The Astrophysical Journal, 598:73–85, 2003 November 20 E

# 2003. The American Astronomical Society. All rights reserved. Printed in U.S.A.

EARLY STRUCTURE FORMATION AND REIONIZATION IN A COSMOLOGICAL MODEL

WITH A RUNNING PRIMORDIAL POWER SPECTRUM
Naoki Yoshida,1,2 Aaron Sokasian,1 Lars Hernquist,1 and Volker Springel3
Received 2003 May 27; accepted 2003 July 31

ABSTRACT
We study high-redshift structure formation and reionization in a
CDM universe under the assumption
that the spectral power index of primordial density fluctuations is a function of length scale. We adopt a
particular formulation of the ‘‘ running ’’ spectral index (RSI) model as suggested by the combined analysis
of the recent Wilkinson Microwave Anisotropy Probe (WMAP) data and two other large-scale structure
observations. We carry out high-resolution cosmological simulations and use them to study the formation of
primordial gas clouds where the first stars are likely to form. While early structure forms hierarchically in the
RSI model, quite similar to the standard power-law
CDM model, the reduced power on small scales causes
a considerable delay in the formation epoch of low-mass (
106 M ) ‘‘ minihalos ’’ compared with the
CDM
model. The abundance of primordial star-forming gas clouds in such halos also differs by more than an order
of magnitude at z > 15 between the two models. The extremely small number of gas clouds in the RSI model
indicates that reionization is initiated later than z < 15, generally resulting in a smaller total Thomson optical
depth than in the
CDM model. By carrying out radiative transfer calculations, we also study reionization
by stellar populations formed in galaxies. We show that, in order to reionize the universe by z
7, the escape
fraction of ultraviolet photons from galaxies in the RSI model must be as high as 0.6 throughout the redshift
range 5 < z < 18 for a stellar population similar to that of the local universe. Even with a top-heavy initial
mass function representing an early population of massive stars and/or an extraordinarily high photon
emission rate from galaxies, the total optical depth can only be as large as
e
0:1 for reasonable models of
early star formation. The RSI model is thus in conflict with the large Thomson optical depth inferred by the
WMAP satellite.
Subject headings: cosmology: theory — early universe — galaxies: formation —
large-scale structure of universe — methods: n-body simulations
On-line material: color figure

1. INTRODUCTION 2002; Seljak, McDonald, & Makarov 2003), and in the

context of structure formation (Zentner & Bullock 2002).
The origin of matter density fluctuations in the universe is
The spectral index of the primordial power, n, may even
one of the fundamental problems in cosmology. The so-
be a function of length scale. This ‘‘ running ’’ of the spectral
called standard theory of structure formation posits that the
index modulates the relative amplitudes of density fluctua-
present-day clumpy appearance of the universe developed
tions on large to small scales. Interestingly, the combined
through gravitational amplification of an initially very
analysis of the first-year Wilkinson Microwave Anisotropy
smooth, but perturbed matter distribution. Recent obser-
Probe (WMAP) data, the Two-Degree Field (2dF) galaxy
vations provide a consistent picture of the large-scale matter
distribution in the early universe; fluctuations arise from redshift survey, and Ly forest observations favors a cos-
mological model with a ‘‘ running ’’ of the primordial power
adiabatic perturbations whose statistics are described by a
spectrum (Spergel et al. 2003; Peiris 2003). Although the
Gaussian field, as predicted by popular inflationary theo-
WMAP data alone give a best-fit power-law
CDM model
ries. The inflationary models also predict a scale-invariant
with the spectral index n  1 and the fluctuation amplitude
power spectrum of primordial density fluctuations. In fact,
8 ¼ 0:9, the Ly forests observations (Croft et al. 1998,
the generic inflationary prediction is nearly but not precisely
2002; McDonald et al. 2000; Gnedin & Hamilton 2002)
scale-invariant density fluctuations; i.e., the fluctuation
consistently

favor a low-amplitude model on scales
power spectrum scales as PðkÞ / kn with n  1 (see Lyth &
1dk Mpc1 d10, which can be reconciled if the
Riotto 1999 for a recent review). The tilt, or the exact value
primordial power spectrum has a mild tilt (n < 1) or a nega-
for n, has attracted considerable attention in theories of the
physics of the early universe (Covi & Lyth 1999), in astro- tive running (dn=d ln k < 0). For a given normalization on a
large scale, models with negative running predict progres-
nomical observations (Croft et al. 2002; Hannestad et al.
sively reduced linear power on small scales (Kosowsky &
Turner 1995; Hannestad et al. 2002).
Suppressing linear power on small, galactic to subgalactic
1 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, scales may offer a resolution to the often-claimed problems
Cambridge, MA 02138; [email protected], of cold dark matter (CDM) models (e.g., Moore et al. 1999).
[email protected], [email protected]. It could alleviate apparent discrepancies between the
2 National Astronomical Observatory Japan, Mitaka, Tokyo 186-8588,
Japan.
observed structure of dark halos and predictions from
3 Max-Planck-Institut für Astrophysik, Karl-Schwarzschild-Strasse 1, CDM models (Zentner & Bullock 2003). On the other hand,
Garching bei München, Germany; [email protected]. too much reduction of small-scale power may result in a
73
74 YOSHIDA ET AL. Vol. 598

conflict with other observations. Dalal & Kochanek (2002) Unlike semianalytic methods, direct hydrodynamic simula-
and Chiba (2002) argue that the abundance of dark matter tions enable us to locate plausible star-forming sites
substructure in several galactic lens systems inferred from robustly in a cosmological volume. As we discuss later in
observed flux anomalies is consistent with predictions from the present paper, the distribution of the sources strongly
the standard
CDM model (see, however, Bullock & affects the overall topology of the ionized regions.
Zentner 2003 for an attempt to constrain some variant We show that the reduced small-scale power in the RSI
models). In models with reduced small-scale power, non- model causes a considerable delay in the formation epoch of
linear objects with characteristic mass 106–109 M form low-mass halos and the primordial gas clouds within them
late, and hence reionization due to photons from stellar compared to a
CDM counterpart. In view of this result, it
sources formed in low-mass systems is expected to occur late appears unlikely that massive stars in minihalos can con-
(Somerville, Bullock, & Livio 2003). To the contrary, the tribute significantly to reionization in cosmologies with a
measurement of TE-polarization by WMAP suggests a running spectral index, at least for those with parameters
large Thomson optical depth
e ¼ 0:17  0:04, implying similar to what is inferred from the joint WMAPext + 2dF
that reionization took place as early as zreion
17 (Kogut GRS + Ly forest analysis (Spergel et al. 2003). To study
et al. 2003). the impact of ordinary (‘‘ Population II ’’) stars formed in
Somerville et al. (2003) argue that models like the running galaxies and protogalaxies, we carry out ray tracing calcula-
spectral index (RSI) one we consider here are inconsistent tions of reionization by stellar sources. For an initial mass
with the inferred early reionization epoch. Yoshida et al. function (IMF) similar to that in local galaxies, reionization
(2003c) studied early structure formation in a warm dark can be completed at sufficiently early times to account for
matter model and showed that suppressing the linear power the Ly optical depths measured in quasars at z > 6, but in
even on very small scales ðk > 100 Mpc1) can considerably the case of the RSI model only if a large fraction of ionizing
affect the formation of high-redshift star-forming gas photons are able to escape from galaxies. Furthermore,
clouds, making early reionization by massive, metal-free because reionization occurs late in this context, the
stars unlikely to occur early on. Chiu, Fan, & Ostriker Thomson optical depth is too small to be compatible with
(2003) argue that a large optical depth
e > 0:1 may require that suggested by the WMAP measurements.
exotic radiation sources other than a normal stellar popula- Finally, we consider the possibility that the production
tion. Avelino & Liddle (2003) also conclude that running of rate of ionizing photons could be boosted in galaxies and
the primordial power systematically delays the reionization protogalaxies with a top-heavy IMF. In this case, Thomson
epoch. Overall, these theoretical studies suggest that, if the optical depths approaching that favored by the WMAP
large Thomson optical depth observed by the WMAP satel- analysis are possible, but only if the parameters describing
lite is confirmed, the standard power-law (n  1)
CDM these stars are pushed to extreme values.
model is favored over other models with reduced small-scale The paper is organized as follows. We first describe in x 2
power. Thus, the WMAP results appear to suggest a some- the cosmological simulations we used to study the abun-
what contradictory cosmological model. Since the major dance of high-redshift star-forming systems. We present the
reason to favor power-law
CDM in this context is the simulation results in x 3. In x 4, we discuss the possible effect
inferred early epoch of reionization, it is clearly important of the finite size of our simulation volume. In x 5, the abun-
to study and quantitatively compare details of the reioniza- dance of primordial gas clouds and implications are
tion process in the two cosmologies, the RSI model and a described. There, we also give a brief description of the basic
power-law
CDM universe. relation between reionization epoch and the total Thomson
In the present paper, we study early structure formation optical depth. In x 6 we show the results of radiative transfer
in RSI and
CDM universes using cosmological N-body/ calculations in which we consider ordinary stellar popula-
hydrodynamic simulations. By combining these dynamical tions in galaxies and protogalaxies as radiation sources. A
models with radiative transfer calculations, we compute the summary and discussion are given in x 7.
epoch of reionization by various stellar sources in the two
cosmologies and compare the Thomson optical depth to
2. THE N-BODY/HYDRODYNAMIC SIMULATIONS
that measured by the WMAP satellite.
In particular, at high redshifts, ze15, we consider the We use the parallel tree particle-mesh (tree-PM) solver
implications of an early population of massive, metal-free GADGET2 combined with smoothed particle hydrody-
stars (the ‘‘ first stars ’’), which have been hypothesized to namics (SPH) in our simulations. Our treatment of SPH
form in ‘‘ minihalos ’’ (
106 M ). We refer to the sites where employs the conservative entropy formulation of Springel &
these stars can form as ‘‘ primordial gas clouds ’’; i.e., dense Hernquist (2002), which offers several distinct advantages
regions of cold gas that accumulate through the action of over previous versions of SPH. In particular, because the
cooling by molecular hydrogen. In order to identify these energy equation is written with the entropy as the indepen-
primordial gas clouds, our simulations follow the nonequili- dent thermodynamic variable, as opposed to the thermal
brium chemical reactions of a chemically pristine gas and energy, the ‘‘ p dV ’’ term is not evaluated explicitly, reduc-
molecular hydrogen cooling. In an earlier study (Yoshida ing noise from smoothed estimates of, e.g., the density. By
et al. 2003a), we found that nonequilibrium effects associ- including terms involving derivatives of the density with
ated with dynamical heating play a critical role in regulating respect to the particle smoothing lengths, this approach
the supply of molecular gas, contrary to the assumptions explicitly conserves entropy (in regions without shocks),
commonly made in semianalytic studies of this process. even when smoothing lengths evolve adaptively, avoiding
Hence, it is necessary to explicitly follow the dynamics of the problems noted by, e.g., Hernquist (1993). Furthermore,
structure formation to reliably estimate the abundance of this formulation moderates the overcooling problem
early stars. It is also important to specify the source location present in earlier formulations of SPH (Pearce et al. 1999;
accurately when carrying out radiative transfer calculations. Croft et al. 2001; see also Yoshida et al. 2002).
No. 1, 2003 REIONIZATION IN RUNNING SPECTRAL INDEX MODEL 75

The simulations follow the nonequilibrium reactions of

nine chemical species (e, H, H+, He, He+, He++, H2, Hþ 106
2,
H) using the reaction coefficients compiled by Abel et al.
(1997). We use the cooling rate of Galli & Palla (1998) for 104
molecular hydrogen cooling. Further simulation details are LCDM
found in Yoshida et al. (2003a, 2003c). The simulations 102 RSI
employ 2  3243 particles in a comoving volume of 1 Mpc
on a side. We work with
-dominated cosmologies with

P(k) [Mpc3]
matter density m ¼ 0:3, cosmological constant 
¼ 0:7, 100
and the Hubble constant at the present time h ¼ 0:7 in units
of 100 km s1 Mpc1. We set the baryon density to 10-2
b ¼ 0:04 and the normalization parameter to 8 ¼ 0:9.
The transfer functions are computed by the Boltzmann code
of Sugiyama (1995) for the adopted cosmology, and initial 10-4
conditions are realized as described in Yoshida, Sugiyama,
& Hernquist (2003d). The primordial power spectrum for 10-6
the
CDM case is assumed to be PðkÞ / k, whereas that of
the RSI model is given by PðkÞ / kns with 10-8
  0.001 0.01 0.1 1 10 100 1000
1 dns k
ns ¼ ns ðk0 Þ þ ln : ð1Þ k [Mpc-1]
2 d ln k k0
Specifically, we choose k0 ¼ 0:05 Mpc1, ns ðk0 Þ ¼ 0:93, and 1000.0
dns =d ln k ¼ 0:03, as indicated by the WMAPext + 2dF
GRS + Ly analysis (Spergel et al. 2003). Note that we
adopt the same normalization, 8 ¼ 0:9, for both the
power-law
CDM model and the RSI model. Hereafter, we 100.0
refer to the conventional power-law
CDM model simply LCDM
as the ‘‘
CDM model,’’ and to the other as the ‘‘ RSI
model.’’
Figure 1 shows the linear power spectra PðkÞ for the two
∆2(k)

models. In the bottom panel, we plot the variance per unit 10.0 RSI
logarithmic interval in k, D2 ðkÞ ¼ k3 PðkÞ=ð22 Þ, to better
describe
the difference
 between the two models in the range
0:1 < k Mpc1 < 1000. Since the input power spectra are
normalized at a length scale larger than the simulation box 1.0 simulation box

size, the amplitudes of the density fluctuations realized in

the RSI simulation are smaller than in the
CDM simula-
tion on all length scales. Hence, the density fluctuations on
large scales (k
10) in the RSI model should evolve simi- 0.1
larly to those in
CDM but with a slight delay in time. We 0.1 1.0 10.0 100.0 1000.0
note that while the box size of our simulations is sufficiently k [Mpc-1]
large to ensure that the fundamental fluctuation mode in the
simulations remains in the linear regime at the early epochs Fig. 1.—Linear power spectra for the power-law
CDM model (solid
we simulate, it may not sample a fair volume of the universe. line) and for the RSI model (dashed line). In the bottom panel we plot the
We will discuss finite box size effects in x 4. variance D2 ðkÞ linearly extrapolated to the present epoch for the two
models. Note the difference in the plotted range in k between the top and
Figure 2 shows the projected gas distribution at z ¼ 22 bottom panels. The arrow in the bottom panel indicates the length scales
(top panels) and at z ¼ 17 (bottom panels) for the two mod- our simulations probe.
els. We choose these output redshifts such that the density
fluctuation on the largest length scale in the RSI model at
z ¼ 17 resembles that of the
CDM model at z ¼ 22. As dark matter halos. Whereas locating dark halos in pure
expected, the RSI model appears as a delayed
CDM model N-body simulations is relatively simple using conventional
on these scales. Also, a rich small-scale structure develops techniques such as the friends-of-friends (FOF) algorithm,
early on only in the
CDM model, as can be inferred from there are a few subtleties in identifying and assigning prop-
the linear power spectrum (Fig. 1). erties to halos in N-body/hydrodynamic simulations. Since
our main objective is to compare the halo abundance
between the two models, we avoid ambiguities by adopting
3. HALO ABUNDANCE
a simple procedure, as follows.
Since the typical masses of the halos that host early We locate halos by running a FOF groupfinder with link-
primordial gas clouds are 5  105 107 M (Tegmark et al. ing parameter b ¼ 0:2 on the dark matter particles in our
1997; Machacek, Bryan, & Abel 2001; Yoshida et al. simulations. We then assign the total mass to each group as
2003a), the abundance of such small-mass halos is an impor- the number of linked member particles multiplied by the
tant quantity to compare. In order to quantify the difference scaled particle mass mpart ¼ mdm þ mgas , where mdm is the
between the two models, we measure the mass function of dark matter particle mass and mgas is the gas particle mass.
76 YOSHIDA ET AL. Vol. 598

Fig. 2.—Projected gas density distribution at z ¼ 22 (top) and z ¼ 17 (bottom) for the
CDM (left) and RSI (right) models. [See the electronic edition of the
Journal for a color version of this figure.]

We compute the Press-Schechter (PS) mass function to feature is also seen in the
CDM model, the deviation from
which we input the linear power spectrum for the dark mat- the PS mass function is more clearly noticeable for the RSI
ter component. (Note that, for our simulations, separate model. We discuss this issue together with finite box size
transfer functions are used for the dark matter and gas com- effects in x 4.
ponents in setting up the initial conditions; for details, see
Yoshida, Sugiyama, & Hernquist 2003d.) We compare the
mass function of the halos identified in this manner at 4. MASS VARIANCE
z ¼ 17 and 22 in Figure 3.4 For both models, the measured
mass function is overall well fitted by the PS mass function It is well known that structure in CDM models grows
(shown by the solid line for the
CDM model and the hierarchically; i.e., smaller mass objects form earlier and
dashed line for the RSI model). Again, the RSI model at merge to form bigger objects. While this can easily be
z ¼ 17 resembles the
CDM model at z ¼ 22 with the inferred from the variance D2 ðkÞ for the
CDM model,
abundance of low-mass halos being appreciably small. which has larger amplitudes on smaller length scales (see
The halo abundance in the RSI model appears to deviate Fig. 1), the situation is less clear for the RSI model. As
slightly from the PS mass function. Although a similar Figure 1 shows, the variance in the RSI model appears
approximately

 constant over a range of scales 10 <
k Mpc1 < 1000, indicating that nonlinear objects with
4 Only in Figs. 3 and 11, we use units that include the Hubble constant h, widely different masses could form at nearly the same epoch.
to be consistent with those commonly used, h1 M for mass, and It is also worth mentioning that it is not clear whether the
ðh1 MpcÞ3 for abundance. Press-Schechter theory can be used for such models. For
No. 1, 2003 REIONIZATION IN RUNNING SPECTRAL INDEX MODEL 77

105 105

LCDM z=22 RSI z=22

104 104

103 103
dN/dlog10M

dN/dlog10M
102 102

101 101

100 100

10-1 10-1
104 105 106 107 108 109 104 105 106 107 108 109
Msun/h Msun/h

105 105

LCDM z=17 RSI z=17

104 104

103 103
dN/dlog10M
dN/dlog10M

102 102

101 101

100 100

10-1 10-1
104 105 106 107 108 109 104 105 106 107 108 109
Msun/h Msun/h

Fig. 3.—Mass function for the
CDM model (left) and for the RSI model (right) at z ¼ 17 and 22. The solid line is the Press-Schechter mass function for
the
CDM model, and the dashed line is that for the RSI model.

the comparison shown in Figure 3, we calculated the PS the RSI model monotonically increases toward smaller
mass function by formally inputting the linear power mass scales, indicating that less massive objects form at
spectrum for the RSI model. Although the agreement earlier epochs; i.e., structure formation is still expected to be
shown in Figure 3 is reasonably good, it is not trivially ‘‘ bottom-up.’’ Note that the RSI model adopted in the
expected. The flatness of the variance of the RSI model also present paper predicts an effective slope of the power spec-
deserves particular attention for another reason; the hier- trum smaller than 3 for k4200. On very small length
archical nature of the formation of dark matter halos has a scales (k > 1000), or equivalently on low-mass scales
strong influence on the formation of primordial gas clouds (M < 1000 M ), structure formation should then proceed
(Yoshida et al. 2003a). Hence, it is important to verify differently from the ‘‘ bottom-up ’’ picture. Owing to the
whether small-mass (
106 M ) nonlinear objects in the RSI flattening of the variance, the characteristic mass scale of
model grow hierarchically early on. For this purpose, it is the first nonlinear objects may not be well defined. Halos
more appropriate to work with a properly defined mass with masses 10 1000 M should collapse first, and larger
variance rather than the power spectrum. We compute the objects will be formed through mergers of these small halos.
mass variance as Simulations with a substantially higher mass resolution are
Z needed to reveal how halos with such very low masses form,
1
2 ðMÞ ¼ 2 PðkÞW 2 ðkRÞk2 dk ; ð2Þ although it will not be important for the formation of halos
2 with masses
106 M that are crucial for primordial gas
where the top-hat window function is given by cloud formation.
W ðxÞ ¼ 3½sinðxÞ=x3  cosðxÞ=x2 . Figure 4 shows ðMÞ for Let us now examine the consequences of the finite box size
the two models. In the plotted range, the mass variance in of our simulations. We quantify the effect by computing an
78 YOSHIDA ET AL. Vol. 598

1000 LCDM
LCDM
( LCDM kcut )
RSI
RSI

N halo (>Mcrit)
( RSI kcut ) 100
10
σ(M)

10

1
1 15 20 25 30
103 104 105 106 107 108 redshift
M [Msun]
Fig. 5.—Number of halos with mass greater than 7  105 M per cubic
Fig. 4.—Mass variance computed from the power spectra linearly Mpc (comoving) volume. The solid line is the analytic estimate obtained
extrapolated to the present epoch. The thick solid line is for the
CDM from the Press-Schechter mass function for the
CDM model, and the
model, and the thick dashed line is for the RSI model. The thin lines show dashed line is for the RSI model.
the effective mass variance as defined in eq. (3) for each model.

7  105 M . We chose the threshold mass Mcrit based on

effective variance the results of Yoshida et al. (2003a) for star-forming sys-
Z tems. Although the incomplete sampling of the halo mass
1 function owing to the finite box size is appreciable at high
2eff ðMÞ ¼ 2 PðkÞW 2 ðkRÞk2 dk ; ð3Þ
2 kmin redshift, z > 27 for the
CDM model and z > 20 for the
RSI model, overall the halo abundance is in agreement with
where we set the lower boundary kmin to exclude the contri- the PS prediction. Thus, we conclude that the statistics we
bution from large-scale density fluctuations that are missing discuss in the present paper, such as the abundance of
in our simulations. We simply set kmin ¼ 2=Lbox , with Lbox primordial gas clouds, are not significantly affected by the
being the simulation box side length. The thin lines in Figure finite simulation box size. The apparent discrepancy in Fig-
4 are the effective mass variance computed in this manner. ure 5 may be partly due to some inaccuracies of the analytic
For both the models, eff ðMÞ is slightly smaller than ðMÞ mass function itself (see Jenkins et al. 2001; Reed et al.
on the relevant mass scales (
105–106 M ), and thus a sys- 2003), although the simulation results of Jang-Condell &
tematic delay in the characteristic formation epoch (or a Hernquist (2001) indicate that the PS mass function is a
slightly lower abundance at a given epoch) is expected. This good approximation on these mass scales. An important
may account for the small discrepancy in the mass function point to note is that both the analytic estimate and the simu-
plotted in Figure 3. Since we use linearly extrapolated power lation results suggest that the number of halos with mass
spectra to compute ðMÞ and eff ðMÞ, the precise effect of a
106 M , which are crucial for the primordial gas cloud for-
finite box size still remains uncertain. The apparent differ- mation, is an order of magnitude smaller in the RSI model.
ence shown in Figure 4 should be interpreted as qualitatively Finally, we mention that a promising way of compensat-
representing the finite box size effect. Note that the RSI ing for the missing large-scale power would be to employ
model is more affected by the finite box size because the the mode-adding procedure developed by Tormen &
relative fractional power on large scales is larger than in the Bertschinger (1996) and Cole (1997). The method can effec-

CDM model (see Fig. 1). Systematic studies using larger tively extend the dynamic range of cosmological simulations
box simulations are clearly needed to address this issue and thus will be useful for studies of early structure
further. formation in a cosmological context.
By constructing a halo merger history using 30 simulation
outputs dumped frequently between 17 < z < 25, we also
verified that almost all the halos that have masses greater
5. CONTRIBUTION OF THE ‘‘ FIRST STARS ’’
than 7  105 M at z ¼ 17 grew monotonically in mass by
TO REIONIZATION
repeated mergers and continuous accretion of smaller mass
elements, quite similar to what is found in the
CDM model Numerical simulations (e.g., Abel, Bryan, & Norman
at somewhat higher redshifts. As long as we are concerned 2002; Bromm, Coppi, & Larson 2002) suggest that the first
with the abundance of low-mass halos, the results of our stars in the universe were unusually massive and formed in
small box simulations appear to be in reasonable agreement ‘‘ minihalos ’’ at redshifts z > 20, as a consequence of cool-
with the Press-Schechter mass function. In Figure 5 we show ing by molecular hydrogen. It is customary to refer to early,
the number of halos with masses greater than Mcrit ¼ massive stars as Population III. In what follows, we
No. 1, 2003 REIONIZATION IN RUNNING SPECTRAL INDEX MODEL 79

distinguish between massive stars that may have formed in run down to z ¼ 15 to see whether there is a rapid increase
minihalos by molecular cooling (the ‘‘ first stars ’’) from in the number of gas clouds. Although the increase in the
those that could have originated in larger halos where gas number of gas clouds at z
14 17 appears large, the total
can cool by atomic processes. While the nature of the ‘‘ first number is still much smaller than that of the
CDM model.
stars ’’ has been studied in some detail, the stellar As we argued in x 4, some halos grow very rapidly in the
populations expected in slightly more massive halos (the RSI model, which causes considerable perturbation to the
‘‘ first galaxies ’’) is unclear. With modest uncertainty, we formation and the growth of the primordial gas clouds
can quantify the importance of the former to reionization, within them. This may partly explain the extremely small
but numerous assumptions are required to account for number of gas clouds at z
15 in the RSI model. It is inter-
‘‘ nonstandard ’’ IMFs in early protogalaxies, as we discuss esting that the number of the gas clouds in the RSI model is
later in the paper. even smaller than, or at most comparable to, the warm dark
matter case we studied in Yoshida et al. (2003c). For the spe-
5.1. Primordial Gas Cloud Formation cific model employed in Yoshida et al. (2003c), which
assumes a single massive (300 M ) metal-free star is formed
Owing to the delayed formation of ‘‘ minihalos ’’ in the
per gas cloud, reionization by the ‘‘ first stars ’’ is not com-
RSI model, the formation of primordial gas clouds is also pleted at z > 15 in the RSI model. In fact, with the very
expected to occur late. It is in these regions that significant small number of sources, the ionized volume fraction rises
quantities of molecular gas can accumulate, leading to the
only up to a few percent by z ¼ 15. This is clearly in dis-
formation of massive, metal-free stars of the type modeled agreement with the WMAP result of zreion
17. Since the
by, e.g., Abel et al. (2002) and Bromm et al. (2002). We estimated uncertainty in the WMAP result of the reioniza-
emphasize that nonequilibrium processes not typically
tion epoch is rather large (Kogut et al. 2003), it still remains
accounted for in semianalytic estimates, like the dynamical
to be seen whether the particular RSI model we consider is
heating effect identified by Yoshida et al. (2003a), signifi- compatible with the WMAP data. Nevertheless, it is clear
cantly influence the abundance of these star-forming
that the running of the power spectrum is not favored in
regions.
terms of early reionization.
In Figure 6 we plot the number of gas clouds identified in
the two simulations. We define groups of cold (T < 500 K),
dense (nH > 500 cm3) gas particles as ‘‘ gas clouds.’’ We
5.2. Reionization Epoch and the Thomson Optical Depth
locate the gas clouds by running a FOF groupfinder to the
simulation gas particles with a small linking parameter The differential optical depth to Thomson scattering in a
b ¼ 0:05. We then discard from the groups gas particles that small redshift interval dz is
do not satisfy the above conditions. Figure 6 clearly shows
dt
that the total number of gas clouds in the simulated volume d
e ¼ T ne ðzÞc dz ; ð4Þ
differs by more than an order of magnitude between the two dz
models in the redshift range plotted. At z ¼ 17, we identified where T ¼ 6:65  1025 cm2 is the Thomson scattering
66 gas clouds in the CDM model, whereas there is only one cross section, ne ðzÞ is the mean electron number density at z,
gas cloud found in the RSI model. We continued the RSI and c is the speed of light. The mean electron number
density is given by
ne ðzÞ ¼ ne ð0Þð1 þ zÞ3 xe ðzÞ ; ð5Þ
where xe is the ionization fraction. For a flat
CDM model,
100 LCDM HðzÞ ¼ H0 ½
þ m ð1 þ zÞ3 1=2 . Putting xe ¼ 1 for an
interval between z1 and z2 , we obtain the contribution as
RSI
 1=2 z2
2 T ne ð0Þc 3 

d
e jz1 !z2 ¼ ð1 þ zÞ þ  : ð6Þ
3 ðm H 2 Þ1=2 m 
0 z1
N cloud

Consider a simple case in which abrupt and complete

10 reionization occurs at z. Using the WMAP results for the
cosmological parameters, b h2 ¼ 0:0224 and m h2 ¼ 0:135,
and assuming a plasma of primordial abundance with, for
example, fully ionized hydrogen and singly ionized helium,
we obtain the total optical depth to redshift z,
Z z
 1=2
d
e  0:0023 ð1 þ zÞ3 þ2:7  1:93 : ð7Þ
0
1
Figure 7 shows the total optical depth against reioniza-
tion epoch at z ¼ zreion . The WMAP result is indicated at
14 16 18 20 22 24 26 28
e ¼ 0:17 (solid line) with a 1  range 0:17  0:04 (dotted
redshift line). The actual reionization history could be more compli-
cated, but Figure 7 provides the useful insight that a large
Fig. 6.—Number of primordial ‘‘ gas clouds ’’ where massive, metal-free Thomson optical depth, say
e > 0:13, can be achieved only
stars can form by molecular hydrogen cooling. if reionization takes place at an early epoch, zreion e14.
80 YOSHIDA ET AL. Vol. 598

0.30 assumptions about the stellar populations. In this section,

we study reionization by stars in galaxies and protogalaxies
with total masses larger than
4  107 M , where the gas
0.25 can cool rapidly by atomic line transitions. The nature of
the stars forming early in these environments is uncertain. It
is possible, for example, that the IMF will be top-heavy, at
0.20 least initially, perhaps increasing the production rate of ion-
izing photons per unit stellar mass compared to a conven-
WMAP tional IMF inferred for stellar populations in the local
universe.
τe

0.15
In what follows, we begin by considering star formation
in halos in the RSI model by assuming that the stellar popu-
lation is similar to that locally and is unevolving. Namely,
0.10 we adopt a Salpeter (1955) IMF and normalize the star for-
mation rate (SFR) so that it reproduces the Kennicutt
(1998) law. For simplicity, we refer to this description as a
0.05 ‘‘ Population II ’’ SFR. In x 6.4 we examine the overall effect
of relaxing these assumptions. However, it is already clear
that the RSI model will have difficulty accounting for the
0.00 WMAP estimate of the Thomson optical depth for any
0 5 10 15 20 reasonable assumptions about star formation simply
zREION because structure forms too late in this scenario to produce
a large
e , as suggested by Figure 7.
Fig. 7.—Optical depth to Thomson scattering to zreion when abrupt and To study how reionization proceeds in the RSI model, we
complete reionization at zreion is assumed.
run multisource radiative transfer simulations using the
technique of Sokasian, Abel, & Hernquist (2001) and
Our simulation of early structure formation in the RSI Sokasian et al. (2003a, 2003b). In order to locate radiation
model showed that there are only five gas clouds at z ¼ 15 in sources and to compute the IGM density field, we utilize
the simulated volume. As Yoshida et al. (2003c) demon- existing outputs of a large volume
CDM simulation from
strate, approximately 100 very massive stars must be turned Springel & Hernquist (2003b). We normalize the star forma-
on within
1 recombination time in a (1 Mpc)3 volume to tion rate expected for the RSI model using the analytic
cause complete reionization. Owing to the strong radiative description of Hernquist & Springel (2003), which can be
feedback from the first stars, the number of stars formed in applied to any cosmology.
the primordial gas clouds is likely to be limited to 1 (Omukai
& Nishi 1999; Oh et al. 2001). Hence, the number of ‘‘ first 6.1. The Star Formation Rate in the RSI Model
stars ’’ in the RSI model is far too small at z > 15 to reionize
First, we compute the ‘‘ generic ’’ star formation rate for
the universe. The epoch of complete reionization is certainly
the RSI model using the analytic model of Hernquist &
much later than z
15, implying a small total optical depth
Springel (2003). Their approach describes the formation of

e < 0:13.
dark matter halos using an analytic halo mass function
In summary, primordial gas cloud formation that relies
(Sheth & Tormen 1999, 2002) and employs simplified pre-
on molecular hydrogen cooling in low-mass, ‘‘ minihalos ’’
scriptions for the cooling of gas and star formation that are
appears to be extremely inefficient at z > 15 in the RSI
commonly used in semianalytic models of galaxy formation
model, and thus other sources, such as early protogalaxies,
(Kauffmann et al. 1999; Springel et al. 2001; Yoshida et al.
must contribute to reionization for this model to be compat-
2002). This analytic formalism has been tested in detail by
ible with the WMAP results. Furthermore, it is important to
comparing its predictions with those from cosmological
establish whether ordinary stellar sources in galaxies in the
simulations that include hydrodynamics and star formation
RSI model can account for reionization early enough to
(Springel & Hernquist 2003b). A thorough description of
even be consistent with measurements of the Ly optical
the model is given by Hernquist & Springel (2003) and
depth in quasars at z > 6.
references therein.
In particular, Hernquist & Springel (2003) show that
under rather general circumstances the star formation rate
6. REIONIZATION BY STARS IN GALAXIES density is well approximated by the simple form
In x 5, we have shown that low-mass minihalos and pri- p
mordial gas clouds that can harbor massive, metal-free stars _  ðzÞ ¼ _  ð0Þ ; ð8Þ
(the ‘‘ first stars ’’) do not form in sufficient numbers at high 1 þ ð  1Þq expðr Þ
redshift in the RSI model to account for reionization at where
ze15. As we emphasize in x 5.2 and Figure 7, reionization  2=3
must occur early if the WMAP measurement of the HðzÞ
Thomson optical depth is to be accurate. ðzÞ : ð9Þ
H0
The number of larger mass (
107–109 M ) halos is also
predicted to be small for the RSI model. Thus, reionization In equation (8), the factors  and r are fixed by the under-
by stars in systems that are the likely progenitors of galaxies lying cosmological power spectrum, the choices for p and q
will be less efficient than for a
CDM model, for similar reflect the interplay between radiative cooling in halos and
No. 1, 2003 REIONIZATION IN RUNNING SPECTRAL INDEX MODEL 81

1.0000 side. We first calibrate the SFRs of individual sources iden-

tified in the simulation and then convert these into luminosi-
ties. As in Sokasian et al. (2003a), we use the derived SFR
(eq. [8]) as a reference to match the total SFR of the sources.
We carry out the calibration procedure after selecting
0.1000 sources in the simulation outputs in the following two ways.
SFR [MSUNyr-1Mpc-3]

LCDM 6.2.1. Dim-Source Model

In this approach, we identify all the halos found in each
0.0100 output as ‘‘ galaxies ’’ and assign luminosities to them based
RSI on the actual star formation rates, which are computed in a
dynamically consistent way in the cosmological simulation.
The minimum mass of the sources is set by the resolution of
0.0010 the simulation, which is Mmin ¼ 4:3  107 h1 M in our
case. The SFRs of the sources with masses close to the reso-
lution limit are corrected in the same manner as in Sokasian
et al. (2003a; see their x 4.2.1). We then scale the SFRs of all
0.0001 the galaxies by multiplying by a constant factor CL ðzÞ
such that the total SFR within the simulated volume
5 10 15 20 matches that for the RSI model obtained from equation (8).
redshift Note that the scaling factor CL ðzÞ is kept constant for all
Fig. 8.—Star formation rate density in galaxies (‘‘ Population II ’’) as a
the galaxies within one simulation output, but it is
function of redshift for the
CDM (solid line) and RSI (dashed line) models. time-dependent.

6.2.2. Massive-Halo Model

the expansion of the universe, and the normalizing con-
Since the halo abundance itself is lower in the RSI model
stants _  ð0Þ and  depend on the rate at which cold, dense
than in the
CDM cosmology, it may be more appropriate
gas is converted into stars.
to populate a smaller number of objects as radiation sources
The form of equation (8) is motivated by two competing
in the RSI model. Motivated by this, we first compute the
effects that regulate star formation. At sufficiently high red-
number density of halos with virial temperature greater
shifts, collapsed gas is dense and cooling times are short, so
than 104 K in which the gas can cool by atomic line
that star formation is only limited by the gravitational
transitions. The threshold mass is given by
growth of massive halos. At low redshifts, the cooling time
in halos increases owing to the decline in the mean density  3=2
8 1þz
of the universe by cosmic expansion. In this regime, the Mmin ¼ 10 M : ð10Þ
supply of star-forming gas is throttled by the expansion 11
of the universe. Then we use the halo mass function to compute the expected
Based on the multiphase description of star-forming gas abundance of halos with mass greater than Mmin . Let us
described by Springel & Hernquist (2003a), Hernquist & denote this number by Nsource ðzÞ. In the simulation output
Springel (2003) show that the appropriate values for the produced at z, we select Nsource ðzÞ halos in order of their
parameters in equation (8) for the
CDM model employed mass. Luminosities are assigned to the selected sources in
here are p ¼ 2, q ¼ 3, r ¼ 7=4,  ¼ 0:012,  ¼ 0:041, and proportion to their mass, L / Mhalo , such that the total
_  ð0Þ ¼ 0:013 M yr1 Mpc3. SFR summed over the selected sources matches the global
For the RSI cosmology, we use the RSI linear power SFR given by equation (8), with the parameter values
spectrum as input and follow the procedure described in appropriate for the RSI model.
Hernquist & Springel (2003). We find that the resulting SFR
density is again well fitted by equation (8) for the choices 6.2.3. Escape Fraction
p ¼ 1:5, q ¼ 3:05, r ¼ 1:85,  ¼ 0:01,  ¼ 0:057, and
_  ð0Þ ¼ 0:0176 M yr1 Mpc3. Finally, the ionizing flux from the sources is modulated
In Figure 8, we compare the SFR density estimated in this by the ‘‘ escape fraction,’’ fesc , which describes the probabil-
manner for the two cosmologies. The RSI model predicts a ity that individual photons can escape from the galaxy in
significantly smaller SFR than the
CDM model, in which they were produced into the surrounding inter-
qualitative agreement with the semianalytic estimates of galactic medium (IGM). Sokasian et al. (2003a) studied
Somerville et al. (2003). We use the forms for the SFR den- reionization by stellar sources in the
CDM universe by
sity in our ray-tracing calculations only within the restricted adopting several values for fesc . Based on a detailed compar-
redshift range 5 < z < 20 to compute radiation source ison with the observational results of the z ¼ 6:28 quasar
luminosities. (SDSS 1030.10+0524) in Becker et al. (2001), they conclude
that fesc ¼ 0:2 is most consistent with the observed Ly flux
transmittance. We take this fesc ¼ 0:2 model in Sokasian
6.2. Source Luminosities
et al. (2003a) as our fiducial case for the
CDM model. For
The second step is to prescribe the luminosity of the the RSI model, our zeroth-order requirement is that reioni-
sources (‘‘ galaxies ’’) located in the simulations. Specifi- zation should be completed by z ¼ 7. After several experi-
cally, we use the outputs from run Q5 of Springel & ments, we found that setting fesc > 0:6 is necessary for the
Hernquist (2003b), which simulates a cube of 14.3 Mpc on a RSI model to satisfy this condition. We thus set fesc ¼ 0:6 as
82 YOSHIDA ET AL. Vol. 598

our basic choice for the RSI model and compare the results 0.10
with the fiducial
CDM model.

6.3. Numerical Results for ‘‘ Population II ’’ Star Formation 0.08 LCDM (f_esc = 0.2)
We compute the total volume filling factor of the ionized RSI (f_esc = 0.6)
medium QH ii from the results of our ray-tracing calcula-
tions. Figure 9 shows the resulting ionized volume fraction, 0.06
and Figure 10 shows the Thomson optical depth
e as a

τe
function of redshift. In the figures, we show only the results
for our ‘‘ massive-halo model ’’ of source selection for the 0.04
RSI model. Comparing the results for the dim-source model
and the massive-halo model, we found that there is no sub-
stantial difference between the two in either
e ðzÞ or QH ii ðzÞ.
We note, however, that the overall morphology of the ion- 0.02
ized regions looks somewhat different between the two
cases, because there are more sources in the dim-source
model that reside in less dense regions. Nevertheless, for
0.00
global quantities such as the total optical depth, the two 0 2 4 6 8 10 12 14
models are found to be quite similar. redshift
On the other hand, the results for the
CDM and the RSI
models differ substantially. As Figure 9 shows, the ionized Fig. 10.—Thomson optical depth computed from the outputs of our
simulations for the
CDM model with fesc ¼ 0:2 (solid line) and for the RSI
volume fraction in the RSI model remains small until z
9 model with fesc ¼ 0:6 (dashed line) for ‘‘ Population II ’’ stars.
and then rapidly increases close to unity at z
7. At later
epochs, zd9, the abundance of high-mass (
109 M ) halos
rapidly increases, and the total star formation (and hence shown in Figure 10, the resulting total optical depth
the photon production) is dominated by such large halos.
e ¼ 0:055 for the RSI model is still small, owing to the
This can also be understood in terms of the halo mass func- small QH ii at earlier epochs. We emphasize that a large
tion at z ¼ 7 shown in Figure 11. By this time, the mass Thomson optical depth can only be achieved if reionization
functions for the two models are already quite similar, with occurs early (x 6). In other words, after reionization is com-
the difference being only a constant factor of roughly 2 pleted, the additional photon supply does not matter in the
(compared with the much larger difference at z ¼ 17 at total optical depth as long as the IGM is kept almost fully
smaller mass scales shown in Fig. 3). Namely, the RSI ionized thereafter, as implied by Figure 7. (Note that here
model quickly catches up with the
CDM model in the we discuss only hydrogen reionization.)
abundance of halos around z
7 11 (see also the increasing Our fiducial
CDM case with fesc ¼ 0:2 predicts a larger
SFR in Fig. 8), which causes the rapid increase in the ionized total optical depth than the RSI model with fesc ¼ 0:6 by
volume fraction as seen in Figure 9. Reionization is nearly D
e
0:01. The difference mostly comes from the fact that
completed (QH ii > 0:95) at z ¼ 8 in the
CDM model, QH ii is still quite small at z > 9 in the RSI model.
whereas it occurs at z  7:3 in the RSI model. However, as Interestingly, none of the ‘‘ conventional ’’ scenarios we

1.0
103
Tvir4 sources
0.8 102 LCDM
ionized volume fraction

LCDM (f_esc = 0.2) RSI

(f_esc = 0.6)
101
0.6 RSI
dN/dlog10M

100
0.4
10-1

0.2 10-2

10-3 z=7
0.0
6 8 10 12 14
10-4
redshift
107 108 109 1010 1011 1012
Fig. 9.—Ionized volume fraction for the
CDM model with fesc ¼ 0:2 Msun/h
(solid line) and for the RSI model with fesc ¼ 0:6 (dashed line) for
‘‘ Population II ’’ stars. Fig. 11.—Press-Schechter mass function at z ¼ 7
No. 1, 2003 REIONIZATION IN RUNNING SPECTRAL INDEX MODEL 83

have considered for ‘‘ Population II ’’ star formation repro- 0.20

duces the claimed high optical depth from WMAP of
0:13 <
e < 0:21 (1 ; Kogut et al. 2003). It is clear that
these models are inconsistent with the WMAP measurement RSI boost
and thus may need, for example, additional radiation 0.15
sources to ionize the IGM early on.
In related work, we have shown that an early generation
of massive, metal-free stars in minihalos supplemented by
‘‘ ordinary ’’ star formation at lower redshifts to account for

τe
0.10
the spectra of high-redshift quasars can yield a Thomson
optical depth approaching that of the WMAP measurement
in a
CDM cosmology (Yoshida et al. 2003c; Sokasian
et al. 2003b), along the lines suggested by Haiman & Holder
(2003) and Cen (2003). However, it does not appear that a 0.05
similar conclusion will obtain for the RSI model because of
the low abundance of minihalos at high redshifts, as we
found in the case of a warm dark matter model with particle 0.00
mass 10 keV (Yoshida et al. 2003c).
0 5 10 15 20
redshift
6.4. Star Formation in Galaxies with
Boosted Photon Emission Fig. 12.—Thomson optical depth for the RSI model in which the photon
emission rate is ‘‘ boosted ’’ according to eq. (11). The filled square is the
The results described above indicate that the RSI model is WMAP result with its 1  error bar.
incompatible with a high Thomson optical depth like that
suggested by the WMAP measurements. A remaining loop- ing photons per second, a 50 times higher photon emission
hole in this argument is that the stellar population in rate, can be maintained in this case, assuming such very mas-
galaxies at early times may have differed from that in our sive stars are continuously formed. (Note that these massive
underlying model so that a larger number of ionizing stars are typically short-lived, with a main-sequence lifetime
photons were produced from galaxies at z > 10 than we of about 2 million years.) We thus conclude that a 100 times
assumed above (e.g., Ciardi, Ferrara, & White 2003; Wyithe higher photon production rate than assumed in our original
& Loeb 2003; Sokasian et al. 2003a). We first discuss such RSI model serves as a conservative upper limit.
possibilities quantitatively and then show numerical results We examine the overall effect of such an additional
for a model that enhances the photon production rate in photon supply by ‘‘ boosting ’’ the photon emission rate of
galaxies. the sources in the RSI massive-halo model. Sokasian et al.
For a Salpeter-type power-law IMF, the hydrogen ioniz- (2003a) found that boosting the photon emission rate at
ing photon emission rate of zero-metallicity stars is larger high redshifts progressively shifts the reionization epoch
than that of a Population II (i.e., having metals) counterpart (see their Fig. 13). Motivated by their result, we employ a
only by about 50% (Tumlinson & Shull 2000; see Leitherer similar prescription that increases the emission rate of
et al. 1999 for low-metallicity cases). The stellar IMF in the individual sources according to
first galaxies may even have had an extreme top-heavy
shape, with massive stars exclusively forming out of a N_ ph;eff ¼ exp½0:53ðz  6Þ  N_ ph ; ð11Þ
primeval gas (Schwarzschild & Spitzer 1953; Matsuda, Sato,
& Takeda 1969; Yoshii & Saio 1986; Larson 1998; Abel et where N_ ph and N_ ph;eff are the original and the boosted pho-
al. 2002; Bromm et al. 2002). Very massive Population III ton emission rates, respectively. With this parameterization,
stars are efficient emitters of UV photons, with the hydrogen the ionizing photon emission rate of the galaxies steeply
ionizing photon emission rate per stellar mass being up to increases toward higher redshift, modeling either a transi-
10–20 times higher than that for ordinary stellar popula- tion of the stellar IMF or a contribution from massive
tions (Bromm, Kudritzki, & Loeb 2001; Schaerer 2003). It is Population III stars or both. This rather extreme model
thus conceivable that, in early galaxies, the photon emission requires effectively more than a 10 times higher photon
rate per stellar mass was
10 times higher than that of the emission rate at z > 10 and more than 100 times higher at
‘‘ Population II ’’ case we considered above. z > 15. We carry out radiative transfer simulations for the
More specifically, the predicted star formation rate for RSI model with this luminosity ‘‘ boost.’’ Figure 12 shows
the RSI model shown in Figure 8 is
5  104 M the resulting
e ðzÞ. The reionization epoch is found to be
yr1 Mpc3 at z ¼ 15. Assuming a constant conversion shifted to earlier times, z
10, than in the case considered
factor of 1053 s1 ionizing photons per star formation rate of in x 6.3, and consequently
e is greatly increased compared
1 M yr1 for an ordinary stellar population with solar met- to the fiducial RSI model. However, the total optical depth
allicity (Madau, Haardt, & Rees 1999), we obtain a total for this boost model reaches
e
0:12 at z ¼ 0, still lower
rate of
5  1049 ionizing photons per second per cubic than the 1  range of the WMAP result.
comoving 1 Mpc volume at z ¼ 15 for our fiducial RSI
model. If all the formed stars have a mass of 300 M , the
7. SUMMARY
star formation rate of ~5  104 M yr1 Mpc3 corre-
sponds to the formation of five such stars every 3 million Our numerical simulations of early structure formation
years within the volume. The calculation of Bromm et al. show that primordial gas cloud formation in low-mass
(2001) suggests that a total emission rate of 2:5  1051 ioniz- ‘‘ minihalos ’’ is very inefficient at z > 15 in the RSI model,
84 YOSHIDA ET AL. Vol. 598

making it unlikely that the ‘‘ first stars ’’ contribute signifi- In the end, we are left with a contradiction: the running
cantly to reionization in this scenario. Using radiative of the primordial power spectrum favored by the
transfer calculations, we also show that reionization by WMAPext + 2dF + Ly analysis and the high Thomson
ordinary (‘‘ Population II ’’) stellar sources in galaxies is optical depth measured by the WMAP satellite appear to be
completed late in the RSI model if we employ conventional inconsistent with one other. We have discussed various pos-
models of star formation. In order to be compatible with the sibilities to resolve the conflict and conclude that a rather
epoch of the end stage of reionization inferred from high- radical solution or a fine-tuned combination, if any is possi-
redshift quasar observations, the photon escape fraction ble, of the proposed resolutions are necessary. Models with
from galaxies must be large, fesc e0:6. We found that the a weaker running of the primordial power spectrum will
resulting total Thomson optical depth in this case is alleviate the discrepancy slightly, but the overall results

e  0:055, in apparent conflict with the recent WMAP would not be changed. If the high Thomson optical depth is
measurements of CMB polarization. confirmed and the running of the primordial power spec-
In order for the total optical depth to be as large as trum turns out to be real, it may be necessary to invoke

e > 0:1, reionization must be completed before z
10 (Fig. exotic radiation sources other than stellar populations or
7). Ciardi et al. (2003) and Wyithe & Loeb (2003) suggest some ionization mechanisms. Such scenarios include decay-
that employing an effectively top-heavy IMF may be neces- ing particles (Sciama 1982; Dodelson & Jubas 1992; Hansen
sary to cause early reionization. Cen (2003) argues that a & Haiman 2003) or the formation of early miniquasars
large Thomson optical depth could be obtained if the (Eisenstein & Loeb 1995; Sasaki & Umemura 1996).
‘‘ Population III era ’’ is prolonged so that massive stars The prospects for observationally resolving this discrep-
continuously form until quite low redshift. If a large volume ancy in the near future appear bright. Data from planned
of the IGM could be kept chemically pristine until low red- CMB polarization experiments, by the continued operation
shift (z
10) as a result of the very low global star forma- of WMAP, and post-WMAP observatories such as Planck,
tion rate in the RSI model, the stellar IMF in galaxies may will pin down a precise value for
e . Ly forest observations
be top-heavy (Schneider et al. 2002; Mackey, Bromm, & exploiting a large sample of SDSS quasars will place a
Hernquist 2003). Unfortunately, little is presently known tighter constraint on the matter power spectrum on large
about star formation in a primordial gas in high-mass scales, as will the SDSS galaxy redshift survey. In the longer
halos—the ‘‘ first galaxies ’’—and clearly needs to be term, it may be possible to map out the evolution of reioniza-
addressed using numerical simulations as well as theoretical tion from redshifted 21 cm emission using instruments such
modeling (Larson 1998; Omukai 2001; Oh & Haiman 2002). as the Square Kilometer Array5 or the Low Frequency
We have explored the possibility of enhancing the photon Array6 (e.g., Madau, Meiksin, & Rees 1997; Tozzi et al.
emission rate by employing a simple prescription. The 2000; Iliev et al. 2002; Ciardi & Madau 2003). For example,
model parameters were chosen such that the photon emis- Furlanetto, Sokasian, & Hernquist (2003) show that fre-
sion rate is close to a plausible maximum bound. Our analy- quency fluctuations can be used to distinguish between
sis has shown that, whereas the extreme boost model does models with a different number of reionization epochs.
predict an earlier reionization epoch zreion
10 and a larger Analyses of these various high-precision data promise to
total optical depth
e
0:12, the result is still marginally provide a more complete picture of the matter density distri-
inconsistent with the WMAP data. bution in the early universe over a wide range of scales and
Recent numerical simulations by Bromm, Yoshida, & its relationship to the formation of stars and galaxies.
Hernquist (2003) show that early Population III supernovae
quickly pollute the surrounding IGM with metals. If many We thank James Bullock, Volker Bromm, and Ravi Sheth
massive population III stars are formed in ‘‘ minihalos ’’ for helpful comments on the earlier draft. N. Y. acknowl-
and explode as hyperenergetic supernovae, the global metal edges support from the Japan Society of Promotion of
enrichment of the IGM is quickly achieved to such a degree Science Special Research Fellowship. This work was sup-
that a transition of the stellar IMF is caused at a very early ported in part by NSF grants ACI 96-19019, AST 98-02568,
epoch (N. Yoshida, V. Bromm, & L. Hernquist 2003b, in AST 99-00877, and AST 00-71019. The simulations were per-
preparation). Then, the large photon production rate at formed at the Center for Parallel Astrophysical Computing
z > 6 we hypothesized as coming from metal-free stars at the Harvard-Smithsonian Center for Astrophysics.
cannot be maintained for a sufficiently long time, and the
reionization history in the RSI model will be just as shown
in Figure 9, with a small total optical depth (
e < 0:06) as 5 Additional information is available at http://www.usska.org.
shown in Figure 10. 6 Additional information is available at http://www.lofar.org.

REFERENCES
Abel, T., Anninos, P., Norman, M. L., & Zhang, Y. 1997, NewA, 2, 181 Ciardi, B., Ferrara, A., & White, S. D. M. 2003, MNRAS, submitted
Abel, T., Bryan, G. L., & Norman, M. L. 2002, Science, 295, 93 (astro-ph/0302451)
Avelino, P. P., & Liddle, A. R. 2003, preprint (astro-ph/0305357) Ciardi, B., & Madau, P. 2003, ApJ, 596, 1
Becker, R. H., et al. 2001, AJ, 122, 2850 Cole, S. 1997, MNRAS, 286, 38
Bromm, V., Coppi, P. S., & Larson, R. B. 2002, ApJ, 564, 23 Covi, L., & Lyth, D. H. 1999, Phys. Rev. D, 59, 63515
Bromm, V., Kudritzki, R. P., & Loeb, A. 2001, ApJ, 552, 464 Croft, R. A. C., Di Matteo, T., Davé, R., Hernquist, L., Katz, N., Fardal,
Bromm, V., Yoshida, N., & Hernquist, L. 2003, ApJ, in press M. A., & Weinberg, D. H. 2001, ApJ, 557, 67
Bullock, J. S., & Zentner, A. R. 2003, in Dark Matter 2002: Proc. of the 5th Croft, R. A. C., Weinberg, D. H., Bolte, M., Burles, S., Hernquist, L., Katz,
International UCLA Symposium on Sources and Detection of Dark N., Kirkman, D., & Tytler, D. 2002, ApJ, 581, 20
Matter and Dark Energy in the Universe, ed. D. Cline (Amsterdam: Croft, R. A. C., Weinberg, D. H., Katz, N., & Hernquist, L. 1998, ApJ,
Elsevier Science) (astro-ph/0207534) 495, 44
Cen, R. 2003, ApJ, 591, 12 Dalal, N., & Kochanek, C. S. 2002, ApJ, 572, 25
Chiba, M. 2002, ApJ, 565, 17 Dodelson, S, & Jubas, J. M. 1992, Phys. Rev. D, 45, 1076
Chiu, W. A., Fan, X., & Ostriker, J. P. 2003, preprint (astro-ph/0304234) Eisenstein, D. J., & Loeb, A. 1995, ApJ, 443, 11
No. 1, 2003 REIONIZATION IN RUNNING SPECTRAL INDEX MODEL 85

Furlanetto, S., Sokasian, A., & Hernquist, L. 2003, MNRAS, submitted Salpeter, E. E. 1955, ApJ, 121, 161
(astro-ph/0305065) Sasaki, S., & Umemura, M. 1996, ApJ, 462, 104
Galli, D., & Palla, F. 1998, A&A, 335, 403 Schaerer, D. 2003, A&A, 397, 527
Gnedin, N. Y., & Hamilton, A. J. S. 2002, MNRAS, 334, 107 Schneider, R., Ferrara, A., Natarayan, P., & Omukai, K. 2002, ApJ, 571,
Haiman, Z., & Holder, G. 2003, preprint (astro-ph/0302403) 30
Hannestad, S., Hansen, S. H., Villante, F. L., & Hamilton, A. J. S. 2002, Schwarzschild, M., & Spitzer, L. 1953, Observatory, 73, 77
Astropart. Phys., 17, 375 Sciama, D. W. 1982, MNRAS, 198, 1
Hansen, S. H., & Haiman, Z. 2003, preprint (astro-ph/0305126) Seljak, U., McDonald, P., & Makarov, A. 2003, MNRAS, 342, L79
Hernquist, L. 1993, ApJ, 404, 717 Sheth, R. K., & Tormen, G. 1999, MNRAS, 308, 119
Hernquist, L., & Springel, V. 2003, MNRAS, 341, 1253 ———. 2002, MNRAS, 329, 61
Iliev, I. T., Shapiro, P. R., Ferrara, A., & Martel, H. 2002, ApJ, 572, L123 Sokasian, A., Abel, T., & Hernquist, L. 2001, NewA, 6, 359
Jang-Condell, H., & Hernquist, L. 2001, ApJ, 548, 68 Sokasian, A., Abel, T., Hernquist, L., & Springel, V. 2003a, MNRAS, 344,
Jenkins, A., Frenk, C. S., White, S. D. M., Colberg, J. M., Cole, S., Evrard, 607
A. E., Couchman, H. M. P., & Yoshida, N. 2001, MNRAS, 321, 372 Sokasian, A., Yoshida, N., Abel, T., Hernquist, L., & Springel, V. 2003b,
Kauffmann, G., Colberg, J. M., Diaferio, A., & White, S. D. M. 1999, MNRAS, submitted (astro-ph/0307451)
MNRAS, 303, 188 Somerville, R., Bullock, J. S., & Livio, M. 2003, ApJ, 593, 616
Kennicutt, R. C. 1998, ApJ, 498, 541 Spergel, D. N., et al. 2003, ApJS, 148, 175
Kogut, A., et al. 2003, ApJS, 148, 161 Springel, V., & Hernquist, L. 2002, MNRAS, 333, 649
Kosowsky, A., & Turner, M. S. 1995, Phys. Rev. D, 52, 1739 ———. 2003a, MNRAS, 339, 289
Larson, R. B. 1998, MNRAS, 301, 569 ———. 2003b, MNRAS, 339, 312
Leitherer, C., et al. 1999, ApJS, 123, 3 Springel, V., White, S. D. M., Kauffmann, G., & Tormen, G. 2001,
Lyth, D. H., & Riotto, A. A. 1999, Phys. Rep., 314, 1 MNRAS, 328, 726
Machacek, M., Bryan, G. L., & Abel, T. 2001, ApJ, 548, 509 Sugiyama, N. 1995, ApJS, 100, 281
Mackey, J., Bromm, V., & Hernquist, L. 2003, ApJ, 586, 1 Tegmark, M., Silk, J., Rees, M., Blanchard, A., Abel, T., & Palla, F. 1997,
Madau, P., Haardt, F., & Rees, M. J. 1999, ApJ, 514, 648 ApJ, 474, 1
Madau, P., Meiksin, A., & Rees, M. J. 1997, ApJ, 475, 429 Tormen, G., & Bertschinger, E. 1996, ApJ, 472, 14
Matsuda, T., Sato, H., & Takeda, H. 1969, Prog. Theor. Phys., 42, 219 Tozzi, P., Madau, P., Meiksin, A., & Rees, M. 2000, ApJ, 528, 597
McDonald, P., et al. 2000, ApJ, 543, 1 Tumlinson, J., & Shull, J. M. 2000, ApJ, 528, L65
Moore, B., Ghigna, S., Governato, F., Lake, G., Quinn, T., Stadel, J., & Wyithe, J. S., & Loeb, A. 2003, ApJ, 588, L69
Tozzi, P. 1999, ApJ, 524, L19 Yoshida, N., Abel, T., Hernquist, L., & Sugiyama, N. 2003a, ApJ, 592, 645
Oh, S.-P., & Haiman, Z. 2002, ApJ, 569, 558 Yoshida, N., Sokasian, A., Hernquist, L., & Springel, V. 2003c, ApJ, 591,
Oh, S.-P., Nollett, K. M., Madau, P., & Wasserburg, G. J. 2001, ApJ, 562, L1
L1 Yoshida, N., Stoehr, F., Springel, V., & White, S. D. M. 2002, MNRAS,
Omukai, K. 2001, ApJ, 546, 635 335, 762
Omukai, K., & Nishi, R. 1999, ApJ, 518, 64 Yoshida, N., Sugiyama, N., & Hernquist, L. 2003d, MNRAS, 344, 481
Pearce, F. R., et al. 1999, ApJ, 521, L99 Yoshii, Y., & Saio, H. 1986, ApJ, 301, 587
Peiris, H. V. 2003, ApJS, 148, 213 Zentner, A. R., & Bullock, J. S. 2002, Phys. Rev. D, 66, 043003
Reed, D., et al. 2003, MNRAS, in press (astro-ph/0301270) ———. 2003, ApJ, 598, 49