1 m_2 m_3} 
121: \frac{b_{\ell_1, \ell_2, \ell_3}}{C^T_{\ell_1} C^T_{\ell_2} C^T_{\ell_3}}
122: a_{\ell_1 m_1} a_{\ell_2 m_2} a_{\ell_3 m_3} \right], 
123: \end{equation}
124: where the normalization is
125: \begin{equation}
126: A  = \sum \frac{(2 \ell_1 + 1)(2 \ell_2 + 1)(2 \ell_3 + 1)}{4\pi}
127: \left(\begin{array}{ccc} \ell_1 & \ell_2 & \ell_3 \\ 0 & 0 & 0 \end{array}\right)^2
128: \frac{b^2_{\ell_1, \ell_2, \ell_3}}{C^T_{\ell_1} C^T_{\ell_2} C^T_{\ell_3}}, 
129: \end{equation}
130: here $C^T_{\ell} = C_{\ell} + C^N_{\ell}$ is the sum of the CMB signal and noise. The CMB experimental noise parameters
131: are described in Table \ref{table:exp_info}.
132: 
133: We are ignoring the additional linear term in the estimator because the contribution of radio point source will not bias 
134: this piece of the estimator if the signal and noise covariance matrices well represent the real data.
135: 
136: The weight functions used in the estimator are optimized for a bispectrum produced by the local model. The estimator bias 
137: will be determined by substituting the various forms of the point sources bispectra into the estimator 
138: \begin{equation}
139: \Delta f^{\alpha}_{NL} = \frac{1}{A} \sum \frac{(2 \ell_1 + 1)(2 \ell_2 + 1)(2 \ell_3 + 1)}{4\pi}
140: \left(\begin{array}{ccc} \ell_1 & \ell_2 & \ell_3 \\ 0 & 0 & 0 \end{array}\right)^2
141: \frac{b_{\ell_1, \ell_2, \ell_3} b^{\alpha}_{\ell_1, \ell_2, \ell_3}}{C^T_{\ell_1} C^T_{\ell_2} C^T_{\ell_3}}.
142: \end{equation}
143: Here $b^{\alpha}_{\ell_1, \ell_2, \ell_3}$ is the one of the reduced bispectrum produced by point sources. The CMB bispectrum
144: produced by the local model is generally negative. The collapsed triangle modes, which have the highest signal-to-noise, are
145: always negative. So a positive point source bispectrum will cause a negative estimator bias.
146: We will now discuss the possible bispectrum terms. 
147: 
148: \section{Point Source Bispectra} \label{sec:ps}
149: 
150: The observed signal is the sum of the primordial and secondary temperature anisotropies, foreground emission and 
151: instrument noise. While the secondary anisotropies and extra-galactic foregrounds, which maybe quite non-Gaussian,
152: are important on small angular scales, the signal on the large angular scales relevant for WMAP is dominated by
153: primary anisotropies. Thus it is assumed that any measured non-Gaussianity by WMAP is primordial in nature. We will
154: argue that cross-correlations between some of these signals may induce bispectra on large angular scales. 
155: 
156: Moreover these bispectra may have similar forms as the local model if the signal power spectrum becomes spatial 
157: inhomogeneous in manner that then correlates with a second component in the observed signal. For example, the matter 
158: overdensity will bias the local radio point source power spectrum and it will be correlated with the CMB temperature 
159: anisotropies produced via the ISW effect. This will produce a bispectrum similar in form to the local model. 
160: 
161: This is not a coincidence as the non-Gaussianity in the local model is produced by the modulation of the small scale 
162: inflaton power spectrum by the large scale inflaton fluctuations that have already left the horizon and frozen out. 
163: The parameter $f_{NL}$ is a measure of the non-linear coupling between these different scales. Likewise the large 
164: scale matter overdensity modulates the small scale Poisson fluctuation power spectrum by altering the local number 
165: density of radio point sources. In an analogous fashion the bias describes the coupling between the different scales.
166: As discussed in the previous section any bispectrum present in the data can bias the estimator.
167: 
168: \section{Anisotropy Mechanisms}
169: 
170: Now we will discuss the various physical effects considered in this paper -- radio point source emission, the thermal
171: Sunyaev-Zeldovich effect and the integrated Sachs-Wolfe effect.
172: 
173: \subsubsection{Radio Point Sources}
174: 
175: The temperature anisotropy induced by unresolved radio point sources can be expressed as an integral over
176: their flux distribution function
177: \begin{equation}\label{eq:flux}
178: \frac{\Delta T}{T}(\hat{\bm n},\nu) = \frac{1}{c_{\nu}}
179: \int_0^{\bar{S}(\hat{\bm n})} dS S \frac{dN}{dS}(S,\nu;\hat{\bm n}).
180: \end{equation}
181: The conversion between the temperature and intensity fluctuations is 
182: \begin{equation}
183: c_{\nu} = \frac{\partial B_{\nu}}{\partial \ln{T}}(T_{CMB}), 
184: \end{equation}
185: where $T_{CMB} = 2.728$ K and $B_{\nu}(T)$ is a blackbody frequency distribution.
186: 
187: Note that we have allowed both the upper flux limit for unresolved radio point sources and their number density to be spatially 
188: inhomogeneous. As we will describe below, these spatial inhomogeneities will correlate with other signals present in the data to 
189: produce bispectra in the observed CMB data.
190: 
191: The actual values for the bispectra calculated in this paper depend on the radio point sources properties (both flux and redshift distributions) 
192: and on the flux cut at a given frequency for a specific experiment. In order to evaluate the residual 
193: point source contribution to the total estimated bispectrum in the WMAP--3yrs data, we must consider the technique used by WMAP to 
194: identify and substract radio point sources \cite{Hinshaw07} and the point source mask applied to the data. The WMAP source 
195: selection criterion does not correspond to a single flux threshold at a given frequency, rather it requires that a candidate 
196: source should be seen with a minimal statistical significance in all channels \citep{Pierpa03}.  As Yadav \& Wandelt's 
197: results \citep{Yadav07}, which are the motivation for this paper, were derived using the V and W frequencies bands, the radio
198: point source populations at these frequencies is the most relevant. Unfortunately, as most radio sources are stronger at low frequencies, 
199: they tend to be detected with higher significance in the K -- Q bands than in the V -- W ones. 
200: 
201: In addition from a blind search,  at 20--30 GHz it is possible to use lower--frequency catalogs of point sources as tracers for 
202: detection \citep{LopezCaniego07}. As a result, the source population at such frequencies is much 
203: better characterized than the one at higher frequencies. By considering flux number counts at low frequencies it is possible 
204: to give an estimate of the flux above which the detected point sources create a complete catalog. 
205: This flux threshold is estimated to be above 1.1 Jy at 23 GHz (K Band) \cite{LopezCaniego07,Gonzalez08}, while in the W band 
206: (94 GHz) the number counts are not sufficiently well determined to allow such estimate.
207: An alternative blind search technique applied to the WMAP V and W band data increases the number of sources found in the V band 
208: by 50\% compared to the WMAP team results \citep{ChenWright07}. These results include some sources that were not contained in 
209: the WMAP point source mask. The new sources found, however, do not seem to represent a different population than the ones
210: previously detected. As the WMAP 5 year data is now available, more work on point source characterization at the frequencies 
211: where the CMB science is derived should be possible. 
212: 
213: The actual residual contribution of point sources depends upon the mask applied to the observed map. In the case of WMAP, this 
214: mask considers the WMAP detected sources (about 300) as well as some bright sources from other low--frequencies catalogs, for 
215: a total of approximately seven hundred sources masked. The actual selection function that this procedure imposes at the V and 
216: W bands is poorly understood, and indeed some sources detected by \cite{ChenWright07} were not masked. However, while the 
217: detection threshold implied by this whole procedure is poorly defined, there is good agreement in the residual point sources 
218: power spectrum contribution as derived by different authors \citep{Huffenberger06,ChenWright07,Hinshaw07}. This can be 
219: approximately translated in a flux threshold of 0.6 Jy in the Q band \citep{Huffenberger06} and 0.75 Jy in the V band 
220: \cite{ChenWright07}. For illustrative purposes, we will adopt here an approximate estimate for the flux cut-off of 0.7 Jy 
221: in all WMAP bands.  
222: In \S \ref{sec:results} we discuss the dependence of our results on this choice.
223: 
224: The Planck satellite will have better resolution and sensitivity, resulting in a lower detection threshold for point sources. 
225: In the following, we make predictions of the bispectrum expected in the final Planck maps, considering the detection thresholds 
226: for a 95\% complete sample derived  by \cite{LopezCarniego06} using realistic Planck sky simulations. The flux cut-offs for both
227: WMAP and Planck, as well as, the adopted instrument noise parameters are given in Table \ref{table:exp_info}.
228: 
229: \begin{table}
230: \begin{center}
231: \begin{tabular}{c|c|c|c}
232: \hline
233: \hline
234: Frequency (GHz) & $\bar{S}$ (Jy) & FWHM (arcmin) & $\Delta T/T$ \\
235: \hline
236: WMAP -- 33 (Ka) & 0.7 & 41 & 5.7  \\
237: WMAP -- 41 (Q)  & 0.7 & 28 & 8.2 \\
238: WMAP -- 61 (V)  & 0.7 & 21 & 11.0  \\
239: WAMP -- 94 (W)  & 0.7 & 13 & 18.3 \\
240: \hline
241: Planck -- 30  & 0.33 & 33 & 1.6 \\
242: Planck -- 44  & 0.36 & 23 & 2.4  \\
243: Planck -- 70  & 0.34 & 14 & 3.6 \\
244: Planck -- 100 & 0.13 & 11 & 1.6  \\
245: \hline
246: \hline
247: \end{tabular}\\
248: \end{center}
249: \caption{\label{table:exp_info} Radio point source flux threshold, FWHM, and instrument pixel noise (in $10^{-6}$)
250: for the relevant WMAP and Planck frequency bands. For WMAP the various band names are also listed.}
251: \end{table}
252: 
253: Finally, in order to compute the bispectra implied by point sources below a given flux, we adopt the source counts predictions 
254: of Toffolatti et al \cite{Toffolatti98} rescaled by a factor 0.8, as suggested by the matching of these predictions with the 
255: actual number counts at fluxes above 1.1 at 41 GHz obtained by \cite{Gonzalez08}. We will use the radio point source bias $b^{PS} 
256: \simeq 1.7$ \cite{Smith07, Blake04, Boughn02} as inferred for low frequency radio point sources. This value is uncertain and almost
257: definite varies for the various population type that constitute the high frequency sample.
258: 
259: It is difficult to constrain the redshift distribution due to the lack of optical studies of $60-96 {\rm GHz}$ source population, 
260: work at $23 {\rm GHz}$ has been conducted by 
261: Gonzalez et al. for fluxes above 1 Jy \cite{Gonzalez08}. At these fluxes and frequencies the number counts are dominated by 
262: QSOs for relatively high redshifts. This population is well approximated by the following analytical formula:
263: \begin{equation}\label{eq:n_PS}
264: n^{PS}(z)  \propto 0.75 \times e^{-(z-z_0)^2/(2 \sigma)^2},
265: \end{equation}
266: with $z_0 = 0.95$ and $\sigma$ is 0.4 (0.9) for $z < 1$ ($z > 1$). This redshift distribution is in agreement with recent 
267: studies of radio sources populations at ~90 GHz with ATCA \citep{Sadler07}. In addition, there is a small (10--15 \%) of 
268: the total population that consists of radio loud galaxies which peaks at much lower redshifts ($z \le 0.1$). Given the small 
269: number of sources, it is difficult to derive an appropriate fitting formula for this other population. In the following, 
270: we will adopt the following distribution of low redshift sources
271: \begin{equation}\label{eq:n_PSlow}
272: n^{PS}(z) \propto 0.25 \times 10^{-3z},
273: \end{equation}
274: that provides a better fit to counts found by Gonzalez et al. \cite{Gonzalez08}.
275: We will take the sum of Eqs.\ref{eq:n_PS} \& \ref{eq:n_PSlow} with a common proportionality factor determined by  requiring 
276: $n^{PS}(z)$ to integrate to unity over the range $z=0$ to $z=3.1$. The relative amplitude of the QSO and radio galaxy contribution
277: to the sources is derived from the optical indentifications of Gonzalez et al \cite{Gonzalez08}.
278: This fit will be called model 1. Gonzalez et al. \cite{Gonzalez08} provide a fit to a theoretical model of the luminosity functions of 
279: the various source populations as derived from \cite{DeZotti05}. To understand how the uncertainty in the source redshift distribution 
280: affects our results, we also do calculations with the model of Gonzalez \cite{Gonzalez08}; this will be called model 2.
281: 
282: WMAP only resolves the high-flux sources which are typically dominated by AGN. Most likely lower flux sources consist of a 
283: different population (e.g. \cite{PierpaPerna04, DeZotti05}) and therefore have a different redshift distribution with possibly 
284: more weight either at lower or higher redshifts. An analogous redshift analysis on higher frequencies catalog is strongly needed, but 
285: is not available at this time. Future investigations of radio point sources catalogs plus WMAP and Planck results are higher
286: frqeuencies will help clarify this issue. For the aims of this paper we take Eq.~\ref{eq:n_PS} to be the redshift distribution at 
287: all frequencies and fluxes and keep in mind the potential uncertainty that this assumption introduces.
288: 
289: \subsubsection{Thermal Sunyaev-Zeldovich Effect}
290: 
291: The hot plasma in the intra-cluster medium will produce temperature anisotropies via Thomson scattering of the 
292: incident CMB photons; this is the well-known thermal Sunyaev-Zeldovich (SZ) effect (see \citet{Carlstrom02} for 
293: a review). Following the model of Komatsu \& Seljak \cite{Komatsu02a,Komatsu01}, the temperature anisotropies 
294: produced by the SZ effect can be expressed as an integral over the cluster mass distribution function
295: \begin{equation}\label{eq:sz}
296: \frac{\Delta T}{T}(\hat{\bm n},\nu) = g_{\nu} \int dz \frac{dV}{dz} \int_0^{\infty} dM y(M,z)
297: \frac{dn}{dM}[M,z;\hat{\bm n \chi(z)}],
298: \end{equation}
299: here the frequency dependence of the thermal SZ effect is given by
300: \begin{equation}
301: g_{\nu} = x\frac{e^x+1}{e^x-1} - 4,
302: \end{equation}
303: where $x = h\nu/k_B T_{\rm CMB}$. The Compton y-parameter is related to the line-of-sight integral of the the 
304: cluster's thermal pressure, $dn/dM$ is the cluster mass function and the volume element is
305: \begin{equation}
306: \frac{dV}{dz} = \frac{c}{H(z)} \chi^2(z),
307: \end{equation}
308: where $\chi(z)$ is comoving distance to redshift z. The details of the implementation of this model are extensively 
309: discussed in \cite{Komatsu02a}. Ignoring clustering terms in the power spectrum, we find the thermal SZ point source 
310: power spectrum
311: \begin{equation}\label{eq:cl_sz}
312: C^{SZ}_{\ell}(\nu) = g^2_{\nu} \int dz \frac{dV}{dz} \int_0^{\infty} dM y_{\ell}^2(M,z) \frac{dn}{dM}(M,z).
313: \end{equation}
314: 
315: In a similar fashion to the radio point sources, the number density of massive clusters will be changed by both the
316: biasing effect of the large scale matter overdensity and gravitational lensing magnification. 
317: In order to determine how these processes will affect the SZ power spectrum we need to know the redshift distribution
318: of power in the SZ effect. The weighted redshift distribution of power produced by the SZ effect can be expressed 
319: as the integral over halo mass of the Sheth-Tormen \citep{Sheth99} halo mass function 
320: \begin{equation}\label{eq:n_SZ}
321: n^{SZ}(z) \propto  \frac{dV}{dz} \int_{M_{\rm min}}^{M_{\rm max}} dM M^{2\alpha} \frac{dn}{dM}(M,z).
322: \end{equation}
323: The cluster y-parameter -- mass scaling relationship
324: slope is taken to be $\alpha = 1.6$ \cite{Nagai06}. The limits of integration are taken to be $M_{\rm min}= 10^{14} M_{\odot}$ 
325: and $M_{\rm max} = 5\times 10^{15} M_{\odot}$. The redshift distribution is normalized so it will integrate to unity. We also 
326: need the bias weighted redshift distribution of power produced by the SZ effect can be expressed as the integral over the 
327: Sheth-Tormen halo mass function 
328: \begin{equation}\label{eq:bn_SZ}
329: (bn)^{SZ}(z) \propto  \frac{dV}{dz} \int_{M_{\rm min}}^{M_{\rm max}} dM M^{2\alpha} \frac{dn}{dM}(M,z)~b(M,z),
330: \end{equation}
331: where $b(M,z)$ is the standard bias of the Sheth-Tormen mass function. 
332: 
333: \subsubsection{Integrated Sachs-Wolfe Effect}
334: 
335: Most of the power in the CMB temperature anisotropies is produced at high redshift during recombination. There will be
336: very little cross-correlation between these temperature anisotropies and the low redshift matter overdensity responsible
337: for altering the small scale point source power spectrum. However additional CMB temperature anisotropies can be generated
338: at low redshift via the Integrated Sachs-Wolfe (ISW) effect if the gravitational potential fluctuations are evolving in time.
339: In a fully matter dominated regime the gravitational potential fluctuations are static, however as the universe becomes dark
340: energy dominated the ISW effect can occur.
341: 
342: The temperature anisotropy produced by the ISW effect can be expressed as
343: \begin{eqnarray}
344: \frac{\Delta T}{T}(\hat{\bm n}) &=& -2 \int dz \frac{\partial \Phi}{\partial z}, \\
345: &=& 3 H^2_0 \Omega_M \int dz [(1+z) D(z)]' \int \frac{d^3{\bm k}}{(2\pi)^3} e^{i{\bm k} \cdot \hat{\bm n} \chi} 
346: \frac{\delta({\bm k})}{k^2},
347: \end{eqnarray}
348: where $D(z)$ is the linear theory growth function and the prime denotes differentiation with respect to $z$.
349: 
350: \begin{figure}
351: \centering
352: \includegraphics[width=10cm,height=10cm]{overlap.eps}
353: \caption{\label{fig:nz} Redshift weight functions for the ISW effect (green, dot-dashed); radio point 
354: sources -- model 1 (red, dashed) and model 2 (blue, long-dashed); and thermal SZ effect (solid, black).}
355: \end{figure}
356: 
357: In Fig. \ref{fig:nz} we show the redshift weight functions for the ISW effect $[(1+z)D(z)]'$ (green, dot-dashed); the 
358: radio point sources $b^{PS} n^{PS}(z) D(z)$  -- model 1 (red, dashed) and  model 2 (blue, long-dashed); and the thermal 
359: SZ effect $(bn)^{SZ}(z) D(z)$ (solid, black). The overlap of these weight fuctions will determine the amplitude of the 
360: cross-correlation spectra as described in \S \ref{sec:cross} \& \ref{sec:mag} and shown in Figs. \ref{fig:cross_power} 
361: \& \ref{fig:mag_power}.
362: 
363: \subsection{Point Source Bispectrum}\label{sec:ps_bisp}
364: 
365: The simplest bispectrum form produced in the CMB is due to radio point source Poisson fluctuations. 
366: The reduced bispectrum is independent of scale and can be written as
367: \begin{equation}
368: b_{\ell_1, \ell_2, \ell_3} = c_{\nu}^{-3} \int_0^{\bar{S}} dS S^3 \frac{\bar{dN}}{dS}(S,\nu).
369: \end{equation}
370: This bispectrum component has been detected in the WMAP data \cite{Komatsu03}. Since its functional form is not similar 
371: to the local model it will not significantly bias the estimator and we will ignore it as did Yadav \& Wandelt. The five-year
372: WMAP analysis \cite{Komatsu08} includes estimates (and corrections) for the estimator bias produced by this bispectrum form.
373: 
374: \subsection{Number Density Modulation}\label{sec:cross}
375: 
376: The Poisson fluctuation power spectrum in a certain region of the sky can be written as integral over the distribution of 
377: sources in that region. If the anisotropic component of the power spectrum correlates with any other signal present in the 
378: data, non-Gaussian correlation functions will be generated. In this subsection we will focus on correlation of the large 
379: scale matter overdensity, which bias the number density of point sources, with the ISW effect.
380: 
381: The power spectrum produced by radio point sources in direction $\hat{\bm n}$ is
382: \begin{equation}
383: C^{PS}(\hat{\bm n}) = c_{\nu}^{-2} \int_0^{\bar{S}} dS S^2 \frac{dN}{dS}(S,\nu;\hat{\bm n}),
384: \end{equation}
385: where the anisotropic point source distribution can be expressed in terms of the matter overdensity as
386: \begin{equation}\label{eq:ps_n}
387: \frac{dN}{dS}(S,\nu;\hat{\bm n}) = \frac{\bar{dN}}{dS}(S,\nu) \left[ 1 + b^{PS}\int dz 
388: n^{PS}(z) \delta(\hat{\bm n},z) \right],
389: \end{equation}
390: the mean point source distribution was described in \S \ref{sec:ps}. The large scale matter overdensity, which biases the 
391: local number density of point sources as shown in Eq. \ref{eq:ps_n}, will be correlated with the large scale CMB temperature 
392: anisotropies produced by the ISW effect and the following reduced bispectrum will be induced
393: \begin{equation}
394: b_{\ell_1, \ell_2, \ell_3} = 2 \bar{C}^{PS} (X^{PS}_{\ell_1}  + \mathrm{cyc.}).
395: \end{equation}
396: Here the isotropic source distribution $\bar{dN}/dS$ leads to an isotropic power spectrum $\bar{C}^{PS}$. The matter-ISW 
397: cross correlation spectrum can be expressed as
398: \begin{equation}\label{eq:x_ps}
399: X^{PS}_{\ell} = \frac{3 H^2_0 \Omega_M b^{PS}}{\ell^2} \int \frac{H(z) dz}{c} P\left[\frac{\ell}{\chi(z)}\right]
400:  D(z) n^{PS}(z) [(1+z) D(z)]' 
401: \end{equation}
402: where $P(k)$ is the matter power spectrum and $b^{PS} \simeq 1.7$ is the radio point source bias. The cross-correlation
403: spectrum is positive because matter overdensities correspond to potential wells. At low redshift as the amplitude of the 
404: potential wells decay the CMB photons experience a net blue-shift, thus the cross-correlation is positive. In Eq. \ref{eq:x_ps} 
405: we have employed Limber's approximation to simplify the cross-correlation spectrum. The function $n^{PS}(z)$ is the redshift 
406: probability distribution function of the radio point sources defined in Eq. \ref{eq:n_PS}. The large scale matter overdensity 
407: affects the thermal SZ power spectrum in an analogous fashion. 
408: 
409: In Fig. \ref{fig:cross_power} we show the matter-ISW cross correlation spectrum for the thermal SZ effect (solid, black) and
410: the radio point sources -- model 1 (red, dashed), model 2 (blue, long-dashed). The thermal SZ effect has the largest cross
411: correlation because the clusters tend to be located at lower redshift than the radio point sources. The ISW effect primarily
412: occurs at low redshift once the universe is strongly dark energy dominated, so it most strongly overlaps with the SZ effect.
413: 
414: \begin{figure}
415: \centering
416: \includegraphics[width=10cm,height=10cm]{cross.eps}
417: \caption{\label{fig:cross_power} Density-ISW cross-correlation spectrum for thermal SZ (solid, black); radio point 
418: sources -- model 1 (red, dashed) and model 2 (blue, long-dashed).}
419: \end{figure}
420: 
421: In Fig. \ref{fig:cross_bispect} we show the number density modulation bispectrum for the thermal SZ effect (solid, black) and
422: the radio point sources -- model 1 (red, dashed), model 2 (blue, long-dashed). Also shown for reference is the equilateral shape 
423: $(\ell_1 = \ell_2 = \ell_3 = \ell)$ of the primordial local model bispectrum ($f_{NL} = 1$) (green, dot-dashed). The equilateral 
424: shape of the local model bispectrum changes signs, the zero-crossing are obvious from the plot. This is a consequence of the radiative 
425: transfer functions producing both hot and cold regions on the sky. The collapsed shape always has the same sign, opposite the sign 
426: of $f_{NL}$. For the local model the collapsed bispectra have the highest signal-to-noise. 
427: 
428: \begin{figure}
429: \centering
430: \includegraphics[width=10cm,height=10cm]{bispect_x.eps}
431: \caption{\label{fig:cross_bispect} Number density modulation bispectrum for thermal SZ (solid, black); radio point 
432: sources -- model 1 (red, dashed) and model 2 (blue, long-dashed). The equilateral shape $(\ell_1 = \ell_2 = \ell_3 = \ell)$ 
433: of the local model ($f_{NL} = 1$) (green, dot-dashed) is also shown for reference.}
434: \end{figure}
435: 
436: \subsection{Magnification Modulation}\label{sec:mag}
437: 
438: The distribution of matter along the line-of-sight between the observer and the sources will gravitationally lens these
439: point sources. The priniciple effect of gravitational lensing will be to change the source density by magnifying
440: and de-magnifying certain regions of the sky. The magnification in a given direction can be written, to first order, as 
441: \begin{equation}
442: \mu(\hat{\bm n}) \simeq 1 + 2 \kappa(\hat{\bm n}). 
443: \end{equation}
444: The convergence field distorting a background source at $z$ is related to matter overdensity as
445: \begin{equation}
446: \kappa(\hat{\bm n}, z) = \frac{3 \Omega_M H^2_0}{2} \int \frac{c dz'}{H(z')} (1+z') \frac{\chi(z')}{\chi(z)} [\chi(z)-\chi(z')]
447: \delta(\hat{\bm n},z'),
448: \end{equation}
449: and the average magnification of SZ point sources in a given direction is then given by
450: \begin{equation}
451: \kappa(\hat{\bm n}) = \int dz n^{SZ}(z) \kappa(\hat{\bm n}, z).
452: \end{equation}
453: The observed SZ cluster number density in a given direction will be related to the intrinsic number density as
454: \begin{eqnarray}
455: \frac{dn^{obs}}{dM}(M,\hat{\bm n}) &=& \frac{1}{\mu(\hat{\bm n})} \frac{dn}{dM}(M,\hat{\bm n}), 
456: \\
457: &\simeq& [1 - 2\kappa(\hat{\bm n})] \frac{dn}{dM}(M,\hat{\bm n}).
458: \end{eqnarray}
459: The convergence field is correlated with the CMB temperature anisotropies produced via the ISW effect,
460: \begin{equation}\label{eq:m_sz}
461: M^{SZ}_{\ell} = \frac{9 \Omega^2_M H^4_0}{2\ell^2} \int_0^{\infty} dz n^{SZ}(z) 
462: \int_0^{z} dz_1 P\left[\frac{\ell}{\chi(z_1)}\right] (1+z_1) D(z_1) [(1+z_1) D(z_1)]' \frac{\chi(z_1)}{\chi(z)} [\chi(z)-\chi(z_1)].
463: \end{equation}
464: Again we have evaluated the cross-correlation according to Limber's approximation and the convergence-ISW cross-correlation spectrum
465: is positive. This magnification effect will result in the following reduced bispectrum
466: \begin{equation}\label{eq:lensing_sz}
467: b_{\ell_1, \ell_2, \ell_3} = -2 [M^{SZ}_{\ell_1} (C^{SZ}_{\ell_2} + C^{SZ}_{\ell_3}) 
468: + M^{SZ}_{\ell_2} (C^{SZ}_{\ell_1} + C^{SZ}_{\ell_3}) 
469: M^{SZ}_{\ell_3} (C^{SZ}_{\ell_1} + C^{SZ}_{\ell_2})].
470: \end{equation}
471: The bispectrum is negative because large scale CMB hot spots (positive ISW effect) correlate with positive magnification which always
472: reduces the amplitude of point source Poisson fluctuations.
473: 
474: The radio point source selection function is expressed in terms of the observed flux which can be affected by gravitational 
475: lensing due matter along the line-of-sight. The gravitational lensing magnification will modulate the flux cut-off and be 
476: correlated with the ISW temperature anisotropies. The point source dilution effect discussed above will also occur. The 
477: radio point source Poisson fluctuation power spectrum can be expressed as
478: \begin{equation}
479: C^{PS}(\hat{\bm n}) = c_{\nu}^{-2} \int_0^{\bar{S}/\mu(\hat{\bm n})} dS S^2 \frac{1}{\mu(\hat{\bm n})} \frac{dN}{dS}(S,\nu).
480: \end{equation}
481: Linearizing in the convergence field, we find that the Poisson fluctuation power spectrum becomes anisotropic
482: \begin{equation}
483: C^{PS}(\hat{\bm n}) = 2 c_{\nu}^{-2} \kappa(\hat{\bm n}) \left[ \bar{S}^3 \frac{dN}{dS}(\bar{S},\nu) + \bar{C}^{PS} \right].
484: \end{equation}
485: This will result in the following reduced bispectrum
486: \begin{equation}\label{eq:lensing}
487: b_{\ell_1, \ell_2, \ell_3} =  -4(M^{PS}_{\ell_1} + {\rm cyc.}) \left[ \bar{S}^3 \frac{dN}{dS} (\bar{S}, \nu) + \bar{C}^{PS} \right],
488: \end{equation}
489: where $M^{PS}_{\ell}$ is the radio point source version of Eq. \ref{eq:m_sz}.
490: Note that gravitational lensing affects the radio point source power spectrum by changing both the upper flux cut-off
491: and the source counts, whereas the SZ power spectrum is only altered by changes in the local cluster counts. If the SZ
492: clusters are detected with high signal-to-noise and removed from the CMB maps then flux cut-off modulation effect will
493: also produce an additional bispectrum term.
494: 
495: In Fig. \ref{fig:mag_power} we show the convergence-ISW cross correlation spectrum for the thermal SZ effect (solid, black) 
496: and the radio point sources -- model 1 (red, dashed), model 2 (blue, long-dashed). The model 2 of the radio point sources 
497: has the largest cross correlation because it predicts that the point sources tend to be located at higher redshift. As 
498: opposed to the matter-ISW cross-correlation, which requires that the sources lie in the same redshift range over which the 
499: ISW effect occurs, in this case the point source do not have to be at the same redshift at which the  ISW
500: occurs  in order for the contribution to be relevant. In fact the point sources 
501: experience greater magnification if they are at substantially higher redshift than the matter distribution that is 
502: simultaneously magnifying them and is correlated with the ISW effect. 
503: 
504: \begin{figure}
505: \centering
506: \includegraphics[width=10cm,height=10cm]{mag.eps}
507: \caption{\label{fig:mag_power} Convergence-ISW cross-correlation spectrum for thermal SZ (solid, black); radio point 
508: sources -- model 1 (red, dashed) and model 2 (blue, long dashed)}
509: \end{figure}
510: 
511: In Fig. \ref{fig:mag_bispect} we show the magnification modulation bispectrum for the thermal SZ effect (solid, black);
512: radio point sources -- model 1 (red, dashed) and model 2 (blue, long-dashed). The local model bispectrum ($f_{NL} = 1$) 
513: (green, dot-dashed) is also shown for reference.
514: 
515: \begin{figure}
516: \centering
517: \includegraphics[width=10cm,height=10cm]{bispect_m.eps}
518: \caption{\label{fig:mag_bispect} Magnification modulation bispectrum for thermal SZ effect (solid, black); radio point 
519: sources -- model 1 (red, dashed) and model 2 (blue, long-dashed). The local model bispectrum ($f_{NL} = 1$) --
520: equilateral shape $(\ell_1 = \ell_2 = \ell_3 = \ell)$ (green, dot-dashed) is also shown for reference.}
521: \end{figure}
522: 
523: \subsection{Selection Modulation}
524: 
525: The amplitude of the Poisson fluctuation power spectrum depends on the number density of radio point sources 
526: below some flux limit determined by the radio point source removal technique. If the selection criterion used
527: produces an anisotropic flux limit that is correlated with either the large scale temperature anisotropy
528: or the instrument noise, then a bispectrum will be produced. If the radio point sources are identified via external 
529: catalogs, such as NVSS or PMN, then there will be no cross-correlation and therefore no bispectrum. And if
530: multi-wavelength data is differenced to isolate the power-law frequency dependence of the radio point source, 
531: then the modulated selection function will produce correlations between the data in different frequency bands.
532: 
533: The simplest method is to removal pixels above some multiple ($\gamma$) of the pixel variance
534: \begin{equation}
535: \bar{S}(\hat{\bm n}) = c_{\nu}\left[\gamma \sigma_p - \Delta \Omega \left( \frac{\Delta T}{T}(\hat{\bm n}) 
536: + N(\hat{\bm n}) \right)\right]
537: \end{equation}
538: The temperature anisotropy and instrument noise have vanishing expectation values, so the mean flux cut-off is
539: \begin{equation}
540: \bar{S}_0 = c_{\nu} \gamma \sigma_p,
541: \end{equation}
542: where the pixel variance is defined as
543: \begin{equation}
544: \sigma^2_p = (\Delta \Omega)^2 \sum_{\ell} \frac{(2\ell + 1)}{4\pi} C^T_{\ell},
545: \end{equation}
546: here $C^T_{\ell} = C_{\ell} + C^N_{\ell}$ is the sum of the CMB signal and noise and $\Delta \Omega$ is the pixel 
547: size. There will be fluctuations about the expected flux cut-off that are correlated with either the temperature 
548: anisotropy or the instrument noise. Linearizing in these fluctuations, we find the following reduced bispectrum
549: \begin{equation}
550: b_{\ell_1, \ell_2, \ell_3} = - \frac{\Delta \Omega}{c_{\nu}} (C^T_{\ell_1} + {\rm cyc.}) \bar{S}^2_0 \frac{dN}{dS}(\bar{S},\nu). 
551: \end{equation}
552: 
553: The selection criterion used by WMAP is based on an algorithm developed by Tegmark \& de Oliveira-Costa \cite{Tegmark98}.
554: The map is filtered in order to reduce the importance of the long-wavelength CMB modes. The filtered total temperature
555: fluctuation in some pixel is
556: \begin{equation}
557: y(\hat{\bm n}) = \sum_{\ell m} \frac{1}{C^T_{\ell}} Y_{\ell m}(\hat{\bm n}) (a_{\ell m} + n_{\ell m}) \Delta \Omega 
558: + \sum_{\ell} \frac{(2\ell + 1)}{4\pi} \frac{S(\hat{\bm n})}{c_{\nu} C^T_{\ell}}.
559: \end{equation}
560: The algorithm removes any pixel that has a value greater than some multiple ($\gamma$) of the filtered map pixel variance
561: \begin{equation}
562: \tilde{\sigma}^{-2}_p = (\Delta \Omega)^{-2} \sum_{\ell} \frac{(2\ell + 1)}{4\pi} \frac{1}{C^T_{\ell}}.
563: \end{equation}
564: The threshold $\gamma$ is chosen according to some compromise between false positive and negatives.
565: Since the thresholding is applied to the total signal in a pixel, the corresponding radio point source flux 
566: cut-off in a pixel is
567: \begin{equation}
568: \bar{S}(\hat{\bm n}) = \frac{c_{\nu}}{F} \left[ \gamma \tilde{\sigma}_p 
569: - \Delta \Omega \sum_{\ell m} \frac{1}{C^T_{\ell}} Y_{\ell m}(\hat{\bm n}) (a_{\ell m} + n_{\ell m}) \right],
570: \end{equation}
571: where the normalization is
572: \begin{equation}
573: F = \sum_{\ell} \frac{(2\ell + 1)}{4\pi} \frac{1}{C^T_{\ell}}.
574: \end{equation}
575: The temperature anisotropy and instrument noise have vanishing expectation values, so the mean flux cut-off is
576: \begin{equation}
577: \bar{S}_0 = \frac{c_{\nu}}{F} \gamma \tilde{\sigma}_p.
578: \end{equation}
579: There will be fluctuations about this expected value and these fluctuations will be correlated with either the
580: temperature anisotropy or the instrument noise. Linearizing in these fluctuations, we find the following reduced 
581: bispectrum
582: \begin{equation}\label{eq:select}
583: b_{\ell_1, \ell_2, \ell_3} = - \frac{3 \Delta \Omega}{c_{\nu} F}  \bar{S}^2_0 \frac{dN}{dS} (\nu, \bar{S}_0). 
584: \end{equation}
585: In regions with large instrument noise or CMB temperature anisotropy, the radio point source flux cut-off will be 
586: lowered. This reduces the total number density of radio point sources in that region and therefore the Poisson 
587: fluctuation power spectrum. This effect explains the negative sign in the reduced bispectrum, Eq. \ref{eq:select}.
588: 
589: Since the filtering applied to the maps produces a bispectrum independent of scale, similar to the point source 
590: bispectrum described in \S \ref{sec:ps_bisp}, we will ignore it. In reality the actual selection function
591: is more complicate than this simple filter technique predicts and it might produce a bispectrum of a much different form.
592: Numerical simulations incorporating the exact selection procedure will need to be done in order to fully determine
593: its effect on the estimator.
594: 
595: \section{Numerical Results}\label{sec:results}
596: 
597: In this section we will present numerical results for the contamination of the standard non-Gaussianity estimator 
598: by the various bispectra discussed in this paper. This is done for both the WMAP and Planck instrument noise levels, 
599: frequency bands and flux cut-offs given in Table \ref{table:exp_info}.
600: 
601: In Fig. \ref{fig:wmap} the $f_{NL}$ estimator bias as a function of $\ell_{max}$ is shown for the different bispectra 
602: -- radio point source number density modulation (solid, black); SZ number density modulation (dotted, red); radio point 
603: source gravitational lensing magnification modulation (dashed, blue); and SZ gravitational lensing magnification 
604: modulation (long-dashed, green) with WMAP instrument noise. The bias plots are shown for the four relevant WMAP frequency
605: bands -- upper left Ka -- $33~{\rm GHz}$; upper right Q -- $40~{\rm GHz}$; lower left V -- $61~{\rm GHz}$; 
606: lower right W -- $94~{\rm GHz}$. The magnification modulation effect produces a positive bias since its bispectrum is 
607: negative, while the density modulation effect produces a negative bias. 
608: 
609: \begin{figure}
610: \centering
611: \includegraphics{wmap_bias.eps}
612: \caption{\label{fig:wmap} The WMAP estimator bias terms $\Delta f^{\alpha}_{NL}$ as a function of $\ell_{max}$ for radio point 
613: source density modulation (solid, black); SZ number density modulation (dotted, red); radio point source gravitational lensing 
614: magnification modulation (dashed, blue); and SZ gravitational lensing magnification modulation (long-dashed, green) in
615: Ka -- $33~{\rm GHz}$ (upper left); Q -- $40~{\rm GHz}$ (upper right); V -- $61~{\rm GHz}$; and W -- $94~{\rm GHz}$. The density
616: modulation terms produce a negative bias, while the magnification bias produce a positive bias.}
617: \end{figure}
618: 
619: Yadav \& Wandelt claim a central value of $f_{NL} = 86.8$ with a standard deviation of $\sigma = 30.0$ for a $2.9 \sigma$
620: detection of non-Gaussianity \cite{Yadav07}. In the five-year WMAP data the central value is found to be $f_{NL} = 67$ with
621: a standard deviation of $\sigma = 31$ \cite{Komatsu08}.
622: At $\ell_{max} = 750$ the total estimator bias $\Delta f_{NL} = 0.35$ in the Ka band, $\Delta f_{NL} = 0.24$ in the Q band,
623: $\Delta f_{NL} = -0.097$ in the V band and $\Delta f_{NL} = -0.13$ in W band. Since the density modulation and the magnification 
624: modulation bispectra have different signs they partially cancel and reduce the overall effect. At low frequency the radio point
625: source magnification modulation bispectrum is the most important so the bias is positive. At higher frequencies the SZ density
626: modulation bispectrum dominates which makes the bias negative. These numbers should be compared to the estimator bias produced 
627: by the radio point source Poisson fluctuation bispectrum. Komatsu et al \cite{Komatsu08} have estimated this bias to be 
628: $\Delta f_{NL} \simeq -(3-5)$ at $\ell_{max} = 700$.
629: 
630: In Fig. \ref{fig:planck} the $f_{NL}$ estimator bias as a function of $\ell_{max}$ is shown for the different bispectra 
631: -- radio point source number density modulation (solid, black); SZ number density modulation (dotted, red); radio point 
632: source gravitational lensing magnification modulation (dashed, blue); and SZ gravitational lensing magnification modulation 
633: (long-dashed, green) for Planck at $30~{\rm GHz}$ (upper left); $44~{\rm GHz}$ (upper left); $70~{\rm GHz}$ (lower left) 
634: and $100~{\rm GHz}$ (lower right).
635: 
636: \begin{figure}
637: \centering
638: \includegraphics{planck_bias.eps}
639: \caption{\label{fig:planck} The Planck estimator bias terms $\Delta f^{\alpha}_{NL}$ for $30~{\rm GHz}$ (upper left);
640: $44~{\rm GHz}$ (upper left); $70~{\rm GHz}$ (lower left) and $100~{\rm GHz}$ (lower right). The curves are the same 
641: as Fig. \ref{fig:wmap}. The density modulation terms produce a negative bias, while the magnification bias produce a positive bias.}
642: \end{figure}
643: 
644: Estimates of the sensitivity on $f_{NL}$ achievable with Planck suggest that $\Delta f_{NL} \simeq 10$ at 95 \% C.L.using just 
645: temperature information and $\Delta f_{NL} \simeq 5$ at 95 \% C.L also including polarization. Summing the various biases we find
646: $\Delta f_{NL} = 1.3$ at $\nu = 30$ GHz, $\Delta f_{NL} = 0.34$ at $\nu = 44$ GHz, $\Delta f_{NL} = -0.25$ at $\nu = 70$ GHz
647: and $\Delta f_{NL} = -0.48$ at $\nu = 100$ GHz. These results imply that a good knowledge of point source
648: properties is important if Planck will be able to achieve its full potential in constraining primordial non-Gaussianity.
649: 
650: There are uncertainties in the models we have used to describe the radio point sources. These model uncertainities will 
651: directly lead to uncertainities in the above predictions. As can be directly seen in Figs. \ref{fig:cross_power} -- 
652: \ref{fig:mag_bispect}, the differences between the radio point source redshift distributions, model 1 and model 2, are 
653: not significant. This is not surprising as the ISW kernel, with which the radio point source redshift distributions are 
654: being cross-correlated, is quite broad. The redshift distributions are supposed to trace both the high--flux source 
655: populations at low frequencies. However, they may not be representative of the source populations at all frequencies 
656: considered here for lower flux cut thresholds. If a lower flux cut implies a higher population of low--redshift objects, 
657: the amplitude of the ISW-density cross-correlation spectrum would be increased. The decrease in the flux cut-off will 
658: decrease the Poisson fluctuation power spectrum, so the change in the bias of the $f_{NL}$ estimator is not clear. We 
659: also note that reducing the flux cut--off from the WMAP to the Planck level does not reduce the bias implied on $f_{NL}$ 
660: as the Planck noise levels and beam sizes are also smaller. 
661: 
662: \section{Conclusion}\label{sec:conc}
663: 
664: In this paper we analyzed the effect of point sources, both due to radio emission and the cluster SZ effect, on the 
665: estimation of primordial non-Gaussianity in the cosmic microwave background. The standard non-Gaussianity estimator is 
666: sensitive to any bispectrum present in the data. In addition to the standard Poisson fluctuation bispectrum, we found 
667: that cross-correlations between the radio point source and SZ power spectra and either the CMB temperature anisotropies 
668: or instrument noise can produce bispectra. These bispectra have forms somewhat similar to the local model, which is the 
669: standard bispectrum form used to search for primordial non-Gaussianity in CMB data. These similarities are not accidental, 
670: but occur because the same basic principle generates the non-Gaussianity in these cases. Due to this similarity it will 
671: be much more difficult to distinguish this non-Gaussianity from the primordial signal than other secondary bispectra with 
672: shapes which can be quite different.
673:  
674: A related paper by Serra \& Cooray \cite{Serra08} has examined different secondary bispectra, but has reached similar conclusions. 
675: They examined the bispectra produced by the cross-correlation of the thermal SZ effect and the gravitational lensing of the 
676: primary CMB anisotropies. They concluded that the effects are too small to account for the infered values of $f_{NL}$ from
677: the WMAP data and will start to become important for the Planck dataset.
678: 
679: The estimator bias that we have calculated is quite small and is not able to explain the results found by Yadav \& Wandelt,
680: although the bispectra considered in our paper will start to become important for the Planck non-Gaussianity analysis. 
681: Due to the extremely important nature of the detection of primordial non-Gaussianity in the WMAP data, all possible 
682: alternative explanations should be considered. 
683: 
684: \begin{acknowledgments}
685: DB acknowledges financial support from the Betty and Gordon Moore Foundation and would like to thank Sterl Phinney, 
686: Daisuke Nagai and Eiichiro Komatsu for helpful conversations. EP is and NSF--ADVANCE fellow (AST--0649899) also supported 
687: by NASA grant NNX07AH59G, Planck subcontract 1290790 and JPL SURP award 1314616.
688: She would also like to thank Kevin Huffenberger and Joachin Gonzales-Nuevo for useful conversations.
689: We also thank Kendrick Smith for helpful comments on a draft of the paper.
690: \end{acknowledgments}
691: 
692: %\bibliography{elebib}
693: 
694: \begin{thebibliography}{27}
695: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
696: \expandafter\ifx\csname bibnamefont\endcsname\relax
697:   \def\bibnamefont#1{#1}\fi
698: \expandafter\ifx\csname bibfnamefont\endcsname\relax
699:   \def\bibfnamefont#1{#1}\fi
700: \expandafter\ifx\csname citenamefont\endcsname\relax
701:   \def\citenamefont#1{#1}\fi
702: \expandafter\ifx\csname url\endcsname\relax
703:   \def\url#1{\texttt{#1}}\fi
704: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
705: \providecommand{\bibinfo}[2]{#2}
706: \providecommand{\eprint}[2][]{\url{#2}}
707: 
708: \bibitem[{\citenamefont{{Yadav} and {Wandelt}}(2007)}]{Yadav07}
709: \bibinfo{author}{\bibfnamefont{A.~P.~S.} \bibnamefont{{Yadav}}}
710:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{B.~D.}
711:   \bibnamefont{{Wandelt}}}, \bibinfo{journal}{ArXiv e-prints}
712:   \textbf{\bibinfo{volume}{712}} (\bibinfo{year}{2007}), \eprint{0712.1148}.
713: 
714: \bibitem[{\citenamefont{{Spergel} et~al.}(2007)\citenamefont{{Spergel}, {Bean},
715:   {Dor{\'e}}, {Nolta}, {Bennett}, {Dunkley}, {Hinshaw}, {Jarosik}, {Komatsu},
716:   {Page} et~al.}}]{Spergel07}
717: \bibinfo{author}{\bibfnamefont{D.~N.} \bibnamefont{{Spergel}}},
718:   \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{{Bean}}},
719:   \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{{Dor{\'e}}}},
720:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{{Nolta}}},
721:   \bibinfo{author}{\bibfnamefont{C.~L.} \bibnamefont{{Bennett}}},
722:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Dunkley}}},
723:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{Hinshaw}}},
724:   \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{{Jarosik}}},
725:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Komatsu}}},
726:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Page}}},
727:   \bibnamefont{et~al.}, \bibinfo{journal}{ApJS} \textbf{\bibinfo{volume}{170}},
728:   \bibinfo{pages}{377} (\bibinfo{year}{2007}), \eprint{arXiv:astro-ph/0603449}.
729: 
730: \bibitem[{\citenamefont{{Maldacena}}(2003)}]{Maldacena03}
731: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Maldacena}}},
732:   \bibinfo{journal}{Journal of High Energy Physics}
733:   \textbf{\bibinfo{volume}{5}}, \bibinfo{pages}{13} (\bibinfo{year}{2003}),
734:   \eprint{arXiv:astro-ph/0210603}.
735: 
736: \bibitem[{\citenamefont{{Komatsu}}(2002)}]{Komatsu02}
737: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Komatsu}}},
738:   \bibinfo{journal}{ArXiv Astrophysics e-prints}  (\bibinfo{year}{2002}),
739:   \eprint{astro-ph/0206039}.
740: 
741: \bibitem[{\citenamefont{{Babich}}(2005)}]{Babich05}
742: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{{Babich}}},
743:   \bibinfo{journal}{\prd} \textbf{\bibinfo{volume}{72}},
744:   \bibinfo{pages}{043003} (\bibinfo{year}{2005}),
745:   \eprint{arXiv:astro-ph/0503375}.
746: 
747: \bibitem[{\citenamefont{{Creminelli} et~al.}(2006)\citenamefont{{Creminelli},
748:   {Nicolis}, {Senatore}, {Tegmark}, and {Zaldarriaga}}}]{Creminelli06}
749: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{{Creminelli}}},
750:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{{Nicolis}}},
751:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Senatore}}},
752:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Tegmark}}},
753:   \bibnamefont{and}
754:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Zaldarriaga}}},
755:   \bibinfo{journal}{Journal of Cosmology and Astro-Particle Physics}
756:   \textbf{\bibinfo{volume}{5}}, \bibinfo{pages}{4} (\bibinfo{year}{2006}),
757:   \eprint{arXiv:astro-ph/0509029}.
758: 
759: \bibitem[{\citenamefont{{Hinshaw} et~al.}(2007)\citenamefont{{Hinshaw},
760:   {Nolta}, {Bennett}, {Bean}, {Dor{\'e}}, {Greason}, {Halpern}, {Hill},
761:   {Jarosik}, {Kogut} et~al.}}]{Hinshaw07}
762: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{Hinshaw}}},
763:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{{Nolta}}},
764:   \bibinfo{author}{\bibfnamefont{C.~L.} \bibnamefont{{Bennett}}},
765:   \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{{Bean}}},
766:   \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{{Dor{\'e}}}},
767:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{{Greason}}},
768:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Halpern}}},
769:   \bibinfo{author}{\bibfnamefont{R.~S.} \bibnamefont{{Hill}}},
770:   \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{{Jarosik}}},
771:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{{Kogut}}},
772:   \bibnamefont{et~al.}, \bibinfo{journal}{ApJS} \textbf{\bibinfo{volume}{170}},
773:   \bibinfo{pages}{288} (\bibinfo{year}{2007}), \eprint{arXiv:astro-ph/0603451}.
774: 
775: \bibitem[{\citenamefont{{Pierpaoli}}(2003)}]{Pierpa03}
776: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Pierpaoli}}},
777:   \bibinfo{journal}{\apj} \textbf{\bibinfo{volume}{589}}, \bibinfo{pages}{58}
778:   (\bibinfo{year}{2003}), \eprint{arXiv:astro-ph/0301563}.
779: 
780: \bibitem[{\citenamefont{{L{\'o}pez-Caniego}
781:   et~al.}(2007)\citenamefont{{L{\'o}pez-Caniego}, {Gonz{\'a}lez-Nuevo},
782:   {Herranz}, {Massardi}, {Sanz}, {De Zotti}, {Toffolatti}, and
783:   {Arg{\"u}eso}}}]{LopezCaniego07}
784: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{L{\'o}pez-Caniego}}},
785:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Gonz{\'a}lez-Nuevo}}},
786:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{{Herranz}}},
787:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Massardi}}},
788:   \bibinfo{author}{\bibfnamefont{J.~L.} \bibnamefont{{Sanz}}},
789:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{De Zotti}}},
790:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Toffolatti}}},
791:   \bibnamefont{and}
792:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{{Arg{\"u}eso}}},
793:   \bibinfo{journal}{ApJS} \textbf{\bibinfo{volume}{170}}, \bibinfo{pages}{108}
794:   (\bibinfo{year}{2007}), \eprint{arXiv:astro-ph/0701473}.
795: 
796: \bibitem[{\citenamefont{{Gonz{\'a}lez-Nuevo}
797:   et~al.}(2008)\citenamefont{{Gonz{\'a}lez-Nuevo}, {Massardi}, {Arg{\"u}eso},
798:   {Herranz}, {Toffolatti}, {Sanz}, {L{\'o}pez-Caniego}, and {de
799:   Zotti}}}]{Gonzalez08}
800: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Gonz{\'a}lez-Nuevo}}},
801:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Massardi}}},
802:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{{Arg{\"u}eso}}},
803:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{{Herranz}}},
804:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Toffolatti}}},
805:   \bibinfo{author}{\bibfnamefont{J.~L.} \bibnamefont{{Sanz}}},
806:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{L{\'o}pez-Caniego}}},
807:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{de
808:   Zotti}}}, \bibinfo{journal}{MNRAS} pp. \bibinfo{pages}{48--+}
809:   (\bibinfo{year}{2008}), \eprint{arXiv:0711.2631}.
810: 
811: \bibitem[{\citenamefont{{Chen} and {Wright}}(2007)}]{ChenWright07}
812: \bibinfo{author}{\bibfnamefont{X.}~\bibnamefont{{Chen}}} \bibnamefont{and}
813:   \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{{Wright}}},
814:   \bibinfo{journal}{ArXiv e-prints} \textbf{\bibinfo{volume}{712}}
815:   (\bibinfo{year}{2007}), \eprint{0712.3594}.
816: 
817: \bibitem[{\citenamefont{{Huffenberger}
818:   et~al.}(2006)\citenamefont{{Huffenberger}, {Eriksen}, and
819:   {Hansen}}}]{Huffenberger06}
820: \bibinfo{author}{\bibfnamefont{K.~M.} \bibnamefont{{Huffenberger}}},
821:   \bibinfo{author}{\bibfnamefont{H.~K.} \bibnamefont{{Eriksen}}},
822:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{F.~K.}
823:   \bibnamefont{{Hansen}}}, \bibinfo{journal}{ApJL}
824:   \textbf{\bibinfo{volume}{651}}, \bibinfo{pages}{L81} (\bibinfo{year}{2006}),
825:   \eprint{arXiv:astro-ph/0606538}.
826: 
827: \bibitem[{\citenamefont{{L{\'o}pez-Caniego}
828:   et~al.}(2006)\citenamefont{{L{\'o}pez-Caniego}, {Herranz},
829:   {Gonz{\'a}lez-Nuevo}, {Sanz}, {Barreiro}, {Vielva}, {Arg{\"u}eso}, and
830:   {Toffolatti}}}]{LopezCarniego06}
831: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{L{\'o}pez-Caniego}}},
832:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{{Herranz}}},
833:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Gonz{\'a}lez-Nuevo}}},
834:   \bibinfo{author}{\bibfnamefont{J.~L.} \bibnamefont{{Sanz}}},
835:   \bibinfo{author}{\bibfnamefont{R.~B.} \bibnamefont{{Barreiro}}},
836:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{{Vielva}}},
837:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{{Arg{\"u}eso}}},
838:   \bibnamefont{and}
839:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Toffolatti}}},
840:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{370}},
841:   \bibinfo{pages}{2047} (\bibinfo{year}{2006}),
842:   \eprint{arXiv:astro-ph/0606199}.
843: 
844: \bibitem[{\citenamefont{{Toffolatti} et~al.}(1998)\citenamefont{{Toffolatti},
845:   {Argueso Gomez}, {de Zotti}, {Mazzei}, {Franceschini}, {Danese}, and
846:   {Burigana}}}]{Toffolatti98}
847: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Toffolatti}}},
848:   \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{{Argueso Gomez}}},
849:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{de Zotti}}},
850:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{{Mazzei}}},
851:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{{Franceschini}}},
852:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Danese}}}, \bibnamefont{and}
853:   \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{{Burigana}}},
854:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{297}}, \bibinfo{pages}{117}
855:   (\bibinfo{year}{1998}), \eprint{arXiv:astro-ph/9711085}.
856: 
857: \bibitem[{\citenamefont{{Smith} et~al.}(2007)\citenamefont{{Smith}, {Zahn}, and
858:   {Dor{\'e}}}}]{Smith07}
859: \bibinfo{author}{\bibfnamefont{K.~M.} \bibnamefont{{Smith}}},
860:   \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{{Zahn}}}, \bibnamefont{and}
861:   \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{{Dor{\'e}}}},
862:   \bibinfo{journal}{\prd} \textbf{\bibinfo{volume}{76}},
863:   \bibinfo{pages}{043510} (\bibinfo{year}{2007}), \eprint{arXiv:0705.3980}.
864: 
865: \bibitem[{\citenamefont{{Blake} et~al.}(2004)\citenamefont{{Blake}, {Ferreira},
866:   and {Borrill}}}]{Blake04}
867: \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{{Blake}}},
868:   \bibinfo{author}{\bibfnamefont{P.~G.} \bibnamefont{{Ferreira}}},
869:   \bibnamefont{and}
870:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Borrill}}},
871:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{351}}, \bibinfo{pages}{923}
872:   (\bibinfo{year}{2004}), \eprint{arXiv:astro-ph/0404085}.
873: 
874: \bibitem[{\citenamefont{{Boughn} and {Crittenden}}(2002)}]{Boughn02}
875: \bibinfo{author}{\bibfnamefont{S.~P.} \bibnamefont{{Boughn}}} \bibnamefont{and}
876:   \bibinfo{author}{\bibfnamefont{R.~G.} \bibnamefont{{Crittenden}}},
877:   \bibinfo{journal}{Physical Review Letters} \textbf{\bibinfo{volume}{88}},
878:   \bibinfo{pages}{021302} (\bibinfo{year}{2002}),
879:   \eprint{arXiv:astro-ph/0111281}.
880: 
881: \bibitem[{\citenamefont{{Sadler} et~al.}(2007)\citenamefont{{Sadler}, {Ricci},
882:   {Ekers}, {Sault}, {Jackson}, and {De Zotti}}}]{Sadler07}
883: \bibinfo{author}{\bibfnamefont{E.~M.} \bibnamefont{{Sadler}}},
884:   \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{{Ricci}}},
885:   \bibinfo{author}{\bibfnamefont{R.~D.} \bibnamefont{{Ekers}}},
886:   \bibinfo{author}{\bibfnamefont{R.~J.} \bibnamefont{{Sault}}},
887:   \bibinfo{author}{\bibfnamefont{C.~A.} \bibnamefont{{Jackson}}},
888:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{De
889:   Zotti}}}, \bibinfo{journal}{ArXiv e-prints} \textbf{\bibinfo{volume}{709}}
890:   (\bibinfo{year}{2007}), \eprint{0709.3563}.
891: 
892: \bibitem[{\citenamefont{{Pierpaoli} and {Perna}}(2004)}]{PierpaPerna04}
893: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Pierpaoli}}} \bibnamefont{and}
894:   \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{{Perna}}},
895:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{354}},
896:   \bibinfo{pages}{1005} (\bibinfo{year}{2004}),
897:   \eprint{arXiv:astro-ph/0407561}.
898: 
899: \bibitem[{\citenamefont{{de Zotti} et~al.}(2005)\citenamefont{{de Zotti},
900:   {Ricci}, {Mesa}, {Silva}, {Mazzotta}, {Toffolatti}, and
901:   {Gonz{\'a}lez-Nuevo}}}]{DeZotti05}
902: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{de Zotti}}},
903:   \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{{Ricci}}},
904:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{{Mesa}}},
905:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Silva}}},
906:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{{Mazzotta}}},
907:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Toffolatti}}},
908:   \bibnamefont{and}
909:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Gonz{\'a}lez-Nuevo}}},
910:   \bibinfo{journal}{AA} \textbf{\bibinfo{volume}{431}}, \bibinfo{pages}{893}
911:   (\bibinfo{year}{2005}), \eprint{arXiv:astro-ph/0410709}.
912: 
913: \bibitem[{\citenamefont{{Carlstrom} et~al.}(2002)\citenamefont{{Carlstrom},
914:   {Holder}, and {Reese}}}]{Carlstrom02}
915: \bibinfo{author}{\bibfnamefont{J.~E.} \bibnamefont{{Carlstrom}}},
916:   \bibinfo{author}{\bibfnamefont{G.~P.} \bibnamefont{{Holder}}},
917:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.~D.}
918:   \bibnamefont{{Reese}}}, \bibinfo{journal}{ARAA}
919:   \textbf{\bibinfo{volume}{40}}, \bibinfo{pages}{643} (\bibinfo{year}{2002}),
920:   \eprint{arXiv:astro-ph/0208192}.
921: 
922: \bibitem[{\citenamefont{{Komatsu} and {Seljak}}(2002)}]{Komatsu02a}
923: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Komatsu}}} \bibnamefont{and}
924:   \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{{Seljak}}},
925:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{336}},
926:   \bibinfo{pages}{1256} (\bibinfo{year}{2002}),
927:   \eprint{arXiv:astro-ph/0205468}.
928: 
929: \bibitem[{\citenamefont{{Komatsu} and {Seljak}}(2001)}]{Komatsu01}
930: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Komatsu}}} \bibnamefont{and}
931:   \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{{Seljak}}},
932:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{327}},
933:   \bibinfo{pages}{1353} (\bibinfo{year}{2001}),
934:   \eprint{arXiv:astro-ph/0106151}.
935: 
936: \bibitem[{\citenamefont{{Sheth} and {Tormen}}(1999)}]{Sheth99}
937: \bibinfo{author}{\bibfnamefont{R.~K.} \bibnamefont{{Sheth}}} \bibnamefont{and}
938:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{Tormen}}},
939:   \bibinfo{journal}{MNRAS} \textbf{\bibinfo{volume}{308}}, \bibinfo{pages}{119}
940:   (\bibinfo{year}{1999}), \eprint{arXiv:astro-ph/9901122}.
941: 
942: \bibitem[{\citenamefont{{Nagai}}(2006)}]{Nagai06}
943: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{{Nagai}}},
944:   \bibinfo{journal}{\apj} \textbf{\bibinfo{volume}{650}}, \bibinfo{pages}{538}
945:   (\bibinfo{year}{2006}), \eprint{arXiv:astro-ph/0512208}.
946: 
947: \bibitem[{\citenamefont{{Komatsu} et~al.}(2003)\citenamefont{{Komatsu},
948:   {Kogut}, {Nolta}, {Bennett}, {Halpern}, {Hinshaw}, {Jarosik}, {Limon},
949:   {Meyer}, {Page} et~al.}}]{Komatsu03}
950: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Komatsu}}},
951:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{{Kogut}}},
952:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{{Nolta}}},
953:   \bibinfo{author}{\bibfnamefont{C.~L.} \bibnamefont{{Bennett}}},
954:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Halpern}}},
955:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{Hinshaw}}},
956:   \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{{Jarosik}}},
957:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Limon}}},
958:   \bibinfo{author}{\bibfnamefont{S.~S.} \bibnamefont{{Meyer}}},
959:   \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{{Page}}},
960:   \bibnamefont{et~al.}, \bibinfo{journal}{ApJS} \textbf{\bibinfo{volume}{148}},
961:   \bibinfo{pages}{119} (\bibinfo{year}{2003}), \eprint{arXiv:astro-ph/0302223}.
962: 
963: \bibitem[{\citenamefont{{Tegmark} and {de Oliveira-Costa}}(1998)}]{Tegmark98}
964: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Tegmark}}} \bibnamefont{and}
965:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{{de Oliveira-Costa}}},
966:   \bibinfo{journal}{ApJL} \textbf{\bibinfo{volume}{500}}, \bibinfo{pages}{L83+}
967:   (\bibinfo{year}{1998}), \eprint{arXiv:astro-ph/9802123}.
968: 
969: \bibitem[{\citenamefont{{Serra} and {Cooray}}(2008)}]{Serra08}
970: \bibinfo{author}{\bibfnamefont{P.} \bibnamefont{{Serra}}} \bibnamefont{and}
971:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{{Cooray}}},
972:   \bibinfo{journal}{.} \textbf{\bibinfo{volume}{.}}, \bibinfo{pages}{.}
973:   (\bibinfo{year}{2008}), \eprint{arXiv0801.3276}.
974: 
975: \bibitem[{\citenamefont{{Komatsu} et~al.}(2008)\citenamefont{{Komatsu},
976:   {Dunkley}, {Nolta}, {Bennett}, {Gold}, {Hinshaw}, {Jarosik}, {Limon},
977:   {Larson},  et~al.}}]{Komatsu08}
978: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{{Komatsu}}},
979:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{{Dunkely}}},
980:   \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{{Nolta}}},
981:   \bibinfo{author}{\bibfnamefont{C.~L.} \bibnamefont{{Bennett}}},
982:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Gold}}},
983:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{{Hinshaw}}},
984:   \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{{Jarosik}}},
985:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{{Limon}}},
986:   \bibinfo{author}{\bibfnamefont{D.} \bibnamefont{{Larson}}},
987:   \bibnamefont{et~al.}, \bibinfo{journal}{ApJS} \textbf{\bibinfo{volume}{.}},
988:   \bibinfo{pages}{.} (\bibinfo{year}{2000}), \eprint{arXiv:astro-ph/0803.0547}.
989: 
990: 
991: 
992: \end{thebibliography}
993: 
994: \end{document}
995: