1: %\documentclass[manuscript]{aastex}
2: %\documentclass[12pt,preprint]{aastex}
3: \documentclass[11pt,preprint]{aastex}
4: %\documentclass[preprint]{aastex}
5: %\documentclass[preprint2]{aastex}
6:
7: %\usepackage[onecolumn]{emulateapj5}
8: %\usepackage[twocolumn]{emulateapj5}
9: %\usepackage{emulateapj5}
10:
11: \newcommand{\bfm}[1]{{\mbox{\boldmath $#1$}}}
12: \newcommand{\sbfm}[1]{{\mbox{\scriptsize\boldmath $#1$}}}
13: \newcommand{\ssbfm}[1]{{\mbox{\tiny\boldmath $#1$}}}
14:
15: %comments for draft
16:
17: \newcommand{\notes}[1]{{\bf (note: #1)}}
18:
19: %\slugcomment{Submitted to Astrophys. J.}
20:
21: \begin{document}
22:
23: %\renewcommand{\thefootnote}{\fnsymbol{footnote}}
24: %\renewcommand{\theequation}{\mbox{\rm
25: %{\arabic{section}.\arabic{equation}}}}
26: %\renewcommand{\theequation}{\mbox{\rm {\arabic{equation}}}}
27:
28: %MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM
29:
30: \title{Phase Correlations in Non-Gaussian Fields}
31:
32: \author{Takahiko Matsubara}
33: \affil{Department of Physics and Astrophysics,
34: Nagoya University, Chikusa, Nagoya 464-8602, Japan}
35:
36: \email{taka@a.phys.nagoya-u.ac.jp}
37:
38: \begin{abstract}
39: We present the general relationship between phase correlations and the
40: hierarchy of polyspectra in the Fourier space, and the new theoretical
41: understanding of the phase information is provided. Phase correlations
42: are related to the polyspectra only through the non-uniform
43: distributions of the phase sum $\theta_{\sbfm{k}_1} + \cdots +
44: \theta_{\sbfm{k}_N}$ with closed wave vectors, $\bfm{k}_1 + \cdots +
45: \bfm{k}_N = 0$. The exact relationship is given by the infinite
46: series, which one can truncate in a consistent manner. The method to
47: calculate the series to arbitrary order is explained, and the explicit
48: expression of the first-order approximation is given. A numerical
49: demonstration proves that the distribution of the phase sum is a
50: robust estimator and provides an alternative statistic to search for
51: the non-Gaussianity.
52: \end{abstract}
53:
54: %MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM
55:
56: \keywords{cosmology: theory --- large-scale structure of universe ---
57: methods: statistical}
58:
59: \section{Introduction}
60: \label{sec1}
61:
62: Quantifying the cosmic fields, such as the density fields, velocity
63: fields, gravitational lensing fields, temperature fluctuations in the
64: cosmic microwave background, etc.~is undoubtedly crucial to study the
65: origin and dynamics of the structure in the universe. The structure of
66: these fields are believed to be emerged from primordial random
67: Gaussian perturbations, as most of the inflationary models naturally
68: predict nearly scale-invariant Gaussian fluctuations
69: \citep{gut82,sta82,haw82,bar83}.
70:
71: Even if the primordial perturbations are random Gaussian, the
72: gravitationally nonlinear evolution produces non-Gaussianity in the
73: cosmic fields. Quantifying the non-Gaussianity is not trivial since it
74: depends on the full hierarchy of the higher-order correlation
75: functions in real space, or of the polyspectra in Fourier space. First
76: several members of such hierarchy can be observationally determined,
77: with which only partial information on non-Gaussianity is quantified.
78: Therefore, alternative statistics, such as the void probability
79: function \citep{whi79}, the genus statistic \citep{got86}, the
80: Minkowski functionals \citep{min03,mec94,sch97}, etc.~which contain
81: the information on the full hierarchy of higher-order statistics
82: should be useful.
83:
84: Non-Gaussianity is frequently termed ``phase correlations''. This term
85: reflects the fact that the Fourier phases of a random Gaussian field
86: are randomly distributed without any correlation among different
87: modes. Therefore, phase correlations, if any, obviously characterize
88: the non-Gaussianity. However, what kind of phase correlations arise in
89: a given non-Gaussian field have been far from obvious. Investigations
90: along this line are quite limited in the literatures despite its
91: importance, apparently because of the lack of theoretical guidelines.
92: Most of the earlier work \citep{ryd91,sod92,ber98} only assess the
93: nonlinear evolution of phases in individual Fourier modes without
94: statistics. Phenomenological studies of $N$-body simulations have
95: revealed that the one-point phase distribution remains uniform even in
96: non-Gaussian fields \citep{sug91}, and that the phase difference
97: between neighboring Fourier modes is non-uniformly distributed
98: \citep{sch91,col00,chi01,chi02,wat03b}. However, the meaning of the
99: discovered phase correlations is obscure in those literatures.
100:
101: Since the hierarchy of the higher-order statistics contains
102: statistically all information on the distribution \citep{ber92}, there
103: should be some connection between phase correlations and polyspectra,
104: which is the key to theoretically understand the phase correlations.
105: Examining a toy model, \citet{wat03a} realized the importance of the
106: phase sums with closed wave vectors in this connection, although they
107: have never derived the exact relations. In this Letter, the connection
108: in the general form is discovered for the first time. As a result, we
109: will have much better theoretical understanding of the phase
110: information than before.
111:
112: \section{Phase correlations and polyspectra}
113: \label{sec2}
114:
115: Although the real part ${\rm Re} f_\sbfm{k}$ and the imaginary part
116: ${\rm Im} f_\sbfm{k}$ of the Fourier transform $f_\sbfm{k}$ of a
117: random field $f$ are naturally the independent variables, one can also
118: take their linear combinations $f_\sbfm{k} = {\rm Re} f_\sbfm{k} + i
119: {\rm Im} f_\sbfm{k}$ and $f_\sbfm{k}^* = {\rm Re} f_\sbfm{k} - i {\rm
120: Im} f_\sbfm{k}$ as another set of mutually independent variables. For
121: calculational advantages, we use the latter choice. In this Letter,
122: the reality of the random field $f$ is assumed since most of the
123: cosmic fields are real, although one can readily generalize the
124: following analysis to the complex fields. Because of the reality
125: condition, $f_\sbfm{k}^* = f_{-\sbfm{k}}$, $f_\sbfm{k}^*$'s are
126: actually not independent variables, and $f_\sbfm{k}$'s of all modes
127: $\bfm{k}$ are taken as independent variables.
128:
129: It is useful to define the normalized quantity $\alpha_\sbfm{k} \equiv
130: f_\sbfm{k}/\sqrt{P(k)}$, where $P(k) = \langle |f_\sbfm{k}|^2 \rangle$
131: is the power spectrum of the random field. The key technique to derive
132: the relation between phase correlations and polyspectra is given by
133: previous work \citep{mat95,mat03}: the joint probability function
134: $P(\{\alpha_\sbfm{k}\})$ of having particular set of $\alpha_\sbfm{k}$
135: is formally represented by
136: \begin{equation}
137: {\cal P}(\{\alpha_\sbfm{k}\}) =
138: \exp
139: \left(
140: \sum_{N=3}^\infty \frac{(-)^N}{N!}
141: \sum_{\sbfm{k}_1,\ldots,\sbfm{k}_N}
142: \left\langle
143: \alpha_{\sbfm{k}_1}\cdots\alpha_{\sbfm{k}_N}
144: \right\rangle_{\rm c}
145: \frac{\partial^N}
146: {\partial\alpha_{\sbfm{k}_1}\cdots\partial\alpha_{\sbfm{k}_N}}
147: \right)
148: {\cal P}_{\rm G}(\{\alpha_\sbfm{k}\}),
149: \label{eq01}
150: \end{equation}
151: where $\langle\cdots\rangle_{\rm c}$ indicates the cumulants, and
152: ${\cal P}_{\rm G}(\{\alpha_\sbfm{k}\})$ is the multivariate Gaussian
153: distribution function of variables $\{\alpha_\sbfm{k}\}$. In the
154: present case, ${\cal P}_{\rm G}(\{\alpha_\sbfm{k}\}) \propto
155: \exp\left(-\frac12 \sum_\sbfm{k} \alpha_\sbfm{k}
156: \alpha_{-\sbfm{k}}\right)$, since the covariance matrix is $\langle
157: \alpha_\sbfm{k} \alpha_{\sbfm{k}'}\rangle = \delta^{\rm
158: K}_{\sbfm{k}+\sbfm{k}'}$, where the symbol $\delta^{\rm K}_{\sbfm{k}}$
159: is defined by $\delta^{\rm K}_{\sbfm{k}} = 1$ for $\bfm{k} = 0$ and
160: $\delta^{\rm K}_{\sbfm{k}} = 0$ for $\bfm{k} \ne 0$. The periodic
161: boundary condition with boxsize $V=L^3$ is assumed.
162:
163: Since the polyspectra $P^{(N)}(\bfm{k}_1,\ldots,\bfm{k}_N)$ are
164: defined from the cumulants by
165: \begin{equation}
166: \langle f_{\sbfm{k}_1}\cdots f_{\sbfm{k}_N} \rangle_{\rm c} =
167: V^{1-N/2}
168: \delta^{\rm K}_{\sbfm{k}_1 + \cdots + \sbfm{k}_N}
169: P^{(N)}(\bfm{k}_1,\ldots,\bfm{k}_{N-1}),
170: \label{eq02}
171: \end{equation}
172: the above formula (\ref{eq01}) provides the relation between
173: polyspectra and joint distribution of the Fourier coefficients.
174: Expanding the exponential in equation (\ref{eq01}), each term in this
175: expansion consists of the products of polyspectra times derivatives of
176: ${\cal P}_{\rm G}$. The derivatives of ${\cal P}_{\rm G}$ are given by
177: a simple polynomial of $\alpha_{\sbfm{k}}$'s times ${\cal P}_{\rm G}$.
178: The general term in the expansion has the form
179: \begin{equation}
180: \sum_{\sbfm{k}'s}
181: \delta^{\rm K}_{\sbfm{k}_1 + \sbfm{k}_2 + \cdots}
182: \delta^{\rm K}_{\sbfm{k}'_1 + \sbfm{k}'_2 + \cdots} \cdots
183: p^{(N)}(\bfm{k}_1,\bfm{k}_2,\ldots)
184: p^{(N)}(\bfm{k}'_1,\bfm{k}'_2,\ldots) \cdots
185: H_{\sbfm{k}_1 \sbfm{k}_2 \cdots \sbfm{k}'_1 \sbfm{k}'_2 \cdots}
186: {\cal P}_{\rm G},
187: \label{eq03}
188: \end{equation}
189: with appropriate coefficients, where
190: \begin{equation}
191: p^{(N)}(\bfm{k}_1,\bfm{k}_2,\ldots,\bfm{k}_{N-1}) =
192: \frac{P^{(N)}(\bfm{k}_1,\bfm{k}_2,\ldots,\bfm{k}_{N-1})}
193: {\sqrt{V^{N-2} P(k_1)P(k_2)\cdots P(k_{N-1})
194: P\left(|\bfm{k}_1 + \cdots + \bfm{k}_{N-1}|\right)}},
195: \label{eq04}
196: \end{equation}
197: are the dimensionless, normalized polyspectra of $\alpha_{\sbfm{k}}$,
198: and
199: \begin{equation}
200: H_{\sbfm{k}_1 \sbfm{k}_2 \cdots} =
201: \frac{1}{{\cal P}_{\rm G}}
202: \left(-\frac{\partial}{\partial\alpha_{\sbfm{k}_1}}\right)
203: \left(-\frac{\partial}{\partial\alpha_{\sbfm{k}_2}}\right)
204: \cdots
205: {\cal P}_{\rm G},
206: \label{eq05}
207: \end{equation}
208: is a generalization of Hermite polynomials and is given by polynomials
209: of $\alpha_{\sbfm{k}}$'s and $\delta^{\rm K}_{\sbfm{k}}$'s. For
210: example,
211: \begin{equation}
212: H_{\sbfm{k}_1 \sbfm{k}_2 \sbfm{k}_3} =
213: \alpha_{-\sbfm{k}_1}\alpha_{-\sbfm{k}_2}\alpha_{-\sbfm{k}_3} -
214: \delta^{\rm K}_{\sbfm{k}_1 + \sbfm{k}_2}\alpha_{-\sbfm{k}_3} -
215: \delta^{\rm K}_{\sbfm{k}_2 + \sbfm{k}_3}\alpha_{-\sbfm{k}_1} -
216: \delta^{\rm K}_{\sbfm{k}_3 + \sbfm{k}_1}\alpha_{-\sbfm{k}_2},
217: \label{eq06}
218: \end{equation}
219: and so on. Thus, the joint probability function ${\cal
220: P}(\{\alpha_{\sbfm{k}}\})$ is represented by ${\cal P}_{\rm G}$ times
221: infinite sum of products by $\alpha_{\sbfm{k}}$'s, $\delta^{\rm
222: K}_{\sbfm{k}}$'s, and normalized polyspectra.
223:
224: Next step is to transform the complex variable $\alpha_{\sbfm{k}}$
225: into the modulus $|\alpha_{\sbfm{k}}|$ and the phase
226: $\theta_{\sbfm{k}}$ by $\alpha_{\sbfm{k}} = |\alpha_{\sbfm{k}}|
227: e^{i\theta_{\ssbfm{k}}}$. This should be carefully done, since
228: $\alpha_{\sbfm{k}}$ is considered independent on
229: $\alpha_{\sbfm{k}}^*$. At this point, we restrict the wavenumber
230: $\bfm{k}$ in the upper half sphere (uhs), $k_z \ge 0$, and the degrees
231: of freedom in the lower half sphere is relabeled by the reality
232: relation, $\alpha_{\sbfm{k}} = \alpha_{-\sbfm{k}}^*$ for $k_z < 0$.
233: The mode $\bfm{k}=0$ is excluded which ensures zero mean of the
234: original field $f$. The term (\ref{eq03}) is accordingly relabeled,
235: resulting in the sum of the products of $\alpha_{\sbfm{k}}$'s,
236: $\alpha_{\sbfm{k}}^*$'s, $\delta^{\rm K}_{\sbfm{k}}$'s, where $\bfm{k}
237: \in {\rm uhs}$, and normalized polyspectra. With the above procedures,
238: one can express the ratio ${\cal P}(\{\alpha_{\sbfm{k}}\})/{\cal
239: P}_{\rm G}(\{\alpha_{\sbfm{k}}\})$ in terms of the normalized
240: polyspectra, the modulus $|\alpha_{\sbfm{k}}|$ and the phase
241: $\theta_{\sbfm{k}}$. The Jacobian of the transform from
242: $\alpha_{\sbfm{k}}$ to $(|\alpha_{\sbfm{k}}|, \theta_{\sbfm{k}})$ is
243: the same for the probability functions ${\cal
244: P}(\{\alpha_{\sbfm{k}}\})$ and ${\cal P}_{\rm
245: G}(\{\alpha_{\sbfm{k}}\})$. Therefore this meets our ends to relate
246: the phase correlations and polyspectra, which is a completely new
247: result.
248:
249: Practically, one needs to truncate the infinite series by a consistent
250: manner. Fortunately, the non-Gaussianity generated by gravitationally
251: nonlinear evolution is known to approximately follow the hierarchical
252: model of the higher-order correlations, in which the polyspectra
253: $P^{(N)}$ have the order, $P^{(N)} \sim {\cal O}[P(k)^{N-1}]$
254: \citep[e.g.,][]{ber02}. This means $p^{(N)} \sim {\cal O}(\epsilon^{N-2})$,
255: where $\epsilon \sim \sqrt{P(k)/V}$. Therefore one can evaluate the
256: phase correlations in perturbative manner as long as the expansion
257: parameter $\epsilon$ is small. It is straightforward to perform the
258: above procedure to express ${\cal P}(\{|\alpha_{\sbfm{k}}|,
259: \theta_{\sbfm{k}}\})$ in terms of normalized polyspectra to arbitrary
260: order in $\epsilon$. In the lowest order approximation, only the
261: normalized bispectrum $p^{(3)}$ gives the term of order ${\cal
262: O}(\epsilon^1)$. The result is
263: \begin{eqnarray}
264: && {\cal P}(\{|\alpha_{\sbfm{k}}|, \theta_{\sbfm{k}}\})
265: \prod_{\sbfm{k}\in{\rm uhs}} d|\alpha_{\sbfm{k}}|
266: d\theta_{\sbfm{k}}
267: \nonumber\\
268: &&\qquad = \left[1 +
269: \sum_{\sbfm{k}_1,\sbfm{k}_2 \in {\rm uhs}}
270: |\alpha_{\sbfm{k}_1}||\alpha_{\sbfm{k}_2}| |\alpha_{\sbfm{k}_1 +
271: \sbfm{k}_2}|
272: \cos\left( \theta_{\sbfm{k}_1} + \theta_{\sbfm{k}_2} -
273: \theta_{\sbfm{k}_1 + \sbfm{k}_2} \right) p^{(3)}(\bfm{k}_1,
274: \bfm{k}_2) \right]
275: \nonumber\\
276: &&\qquad\quad\times
277: \prod_{\sbfm{k}\in{\rm uhs}} 2|\alpha_{\sbfm{k}}|
278: e^{-|\alpha_{\ssbfm{k}}|^2}
279: d|\alpha_{\sbfm{k}}|\frac{d\theta_{\sbfm{k}}}{2\pi}.
280: \label{eq07}
281: \end{eqnarray}
282: Higher-order terms can be similarly calculated, although they are
283: somehow tedious. For example, in the second-order approximation,
284: ${\cal O}(\epsilon^2)$, there appears the square of the first-order
285: term, and terms like
286: \begin{equation}
287: |\alpha_{\sbfm{k}_1}||\alpha_{\sbfm{k}_2}||\alpha_{\sbfm{k}_3}|
288: |\alpha_{\sbfm{k}_1 + \sbfm{k}_2 \pm \sbfm{k}_3}|
289: \cos\left(
290: \theta_{\sbfm{k}_1} +
291: \theta_{\sbfm{k}_2} \pm
292: \theta_{\sbfm{k}_3} -
293: \theta_{\sbfm{k}_1 + \sbfm{k}_2 \pm \sbfm{k}_3}
294: \right)
295: \label{eq08}
296: \end{equation}
297: with appropriate normalized trispectrum or the product of normalized
298: bispectra multiplied, and other terms which do not depend on phases.
299:
300: The phases always contribute to the probability distribution by the
301: combination of the form, $\cos(\theta_{\sbfm{k}_1} + \cdots +
302: \theta_{\sbfm{k}_N})$, with closed wavevectors: $\bfm{k}_1 + \cdots +
303: \bfm{k}_N = 0$. This is generally true because the phase dependence in
304: equation (\ref{eq03}) is the exponential of the sum of phases, and the
305: probability is the real number so that taking real parts gives the
306: cosine function. The reason that phase correlations exist only among
307: modes with closed wavevectors comes from the translational invariance.
308: In equations (\ref{eq07}) and (\ref{eq08}), wavenumbers are restricted
309: to the uhs so that the modes in the lower half sphere are relabeled by
310: $\theta_{\sbfm{k}} = - \theta_{-\sbfm{k}}$.
311:
312: The moduli $|\alpha_{\sbfm{k}}|$'s are easily integrated in the
313: first-order approximation of equation (\ref{eq07}), resulting in
314: %% \begin{eqnarray}
315: %% &&
316: %% {\cal P}(\{\theta_{\sbfm{k}}\}) \propto
317: %% 1 + \frac{\sqrt{\pi}}{2} \sum_{\sbfm{k}}^{\rm uhs}
318: %% p^{(3)}(\bfm{k},\bfm{k})
319: %% \cos(2\theta_{\sbfm{k}} - \theta_{2\sbfm{k}})
320: %% \nonumber\\
321: %% &&\qquad +
322: %% \left(\frac{\sqrt{\pi}}{2}\right)^3
323: %% \sum_{\sbfm{k}\ne \sbfm{k}'}^{\rm uhs}
324: %% p^{(3)}(\bfm{k},\bfm{k}')
325: %% \cos(\theta_{\sbfm{k}} + \theta_{\sbfm{k}'} -
326: %% \theta_{\sbfm{k} + \sbfm{k}'}).
327: %% \nonumber\\
328: %% \label{eq09}
329: %% \end{eqnarray}
330: \begin{equation}
331: {\cal P}(\{\theta_{\sbfm{k}}\}) \propto
332: 1 + \frac{\sqrt{\pi}}{2} \sum_{\sbfm{k}}^{\rm uhs}
333: p^{(3)}(\bfm{k},\bfm{k})
334: \cos(2\theta_{\sbfm{k}} - \theta_{2\sbfm{k}})
335: +
336: \left(\frac{\sqrt{\pi}}{2}\right)^3
337: \sum_{\sbfm{k}\ne \sbfm{k}'}^{\rm uhs}
338: p^{(3)}(\bfm{k},\bfm{k}')
339: \cos(\theta_{\sbfm{k}} + \theta_{\sbfm{k}'} -
340: \theta_{\sbfm{k} + \sbfm{k}'}).
341: \label{eq09}
342: \end{equation}
343: The practically useful relations between phase correlations and the
344: bispectrum are obtained by further integrating some phases in equation
345: (\ref{eq09}). One obtains
346: \begin{eqnarray}
347: &&
348: {\cal P}(\theta_\sbfm{k},\theta_{2\sbfm{k}}) \propto
349: 1 + \frac{\sqrt{\pi}}{2}
350: p^{(3)}(\bfm{k},\bfm{k})
351: \cos(2\theta_{\sbfm{k}} - \theta_{2\sbfm{k}}),
352: \label{eq10}\\
353: && {\cal P}(\theta_\sbfm{k},\theta_{\sbfm{k}'},\theta_{\sbfm{k} +
354: \sbfm{k}'}) \propto 1 + \frac{\pi^{3/2}}{4}
355: p^{(3)}(\bfm{k},\bfm{k}') \cos(\theta_{\sbfm{k}} +
356: \theta_{\sbfm{k}'} - \theta_{\sbfm{k} + \sbfm{k}'}),
357: \label{eq11}
358: \end{eqnarray}
359: where $\bfm{k} \ne \bfm{k}'$. These are the explicit forms of the
360: relation between phase correlations and the bispectrum in the
361: first-order approximation. We find that the distribution of the `phase
362: sum' $\theta_{\sbfm{k}} + \theta_{\sbfm{k}'} - \theta_{\sbfm{k} +
363: \sbfm{k}'}$ is determined only by the normalized bispectrum at the
364: first-order level, although higher-order normalized polyspectra can
365: contribute in general. The higher-order calculations show that the
366: distribution of the phase sum $\theta_{\sbfm{k}_1} + \cdots +
367: \theta_{\sbfm{k}_N}$ for the modes with closed wavevectors $\bfm{k}_1
368: + \cdots + \bfm{k}_N = 0$ is determined by normalized polyspectra of
369: order 3 to $N$ in the lowest-order approximation, where the
370: identification $\theta_{\sbfm{k}} = - \theta_{-\sbfm{k}}$ is
371: understood. It was vaguely suggested that there is some relationship
372: between the phase sum and polyspectra based on a particular
373: non-Gaussian model by \citet{wat03a}. We now find the explicit
374: relationship between them in general non-Gaussian fields.
375:
376: If we further integrate all phases but one particular
377: $\theta_{\sbfm{k}}$, the one-point probability function of a phase is
378: uniform, ${\cal P}(\theta_{\sbfm{k}}) = 1/2\pi$, which is consistent
379: with the previous $N$-body analysis \citep{sug91}. This conclusion
380: does not depend on the first-order approximation, since a single
381: wavevector can not be closed unless $\bfm{k} = 0$. Similarly, the
382: two-point probability function ${\cal P}(\theta_{\sbfm{k}},
383: \theta_{\sbfm{k}'})$ is also uniform unless $\bfm{k} = 2\bfm{k}'$. At
384: first glance, this conclusion seems to contradict the reported
385: non-uniform distribution of the phase difference of neighboring
386: wavevectors $D_{\sbfm{k}} \equiv \theta_{\sbfm{k}+\Delta\sbfm{k}} -
387: \theta_{\sbfm{k}}$ in $N$-body data
388: \citep{sch91,col00,chi01,chi02,wat03b}, where $\Delta\bfm{k}$ is a
389: fixed small vector. The same arguments are also applied to higher-order
390: approximations, so that the phase correlations between neighboring
391: wavenumbers should not appear even in strongly non-Gaussian fields in
392: a statistical sense.
393:
394: To resove this puzzle, it is useful to consider the conditional
395: probability function given a Fourier coefficient of a small wavenumber
396: $\alpha_{\Delta\sbfm{k}}$. In the first-order approximation, the joint
397: probability of having phases $\theta_\sbfm{k},\theta_{\sbfm{k} +
398: \Delta\sbfm{k}}$ with fixed $\alpha_{\Delta\sbfm{k}}$ is given by
399: \begin{equation}
400: {\cal P} \left( \theta_\sbfm{k},\theta_{\sbfm{k} +
401: \Delta\sbfm{k}} \left|\alpha_{\Delta\sbfm{k}}\right. \right)
402: \propto 1 + \frac{\pi}{2}
403: |\alpha_{\Delta\sbfm{k}}| \cos\left( \theta_{\sbfm{k} +
404: \Delta\sbfm{k}} - \theta_\sbfm{k} - \theta_{\Delta\sbfm{k}} \right)
405: p^{(3)}(\Delta\bfm{k}, \bfm{k}),
406: \label{eq12}
407: \end{equation}
408: which arise the non-uniform distribution pattern of the phase
409: difference $D_\sbfm{k}$. The pattern depends on the fixed phase
410: $\theta_{\Delta\sbfm{k}}$, which means the pattern varies from sample
411: to sample, and this is exactly what is reported in the $N$-body
412: analyses. The pattern of the phase difference should be significant
413: for red power spectrum, which is also consistent with $N$-body
414: analyses. The functional form of equation (\ref{eq12}) also agrees
415: with the $N$-body analysis \citep{wat03b}. The statistics of phase
416: difference is thus the manifestation of the large-scale patterns of
417: individual realizations. The position of the trough in the
418: distribution of the phase difference corresponds to the phase of the
419: mode $\Delta \bfm{k}$, and the degree of deviations from the uniform
420: distribution depends on the specific amplitude of the mode
421: $\Delta\bfm{k}$ and also on the normalized bispectrum.
422:
423:
424: \section{A numerical demonstration}
425: \label{sec3}
426:
427: The equations (\ref{eq10}) and (\ref{eq11}) relate the bispectrum to
428: the distribution of the phase sum $\theta_{\sbfm{k}} +
429: \theta_{\sbfm{k}'} - \theta_{\sbfm{k} + \sbfm{k}'}$. To see if this
430: kind of phase information is practically robust, we numerically
431: examine simple examples of non-Gaussian fields. Instead of examining
432: cosmological simulations, the following simple example is enough to
433: compare the numerical phase distributions and theoretical predictions.
434: Series of non-Gaussian fields are simply generated by exponential
435: mapping of a random Gaussian field:
436: \begin{equation}
437: f(\bfm{x}) =
438: \exp\left(g\phi(\bfm{x}) - g^2/2\right) - 1,
439: \label{eq13}
440: \end{equation}
441: where $\phi$ is a random Gaussian field with zero mean, unit variance,
442: and $g$ is the non-Gaussian parameter. We simply take a flat power
443: spectrum for the Gaussian field $\phi$. The field $f$ has zero mean
444: and variance $\langle f^2 \rangle = \exp(g^2) - 1$, and is called the
445: lognormal field \citep{col91}. This field has quite similar
446: statistical properties to gravitationally evolved non-Gaussian fields
447: and approximately follows the hierarchical model of higher-order
448: correlations. The parameter $g$ controls the non-Gaussianity, and the
449: random Gaussian field is recovered by taking the limit $g\rightarrow
450: 0$. The random field $f$ is generated on $64^3$ grids in a rectangular
451: box with the periodic boundary condition.
452:
453: In Fig.~\ref{fig1}, the distribution of the phase sum
454: $\theta_{\sbfm{k}_1} + \theta_{\sbfm{k}_2} - \theta_{\sbfm{k}_1 +
455: \sbfm{k}_2}$ is plotted for a binned configuration of the wavevectors,
456: $|\bfm{k}_1| = [0.4,0.5]$, $|\bfm{k}_2| = [0.5,0.6]$, $\theta_{12} =
457: [50^\circ,60^\circ]$, as an example, where $\theta_{12}$ is the angle
458: between $\bfm{k}_1$ and $\bfm{k}_2$, and the magnitudes of the
459: wavenumber are in units of the Nyquist wavenumber.
460: \begin{figure}
461: \epsscale{0.6}
462: \plotone{f1.eps}
463: \caption
464: {The distribution of the phase sum for a particular configuration of
465: wavevectors. Five non-Gaussian fields are shown, where $g$ is the
466: non-Gaussian parameter and $p^{(3)}$ is the normalized bispectrum for
467: the particular configuration. Theoretical predictions by normalized
468: bispectra in the first-order approximation are shown by solid
469: curves.
470: \label{fig1}}
471: \end{figure}
472: The phase sum is averaged over the wavevectors in a configuration bin.
473: The points represent the distributions of the phase sum in each
474: realization. Poisson errorbars are smaller than the size of the
475: points. The normalized bispectra $p^{(3)}(\bfm{k}_1,\bfm{k}_2)$ are
476: numerically evaluated from each realization, which are used to draw
477: the theoretical curves in the first-order approximation of equation
478: (\ref{eq11}). There is not any fitting parameter at all. The agreement
479: is remarkable in weakly non-Gaussian fields. When the non-Gaussianity
480: becomes high, the data points deviate from the first-order
481: approximation, and the distribution of the phase sum is sharply peaked
482: at $\theta_{\sbfm{k}_1} + \theta_{\sbfm{k}_2} - \theta_{\sbfm{k}_1 +
483: \sbfm{k}_2} = 0\ {\rm mod}\ 2\pi$. Up to $g \sim 3.0$, or $\langle
484: f^2\rangle^{1/2} \simeq 100$, the distribution of the phase sum is
485: accurately described by the first-order approximation, and is
486: determined only by the normalized bispectrum. Even though the
487: non-Gaussianity $g \sim 3.0$ on scales of the Nyquist wavenumber is
488: beyond the perturbative regime, the normalized bispectrum on scales of
489: the presently tested configuration is still within the perturbative
490: regime, $p^{(3)} \sim 0.25$. This means that the phase sum is well
491: approximated by first-order formula of the present work even when the
492: field is strongly nonlinear in dynamics, as long as the parameter
493: $P(k)/V$ on the relevant scales is small. Increasing the power on
494: relevant scales and/or decreasing the volume drive the phase
495: correlation large, due to the fact that the phase correlations are
496: particularly dependent on significant features in the sample.
497:
498: \section{Summary}
499: \label{sec4}
500:
501: The structure of the phase correlations in non-Gaussian fields is
502: elucidated. The method to relate the joint distribution of phases to
503: polyspectra is newly found and developed. The distribution of the
504: phase sum of closed wavevectors is represented by the polyspectra. We
505: found the statistics of the phase difference reflect the particular
506: phase of the mode within an individual sample. The distribution of the
507: phase sum of three or more modes carries the statistically useful
508: information. The understanding of the phase correlations in
509: non-Gaussian fields is now reached unprecedented level in this Letter,
510: so that many investigations to make use of the phase information will
511: be followed, such as the analysis of the non-Gaussianity of all kinds
512: of cosmic fields, the nonlinear gravitational evolution of the density
513: fields, the biasing and redshift-space distortion effects on the
514: galaxy clustering, the primordial non-Gaussianity from inflationary
515: models, and so forth. One may also hope that phase information can be
516: useful in statistical analyses of all kinds of non-Gaussian fields,
517: from various phenomena of pattern formations to human brain mapping,
518: etc.
519:
520: \acknowledgements
521:
522: I acknowledge support from grants MEXT 13740150.
523:
524: \begin{thebibliography}{}
525:
526: \bibitem[Bardeen, Steinhardt \& Turner(1983)]{bar83}
527: Bardeen, J.~M., Steinhardt, P.~J., \& Turner, M.~S.\ 1983, \prd, 28, 679
528: \bibitem[Bernardeau, Colombi, Gaztanaga \&
529: Scoccimarro(2002)]{ber02} Bernardeau, F., Colombi, S.,
530: Gaztanaga, E., \& Scoccimarro, R.\ 2002, \physrep, 367, 1
531: \bibitem[Bertschinger(1992)]{ber92} Bertschinger, E.\ in Lecture Notes
532: in Physics, 408, {\it New Insights into the Universe}, ed.
533: V.~J.~Martinez, M.~Portilla, and D.~Saez.~, p65 (Springer-Verlag,
534: Berlin, 1992).
535: \bibitem[Chiang(2001)]{chi01} Chiang, L.\ 2001, \mnras, 325,
536: 405
537: \bibitem[Chiang, Naselsky \& Coles(2002)]{chi02} Chiang, L.,
538: Naselsky, P., \& Coles, P.\ 2002, ArXiv arXiv:astro-ph/0208235
539: \bibitem[Coles \& Jones(1991)]{col91} Coles, P.~\& Jones, B.\
540: 1991, \mnras, 248, 1
541: \bibitem[Coles \& Chiang(2000)]{col00} Coles, P.~\& Chiang,
542: L.\ 2000, \nat, 406, 376
543: \bibitem[Gott, Dickinson, \& Melott(1986)]{got86} Gott,
544: J.~R., Dickinson, M., \& Melott, A.~L.\ 1986, \apj, 306, 341
545: \bibitem[Guth \& Pi(1982)]{gut82} Guth, A.~H.~\& Pi, S.-Y.\
546: 1982, Physical Review Letters, 49, 1110
547: \bibitem[Hawking(1982)]{haw82} Hawking, S.~W.\
548: 1982, \prb, 115, 295
549: \bibitem[Jain \& Bertschinger(1998)]{ber98} Jain, B.~\&
550: Bertschinger, E.\ 1998, \apj, 509, 517
551: \bibitem[Matsubara(1995)]{mat95} Matsubara, T.\ 1995, \apjs,
552: 101, 1
553: \bibitem[Matsubara(2003)]{mat03} Matsubara, T.\ 2003, \apj,
554: 584, 1
555: \bibitem[Mecke, Buchert \& Wagner(1994)]{mec94} Mecke,
556: K.~R., Buchert, T., \& Wagner, H.\ 1994, \aap, 288, 697
557: \bibitem[Minkowski(1903)]{min03} Minkowski, H.\ 1903,
558: Math.~Ann., 57, 447
559: \bibitem[Ryden \& Gramann(1991)]{ryd91} Ryden, B.~S.~\&
560: Gramann, M.\ 1991, \apjl, 383, L33
561: \bibitem[Scherrer, Melott, \& Shandarin(1991)]{sch91}
562: Scherrer, R.~J., Melott, A.~L., \& Shandarin, S.~F.\ 1991, \apj, 377,
563: 29
564: \bibitem[Schmalzing \& Buchert(1997)]{sch97} Schmalzing,
565: J.~\& Buchert, T.\ 1997, \apjl, 482, L1
566: \bibitem[Soda \& Suto(1992)]{sod92} Soda, J.~\& Suto, Y.\
567: 1992, \apj, 396, 379
568: \bibitem[Starobinskii(1982)]{sta82} Starobinskii, A.~A.\
569: 1982, \prb, 117, 175
570: \bibitem[Suginohara \& Suto(1991)]{sug91} Suginohara, T.~\&
571: Suto, Y.\ 1991, \apj, 371, 470
572: \bibitem[Watts \& Coles(2003)]{wat03a} Watts, P.~\& Coles, P.\
573: 2003, \mnras, 338, 806
574: \bibitem[Watts, Coles, \& Melott(2003)]{wat03b} Watts, P.,
575: Coles, P., \& Melott, A.\ 2003, \apjl, in press, arXiv:astro-ph/0211408.
576: \bibitem[White(1979)]{whi79} White, S.~D.~M.\ 1979, \mnras,
577: 186, 145
578:
579:
580: \end{thebibliography}
581:
582: \clearpage
583:
584: %\begin{figure}
585: %\epsscale{0.6} \plotone{mgcomb.2.eps} \figcaption[fig2.eps]{ The
586: %marginalized concentration ellipses for four cosmological parameters.
587: %\label{fig2}}
588: %\end{figure}
589:
590: \clearpage
591:
592: \end{document}
593: