1: \documentclass[12pt,preprint]{aastex}
2:
3:
4: \newcommand{\simgt} {\,\hbox{\lower0.6ex\hbox{$\sim$}\llap{\raise0.6ex\hbox{$>$}}}\,}
5: \newcommand{\lya} {{\rm Ly}$\alpha$\ }
6: \newcommand{\msun} {{\rm\,M_\odot}}
7:
8: \shorttitle{HII Bubble Growth during Reionization}
9: \shortauthors{Shin et al.}
10:
11:
12: \begin{document}
13:
14:
15: \title{COSMOLOGICAL HII BUBBLE GROWTH DURING REIONIZATION}
16: \author{MIN-SU SHIN, HY TRAC AND RENYUE CEN}
17: \affil{Princeton University Observatory, Peyton Hall, Princeton, NJ 08544-1001}
18:
19: \begin{abstract}
20: We present general properties of ionized hydrogen (HII) bubbles and
21: their growth based on a state-of-the-art
22: large-scale ($100$ Mpc/h) cosmological radiative transfer simulation.
23: The simulation resolves all halos with atomic cooling
24: at the relevant redshifts and simultaneously performs
25: radiative transfer and dynamical evolution of structure formation.
26: Our major conclusions include:
27: (1) for significant HII bubbles, the number distribution is peaked at
28: a volume of $\sim 0.6\ {\rm Mpc^{3}/h^{3}}$ at all redshifts.
29: But, at $z\le 10$, one large, connected network of bubbles dominates the entire HII volume.
30: (2) HII bubbles are highly non-spherical.
31: (3) The HII regions are highly biased with respect to the underlying matter distribution
32: with the bias decreasing with time.
33: (4) The non-gaussianity of the HII region is small when the universe
34: becomes 50\% ionized. The non-gaussianity reaches its maximal near
35: the end of the reionization epoch $z\sim 6$.
36: But at all redshifts of interest there is a significant non-gaussianity in the HII field.
37: (5) Population III galaxies may play a significant role in the reionization process.
38: Small bubbles are initially largely produced by Pop III stars.
39: At $z\ge 10$ even the largest HII bubbles have a balanced
40: ionizing photon contribution from Pop II and Pop III stars,
41: while at $z\le 8$ Pop II stars start to dominate the overall ionizing photon production
42: for large bubbles, although Pop III stars continue to make a non-negligible contribution.
43: (6) The relationship between halo number density and
44: bubble size is complicated but a strong correlation
45: is found between halo number density and bubble size for large bubbles.
46: \end{abstract}
47:
48:
49: \keywords{cosmology: theory---large-scale structure of universe---early universe
50: ---galaxies: intergalactic medium---methods: numerical---radiative transfer}
51:
52:
53: \section{INTRODUCTION}
54:
55: Understanding the reionization process is a key challenge in cosmology.
56: Present observations provide extremely useful but still limited information with respect
57: to the epoch of reionization.
58: On the one hand, the absorption spectrum observations of high redshift
59: quasars from the Sloan Digital Sky Survey (SDSS)
60: strongly suggest that the final reionization episode
61: comes to completion at $z\sim 6$ \citep[e.g.,][]{becker01,fan02,barkana02b,
62: cen02}.
63: On the other hand, the Wilkinson Microwave Anisotropy Probe (WMAP)
64: observations \citep{page06,spergel06}
65: infer that the intergalactic medium
66: is already significantly reionized by
67: some very early time, possibly $z\sim 15$.
68: In combination, it suggests that reionization process may be
69: quite complex and perhaps non-monotonic \citep[e.g.,][]{cen03,
70: haiman03,wyithe07a}.
71:
72: In addition, some other observations and analyses have also provided useful
73: constraints on the reionization process.
74: \citet{theuns02,hui03}
75: have shown that the observed temperature of the \lya forest
76: at redshift $z = 2-4$ requires that the cosmological reionization
77: occurred no earlier than $z = 9-10$,
78: although He II reionization somewhat
79: complicates this constraint.
80: Analysis based on the SDSS quasar Stromgren sphere size
81: suggest that the neutral hydrogen fraction
82: at $z\sim 6.3$ is a few tens of percent \citep{wyithe04,mesinger04},
83: whereas analyses based on the evolution of Lyman alpha emitters
84: from z = 6.5 to 5.7 imply that
85: a partially neutral IGM of neutral fraction of
86: $\sim 0.25$ is consistent with observations
87: \citep{malhotra04,stern05,haiman05,malhotra06}. But the recent
88: observations of Lyman alpha emitters begin to support the complete
89: reionization at $z\sim6$ as the SDSS quasar observations imply
90: \citep[e.g.,][]{kashikawa06,lidz07a}.
91:
92: While a fully self-consistent reionization picture
93: is far from being painted \citep[see][for a review]{barkana01,loeb06,fan06araa},
94: some key breakthroughs may rest in the redshift range $z = 7-15$,
95: where upcoming observations, including 21cm radio and CMB observations,
96: may be able to provide useful constraints.
97: To understand the reionization process several elegant analytical and semi-analytic
98: models have been developed and predictions made
99: \citep{miraldaescude00,barkana02a,furlanetto04,furlanetto05,furlanetto06,
100: zahn07,cohn07a}.
101: These methods provide ways to economically explore the large parameter space
102: and have provided very important insights with respect to
103: HII bubble evolution as well as
104: the evolution of global quantities such as the ionization fraction.
105: These methods, however, need to make necessary simplifying
106: assumptions, which would limit the scope of their applicability or
107: the accuracy of the predictions.
108:
109: We will take a complementary approach by making detailed numerical simulations
110: with fewer assumptions. In earlier simulations,
111: a choice is forced to be made between simulating a large volume
112: with very limited resolution or a sufficient resolution with too small volume
113: \citep[e.g.,][]{gnedin00,ciardi00,razoumov02,sokasian03,sokasian04,gnedin04}.
114: However, both a large volume ($\sim 100$ Mpc/h) and an adequate resolution
115: are necessary in order to follow the reionization sources
116: and sinks properly. A large simulation box is required because the sources in question
117: are extremely highly biased \citep{barkana04,trac06b}
118: and the bubbles have sizes of tens of Mpc prior to complete percolation
119: \citep[e.g.,][]{furlanetto04},
120: whereas resolving the bulk of sources of halo mass $\sim 10^8\msun$
121: dictates, at least, the mass resolution of $\sim 10^6\msun$ and spatial resolution
122: of a few kpc. With a unique hybrid (dark matter + baryons + stars + radiation) computational code
123: \citep{trac06b}
124: we have crossed the threshold to be able to simultaneously simulate a large volume
125: and have an adequate resolution to identify the bulk of the ionizing sources at
126: high redshift as well as to have an approximate yet adequate treatment of radiation transfer.
127: Recent direct simulations performed by other groups \citep{iliev06a,zahn07}
128: have yielded very important results but the inability
129: to adequately resolve dark matter halos of $\sim 10^8\msun$
130: renders the results uncertain. Additional features of our method include
131: (1) following simultaneously the evolution of structure formation and radiative transfer,
132: instead of performing radiative transfer as a post-processing step \citep{iliev06b}
133: or adding unresolved halos analytically \citep{mcquinn07},
134: (2) a self-consistent, albeit still uncertain, treatment of
135: metal-enrichment process, hence the spatially varying transition from Population III
136: (Pop III) to Population II (Pop II) initial mass function (IMF). Therefore, our
137: simulations have more direct numerical treatments of radiation sources, small-scale clumping,
138: and self-shielding than both post-processing and semi-analytic models.
139:
140: In this paper, we analyze the evolution of HII bubbles in a $100$ Mpc/h simulation box.
141: In particular, we study how HII bubbles grow and what physics determines the growth. Understanding
142: the evolution of HII bubbles is closely related to predicting future observation results of the
143: redshifted 21cm line and cosmic microwave background \citep[see][for a review]{fan06araa}. Among
144: large volume reionization simulations the morphology of HII regions were studied in \citet{iliev06a,
145: zahn07,mcquinn07}.
146: In \S2, the details of our simulation are described.
147: Results of simulations are given in \S3, followed by discussion and conclusions in \S4.
148: We use the following cosmological parameters based on the WMAP3 results \citep[see]
149: [and references therein]{spergel06}: $\Omega_{m}$ = 0.26, $\Omega_{\Lambda}$ = 0.74,
150: $\Omega_{b}$ = 0.044, h = 0.72, $\sigma_{8}$ = 0.77, and ${\rm n_{s}}$ = 0.95.
151: Throughout the paper both length and volume are given in comoving units.
152:
153:
154: \section{SIMULATIONS}
155:
156:
157:
158: Our simulation is generally based on the numerical methodology described in \citet{trac06b} and similarly utilized an N-body algorithm for dark matter, a star formation prescription, and a radiative transfer (RT) algorithm for ionizing radiation. However, we have taken some simpler approaches in this initial step to satisfy the computational challenge of large volume, high resolution simulations of cosmic reionization. In particular, we use an alternative RT algorithm and do not use a halo model for prescribing baryons and star formation. Here we summarize the main components and describe the modifications.
159:
160: In a ${\rm L = 100 ~ Mpc/h}$ simulation box, a high resolution N-body calculation for $2880^3$ dark matter particles on an
161: effective grid with $11520^3$ cells is performed using a particle-multi-mesh code \citep{trac06a}. The particle mass
162: resolution is ${\rm 3.02 \times 10^{6}\msun}$ and approximately 33 particles make up a ${\rm 10^{8}\msun/h}$
163: halo. The comoving grid spacing is 8.68 kpc/h and approximately 12 cells make up the virial volume of
164: a ${\rm 10^{8}\msun/h}$ halo. We identify dark matter halos in post-processing rather than during the
165: course of the simulation.
166:
167: We do not use the halo model of \citet{trac06b} for prescribing baryons and star formation. Instead, an alternative
168: approach is taken where we calculate the local matter density $\rho_m$ and velocity dispersion $\sigma_v$ for
169: each particle. The baryons are assumed to trace the dark matter distribution on all scales and we obtain
170: the local baryon density $\rho_b=\rho_m(\Omega_b/\Omega_m)$ and gas temperature $T=\mu\sigma_v^2/(3k)$. Star
171: formation is only allowed to occur in particles with densities $\rho_m>100\rho_{\rm crit}(z)$ and temperatures $T>10^4$ K.
172: This cut in the temperature-density phase-space effectively restricts star formation to regions within the virial
173: radius of halos where efficient atomic line cooling allows the gas to dissipate energy and further collapse to very
174: high densities. We also differentiate between the first generation Pop III stars and the second generation Pop II
175: stars by following the chemical enrichment of the interstellar medium and intergalactic medium as described in \citet{trac06b}.
176:
177: Our radiative transfer algorithm for ionizing radiation is based on a photon-advection scheme which is much less
178: computationally expensive than ray tracing. Particles, sources, and photons are collected on a RT grid with $360^3$
179: cells. Each cell spans 278 comoving kpc/h and the RT time step is set by the light-crossing time. For a source cell,
180: the excess source photons are propagated to the 26 neighboring cells. The advection is photon-conserving and the
181: isotropic redistribution uses a weighting function that is proportional to $1/r^2$. However, for a non-source cell
182: with excess radiation, the advection is generally anisotropic. For a HII region, photons originating from a central
183: source propagate in the direction coinciding with decreasing radiation flux. Therefore, we propagate photons by
184: looking for gradients in the radiation field. Consider a non-source cell with cell indices ${\rm (i, j, k)}$ and
185: radiation density $n_\gamma{\rm (i, j, k)}$. Radiation can propagate to an adjacent cell with
186: indices ${\rm(i+di, j+dj, k+dk)}$ if either of the gradient terms,
187: \begin{eqnarray}
188: \Delta_+ & \equiv n_\gamma{\rm (i+di, j+dj, k+dk)}-n_\gamma{\rm (i, j, k)}, \\
189: \Delta_- & \equiv n_\gamma{\rm (i, j, k)}-n_\gamma{\rm (i-di, j-dj, k-dk)},
190: \end{eqnarray}
191: is negative. The first gradient term indicates the downstream or expansion direction while the second
192: term indicates the upstream or source direction. For the 26 possible neighbor cells, we count how many
193: cells satisfy the above criteria and redistribute the excess photons equally among them. If none of the
194: 26 neighbor cells satisfy the above criteria, then the central cell is a convergent point and we redistribute
195: the photons equally among the neighbors.
196:
197: For an isolated HII region with one central source, if one of gradient terms is false, the other is generally
198: false too. However, this may not be the case near the interface of merging or overlapping HII regions. If
199: a weaker radiation field is trying to expand into the vicinity of a stronger radiation field, then the first
200: gradient term will be positive even though the second gradient term will be negative. Therefore, it is
201: necessary to use either of the gradient terms to determine the direction of radiation propagation. In the appendix, we compare the photon-advection scheme with the ray-tracing scheme of \citet{trac06b} and show that this simpler approach correctly captures the radiative transfer for a significantly majority of the reionization epoch.
202:
203: For each particle, we store 12 floating-point variables: three dimensional position, three dimensional velocity,
204: matter density, baryon density, temperature, stellar fraction, HI fraction, HeI fraction, and HeII fraction. We
205: calculate the ionization and recombination for each particle individually rather than using the lower
206: resolution density field defined on the RT grid. This allows us to correctly account for the clumping factor
207: and self-shielding of dense gas down to small scales of several comoving kpc/h.
208:
209: The simulation was run at the National Center for Supercomputing Applications (NCSA) on a shared-memory SGI Altix with
210: Itanium 2 processors. We used 512 processors, 2 TB of memory, and approximately 80000 cpu hours. With nearly 24
211: billion particles, this is the largest cosmological N-body simulation run to date. This is also the first
212: reionization simulation with a ${\rm L = 100 ~ Mpc/h}$ simulation box that can
213: directly resolve dark matter halos down to virial temperatures of $10^4$ K.
214:
215: In post-processing, we have identified dark matter halos using a friends-of-friends (FoF) algorithm
216: \citep{davis85} with a standard linking length of $b = {0.2 n^{-1/3}}$, where $n$ is the mean
217: particle number density. Figure 1 compares our mass functions with the analytic prediction of
218: \citet{press74} and the empirical prediction of \citet{warren06}. For redshifts
219: $z\lesssim10$, our results are in very good agreement with recent works on the mass function of
220: high redshift dark matter halos \citep{reed07, lukic07, cohn07b}. For higher redshifts
221: $z\gtrsim15$, we systematically under-resolve halos because our starting redshift of $z = 60$ is
222: too late to capture the nonlinear gravitational collapse. \citet{reed07} have suggested that
223: simulations must start $\sim10-20$ expansion factors before the redshift at which results converge.
224: We have found that we can accurately capture the formation of halos at $z\sim15-20$ with a smaller simulation
225: starting at $z = 300$ \citep{trac06b}.
226:
227: Our star formation prescription resembles those used in hydrodynamic simulations
228: \citep[e.g.,][]{springel03} and we obtain very similar results. In Figure 2, we compare
229: the star formation rate $\dot{\rho}(z)$ with an analytical model calibrated using hydrodynamic
230: simulations \citep{hernquist03}, corrected for our cosmology. The overall shape is
231: in very good agreement at all relevant redshifts. At $z = 6$, our amplitude is 1.25 times larger
232: and this difference is simply due to the fact that we have chosen a coarse value of $c_*=0.06$
233: for the star formation efficiency. For our purposes, the overall amplitude of the star
234: formation rate is unimportant since it is degenerate with the radiation escape fraction. We
235: have correspondingly chosen a radiation escape fraction of $f_{\rm esc}=0.15$ in order to have
236: the redshift of complete overlap occur at $z\approx6$.
237:
238:
239:
240:
241: \section{RESULTS}
242:
243: The first reionization region appears when regions of dense baryons turn on star formation.
244: Once the radiation sources begin ionization around them, the global ionization regions appear
245: as shown in Figure 3. At the early stage of reionization, isolated ionization regions are
246: easily found (see z = 11.228 in Figure 3).
247: As more radiation sources develop, the isolated bubbles get
248: connected together along filaments.
249: The process of HII bubble merging is
250: very complex and difficult to treat without detailed simulation.
251: Visually, in Figure 3, the computed HII bubbles
252: do not appear to be close to spheres,
253: as will be confirmed quantitatively below.
254: After $z \sim 10$
255: one large connected network of HII bubbles begins to dominate the simulation volume,
256: the HII percolation process
257: enters the ``cannibalistic" phase where
258: the dominant HII bubble rapidly swallows
259: other HII bubbles, as evident
260: in the z = 9.3, and 7.2 panels.
261: In the following subsections, we will present quantitative results.
262:
263:
264: \subsection{Size of HII bubbles}
265:
266: \subsubsection{Characteristic size}
267:
268:
269:
270: The size of HII bubbles is a basic quantity that has
271: received significant attention
272: \citep{furlanetto05,furlanetto06,iliev06a,mcquinn07,zahn07}.
273: We define HII bubbles as follows.
274: First, we mark all the cells
275: that are ionized at $>50$\% level.
276: Then, a bubble is defined by all such cells that are connected
277: at least by one face;
278: this is practically done by
279: grouping cells using a friends-of-friends (FoF) algorithm
280: with a linking length equal to simulation data ouput cell size, i.e. $\sim 0.14\ {\rm Mpc/h}$.
281: In addition, at least two cells are needed to make up a single bubble.
282:
283:
284:
285: Since the shape of HII bubbles is not close to a sphere as shown in \S3.2,
286: we present, in Figure 4, the volume
287: distribution of the found HII bubbles instead of the radius of a sphere.
288: Note that in previous
289: analytical studies HII bubbles are often assumed to be
290: spherical \citep[e.g.,][]{pritchard06} or have a characteristic size
291: \citep[e.g.,][]{mcquinn07}.
292: We present the size distribution of z = 5.7, 6.3, 8.9, 13.5, 17.2, and 20.6.
293:
294: In this paper, we define the
295: characteristic size of bubbles based on their number fraction not volume fraction as used
296: in \citet{iliev06a}. Therefore, our characteristic sizes well describe the existence of many small
297: non-connected bubbles. Meanwhile,
298: the definition of \citet{zahn07} is affected by volume occupied by HII bubbles because it measures
299: an ionization fraction within a certain smoothing radius.
300:
301: The volume distribution by the number fraction shows that there are three characteristic sizes of the HII
302: bubbles before the simulation box is dominated by a single bubble.
303: The characteristic volumes are 0.6,\ 0.03,\ and 0.006
304: ${\rm Mpc^{3}/h^{3}}$. These characteristic volumes do not change even though more regions become
305: ionized. But the least volume of the possible bubbles is $\sim 0.005\ {\rm Mpc^{3}/h^{3}}$ because
306: our bubble finding method defines the smallest bubble as at least two connected simulation cells.
307: Therefore, the smallest-volume peak in the HII bubble distribution
308: may be limited by our radiation cell size and should be considered as an upper limit
309: for the peak at the smallest scales.
310:
311: Other two characteristic sizes of $\sim 0.03$ and $\sim {\rm 0.6\ Mpc^{3}/h^{3}}$ are
312: real features of the size distribution. The peak at $\sim 0.03\ {\rm Mpc^{3}/h^{3}}$ represents
313: HII bubbles that are not limited by our radiation transfer resolution and should be real.
314: The peak at $\sim 0.6\ {\rm Mpc^{3}/h^{3}}$
315: probably represents typical mature HII bubbles produced by individual galaxies
316: (probably with their satellite galaxies),
317: which may subsequently merge with other bubbles.
318: This peak is maintained at all times until
319: when the percolation permeates the entire volume to complete
320: the reionization, as seen from the difference
321: between z = 6.3 and z = 5.7 in Figure 4.
322: The properties of these characteristic sizes
323: will be further analyzed and understood in \S3.1.3 and \S3.1.4.
324:
325:
326: The appearance of few dominant large bubbles is easily found by the change of volume fraction for
327: different sizes of bubbles. As shown in Figure 4, the volume fraction is dominated by
328: few large bubbles after z = 13.5. This result is also visually recognized in Figure 3.
329: In the plot of z $\sim$ 11.2, large bubbles begin to appear while many small bubbles
330: around newly formed sources dominate the number fraction in Figure 4. Therefore, the
331: characteristic sizes of 0.6,\ 0.03,\ and 0.006 ${\rm Mpc^{3}/h^{3}}$ can be maintained
332: until the scale of large bubbles is big enough to percolate small bubbles that are distributed
333: between large bubbles.
334:
335:
336: The volume distribution of HII bubbles in \citet{iliev06a} can be compared to our results.
337: Both simulations use the same definition of HII bubbles and the bubbles
338: are found by FoF method. The first difference is that in our simulation
339: large bubbles of $\simgt 10^{4}\ {\rm Mpc^{3}/h^{3}}$ appear later than z $\sim$ 15 when
340: the global volume-weighted ionization fraction is about 0.04. But
341: \citet{iliev06a} found the existence of those large bubbles around z $\sim$ 15 when
342: in their simulation the global volume-weighted ionization fraction is 0.05. This
343: difference is also related to different characteristic sizes that found in our simulation.
344: In our simulation, most HII bubbles are much smaller than $\simgt 10^{4}\ {\rm Mpc^{3}/h^{3}}$.
345: Even the found characteristic size of intermediate-size bubbles is $\sim 0.6\ {\rm Mpc^{3}/h^{3}}$.
346: \citet{iliev06a} generally shows larger bubbles for all ranges of the global ionization fraction
347: and the formation of large bubbles at
348: the earlier times than ours. Moreover, the largest fraction of bubbles is explained by
349: intermediate-size bubbles in \citet{iliev06a} while our results show the dominance of small bubbles
350: for almost all simulation time.
351:
352: We think this is probably due to the difference in the adopted
353: cosmological model in which the model used by \citet{iliev06a} has a higher $\sigma_8$
354: and the fact that they require the universe to be ionized much earlier such that
355: their computed $\tau_e$ matches the results from the first-year WMAP results. The low mass resolution
356: of \citet{iliev06a} also does not permit the formation of small bubbles because it cannot
357: capture the formation of low-mass DM halos \citep{zahn07}. Finally, the implementation of
358: baryon physics like ionizing sources in their simulation is quite different from our prescription
359: given in \S2. The importance of this difference is explained in \S3.1.3 and \S3.1.4.
360:
361:
362: %We note that our result is different from that of analytical studies and other radiative transfer models too.
363: %For example, \citet{zahn07} suggest that defining one characteristic size is possible at each epoch
364: %with their bubble boundary definition. But our results shows the fraction of smallest bubbles increases
365: %as reionization continues. Moreover, the characteristic sizes of HII bubbles in our simulation
366: %are maintained until the late stage of reionization. In particular, the peak of ${\rm \sim 0.6\ Mpc^{3}/h^{3}}$
367: %disappears only after z $<$ 6. Even considering a different boundary ionization
368: %fraction threshold of \citet{zahn07}, i.e. 90\%, our result is inconsistent with theirs,
369: %because a 90\% threshold would make the characteristic sizes of HII bubbles from our simulation still smaller.
370: %The reason for this difference is a different definition of a bubble size. In our definition, the
371: %characteristic size of bubbles is found based on their number fraction not volume fraction as used
372: %in \citet{iliev06a}. Therefore, our characteristic sizes well describe the existence of many small
373: %non-connected bubbles. Meanwhile,
374: %the definition of \citet{zahn07} is affected by volume occupied by HII bubbles because it measures
375: %an ionization fraction within a certain smoothing radius.
376:
377:
378: In Figure 4, we also present the bubble size distribution based on the definition of \citet{zahn07}
379: for comparison.
380: The same bubble boundary threshold of 90\% is adopted in this measurement. For similar
381: global volume-weighted ionization fractions as \citet{zahn07}, the size distribution from our
382: simulation has its peak at the smaller radius than their size distribution. The existence
383: of many resolved small bubbles results in the difference of the size distribution even though
384: we adopt the same definition as \citet{zahn07}. For example, we find two peaks of size distribution
385: at $\sim 0.1$ and $0.5$ Mpc/h for the volume-weighted ionization fraction of 0.12, while
386: \citet{zahn07} found a single peak at $\sim 0.7$ Mpc/h for the volume-weighted ionization
387: fraction of 0.11. Comparing the results for the volume-weighted ionization fraction
388: of 0.52, our distribution shows a single peak size, but the size is $\sim 2.5$ Mpc/h that is
389: smaller than $\sim 4.5$ Mpc/h of \citet{zahn07}. In particular, the resolved small bubbles makes the
390: peak probability of the size distribution, i.e. the height of the peak, insensitive to the change
391: of a global ionization fraction. Therefore, even with the same definition of the bubble size,
392: our result shows a slightly different result although a general result is consistent with each other.
393:
394:
395: \subsubsection{Bubble merger history}
396:
397: As it turns out, the percolation of bubbles is also related to the transition of major
398: ionizing photon production from Pop III to Pop II stars.
399:
400: To shed light on the overall reionization process,
401: we follow the change of ionizing environment at four different location
402: in the simulation box that are occupied by four individual bubbles that
403: all started at $z\sim 21$, shown in Figure 5. In the plot, we track the
404: size change of bubbles that occupy the four locations.
405: We see that these bubbles initially, formed at about the same time but
406: at four different locations, display a variety of histories, clearly due to the rich, disparate
407: structure formation histories and the evolution of ionizing environment.
408: One striking and very clear signature is some intermittent, rapid
409: downturns in the HII volume during an individual location's evolution history.
410: This is due to a rapid transition
411: of the IMF from Pop III to Pop II in the bubble, which in general occur at
412: the redshift range of $z = 10 - 15$.
413: This is consistent with the overall contributions of Pop II stars and and Pop III stars
414: as a function of redshift shown in Figure 6,
415: where we see that ionizing photons from Pop II stars begin to become more dominant between z = 10 and 15.
416: But this transition occurs spatially at different time such that each bubble position has
417: a slight different history.
418: We note that this transition time is not the same as the epoch when star formation rate is dominated by
419: Pop II population, since Pop III stars are more efficient producers of ionizing photons than Pop II stars
420: for a given amount of star formation rate. \citet{mcquinn07} pointed out that the nature of the ionizing sources
421: plays an important role in shaping the bubbles and our results
422: quantitatively confirm their conclusion by being able to perform
423: a detailed calculation that allows for a spatially resolved transition
424: from Pop III to Pop II stars.
425:
426: Merger history of the bubbles is also consistent with the changing total number of bubbles and
427: the dominance of the largest bubble over small bubbles after z $<$ 10. As percolation of the
428: bubbles proceeds, the total volume of HII bubbles increases while the number of individual bubbles
429: decreases as shown in Figure 7. That is, percolation of the bubbles becomes so important that
430: the largest network of connected bubbles dominates the simulation box. Although the number of
431: small bubbles also begins to decrease after z $<$ 10, the number fraction of small bubbles increases
432: because the size of the small bubbles is not large enough to be connected to other bubbles (see
433: Figure 4).
434:
435: \subsubsection{Dark matter halo and bubble size distribution}
436:
437: Basically, the clustering of radiation sources affects the percolation
438: domination and the spatially varying different transition epoch from Pop III star formation
439: to Pop II star formation. Because the two small characteristic size peaks found in Figure 4
440: are maintained by the formation of new radiation sources, the correlation between the bubble sizes
441: and the DM halos can be expected to explain a physical reason behind the bubble size distribution.
442:
443:
444: In Figure 8 we show the number density of halos as a function of HII bubble volume.
445: All DM halos whose centers are inside the bubble are said to belong to the bubble.
446: We separate the member DM halos to three separate mass ranges, ${\rm M_{halo} <
447: 10^{8}\ M_{\odot}/h,\ 10^{8}\ M_{\odot}/h \leq M_{halo} < 10^{9}\ M_{\odot}/h}$, and
448: ${\rm 10^{9}\ M_{\odot}/h \leq M_{halo}}$.
449: We see that there is an upturn of halo number density for all mass ranges for the smallest bubbles.
450: This indicates that these bubbles are rapidly expanding due to newly formed sources while the
451: number of halos stays approximately constant. These smallest bubbles do not contain
452: DM halos of ${\rm M_{halo} > 10^{9} M_{\odot}}$, corresponding to the peaks of
453: ${\rm \sim 0.11\ and\ 0.19\ Mpc/h}$
454: as found in \S3.1.1.
455: The halo number density decreases, as one moves to the right,
456: and then levels out at $\sim -1$ to $0$ of the displayed x-axis.
457: The most interesting feature is that there is a sudden rise
458: of halo number density of all masses toward the largest bubbles,
459: and points to the direction that percolation is expected to continue
460: with time, consistent with our analysis of the actual bubble size evolution.
461: The boundary between the percolation dominance and the source formation dominance is
462: the characteristic volume of ${\rm \sim 0.6\ Mpc^{3}/h^{3}}$.
463: This feature is described as the beginning of the flat
464: DM halo density distribution in Figure 8. A large drop in the minimum number density is
465: caused by difference of bubble age among DM halos of the same mass. For example,
466: the same mass DM halos of ${\rm 10^{8.5}\ M_{\odot}/h}$ collapse at different epoch and
467: it results in the differently matured HII bubbles around them. So, the number density
468: of DM halos can have a large dispersion, in particular, for the intermediate-mass DM halos.
469:
470:
471: As characteristic sizes of HII bubbles do not change much until z $\sim$ 6,
472: the DM halo distribution does not change as shown for z $\sim$ 8 and 10. So, we find that the size distribution
473: of HII bubbles are mainly regulated by an overall mass density that determines the evolution of baryons.
474: The latter property indicates highly biased galaxy formation and
475: suggests that stellar mass density is expected to rapidly rise in high density regions.
476: Whether this property may be preserved requires more detailed analysis.
477:
478:
479:
480:
481: \subsubsection{Dependence of bubble size on ionizing photon production}
482:
483: The rate of ionizing photon production per unit volume is likely
484: the main physical factor to determine the evolution of HII bubbles,
485: while the broad correlation between star formation rate and DM halo
486: density has likely produced some of the interesting
487: features presented in the previous sub-section,
488: because the latter is essentially proportional to
489: the total, integrated ionizing photon number density.
490: An ionizing photon production rate inside HII bubbles is estimated
491: to be proportional to star formation rate times
492: the ionizing photon production efficiency that
493: is a function of the IMF.
494: Figure 9 shows the ionizing photon production rate, separately from
495: Pop III stars and Pop II stars, as a function of the bubble size
496: at $z=8$ and $z=10$, respectively. The flat part of the average
497: ionizing photon production rate represents the value which is
498: approximately set by the average baryon number density and
499: recombination rate inside the bubbles.
500: It is clear that for the smallest bubbles with size $\le 0.3$ Mpc/h
501: Pop III stars dominate the ionizing photon production rate at both $z=10$ and $z=8$,
502: indicating that these bubbles are produced by
503: the first generation of stars formed in those locations
504: which are likely relatively removed from large halos where
505: star formation has gone through more than one generation and
506: the gas has been significantly enriched with metals,
507: consistent with Figure 8.
508: We expect that Pop III galaxies may be present in small bubbles
509: until $z\sim 6$.
510: For the large bubbles (size $\ge 10$ Mpc/h)
511: the contributions from Pop III stars and Pop II stars
512: seem balanced at $z\sim 10$.
513: But, by $z\sim 8$, these large bubbles begin to be dominated by
514: Pop II stars, although Pop III stars continue to make
515: significant contributions of ionizing photons.
516:
517: We show the careful
518: consideration of this transition is needed to understand the reason why we find the bubble size
519: distribution given in Figure 4.
520: We believe that the overall evolution of HII bubbles
521: depends on detailed modeling
522: of Pop II and Pop III stars.
523: A neglect or non-detailed treatment
524: \citep[e.g.,][]{zahn07,furlanetto05,iliev06b}
525: of the Pop III stars
526: would translate to a significant
527: change in the characteristics of bubble evolution.
528:
529:
530:
531:
532: \subsection{Shape of HII bubbles}
533:
534:
535: The shape of HII bubbles are far from a sphere unlike what is assumed in most analytical studies. Our
536: result is the first effort to present a quantitative measurement of HII bubble shape in the research of
537: cosmic reionization.
538: In order to quantify the shape of the bubbles as ellipsoid, first, we calculate
539: the inertia tensor:
540: \begin{equation}
541: I_{ij} = \Sigma (x_{i} - x_{i}^{c}) (x_{j} - x_{j}^{c}),
542: \end{equation}
543: where $x$ is the coordinate of every cell that is included in each HII bubble, and $x^{c}$ is
544: a geometrical center of the bubble. From the above defined $I_{ij}$, we obtain the square roots of
545: the three eigenvalues, $a > b > c$. We measure the shape of the bubbles that have at least six member
546: cells. Therefore, the smallest volume of the bubble is $\sim 0.016 {\rm \ Mpc^{3}/h^{3}}$.
547: The shape is expressed as two parameters $e_{1}$ and $e_{2}$:
548: \begin{equation}
549: e_{1} = \sqrt{1 - \frac{b^{2}}{a^{2}}},\ e_{2} = \sqrt{1 - \frac{c^{2}}{b^{2}}}.
550: \end{equation}
551: After percolation becomes a main process in growth of HII bubbles, the above measurement of
552: shape is useless. At late stage of reionization, the complicatedly connected bubbles are not
553: similar to ellipsoids. Moreover, neutral hydrogen regions can be surrounded by HII bubbles
554: such as a grape-like structure. Therefore, we derive the $e_{1}$ and $e_{2}$ for bubbles
555: before the largest bubble is comparable to the simulation volume.
556:
557:
558:
559: The bubble volume-weighted shape distribution is estimated in the following way:
560: \begin{equation}
561: P(e_{1}, e_{2}) = \int_{V} P(e_{1}, e_{2} | V) P(V) dV.
562: \end{equation}
563: Figure 10 shows
564: the concentration of volume-weighted HII bubbles in $(e_1,e_2)$ plane.
565: Note that a spherical bubble would have $e_1=e_2=0$.
566: In Figure 11 we show the bubble shape as a function of
567: bubble size, with the largest bubble separately shown as the star symbol,
568: for $z=13.8$ and $z=10$.
569: The shape of the bubbles follows a complicated dependence on their sizes.
570: The shapes of small bubbles are, however, less accurately calculated because of limited resolution.
571: Clearly, most of the volume in HII bubbles
572: is far from spherical.
573: The fact that $e_1$ is quite close to $1$ and $e_2$ has a wide range indicates
574: that bubbles tend to have shapes ranging from a cigar to a ruler.
575: Note that at $z<12$ the shape distribution is basically dominated by the largest bubble.
576: For example, at z = 10.0, the most probable shape parameter is close to $e_{1} \sim
577: 0.9$ and $e_{2} \sim 0.6$ of the largest bubble,
578: corresponding to axial ratios $b/a=0.4$ and $c/b=0.8$.
579: This high complex non-sphericality would introduce
580: substantial inaccuracies in calculations based on
581: analytic modeling of observables of the high-redshift universe \citep[see][for a review]{fan06araa}.
582: The complicated shape of HII bubbles
583: is largely determined by the clustering of ionizing sources,
584: as bubbles grow around large-scale structure, primarily filamentary networks.
585: Figure 12 gives the three orthogonal projections of a single randomly chosen bubble
586: which show characteristic of bubbles resulting from mergers of smaller bubbles.
587:
588:
589:
590: \subsection{Non-Gaussianity of reionization}
591:
592:
593:
594: We sample 5000 spheres randomly placed
595: in the simulation box and measure the mass-weighted ionization fraction of the spheres.
596: That is, we measure $x_{bubble}$ = (mass of HII) / (mass of H) for each sampled sphere. It is a
597: slightly different way to find a PDF compared to that of \citet{iliev06a}
598: in which they sample independent sub-cubes in their simulation box.
599: From these sampled spheres,
600: we estimate mean ($\mu$), variance ($\sigma^{2}$), skewness ($\gamma_{1}$), and kurtosis
601: ($\gamma_{2}$).
602: We use three different sphere sizes of comoving radius $5$, $10$, and $20\ {\rm Mpc/h}$, respectively.
603: Figure 13 shows measurements
604: of the underlying HII probability distribution functions.
605: First of all, we see that the mean ionization fraction reaches
606: 50\% at $z\ \sim\ 9.5$ (hereafter, ${\rm z_{50\%}}$).
607: Because the size of sampled spheres is much larger than the characteristic bubble sizes,
608: the measured $\mu$ reflects the global mean ionization fraction.
609: The largest dispersion of ionization fraction appears at
610: $z\ \sim\ 8.9$,
611: close to ${\rm z_{50\%}}$.
612: It is easily seen that
613: the degree of non-Gaussianity decreases with increasing size of the sphere,
614: as shown in all three bottom panels,
615: as expected.
616: While the variance starts to fall at $z< 8.9$,
617: the non-Gaussianity as measured by
618: $\gamma_{1}$ and $\gamma_{2}$
619: continue to rise, peaking at $z\sim 5.7$ when the largest bubble starts to fill
620: the entire simulation box.
621: Interestingly, it might be of some comfort for analytic modelers
622: that at ${\rm z_{50\%}}$, the HII field
623: is weakly non-Gaussian with $\gamma_1=(0.22, 0.17, 0.01)$ and $\gamma_2=(-0.66, -0.26, -0.46)$,
624: for $r=(5,10,20)$ Mpc/h, respectively. But $\gamma_1$ and $\gamma_2$ are most close to
625: 0 at $z \sim 9$ and $\sim 10$, respectively. Because the boundary between fully neutral and fully
626: ionized regions is much more smaller than the sizes of the sampling spheres, the measured
627: non-Gaussianity is quite small around ${\rm z_{50\%}}$ \citep{wyithe07b,lidz07b}.
628:
629:
630:
631: \subsection{Power spectrum of ionized hydrogen density fluctuation}
632:
633:
634:
635: We present the 3D power spectrum of ionized hydrogen density fluctuation field
636: (${\rm \delta_{HII}}$), neutral hydrogen density fluctuation field (${\rm \delta_{HII}}$),
637: and matter density fluctuation field (${\rm \delta_{mat}}$) in Figure 14. We note that
638: the power spectrum is not measured for ionization fraction fluctuation but for ionized density
639: fluctuation. The dimensionless power spectrum is defined here as
640:
641: \begin{equation}
642: \Delta_{k}^{2} = {\rm \frac{V}{(2 \pi^{2})} 4 \pi k^{3} ~ P(k)}.
643: \end{equation}
644:
645: We see that the ionized hydrogen density field shows a strong bias with respect to
646: the matter power spectrum,
647: where the bias is the strongest on small scales
648: and decreases toward large scales.
649: In general, the bias of ionized regions always decreases with time,
650: whereas the bias of the neutral regions does the opposite.
651: We find that ${\rm \Delta_{HI}^2}$ matches to that of ${\rm \delta_{mat}}$ at the early time when
652: most hydrogen is still neutral, as expected,
653: whereas
654: ${\rm \Delta_{HII}^2}$ becomes equal to that of ${\rm \delta_{mat}}$ when
655: the universe gets completely ionized.
656: This is in agreement with the underlying physics that the earlier bubbles are produced
657: by highly biased galaxies formed inside DM halos of high $\sigma$ peaks,
658: whereas only those highly biased hence large galaxies host
659: most of the neutral hydrogen when the universe becomes highly ionized.
660: It is interesting to note that
661: at around ${\rm z_{50\%}}$
662: ${\rm \Delta_{HI}^2}$ and ${\rm \Delta_{HII}^2}$
663: appear to have roughly the same shape,
664: although the latter has somewhat more power than the former,
665: while both have more power than the total matter,
666: suggesting a significant non-gaussian distribution of both.
667: We also note that there are significant difference
668: between the nonlinear (actual) power spectrum of total matter and
669: the linear power spectrum at $k\ge 0.3$ at $z\le 10$, indicating
670: the necessity to take nonlinear effects into account.
671:
672:
673: \section{CONCLUSIONS AND DISCUSSION}
674:
675:
676: Using the state-of-the-art, largest cosmological simulation of box size $100$ Mpc/h
677: with detailed radiative transfer of ionizing photons,
678: we compute the evolution
679: of HII bubbles during cosmological reionization from $z\sim25$ to $z\sim6$.
680: Our simulation resolves galaxy sources that produce most of the
681: ionizing photons in the universe in the concerned redshift range.
682: Here are a few major findings from our analysis.
683:
684: (1) We find that, for significant HII bubbles,
685: their number distribution is peaked at
686: a volume of $\sim 0.6$ ${\rm Mpc^{3}/h^{3}}$ at all redshifts.
687: But, at $z\le 10$, one large, connected network of bubbles dominates the entire HII volume.
688:
689: (2) HII bubbles are highly non-spherical. This result is not totally unexpected,
690: since one is quite familiar with the generic filamentary nature
691: of cosmic structure formation.
692: The mergers of HII bubbles blown by galaxies formed along filaments
693: consequently produce filamentary (fatter) HII bubbles.
694:
695: (3) The HII regions are highly biased with respect to the underlying matter distribution
696: with the bias decreasing with time as more and more less biased structures
697: begin to dominate the ionizing photon production rate. The opposite is true for
698: the neutral hydrogen region in the sense that
699: the bias of the neutral regions increases with time.
700:
701: (4) The universe becomes 50\% ionized at redshfit $z\sim 9.5$,
702: when the HII region is actually the least non-gaussian
703: in the redshift range concerned.
704: The non-gaussianity of the HII region reaches its maximal near the end of the reionization epoch $z\sim 6$.
705: But at all redshifts of interest there is a significant non-gaussianity in the HII field.
706:
707: (5) Population III galaxies play a very important role in the reionization process.
708: With our spatially resolved treatment of metal enrichment,
709: we show that small bubbles are initially largely produced by Pop III stars.
710: At $z\ge 10$ even the largest HII bubbles have a balanced
711: ionizing photon contribution from Pop II and Pop III stars.
712: At $z\le 8$, however, Pop II stars start to dominate the overall ionizing photon
713: budget for large bubbles, although Pop III stars continue to make
714: a non-negligible contribution.
715:
716: (6) The relationship between halo number density and
717: bubble size is complicated, although there is a strong correlation
718: for large bubbles in which larger bubbles tend to have higher halo number density.
719: In addition, we find that only the large bubbles (size $\ge 1$ Mpc/h)
720: contain halos of mass in excess of $10^9\msun$/h.
721:
722: Some of the results depend on modeling the formation of Pop III stars and the transition
723: from Pop III to Pop II star formation that needs more careful studies
724: \citep[e.g.][]{schneider06,smith07}. First
725: of all, the correct ionizing photon generation rate from the unit mass of Pop III stars or Pop II
726: stars is as important as a careful treatment of recombination process in reionization simulation.
727: As shown in \S3.1.4, the formation of small bubbles is closely connected to the ionizing efficiency
728: of the Pop III stars that is subject to the uncertain IMF of the Pop III stars.
729: Our reionization simulation is the first trial to simulate the transition of Pop III star formation to
730: Pop II star formation in a large simulation box of 100 Mpc/h. Improving our understanding
731: of chemical enrichment during reionization \citep[e.g.][]{greif07}, the transition of
732: a dominant radiation source will be more certainly simulated in future.
733:
734: In addition to the uncertain physics of Pop III population, the numerical treatment of
735: self-shielding and recombination has to be improved in future simulations. We used
736: information of simulation particles for calculating the recombination rate that has
737: better resolution than grid-based approach. But out approach also has limitation of
738: resolving small-scale physics like self-shielding. That must be improved in the reionization
739: simulation with a sub-grid model.
740:
741:
742: \acknowledgments
743: We thank Paul Bode for allowing us to use his FoF code and Marcelo Alvarez for
744: his bubble size measurement code. We also like to thank the referee for
745: many constructive comments that improved this paper. This research is
746: supported in part by grants AST-0407176 and NNG06GI09G.
747: HT is additionally supported in part by NASA grant LTSA-03-000-0090.
748:
749: \appendix
750: \section{RADIATIVE TRANSFER: PHOTON-ADVECTION VERSUS RAY-TRACING}
751:
752: We compare the photon-advection scheme with the ray-tracing scheme of \citet{trac06b} by
753: applying these RT algorithms to two reionization simulations with identical initial
754: conditions and source prescriptions. The test is conducted in a 25 Mpc/h box with
755: $720^3$ dark matter particles and $90^3$ RT cells and has the same spatial, mass,
756: and temporal resolution as the large 100 Mpc/h simulation.
757:
758: We find that the photon-advection scheme produces very similar results in terms of
759: the spatial and temporal evolution of HI for a substantial majority of the
760: reionization epoch. Figure 15 shows very good agreement in the redshift evolution of
761: the volume-averaged HI fraction $f_{\rm HI}$ and deviations are only found when the box
762: is already significantly ionized. The timing starts to differ at late stages near
763: complete overlap. There are small delays of $\Delta z\sim0.1$ and 0.2 when the neutral
764: fraction drops to 0.1 and 0.01, respectively.
765:
766:
767: In order to quantify the agreement in the spatial evolution of HI, we
768: cross-correlated the $f_{\rm HI}(x)$ fields from the two RT simulations and
769: plotted the results in Figure 16. From the two power spectra, $P_{\rm adv}(k)$
770: and $P_{\rm ray}(k)$, and the cross-power spectrum $P_{\rm adv-ray}(k)$, we
771: can define the bias function,
772: \begin{equation}
773: b(k)\equiv\sqrt{\frac{P_{\rm adv}(k)}{P_{\rm ray}(k)}},
774: \end{equation}
775: and cross-correlation function,
776: \begin{equation}
777: r(k)\equiv\frac{P_{\rm adv-ray}(k)}{\sqrt{P_{\rm adv}(k)P_{\rm ray}(k)}}.
778: \end{equation}
779: and use them to quantify the statistical correlation between the two fields.
780: In both simulations, the neutral fraction drops to 0.5 and 0.25 at $z=8.2$ and 7.5,
781: respectively. At these two stages of reionization, the bias $b(k)$ is very close to
782: unity and only deviates by a maximum of $\sim0.1$ near the RT grid Nyquist frequency
783: $k=11.3$ h/Mpc. Similarly, the cross-correlation $r(k)$ shows very good agreement
784: even down to the smallest scale. In the ray-tracing simulation, the neutral fraction
785: drops to 0.1 at $z=7.1$, but $f_{\rm HI}$ is slightly higher by $0.04$ in the
786: photon-advection case. At this redshift, the bias and cross-correlation
787: functions show that the two fields differ appreciably at all scales.
788: However, when the two simulations are compared at the same neutral fraction
789: of 0.1, the correlation is much better, particularly on large scales.
790: The deviations from unity at the smallest scales are due to the appearance of
791: new sources in the photon-advection simulation taken at a slightly later redshift.
792:
793:
794: In summary, we find that the photon-advection scheme correctly captures
795: the reionization process up until it is $\sim75\%$ completed by volume. At
796: earlier stages, the radiation field is highly nonuniform, even within the HII
797: regions, and the propagation of photons in the direction of decreasing radiation
798: flux is a good description. However, at later stages of reionization near
799: complete overlap, the radiation field is much more uniform and the weak
800: gradients in the radiation density do not provide accurate directions for photon
801: propagation. We conclude that the photon-advection scheme provides a cost-effective
802: approach to radiative transfer for a significant majority of the reionization epoch.
803: However, the stages just before and after reionization should be simulated
804: using more accurate approaches like ray-tracing. Our results in this
805: paper are valid at all stages when the the ionization fraction, rather
806: than the redshift, is used as an indicator of the progress of reionization.
807:
808:
809: \begin{thebibliography}{48}
810: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
811:
812: \bibitem[{{Barkana}(2002)}]{barkana02b}
813: {Barkana}, R. 2002, New Astronomy, 7, 85
814:
815: \bibitem[{{Barkana} \& {Loeb}(2001)}]{barkana01}
816: {Barkana}, R., \& {Loeb}, A. 2001, \physrep, 349, 125
817:
818: \bibitem[{{Barkana} \& {Loeb}(2002)}]{barkana02a}
819: ---. 2002, \apj, 578, 1
820:
821: \bibitem[{{Barkana} \& {Loeb}(2004)}]{barkana04}
822: ---. 2004, \apj, 609, 474
823:
824: \bibitem[{{Becker} {et~al.}(2001){Becker}, {Fan}, {White}, {Strauss},
825: {Narayanan}, {Lupton}, {Gunn}, {Annis}, {Bahcall}, {Brinkmann}, {Connolly},
826: {Csabai}, {Czarapata}, {Doi}, {Heckman}, {Hennessy}, {Ivezi{\'c}}, {Knapp},
827: {Lamb}, {McKay}, {Munn}, {Nash}, {Nichol}, {Pier}, {Richards}, {Schneider},
828: {Stoughton}, {Szalay}, {Thakar}, \& {York}}]{becker01}
829: {Becker}, R.~H., {Fan}, X., {White}, R.~L., {Strauss}, M.~A., {Narayanan},
830: V.~K., {Lupton}, R.~H., {Gunn}, J.~E., {Annis}, J., {Bahcall}, N.~A.,
831: {Brinkmann}, J., {Connolly}, A.~J., {Csabai}, I., {Czarapata}, P.~C., {Doi},
832: M., {Heckman}, T.~M., {Hennessy}, G.~S., {Ivezi{\'c}}, {\v Z}., {Knapp},
833: G.~R., {Lamb}, D.~Q., {McKay}, T.~A., {Munn}, J.~A., {Nash}, T., {Nichol},
834: R., {Pier}, J.~R., {Richards}, G.~T., {Schneider}, D.~P., {Stoughton}, C.,
835: {Szalay}, A.~S., {Thakar}, A.~R., \& {York}, D.~G. 2001, \aj, 122, 2850
836:
837: \bibitem[{{Cen}(2003)}]{cen03}
838: {Cen}, R. 2003, \apj, 591, 12
839:
840: \bibitem[{{Cen} \& {McDonald}(2002)}]{cen02}
841: {Cen}, R., \& {McDonald}, P. 2002, \apj, 570, 457
842:
843: \bibitem[{{Ciardi} {et~al.}(2000){Ciardi}, {Ferrara}, {Governato}, \&
844: {Jenkins}}]{ciardi00}
845: {Ciardi}, B., {Ferrara}, A., {Governato}, F., \& {Jenkins}, A. 2000, \mnras,
846: 314, 611
847:
848: \bibitem[{{Cohn} \& {Chang}(2007)}]{cohn07a}
849: {Cohn}, J.~D., \& {Chang}, T.-C. 2007, \mnras, 374, 72
850:
851: \bibitem[{{Cohn} \& {White}(2007)}]{cohn07b}
852: {Cohn}, J.~D., \& {White}, M. 2007, preprint (arXiv: 0706.0208)
853:
854: \bibitem[{{Davis} {et~al.}(1985){Davis}, {Efstathiou}, {Frenk}, \&
855: {White}}]{davis85}
856: {Davis}, M., {Efstathiou}, G., {Frenk}, C.~S., \& {White}, S.~D.~M. 1985, \apj,
857: 292, 371
858:
859: \bibitem[{{Fan} {et~al.}(2006){Fan}, {Carilli}, \& {Keating}}]{fan06araa}
860: {Fan}, X., {Carilli}, C.~L., \& {Keating}, B. 2006, \araa, 44, 415
861:
862: \bibitem[{{Fan} {et~al.}(2002){Fan}, {Narayanan}, {Strauss}, {White}, {Becker},
863: {Pentericci}, \& {Rix}}]{fan02}
864: {Fan}, X., {Narayanan}, V.~K., {Strauss}, M.~A., {White}, R.~L., {Becker},
865: R.~H., {Pentericci}, L., \& {Rix}, H.-W. 2002, \aj, 123, 1247
866:
867: \bibitem[{{Furlanetto} {et~al.}(2006){Furlanetto}, {McQuinn}, \&
868: {Hernquist}}]{furlanetto06}
869: {Furlanetto}, S.~R., {McQuinn}, M., \& {Hernquist}, L. 2006, \mnras, 365, 115
870:
871: \bibitem[{{Furlanetto} \& {Oh}(2005)}]{furlanetto05}
872: {Furlanetto}, S.~R., \& {Oh}, S.~P. 2005, \mnras, 363, 1031
873:
874: \bibitem[{{Furlanetto} {et~al.}(2004){Furlanetto}, {Zaldarriaga}, \&
875: {Hernquist}}]{furlanetto04}
876: {Furlanetto}, S.~R., {Zaldarriaga}, M., \& {Hernquist}, L. 2004, \apj, 613, 1
877:
878: \bibitem[{{Gnedin}(2000)}]{gnedin00}
879: {Gnedin}, N.~Y. 2000, \apj, 535, 530
880:
881: \bibitem[{{Gnedin}(2004)}]{gnedin04}
882: ---. 2004, \apj, 610, 9
883:
884: \bibitem[{{Greif} {et~al.}(2007){Greif}, {Johnson}, {Bromm}, \&
885: {Klessen}}]{greif07}
886: {Greif}, T.~H., {Johnson}, J.~L., {Bromm}, V., \& {Klessen}, R.~S.
887: 2007, preprint (astro-ph/0705.3048)
888:
889: \bibitem[{{Haiman} \& {Cen}(2005)}]{haiman05}
890: {Haiman}, Z., \& {Cen}, R. 2005, \apj, 623, 627
891:
892: \bibitem[{{Haiman} \& {Holder}(2003)}]{haiman03}
893: {Haiman}, Z., \& {Holder}, G.~P. 2003, \apj, 595, 1
894:
895: \bibitem[{{Hernquist} \& {Springel}(2003)}]{hernquist03}
896: {Hernquist}, L., \& {Springel}, V. 2003, \mnras, 341, 1253
897:
898: \bibitem[{{Hui} \& {Haiman}(2003)}]{hui03}
899: {Hui}, L., \& {Haiman}, Z. 2003, \apj, 596, 9
900:
901: \bibitem[{{Iliev} {et~al.}(2006{\natexlab{a}}){Iliev}, {Mellema}, {Pen},
902: {Merz}, {Shapiro}, \& {Alvarez}}]{iliev06a}
903: {Iliev}, I.~T., {Mellema}, G., {Pen}, U.-L., {Merz}, H., {Shapiro}, P.~R., \&
904: {Alvarez}, M.~A. 2006{\natexlab{a}}, \mnras, 369, 1625
905:
906: \bibitem[{{Iliev} {et~al.}(2006{\natexlab{b}}){Iliev}, {Mellema}, {Shapiro}, \&
907: {Pen}}]{iliev06b}
908: {Iliev}, I.~T., {Mellema}, G., {Shapiro}, P.~R., \& {Pen}, U.-L.
909: 2006{\natexlab{b}}, preprint (astro-ph/0607517)
910:
911: \bibitem[{{Kashikawa} {et~al.}(2006){Kashikawa}, {Shimasaku}, {Malkan},
912: {Doi}, {Matsuda}, {Ouchi}, {Taniguchi}, {Ly}, {Nagao}, {Iye}, {Motohara},
913: {Murayama}, {Murozono}, {Nariai}, {Ohta}, {Okamura}, {Sasaki}, {Shioya}, \&
914: {Umemura}}]{kashikawa06}
915: {Kashikawa}, N., {Shimasaku}, K., {Malkan}, M.~A.,
916: {Doi}, M., {Matsuda}, Y., {Ouchi}, M., {Taniguchi}, Y.,
917: {Ly}, C., {Nagao}, T., {Iye}, M., {Motohara}, K.,
918: {Murayama}, T., {Murozono}, K., {Nariai}, K., {Ohta}, K.,
919: {Okamura}, S., {Sasaki}, T., {Shioya}, Y., \& {Umemura}, M.
920: 2006, \apj, 648, 7
921:
922: \bibitem[{{Lidz} {et~al.}(2007){Lidz}, {McQuinn}, {Zaldarriaga}, {Hernquist}, \&
923: {Dutta}}]{lidz07a}
924: {Lidz}, A., {McQuinn}, M., {Zaldarriaga}, M., {Hernquist}, L., \&
925: {Dutta}, S. 2007, \apj, 670, 39
926:
927: \bibitem[Lidz et al.(2007)]{lidz07b} Lidz, A., Zahn, O.,
928: McQuinn, M., Zaldarriaga, M., Dutta, S.,
929: \& Hernquist, L.\ 2007, \apj, 659, 865
930:
931: \bibitem[{{Loeb}(2006)}]{loeb06}
932: {Loeb}, A. 2006, preprint (astro-ph/0603360)
933:
934: \bibitem[{{Lukic} {et~al.}(2007){Lukic}, {Heitmann}, {Habib}, {Bashinsky}, \&
935: {Ricker}}]{lukic07}
936: {Lukic}, Z., {Heitmann}, K., {Habib}, S., {Bashinsky}, S., \& {Ricker}, P.~M.
937: 2007, preprint (astro-ph/0702360)
938:
939: \bibitem[{{Malhotra} \& {Rhoads}(2004)}]{malhotra04}
940: {Malhotra}, S., \& {Rhoads}, J.~E. 2004, \apjl, 617, L5
941:
942: \bibitem[{{Malhotra} \& {Rhoads}(2006)}]{malhotra06}
943: ---. 2006, \apjl, 647, L95
944:
945: \bibitem[{{McQuinn} {et~al.}(2007){McQuinn}, {Lidz}, {Zahn}, {Dutta},
946: {Hernquist}, \& {Zaldarriaga}}]{mcquinn07}
947: {McQuinn}, M., {Lidz}, A., {Zahn}, O., {Dutta}, S., {Hernquist}, L., \&
948: {Zaldarriaga}, M. 2007, \mnras, 377, 1043
949:
950: \bibitem[{{Mesinger} \& {Haiman}(2004)}]{mesinger04}
951: {Mesinger}, A., \& {Haiman}, Z. 2004, \apjl, 611, L69
952:
953: \bibitem[{{Miralda-Escud{\'e}} {et~al.}(2000){Miralda-Escud{\'e}}, {Haehnelt},
954: \& {Rees}}]{miraldaescude00}
955: {Miralda-Escud{\'e}}, J., {Haehnelt}, M., \& {Rees}, M.~J. 2000, \apj, 530, 1
956:
957: \bibitem[{{Page} {et~al.}(2006){Page}, {Hinshaw}, {Komatsu}, {Nolta},
958: {Spergel}, {Bennett}, {Barnes}, {Bean}, {Dore'}, {Halpern}, {Hill},
959: {Jarosik}, {Kogut}, {Limon}, {Meyer}, {Odegard}, {Peiris}, {Tucker}, {Verde},
960: {Weiland}, {Wollack}, \& {Wright}}]{page06}
961: {Page}, L., {Hinshaw}, G., {Komatsu}, E., {Nolta}, M.~R., {Spergel}, D.~N.,
962: {Bennett}, C.~L., {Barnes}, C., {Bean}, R., {Dore'}, O., {Halpern}, M.,
963: {Hill}, R.~S., {Jarosik}, N., {Kogut}, A., {Limon}, M., {Meyer}, S.~S.,
964: {Odegard}, N., {Peiris}, H.~V., {Tucker}, G.~S., {Verde}, L., {Weiland},
965: J.~L., {Wollack}, E., \& {Wright}, E.~L. 2006, preprint (astro-ph/0603449)
966:
967: \bibitem[{{Press} \& {Schechter}(1974)}]{press74}
968: {Press}, W.~H., \& {Schechter}, P. 1974, \apj, 187, 425
969:
970: \bibitem[{{Pritchard} {et~al.}(2007){Pritchard}, {Furlanetto}, \&
971: {Kamionkowski}}]{pritchard06}
972: {Pritchard}, J.~R., {Furlanetto}, S.~R., \& {Kamionkowski}, M. 2007, \mnras,
973: 374, 159
974:
975: \bibitem[{{Razoumov} {et~al.}(2002){Razoumov}, {Norman}, {Abel}, \&
976: {Scott}}]{razoumov02}
977: {Razoumov}, A.~O., {Norman}, M.~L., {Abel}, T., \& {Scott}, D. 2002, \apj, 572,
978: 695
979:
980: \bibitem[{{Reed} {et~al.}(2007){Reed}, {Bower}, {Frenk}, {Jenkins}, \&
981: {Theuns}}]{reed07}
982: {Reed}, D.~S., {Bower}, R., {Frenk}, C.~S., {Jenkins}, A., \& {Theuns}, T.
983: 2007, \mnras, 374, 2
984:
985: \bibitem[{{Schneider} {et~al.}(2006){Schneider}, {Salvaterra}, {Ferrara}, \&
986: {Ciardi}}]{schneider06}
987: {Schneider}, R., {Salvaterra}, R., {Ferrara}, A., \& {Ciardi}, B. 2006, \mnras,
988: 369, 825
989:
990: \bibitem[{{Smith} \& {Sigurdsson}(2007)}]{smith07}
991: {Smith}, B.~D. \& {Sigurdsson}, S. 2007, \apjl, 661, L5
992:
993: \bibitem[{{Sokasian} {et~al.}(2003){Sokasian}, {Abel}, {Hernquist}, \&
994: {Springel}}]{sokasian03}
995: {Sokasian}, A., {Abel}, T., {Hernquist}, L., \& {Springel}, V. 2003, \mnras,
996: 344, 607
997:
998: \bibitem[{{Sokasian} {et~al.}(2004){Sokasian}, {Yoshida}, {Abel}, {Hernquist},
999: \& {Springel}}]{sokasian04}
1000: {Sokasian}, A., {Yoshida}, N., {Abel}, T., {Hernquist}, L., \& {Springel}, V.
1001: 2004, \mnras, 350, 47
1002:
1003: \bibitem[{{Spergel} {et~al.}(2006){Spergel}, {Bean}, {Dore'}, {Nolta},
1004: {Bennett}, {Hinshaw}, {Jarosik}, {Komatsu}, {Page}, {Peiris}, {Verde},
1005: {Barnes}, {Halpern}, {Hill}, {Kogut}, {Limon}, {Meyer}, {Odegard}, {Tucker},
1006: {Weiland}, {Wollack}, \& {Wright}}]{spergel06}
1007: {Spergel}, D.~N., {Bean}, R., {Dore'}, O., {Nolta}, M.~R., {Bennett}, C.~L.,
1008: {Hinshaw}, G., {Jarosik}, N., {Komatsu}, E., {Page}, L., {Peiris}, H.~V.,
1009: {Verde}, L., {Barnes}, C., {Halpern}, M., {Hill}, R.~S., {Kogut}, A.,
1010: {Limon}, M., {Meyer}, S.~S., {Odegard}, N., {Tucker}, G.~S., {Weiland},
1011: J.~L., {Wollack}, E., \& {Wright}, E.~L. 2007, \apjs, 170, 377
1012:
1013: \bibitem[{{Springel} \& {Hernquist}(2003)}]{springel03}
1014: {Springel}, V., \& {Hernquist}, L. 2003, \mnras, 339, 289
1015:
1016: \bibitem[{{Stern} {et~al.}(2005){Stern}, {Yost}, {Eckart}, {Harrison},
1017: {Helfand}, {Djorgovski}, {Malhotra}, \& {Rhoads}}]{stern05}
1018: {Stern}, D., {Yost}, S.~A., {Eckart}, M.~E., {Harrison}, F.~A., {Helfand},
1019: D.~J., {Djorgovski}, S.~G., {Malhotra}, S., \& {Rhoads}, J.~E. 2005, \apj,
1020: 619, 12
1021:
1022: \bibitem[{{Theuns} {et~al.}(2002){Theuns}, {Schaye}, {Zaroubi}, {Kim},
1023: {Tzanavaris}, \& {Carswell}}]{theuns02}
1024: {Theuns}, T., {Schaye}, J., {Zaroubi}, S., {Kim}, T.-S., {Tzanavaris}, P., \&
1025: {Carswell}, B. 2002, \apjl, 567, L103
1026:
1027: \bibitem[{{Trac} \& {Cen}(2006)}]{trac06b}
1028: {Trac}, H., \& {Cen}, R. 2006, \apj, in press
1029:
1030: \bibitem[{{Trac} \& {Pen}(2006)}]{trac06a}
1031: {Trac}, H., \& {Pen}, U.-L. 2006, New Astronomy, 11, 273
1032:
1033: \bibitem[{{Warren} {et~al.}(2006){Warren}, {Abazajian}, {Holz}, \&
1034: {Teodoro}}]{warren06}
1035: {Warren}, M.~S., {Abazajian}, K., {Holz}, D.~E., \& {Teodoro}, L. 2006, \apj,
1036: 646, 881
1037:
1038: \bibitem[{{Wyithe} \& {Loeb}(2004)}]{wyithe04}
1039: {Wyithe}, J.~S.~B., \& {Loeb}, A. 2004, \nat, 427, 815
1040:
1041: \bibitem[{{Wyithe} \& {Cen}(2007)}]{wyithe07a}
1042: {Wyithe}, J.~S.~B., \& {Cen}, R. 2007, \apj, 659, 890
1043:
1044: \bibitem[Wyithe \& Morales(2007)]{wyithe07b} Wyithe, J.~S.~B., \& Morales, M.~F.\ 2007, \mnras, 379, 1647
1045:
1046: \bibitem[{{Zahn} {et~al.}(2007){Zahn}, {Lidz}, {McQuinn}, {Dutta}, {Hernquist},
1047: {Zaldarriaga}, \& {Furlanetto}}]{zahn07}
1048: {Zahn}, O., {Lidz}, A., {McQuinn}, M., {Dutta}, S., {Hernquist}, L.,
1049: {Zaldarriaga}, M., \& {Furlanetto}, S.~R. 2007, \apj, 654, 12
1050:
1051: \end{thebibliography}
1052:
1053:
1054: \begin{figure}
1055: \plotone{f1.eps}
1056: \caption{Dark matter halo mass functions. Dark matter halos are identified
1057: using a friends-of-friends algorithm with a standard linking
1058: length of 0.2 times the mean inter-particle spacing. Our mass functions are in better agreement with
1059: \citet{warren06} {\it (solid)} than with \citet{press74} {\it (dash)} for $z\lesssim10$.
1060: At $z\gtrsim15$, we under-resolve the halos because of the late starting redshift of $z = 60$.
1061: A smaller simulation starting at $z = 300$ correctly captures the abundance of high redshift
1062: halos {\it (open circles)}.}
1063: \label{fig:halo_mass_func}
1064: \end{figure}
1065:
1066: \begin{figure}
1067: \plotone{f2.eps}
1068: \caption{Comoving star formation rate (SFR). The total SFR {\it (solid)}, from Pop III {\it (long dash)}
1069: and Pop II {\it (dash)} is consistent with that of \citet{hernquist03} {\it (dotted)}.}
1070: \label{fig:star_formation_rate}
1071: \end{figure}
1072:
1073: \begin{figure}
1074: \plotone{f3_small.eps}
1075: \caption{
1076: Distributions of ionization fraction at redshifts z\ $\sim$\ 13.5,\ 11.2,\ 9.3,\
1077: and 7.2. Highly ionized regions are represented by black while ionization fraction
1078: below 50 percent is shown as white. This is a plot of one slice in the simulation box.
1079: Each side of the plot is 100 Mpc/h. The global ionization fractions are $\sim$ 10, 30, 50,
1080: and 90 percents for z\ $\sim$\ 13.5,\ 11.2,\ 9.3,\ and 7.2, respectively.}
1081: \label{fig:HII_evolution}
1082: \end{figure}
1083:
1084: \begin{figure}
1085: \plottwo{f4a.eps}{f4b.eps}
1086: \caption{Number and accumulated volume fraction distributions of HII
1087: bubble volumes {\it (left)} and bubble size distribution {\it (right)}.
1088: At redshifts z = 20.6, 17.2, 13.5, 8.9, 6.3, and 5.7, respectively, the number {\it (solid)}
1089: and accumulated volume distribution {\it (dash)} is given with
1090: the total number of bubbles, ${\rm N_{tot}}$, and the global volume-weighted mean ionization fraction,
1091: ${\rm <x>}$. Before large bubbles that are comparable to the simulation box size appear around
1092: the redshift 6, bubble size distributions show three peaks that have the volumes
1093: of about 0.6, 0.03, and 0.006 ${\rm Mpc^{3}/h^{3}}$. The domination of one large bubble appears
1094: at z $<$ 10 as shown in the accumulated volume fractions of bubbles. In the right plot,
1095: we present the bubble size distribution that is measured as \citet{zahn07} at z =
1096: 13.2, 12.1, 10.5, 9.0, and 8.0 that are correspondent to the global volume-weighted mean ionization
1097: fraction of 0.12, 0.18, 0.32, 0.52, and 0.75 from the small to large peak size.}
1098: \label{fig:size_distribution}
1099: \end{figure}
1100:
1101: \begin{figure}
1102: \plotone{f5.eps}
1103: \caption{
1104: Bubble merger history. The evolution of the bubble size is given
1105: for four positions that are occupied by the four randomly selected
1106: smallest bubbles at z = 21.5. The bubbles that occupy the
1107: four positions are tracked and their volumes are measured.
1108: This bubble merger history is well explained by the change of
1109: a dominant stellar population and its ionizing photon production rate
1110: between z $\sim$ 10 and $\sim$ 15 as shown in Figure 7. During that
1111: transition bubbles experience reconnection to other bubbles.}
1112: \label{fig:merger}
1113: \end{figure}
1114:
1115: \begin{figure}
1116: \plotone{f6.eps}
1117: \caption{Evolution of ionizing photon production rate from Pop III and Pop II stars.
1118: The ionizing photon production rate is estimated by a star forming rate times an
1119: ionizing photon of ${\rm h \nu \ge 13.6 eV}$ production rate per unit stellar mass.}
1120: \label{fig:photon_production}
1121: \end{figure}
1122:
1123: \begin{figure}
1124: \plotone{f7.eps}
1125: \caption{Change of the number and volume fraction of HII bubbles. {\it(upper)} The total
1126: number of the HII bubbles increases until one large bubble dominates. Although the volumes of
1127: the smallest bubbles do not occupy a large fraction of the total volume, their number fraction
1128: dominates other sizes and shows a slightly later decline. {\it(lower)} Although the number
1129: of bubbles shows decrease after ${\rm z} \sim 10$, the total volume fraction of all HII bubbles
1130: increases as cosmic reionization continues. The volume fraction of the largest bubble
1131: at each epoch explains the change of the bubble number is mainly caused by the fact that
1132: intermediate-size bubbles are merged to the largest bubble so that even the second largest bubble
1133: occupy a small fraction of the total volume.}
1134: \label{fig:number}
1135: \end{figure}
1136:
1137: \begin{figure}
1138: \plotone{f8_color.eps}
1139: \caption{
1140: Number density distribution of dark matter halos inside HII bubbles at z $\sim$ 8 and 10. Open triangle
1141: (red),
1142: dot (black), and open rectangle (blue) represent maximum, average, and minimum number density of halos for
1143: a given bubble size, respectively. The ranges of DM halo masses
1144: are ${\rm M_{halo} < 10^{8} ~ M_{\odot}/h, 10^{8}\ M_{\odot}/h \leq M_{halo} < 10^{9}\ M_{\odot}/h}$,
1145: and ${\rm 10^{9}\ M_{\odot}/h \leq M_{halo}}$. [{\it See the electronic edition of the Journal
1146: for a color version of this figure.}]}
1147: \label{fig:halo_size}
1148: \end{figure}
1149:
1150: \begin{figure}
1151: \plotone{f9_color.eps}
1152: \caption{
1153: Ionizing photon production rate per unit volume inside HII bubbles at redshifts 8.3 and 10.2.
1154: The ionizing photon production rate is calculated as used in Figure 6.
1155: Open triangle (red), dot (black), and open rectangle (blue) data points
1156: are maximum, average, and minimum values for different sizes of bubbles, respectively. Here we
1157: plot only bubbles that have stellar masses inside them. Note that ionizing photon production
1158: rate of Pop II stars is comparable to that of Pop III stars for bubbles of
1159: ${\rm V\ \simgt\ 0.6\ Mpc^{3}/h^{3}}$.
1160: [{\it See the electronic edition of the Journal for a color version of this figure.}]}
1161: \label{fig:photon_size}
1162: \end{figure}
1163:
1164: \begin{figure}
1165: \plotone{f10_color.eps}
1166: \caption{Distribution probability of the HII bubble shapes expressed as $e_{1}$ and $e_{2}$ at z = 13.8 and 6.3.
1167: Most bubbles are far from a sphere, i.e. $e_{1}$ and $e_{2} \gg 0$. This trend does not change even though
1168: reionized volume becomes increasing. The distribution is derived from Equation 5.
1169: [{\it See the electronic edition of the Journal for a color version of this figure.}]}
1170: \label{fig:bubble_shape}
1171: \end{figure}
1172:
1173: \begin{figure}
1174: \plotone{f11_color.eps}
1175: \caption{
1176: Distribution probability of bubble shape parameters and volume at z = 13.8 and 10. The bubble shape
1177: has a distribution that does not change even though the fraction of
1178: reionized regions increases between z = 13.8 and 10. While small bubbles have the narrow range of shapes
1179: around $e_{1}$ or $e_{2} \sim 0$ or 1, the bubbles of intermediate size show
1180: the broad range of shapes. The star symbols represent the shape of the largest bubble.
1181: [{\it See the electronic edition of the Journal for a color version of this figure.}]}
1182: \label{fig:shape_size}
1183: \end{figure}
1184:
1185:
1186: \begin{figure}
1187: \plotone{f12.eps}
1188: \caption{Projected mean ionization fraction distribution of a single bubble at z = 13.8.
1189: Its derived shape parameters, $e_{1} \sim 0.93$ and $e_{2} \sim 0.5$, are consistent with
1190: this ionization fraction distribution. Here we set the background around the bubble to have
1191: zero ionization fraction for visualization.}
1192: \label{fig:one_bubble}
1193: \end{figure}
1194:
1195:
1196: \begin{figure}
1197: \plotone{f13_color.eps}
1198: \caption{Non-Gaussianity of reionization field. The statistics of mass-weighted ionization fraction
1199: are given for mean, variance, skewness, and kurtosis from top to bottom. Each statistic is calculated
1200: by sampling 5000 spheres of the comoving radius $5$ {\it (dot line, red)}, $10$
1201: {\it (dash line, green)}, and $20$ {\it (solid line, blue)}
1202: Mpc/h, respectively.
1203: [{\it See the electronic edition of the Journal for a color version of this figure.}]}
1204: \label{fig:non_gaussianity}
1205: \end{figure}
1206:
1207: \begin{figure}
1208: \plotone{f14_color.eps}
1209: \caption{Power spectrum of ${\rm \delta_{HI}}$, ${\rm \delta_{HII}}$, and
1210: ${\rm \delta_{mat}}$. At high redshift the power spectrum of a neutral hydrogen
1211: {\it(dot line, red)} is well matched to that of matter {\it(dash line, green)}
1212: while an ionized hydrogen
1213: {\it(dot-dash line, blue)} shows the match to the matter power spectrum at low redshift. A
1214: linearly developed matter power spectrum {\it(solid line, black)} does not match to a
1215: matter power spectrum at a small scale as a non-linear structure forms.
1216: [{\it See the electronic edition of the Journal for a color version of this figure.}]}
1217: \label{fig:power_spectrum}
1218: \end{figure}
1219:
1220: \begin{figure}
1221: \plotone{f15_color.eps}
1222: \caption{A comparison of the volume-weighted neutral hydrogen fraction $f_{\rm HI}$
1223: from the photon-advection scheme (blue, solid) and ray-tracing scheme (red, dashed)
1224: for radiative transfer. The photon-advection scheme correctly captures the reionization
1225: process up until it is $\sim75\%$ completed by volume. The magnitude of the difference
1226: in $f_{\rm HI}$ (black, dotted) is generally very small and only reaches a maximum value
1227: of 0.05 near complete overlap. [{\it See the electronic edition of the Journal
1228: for a color version of this figure.}]}
1229: \end{figure}
1230:
1231: \begin{figure}
1232: \plotone{f16_color.eps}
1233: \caption{A comparison of the photon-advection scheme and ray-tracing scheme for radiative
1234: transfer using the bias $b(k)$ and cross-correlation $r(k)$ of the neutral hydrogen
1235: fraction $f_{\rm HI}(x)$ field. At $z=8.2$ (red, solid) and 7.5 (green, long-dashed),
1236: both simulations have volume-averaged neutral fractions of 0.5 and 0.25, respectively, and
1237: the spatial evolution of HI is very similar. At $=7.1$ (blue, short-dashed), the correlation
1238: is poor because reionization has progressed slightly faster in the ray-tracing scheme.
1239: However, when the two simulations are compared at the same neutral fraction of 0.1
1240: (magenta, dotted), the correlation is much better, particularly on large scales.
1241: [{\it See the electronic edition of the Journal
1242: for a color version of this figure.}]}
1243: \end{figure}
1244:
1245: \end{document}
1246: