Cosmic Microwave Background Anisotropies

23 Jul 2002 14:3 AR AR166-AA40-06.tex AR166-AA40-06.
SGM LaTeX2e(2002/01/18) P1: GJC

10.1146/annurev.astro.40.060401.093926
Annu. Rev. Astron. Astrophys. 2002. 40:171216
doi: 10.1146/annurev.astro.40.060401.093926
Copyright c 2002 by Annual Reviews. All rights reserved
COSMIC MICROWAVE BACKGROUND
ANISOTROPIES
Wayne Hu
1,2,3
and Scott Dodelson
2,3
1
Center for Cosmological Physics, University of Chicago, Chicago, Illinois 60637;
email: [email protected]
2
NASA/Fermilab Astrophysics Center, P.O. Box 500, Batavia, Illinois 60510;
email: [email protected]
3
Department of Astronomy and Astrophysics, University of Chicago, Chicago,
Illinois 60637
Key Words background radiation, cosmology, theory, dark matter, early universe
I Abstract Cosmic microwave background (CMB) temperature anisotropies have
and will continue to revolutionize our understanding of cosmology. The recent discov-
ery of the previously predicted acoustic peaks in the power spectrum has established a
working cosmological model: a critical density universe consisting of mainly dark mat-
ter and dark energy, which formed its structure through gravitational instability from
quantumuctuations during an inationary epoch. Future observations should test this
model and measure its key cosmological parameters with unprecedented precision. The
phenomenology and cosmological implications of the acoustic peaks are developed in
detail. Beyond the peaks, the yet to be detected secondary anisotropies and polarization
present opportunities to study the physics of ination and the dark energy. The analysis
techniques devised to extract cosmological information from voluminous CMB data
sets are outlined, given their increasing importance in experimental cosmology as a
whole.
INTRODUCTION
The eld of cosmic microwave background (CMB) anisotropies has dramatically
advanced over the past decade (see White et al. 1994), especially on its obser-
vational front. The observations have turned some of our boldest speculations
about our Universe into a working cosmological model: namely, that the Universe
is spatially at, consists mainly of dark matter and dark energy, with the small
amount of ordinary matter necessary to explain the light element abundances, and
all the rich structure in it formed through gravitational instability from quantum
mechanical uctuations when the Universe was a fraction of a second old. Obser-
vations over the coming decade should pin down certain key cosmological param-
eters with unprecedented accuracy (Knox 1995, Jungman et al. 1996, Bond et al.
1997, Zaldarriaga et al. 1997, Eisenstein et al. 1999). These determinations will
0066-4146/02/0922-0171$14.00 171
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
23 Jul 2002 14:3 AR AR166-AA40-06.tex AR166-AA40-06.SGM LaTeX2e(2002/01/18) P1: GJC
172 HU

DODELSON
have profound implications for astrophysics, as well as other disciplines. Particle
physicists, for example, will be able to study neutrino masses, theories of ination
impossible to test at accelerators, and the mysterious dark energy or cosmological
constant.
For the 28 years between the discovery of the CMB (Penzias & Wilson 1965)
and the COBE DMR detection of 10
5
uctuations in its temperature eld across
the sky (Smoot et al. 1992), observers searched for these anisotropies but found
none except the dipole induced by our own motion (Smoot et al., 1977). They
learned the hard way that the CMB is remarkably uniform. This uniformity is in
stark contrast to the matter in the Universe, organized in very nonlinear struc-
tures like galaxies and clusters. The disparity between the smooth photon distri-
bution and the clumpy matter distribution is due to radiation pressure. Matter
inhomogeneities grow owing to gravitational instability, but pressure prevents the
same process from occuring in the photons. Thus, even though both inhomo-
geneities in the matter in the Universe and anisotropies in the CMB apparently
originated from the same source, these appear very different today.
Because the photon distribution is very uniform, perturbations are small, and
linear response theory applies. This is perhaps the most important fact about CMB
anisotropies. Because they are linear, predictions can be made as precisely as their
sources are specied. If the sources of the anisotropies are also linear uctuations,
anisotropy formation falls in the domain of linear perturbation theory. There are
then essentially no phenomenological parameters that need to be introduced to
account for nonlinearities or gas dynamics or any other of a host of astrophysical
processes that typically afict cosmological observations.
CMBanisotropies in the working cosmological model, which we briey review
in Observables, fall almost entirely under linear perturbation theory. The most
important observables of the CMB are the power spectra of the temperature and
polarization maps. Theory predicted, and now observations conrm, that the tem-
perature power spectrumhas a series of peaks and troughs. In Acoustic Peaks, we
discuss the origin of these acoustic peaks and their cosmological uses. Although
they are the most prominent features in the spectrum and are the focus of the cur-
rent generation of experiments, future observations will turn to even ner details,
potentially revealing the physics at the two opposite ends of time. Some of these
are discussed in Beyond the Peaks. Finally, the past few years have witnessed
important new advances, introduced in Data Analysis, from a growing body of
CMB data analysts on how best to extract the information contained in CMB data.
Some of the fruits of this labor have already spread to other elds of astronomy.
OBSERVABLES
Standard Cosmological Paradigm
Whereas a review of the standard cosmological paradigm is not our intention (see
Narkilar & Padmanabhan 2001 for a critical appraisal), we briey introduce the
observables necessary to parameterize it.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
CMB ANISOTROPIES 173
The expansion of the Universe is described by the scale factor a(t), set to
unity today, and by the current expansion rate, the Hubble constant H
0
=100 h km
sec
1
Mpc
1
, with h .0.7 (Freedman et al. 2001; compare Theureau et al. 1997,
Sandage et al. 2000). The Universe is at (no spatial curvature) if the total density
is equal to the critical density,
c
=1.88 h
2
10
29
g cm
3
; it is open (negative
curvature) if the density is less than this and closed (positive curvature) if greater.
The mean densities of different components of the Universe control a(t) and are
typically expressed today in units of the critical density O
i
, with an evolution with
a specied by equations of state n
i
=p
i
,
i
, where p
i
is the pressure of the ith
component. Density uctuations are determined by these parameters through the
gravitational instability of an initial spectrum of uctuations.
The working cosmological model contains photons, neutrinos, baryons, cold
dark matter, and dark energy with densities proscribed within a relatively tight
range. For the radiation, O
r
=4.17 10
5
h
2
(n
r
=1,3). The photon contribution
to the radiation is determined to high precision by the measured CMBtemperature,
T =2.728 0.004 K (Fixsen et al. 1996). The neutrino contribution follows from
the assumption of three neutrino species, a standard thermal history, and a negligi-
ble mass m
_1 eV. Massive neutrinos have an equation of state n
=1,3 0 as
the particles become nonrelativistic. For m
1 eVthis occurs at a 10
3
and can
leave a small but potentially measurable effect on the CMB anisotropies (Ma &
Bertschinger 1995, Dodelson et al. 1996).
For the ordinary matter or baryons, O
b
0.02 h
2
(n
b
0), with statistical un-
certainties at about the 10% level determined through studies of the light element
abundances (for reviews, see Boesgaard & Steigman 1985, Schramm & Turner
1998, Tytler et al. 2000). This value is in strikingly good agreement with that im-
plied by the CMBanisotropies themselves, as we shall see. There is very strong ev-
idence that there is also substantial nonbaryonic dark matter. This dark matter must
be close to cold (n
m
=0) for the gravitational instability paradigmto work (Peebles
1982), and when added to the baryons gives a total in nonrelativistic matter of O
m
.
1,3. Because the Universe appears to be at, the total O
tot
must equal one. Thus,
there is a missing component to the inventory, dubbed dark energy, with O
A
.2,3.
The cosmological constant (n
A
=1) is only one of several possible candidates,
but we generally assume this form unless otherwise specied. Measurements of
an accelerated expansion from distant supernovae (Riess et al. 1998, Perlmutter
et al. 1999) provide entirely independent evidence for dark energy in this amount.
The initial spectrum of density perturbations is assumed to be a power law
with a power law index or tilt of n 1 corresponding to a scale-invariant spec-
trum. Likewise the initial spectrum of gravitational waves is assumed to be scale-
invariant, with an amplitude parameterized by the energy scale of ination E
i
and
compared with the initial density spectrum (see Gravitational Waves). Finally
the formation of structure will eventually reionize the Universe at some redshift
6 z
ri
20.
Many of the features of the anisotropies will be produced even if these param-
eters fall outside the expected range or even if the standard paradigm is incorrect.
Where appropriate, we try to point these out.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
1 Aug 2002 12:36 AR AR166-AA40-06.tex AR166-AA40-06.SGM LaTeX2e(2002/01/18) P1: GJC
174 HU

DODELSON
Cosmic Microwave Background Temperature Field
The basic observable of the CMB is its intensity as a function of frequency and
direction on the sky n. Because the CMBspectrumis an extremely good blackbody
(Fixsen et al. 1996) with a nearly constant temperature across the sky T, we gener-
ally describe this observable in terms of a temperature uctuation O( n) =LT,T.
If these uctuations are Gaussian, then the multipole moments of the tempera-
ture eld
O
m
=
d nY
m
( n)O( n) (1)
are fully characterized by their power spectrum
m
O

mm
C
. (2)
whose values as a function of are independent in a given realization. For this rea-
son predictions and analyses are typically performed in harmonic space. On small
sections of the sky where its curvature can be neglected, the spherical harmonic
analysis becomes ordinary Fourier analysis in two dimensions. In this limit be-
comes the Fourier wavenumber. Because the angular wavelength =2,, large
multipole moments corresponds to small angular scales with 10
2
representing
degree scale separations. Likewise, because in this limit the variance of the eld
is

d
2
C
,(2)
2
, the power spectrum is usually displayed as
L
2
T

( +1)
2
C
T
2
. (3)
the power per logarithmic interval in wavenumber for 1.
Figure 1 (top) shows observations of L
T
along with the prediction of the
working cosmological model, complete with the acoustic peaks mentioned in the
Introduction and discussed extensively in Acoustic Peaks, below. Whereas
COBE rst detected anisotropy on the largest scales (inset), observations
in the past decade have pushed the frontier to smaller and smaller scales (left to
right in the gure). The Microwave Anisotropy Probe (MAP) satellite, launched
in June 2001, will go out to 1000, whereas the European satellite, Planck,
scheduled for launch in 2007, will go a factor or two higher (see Figure 1,
bottom).
The power spectra shown in Figure 1 all begin at =2 and exhibit large errors
at low multipoles. This is because the predicted power spectrum is the average
power in the multipole moment an observer would see in an ensemble of uni-
verses. However, a real observer is limited to one universe and one sky with its
one set of O
m
s, 2 +1 numbers for each . This is particularly problematic for
the monopole and dipole ( =0. 1). If the monopole were larger in our vicinity
than its average value, we would have no way of knowing it. Likewise for the
dipole, we have no way of distinguishing a cosmological dipole from our own pe-
culiar motion with respect to the CMB rest frame. Nonetheless, the monopole and
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
dipolewhich we will often call simply Oand :
are of the utmost signicance

in the early Universe. It is precisely the spatial and temporal variation of these
quantities, especially the monopole, which determine the pattern of anisotropies
we observe today. A distant observer sees spatial variations in the local tempera-
ture or monopole, at a distance given by the lookback time, as a ne-scale angular
anisotropy. Similarly, local dipoles appear as a Doppler-shifted temperature that is
viewed analogously. In the jargon of the eld this simple projection is referred to
as the freestreaming of power from the monopole and dipole to higher multipole
moments.
How accurately can the spectra ultimately be measured? As alluded to above,
the fundamental limitation is set by cosmic variance, the fact that there are
only 2 1 m-samples of the power in each multipole moment. This leads to an
inevitable error of
LC
=
_
2
2 1
C
. (4)
Allowing for further averaging over in bands of L , we see that the precision
in the power spectrum determination scales as
1
, i.e., 1% at =100 and
0.1% at =1000. It is the combination of precision predictions and prospects
for precision measurements that gives CMB anisotropies their unique stature.
There are two general caveats to these scalings. The rst is that any source
of noise, instrumental or astrophysical, increases the errors. If the noise is also
Gaussian and has a known power spectrum, one simply replaces the power spec-
trum on the right-hand side of Equation 4 with the sum of the signal and noise
power spectra (Knox 1995). This is why the errors for the Planck satellite increase
near its resolution scale in Figure 1 (bottom). Because astrophysical foregrounds
are typically non-Gaussian, it is usually also necessary to remove heavily con-
taminated regions, e.g., the galaxy. If the fraction of sky covered is f
sky
, then the
errors increase by a factor of f
1,2
sky
and the resulting variance is usually dubbed
sample variance (Scott et al. 1994). An f
sky
=0.65 was chosen for the Planck
satellite in Figure 1.
Cosmic Microwave Background Polarization Field
Whereas no polarization has yet been detected, general considerations of Thomson
scattering suggest that up to 10% of the anisotropies at a given scale are polar-
ized. Experimenters are currently hot on the trail, with upper limits approaching
the expected level (Hedman et al. 2001, Keating et al. 2001). Thus, we expect
polarization to be an extremely exciting eld of study in the coming decade.
The polarization eld can be analyzed in a way very similar to the temperature
eld, save for one complication. In addition to its strength, polarization also has an
orientation, depending on relative strength of two linear polarization states. The
classical literature has tended to describe polarization locally in terms of the Stokes
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
176 HU

DODELSON
parameters Q and U
1
, but recently cosmologists (Seljak 1997, Kamionkowski
et al. 1997, Zaldarriaga & Seljak 1997) have found that the scalar E and pseudo-
scalar B, linear but nonlocal combinations of Q and U, provide a more useful
description. Postponing the precise denition of E and B until Polarization,
below, we can, in complete analogy with Equation 1, decompose each of them in
terms of multipole moments and then, following Equation 2, consider the power
spectra,
_
E
m
E
/
m
/
_
=
/
mm
/ C
EE
.
_
B
m
B
/
m
/
_
=
/
mm
/ C
BB
.
_
O
m
E
/
m
/
_
=
/
mm
/ C
OE
. (5)
Parity invariance demands that the cross correlation between the pseudoscalar B
and the scalars O or E vanishes.
The polarization spectra shown in Figure 1 (bottom, plotted in jK following
Equation3) have several notable features. First, the amplitude of the EEspectrumof
Equation 5 is indeed down fromthe temperature spectrumbya factor of 10. Second,
the oscillatory structure of the EE spectrum is very similar to the temperature
oscillations, although they are apparently out of phase but correlated with each
other. Both of these features are a direct result of the simple physics of acoustic
oscillations, as shown in Acoustic Peaks below. The nal feature of the po-
larization spectra is the comparative smallness of the BB signal. Indeed, density
perturbations do not produce B modes to rst order. A detection of substantial
B polarization, therefore, would be momentous. Whereas E polarization effec-
tively doubles our cosmological information, supplementing that contained in C
,
B detection would push us qualitatively forward into new areas of physics.
ACOUSTIC PEAKS
When the temperature of the Universe was 3000 K at a redshift z
10
3
, elec-
trons and protons combined to form neutral hydrogen, an event usually known
as recombination (Peebles 1968, Zeldovich et al. 1969; see Seager et al. 2000
for recent renements). Before this epoch, free electrons acted as glue between
the photons and the baryons through Thomson and Coulomb scattering, so the
cosmological plasma was a tightly coupled photon-baryon uid (Peebles & Yu
1970). The spectrum depicted in Figure 1 can be explained almost completely by
analyzing the behavior of this prerecombination uid.
In Basics, below, we start from the two basic equations of uid mechanics
and derive the salient characteristics of the anisotropy spectrum: the existence of
1
There is also the possibility in general of circular polarization, described by the Stokes
parameter V, but this is absent in cosmological settings.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
peaks and troughs, the spacing between adjacent peaks, and the location of the
peaks. These properties depend in decreasing order of importance on the initial
conditions, the energy contents of the Universe before recombination, and those
after recombination. Ironically, the observational milestones have been reached
in almost the opposite order. Throughout the 1990s constraints on the location of
the rst peak steadily improved, culminating with precise determinations from the
Toco (Miller et al. 1999), Boomerang (de Bernardis et al. 2000), and Maxima-1
(Hanany et al. 2000) experiments (see Figure 1, top). In the working cosmological
model it shows up right where it should be if the present energy density of the
Universe is equal to the critical density, i.e., if the Universe is at. The skeptic
should note that the working cosmological model assumes a particular form for
the initial conditions and energy contents of the Universe before recombination,
which we shall see have only recently been tested directly (with an as yet much
lower level of statistical condence) with the higher peaks.
In Initial Conditions, below, we introduce the initial conditions that appar-
ently are the source of all clumpiness in the Universe. In the context of ab initio
models the term initial conditions refers to the physical mechanism that gener-
ates the primordial small perturbations. In the working cosmological model this
mechanism is ination and it sets the initial phase of the oscillations to be the
same across all Fourier modes. Remarkably, from this fact alone comes the pre-
diction that there will be peaks and troughs in the amplitude of the oscillations
as a function of wavenumber. Additionally, the inationary prediction of an ap-
proximately scale-invariant amplitude of the initial perturbations implies roughly
scale-invariant oscillations in the power spectrum, and ination generically pre-
dicts a at Universe. These are all falsiable predictions of the simplest inationary
models, and they have withstood the test against observations to date.
The energy contents of the Universe before recombination all leave their distinct
signatures on the oscillations, as discussed in Gravitational Forcing and the two
following sections. In particular, the cold dark matter and baryon signatures have
now been seen in the data (Halverson et al. 2001, Nettereld et al. 2001, Lee et al.
2001). The coupling between electrons and photons is not perfect, especially as
one approaches the epoch of recombination. As discussed in Damping, below,
this imperfect coupling leads to damping in the anisotropy spectrum: Very small
scale inhomogeneities are smoothed out. The damping phenomenon has now been
observed by the Cosmic Background Imager (CBI) experiment (Padin et al. 2001).
Importantly, uid imperfections also generate linear polarization as covered in
Polarization, below. Because the imperfection is minimal and appears only at
small scales, the polarization generated is small and has not been detected to date.
After recombination the photons essentially travel freely to us today, so the
problem of translating the acoustic inhomogeneities in the photon distribution at
recombination to the anisotropy spectrum today is simply one of projection. This
projection depends almost completely on one number, the angular diameter dis-
tance between us and the surface of last scattering. This number depends on the
energy contents of the Universe after recombination through the expansion rate.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
178 HU

DODELSON
The hand-waving projection argument of Basics is formalized in Integral Ap-
proach in the process introducing the popular code used to compute anisotropies,
CMBFAST. Finally, we discuss the sensitivity of the acoustic peaks to cosmolog-
ical parameters in Parameter Sensitivity.
Basics
For pedagogical purposes, let us begin with an idealization of a perfect photon-
baryon uid and neglect the dynamical effects of gravity and the baryons. Pertur-
bations in this perfect uid can be described by a simple continuity and a Euler
equation that encapsulate the basic properties of acoustic oscillations.
The discussion of acoustic oscillations takes place exclusively in Fourier space.
For example, we decompose the monopole of the temperature eld into
O
=0.m=0
(x) =
_
d
3
k
(2)
3
e
i kx
O(k) (6)
and omit the subscript
00
on the Fourier amplitude. Because perturbations are very
small, the evolution equations are linear, and different Fourier modes evolve in-
dependently. Therefore, instead of partial differential equations for a eld O(x),
we have ordinary differential equations for O(k). In fact, owing to rotational sym-
metry, all O(k) for a given k obey the same equations. Here and in the following
sections we omit the wavenumber argument k when no confusion with physical
space quantities will arise.
Temperature perturbations in Fourier space obey
O =
1
3
k:
. (7)
This equation for the photon temperature O, which does indeed look like the
familiar continuity equation in Fourier space (derivatives become wavenumbers
ik), has a number of subtleties hidden in it, owing to the cosmological setting.
First, the over-dot derivative here is with respect to conformal time
_
dt,a(t).
Because we are working in units in which the speed of light c =1, is also the
maximum comoving distance a particle could have traveled since t =0. It is often
called the comoving horizon or more specically the comoving particle horizon.
The physical horizon is a times the comoving horizon.
Second, the photon uid velocity, :
, has been written as a scalar instead of a

vector. In the early universe only the velocity component parallel to the wavevector
k is expected to be important because it alone has a source in gravity. Speci-
cally, v
=i:

k. In terms of the moments introduced in Observables, above,
:
represents a dipole moment directed along k. The factor of 1,3 comes about
because continuity conserves photon number, not temperature, and the number
density n
T
3
. Finally, we reiterate that, for the time being, we are neglecting the
effects of gravity.
The Euler equation for a uid is an expression of momentum conservation.
The momentum density of the photons is (
):
, where the photon pressure

A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
p
,3. In the absence of gravity and viscous uid imperfections, pressure

gradients p
,3 supply the only force. Because
T
4
, this becomes
4kO
,3 in Fourier space. The Euler equation then becomes

:
= kO. (8)
Differentiating the continuity equation and inserting the Euler equation yields
the most basic form of the oscillator equation,
Oc
2
s
k
2
O = 0. (9)
where c
s
p, =1,
3 is the sound speed in the (dynamically baryon-free)

uid. What this equation says is that pressure gradients act as a restoring force to
any initial perturbation in the system, which thereafter oscillates at the speed of
sound. Physically these temperature oscillations represent the heating and cooling
of a uid that is compressed and rareed by a standing sound or acoustic wave.
This behavior continues until recombination. Assuming negligible initial velocity
perturbations, we have a temperature distribution at recombination of
O(
) = O(0) cos(ks
). (10)
where s =
_
c
s
d ,
3 is the distance sound can travel by , usually called the

sound horizon. Asterisks denote evaluation at recombination z
.
In the limit of scales that are large compared with the sound horizon ks _1, the
perturbation is frozen into its initial conditions. This is the gist of the statement
that the large-scale anisotropies measured by COBE directly measure the initial
conditions. On small scales the amplitude of the Fourier modes exhibit temporal
oscillations, as shown in Figure 2 [with + =0, +
i
=3O(0) for this idealization].
Modes that are caught at maxima or minima of their oscillation at recombination
correspond to peaks in the power, i.e., the variance of O(k,
). Because sound
takes half as long to travel half as far, modes corresponding to peaks follow a
harmonic relationship k
n
=n,s
, where n is an integer (see Figure 2a).

How does this spectrum of inhomogeneities at recombination appear to us
today? Roughly speaking, a spatial inhomogeneity in the CMB temperature of
wavelength appears as an angular anisotropy of scale ,D, where D(z)
is the comoving angular diameter distance from the observer to redshift z. We
address this issue more formally in Integral Approach, below. In a at universe
D
=
0
0
, where
0
(z =0). Inharmonic space the relationshipimplies
a coherent series of acoustic peaks in the anisotropy spectrum, located at
n
n
a
.
a
D
,s
. (11)
To get a feel for where these features should appear, note that in a at
matter-dominateduniverse (1 z)
1,2
sothat
,
0
1,30 2
. Equivalently,
1
200. Notice that because we are measuring ratios of distances, the absolute
distance scale drops out; we see in Radiation Driving, below, that the Hubble
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
180 HU

DODELSON
Figure 2 Idealized acoustic oscillations with time measured as the extent of the sound
horizon relative to its value at recombination (s/s
). (a) Peak scales: The wavemode that

completes half an oscillation by recombination sets the physical scale of the rst peak. Both
minima and maxima correspond to peaks in power (dashed lines, absolute value), so higher
peaks are integral multiples of this scale with equal height. Plotted here is the idealization
of Equation 15 (constant potentials, no baryon loading). (b) Baryon loading: Baryon loading
boosts the amplitudes of every other oscillation. Plotted here is the idealization of Equation
16 (constant potentials and baryon loading R=1,6) for the third peak.
constant sneaks back into the problem because the Universe is not fully matter-
dominated at recombination.
In a spatially curved universe the angular diameter distance no longer equals the
coordinate distance, making the peak locations sensitive to the spatial curvature of
the Universe (Doroshkevich et al. 1978, Kamionkowski et al. 1994). Consider rst
a closed universe with radius of curvature R = H
1
0
[O
tot
1[
1,2
. Suppressing one
spatial coordinate yields a two-sphere geometry with the observer situated at the
pole (see Figure 3). Light travels on lines of longitude. A physical scale at xed
latitude given by the polar angle subtends an angle =,Rsin . For _1, a
Euclidean analysis would infer a distance D=Rsin , even though the coordinate
distance along the arc is d =R; thus
D = R sin(d,R). (12)
For open universes, simply replace sin with sinh. The result is that objects in an
open (closed) universe are closer (further) than they appear, as if seen through a
lens. In fact one way of viewing this effect is as the gravitational lensing caused by
the background density (see Gravitational Lensing, below). A given comoving
scale at a xed distance subtends a larger (smaller) angle in a closed (open) universe
than a at universe. This strong scaling with spatial curvature indicates that the
observed rst peak at
1
200 constrains the geometry to be nearly spatially at.
We implicitly assume spatial atness in the following sections unless otherwise
stated.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Figure 3 Angular diameter distance. In a closed universe objects are further than
they appear to be from Euclidean (at) expectations, corresponding to the difference
between coordinate distance d and angular diameter distance D. Consequently, at a
xed coordinate distance, a given angle corresponds to a smaller spatial scale in a
closed universe. Acoustic peaks therefore appear at larger angles or lower in a closed
universe. The converse is true for an open universe.
Finally, in a at dark energy-dominated universe the conformal age of the
universe decreases approximately as
0
0
(1 ln O
0.085
m
). For reasonable O
m
,
this causes only a small shift of
1
to lower multipoles (see Figure 4, color insert)
relative to the effect of curvature. Combined with the effect of the radiation near
recombination, the peak locations provide a means to measure the physical age t
0
of a at universe (Hu et al. 2001).
Initial Conditions
As suggested above, observations of the location of the rst peak strongly point
to a at universe. This is encouraging news for adherents of ination, a theory
that initially predicted O
tot
=1 at a time when few astronomers would sign on to
such a high value (see Liddle & Lyth 1993 for a review). However, the argument
for ination goes beyond the conrmation of atness. In particular, the discussion
of the previous subsection begs the question: Whence O(0), the initial conditions
of the temperature uctuations? The answer requires the inclusion of gravity and
considerations of causality that point to ination as the origin of structure in the
Universe.
The calculations of the typical angular scale of the acoustic oscillations in the
previous section are familiar in another context: the horizon problem. Because the
sound speed is near the speed of light, the degree scale also marks the extent of a
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
182 HU

DODELSON
causally connected region or particle horizon at recombination. For the picture in
the previous section to hold, the perturbations must have been laid down while the
scales in question were still far outside the particle horizon
2
. The recent observa-
tional verication of this basic peak structure presents a problem potentially more
serious than the original horizon problem of approximate isotropy: The mecha-
nism that smooths uctuations in the Universe must also regenerate them with
superhorizon-sized correlations at the 10
5
level. Ination is an idea that solves
both problems simultaneously.
The inationary paradigm postulates that an early phase of near exponential
expansion of the Universe was driven by a formof energy with negative pressure. In
most models this energy is usually provided by the potential energy of a scalar eld.
The inationary era brings the observable universe to a nearly smooth and spatially
at state. Nonetheless, quantum uctuations in the scalar eld are unavoidable
and are carried to large physical scales by the expansion. Because an exponential
expansion is self-similar in time, the uctuations are scale-invariant, i.e., in each
logarithmic interval in scale the contribution to the variance of the uctuations is
equal. Because the scalar eld carries the energy density of the Universe during
ination, its uctuations induce variations in the spatial curvature (Guth & Pi
1985, Hawking 1982, Bardeen et al. 1983). Instead of perfect atness, ination
predicts that each scale will resemble a very slightly open or closed universe.
This uctuation in the geometry of the Universe is essentially frozen while the
perturbation is outside the horizon (Bardeen 1980).
Formally, curvature uctuations are perturbations to the space-space piece of
the metric. In a Newtonian coordinate system, or gauge, where the metric is di-
agonal the spatial curvature uctuation is called g
i j
=2a
2
+
i j
(see, e.g., Ma &
Bertschinger 1995). The more familiar Newtonian potential is the time-time uc-
tuation g
t t
=2+ and is approximately + +. Approximate scale invariance
then says that L
2
+
k
3
P
+
(k),2
2
k
n1
, where P
+
(k) is the power spectrum of +
and the tilt n 1.
Now let us relate the inationary prediction of scale-invariant curvature uc-
tuations to the initial temperature uctuations. Newtonian intuition based on the
Poisson equation k
2
+=4Ga
2
tells us that on large scales (small k) density and
hence temperature uctuations should be negligible compared with Newtonian
potential. General relativity says otherwise because the Newtonian potential is
also a time-time uctuation in the metric. It corresponds to a temporal shift of
t,t =+. The CMB temperature varies as the inverse of the scale factor, which
in turn depends on time as a t
2,[3(1p,)]
. Therefore, the fractional change in the
CMB temperature
O =
a
a
=
2
3
_
1
p
_
1
t
t
. (13)
2
Recall that the comoving scale k does not vary with time. At very early times, then, the
wavelength k
1
is much larger than the horizon .
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Thus, a temporal shift produces a temperature perturbation of +,2 in the
radiation-dominated era (when p =,3) and 2+,3 in the matter-dominated
epoch (p =0) (Sachs & Wolfe 1967, Peacock 1991, White & Hu 1997). The initial
temperature perturbation is therefore inextricably linked with the initial gravita-
tional potential perturbation. Ination predicts scale-invariant initial uctuations
in both the CMB temperature and the spatial curvature in the Newtonian gauge.
Alternate models that seek to obey causality can generate curvature uctuations
only inside the particle horizon. Because the perturbations are then not generated
at the same epoch independent of scale, there is no longer a unique relationship
between the phase of the oscillators. That is, the argument of the cosine in Equa-
tion 10 becomes ks
(k), where is a phase that can in principle be different

for different wavevectors, even those with the same magnitude k. This leads to
temporal incoherence in the oscillations and hence a washing out of the acoustic
peaks (Albrecht et al. 1996), most notably in cosmological defect models (Allen
et al. 1997, Seljak et al. 1997). Complete incoherence is not a strict requirement of
causality because there are other ways to synch up the oscillations. For example,
many isocurvature models, in which the initial spatial curvature is unperturbed, are
coherent because their oscillations begin with the generation of curvature uctua-
tions at horizon crossing (Hu &White 1996). Still they typically have ,=0 (com-
pare Turok 1996). Independent of the angular diameter distance D
, the ratio of the

peak locations gives the phase:
1
:
2
:
3
1 : 2 : 3 for =0. Likewise indepen-
dent of a constant phase, the spacing of the peaks
n
n1
=
A
gives a measure of
the angular diameter distance (Hu & White 1996). The observations, which indi-
cate coherent oscillations with =0, therefore, have provided a nontrivial test of
the inationary paradigm and supplied a substantially more stringent version of
the horizon problem for contenders to solve.
Gravitational Forcing
We saw above that uctuations in a scalar eld during ination get turned into
temperature uctuations via the intermediary of gravity. Gravity affects Oin more
ways than this. The Newtonian potential and spatial curvature alter the acoustic
oscillations by providing a gravitational force on the oscillator. The Euler equa-
tion (8) gains a term on the right-hand side owing to the gradient of the potential
k+. The main effect of gravity then is to make the oscillations a competition be-
tween pressure gradients kOand potential gradients k+ with an equilibrium when
O+ =0.
Gravity also changes the continuity equation. Because the Newtonian curvature
is essentially a perturbation to the scale factor, changes in its value also generate
temperature perturbations by analogy to the cosmological redshift O=+, and
so the continuity equation (7) gains a contribution of
+ on the right-hand side.

These two effects bring the oscillator equation (9) to
Oc
2
s
k
2
O =
k
2
3
+

+. (14)
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
184 HU

DODELSON
In a at universe and in the absence of pressure + and + are constant. Also, in
the absence of baryons, c
2
s
=1,3, so the new oscillator equation is identical to
Equation 9 with O replaced by O+. The matter-dominated epoch is then
[O+]() = [O+](
md
) cos(ks)
=
1
3
+(
md
) cos(ks). (15)
where
md
represents the start of the matter-dominated epoch (see Figure 2a). We
have used the matter-dominated initial conditions for O given in the previous
section, assuming large scales, ks
md
_1.
The results from the idealization of Basics, above, carry through with a few
exceptions. Even without an initial temperature uctuation to displace the oscilla-
tor, acoustic oscillations would arise by the infall and compression of the uid into
gravitational potential wells. Because it is the effective temperature O+ that
oscillates, they occur even if O(0) =0. The quantity O+ can be thought of as
an effective temperature in another way: After recombination, photons must climb
out of the potential well to the observer and thus suffer a gravitational redshift of
LT,T =+. The effective temperature uctuation is therefore also the observed
temperature uctuation. We now see that the large scale limit of Equation 15 re-
covers the famous Sachs-Wolfe result that the observed temperature perturbation
is +,3 and overdense regions correspond to cold spots on the sky (Sachs & Wolfe
1967). When + -0, although O is positive, the effective temperature O+ is
negative. The plasma begins effectively rareed in gravitational potential wells. As
gravity compresses the uid and pressure resists, rarefaction becomes compres-
sion and rarefaction again. The rst peak corresponds to the mode that is caught
in its rst compression by recombination. The second peak at roughly half the
wavelength corresponds to the mode that went through a full cycle of compression
and rarefaction by recombination. We will use this language of the compression
and rarefaction phase inside initially overdense regions, but one should bear
in mind that there are an equal number of initially underdense regions with the
opposite phase.
Baryon Loading
So far we have been neglecting the baryons in the dynamics of the acoustic os-
cillations. To see whether this is a reasonable approximation consider the photon-
baryon momentumdensity ratio R=( p
b
b
),( p
) 30O
b
h
2
(z,10
3
)
1
. For
typical values of the baryon density this number is of order unity at recombination,
and so we expect baryonic effects to begin appearing in the oscillations just as they
are frozen in.
Baryons are conceptually easy to include in the evolution equations because
their momentum density provides extra inertia in the joint Euler equation for
pressure and potential gradients to overcome. Because inertial and gravitational
mass are equal, all terms in the Euler equation except the pressure gradient are
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
multiplied by 1 R, leading to the revised oscillator equation (Hu & Sugiyama,
1995)
c
2
s
d
d
_
c
2
s

O
_
c
2
s
k
2
O =
k
2
3
+ c
2
s
d
d
_
c
2
s

+
_
. (16)
where we have used the fact that the sound speed is reduced by the baryons to
c
s
=1,
3(1 R).
To get a feel for the implications of the baryons take the limit of constant R, +,
and +. Then d
2
(R+),d
2
(= 0) may be added to the left-hand side to again put
the oscillator equation in the form of Equation 9 with OO(1 R)+. The
solution then becomes
[O(1 R)+]() = [O(1 R)+](
md
) cos(ks). (17)
Aside from the lowering of the sound speed, which decreases the sound horizon,
baryons have two distinguishing effects: They enhance the amplitude of the oscil-
lations and shift the equilibrium point to O=(1 R)+ (see Figure 2b). These
two effects are intimately related and are easy to understand because the equations
are exactly those of a mass m=1 R on a spring in a constant gravitational eld.
For the same initial conditions, increasing the mass causes the oscillator to fall
further in the gravitational eld, leading to larger oscillations and a shifted zero
point.
The shifting of the zero point of the oscillator has signicant phenomenolog-
ical consequences. Because it is still the effective temperature O+ that is the
observed temperature, the zero-point shift breaks the symmetry of the oscillations.
The baryons enhance only the compressional phase, i.e., every other peak. For
the working cosmological model these are the rst, third, fth . . .. Physically, the
extra gravity provided by the baryons enhances compression into potential wells.
These qualitative results remain true in the presence of a time-variable R. An
additional effect arises owing to the adiabatic damping of an oscillator with a
time-variable mass. Because the energy/frequency of an oscillator is an adiabatic
invariant, the amplitude must decay as (1 R)
1,4
. This can also be understood
by expanding the time derivatives in Equation 16 and identifying the

R

O term as
the remnant of the familiar expansion drag on baryon velocities.
Radiation Driving
We have hitherto also been neglecting the energy density of the radiation in
comparison to the matter. The matter-to-radiation energy-density ratio scales as
m
,
r
24O
m
h
2
(z,10
3
)
1
andsois alsoof order unityat recombinationfor reason-
able parameters. Moreover, uctuations corresponding to the higher peaks entered
the sound horizon at an earlier time, during radiation domination.
Including the radiation changes the expansion rate of the Universe and hence
the physical scale of the sound horizon at recombination. It introduces yet another
potential ambiguity in the interpretation of the location of the peaks. Fortunately,
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
186 HU

DODELSON
Figure 5 Radiation driving and diffusion damping. The decay of the potential + drives
the oscillator in the radiation-dominated epoch. Diffusion generates viscosity
, i.e., a
quadrupole moment in the temperature, which damps oscillations and generates polarization.
Plotted here is the numerical solution to Equation 18 and Equation 19 for a mode with
wavelength much smaller than the sound horizon at decoupling, ks
1.
the matter-radiation ratio has another effect in the power spectrum by which it
can be distinguished. Radiation drives the acoustic oscillations by making the
gravitational force evolve with time (Hu & Sugiyama 1995). Matter does not.
The exact evolution of the potentials is determined by the relativistic Poisson
equation. However, qualitatively, we know that the background density is de-
creasing with time, so unless the density uctuations in the dominant component
grow unimpeded by pressure, potentials will decay. In particular, in the radiation-
dominated era once pressure begins to ght gravity at the rst compressional
maxima of the wave, the Newtonian gravitational potential and spatial curvature
must decay (see Figure 5).
This decay actually drives the oscillations: It is timed to leave the uid maxi-
mally compressed with no gravitational potential to ght as it turns around. The net
effect is doubled because the redshifting from the spatial metric uctuation +also
goes away at the same time. When the Universe becomes matter dominated, the
gravitational potential is no longer determined by photon-baryon density pertur-
bations but by the pressureless cold dark matter. Therefore, the amplitudes of the
acoustic peaks increase as the cold dark matter-to-radiation ratio decreases (Seljak
1994, Hu & Sugiyama 1995). Density perturbations in any form of radiation will
stop growing around horizon crossing and lead to this effect. The net result is
that across the horizon scale at matter-radiation equality (k
eq
(4 2
2),
eq
)
the acoustic amplitude increases by a factor of 45 (Hu & Sugiyama 1996). By
eliminating gravitational potentials, photon-baryon acoustic oscillations eliminate
the alternating peak heights from baryon loading. The observed high third peak
(Halverson et al. 2001) is a good indication that cold dark matter both exists and
dominates the energy density at recombination.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Damping
The photon-baryon uid has slight imperfections corresponding to shear viscosity
and heat conduction in the uid (Silk 1968, Weinberg 1971). These imperfections
damp acoustic oscillations. To consider these effects, we nowpresent the equations
of motion of the system in their full form, including separate continuity and Euler
equations for the baryons. Formally the continuity and Euler equations followfrom
the covariant conservation of the joint stress-energy tensor of the photon-baryon
uid. Because photon and baryon numbers are separately conserved, the continuity
equations are unchanged,
O =
k
3
:

+.

b
= k:
b
3

+. (18)
where
b
and :
b
are the density perturbation and uid velocity of the baryons. The
Euler equations contain qualitatively new terms:
:
= k(O+)
k
6
(:
:
b
).
:
b
=
a
a
:
b
k+ (:
:
b
),R. (19)
For the baryons the rst term on the right accounts for cosmological expansion,
which makes momenta decay as a
1
. The third term on the right accounts for mo-
mentum exchange in the Thomson scattering between photons and electrons (pro-
tons are very tightly coupled to electrons via Coulomb scattering), with n
e
T
a,
the differential Thomson optical depth, and is compensated by its opposite in the
photonEuler equation. These terms are the originof heat conductionimperfections.
If the medium is optically thick across a wavelength, ,k 1 and the photons and
baryons cannot slip past each other. As it becomes optically thin, slippage dissi-
pates the uctuations.
In the photon Euler equation there is an extra force on the right-hand side
owing to anisotropic stress gradients or radiation viscosity in the uid,
. The
anisotropic stress is directly proportional to the quadrupole moment of the photon
temperature distribution. A quadrupole moment is established by gradients in :
as photons from, for example, neighboring temperature crests meet at a trough

(see Figure 6, damping and polarization). However, it is destroyed by scatter-
ing. Thus,
=2(k:
, )A
v
, where the order unity constant can be derived from
the Boltzmann equation as A
v
=16,15 (Kaiser 1983). Its evolution is shown in
Figure 5. With the continuity Equation 7, k:
3

O, so viscosity takes the form
of a damping term. The heat conduction term can be shown to have a similar effect
by expanding the Euler equations in k, . The nal oscillator equation including
both terms becomes
c
2
s
d
d
c
2
s

O
k
2
c
2
s

[A
v
A
h
]

Oc
2
s
k
2
O =
k
2
3
+ c
2
s
d
d
c
2
s

+
. (20)
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
188 HU

DODELSON
where the heat conduction coefcient A
h
=R
2
,(1 R). Thus, we expect the inho-
mogeneities to be damped by an exponential factor of order e
k
2
,
(see Figure 5).
The damping scale k
d
is thus of order
,, corresponding to the geometric mean

of the horizon and the mean free path. Damping can be thought of as the result of
the random walk in the baryons that takes photons from hot regions into cold and
vice versa (Silk 1968). Detailed numerical integration of the equations of motion
are required to track the rapid growth of the mean free path and damping length
through recombination itself. These calculations show that the damping scale is
of order k
d
s
10, leading to a substantial suppression of the oscillations beyond

the third peak.
How does this suppression depend on the cosmological parameters? As the
matter density O
m
h
2
increases, the horizon
decreases because the expansion rate

goes up. Because the diffusion length is proportional to
, it too decreases as the

matter densitygoes upbut not as muchas the angular diameter distance D
, whichis
also inversely proportional to the expansion rate. Thus, more matter translates into
more damping at a xed multipole moment; conversely, it corresponds to slightly
less damping at a xed peak number. The dependence on baryons is controlled by
the mean free path, which is in turn controlled by the free electron density: The
increase in electron density caused by an increase in the baryons is partially offset
by a decrease in the ionization fraction caused by recombination. The net result
under the Saha approximation is that the damping length scales approximately
as (O
b
h
2
)
1,4
. Accurate tting formulae for this scale in terms of cosmological
parameters can be found in (Hu & White 1997c).
Polarization
The dissipation of the acoustic oscillations leaves a signature in the polarization
of CMB in its wake (see e.g., Hu & White 1997a and references therein for a
more complete treatment). Much like reection off of a surface, Thomson scat-
tering induces a linear polarization in the scattered radiation. Consider incoming
radiation in the x direction scattered at right angles into the z direction (see
Figure 7, top left). Heuristically, incoming radiation shakes an electron in the
direction of its electric eld vector or polarization c
/
. causing it to radiate with
an outgoing polarization parallel to that direction. However, because the outgo-
ing polarization c must be orthogonal to the outgoing direction, incoming radi-
ation that is polarized parallel to the outgoing direction cannot scatter, leaving
only one polarization state, more generally, the Thomson differential cross section
d
T
,dO[ c
/
c[
2
.
Unlike the reection of sunlight off a surface, the incoming radiation comes
fromall angles. If it were completely isotropic in intensity, radiation coming along
the y would provide the polarization state that is missing from that coming along
x. leaving the net outgoing radiation unpolarized. Only a quadrupole tempera-
ture anisotropy in the radiation generates a net linear polarization from Thomson
scattering. As we have seen, a quadrupole can only be generated causally by the
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
motion of photons and only if the Universe is optically thin to Thomson scattering
across this scale (i.e., it is inversely proportional to ). Polarization generation
suffers froma Catch-22: The scattering that generates polarization also suppresses
its quadrupole source.
The fact that the polarization strength is of order the quadrupole in amplitude ex-
plains the shape and height of the polarization spectra in Figure 1b. The monopole
and dipole O and :
are of the same order of magnitude at recombination, but

their oscillations are ,2 out of phase as follows fromEquation 9 and Equation 10.
Because the quadrupole is of order k:
, (see Figure 5), the polarization spectrum

should be smaller than the temperature spectrumby a factor of order k, at recom-
bination. As in the case of the damping, the precise value requires numerical work
(Bond & Efstathiou 1987) because changes so rapidly near recombination. Cal-
culations showa steady rise in the polarized fraction with increasingl or k to a maxi-
mum of about 10% before damping destroys the oscillations and hence the dipole
source. Because :
is out of phase with the monopole, the polarization peaks should

also be out of phase with the temperature peaks. Indeed, Figure 1b shows that this
is the case. Furthermore, the phase relation also tells us that the polarization is
correlated with the temperature perturbations. Because the correlation power C
OE
is the product of the two, it exhibits oscillations at twice the acoustic frequency.
Until now, we have focused on the polarization strength without regard to its
orientation. The orientation, like a two-dimensional vector, is described by two
components, E and B. The E and B decomposition is simplest to visualize in
the small scale limit, where spherical harmonic analysis coincides with Fourier
analysis (Seljak 1997). The wavevector k picks out a preferred direction against
which the polarization direction is measured (see Figure 7, top right). Because
the linear polarization is a headless vector that remains unchanged upon a 180
rotation, the two numbers E and B that dene it represent polarization aligned or
orthogonal with the wavevector (positive and negative E) and crossed at 45
(positive and negative B).

In linear theory scalar perturbations such as the gravitational potential and
temperature perturbations have only one intrinsic direction associated with them,
that provided by k, and the orientation of the polarization inevitably takes it cue
fromthat one direction, thereby producing an E-mode. The generalization to an all-
sky characterization of the polarization changes none of these qualitative features.
The E-mode and the B-mode are formally distinguished by the orientation of the
Hessian of the Stokes parameters that dene the direction of the polarization itself.
This geometric distinction is preserved under summation of all Fourier modes as
well as the generalization of Fourier analysis to spherical harmonic analysis.
The acoustic peaks in the polarization appear exclusively in the EE power spec-
trum of Equation 5. This distinction is very useful, as it allows a clean separation
of this effect from those occuring beyond the scope of the linear perturbation
theory of scalar uctuations: in particular, gravitational waves). Moreover, in the
working cosmological model, the polarization peaks and correlation are precise
predictions of the temperature peaks, as they depend on the same physics. As
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
190 HU

DODELSON
such, their detection would represent a sharp test of the implicit assumptions of
the working model, especially its initial conditions and ionization history.
Integral Approach
The discussion in the previous sections sufces for a qualitative understanding
of the acoustic peaks in the power spectra of the temperature and polarization
anisotropies. To rene this treatment we must consider more carefully the sources
of anisotropies and their projection into multipole moments.
Because the description of the acoustic oscillations takes place in Fourier space,
the projection of inhomogeneities at recombination onto anisotropies today has an
added level of complexity. An observer today sees the acoustic oscillations in
effective temperature as they appeared on a spherical shell at x =D
n at recombi-
nation, where n is the direction vector and D
=
0
is the distance light can

travel between recombination and the present (see Figure 6). Having solved for
the Fourier amplitude [O+](k,
), we can expand the exponential in Equation

6 in terms of spherical harmonics, so the observed anisotropy today is
O( n.
0
) = 4
m
Y
m
( n)
_
(i )
_
d
3
k
(2)
3
a
(k)Y
m
(
k)
_
. (21)
where the projected source a
(k) =[O+](k.
) j
(k D
). Because the spherical

harmonics are orthogonal, Equation 1 implies that O
m
today is given by the in-
tegral in square brackets today. A given plane wave actually produces a range of
anisotropies in angular scale, as is obvious from Figure 6. The one-to-one map-
ping between wavenumber and multipole moment described in Basics, above, is
only approximately true and comes fromthe fact that the spherical Bessel function
j
(k D
) is strongly peaked at k D
. Notice that this peak corresponds to con-

tributions in the direction orthogonal to the wavevector where the correspondence
between and k is one-to-one (see Figure 6).
Projection is less straightforward for other sources of anisotropy. We have hith-
erto neglected the fact that the acoustic motion of the photon-baryon uid also
produces a Doppler shift in the radiation that appears to the observer as a temper-
ature anisotropy as well. In fact, we argued above that :
b
:
is of comparable
magnitude but out of phase with the effective temperature. If the Doppler effect
projected in the same way as the effective temperature, it would wash out the
acoustic peaks. However, the Doppler effect has a directional dependence as well
because it is only the line-of-sight velocity that produces the effect. Formally, it is
a dipole source of temperature anisotropies and hence has an =1 structure. The
coupling of the dipole and plane-wave angular momenta implies that the projec-
tion of the Doppler effect involves a combination of j
1
that may be rewritten as
j
/
(x) dj
(x),dx. The structure of j

/
lacks a strong peak at x =. Physically this

corresponds to the fact that the velocity is irrotational and hence has no component
in the direction orthogonal to the wavevector (see Figure 6). Correspondingly, the
Doppler effect cannot produce strong peak structures (Hu & Sugiyama 1995). The
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
observed peaks must be acoustic peaks in the effective temperature, not Doppler
peaks.
There is one more subtlety involved when passing from acoustic oscillations
to anisotropies. Recall from Radiation Driving, above, that radiation leads to
decay of the gravitational potentials. Residual radiation after decoupling therefore
implies that the effective temperature is not precisely [O+](
). The photons
actually have slightly shallower potentials to climb out of and lose the perturbative
analogue of the cosmological redshift, so the [O+](
) overestimates the dif-

ference between the true photon temperature and the observed temperature. This
effect of course is already in the continuity equation for the monopole Equation 18,
so the source in Equation 21 gets generalized to
a
(k) = [O+] (
) j
l
(kD
) :
b
(k.
) j
/
(k D
)
_

0
d(

+

+) j
l
(kD). (22)
The last termvanishes for constant gravitational potentials but is nonzero if residual
radiation driving exists, as it will in lowO
m
h
2
models. Note that residual radiation
driving is particularly important because it adds in phase with the monopole: The
potentials vary in time only near recombination, so the Bessel function can be set
to j
l
(kD
) and removed from the integral. This complication has the effect of
decreasing the multipole value of the rst peak
1
as the matter-radiation ratio at
recombination decreases (Hu & Sugiyama 1995). Finally, we mention that time
varying potentials can also play a role at very late times owing to nonlinearities or,
a cosmological constant. Those contributions, to be discussed more in Integrated
Sachs-Wolfe Effect, are sometimes referred to as late integrated Sachs-Wolfe
effects and do not add coherently with [O+](
).
Putting these expressions together and squaring, we obtain the power spectrum
of the acoustic oscillations
C
=
2
_
dk
k
k
3
a
2
(k). (23)
This formulation of the anisotropies in terms of projections of sources with
specic local angular structure can be completed to include all types of sources
of temperature and polarization anisotropies at any given epoch in time, linear
or nonlinear: the monopole, dipole, and quadrupole sources arising from density
perturbations, vorticity, and gravitational waves (Hu & White 1997b). In a curved
geometry one replaces the spherical Bessel functions with ultraspherical Bessel
functions (Abbott & Schaefer 1986, Hu et al. 1998). Precision in the predictions
of the observables is then limited only by the precision in the prediction of the
sources. This formulation is ideal for cases in which the sources are governed by
nonlinear physics, even though the CMB responds linearly, as we see in Beyond
the Peaks, below.
Perhaps more importantly, the widely used CMBFAST code (Seljak &
Zaldarriaga 1996) exploits these properties to calculate the anisotropies in lin-
ear perturbation efciently. It numerically solves for the smoothly varying sources
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
192 HU

DODELSON
on a sparse grid in wavenumber, interpolating in the integrals for a handful of s
in the smoothly varying C
. It has largely replaced the original ground breaking

codes (Wilson & Silk 1981, Bond & Efstathiou 1984, Vittorio & Silk 1984) based
on tracking the rapid temporal oscillations of the multipole moments that simply
reect structure in the spherical Bessel functions themselves.
Parameter Sensitivity
The phenomenology of the acoustic peaks in the temperature and polarization is
essentially described by four observables and the initial conditions (Hu et al. 1997).
These are the angular extents of the sound horizon
a
D
,s
, the particle hori-

zon at matter radiation equality
eq
k
eq
D
, and the damping scale

d
k
d
D
,
as well as the value of the baryon-photon momentum density ratio R
.
a
sets
the spacing between of the peaks;
eq
and
d
compete to determine their ampli-
tude through radiation driving and diffusion damping. R
sets the baryon loading

and, along with the potential well depths set by
eq
, xes the modulation of the
even and odd peak heights. The initial conditions set the phase, or equivalently
the location of the rst peak in units of
a
, and an overall tilt n in the power
spectrum.
In the model of Figure 1, these numbers are
a
=301 (
1
=0.73
a
).
eq
=149.
d
=1332, R
=0.57, and n =1, and in this family of models the parameter sensi-
tivity is approximately (Hu et al. 2001)
L
a
a
0.24
LO
m
h
2
O
m
h
2
0.07
LO
b
h
2
O
b
h
2
0.17
LO
A
O
A
1.1
LO
tot
O
tot
.
L
eq
eq
0.5
LO
m
h
2
O
m
h
2
0.17
LO
A
O
A
1.1
LO
tot
O
tot
.
L
d
d
0.21
LO
m
h
2
O
m
h
2
0.20
LO
b
h
2
O
b
h
2
0.17
LO
A
O
A
1.1
LO
tot
O
tot
. (24)
and LR
,R
1.0LO
b
h
2
,O
b
h
2
. Current observations indicate that
a
=304 4.
eq
=168 15.
d
=1392 18. R
=0.60 0.06, and n =0.96 0.04 (Knox

et al. 2001; see also Wang et al. 2001, Pryke et al. 2001, de Bernardis et al. 2002) if
gravitational wave contributions are subdominant and the reionization redshift is
low, as assumed in the working cosmological model (see Standard Cosmological
Paradigm, above).
The acoustic peaks therefore contain three rulers for the angular diameter dis-
tance test for curvature, i.e., deviations fromO
tot
=1. However, contrary to popular
belief, any one of these alone is not a standard ruler whose absolute scale is known
even in the working cosmological model. This is reected in the sensitivity of
these scales to other cosmological parameters. For example, the dependence of
a
on O
m
h
2
, and hence on the Hubble constant, is quite strong. However, in combi-
nation with a measurement of the matter-radiation ratio from
eq
, this degeneracy
is broken.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
The weaker degeneracy of
a
on the baryons can likewise be broken from a
measurement of the baryon-photon ratio R
. The damping scale

d
provides an
additional consistency check on the implicit assumptions of the working model,
e.g., recombination and the energy contents of the Universe during this epoch.
What makes the peaks so valuable for this test is that the rulers are standardizable
and contain a built-in consistency check.
There remains a weak but perfect degeneracy between O
tot
and O
A
because they
both appear only in D
. This is called the angular diameter distance degeneracy in

the literature and can readily be generalized to dark energy components beyond
the cosmological constant assumed here. Because the effect of O
A
is intrinsically
so small, it creates a correspondingly small ambiguity in O
tot
for reasonable values
of O
A
. The down side is that dark energy can never be isolated through the peaks
alone because it only takes a small amount of curvature to mimic its effects. The
evidence for dark energy through the CMB comes about by allowing for external
information. The most important is the nearly overwhelming direct evidence for
O
m
-1 from local structures in the Universe. The second is the measurements of a
relatively high Hubble constant h 0.7; combined with a relatively lowO
m
h
2
that
is preferred in the CMB data, it implies O
m
-1, but at low signicance currently.
The upshot is that precise measurements of the acoustic peaks yield precise de-
terminations of four fundamental parameters of the working cosmological model:
O
m
h
2
, O
m
h
2
, D
, andn. More generally, the rst three canbe replacedby

a
.
eq
.
d
,
and R
to extend these results to models in which the underlying assumptions of

the working model are violated.
BEYOND THE PEAKS
Once the acoustic peaks in the temperature and polarization power spectra have
been scaled, the days of splendid isolation of CMB theory, analysis, and experi-
ment will have ended. Beyond and beneath the peaks lies a wealth of information
about the evolution of structure in the Universe and its origin in the early uni-
verse. As CMB photons traverse the large-scale structure of the Universe on their
journey from the recombination epoch, they pick up secondary temperature and
polarization anisotropies. These depend on the intervening dark matter, dark en-
ergy, baryonic gas density, and temperature distributions, and even the existence
of primordial gravity waves, so the potential payoff of their detection is enormous.
The price for this extended reach is the loss of the ability both to make precise
predictions, owing to uncertain and/or nonlinear physics, and to make precise
measurements, owing to the cosmic variance of the primary anisotropies and the
relatively greater importance of galactic and extragalactic foregrounds.
We begin in the following section with a discussion of the matter power spec-
trumto set the framework for the discussion of secondary anisotropies. Secondaries
can be divided into two classes: those caused by gravitational effects and those
induced by scattering off of electrons. The former are treated in Gravitational
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
194 HU

DODELSON
Secondaries and the latter in Scattering Secondaries. Secondary anisotropies
are often non-Gaussian, so they show up not only in the power spectra of Ob-
servables, above, but in higher point functions as well. We briey discuss non-
Gaussian statistics in Non-Gaussianity, below. All of these topics are subjects
of current research to which this review can only serve as an introduction.
Matter Power Spectrum
The same balance between pressure and gravity that is responsible for acoustic
oscillations determines the power spectrumof uctuations in nonrelativistic matter.
This relationship is often obscured by focusing on the density uctuations in
the pressureless cold dark matter itself, so we instead consider the matter power
spectrum from the perspective of the Newtonian potential.
PHYSICAL DESCRIPTION After recombination, without the pressure of the pho-
tons, the baryons simply fall into the Newtonian potential wells with the cold dark
matter, an event usually referred to as the end of the Compton drag epoch. We
claimed in Radiation Driving, above, that above the horizon at matter-radiation
equality the potentials are nearly constant. This follows fromthe following dynam-
ics: Where pressure gradients are negligible, infall into some initial potential causes
a potential owof :
tot
(k)+
i
(see Equation19) andcauses densityenhancements
by continuity of
tot
(k):
tot
(k)
2
+
i
. The Poisson equation says that the
potential at this later time + (k)
2
tot
+
i
, so this rate of growth is exactly
right to keep the potential constant. Formally, this Newtonian argument only ap-
plies in general relativity for a particular choice of coordinates (Bardeen 1980),
but the rule of thumb is that if what is driving the expansion (including spatial
curvature) can also cluster unimpeded by pressure, the gravitational potential will
remain constant.
Because the potential is constant in the matter-dominated epoch, the large-scale
observations of COBE set the overall amplitude of the potential power spectrum
today. Translated into density, this is the well-known COBE normalization. It is
usually expressed in terms of
H
, the matter density perturbation at the Hubble
scale today. Because the observed temperature uctuation is approximately +,3
(Sachs & Wolfe 1967),
L
2
T
T
2

1
9
L
2
+

1
4
2
H
. (25)
where the second equality follows from the Poisson equation in a fully matter-
dominated universe with O
m
=1. The observed COBE uctuation of L
T
28 jK
(Smoot et al. 1992) implies
H
2 10
5
. For corrections for O
m
-1, where the
potential decays because the dominant driver of the expansion cannot cluster, see
Bunn & White (1997).
On scales belowthe horizon at matter-radiation equality, we have seen in Radi-
ation Driving, above, that pressure gradients from the acoustic oscillations them-
selves impede the clustering of the dominant component, i.e., the photons, and lead
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
to decay in the potential. Dark matter density perturbations remain but grow only
logarithmically fromtheir value at horizon crossing, which (just as for large scales)
is approximately the initial potential,
m
+
i
. The potential for modes that have
entered the horizon already will therefore be suppressed by +
m
,k
2
+
i
,k
2
at matter domination (neglecting the logarithmic growth), again according to the
Poisson equation. The ratio of + at late times to its initial value is called the transfer
function. On large scales, then, the transfer function is close to one, and it falls off
as k
2
on small scales. If the baryons fraction
b
,
m
is substantial, baryons alter
the transfer function in two ways. First, their inability to cluster below the sound
horizon causes further decay in the potential between matter-radiation equality and
the end of the Compton drag epoch. Second, the acoustic oscillations in the bary-
onic velocity eld kinematically cause acoustic wiggles in the transfer function
(Hu & Sugiyama 1996). These wiggles in the matter power spectrum are related
to the acoustic peaks in the CMB spectrum, like twins separated at birth, and are
actively being pursued by the largest galaxy surveys (Percival et al. 2001). For
tting formulas for the transfer function that includes these effects see Eisenstein
& Hu (1998).
COSMOLOGICAL IMPLICATIONS The combination of the COBEnormalization, the
matter transfer function, and the near scale-invariant initial spectrumof uctuations
tells us that by the present, uctuations in the cold dark matter or baryon density
elds will have gone nonlinear for all scales k 10
1
h Mpc
1
. It is a great triumph
of the standard cosmological paradigm that there is just enough growth between
z
10
3
and z =0 to explain structures in the Universe across a wide range of
scales.
In particular, because this nonlinear scale also corresponds to galaxy clusters,
measurements of their abundance yielda robust measure of the power near this scale
for a givenmatter densityO
m
. The agreement betweenthe COBEnormalizationand
the cluster abundance at lowO
m
0.30.4 was pointed out immediately following
the COBE result (e.g., White et al. 1993, Bartlett & Silk 1993) and is one of the
strongest pieces of evidence for the parameters in the working cosmological model
(Ostriker & Steinhardt 1995, Krauss & Turner 1995).
More generally, the comparison between large-scale structure and the CMB is
important in that it breaks degeneracies between effects owing to deviations from
power law initial conditions and the dynamics of the matter and energy contents
of the Universe. Any dynamical effect that reduces the amplitude of the matter
power spectrum corresponds to a decay in the Newtonian potential that boosts the
level of anisotropy (see Radiation Driving, above, and Integrated Sachs-Wolfe
Effect, below). Massive neutrinos are a good example of physics that drives the
matter power spectrum down and the CMB spectrum up.
The combination is even more fruitful in the relationship between the acoustic
peaks and the baryon wiggles in the matter power spectrum. Our knowledge of the
physical distance between adjacent wiggles provides our best calibrated standard
ruler for cosmology (Eisenstein et al. 1998). For example, at very low z the radial
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
196 HU

DODELSON
distance out to a galaxy is cz,H
0
. The unit of distance is therefore h
1
Mpc,
and a knowledge of the true physical distance corresponds to a determination of
h. At higher redshifts the radial distance depends sensitively on the background
cosmology (especially the dark energy), so a future measurement of baryonic
wiggles at, for instance, z 1 would be a powerful test of dark energy models. To
a lesser extent, the shape of the transfer function, which mainly depends on the
matter-radiation scale in h Mpc
1
, i.e., O
m
h, is another standard ruler (see, e.g.,
Tegmark et al. 2001 for a recent assessment), more heralded than the wiggles but
less robust owing to degeneracy with other cosmological parameters.
For scales correspondingtok 10
1
hMpc
1
, densityuctuations are nonlinear
by the present. Numerical N-body simulations show that the dark matter is bound
up in a hierarchy of virialized structures or halos (see Bertschinger 1998 for a
review). The statistical properties of the dark matter and the dark matter halos
have been extensively studied in the working cosmological model. Less certain
are the properties of the baryonic gas. We see that both enter into the consideration
of secondary CMB anisotropies.
Gravitational Secondaries
Gravitational secondaries arise from two sources: the differential redshift from
time-variable metric perturbations (Sachs & Wolfe 1967) and gravitational lens-
ing. There are many examples of the former, one of which we have already
encountered in Integral Approach in the context of potential decay in the radia-
tion-dominated era. Such gravitational potential effects are usually called the in-
tegrated Sachs-Wolfe (ISW) effect in linear perturbation theory (see next section),
the Rees-Sciama (Rees-Sciama and Moving Halo Effects, below) effect in the
nonlinear regime, and the gravitational wave effect for tensor perturbations (Grav-
itation Waves, below). Gravitational waves and lensing also produce B-modes in
the polarization (see Polarization, below), by which they may be distinguished
from acoustic polarization.
INTEGRATED SACHS-WOLFE EFFECT As we have seen in the previous section, the
potential on a given scale decays whenever the expansion is dominated by a com-
ponent whose effective density is smooth on that scale. This occurs at late times
in an O
m
-1 model at the end of matter domination and the onset dark energy (or
spatial curvature) domination. If the potential decays between the time a photon
falls into a potential well and when it climbs out, it gets a boost in temperature
of + owing to the differential gravitational redshift and ++ owing to an
accompanying contraction of the wavelength (see Gravitational Forcing, above).
Potential decay owing to residual radiation was introduced in Integral Ap-
proach, above, but that owing to dark energy or curvature at late times induces
much different changes in the anisotropy spectrum. What makes the dark energy
or curvature contributions different from those owing to radiation is the longer
length of time over which the potentials decay, on order of the Hubble time today.
Residual radiation produces its effect quickly, so the distance over which photons
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
feel the effect is much smaller than the wavelength of the potential uctuation.
Recall that this meant that j
l
(kD) in the integral in Equation 23 could be set to
j
l
(kD
) and removed from the integral. The nal effect then is proportional to
j
l
(kD
) and adds in phase with the monopole.

The ISW projection, indeed the projection of all secondaries, is much different
(see Figure 6). Because the duration of the potential change is much longer, photons
typically travel through many peaks and troughs of the perturbation. This cancella-
tion implies that many modes have virtually no impact on the photon temperature.
The only modes that do have an impact are those with wavevectors perpendicular
to the line of sight, so that along the line of sight the photon does not pass through
crests and troughs. What fraction of the modes contributes to the effect then? For
a given wavenumber k and line of sight instead of the full spherical shell at ra-
dius 4k
2
dk, only the ring 2kdk with k n participates. Thus, the anisotropy
induced is suppressed by a factor of k (or =k D in angular space). Mathemati-
cally, this arises in the line-of-sight integral of Equation 23 from the integral over
the oscillatory Bessel function
_
dx j
(x) (,2)
1,2
(see also Figure 6).
The ISW effect thus generically shows up only at the lowest s in the power
spectrum (Kofman & Starobinskii 1985). This spectrum is shown in Figure 7
(bottom left). Secondary anisotropy predictions in this gure are for a model with
O
tot
=1, O
A
=2,3, O
b
h
2
=0.02, O
m
h
2
=0.16, n =1, andz
ri
=7andaninationary
energy scale E
i
_10
16
GeV. The ISW effect is especially important in that it is
extremely sensitive to the dark energy: its amount, equation of state, and clustering
properties (Coble et al. 1997, Caldwell et al. 1998, Hu 1998). Unfortunately, being
conned to the low multipoles, the ISW effect suffers severely from the cosmic
variance in Equation 4 in its detectability. Perhaps more promising is its correlation
with other tracers of the gravitational potential [e.g., X-ray background (Boughn
et al. 1998) and gravitational lensing (see below)].
This type of cancellation behavior and corresponding suppression of small-
scale uctuations is a common feature of secondary temperature and polarization
anisotropies from large-scale structure and is quantied by the Limber equation
(Limber 1954) and its CMB generalization (Hu & White, 1996; Hu, 2000a). It is
the central reason why secondary anisotropies tend to be smaller than the primary
ones from z
10
3
despite the intervening growth of structure.
REES-SCIAMA AND MOVING HALO EFFECTS The ISW effect is linear in the pertur-
bations. Cancellation of the ISW effect on small scales leaves second-order and
nonlinear analogues in its wake (Rees & Sciama 1968). From a single isolated
structure, the potential along the line of sight can change not only from evolution
in the density prole but more importantly from its bulk motion across the line of
sight. In the context of clusters of galaxies, this is called the moving cluster effect
(Birkinshaw & Gull 1983). More generally, the bulk motion of dark matter halos
of all masses contributes to this effect (Tuluie & Laguna 1995, Seljak 1996b), and
their clustering gives rise to a low level of anisotropies on a range of scales but
is never the leading source of secondary anisotropies on any scale (see Figure 7,
bottom left).
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
198 HU

DODELSON
Figure 8 Gravitational waves and the energy scale of ination E
i
. (Left) Temperature
and polarization spectra from an initial scale-invariant gravitational wave spectrum with
power E
4
i
=(4 10
16
GeV)
4
. (Right) 95% condence upper limits statistically achievable
on E
i
and the scalar tilt n by the MAP and Planck satellites as well as an ideal experiment out
to =3000 in the presence of gravitational lensing B-modes.
GRAVITATIONAL WAVES A time-variable tensor metric perturbation similarly
leaves an imprint in the temperature anisotropy (Sachs & Wolfe 1967). A ten-
sor metric perturbation can be viewed as a standing gravitational wave and pro-
duces a quadrupolar distortion in the spatial metric. If its amplitude changes, it
leaves a quadrupolar distortion in the CMB temperature distribution (Polnarev
1985). Ination predicts a nearly scale-invariant spectrum of gravitational waves.
Their amplitude depends strongly on the energy scale of ination,
3
(power E
4
i
)
(Rubakov et al. 1982, Fabbri & Pollock 1983), and its relationship to the curva-
ture uctuations discriminates between particular models for ination. Detection
of gravitational waves in the CMB therefore provides our best hope to study the
particle physics of ination.
Gravitational waves, like scalar elds, obey the Klein-Gordon equation in a at
universe, and their amplitudes begin oscillating and decaying once the perturbation
crosses the horizon. Although this process occurs even before recombination,
rapid Thomson scattering destroys any quadrupole anisotropy that develops (see
Damping, above). This fact dicates the general structure of the contributions to
the power spectrum (see Figure 8, left panel): They are enhanced at =2 or the
present quadrupole and sharply suppressed at a multipole larger than that of the
rst peak (Abbott & Wise 1984, Starobinskii 1985, Crittenden et al. 1993). As is
the case for the ISW effect, connement to the low multipoles means the isolation
of gravitational waves is severely limited by cosmic variance.
The signature of gravitational waves in the polarization is more distinct. Be-
cause gravitational waves cause a quadrupole temperature anisotropy at the end
of recombination, they also generate a polarization. The quadrupole generated by
3
E
4
i
V(), the potential energy density associated with the scalar eld(s) driving ination.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
a gravitational wave has its main angular variation transverse to the wavevector
itself (Hu & White 1997a). The resulting polarization has components directed
both along or orthogonal to the wavevector and at 45
angles to it. Gravitational

waves therefore generate a nearly equal amount of E and B mode polarization
when viewed at a distance that is much greater than a wavelength of the uctua-
tion (Kamionkowski et al. 1997, Zaldarriaga & Seljak 1997). The B-component
presents a promising means of measuring the gravitational waves from ination
and hence the energy scale of ination (see Figure 8, right panel). Models of ina-
tion correspond to points in the n, E
i
plane (Dodelson et al. 1997). Therefore, the
anticipated constraints discriminate among different models of ination, probing
fundamental physics at scales well beyond those accessible in accelerators.
GRAVITATIONAL LENSING The gravitational potentials of large-scale structure also
lens the CMB photons. Because lensing conserves surface brightness, it only af-
fects anisotropies and hence is second order in perturbation theory (Blanchard &
Schneider 1987). The photons are deected according to the angular gradient of the
potential projected along the line of sight with a weighting of 2(D
D),(D
D).
Again the cancellation of parallel modes implies that it is mainly the large-scale
potentials that are responsible for deections. Specically, the angular gradient of
the projected potential peaks at a multipole 60, corresponding to scales of a
k few 10
2
Mpc
1
(Hu 2000b). The deections are therefore coherent below the
degree scale. The coherence of the deection should not be confused with its rms
value, which in the model of Figure 1 is a few arcminutes.
This large coherence and small amplitude ensures that linear theory in the
potential is sufcient to describe the main effects of lensing. Because lensing is a
one-to-one mapping of the source and image planes, it simply distorts the images
formed from the acoustic oscillations in accord with the deection angle. This
warping naturally also distorts the mapping of physical scales in the acoustic peaks
to angular scales (see Integral Approach, above) and hence smooths features
in the temperature and polarization (Seljak 1996a). The smoothing scale is the
coherence scale of the deection angle L 60 and is sufciently wide to alter
the acoustic peaks with L 300. The contributions, shown in Figure 7 (bottom
left), are therefore negative (dashed) on scales corresponding to the peaks.
For the polarization, the remapping not only smooths the acoustic power spec-
trum but actually generates B-mode polarization (see Figure 1 and Zaldarriaga &
Seljak 1998). Remapping by the lenses preserves the orientation of the polarization
but warps its spatial distribution in a Gaussian random fashion and hence does not
preserve the symmetry of the original E-mode. The B-modes from lensing set a
detection threshold for gravitational waves for a nite patch of sky (Hu 2002).
Gravitational lensing also generates a small amount of power in the anisotropies
on its own, but this is only noticable beyond the damping tail, where diffusion has
destroyed the primary anisotropies (see Figure 7, bottom left). On these small
scales, the anisotropy of the CMB is approximately a pure gradient on the sky,
and the inhomogeneous distribution of lenses introduces ripples in the gradient
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
200 HU

DODELSON
on the scale of the lenses (Seljak & Zaldarriaga 2000). In fact, the moving halo
effect of Rees-Sciama and Moving Halo Effects, above, can be described as the
gravitational lensing of the dipole anisotropy owing to the peculiar motion of the
halo (Birkinshaw & Gull, 1983).
Because the lensed CMB distribution is not linear in the uctuations, it is
not completely described by changes in the power spectrum. Much of the recent
work in the literature has been devoted to utilizing the non-Gaussianity to isolate
lensing effects (Bernardeau 1997, 1998; Zaldarriaga & Seljak 1999; Zaldarriaga
2000) and their cross-correlation with the ISW effect (Goldberg & Spergel 1999,
Seljak & Zaldarriaga 1999). In particular, there is a quadratic combination of the
anisotropy data that optimally reconstructs the projected dark matter potentials
for use in this cross-correlation (Hu 2001b). The cross-correlation is especially
important in that in a at universe it is a direct indication of dark energy and can be
used to study the properties of the dark energy beyond a simple equation of state
(Hu 2002).
Scattering Secondaries
From the observations both of the lack of a Gunn-Peterson trough (Gunn &
Peterson 1965) in quasar spectra and its preliminary detection (Becker et al. 2001),
we knowthat hydrogenwas reionizedat z
ri
6. This is thought tooccur throughthe
ionizing radiation of the rst generation of massive stars (see, e.g., Loeb &Barkana
2001 for a review). The consequent recoupling of CMB photons to the baryons
causes a few percent of them to be rescattered. Linearly, rescattering induces three
changes to the photon distribution: suppression of primordial anisotropy, genera-
tion of large angle polarization, and a large angle Doppler effect. The latter two are
suppressed on small scales by the cancellation highlighted in Integrated Sachs-
Wolfe Effect, above. Nonlinear effects can counter this suppression; these are
the subject of active research and are outlined in Modulated Doppler Effect,
below.
PEAK SUPPRESSION Like scattering before recombination, scattering at late times
suppresses anisotropies in the distributions that have already formed. Reionization
therefore suppresses the amplitude of the acoustic peaks by the fraction of photons
rescattered, approximately the optical depth
ri
(see Figure 7, bottom right,
dotted line and negative, dashed line; contributions corresponding to [L
2
T
[
1,2
between the z
ri
=7 and z
ri
=0 models). Unlike the plasma before recombination,
the medium is optically thin, so the mean free path and diffusion length of the
photons is of order the horizon itself. New acoustic oscillations cannot form.
On scales approaching the horizon at reionization, inhomogeneities have yet to
be converted into anisotropies (see Integral Approach, above) so large angle
uctuations are not suppressed. Whereas these effects are relatively large compared
with the expected precision of future experiments, they mimic a change in the
overall normalization of uctuations except at the lowest, cosmic variance limited,
multipoles.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
LARGE-ANGLE POLARIZATION The rescattered radiation becomes polarized be-
cause, as discussed in Integral Approach, temperature inhomogeneities become
anisotropies by projection, passing through quadrupole anisotropies when the per-
turbations are on the horizon scale at any given time. The result is a bump in the
power spectrum of the E-polarization on angular scales corresponding to the hori-
zon at reionization (see Figure 1). Because of the low optical depth of reionization
and the nite range of scales that contribute to the quadrupole, the polarization
contributions are on the order of tenths of jK on scales of few. In a perfect,
foreground free world, this is not beyond the reach of experiments and can be used
to isolate the reionization epoch (Hogan et al. 1982, Zaldarriaga et al. 1997). As
in the ISW effect, cancellation of contributions along the line of sight guarantees
a sharp suppression of contributions at higher multipoles in linear theory. Spatial
modulation of the optical depth owing to density and ionization (see Modulated
Doppler Effects, below) does produce higher order polarization but at an entirely
negligible level in most models (Hu 2000a).
DOPPLER EFFECT Naively, velocity elds of order : 10
3
(c =1; see, e.g.,
Strauss & Willick 1995 for a review) and optical depths of a few percent would
imply a Doppler effect that rivals the acoustic peaks themselves. That this is not the
case is the joint consequence of the cancellation described in Integrated Sachs-
Wolfe Effect, above, and the fact that the acoustic peaks are not Doppler peaks
(see Integral Approach, above). Because the Doppler effect comes from the pe-
culiar velocity along the line of sight, it retains no contributions from linear modes
with wavevectors perpendicular to the line of sight. However, as we have seen,
these are the only modes that survive cancellation (see Figure 6 and Kaiser 1984).
Consequently, the Doppler effect from reionization is strongly suppressed and is
entirely negligible below 10
2
unless the optical depth in the reionization epoch
approaches unity (see Figure 7, bottom right).
MODULATED DOPPLER EFFECTS The Doppler effect can survive cancellation if the
optical depth has modulations in a direction orthogonal to the bulk velocity. This
modulation can be the result of either density or ionization uctuations in the gas.
Examples of the former include the effect in clusters and linear as well as nonlinear
large-scale structures.
Cluster modulation The strongly nonlinear modulation provided by the presence
of a galaxy cluster and its associated gas leads to the kinetic Sunyaev-Zeldovich
effect. Cluster optical depths on order of 10
2
and peculiar velocities of 10
3
im-
ply signals in the 10
5
regime in individual arcminute-scale clusters, which are of
course rare objects. Although this signal is reasonably large, it is generally dwarfed
by the thermal Sunyaev-Zeldovich effect (see below) and has yet to be detected
with high signicance (see Carlstrom et al. 2001 and references therein). The ki-
netic Sunyaev-Zeldovich effect has negligible impact on the power spectrum of
anisotropies, owing to the rarity of clusters, and can be included as part of the
general density modulation effect below.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
202 HU

DODELSON
Linear modulation At the opposite extreme, linear density uctuations modulate
the optical depth and give rise to a Doppler effect, as pointed out by Ostriker &
Vishniac (1986) and calculated by Vishniac (1987) (see also Efstathiou & Bond
1987). The result is a signal at the jKlevel peaking at few10
3
that increases
roughly logarithmically with the reionization redshift (see Figure 7, bottom right).
General density modulation Both the cluster and linear modulations are lim-
iting cases of the more general effect of density modulation by the large-scale
structure of the Universe. For the low reionization redshifts currently expected
(z
ri
67), most of the effect comes neither from clusters nor the linear regime
but from intermediate-scale dark matter halos. An upper limit to the total effect
can be obtained by assuming the gas density traces the dark matter density (Hu
2000a) and implies signals on the order of L
T
few jK at >10
3
(see Figure 7,
bottom right). Based on simulations, this assumption should hold in the outer pro-
les of halos (Pearce et al. 2001, Lewis et al. 2000), but gas pressure will tend to
smooth out the distribution in the cores of halos and reduce small scale contribu-
tions. In the absence of substantial cooling and star formation, these net effects can
be modeled under the assumption of hydrostatic equilibrium (Komatsu & Seljak
2001) in the halos and included in a halo approach to the gas distribution (Cooray
2001).
Ionization modulation Finally, optical depth modulation can also come from
variations in the ionization fraction (Aghanim et al. 1996, Gruzinov & Hu 1998,
Knox et al. 1998). Predictions for this effect are the most uncertain, as it involves
both the formation of the rst ionizing objects and the subsequent radiative transfer
of the ionizing radiation (Bruscoli et al. 2000, Benson et al. 2001). It is, however,
unlikely to dominate the density-modulated effect except perhaps at very high mul-
tipoles of 10
4
(crudely estimated, following Gruzinov & Hu 1998) (Figure 7,
bottom right).
SUNYAEV-ZELDOVICH EFFECT Internal motion of the gas in dark matter halos also
gives rise to Doppler shifts in the CMB photons. As in the linear Doppler effect,
shifts that are rst order in the velocity are canceled as photons scatter off of
electrons moving in different directions. At second order in the velocity, there is a
residual effect. For clusters of galaxies where the temperature of the gas canreachT
e
10 keV, the thermal motions are a substantial fraction of the speed of light :
rms
=
(3 T
e
,m
e
)
1,2
0.2. The second-order effect represents a net transfer of energy
between the hot electron gas and the cooler CMB and leaves a spectral distortion
in the CMB where photons on the Rayleigh-Jeans side are transferred to the Wien
tail. This effect is called the thermal Sunyaev-Zeldovich (SZ) effect (Sunyaev &
Zeldovich1972). Because the net effect is of order
cluster
T
e
,m
e
n
e
T
e
, it is a probe
of the gas pressure. Like all CMB effects, once imprinted, distortions relative
to the redshifting background temperature remain unaffected by cosmological
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
dimming, so one might hope to nd clusters at high redshift using the SZ effect.
However, the main effect comes from the most massive clusters because of the
strong temperature weighting, and these have formed only recently in the standard
cosmological model.
Great strides have recently been made in observing the SZ effect in indi-
vidual clusters, following pioneering attempts that spanned two decades
(Birkinshaw 1999). The theoretical basis has remained largely unchanged save
for small relativistic corrections as T
e
,m
e
approches unity. Both developments are
comprehensively reviewed by Carlstrom et al. (2001). Here we instead consider
the implications of the SZ effect as a source of secondary anisotropies.
The SZ effect from clusters provides the most substantial contribution to tem-
perature anisotropies beyond the damping tail. On scales much larger than an
arcminute, in which clusters are unresolved, contributions to the power spec-
trum appear as uncorrelated shot noise (C
=const. or L
T
). The additional
contribution owing to the spatial correlation of clusters turns out to be almost
negligible in comparison because of the rarity of clusters (Komatsu & Kitayama
1999). Below this scale, contributions turn over as the clusters become resolved.
Though there has been much recent progress in simulations (Refregier et al. 2000,
Seljak et al. 2001, Springel et al. 2001), dynamic range still presents a serious
limitation.
Much recent work has been devoted to semi-analytic modeling following the
technique of Cole & Kaiser (1988), in which the SZ correlations are described
in terms of the pressure proles of clusters, their abundance, and their spatial
correlations [now commonly referred to as an application of the halo model
(see Komatsu & Kitayama 1999, Atrio-Barandela & M ucket 1999, Cooray 2001,
Komatsu & Seljak 2001)]. We show the predictions of a simplied version in
Figure 7, (bottom right, Rayleigh-Jeans limit), where the pressure prole is ap-
proximated by the dark matter halo prole and the virial temperature of the halo.
Although this treatment is comparatively crude, the inaccuracies that result are
dwarfed by missing physics in both the simulations and more sophisticated
modeling, e.g., the nongravitational sources and sinks of energy that change the
temperature and density prole of the cluster, often modeled as a uniform pre-
heating of the intercluster medium (Holder & Carlstrom 2001).
Although the SZeffect is expected to dominate the power spectrumof secondary
anisotropies, it does not necessarily make the other secondaries unmeasurable or
contaminate the acoustic peaks. Its distinct frequency signature can be used to
isolate it from other secondaries (see, e.g., Cooray et al. 2000). Additionally, it
mainly comes from massive clusters that are intrinsically rare. Hence, contribu-
tions to the power spectrum are non-Gaussian and concentrated in rare, spatially
localized regions. Removal of regions identied as clusters through X rays and
optical surveys or ultimately through high-resolution CMB maps themselves can
greatly reduce contributions at large angular scales at which they are unresolved
(Persi et al. 1995, Komatsu & Kitayama 1999).
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
204 HU

DODELSON
Non-Gaussianity
As we have seen, most of the secondary anisotropies are not linear in nature
and hence produce non-Gaussian signatures. Non-Gaussianity in the lensing and
SZ signals is important for their isolation. The same is true for contaminants
such as galactic and extragalactic foregrounds. Finally, the lack of an initial non-
Gaussianity in the uctuations is a testable prediction of the simplest inationary
models (Guth & Pi 1985, Bardeen et al. 1983). Consequently, non-Gaussianity
in the CMB is a very active eld of research. The primary challenge in studies
of non-Gaussianity is in choosing the statistic that quanties it. Non-Gaussianity
tells us what the distribution is not, not what it is. The secondary challenge is to
optimize the statistic against the Gaussian noise of the primary anisotropies and
instrumental or astrophysical systematics.
Early theoretical work on the bispectrum, the harmonic analogue of the three-
point function, addressed its detectability in the presence of the cosmic variance of
the Gaussian uctuations (Luo 1994) and showed that the inationary contribution
is not expected to be detectable in most models (Allen et al. 1987, Falk et al.
1993, Gangui et al. 1994). The bispectrum is dened by a triplet of multipoles, or
conguration, that denes a triangle in harmonic space. The large cosmic variance
in an individual conguration is largely offset by the great number of possible
triplets. Interest was spurred by reports of signicant signals in specic bispectrum
congurations in the COBE maps (Ferreira et al. 1998) that turned out to be caused
by systematic errors (Banday et al. 2000). Recent investigations have focused on
the signatures of secondary anisotropies (Goldberg & Spergel 1999, Cooray &
Hu 2000). These turned out to be detectable with experiments that have both high
resolution and angular dynamic range but require the measurement of a wide range
of congurations of the bispectrum. Data analysis challenges for measuring the full
bispectrumlargely remain to be addressed (see Heavens 1998, Spergel &Goldberg
1999, Phillips & Kogut 2001).
The trispectrum, the harmonic analogue of the four-point function, also has
advantages for the study of secondary anisotropies. Its great number of cong-
urations are specied by a quintuplet of multipoles that correspond to the sides
and diagonal of a quadrilateral in harmonic space (Hu 2001a). The trispectrum is
important in that it quanties the covariance of the power spectrum across mul-
tipoles that are often very strong in nonlinear effects, e.g., the SZ effect (Cooray
2001). It is also intimately related to the power spectra of quadratic combinations
of the temperature eld and has been applied to study gravitational lensing effects
(Bernardeau 1997, Zaldarriaga 2000, Hu 2001a).
The bispectrum and trispectrum quantify non-Gaussianity in harmonic space
and have clear applications for secondary anisotropies. Tests for non-Gaussianity
localized in angular space include the Minkowski functionals (including the genus)
(Winitzki & Kosowsky 1997), the statistics of temperature extrema (Kogut et al.
1996), and wavelet coefcients (Aghanim & Forni 1999). These may be more
useful for examining foreground contamination and trace amounts of topological
defects.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
DATA ANALYSIS
The very large CMB data sets that have begun arriving require new, innovative
tools of analysis. The fundamental tool for analyzing CMB datathe likelihood
functionhas been used since the early days of anisotropy searches (Readhead
et al. 1989, Bond et al. 1991, Dodelson & Jubas 1993). Brute-force likelihood
analyses (Tegmark & Bunn 1995) were performed even on the relatively large
COBE data set, with 6000 pixels in its map. Present data sets are a factor of 10
larger, and this factor will soon increase by yet another factor of 100. The brute-
force approach, the time for which scales as the number of pixels cubed, no longer
sufces.
In response, analysts have devised a host of techniques that move beyond the
early brute-force approach. The simplicity of CMB physicsowing to linearity
is mirrored in analysis by the apparent Gaussianity of both the signal and many
sources of noise. In the Gaussian limit, optimal statistics are easy to identify.
These compress the data so that all of the information is retained, but the subsequent
analysisbecause of the compressionbecomes tractable.
The Gaussianity of the CMB is not shared by other cosmological systems
because gravitational nonlinearities turn an initially Gaussian distribution into a
non-Gaussian one. Nontheless, many of the techniques devised to study the CMB
have been proposed for studying the 3D galaxy distribution (Tegmark et al. 1998),
the 2D galaxy distribution (Efstathiou & Moody 2001, Huterer et al. 2001), the
Lyman alpha forest (Hui et al. 2001), and the shear eld from weak lensing (Hu
& White 2001), among others. Indeed, these techniques are now indispensible,
powerful tools for all cosmologists, and we would be remiss not to at least outline
themin a disussion of the CMB, the context inwhichmanyof themwere developed.
Figure 9 summarizes the path from the data analyis starting point, a timestream
of data points, to the end, the determination of cosmological parameters. Preceding
this starting point comes the calibration and the removal of systematic errors from
the raw data, but being experiment specic, we do not attempt to cover such
Figure 9 Data pipeline and radical compression. Maps are constructed for each frequency
channel from the data timestreams, combined, and cleaned of foreground contamination by
spatial (represented here by excising the galaxy) and frequency information. Bandpowers
are extracted from the maps and cosmological parameters from the bandpowers. Each step
involves a substantial reduction in the number of parameters needed to describe the data,
from potentially 10
10
10 for the Planck satellite.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
206 HU

DODELSON
issues here.
4
Each step radically compresses the data by reducing the number of
parameters used to describe it. Although this data pipeline and our discussion
below focus on temperature anisotropies, similar steps have been elucidated for
polarization (Bunn 2001, Tegmark & de Oliveira-Costa 2001, Lewis et al. 2001).
Mapmaking
An experiment can be characterized by the data d
t
taken at many different times;
a pointing matrix, P
ti
, relating the data timestream to the underlying signal
at pixelized positions indexed by i; and a noise matrix C
d.t t
/ characterizing the
covariance of the noise in the timestream. Amodel for the data then is d
t
=P
ti
O
i
n
t
(with implicit sum over the repeating index i); it is the sum of signal plus noise.
Here n
t
is drawnfroma distribution(oftenGaussian) withmeanzeroandcovariance
n
t
n
t
/ ) =C
d.t t
/ . In its simplest form the pointing matrix P contains rowswhich
correspond to a particular timewith all zeroes in it except for one column with
a one (see Figure 9). That column corresponds to the particular pixel observed
at the time of interest. Typically, a pixel will be scanned many times during an
experiment, so a given column will have many ones in it, corresponding to the
many times the pixel has been observed.
Given this model, a well-posed question is, What is the optimal estimator for
the signal O
i
? In effect, what is the best way to construct a map? The answer stems
from the likelihood function L, dened as the probability of getting the data, given
the theory L P[data[theory]. In this case the theory is the set of parameters O
i
,
L
O
(d
t
) =
1
(2)
N
t
,2
det C
d
exp
_
1
2
(d
t
P
ti
O
i
) C
1
d.t t
/
(d
t
/ P
t
/
j
O
j
)
_
. (26)
That is, the noise, the difference between the data and the modulated signal, is
assumed to be Gaussian with covariance C
d
.
Two important theorems are useful in the construction of a map and more
generally in each step of the data pipeline (Tegmark et al. 1997). The rst is Bayes
theorem. In this context, it says that P[O
i
[d
t
], the probability that the temperatures
are equal to O
i
given the data, is proportional to the likelihood function times a
prior P(O
i
). Thus, with a uniform prior,
P[O
i
[d
t
] P[d
t
[O
i
] L
O
(d
t
). (27)
with the normalization constant determined by requiring the integral of the prob-
ability over all O
i
to be equal to one. The probability on the left is the one of
interest. The most likely values of O
i
therefore are those that maximize the likeli-
hood function. Because the log of the likelihood function in question, Equation 26,
is quadratic in the parameters O
i
, it is straightforward to nd this maximum point.
Differentiating the argument of the exponential with respect to O
i
and setting to
4
Aside from COBE, experiments to date have had a sizable calibration error (510%) that
must be factored into the interpretation of Figure 1.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
zero leads immediately to the estimator
O
i
= C
N.i j
P
j t
C
1
d.t t
/
d
t
/ . (28)
where C
N
(P
tr
C
d
1
P)
1
. As the notation suggests, the mean of the estimator is
equal to the actual O
i
, and the variance is equal to C
N
.
The second theorem states that this maximum likelihood estimator is also the
minimum variance estimator. The Cramer-Rao inequality says no estimator can
measure the O
i
with errors smaller than the diagonal elements of F
1
, where the
Fisher matrix is dened as
F
O.i j

_
2
ln L
O
O
i
O
j
_
. (29)
Inspection of Equation 26 shows that, in this case, the Fisher matrix is precisely
equal to C
1
N
. Therefore, the Cramer-Rao theorem implies that the estimator of
Equation 28 is optimal: It has the smallest possible variance (Tegmark 1997a).
No information is lost if the map is used in subsequent analysis instead of the
timestream data, but huge factors of compression have been gained. For example,
in the recent Boomerang experiment (Nettereld et al. 2001), the timestream con-
tained 2 10
8
numbers, and the map had only 57,000 pixels. The map resulted in
compression by a factor of 3500.
There are numerous complications that must be dealt with in realistic applica-
tions of Equation 28. Perhaps the most difcult is estimation of C
d
, the timestream
noise covariance. This typically must be done from the data itself (Ferreira & Jaffe
2000, Stompor et al. 2001). Even if C
d
were known perfectly, evaluation of the map
involves inverting C
d
, a process that scales as the number of rawdata points cubed.
For both of these problems, the assumed stationarity of C
d,t t
/ (it depends only on
t t
/
) is of considerable utility. Iterative techniques toapproximate matrixinversion
can also assist in this process (Wright et al. 1996). Another issue that has received
much attention is the choice of pixelization. The CMB community has converged
on the Healpix pixelization scheme (http://www.eso.org/science/healpix/), now
freely available.
Perhaps the most dangerous complicationarises fromastrophysical foregrounds,
both within and from outside the Galaxy, the main ones being synchrotron, brem-
msstrahlung, dust, and point source emission. All of the main foregrounds have
spectral shapes different than the blackbody shape of the CMB. Modern experi-
ments typically observe at several different frequencies, so a well-posed question
is, How can we best extract the CMB signal from the different frequency channels
(Bouchet & Gispert 1999)? The blackbody shape of the CMB relates the signal
in all the channels, leaving one free parameter. Similarly, if the foreground shapes
are known, each foreground comes with just one free parameter per pixel. A like-
lihood function for the data can again be written down and the best estimator for
the CMB amplitude determined analytically. In the absence of foregrounds, one
would extract the CMB signal by weighting the frequency channels according to
inverse noise. When foregrounds are present, the optimal combination of different
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
208 HU

DODELSON
frequency maps is a more clever weighting that subtracts out the foreground con-
tribution (Dodelson 1997). One can do better if the pixel-to-pixel correlations of
the foregrounds can also be modeled from power spectra (Tegmark & Efstathiou
1996) or templates derived from external data.
This picture is complicated somewhat because the foreground shapes are not
precisely known, varying across the sky, e.g., froma spatially varying dust temper-
ature. This too can be modeled in the covariance and addressed in the likelihood
analysis (Tegmark 1998, White 1998). The resulting cleaned CMB map is obvi-
ously noisier than if foregrounds were not around, but the multiple channels keep
the degradation managable. For example, the errors on some cosmological para-
meters coming fromPlanck may degrade by almost a factor of 10 as compared with
the no-foreground case. However, many errors will not degrade at all, and even the
degraded parameters will still be determined with unprecedented precision (Knox
1999, Prunet et al. 2000, Tegmark et al. 2000).
Many foregrounds tend to be highly non-Gaussian, in particular well-localized
regions of the map. These pixels can be removed from the map as was done for
the region around the galactic disk for COBE. This technique can also be highly
effective against point sources. Indeed, even if there is only one frequency channel,
external foreground templates set the form of the additional contributions to C
N
,
which, when properly included, immunize the remaining operations in the data
pipeline to such contaminants (Bond et al. 1998). The same technique can be used
with templates of residual systematics or constraints imposed on the data, from,
e.g., the removal of a dipole.
Bandpower Estimation
Figure 9 indicates that the next step in the compression process is extracting
bandpowers from the map. What is a bandpower and how can it be extracted from
the map? To answer these questions, we must construct a new likelihood function,
one in which the estimated O
i
are the data. No theory predicts an individual O
i
, but
all predict the distribution from which the individual temperatures are drawn. For
example, if the theory predicts Gaussian uctuations, then O
i
is distributed as a
Gaussian with mean zero and covariance equal to the sum of the noise covariance
matrix C
N
and the covariance due to the nite sample of the cosmic signal C
S
.
Inverting Equation 1 and using Equation 2 for the ensemble average leads to
C
S.i j
O
i
O
j
) =
L
2
T.
W
.i j
. (30)
where L
2
T.
depends on the theoretical parameters through C
(see Equation 3).

Here W
, the window function, is proportional to the Legendre polynomial

P
( n
i
n
j
) and a beam and pixel smearing factor b
2
. For example, a Gaussian

beam of width dictates that the observed map is actually a smoothed picture
of true signal, insensitive to structure on scales smaller than . If the pixel scale
is much smaller than the beam scale, b
2
e
(1)
2
. Techniques for handling
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
asymmetric beams have also recently been developed (Wu et al. 2001, Wandelt &
Gorski 2001, Souradeep & Ratra 2001, Fosalba et al. 2001). Using bandpowers
corresponds to assuming that L
2
T.
is constant over a nite range, or band, of ,
equal to B
a
for
a
a
,2 - -
a
a
,2. Figure 1 gives a sense of the width
and number of bands N
b
probed by existing experiments.
For Gaussian theories, then, the likelihood function is
L
B
(O
i
) =
1
(2)
N
p
,2
det C
O
exp
_
1
2
O
i
C
1
O.i j
O
j
_
. (31)
where C
O
=C
S
C
N
and N
p
is the number of pixels in the map. As before, L
B
is
Gaussian in the anisotropies O
i
, but in this case O
i
are not the parameters to be
determined; the theoretical parameters are the B
a
, uponwhichthe covariance matrix
depends. Therefore, the likelihood function is not Gaussian in the parameters, and
there is no simple, analytic way to nd the point in parameter space (which is
multidimensional depending on the number of bands being t) at which L
B
is a
maximum. An alternative is to evaluate L
B
numerically at many points in a grid
in parameter space. The maximum of the L
B
on this grid then determines the
best t values of the parameters. Condence levels on, e.g., B
1
can be determined
by nding the region within which
_
b
a
dB
1
[H
N
b
i =2
_
dB
i
]L
B
=0.95, e.g., for 95%
limits.
This possibility is no longer viable owing to the sheer volume of data. Consider
the Boomerang experiment with N
p
=57,000. A single evaluation of L
B
involves
computationof the inverse anddeterminant of the N
p
N
p
matrixC
O
, bothof which
scale as N
3
p
. Whereas this single evaluation might be possible with a powerful
computer, a single evaluation does not sufce. The parameter space consists of
N
b
=19 bandpowers equally spaced froml
a
=100 up to l
a
=1000. Ablindly placed
grid on this space would require at least 10 evaluations in each dimension, so the
time required to adequately evaluate the bandpowers would scale as 10
19
N
3
p
. No
computer can do this. The situation is rapidly getting worse (better) because Planck
will have of order 10
7
pixels and be sensitive to of order 10
3
bands.
It is clear that a smart sampling of the likelihood in parameter space is nec-
essary. The numerical problem, searching for the local maximum of a function,
is well posed, and a number of search algorithms might be used. L
B
tends to be
sufciently structureless that these techniques sufce. Bond et al. (1998) proposed
the Newton-Raphson method, which has become widely used. One expands the
derivative of the log of the likelihood functionwhich vanishes at the true maxi-
mum of L
B
around a trial point in parameter space, B
(0)
a
. Keeping terms second
order in B
a
B
(0)
a
leads to
B
a
=

B
(0)
a

F
1
B.ab
ln L
B
B
b
. (32)
where the curvature matrix

F
B.ab
is the second derivative of ln L
B
with respect
to B
a
and B
b
. Note the subtle distinction between the curvature matrix and the
Fisher matrix in Equation 29, F=
F). In general, the curvature matrix depends

A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
210 HU

DODELSON
on the data, on the O
i
. In practice, though, analysts typically use the inverse of the
Fisher matrix in Equation 32. In that case, the estimator becomes
B
a
=

B
(0)
a

1
2
F
1
B.ab
_
O
i
C
1
O.i j
C
O. j k
B
b
C
1
O.ki
O
i
C
1
O.i j
C
O. j i
B
b
_
. (33)
quadratic in the data O
i
. The Fisher matrix is equal to
F
B.ab
=
1
2
C
1
O.i j
C
O. j k
B
a
C
1
O.kl
C
O.li
B
b
. (34)
In the spirit of the Newton-Raphson method, Equation 33 is used iteratively but
often converges after just a handful of iterations. The usual approximation is then to
take the covariance between the bands as the inverse of the Fisher matrix evaluated
at the convergent point C
B
=F
1
B
. Indeed, Tegmark (1997b) derived the identical
estimator by considering all unbiased quadratic estimators and identifying this one
as the one with the smallest variance.
Although the estimator in Equation 33 represents a 10
N
b
improvement over
brute force coverage of the parameter spaceconverginginjust several iterations
it still requires operations that scale as N
3
p
. One means of speeding up the calcula-
tions is to transform the data from the pixel basis to the so-called signal-to-noise
basis, based on an initial guess as to the signal and throwing out those modes that
have low signal to noise (Bond 1995, Bunn & Sugiyama 1995). The drawback is
that this procedure still requires at least one N
3
p
operation and potentially many,
as the guess at the signal improves by iteration. Methods to truly avoid this pro-
hibitive N
3
p
scaling (Oh et al. 1999, Wandelt & Hansen 2001) have been devised
for experiments with particular scan strategies, but the general problem remains.
A potentially promising approach involves extracting the real space correlation
functions as an intermediate step between the map and the bandpowers (Szapudi
et al. 2001). Another involves consistently analyzing coarsely pixelized maps with
nely pixelized submaps (Dore et al. 2001).
Cosmological Parameter Estimation
The huge advantage of bandpowers is that they represent the natural meeting
ground of theory and experiment. The above two sections outline some of the
steps involved in extracting them from the observations. Once they are extracted,
any theory can be compared with the observations without knowledge of experi-
mental details. The simplest way to estimate the cosmological parameters {c
i
} is
to approximate the likelihood as
L
c
(

B
a
)
1
(2)
N
c
,2
detC
B
exp
_
1
2
(

B
a
B
a
)C
1
B.ab
(

B
b
B
b
)
_
(35)
and evaluate it at many points in parameter space (the bandpowers depend on
the cosmological parameters). Because the number of cosmological parameters
in the working model is N
c
10, this represents a nal radical compression of
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
information in the original timestream, which recall has up to N
t
10
10
data
points.
In the approximation that the bandpower covariance C
B
is independent of the
parameters c, maximizing the likelihood is the same as minimizing
2
the argument
of the exponential. This has been done by dozens of groups over the past few
years, especially since the release of CMBFAST (Seljak &Zaldarriaga, 1996), which
allows fast computation of theoretical spectra. Even after all the compression
summarized in Figure 9, these analyses are still computationally cumbersome
owing to the large numbers of parameters varied. Various methods of speeding up
spectra computation have been proposed (Tegmark & Zaldarriaga 2000), based on
the understanding of the physics of peaks outlined in Acoustic Peaks, above, and
Monte Carlo explorations of the likelihood function (Christensen et al. 2001).
Again the inverse Fisher matrix gives a quick and dirty estimate of the errors.
Here the analogue of Equation 29 for the cosmological parameters becomes
F
c.i j
=
B
a
c
i
C
1
B.ab
B
b
c
j
. (36)
In fact, this estimate has been widely used to forecast the optimal errors on cosmo-
logical parameters, given a proposed experiment and a band covariance matrix C
B
that includes diagonal sample and instrumental noise variance. The reader should
be aware that no experiment to date has even come close to achieving the precision
implied by such a forecast!
As we enter the age of precision cosmology, a number of caveats will become
increasingly important. No theoretical spectra are truly at in a given band, so the
question of how to weight a theoretical spectrum to obtain B
a
can be important. In
principle, one must convolve the theoretical spectra with window functions (Knox
1999) distinct from those in Equation 30 to produce B
a
. Among recent experi-
ments, Degree Angular Scale Interferometer (DASI) (Pryke et al. 2001), among
others, have provided these functions. Another complication arises because the
true likelihood function for B
a
is not Gaussian, i.e., not of the form in Equation 35.
The true distribution is skewed: The cosmic variance of Equation 4 leads to larger
errors for an upward uctuation than for a downward uctuation. The true distri-
bution is closer to log-normal (Bond et al. 2000), and several groups have already
accounted for this in their parameter extractions.
DISCUSSION
Measurements of the acoustic peaks in the CMB temperature spectrum have al-
ready shown that the Universe is nearly spatially at and began with a nearly
scale-invariant spectrum of curvature uctuations, consistent with the simplest
of inationary models. In a remarkable conrmation of a prediction of big bang
nucleosynthesis, the CMB measurements have now veried that baryons account
for about 4% of the critical density. Further, they suggest that the matter
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
212 HU

DODELSON
density is about 10 times higher than this, implying the existence of nonbaryonic
dark matter and dark energy.
Future measurements of the morphology of the peaks in the temperature and
polarization should determine the baryonic and dark matter content of the Universe
with exquisite precision. Beyond the peaks, gravitational wave imprint on the
polarization, the gravitational lensing of the CMB, and gravitational and scattering
secondary anisotropies hold the promise of elucidating the physics of ination and
the impact of dark energy on structure formation.
The once and future success of the CMB anisotropy enterprise rests on three
equally important pillars: advances in experimental technique, precision in theory,
and development of data analysis techniques. The remarkable progress in the eld
over the past decade owes much to the efforts of researchers in all three disciplines.
That muchmore effort will be requiredtofulll the bright promise of CMBsuggests
that the eld will remain active and productive for years to come.
ACKNOWLEDGMENTS
W.H. thanks Fermilab, where this review was written, for their hospitality. W.H.
was supported by NASA NAG5-10840 and the DOE OJI program. S.D. was sup-
ported by the DOE, by NASA grant NAG 5-10842 at Fermilab, and by NSF Grant
PHY-0079251 at Chicago.
The Annual Review of Astronomy and Astrophysics is online at
http://astro.annualreviews.org
LITERATURE CITED
Abbott LF, Schaefer RK. 1986. Ap. J. 308:546
62
Abbott LF, Wise MB. 1984. Nucl. Phys. B244:
54148
Aghanim N, Desert FX, Puget JL, Gispert R.
1996. Astron. Astrophys. 311:111
Aghanim N, Forni O. 1999. Astron. Astrophys.
347:40918
Albrecht A, Coulson D, Ferreira P, Magueijo J.
1996. Phys. Rev. Lett. 76:141316
Allen B, Caldwell RR, Dodelson S, Knox L,
Shellard EPS, et al. 1997. Phys. Rev. Lett.
79:262427
Allen TJ, Grinstein B, Wise MB. 1987. Phys.
Lett. B197:6670
Atrio-Barandela F, M ucket JP. 1999. Ap. J.
515:46570
Banday AJ, Zaroubi S, G orski KM. 2000. Ap.
J. 533:57587
Bardeen JM. 1980. Phys. Rev. D 22:1882905
Bardeen JM, Steinhardt PJ, Turner MS. 1983.
Phys. Rev. D 28:67993
Bartlett JG, Silk J. 1993. Ap. J. Lett. 407:L45
48
Becker R, Fan X, White R, Strauss M, Nara-
yanan V, et al. 2001. Astron. J. 122:2850
57
Benson AJ, Nusser A, Sugiyama N, Lacey CG.
2001. MNRAS 320:15376
BernardeauF. 1997. Astron. Astrophys. 324:15
26
Bernardeau F. 1998. Astron. Astrophys. 338:
76776
Bertschinger E. 1998. Annu. Rev. Astron. Astro-
phys. 36:599654
Birkinshaw M. 1999. Phys. Rep. 310:97195
BirkinshawM, Gull SF. 1983. Nature 302:315
17
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Blanchard A, Schneider J. 1987. Astron. Astro-
phys. 184:12
Boesgaard AM, Steigman G. 1985. Annu. Rev.
Astron. Astrophys. 23:31978
Bond JR, Efstathiou G, Lubin P, Meinhold P.
1991. Phys. Rev. Lett. 66:217982
Bond JR. 1995. Phys. Rev. Lett. 74:436972
Bond JR, Efstathiou G. 1984. Ap. J. Lett. 285:
L4548
BondJR, EfstathiouG. 1987. MNRAS226:655
87
Bond JR, Efstathiou G, Tegmark M. 1997.
MNRAS 291:L3341
Bond JR, Jaffe AH, Knox L. 1998. Phys. Rev.
D 57:211737
Bond JR, Jaffe AH, Knox L. 2000. Ap. J. 533:
1937
Bouchet F, Gispert R. 1999. NewAstron. 4:443
79
Boughn SP, Crittenden RG, Turok NG. 1998.
New Astron. 3:27591
Bruscoli M, Ferrara A, Fabbri R, Ciardi B.
2000. MNRAS 318:106872
Bunn EF. 2001. Phys. Rev. D 65:043003
Bunn EF, Sugiyama N. 1995. Ap. J. 446:4953
Bunn EF, White M. 1997. Ap. J. 480:621
Caldwell RR, Dave R, Steinhardt PJ. 1998.
Phys. Rev. Lett. 80:158285
Carlstrom J, Joy M, Grego L, Holder G, Hol-
zapfel W, et al. 2001. IAP Conf. In press
(astro-ph/0103480)
ChristensenN, Meyer R, KnoxL, LueyB. 2001.
Class. Quantum Gravity 18:2677
Coble K, Dodelson S, Frieman JA. 1997. Phys.
Rev. D 55:185159
Cole S, Kaiser N. 1988. MNRAS 233:63748
Cooray A. 2001. Phys. Rev. D 64:063514
Cooray A, Hu W. 2000. Ap. J. 534:53350
Cooray A, Hu W, Tegmark M. 2000. Ap. J.
540:113
Crittenden R, Bond JR, Davis RL, Efstathiou G,
Steinhardt PJ. 1993. Phys. Rev. Lett. 71:324
27
de Bernardis P, Ade P, Bock J, Bond J, Borrill
J, et al. 2002. Ap. J. 564:55966
de Bernardis P, Ade PAR, Bock JJ, Bond JR,
Borrill J, et al. 2000. Nature 404:95559
Dodelson S. 1997. Ap. J. 482:57787
Dodelson S, Gates E, Stebbins A. 1996. Ap. J.
467:1018
Dodelson S, Jubas JM. 1993. Phys. Rev. Lett.
70:222427
Dodelson S, Kinney WH, Kolb EW. 1997. Phys.
Rev. D 56:320715
Dore O, Knox L, Peel A. 2001. Phys. Rev. D
64:083001
Doroshkevich AG, ZelDovich YB, Sunyaev
RA. 1978. Sov. Astron. 22:52328
Efstathiou G, Bond JR. 1987. MNRAS 227:
33P38
Efstathiou G, Moody SJ. 2001. MNRAS 325:
160315
Eisenstein DJ, Hu W, Tegmark M. 1998. Ap. J.
Lett. 504:L5760
Eisenstein DJ, Hu W. 1998. Ap. J. 496:60514
Eisenstein DJ, Hu W, Tegmark M. 1999. Ap. J.
518:223
Fabbri R, Pollock MD. 1983. Phys. Lett. B125:
44548
Falk T, Rangarajan R, Srednicki M. 1993. Ap.
J. Lett. 403:L13
Ferreira PG, Jaffe AH. 2000. MNRAS 312:89
102
Ferreira PG, Magueijo J, Gorski KM. 1998. Ap.
J. Lett. 503:L14
Fixsen DJ, Cheng ES, Gales JM, Mather JC,
Shafer RA, et al. 1996. Ap. J. 473:57687
Fosalba P, Dore O, Bouchet FR. 2001. Phys.
Rev. D 65:063003
Freedman WL, Madore BF, Gibson BK, Ferra-
rese L, Kelson DD, et al. 2001. Ap. J. 553:47
72
Gangui A, Lucchin F, Matarrese S, Molkenach
S. 1994. Ap. J. 430:44757
Goldberg DM, Spergel DN. 1999. Phys. Rev. D
59:103002
Gruzinov A, Hu W. 1998. Ap. J. 508:43539
Gunn JE, Peterson BA. 1965. Ap. J. 142:1633
36
Guth AH, Pi SY. 1985. Phys. Rev. D 32:1899
920
Halverson NW, Leitch EM, Pryke C, Kovac J,
Carlstrom JE, et al. 2001. Ap. J. In press
(astro-ph/0104489)
Hanany S, Ade P, Balbi A, Bock J, Borrill J,
et al. 2000. Ap. J. Lett. 545:L59
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
214 HU

DODELSON
Hawking SW. 1982. Phys. Lett. B115:29597
Heavens AF. 1998. MNRAS 299:8058
HedmanMM, Barkats D, GundersenJO, Staggs
ST, Winstein B. 2001. Ap. J. Lett. 548:L111
14
Hogan CJ, Kaiser N, Rees MJ. 1982. Philos.
Trans. R. Soc. London Ser. 307:97109
Holder G, CarlstromJ. 2001. Ap. J. 558:51559
Hu W. 1998. Ap. J. 506:48594
Hu W. 2000a. Ap. J. 529:1225
Hu W. 2000b. Phys. Rev. D 62:043007
Hu W. 2001a. Phys. Rev. D 64:083005
Hu W. 2001b. Ap. J. Lett. 557:L7983
Hu W. 2002. Phys. Rev. D 65:023003
Hu W, Fukugita M, Zaldarriaga M, Tegmark M.
2001. Ap. J. 549:66980
Hu W, Seljak U, White M, Zaldarriaga M. 1998.
Phys. Rev. D 57:3290301
Hu W, Sugiyama N. 1995. Ap. J. 444:489
506
Hu W, Sugiyama N. 1996. Ap. J. 471:54270
Hu W, Sugiyama N, Silk J. 1997. Nature 386:
3743
Hu W, White M. 1996. Ap. J. 471:3051
Hu W, White M. 1997a. New Astron. 2:32344
Hu W, White M. 1997b. Phys. Rev. D 56:596
615
Hu W, White M. 1997c. Ap. J. 479:56879
Hu W, White M. 2001. Ap. J. 554:6773
Hui L, Burles S, Seljak U, Rutledge RE, Mag-
nier E, et al. 2001. Ap. J. 552:1535
Huterer D, Knox L, Nichol RC. 2001. Ap. J.
555:54757
Jungman G, Kamionkowski M, Kosowsky A,
Spergel DN. 1996. Phys. Rev. D 54:133244
Kaiser N. 1983. MNRAS 202:116980
Kaiser N. 1984. Ap. J. 282:37481
Kamionkowski M, Kosowsky A, Stebbins A.
1997. Phys. Rev. D 55:736888
Kamionkowski M, Spergel DN, Sugiyama N.
1994. Ap. J. Lett. 426:L5760
Keating B, ODell C, de Oliveira-Costa A,
Klawikowski S, Stebor N, et al. 2001. Ap.
J. Lett. 560:L14
Knox L. 1995. Phys. Rev. D 52:430718
Knox L. 1999. MNRAS 307:97783
Knox L, Christensen N, Skordis C. 2001. Ap. J.
Lett. 563:L9598
Knox L, Scoccimarro R, Dodelson S. 1998.
Kofman LA, Starobinskii AA. 1985. Sov. As-
tron. Lett. 11:27174
Kogut A, Banday AJ, Bennett CL, Gorski KM,
HinshawG, et al. 1996. Ap. J. Lett. 464:L29
33
Komatsu E, Kitayama T. 1999. Ap. J. Lett. 526:
L14
Komatsu E, Seljak U. 2001. MNRAS 327:1353
66
Krauss LM, Turner MS. 1995. Gen. Relativ.
Gravity 27:113744
Lee A, Ade P, Balbi A, Bock J, Borill J, et al.
2001. Ap. J. Lett. 561:L15
Lewis A, Challinor A, Turok N. 2001. Phys.
Rev. D 65:023505
Lewis GF, Babul A, Katz N, Quinn T, Hernquist
L, et al. 2000. Ap. J. 536:62344
Liddle AR, Lyth DH. 1993. Phys. Rep. 231:1
105
Limber DN. 1954. Ap. J. 119:65581
Loeb A, Barkana R. 2001. Annu. Rev. Astron.
Astrophys. 39:1966
Luo X. 1994. Ap. J. Lett. 427:L7174
Ma C, Bertschinger E. 1995. Ap. J. 455:725
Miller AD, Caldwell R, Devlin MJ, Dorwart
WB, Herbig T, et al. 1999. Ap. J. Lett. 524:
L14
Narkilar J, Padmanabhan T. 2001. Annu. Rev.
Nettereld C, Ade P, Bock J, Bond J, Borrill J,
et al. 2001. Ap. J. In press (astro-ph/0104460)
Oh SP, Spergel DN, Hinshaw G. 1999. Ap. J.
510:55163
Ostriker JP, Steinhardt PJ. 1995. Nature 377:
6002
Ostriker JP, Vishniac ET. 1986. Ap. J. Lett. 306:
L5154
Padin S, Cartwright JK, Mason BS, Pearson
TJ, Readhead ACS, et al. 2001. Ap. J. Lett.
549:L15
Parodi BR, Saha A, Sandage A, Tamman GA.
2000. Ap. J. 540:63451
Peacock JA. 1991. MNRAS 253:1P5P
Pearce FR, Jenkins A, Frenk CS, White SDM,
Thomas PA, et al. 2001. MNRAS 326:64966
Peebles PJE. 1968. Ap. J. 153:111
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Peebles PJE. 1982. Ap. J. Lett. 263:L15
Peebles PJE, Yu JT. 1970. Ap. J. 162:81536
Penzias AA, Wilson RW. 1965. Ap. J. 142:419
21
Percival W, Baigj CM, Bland-Hawthorn J,
Bridges T, Cannon R, et al. 2001. MNRAS
327:12971306
Perlmutter S, Aldering G, Goldnater G, Knop
RA, Nugent P, et al. 1999. Ap. J. 517:56586
Persi FM, Spergel DN, CenR, Ostriker JP. 1995.
Ap. J. 442:19
Phillips NG, Kogut A. 2001. Ap. J. 548:54049
Polnarev AG. 1985. Sov. Astron. 29:607
Prunet S, Sethi SK, Bouchet FR. 2000. MNRAS
314:34853
Pryke C, Halverson N, Leitch E, Kovac J,
Carlstrom J, et al. 2001. Ap. J. In press
(astro-ph/0104490)
Readhead A, Lawrence C, Myers S, Sargent W,
Hardebeck H, et al. 1989. Ap. J. 346:56687
Rees M, Sciama D. 1968. Nature 217:511
Refregier A, Komatsu E, Spergel DN, Pen UL.
2000. Phys. Rev. D 61:123001
Riess A, Fillipenko AV, Challis P, Clocchiatti
A, Diercks A, et al. 1998. Ap. J. 116:1009
Rubakov VA, Sazhin MV, Veryaskin AV. 1982.
Phys. Lett. B 115:18992
Sachs RK, Wolfe AM. 1967. Ap. J. 147:7390
Sandage A, Tamman GA, Saha A. 1998. Phys.
Rep. 307:114
Schramm DN, Turner MS. 1998. Rev. Mod.
Phys. 70:30318
Scott D, Srednicki M, White M. 1994. Ap. J.
Lett. 421:L57
Seager S, Sasselov DD, Scott D. 2000. Ap. J.
Suppl. 128:40730
Seljak U. 1994. Ap. J. Lett. 435:L8790
Seljak U. 1996a. Ap. J. 463:17
Seljak U. 1996b. Ap. J. 460:54955
Seljak U. 1997. Ap. J. 482:616
Seljak U, Burwell J, Pen UL. 2001. Phys. Rev.
D 63:063001
Seljak U, Pen U, Turok N. 1997. Phys. Rev. Lett.
79:161518
Seljak U, Zaldarriaga M. 1996. Ap. J. 469:437
44
Seljak U, Zaldarriaga M. 1999. Phys. Rev. D
60:043504
Seljak U, Zaldarriaga M. 2000. Ap. J. 538:57
64
Silk J. 1968. Ap. J. 151:45971
Smoot GF, Bennett CL, Kogut A, Wright EL,
Aymon J, et al. 1992. Ap. J. Lett. 396:L15
Smoot GF, Gorenstein MV, Muller RA. 1977.
Souradeep T, Ratra BV. 2001. Ap. J. 560:2840
Spergel D, Goldberg D. 1999. Phys. Rev. D 59:
103001
Springel V, White M, Hernquist L. 2001. Ap. J.
549:68187
Starobinskii A. 1985. Sov. Astron. Lett. 11:133
36
Stompor R, Balbi A, Borrill J, Ferreira P,
Hanany S, et al. 2001. Ap. J. In press (astro-
ph/0106451)
Strauss MA, Willick JA. 1995. Phys. Rep. 261:
271431
Sunyaev R, Zeldovich Y. 1972. Comments As-
trophys. Space Phys. 4:173
Szapudi I, Prunet S, Pogosyan D, Szalay AS,
Bond JR. 2001. Ap. J. Lett. 548:L115
18
Tegmark M. 1997a. Ap. J. Lett. 480:L8790
Tegmark M. 1997b. Phys. Rev. D 55:5895907
Tegmark M. 1998. Ap. J. 502:16
Tegmark M, Bunn EF. 1995. Ap. J. 455:16
Tegmark M, de Oliveira-Costa A. 2001. Phys.
Rev. D 64:063001
Tegmark M, Efstathiou G. 1996. MNRAS 281:
1297314
Tegmark M, Eisenstein DJ, Hu W, de Oliveira-
Costa A. 2000. Ap. J. 530:13365
Tegmark M, Hamilton AJS, Strauss MA, Voge-
ley MS, Szalay AS. 1998. Ap. J. 499:55576
Tegmark M, Taylor AN, Heavens AF. 1997. Ap.
J. 480:2235
Tegmark M, Zaldarriaga M. 2000. Ap. J. 544:
3042
Tegmark M, Zaldarriaga M, Hamilton AJS.
2001. Phys. Rev. D63:043007
Theureau G, Hanski M, EkholmT, Bottinelli L,
Goriguenheim U, et al. 1997. Astron. Astro-
phys. 322:730
Tuluie R, Laguna P. 1995. Ap. J. Let. 445:L73
76
Turok N. 1996. Phys. Rev. Lett. 77:413841
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
216 HU

DODELSON
Tytler D, OMeara JM, Suzuki N, Lubin D.
2000. Physica Scripta Vol. T 85:1231
Vishniac ET. 1987. Ap. J. 322:597604
Vittorio N, Silk J. 1984. Ap. J. Lett. 285:L3943
Wandelt B, Hansen F. 2001. Phys. Rev. D In
press (astro-ph/0106515)
Wandelt BD, Gorski KM. 2001. Phys. Rev. D
63:123002
Wang X, Tegmark M, Zaldarriaga M. 2001.
Phys. Rev. D In press (astro-ph/0105091)
Weinberg S. 1971. Ap. J. 168:17594
White M. 1998. Phys. Rev. D 57:527375
White M, Hu W. 1997. Astron. Astrophys. 321:
89
White M, Scott D, Silk J. 1994. Annu. Rev.
White SDM, Efstathiou G, Frenk CS. 1993.
MNRAS 262:102328
Wilson ML, Silk J. 1981. Ap. J. 243:1425
Winitzki S, Kosowsky A. 1997. New Astron.
3:75100
Wright EL, Hinshaw G, Bennett CL. 1996. Ap.
J. Lett. 458:L5355
Wu JHP, Balbi A, Borrill J, Ferreira PG, Hanany
S, et al. 2001. Ap. J. Suppl. 132:117
Zaldarriaga M. 2000. Phys. Rev. D 62:063510
Zaldarriaga M, Seljak U. 1997. Phys. Rev. D
55:183040
58:023003
59:123507
Zaldarriaga M, Spergel DN, Seljak U. 1997.
Ap. J. 488:113
Zeldovich Y, Kurt V, Sunyaev R. 1969. Sov.
Phys.JETP 28:146
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
2 Aug 2002 16:7 AR AR166-06-COLOR.tex AR166-06-COLOR.SGM LaTeX2e(2002/01/18) P1: GDL
Figure 1 (Top) Temperature anisotropy data with boxes representing 1- errors and
approximate l-bandwidth. (Bottom) Temperature and polarization spectra for
tot
=1,
=2/3,
b
h
2
=0.02,
m
h
2
=0.16, n =1, z
ri
=7, E
i
=2.2 10
16
GeV. Dashed lines
represent negative cross correlation and boxes represent the statistical errors of the
Planck satellite.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Figure 4 Sensitivity of the acoustic temperature spectrum to four fundamental cos-
mological parameters. (a) The curvature as quantied by
tot
. (b) The dark energy
as quantied by the cosmological constant
(w
=1). (c) The physical baryon

density
b
h
2
. (d) The physical matter density
m
h
2
. All are varied around a ducial
model of
tot
=1,
=0.65,
b
h
2
=0.02,
m
h
2
=0.147, n =1, z
ri
=0, E
i
=0.
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Figure 6 Integral approach. Cosmic Microwave Background anisotropies can be
thought of as the line-of-sight projection of various sources of plane wave temperature
andpolarizationuctuations: the acoustic effective temperature andvelocityor Doppler
effect (see 3.8), the quadrupole sources of polarization (see 3.7) and secondary
sources (see 4.2, 4.3). Secondary contributions differ in that the region over which
they contribute is thick compared with the last scattering surface at recombination and
the typical wavelength of a perturbation. A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
Figure 7 (Top) Polarization generation and classication. Thomson scattering of
quadrupole temperature anisotropies generates linear polarization. The component of
the polarization that is parallel or perpendicular to the wavevector k is called the E-
mode, and the one at 45
angles is called the B-mode. (Bottom) Secondary anisotropies.

(a) Gravitational secondaries: IntegratedSachs-Wolfe, lensing, andRees-Sciama (mov-
ing halo) effects. (b) Scattering secondaries: Doppler, density (), and ionization (i)
modulated Doppler, and the Sunyaev-Zeldovich effects. Curves and model are de-
scribed in the text. A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
P1: FRK
July 25, 2002 16:7 Annual Reviews AR166-FM
Annual Review of Astronomy and Astrophysics
Volume 40, 2002
CONTENTS
FRONTISPIECE, Edwin E. Salpeter xii
A GENERALIST LOOKS BACK, Edwin E. Salpeter 1
ULTRA-COMPACT HII REGIONS AND MASSIVE STAR FORMATION,
Ed Churchwell 27
KUIPER BELT OBJECTS: RELICS FROM THE ACCRETION DISK OF
THE SUN, Jane X. Luu and David C. Jewitt 63
THEORY OF GIANT PLANETS, W. B. Hubbard, A. Burrows,
and J. I. Lunine 103
THEORIES OF GAMMA-RAY BURSTS, P. M esz aros 137
COSMIC MICROWAVE BACKGROUND ANISOTROPIES, Wayne Hu
and Scott Dodelson 171
STELLAR RADIO ASTRONOMY: PROBING STELLAR ATMOSPHERES
FROM PROTOSTARS TO GIANTS, Manuel G udel 217
MODIFIED NEWTONIAN DYNAMICS AS AN ALTERNATIVE TO
DARK MATTER, Robert H. Sanders and Stacy S. McGaugh 263
CLUSTER MAGNETIC FIELDS, C. L. Carilli and G. B. Taylor 319
THE ORIGIN OF BINARY STARS, Joel E. Tohline 349
RADIO EMISSION FROM SUPERNOVAE AND GAMMA-RAY
BURSTERS, Kurt W. Weiler, Nino Panagia, Marcos J. Montes,
and Richard A. Sramek 387
SHAPES AND SHAPING OF PLANETARY NEBULAE, Bruce Balick
and Adam Frank 439
THE NEW GALAXY: SIGNATURES OF ITS FORMATION, Ken Freeman
and Joss Bland-Hawthorn 487
THE EVOLUTION OF X-RAY CLUSTERS OF GALAXIES, Piero Rosati,
Stefano Borgani, and Colin Norman 539
LYMAN-BREAK GALAXIES, Mauro Giavalisco 579
COSMOLOGY WITH THE SUNYAEV-ZELDOVICH EFFECT,
John E. Carlstrom, Gilbert P. Holder, and Erik D. Reese 643
vii
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.
P1: FRK
July 25, 2002 16:7 Annual Reviews AR166-FM
viii CONTENTS
INDEXES
Subject Index 681
Cumulative Index of Contributing Authors, Volumes 2940 709
Cumulative Index of Chapter Titles, Volumes 2940 712
ERRATA
An online log of corrections to Annual Review of Astronomy
and Astrophysics chapters (if any, 1997 to the present)
may be found at http://astro.annualreviews.org/errata.shtml
A
n
n
u
.

R
e
v
.

A
s
t
r
o
.

A
s
t
r
o
p
h
y
s
.

2
0
0
2
.
4
0
:
1
7
1
-
2
1
6
.

D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
n
n
u
a
l
r
e
v
i
e
w
s
.
o
r
g
b
y

U
n
i
v
e
r
s
i
t
y

o
f

D
u
r
h
a
m

o
n

0
7
/
2
6
/
1
2
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
.

Cosmic Microwave Background Anisotropies

Uploaded by

Copyright:

Available Formats

Cosmic Microwave Background Anisotropies

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cosmic Microwave Background Anisotropies

Uploaded by

Copyright:

Available Formats

23 Jul 2002 14:3 AR AR166-AA40-06.tex AR166-AA40-06.

SGM LaTeX2e(2002/01/18) P1: GJC

_1 eV. Massive neutrinos have an equation of state n

are of the utmost signicance

, has been written as a scalar instead of a

, where the photon pressure

,3. In the absence of gravity and viscous uid imperfections, pressure

,3 supply the only force. Because

,3 in Fourier space. The Euler equation then becomes

3 is the sound speed in the (dynamically baryon-free)

3 is the distance sound can travel by , usually called the

, where n is an integer (see Figure 2a).

). (a) Peak scales: The wavemode that

(k), where is a phase that can in principle be different

, the ratio of the

+ on the right-hand side.

as photons from, for example, neighboring temperature crests meet at a trough

,, corresponding to the geometric mean

10, leading to a substantial suppression of the oscillations beyond

decreases because the expansion rate

, it too decreases as the

are of the same order of magnitude at recombination, but

, (see Figure 5), the polarization spectrum

is out of phase with the monopole, the polarization peaks should

(positive and negative B).

is the distance light can

), we can expand the exponential in Equation

). Because the spherical

. Notice that this peak corresponds to con-

(x),dx. The structure of j

lacks a strong peak at x =. Physically this

) overestimates the dif-

. It has largely replaced the original ground breaking

, the particle hori-

, and the damping scale

sets the baryon loading

=0.60 0.06, and n =0.96 0.04 (Knox

. The damping scale

. This is called the angular diameter distance degeneracy in

, andn. More generally, the rst three canbe replacedby

to extend these results to models in which the underlying assumptions of

) and adds in phase with the monopole.

angles to it. Gravitational

(see Equation 3).

, the window function, is proportional to the Legendre polynomial

. For example, a Gaussian

F). In general, the curvature matrix depends

=1). (c) The physical baryon

angles is called the B-mode. (Bottom) Secondary anisotropies.

You might also like