1: \documentstyle[12pt,aasms4]{article}
2: %\documentclass[12pt,aasms4]{aastex}
3: %\usepackage{epsfig}
4: %\documentstyle[aas2pp4]{article}
5:
6: \begin{document}
7:
8: \title{Intermittent behavior of cosmic mass field revealed by QSO's
9: Ly$\alpha$ forests}
10:
11: \author{Priya Jamkhedkar\altaffilmark{1}, Hu Zhan\altaffilmark{2},
12: and Li-Zhi Fang\altaffilmark{1}}
13:
14: \altaffiltext{1}{Department of Physics, University of Arizona, Tucson,
15: AZ 85721}
16:
17: \altaffiltext{2}{Department of Physics and Astronomy, Arizona State
18: University, Tempe, AZ 85287}
19:
20: \begin{abstract}
21:
22: The intermittent behavior of the space-scale distribution of Ly$\alpha$
23: transmitted flux of QSO HS1700+64 has been analyzed via a discrete
24: wavelet transform. We found that there are strong indications of
25: intermittency on scales down to about 10 $h^{-1}$ kpc. These are: 1.) the
26: probability distribution function of the local fluctuations of the flux
27: is significantly long-tailed on small scales, and 2.) the local power
28: spectrum of the flux shows prominent spiky structures on small scales.
29: Moreover, the local power spectrum averaged on regions with different sizes
30: shows similar spiky structures. Therefore, the random mass density field
31: traced by the Ly$\alpha$ forests is rougher on smaller scales, consistent
32: with singular clustering.
33:
34: \end{abstract}
35:
36: \keywords{cosmology: theory - large-scale structure of the universe}
37:
38: \section{Introduction}
39:
40: The structures of the cosmic mass field on scales from sub-Mpc to kpc
41: have attracted a lot of attention recently. High resolution N-body
42: simulations show that the core profiles of massive halos of the cold
43: dark matter (CDM) cosmogony are singular (Navarro, Frenk \& White
44: 1997, Moore et al. 2000, Jing \& Suto 2000), while the halo profiles
45: required by the rotation curves of dwarf galaxies (Flores \& Primack
46: 1994, Burkeret 1995) are shallower than the numerical results.
47: The central cusps of dark halos also disagree with soft halo
48: profiles inferred from low surface brightness galaxies (de Blok \&
49: McGaugh 1997.) The cores of galaxies and clusters are even found to be
50: consistent with the thermal equilibrium model with an ``universal''
51: mass density (Firmani et al. 2000.) In other words, the singular
52: behavior of cosmic clustering has not been detected in the cores
53: of galaxies and clusters.
54:
55: We can now ask the question: can the singular behavior, if it exists, be
56: revealed by methods other than the mass profile of galaxies and
57: clusters? At the first glance, this goal seems to be not achievable,
58: as singular mass density profiles can only be seen in cores of galaxies
59: and clusters. However, a random mass field $\rho(x)$ consisting of rare
60: singular structures randomly scattered in low mass density background
61: typically is intermittent (Zel'dovich et al. 1990.) A basic character
62: of an intermittent field is that the density {\it difference} between two
63: neighboring positions, $|\rho(x+r)-\rho(x)|$, can be ``abnormally''
64: large when $r$ is very small. That is, the rare events of large density
65: difference $|\rho(x+r)-\rho(x)|$ on small scales $r$ have a higher
66: probability than that for a Gaussian field. Such a singular behavior of
67: a random mass field means that the probability density function (PDF)
68: of the density differences on small scales, $r$, has a long tail.
69: Obviously, the effects of long tailed PDF of the density difference are
70: not limited to the singular mass profile. This motivated us to look for
71: the PDF's long tail and its effects by using samples other than the
72: cores of galaxies and clusters.
73:
74: The PDF's long tail have not yet been seriously studied. The most popular
75: statistical measure of large-scale structures -- the power spectrum of
76: a mass field, is insensitive to the PDF's long tail. Furthermore, The
77: density difference, $|\rho(x+r)-\rho(x)|$, is a quantity localized in
78: space $x$ and on scale $r$, and so a space-scale decomposition is
79: necessary. Thus using the power spectrum or any statistic that is not
80: based on proper space-scale decomposition, it is not possible to identify
81: the effects of long-tailed PDF of the density difference.
82:
83: In this {\it Letter}, using a discrete wavelet transform (DWT), we look
84: for the long tail effects from a sample of QSO's Ly$\alpha$ forests. It
85: is believed that the distribution of baryonic diffuse matter is almost
86: point-by-point proportional to the underlying dark matter density.
87: Moreover, the absorption optical depth of Ly$\alpha$ is linearly
88: dependent on the baryonic density. Therefore, the high resolution data
89: of the transmitted flux of QSO's absorption would be a good candidate to
90: reveal the long-tailed PDF of cosmic mass field on small scales.
91:
92: \section{Method}
93:
94: Let us consider a 1-D random mass density field $\rho(x)$ in spatial
95: range $L$. With a DWT space-scale decomposition, the local density
96: difference, $|\rho(x+r)-\rho(x)|$, is represented by the wavelet function
97: coefficients (WFCs) as
98: %eq1
99: \begin{equation}
100: \tilde{\epsilon}_{j,l} =\langle \psi_{j,l}, \rho \rangle,
101: \end{equation}
102: where $\psi_{j,l}(x)$ is the orthonormal and complete basis of the
103: discrete wavelet transform, and $\langle ... \rangle =\int ....dx$ is
104: the inner product (Daubechies, 1992.) We use DAUB4 wavelet (Press et al.
105: 1993, Nielsen 1998) for our analysis through out this {\it Letter}. The WFC,
106: $\tilde{\epsilon}_{j,l}$, is the density fluctuation on the scale $L/2^j$
107: at the position $l=0,...2^{j}-1$, or the mean density difference between
108: nearest neighbors ranging on scale $L/2^j$ at $l$. If the ``fair sample
109: hypothesis" (Peebles 1980) holds, then the $2^j$ values of
110: $\tilde{\epsilon}_{j,l}$ form an ensemble of the density differences on
111: scale $j$, and therefore,
112: the distribution of $\tilde{\epsilon}_{j,l}$ is a reasonable estimate
113: of the PDF of the density differences on scale $j$ (Fang \& Thews 1998.)
114:
115: The second order statistics $|\tilde{\epsilon}_{j,l}|^2$ describes
116: the power of the perturbations of the mode $(j,l)$. In other words,
117: at a given position $l$, the local power spectrum is given by
118: %eq2
119: \begin{equation}
120: P_{j,l} = \tilde{\epsilon}_{j,l}^2.
121: \end{equation}
122: By averaging $P_{j,l}$ over all positions $l$, we have
123: %eq3
124: \begin{equation}
125: P_j =\frac{1}{2^j}\sum_{l=0}^{2^j-1}
126: |\tilde{\epsilon}_{j,l}|^2 = \frac{1}{2^j}\sum_{l=0}^{2^j-1}P_{j,l}.
127: \end{equation}
128: It has been shown that $P_j$ actually is a band-averaged Fourier
129: power spectrum (Pando \& Fang 1998; Fang \& Feng 2000.)
130:
131: The Fourier power spectrum lacks phase information, and therefore,
132: $P_j$ cannot show the phase-related features of clustering. However,
133: $P_{j,l}$ is phase-sensitive. One can search for the phase-related
134: features of the mass field by {\it local} DWT power spectrum $P_{j,l}$.
135:
136: We can generalize the definition of local DWT power spectrum, eq.(2)
137: as follows. First we chop $L$ into $2^{j_s}$ sub-interval, labeled by
138: $l_s= 0, 1...(2^{j_s}-1)$. Each sub-interval has a length $L/2^{j_s}$.
139: Then, the local DWT power spectrum at sub-interval $l_s$ is given by
140: %eq4
141: \begin{equation}
142: P_{j,\{j_s,l_s\}}=\frac{1}{2^{j-j_s}}
143: \sum_{l=l_s2^{j-j_{s}}}^{(l_s+1)2^{j-j_{s}}-1}
144: |\tilde{\epsilon}_{j,l}|^2.
145: \end{equation}
146: It is the power on scale $L/2^j$ localized on $l_s$ with size
147: $L/2^{j_s}$.
148:
149: For a Gaussian field, the local power spectrum $P_{j,l}$ will not
150: show structures with respect to $l$. On the other hand, the singular
151: behavior of a random field is measured by the exponent $\alpha$
152: defined by $|\rho(x+r)-\rho(x)| \sim r^{\alpha}$. The larger the
153: $\alpha$ is, the smoother the field on small scales is, and vice
154: versa. If the exponent $\alpha$ is negative there are an actual
155: singularity of the field. Therefore, the singular behavior can be
156: revealed by the roughness of the local power spectrum on small
157: scales. The WFC local power spectrum can also measure the index
158: $n$ of power-law profile $\rho \sim r^{-n}$ for individual core.
159:
160: \section{Sample and analysis}
161:
162: The sample used for the analysis is the Ly$\alpha$ transmitted
163: flux of QSO HS1700+64. This sample has been employed to study the evolution
164: of structure (Bi \& Davidsen 1997), the Fourier and DWT power spectra
165: (Feng \& Fang 2000.) The recovered power spectrum has been found
166: to be consistent with the CDM model on scales larger than about 0.1
167: $h^{-1}$ Mpc. The data ranges from 3727.012\AA \ to 5523.554\AA \ with a
168: resolution of $3$ kms$^{-1}$, for a total of 55882 pixels. In this paper,
169: we use the first 25000 pixels for analysis, which correspond to $z =$
170: 2.07$\sim$ 2.65, or $\lambda =$ 3727.012\AA \ $\sim$ \ 4434.266\AA. On
171: average, a pixel is about 0.029\AA, equivalent to physical size $\sim 5$
172: $h^{-1}$ kpc at $z \sim 2$ for a flat universe. We pad
173: 7768 null pixels at the end to utilize a fast wavelet transform algorithm,
174: which requires the data size in powers of 2. It does not affect the
175: analysis,
176: because the wavelet transform is localized. Moreover, we subject DWT
177: directly to pixels without transforming them to physical positions.
178: The relation between pixel number and physical position is not linear,
179: but it does not affect structures on small scales. Thus, we ignore the
180: effect of the non-linear relation in our present analysis.
181:
182: Most lines in the Ly$\alpha$ transmitted flux are due to absorptions
183: by gases in cool and low density regions. The pressure gradients are
184: generally less than gravitational forces. That is, the gas, and hence
185: the transmitted flux, should be good tracer of the dark matter.
186: Nevertheless, small scale structures of the dark matter field may be
187: smoothed out by the velocity dispersion of Ly$\alpha$ forest gases.
188: Therefore, to identify the clustering feature, we will statistically
189: compare the real data with its phase-randomized counterpart,
190: which is obtained by taking the inverse transform of the Fourier
191: coefficients of the original data after randomizing their phases uniformly
192: over $[0,2\pi]$ without changing their amplitudes.
193:
194: \subsection{The PDFs of WFCs}
195:
196: In Fig. 1, we show the PDFs of the WFCs for $j=8$, and 14. Each PDF is
197: normalized to have unit variance. For scale $j=8$, the departure from
198: Gaussian distribution is not so significant. Especially, no tail shows
199: in the $j=8$ PDF, i.e. no WFCs found to be
200: $\tilde{\epsilon}_{j,l} \geq 3 \sigma$. For j = 14, the PDF of WFCs has
201: $2^{14}$ events. Therefore, if the PDF were Gaussian, the number of
202: the events larger than 3$\sigma$ would be about 44. However, the data
203: shows 234 events beyond the 3$\sigma$ range. Furthermore, a Gaussian PDF
204: predicts that the number of the events larger than 5$\sigma$ should be 0.01,
205: while the data shows more than 100 events larger than 5$\sigma$. The data
206: extends to beyond 15$\sigma$ on both sides. Therefore, the PDF is
207: indeed significantly long-tailed on small scales. In other words, the
208: field is rougher on smaller scales. This indicates that the field may
209: contain singular structures.
210:
211: The shape of the two PDFs of $j=8$ and 14 is very different from each other.
212: That is, the two stochastic variables $\tilde{\epsilon}_{j,l}$, $j=8,
213: \ 14$ don't relate to each other as
214: %eq5
215: \begin{equation}
216: \tilde{\epsilon}_{j,l}
217: = 2^{\beta (j-j')}
218: \tilde{\epsilon}_{j',l},
219: \end{equation}
220: where $\beta$ is a constant. Therefore, the mass field traced by the
221: QSO HS1700+64 is unlikely to be self-similar.
222:
223: \subsection{Local power spectrum}
224:
225: In Fig. 2 we plot the local DWT power spectra of the HS1700+64
226: transmitted flux and its phase-randomized counterpart.
227: We take $j_s=7$, i.e. chopping the entire sample into 128 sub-intervals,
228: and, in each sub-interval, calculating the power spectra for
229: $j= 8$ $\sim$ 14, which correspond to physical scales
230: $2^{15-j} \times 5$ $h^{-1}$ kpc.
231:
232: Fig. 2 shows that the $j=8$ local power spectrum for real data is
233: not very different from its phase-randomized counterpart. This is
234: consistent with the $j=8$ PDF shown in Fig.1. It is closer-to-Gaussian.
235: While the $j=12$ $\sim$ $14$ local spectra are very rough, showing
236: remarkably spiky structures, that completely disappear in the phase
237: randomized counterpart. The spiky features mean that a significant
238: part of the power is concentrated in some small areas. This feature is
239: a result of the long tailed PDF of the density difference, i.e., higher
240: probability of ``abnormal'' density change. The smaller the scale, the
241: more pronounced the spiky features. This, again, points to a singular
242: clustering of cosmic mass field.
243:
244: It should be emphasized that the spikes in the local power spectrum
245: with high $j$ do {\it not} always correspond to the peaks in the density
246: distribution (or the absorption lines in the optical spectrum). The WFCs
247: $\tilde{\epsilon}_{j,l}$ describe the {\it difference} in density between
248: intervals of length $2^{15-j}\times 5$ $h^{-1}$ Mpc. The average of
249: $\tilde{\epsilon}_{j,l}$ over $l$ generally is zero. The mean power
250: (or variance) at $j=14$ is $P_{14}= 1.2\times10^{-5}$. Thus, even a single event
251: $>10\sigma$ at $j=14$ doesn't always refer to high density, and it can
252: happen in regions other than high density cores. The spikes denote the
253: positions where the density (or the absorption optical depth) undergoes
254: a dramatic change, which is the key indicator of the singular behavior of
255: a random field.
256:
257: The wavelet functions $\psi_{j,l}$ are orthogonal, and
258: therefore, the local power spectrum on scale $j$ doesn't contaminate
259: perturbations on other scales. This is very different from
260: the density distribution smoothed by a window function on scale $j$.
261: The peaks identified from the window-function smoothed density
262: field contain all contributions from perturbations on scales
263: $j' \leq j$. Therefore, the peaks in a smoothed field are
264: actually given by a superposition of perturbations on large and small
265: scales. They may not show singular features, because the PDFs of
266: large scale (or $j' < j$) perturbations are closer to Gaussian.
267:
268: We should estimate the possible distortion of the long tail effects
269: caused by velocity dispersion of gases. We calculated
270: the $j=14$ local power spectrum with sub-interval $j_s=14$, which is
271: plotted in Fig. 3. This local power spectrum has almost the same spiky
272: features identified in the $j=14,\ j_s = 7$ spectrum (see, Fig. 2.)
273: That is, most spikes shown by the power localized in sub-interval with size
274: of about 600 $h^{-1}$ kpc actually are localized in sub-interval
275: with size only about few 10 $h^{-1}$ kpc. This result indicates that the
276: contamination of gas velocity dispersion may not be significant,
277: at least, for prominent spikes. Thus, the spiky structures should mainly
278: come from the underlying mass field.
279:
280: Intermittency can more clearly be seen in Fig. 3. The mean powers
281: [eq.(3)] of the real data (left panel) and randomized counterpart (right
282: panel) of the $j_s=14$ and $j=14$ local spectrum actually are the same,
283: while the spikes of the real data are higher than the mean power by a
284: factor of few tens even hundreds. That is, in the real sample, most power
285: of the $j=14$ perturbations is concentrated in the spikes, and almost no
286: power, i.e. $P_{j,\{j_s,l_s\}}\simeq 0$, in places other than the spikes.
287: This is a typical intermittent distribution.
288:
289: \section{Conclusions}
290:
291: With the PDFs of $\tilde{\epsilon}_{j,l}$ and local DWT power spectra,
292: we show that the mass field traced by the QSO HS1700+64 Ly$\alpha$
293: forests is neither Gaussian, nor self-similar, but intermittent. The
294: spiky features shown in the local DWT power spectrum is remarkably
295: pronounced on scales down to about 10 $h^{-1}$ kpc. Moreover, the long
296: tail and the spiky features are substantial on smaller scales. This
297: indicates that the cosmic mass field is rougher on smaller scales,
298: consistent with singular clustering.
299:
300: A big advantage of intermittency is that one can detect singular
301: clustering using the statistical features of entire random density
302: field, not limited to the cores of galaxies and clusters. The information
303: of intermittency extracted from Ly$\alpha$ forests would be important
304: to test models of cosmic clustering in terms of their singular behavior.
305:
306: \acknowledgments
307:
308: We thank Dr. D. Tytler for kindly providing the data of the Keck spectrum
309: HS1700+64. PJ would also like to thank Dr. Robert Maier for his help. HZ
310: thanks Dr. David Burstein for helpful discussions.
311:
312: \begin{thebibliography}{99}
313:
314: \bibitem{bd} Bi, H.G. \& Davidson, A. F., 1997, \apj, 479, 523
315:
316: \bibitem{bu} Burkert, A., 1995, \apj, 477, L25
317:
318: \bibitem{d} Daubechies I. 1992, {\it Ten Lectures on Wavelets},
319: (Philadelphia, SIAM)
320:
321: \bibitem{de} de Blok, W.J.G. \& McGaugh, S.S. 1997, \mnras, 208, 493
322:
323: \bibitem{ft} Fang, L.Z. \& Thews, R. 1998, {\it Wavelet in Physics},
324: (World Scientific, Singapore)
325:
326: \bibitem{ff2} Fang, L.L. \& Feng, L.Z. 2000, \apj, in press,
327: astro-ph/0003259
328:
329: \bibitem{ff1} Feng, L.L. \& Fang, L.Z. 2000, \apj, 535, 519
330:
331: \bibitem{fda} Firmani, C., D'Onghia, E., Avila-Reese, V., Chincarini, G.
332: \& Her\'andez, X. 2000, astro-ph/0002376
333:
334: \bibitem{fp} Flores, R., Primack, J.R., 1994, \apj, 427, L1
335:
336: \bibitem{js} Jing, Y.P. \& Suto, Y. 2000, \apj, 529, L69
337:
338: \bibitem{moo} Moore, B., Gelato, S., Jenkins, A., Pearce, F.R. \&
339: Quillis, V. 2000, astro-ph/0002308
340:
341: \bibitem{nfw} Navarro, J.F., Frenk, C.S. \& White, S.D.M. 1997,
342: \apj, 490, 493
343:
344: \bibitem{oln} Nielsen, O.M. 1998, Wavelet in Scientific Computing,
345: http://www.inm.dtui.dk
346:
347: \bibitem{pf} Pando, J. \& Fang, L.Z. 1998, Phys. Rev. E57, 3593.
348:
349: \bibitem{pe} Peebles, P.J.E., 1980, {\it The Large Scale Structure of the
350: Universe},(Princeton Univ. Press)
351:
352: \bibitem{ptv} Press, W.H., Teukolsky, S.A., Vetterling, W.T., \& Flannery,
353: B.P., 1993, {\it Numerical Recipes}, (2nd Ed. Cambridge)
354:
355: \bibitem{z} Zel'dovich, Ya.B., Ruzmaikin, A.A. \& Sokoloff, D.D. 1990,
356: {\it The Almighty Chance}, (World Scientific, Singapore)
357:
358: \end{thebibliography}
359:
360: \clearpage
361: \begin{figure}
362: \newpage
363: \epsscale{0.8}
364: \plotone{fig1.ps}
365: \figcaption[fig1.ps] {The PDFs of the WFCs $\tilde{\epsilon}_{j,l}$
366: for $j=8$ and $j=14$ are shown on top right corner of each panel.
367: The tail is shown on the main part each panels. Horizontal axis is
368: for $\tilde{\epsilon}_{j,l}/\sigma$, where $\sigma$ is the variance
369: of the sample. Vertical axis is the probability density. The solid
370: curve is the Gaussian distribution with zero mean and unit variance.
371: The PDF of $j=14$ at zero is $\sim$4.5, so it is not shown in the
372: figures.
373: \label{fig1}}
374: \end{figure}
375:
376: \clearpage
377: \begin{figure}
378: \newpage
379: \epsscale{0.9}
380: \plotone{fig2.ps}
381:
382: \figcaption[fig2.ps] {The vertical axis represents
383: $2^{j-j_s}P_{j,\{j_s,l_s\}}$ with $j_s=7$. The left and right panels
384: are for the HS1700+64 transmitted flux, and its phase-randomized
385: counterpart, respectively. The horizontal axis represents the position
386: $l_s$ in units of $2^7 \times 10$ $h^{-1}$ kpc. The scales of the local
387: power spectra, $j= 8,\ 10,\ 12,\ 14$, correspond to physical scales
388: $2^{15-j} \times 5$ $h^{-1}$ kpc.
389: \label{fig2}}
390:
391: \end{figure}
392:
393: \clearpage
394: \begin{figure}
395: \newpage
396: \epsscale{0.8}
397: \plotone{fig3.ps}
398: \figcaption[fig3.ps]{The same as Fig. 2 with $j=14$, but taking
399: $j_s=14$.
400: \label{fig3}}
401: \end{figure}
402:
403: \end{document}
404: