1: %\documentclass[aps,floatfix,twocolumn,groupedaddress,nobalancelastpage]{revtex4}
2: %\documentclass[preprint2]{emulateapj}
3: %\documentclass[aps,prl,galley]{revtex4}
4: \documentclass[aps,preprint]{revtex4}
5: \usepackage{graphicx}% Include figure files
6: \usepackage{dcolumn}% Align table columns on decimal point
7: \usepackage{bm}% bold math
8: \usepackage{hyperref}
9: \usepackage{subfigure}
10: \usepackage[sort&compress]{natbib}
11: \begin{document}
12:
13:
14: \title{Crinkles in the last scattering surface: Non-Gaussianity from
15: inhomogeneous recombination.}
16: \author{Rishi Khatri}
17: \email{rkhatri2@illinois.edu}
18: \affiliation{Department of Astronomy, University of Illinois at Urbana-Champaign, 1002 W.~Green Street, Urbana, IL 61801}
19: \author{Benjamin D. Wandelt}
20: \email{bwandelt@illinois.edu}
21: \affiliation{Departments of Physics and Astronomy, University of Illinois at Urbana-Champaign, 1002 W.~Green Street, Urbana, IL 61801}
22:
23: \date{\today}
24: \begin{abstract}
25: The perturbations in the electron number density during recombination
26: contributes to the cosmic microwave background bispectrum through second
27: order terms. Perturbations in the electron density can be a factor of
28: $\sim 5$ larger than the baryon density fluctuations on large scales as
29: shown in the calculations by Novosyadlyj. This raises the possibility that
30: the contribution to the bispectrum arising from perturbations in the optical
31: depth may be non-negligible. We calculate this bispectrum and find it to
32: peak for squeezed triangles and of peak amplitude of the order of
33: primordial non-Gaussianity of local type with $f_{NL}\approx 0.05 \sim -1$
34: depending on the $\ell$-modes being considered. This is because the shape of
35: the bispectrum is different from the primordial one although it peaks
36: for squeezed configurations, similar to the local type primordial
37: non-Gaussianity.
38: \end{abstract}
39: %\pacs{06.20.Jr,98.80.-k}
40: \maketitle
41: \begin{section}{Introduction}
42: First order perturbation theory has been of sufficient accuracy for
43: analysis of the Cosmic Microwave
44: Background (CMB) observations so far. However future CMB experiments will
45: have high enough precision that second order effects would need to be taken
46: into account for theory to have similar accuracy. The second order
47: contributions will in particular be important for the higher order statistics
48: like the three point correlation or the bispectrum. Second order effects
49: in CMB have been studied previously \cite{0,1,2,3,4,5,6,7,8,9,10,11,12}.
50: Bartolo et al. have derived the Boltzmann equations at second order and
51: also the analytic solutions for the CMB transfer function at second order
52: with some simplifying assumptions \cite{bar1,bar2}, see also \cite{pitrou}.
53:
54: All numerical and analytic calculations at second order so far have
55: ignored the contribution arising from the perturbations in the electron
56: number density, $\delta_e$. These contributions are expected to be small
57: compared to other second order terms since $\delta_e$ multiplies the
58: collision term which has contributions from the difference of first order
59: radiation and electron dipoles , radiation quadrupole and higher
60: order moments of the radiation transfer function. These terms are small
61: during recombination compared to the monopole terms.
62: Recombination however depends on matter and radiation densities and
63: perturbations in the
64: electron number density can be quite different from the perturbations in
65: the matter and radiation densities. This was calculated by Novosyadlyj
66: \cite{novos} who showed that this is indeed the case and perturbations in
67: the electron number density can be $\sim 5$ times the baryon number density
68: perturbations.
69:
70: We calculate the CMB bispectrum on all scales
71: arising due to the perturbations
72: in electron number density and compare
73: it with the bispectrum expected from a primordial non-Gaussianity of the
74: local type. This is a full numerical calculation without any other
75: approximation except that we only consider terms involving
76: $\delta_e$. Although this bispectrum turns out to be below the
77: detection levels of future experiments like Planck \cite{planck}, there are
78: some important general implications which are discussed in the conclusions
79: section. We use the following cosmological parameters for our calculations
80: (values at redshift $z=0$ unless specified):
81: baryon density $\Omega_b=0.0418$, cold dark matter density
82: $\Omega_c=0.19647$, cosmological constant $\Omega_{\Lambda}=0.76173$,
83: number of massless neutrinos $N_{\nu}=3.04$, Hubble constant $H_0=73$, CMB
84: temperature $T_{CMB}=2.725$, primordial Helium fraction $y_{He}=0.24$,
85: redshift of reionization $z_{ri}=10$, primordial gravitational potential power spectrum $P(k)=2\pi^2/k^3$
86:
87: \end{section}
88:
89: \begin{section}{Inhomogeneous Recombination}
90: We use the code DRECFAST \cite{drecfast} by
91: Novosyadlyj \cite{novos}, which is a modification of the recombination code
92: RECFAST \cite{recfast} to calculate the perturbations in the electron
93: number density $\delta_e=(n_e-\bar{n_e})/\bar{n_e}$ during
94: recombination. $n_e$ is the local electron number density and $\bar{n_e}$
95: is the mean electron density. Perturbations in baryon ($\delta_b$) and photon density
96: ($\delta_{\gamma}$) result in perturbations
97: in the electron density with an amplitude that is amplified or suppressed
98: depending on which terms in the evolution equations prevail. Specifically
99: photo-ionization prevails on superhorizon scales resulting in
100: $\delta_e \sim 5\times \delta_b$ during recombination. We refer the
101: reader to \cite{novos} for further details.
102:
103: We will not consider the full second order Boltzmann equation \cite{bar1}
104: but only the terms involving the perturbed electron density. This is given by:
105: \begin{equation}\label{eq1}
106: \frac{\partial \Theta^{(2)}}{\partial \tau} + \bm{\hat{n}}.\bm{\nabla_x}\Theta^{(2)}
107: -\dot{\kappa}\Theta^{(2)} =
108: -\dot{\kappa}\delta_e\left[\Theta_0^{(1)}-\Theta^{(1)}+\bm{\hat{n}}.\bm{V_b}
109: -\frac{1}{2}\mathcal{P}_2(\bm{\hat{V_b}}.\bm{\hat{n}})\Pi^{(1)}\right],\nonumber
110: \end{equation}
111: where $\tau$ is conformal time and $\tau_0$ its value today. $\Theta = \Delta T/T = \Theta^{(1)}+\Theta^{(2)}+\hspace{4 pt}
112: \rm{higher}\hspace{4 pt} \rm{order}\hspace{4 pt} \rm{terms}$ is the fractional
113: perturbation of CMB temperature, superscripts indicate the order of
114: perturbation while subscripts denote the multipole moment. All other
115: perturbations are of first order and we will omit the superscript for
116: them. Vector quantities are in bold face and their magnitudes in normal
117: face with $\bm{\hat{}}$ denoting unit vectors. We have omitted the factor of $1/2$ usually multiplied with
118: the second order term \cite{bar1} for convenience. $\bm{\hat{n}}$ is the
119: unit vector along the line of sight, $\dot{\kappa}\equiv
120: d\kappa/d\tau=-\bar{n_e}\sigma_Ta$ is the mean differential optical depth
121: due to Compton scattering, $\sigma_T$ is the Thomson scattering cross
122: section and $a$ is the scale factor. We take the electron velocity to be equal
123: to the baryon velocity $\bm{V_b}$. $\mathcal{P}_2$ is the Legendre
124: polynomial of order 2 and
125: $\Pi^{(1)}=\Theta_2^{(1)}+\Theta_{P0}^{(1)}+\Theta_{P2}^{(1)}$ is the
126: polarization term, subscript $P$ denoting the polarization field \cite{ma}.
127: We must caution
128: that this partial equation is gauge dependent because $\delta_e$
129: depends on the gauge. We will be using conformal Newtonian gauge for
130: $\delta_e$. The combinations of terms multiplying $\delta_e$ is gauge invariant. All
131: perturbed quantities are functions of $\tau$ and coordinates on spatial
132: slice $\bm{x}$. $\Theta$ is in addition a function of line of sight angle $\bm{\hat{n}}$.
133:
134: Following standard procedure \cite{cmbfast,bar1}, we
135: take the Fourier transform of Equation \ref{eq1} and integrate formally along
136: the line of sight.
137: \begin{eqnarray}
138: \Theta^{(2)}(\bm{k},\bm{\hat{n}},\tau_0) & = &
139: \int_0^{\tau_0}d\tau e^{ik\mu\left(\tau-\tau_0\right)}g(\tau)\int\frac{d^3\bm{k'}}{\left(2\pi\right)^3}\delta_e(\bm{k}-\bm{k'},\tau)\nonumber\\
140: &\times&\left[\Theta_0^{(1)}(\bm{k'},\tau)-\Theta^{(1)}(\bm{k'},\bm{\hat{n}},\tau)+
141: \bm{\hat{n}}.\bm{\hat{k'}}V_b(\bm{k'},\tau)
142: -\frac{1}{2}\mathcal{P}_2(\bm{\hat{k'}}.\bm{\hat{n}})\Pi^{(1)}(\bm{k'},\tau)\right],\label{fourier}
143: \end{eqnarray}
144: where $g(\tau)=-\dot{\kappa}(\tau)e^{-\kappa(\tau)}$ is the visibility
145: function and $\kappa(\tau)\equiv\int_{\tau}^{\tau_0}d\tau ' \bar{n_e}(\tau
146: ')\sigma_Ta(\tau ')$. Also $\bm{V_b}(\bm{k'},\tau)=\bm{\hat{k'}}V_b(\bm{k'},\tau)$.
147: We now take the spherical harmonic transform of Equation \ref{fourier} to
148: get the multipole moments, $\Theta^{(2)}_{\ell m}$.
149: \begin{eqnarray}
150: \Theta^{(2)}_{\ell m}(\bm{k},\tau_0) & = & \int
151: \Theta^{(2)}(\bm{k},\bm{\hat{n}},\tau_0) Y_{\ell
152: m}^{\ast}(\bm{\hat{n}})d\bm{\hat{n}}\nonumber
153: \end{eqnarray}
154:
155: This integral can be performed after decomposing $\Theta^{(1)}$ into
156: multipole moments,
157: $\Theta^{(1)}(\bm{k'},\bm{\hat{n}},\tau)=\sum_{\ell''}(-i)^{\ell ''}(2\ell
158: ''+1)\mathcal{P}_{\ell ''}(\bm{\hat{n}}.\bm{\hat{k'}})\Theta^{(1)}_{\ell
159: ''}(\bm{k'},\tau)$ and using relations between exponential, spherical
160: harmonics, spherical Bessel functions and Legendre polynomials \cite{var}.
161: Note that $\Theta_0^{(1)}$, which is the dominant term in the multipole
162: expansion of $\Theta^{(1)}$, will cancel out. We will see later that the
163: dipole term partially cancels the effect of ($V_b$), the Vishniac term. So only
164: $\ell \ge 2$ modes in $\Theta^{(1)}$, which are expected to be small
165: compared to monopole, will contribute to the bispectrum.
166: The result is:
167: \begin{eqnarray}
168: &&\Theta^{(2)}_{\ell m}(\bm{k},\tau_0) = \int_0^{\tau_0}d\tau g(\tau)\int
169: \frac{d^3\bm{k'}}{\left(2\pi\right)^3}\delta_e(\bm{k}-\bm{k'},\tau)\biggl[ \nonumber\\
170: && - (4\pi)^2\sum_{\ell 'm' \ell ''\neq 0 m''}i^{\ell '}(-i)^{\ell ''}\sqrt{\frac{(2\ell
171: '+1)(2\ell ''+1)}{4\pi(2\ell +1)}}C^{\ell 0}_{\ell ' 0 \ell ''
172: 0}C^{\ell m}_{\ell ' m' \ell '' m''}j_{\ell '}[k(\tau-\tau_0)]Y_{\ell '
173: m'}^{\ast}(\bm{\hat{k}})Y_{\ell '' m''}^{\ast}(\bm{\hat{k'}})\Theta_{\ell
174: ''}^{(1)}(\bm{k'},\tau)\nonumber\\
175: && +\frac{(4\pi)^2}{3}\sum_{\ell 'm' m''}i^{\ell '}\sqrt{\frac{(2\ell
176: '+1)3}{4\pi(2\ell +1)}}C^{\ell 0}_{\ell ' 0 1
177: 0}C^{\ell m}_{\ell ' m' 1 m''}j_{\ell '}[k(\tau-\tau_0)]Y_{\ell '
178: m'}^{\ast}(\bm{\hat{k}})Y_{1
179: m''}^{\ast}(\bm{\hat{k'}})V_b(\bm{k'},\tau)\nonumber\\
180: \label{theta2lm}&& -\frac{1}{2}\frac{(4\pi)^2}{5}\sum_{\ell 'm' m''}i^{\ell '}\sqrt{\frac{(2\ell
181: '+1)5}{4\pi(2\ell +1)}}C^{\ell 0}_{\ell ' 0 2
182: 0}C^{\ell m}_{\ell ' m' 2 m''}j_{\ell '}[k(\tau-\tau_0)]Y_{\ell '
183: m'}^{\ast}(\bm{\hat{k}})Y_{2
184: m''}^{\ast}(\bm{\hat{k'}})\Pi^{(1)}(\bm{k'},\tau)
185: \biggr]\\
186: &&\equiv \int_0^{\tau_0}d\tau g(\tau)\int
187: \frac{d^3\bm{k'}}{\left(2\pi\right)^3}\delta_e(\bm{k}-\bm{k'},\tau)S^{\ell m}(\bm{k},\bm{\hat{k'}},\bm{k'},\tau)\nonumber
188: \end{eqnarray}
189:
190: $C^{\ell m}_{\ell ' m' \ell '' m''}$ are Clebsch-Gordon coefficients,
191: $j_{\ell}$ are spherical Bessel functions. The sums are over all allowed values
192: of $\ell m$ with the exceptions explicitly specified. The last line defines
193: the function $S^{\ell m}$. Its arguments are written so that we can keep track
194: of the part, $\bm{k'}$, that statistical variables like temperature
195: anisotropy depend on from the part that deterministic functions depend on,
196: $\bm{\hat{k'}}$, separately.
197: \end{section}
198:
199: \begin{section}{Bispectrum}
200: We can now use Equation \ref{theta2lm} to calculate the bispectrum. This is
201: defined as:
202: \begin{eqnarray}\label{bispec}
203: B_{m_1m_2m_3}^{\ell_1\ell_2\ell_3}=\langle
204: a_{\ell_1m_1}^{(1)}(\bm{x},\tau_0)a_{\ell_2m_2}^{(1)}(\bm{x},\tau_0)a_{\ell_3m_3}^{(2)}(\bm{x},\tau_0)\rangle
205: + \hspace{4 pt}2\hspace{4 pt}\rm{permutations},
206: \end{eqnarray}
207: where $a_{\ell m}(\bm{x},\tau_0)$ are the coefficients in the spherical
208: harmonic expansion of the corresponding temperature anisotropy. $\langle
209: \rangle$ denotes the ensemble average. At second
210: order they are just the Fourier transform of $\Theta^{(2)}_{\ell
211: m}(\bm{k},\tau_0)$ while at first order they can be computed from
212: $\Theta^{(1)}_{\ell}(\bm{k},\tau_0)$.
213: \begin{eqnarray}
214: a_{\ell m}^{(2)}(\bm{x},\tau_0)&=&\int\frac{d^3\bm{k}}{(2\pi)^3}e^{i\bm{k}.\bm{x}}
215: \Theta^{(2)}_{\ell m}(\bm{k},\tau_0)\nonumber\\
216: a_{\ell m}^{(1)}(\bm{x},\tau_0)&=&4\pi\int\frac{d^3\bm{k}}{(2\pi)^3}e^{i\bm{k}.\bm{x}}
217: (-i)^{\ell}\Theta^{(1)}_{\ell}(\bm{k},\tau_0)Y_{\ell m}^{\ast}(\bm{\hat{k}})\nonumber
218: \end{eqnarray}
219: We can now calculate the first term of the bispectrum in Equation
220: \ref{bispec}.
221: \begin{eqnarray}
222: \langle 1,1,2 \rangle &\equiv &
223: \langle
224: a_{\ell_1m_1}^{(1)}(\bm{x},\tau_0)a_{\ell_2m_2}^{(1)}(\bm{x},\tau_0)a_{\ell_3m_3}^{(2)}(\bm{x},\tau_0)\rangle\nonumber\\
225: &=& (4\pi)^2\int
226: \frac{d^3\bm{k_1}}{(2\pi)^3}\frac{d^3\bm{k_2}}{(2\pi)^3}\frac{d^3\bm{k_3}}{(2\pi)^3}e^{i(\bm{k_1}+\bm{k_2}+\bm{k_3}).\bm{x}}(-i)^{\ell_1+\ell_2}Y_{\ell_1m_1}^{\ast}(\bm{\hat{k_1}})Y_{\ell_2m_2}^{\ast}(\bm{\hat{k_2}})\nonumber\\
227: &&\int_0^{\tau_0}d\tau g(\tau)\int
228: \frac{d^3\bm{k'}}{(2\pi)^3}\langle \delta_e(\bm{k_3}-\bm{k'},\tau)S^{\ell_3m_3}(\bm{k_3},\bm{\hat{k'}},\bm{k'},\tau)\Theta^{(1)}_{\ell_1}(\bm{k_1},\tau_0)\Theta^{(1)}_{\ell_2}(\bm{k_2},\tau_0)\rangle\nonumber\\
229: \end{eqnarray}
230: We can write each term in the ensemble average as a transfer function times
231: initial gravitational potential perturbation. Thus,
232: \begin{eqnarray}
233: \delta_e(\bm{k_3}-\bm{k'},\tau)&=&\Phi_i(\bm{k_3}-\bm{k'})\delta_e(|\bm{k_3}-\bm{k'}|,\tau)\nonumber\\
234: \Theta^{(1)}_{\ell_1}(\bm{k_1},\tau_0)&=&\Phi_i(\bm{k_1})\Theta^{(1)}_{\ell_1}({k_1},\tau_0)\nonumber\\
235: S^{\ell_3m_3}(\bm{k_3},\bm{\hat{k'}},\bm{k'},\tau)&=&\Phi_i(\bm{k'})S^{\ell_3m_3}(\bm{k_3},\bm{\hat{k'}},k',\tau)\nonumber\\
236: \langle \Phi_i(\bm{k_1})\Phi_i(\bm{k_2})\rangle &=& (2\pi)^3\delta^3(\bm{k_1}+\bm{k_2})P(k_1)\nonumber
237: \end{eqnarray}
238: We are using same symbols for statistical variables and their deterministic
239: transfer function counterparts, with arguments determining which one we
240: mean. $\delta^3$ is the three dimensional Dirac delta distribution and
241: $P(k)=2\pi^2/k^3$ is the initial power spectrum. Since we assume the initial
242: perturbation to be Gaussian, the 4-point ensemble average can be decomposed into
243: 2-point ensemble averages.
244: \begin{eqnarray}
245: \langle
246: \Phi_i(\bm{k_3}-\bm{k'})\Phi_i(\bm{k'})\Phi_i(\bm{k_1})\Phi_i(\bm{k_2})\rangle
247: &=&\langle
248: \Phi_i(\bm{k_3}-\bm{k'})\Phi_i(\bm{k'}))\rangle\langle\Phi_i(\bm{k_1})\Phi_i(\bm{k_2})\rangle\nonumber\\
249: &+&\langle
250: \Phi_i(\bm{k_3}-\bm{k'})\Phi_i(\bm{k_1}))\rangle\langle\Phi_i(\bm{k'})\Phi_i(\bm{k_2})\rangle\nonumber\\
251: &+&\langle
252: \Phi_i(\bm{k_3}-\bm{k'})\Phi_i(\bm{k_2}))\rangle\langle\Phi_i(\bm{k'})\Phi_i(\bm{k_1})\rangle\nonumber\\
253: &=&(2\pi)^6\delta^3(\bm{k_3})P(k')\delta^3(\bm{k_1}+\bm{k_2})P(k_1)\nonumber\\
254: &+&(2\pi)^6\delta^3(\bm{k_3}+\bm{k_1}-\bm{k'})P(k_1)\delta^3(\bm{k_2}+\bm{k'})P(k_2)\nonumber\\
255: \label{wick}&+&(2\pi)^6\delta^3(\bm{k_3}+\bm{k_2}-\bm{k'})P(k_2)\delta^3(\bm{k_1}+\bm{k'})P(k_1)
256: \end{eqnarray}
257: First term in Equation \ref{wick} contributes only for $\bm{k_3}=0$, it is
258: a product of monopole and power spectrum and is unobservable. The other two
259: terms are identical with $\bm{k_1},\bm{k_2}$ terms interchanged. So we need
260: consider only one of these. Denoting the two terms by superscript $(1,2)$
261: and $(2,1)$ we can write the first term of
262: the bispectrum as:
263: \begin{eqnarray}
264: \langle 1,1,2 \rangle &=& \langle 1,1,2 \rangle^{(1,2)}+\langle 1,1,2
265: \rangle^{(2,1)},\nonumber\\
266: \langle 1,1,2 \rangle^{(1,2)}& =& (4\pi)^2\int
267: \frac{d^3\bm{k_1}}{(2\pi)^3}\frac{d^3\bm{k_2}}{(2\pi)^3}\frac{d^3\bm{k_3}}{(2\pi)^3}e^{i(\bm{k_1}+\bm{k_2}+\bm{k_3}).\bm{x}}(-i)^{\ell_1+\ell_2}Y_{\ell_1m_1}^{\ast}(\bm{\hat{k_1}})Y_{\ell_2m_2}^{\ast}(\bm{\hat{k_2}})\nonumber\\
268: && \int_0^{\tau_0}d\tau g(\tau)\int
269: \frac{d^3\bm{k'}}{(2\pi)^3}\delta_e(k_1,\tau)S^{\ell_3m_3}(\bm{k_3},\bm{\hat{k'}},k_2,\tau)\Theta^{(1)}_{\ell_1}(k_1,\tau_0)\Theta^{(1)}_{\ell_2}(k_2,\tau_0)\nonumber\\
270: &&(2\pi)^6\delta^3(\bm{k_3}+\bm{k_1}-\bm{k'})P(k_1)\delta^3(\bm{k_2}+\bm{k'})P(k_2)\nonumber\\
271: &=& (4\pi)^2(2\pi)^3\int_0^{\tau_0}d\tau g(\tau)\int
272: \frac{d^3\bm{k_1}}{(2\pi)^3}\frac{d^3\bm{k_2}}{(2\pi)^3}\frac{d^3\bm{k_3}}{(2\pi)^3}(-i)^{\ell_1+\ell_2}Y_{l_1m_1}^{\ast}(\bm{\hat{k_1}})Y_{\ell_2m_2}^{\ast}(\bm{\hat{k_2}})\nonumber\\
273: &&P(k_1)P(k_2)
274: \delta_e(k_1,\tau)S^{\ell_3m_3}(\bm{k_3},\bm{\hat{-k_2}},k_2,\tau)\Theta^{(1)}_{\ell_1}(k_1,\tau_0)\Theta^{(1)}_{\ell_2}(k_2,\tau_0)\delta^3(\bm{k_1}+\bm{k_2}+\bm{k_3})\nonumber\\\label{112}
275: \end{eqnarray}
276: In the last step we have used one of the Dirac delta distributions to
277: integrate over $\bm{k'}$.
278: To proceed further we use the representation of Dirac delta distribution as
279: Fourier transform of unity and the expansion of exponential function in
280: spherical harmonics.
281: \begin{eqnarray}
282: \delta^3(\bm{k_1}+\bm{k_2}+\bm{k_3})&=&\int
283: \frac{d^3{\bm{r}}}{(2\pi)^3}e^{i(\bm{k_1}+\bm{k_2}+\bm{k_3}).\bm{r}}\nonumber\\
284: e^{i(\bm{k}.\bm{r})}
285: &=&\label{dirac}4\pi\sum_{\ell,m}i^{\ell}j_{\ell}(kr)Y_{\ell m}^{\ast}(\bm{\hat{k}})Y_{\ell m}(\bm{\hat{r}})
286: \end{eqnarray}
287: Using Equations \ref{dirac} in \ref{112} we can perform all angular
288: integrals and all radial integrals except two which involve transfer
289: functions of perturbations and the line of sight integral. The integrals
290: involving spherical harmonics result in Wigner 3jm symbols which can then
291: be summed using, for example, formulas tabulated in \cite{var}.
292: The result after performing these integrals is:
293: \begin{eqnarray}
294: \label{b12}\langle 1,1,2 \rangle^{(1,2)}&=&\sqrt{\frac{(2\ell_1 +1)(2\ell_2 +1)
295: (2\ell_3 +1)}{4\pi}}\left(
296: \begin{array}{lcr}
297: \ell_1 & \ell_2 & \ell_3 \\
298: 0 & 0 & 0
299: \end{array}
300: \right)
301: \left(
302: \begin{array}{lcr}
303: \ell_1 & \ell_2 & \ell_3 \\
304: m_1 & m_2 & m_3
305: \end{array}
306: \right)
307: \int_0^{\tau_0}d\tau
308: g(\tau)B_{\delta\Theta}^{\ell_1}(\tau)B_{\Theta\Theta}^{\ell_2}(\tau)\nonumber\\\hspace{4
309: pt}\\
310: B_{\delta\Theta}^{\ell_1}(\tau)&=&\frac{2}{\pi}\int
311: k_1^2dk_1P(k_1)\Theta_{\ell_1}^{(1)}(k_1,\tau_0)\delta_e(k_1,\tau)j_{\ell_1}[k_1(\tau_0-\tau)]\nonumber\\
312: B_{\Theta\Theta}^{\ell_2}(\tau)&=&\frac{2}{\pi}\int
313: k_2^2dk_2P(k_2)\Theta_{\ell_2}^{(1)}(k_2,\tau_0)\biggl[\nonumber\\
314: &&-\sum_{\ell ''\geq 1, \ell_2'} i^{\ell ''+\ell_2 + \ell_2 '}(-1)^{\ell_2}(2\ell
315: ''+1)(2\ell_2 '+1)\left(
316: \begin{array}{lcr}
317: \ell_2 ' & \ell_2 & \ell '' \\
318: 0 & 0 & 0
319: \end{array}
320: \right)^2
321: \Theta_{\ell ''}^{(1)}(k_2,\tau)j_{\ell_2'}[k_2(\tau_0-\tau)]\nonumber\\
322: && + i V_b(k_2,\tau)j'_{\ell_2}
323: [k_2(\tau_0-\tau)]
324: +\frac{1}{4}\Pi^{(1)}(k_2,\tau)\left\{3j''_{\ell_2}[k_2(\tau_0-\tau)]+j_{\ell_2}\left[k_2(\tau_0-\tau)\right]\right\}
325: \biggr]\nonumber\\
326: &=&\frac{2}{\pi}\int
327: k_2^2dk_2P(k_2)\Theta_{\ell_2}^{(1)}(k_2,\tau_0)\biggl[\nonumber\\
328: &&-\sum_{\ell ''\geq 2, \ell_2'} i^{\ell ''+\ell_2 + \ell_2 '}(-1)^{\ell_2}(2\ell
329: ''+1)(2\ell_2 '+1)\left(
330: \begin{array}{lcr}
331: \ell_2 ' & \ell_2 & \ell '' \\
332: 0 & 0 & 0
333: \end{array}
334: \right)^2
335: \Theta_{\ell ''}^{(1)}(k_2,\tau)j_{\ell_2'}[k_2(\tau_0-\tau)]\nonumber\\
336: & + &\left[\theta_b(k_2,\tau)-\theta_{\gamma}(k_2,\tau)\right]\frac{j'_{\ell_2}
337: [k_2(\tau_0-\tau)]}{k_2}\nonumber\\
338: \label{btt}& +&\frac{1}{4}\Pi^{(1)}(k_2,\tau)\left\{3j''_{\ell_2}[k_2(\tau_0-\tau)]+j_{\ell_2}\left[k_2(\tau_0-\tau)\right]\right\}
339: \biggr]
340: \end{eqnarray}
341: In the last step we have defined $iV_b=\theta_b/k$ and
342: $\theta_{\gamma}=3k\Theta_1$ and evaluated the sum over $\ell_2'$ explicitly for
343: $\Theta_1$. It can be seen from this expression that the effect of Vishniac
344: term $\theta_b$ is partly cancelled out by $\theta_{\gamma}$.
345: In this form the gauge invariance of $B_{\Theta\Theta}^{\ell_2}$ is also
346: apparent. In arriving at these expressions we have also used the identity
347: $j_{\ell}(-x)=(-1)^{\ell}j_{\ell}(x)$. The prime on the Bessel functions
348: denotes the derivative with respect to the argument.
349:
350: We can now write down the final expression for the angular averaged
351: bispectrum defined by:
352: \begin{eqnarray}
353: B^{\ell_1\ell_2\ell_3}&=&\sum_{m_1m_2m_3}\left(
354: \begin{array}{lcr}
355: \ell_1 & \ell_2 & \ell_3 \\
356: m_1 & m_2 & m_3
357: \end{array}
358: \right)
359: B_{m_1m_2m_3}^{\ell_1\ell_2\ell_3}\nonumber\\
360: &=&\sqrt{\frac{(2\ell_1 +1)(2\ell_2 +1)
361: (2\ell_3 +1)}{4\pi}}\left(
362: \begin{array}{lcr}
363: \ell_1 & \ell_2 & \ell_3 \\
364: 0 & 0 & 0
365: \end{array}
366: \right)
367: \int_0^{\tau_0}d\tau
368: g(\tau)\biggl[ B_{\delta\Theta}^{\ell_1}(\tau)B_{\Theta\Theta}^{\ell_2}(\tau)\nonumber\\
369: &&+
370: B_{\delta\Theta}^{\ell_2}(\tau)B_{\Theta\Theta}^{\ell_1}(\tau)+
371: B_{\delta\Theta}^{\ell_2}(\tau)B_{\Theta\Theta}^{\ell_3}(\tau)+
372: B_{\delta\Theta}^{\ell_3}(\tau)B_{\Theta\Theta}^{\ell_2}(\tau)+
373: B_{\delta\Theta}^{\ell_1}(\tau)B_{\Theta\Theta}^{\ell_3}(\tau)+
374: B_{\delta\Theta}^{\ell_3}(\tau)B_{\Theta\Theta}^{\ell_1}(\tau)
375: \biggr]\nonumber\\
376: \end{eqnarray}
377:
378: \end{section}
379: \begin{section}{Primordial Non-Gaussianity of local type}
380: We will compare our results with the bispectrum from primordial
381: non-Gaussianity of local type. This is given by \cite{komatsu01,review}:
382: \begin{eqnarray}
383: B^{\ell_1\ell_2\ell_3}_{prim}&=&2
384: \sqrt{\frac{(2\ell_1 +1)(2\ell_2 +1)
385: (2\ell_3 +1)}{4\pi}}\left(
386: \begin{array}{lcr}
387: \ell_1 & \ell_2 & \ell_3 \\
388: 0 & 0 & 0
389: \end{array}
390: \right)
391: \int_0^{\tau_0}d\tau
392: (\tau_0-\tau)^2\biggl[ \beta_{\ell_1}(\tau)\beta_{\ell_2}(\tau)\alpha_{\ell_3}(\tau)\nonumber\\
393: &&+\beta_{\ell_2}(\tau)\beta_{\ell_3}(\tau)\alpha_{\ell_1}(\tau)
394: +\beta_{\ell_3}(\tau)\beta_{\ell_1}(\tau)\alpha_{\ell_2}(\tau)\biggr],\nonumber\\
395: \beta_{\ell}(\tau)&=&\frac{2}{\pi}\int k^2dk P(k)\Theta_{\ell}(k,\tau_0)j_{\ell}[k(\tau_0-\tau)],\nonumber\\
396: \alpha_{\ell}(\tau)&=&\frac{2}{\pi}\int k^2dk f_{NL}\Theta_{\ell}(k,\tau_0)j_{\ell}[k(\tau_0-\tau)],
397: \end{eqnarray}
398: where $f_{NL}$ is the non-Gaussianity parameter defined by the following
399: form for the primordial potential,
400: $\Phi_i(\bm{x})=\Phi_L(\bm{x})+f_{NL}\left(\Phi^2_L(\bm{x})-\left\langle\Phi^2_L(\bm{x})\right\rangle\right)$
401: with $\Phi_L(\bm{x})$ Gaussian. Note that the expression for $\beta_{\ell}$
402: is similar to $B_{\delta\Theta}^{\ell}$ and $B_{\Theta\Theta}^{\ell}$,
403: the difference being the additional modulation by the terms at recombination
404: in the later case. As we will see later, $\alpha_{\ell}$ is similar in
405: shape to the visibility function $g({\tau})$ but peaks at an earlier time.
406: All plots and results are for $f_{NL}=1$.
407: \end{section}
408: \begin{section}{Numerical calculation and Results}
409: We calculate $\delta_e$ in conformal Newtonian gauge using DRECFAST \cite{novos}. All
410: other first order terms are calculated using CMBFAST \cite{cmbfast}. In
411: particular $\Theta_{\ell ''}(k,\tau)$ is given by the line of sight
412: integral:
413: \begin{equation}
414: \Theta_{\ell ''}(k,\tau) = e^{\kappa(\tau)}\int_0^{\tau}d\tau
415: 'S^{(1)}(k,\tau ')j_{\ell ''}[k(\tau - \tau ')]
416: \end{equation}
417: Here $S^{(1)}(k,\tau ')$ is the usual first order source term. Since we are
418: evaluating the transfer function at $\tau < \tau_0$, we get an extra factor
419: of $e^{\kappa(\tau)}$, otherwise this is same as the standard line of sight
420: formula \cite{cmbfast}.
421: $\Theta_{\ell ''}$ becomes smaller with increasing $\ell ''$
422: and we cut off the sum in Equation \ref{btt} at $\ell '' = 30$. This is accurate for $\tau <
423: 1000 \rm{Mpc}$ which is sufficient for the present calculation since the
424: visibility $g(\tau)$ is non-negligible only for $240\rm{Mpc} \lesssim
425: \tau \lesssim 800 \rm{Mpc}$ (Figure \ref{alphaeps}). Wigner 3jm symbols
426: are calculated using the code by Gordon and Schulten \cite{gordon} which
427: is publicly available at SLATEC common mathematical library \cite{slatec}.
428:
429:
430: \begin{figure}
431: \includegraphics{beta.eps}
432: \caption{\label{betaeps}$\beta_{\ell}(\tau)$ and
433: $B_{\delta\Theta}^{\ell}(\tau)$ is shown as a function of $\tau$ for
434: several values of $\ell$. }
435: \end{figure}
436:
437: \begin{figure}
438: \includegraphics{alpha.eps}
439: \caption{\label{alphaeps}$(\tau_0-\tau)^2\alpha_{\ell}(\tau)$ for
440: several values of $\ell$ and
441: the visibility function $g(\tau)$ as a function of conformal time
442: $\tau$. $(\tau_0-\tau)^2\alpha_{\ell}(\tau)$ peaks earlier than
443: $g(\tau)$.}
444: \end{figure}
445:
446: \begin{figure}
447: \includegraphics{btt.eps}
448: \caption{\label{btteps} $B_{\Theta\Theta}^{\ell}(\tau)$ is shown for several
449: values of $\ell$. Also shown are contributions from the polarization term
450: $\Pi$, slip term $\theta_b-\theta_g$ and from all the other terms
451: $\sum_{\ell \geq 2}\Theta^{(1)}_{\ell}$.
452: }
453: \end{figure}
454:
455: \begin{figure}
456: \includegraphics{b_l.eps}
457: \caption{\label{bleps} $0.1\times
458: \ell(\ell+1)\beta_{\ell}(\tau_{\ast})$,$0.01\times
459: \ell(\ell+1)B^{\ell}_{\delta\Theta}(\tau_{\ast})$,
460: $\ell(\ell+1)B^{\ell}_{\Theta\Theta}(\tau_{\ast})$ and
461: contributions to it from polarization, slip and rest of the terms is
462: shown as a function of multipole moments $\ell$. Some of the functions have been scaled as specified above. }
463: \end{figure}
464: Figure \ref{betaeps} shows a comparison of $\beta_{\ell}(\tau)$ and
465: $B_{\delta\Theta}^{\ell}(\tau)$. The modulation by $\delta_e$ results in
466: shifting the peak to later times. Also visibility $g(\tau)$
467: can be compared to primordial term $\alpha_{\ell}$. They are similar in
468: magnitude but have a different shape (Figure \ref{alphaeps}).
469: $B_{\Theta\Theta}^{\ell}$ is however much smaller in magnitude than the
470: other terms, $B_{\delta\Theta}$ and $\beta_{\ell}$, as can be seen from
471: Figures \ref{btteps} and \ref{bleps} at low $\ell$ but become comparable
472: at high $\ell$. This results in a much smaller bispectrum
473: from recombination at low $\ell$ compared to the primordial one. Figures
474: \ref{b10}, \ref{b200}, \ref{b1000} and \ref{b2000} show the absolute
475: value of the bispectrum from the primordial
476: non-Gaussianity with $f_{NL}=1$ and that due to inhomogeneous
477: recombination for $\ell_3 = 10,200,1000,2000$ as a function of
478: $\ell_1,\ell_2$. Z-axis is on linear scale while the color map is on log
479: scale. They are almost identical at the peaks but differ considerably
480: away from the peaks which occur when either $\ell_1$ or $\ell_2$ is
481: equal to $\ell_3$ and the other is small, a signature of the local
482: nature of the non-Gaussianity. At low $\ell$ the amplitude $B^{\ell_1
483: \ell_2 \ell_3}$ is much smaller than $B^{\ell_1
484: \ell_2 \ell_3}_{prim}$ but they become comparable at high $\ell$. Their
485: signs are however different and this will become apparent when we estimate
486: the confusion in $f_{NL}$ due to $B^{\ell_1
487: \ell_2 \ell_3}$.
488: \begin{figure}
489: \includegraphics{B10out.eps}
490: \caption{\label{b10} Absolute value of $B^{\ell_1
491: \ell_2 \ell_3}_{prim}$ labeled ``Primordial'' and $B^{\ell_1
492: \ell_2 \ell_3}$ labeled ``Recombination'' for $\ell_3=10$. Z axis
493: is on linear scale while color plot shows the same on log scale.
494: }
495: \end{figure}
496:
497: \begin{figure}
498: \includegraphics{B200out.eps}
499: \caption{\label{b200} Absolute value of $B^{\ell_1
500: \ell_2 \ell_3}_{prim}$ labeled ``Primordial'' and $B^{\ell_1
501: \ell_2 \ell_3}$ labeled ``Recombination'' for $\ell_3=200$. Z axis
502: is on linear scale while color plot shows the same on log scale.
503: }
504: \end{figure}
505:
506: \begin{figure}
507: \includegraphics{B1000out.eps}
508: \caption{\label{b1000} Absolute value of $B^{\ell_1
509: \ell_2 \ell_3}_{prim}$ labeled ``Primordial'' and $B^{\ell_1
510: \ell_2 \ell_3}$ labeled ``Recombination'' for $\ell_3=1000$. Z axis
511: is on linear scale while color plot shows the same on log scale.
512: }
513: \end{figure}
514:
515: \begin{figure}
516: \includegraphics{B2000out.eps}
517: \caption{\label{b2000} Absolute value of $B^{\ell_1
518: \ell_2 \ell_3}_{prim}$ labeled ``Primordial'' and $B^{\ell_1
519: \ell_2 \ell_3}$ labeled ``Recombination'' for $\ell_3=2000$. Z axis
520: is on linear scale while color plot shows the same on log scale.
521: }
522: \end{figure}
523:
524: To estimate the confusion to the estimate of $f_{NL}$ we follow \cite{fast}
525: and define the statistic
526: \begin{eqnarray}
527: S_{rec}&=&\sum_{\ell_1\leq \ell_2 \leq \ell_3}\frac{B^{\ell_1 \ell_2
528: \ell_3}B^{\ell_1 \ell_2
529: \ell_3}_{prim}}{C_{\ell_1}C_{\ell_2}C_{\ell_3}}\nonumber\\\label{fnl}
530: &\simeq&f_{NL}\sum_{\ell_1\leq \ell_2 \leq \ell_3}\frac{(B^{\ell_1 \ell_2
531: \ell_3}_{prim})^2}{C_{\ell_1}C_{\ell_2}C_{\ell_3}}
532: \end{eqnarray}
533: The result of solving Equation \ref{fnl} for $f_{NL}$ is shown in Figure
534: \ref{fnlfig} as a function of $\ell_{max}$, where $\ell_{max}$ is the
535: maximum value of $\ell$ included in the sum in Equation \ref{fnl}. As
536: expected from the examination of bispectra, $f_{NL}$ is small
537: and positive at low $\ell_{max}$ but $\sim -1$ at high $\ell_{max}$.
538:
539: \begin{figure}
540: \includegraphics{fnl42.eps}
541: \caption{\label{fnlfig} Comparison of primordial bispectrum from local type non-Gaussianity
542: with bispectrum due to inhomogeneous recombination in terms of parameter
543: $f_{NL}$ as a function of $\ell_{max}$, the maximum $\ell$ mode
544: considered.}
545: \end{figure}
546: \end{section}
547:
548: \begin{section}{Conclusions}
549: We have calculated the CMB bispectrum due to inhomogeneous
550: recombination. This was expected to be small because the combination of
551: terms multiplying $\delta_e$ is small. However calculations by Novosyadlyj
552: \cite{novos} showed that $\delta_e$ could be large and this suggested that the
553: CMB bispectrum could be non-negligible. Although it turns out to be small it
554: is still larger than what one might have expected from making an
555: estimate based on tight coupling or instantaneous recombination approximation \cite{bar2} and ignoring the
556: perturbations due to inhomogeneous recombination. This is especially evident
557: at high $\ell_{max}$.
558: Also the bispectrum from recombination looks
559: remarkably like the local type primordial bispectrum, which is not entirely
560: unexpected since both arise due to product of two first order terms.
561: Since the other second order terms in the Boltzmann equation \cite{bar1} are expected to
562: be larger than the ones we considered, our calculation motivates a full
563: second order numerical calculation of these terms in order to assess their
564: effect on future experiments such as Planck \cite{planck} and the level to which they
565: cause confusion when probing for primordial non-Gaussianity.
566: \end{section}
567: \begin{acknowledgments}
568: We thank Amit Yadav for help in the calculation of
569: the primordial bispectrum.
570: \end{acknowledgments}
571:
572: \bibliographystyle{apsrev}
573: \bibliography{ng}
574: \end{document}
575:
576: