The Astrophysical Journal, 664:660Y674, 2007 August 1 A

# 2007. The American Astronomical Society. All rights reserved. Printed in U.S.A.

ON THE ROBUSTNESS OF THE ACOUSTIC SCALE IN THE LOW-REDSHIFT CLUSTERING OF MATTER

Daniel J. Eisenstein,1,2 Hee-Jong Seo,1 and Martin White3
Received 2006 April 17; accepted 2006 May 9

ABSTRACT
We discuss the effects of nonlinear structure formation on the signature of acoustic oscillations in the late-time
galaxy distribution. We argue that the dominant nonlinear effect is the differential motion of pairs of tracers separated
by 150 Mpc. These motions are driven by bulk flows and cluster formation and are much smaller than the acoustic
scale itself. We present a model for the nonlinear evolution based on the distribution of pairwise Lagrangian dis-
placements that provides a quantitative model for the degradation of the acoustic signature, even for biased tracers in
redshift space. The Lagrangian displacement distribution can be calibrated with a significantly smaller set of simula-
tions than would be needed to construct a precise power spectrum. By connecting the acoustic signature in the Fourier
basis with that in the configuration basis, we show that the acoustic signature is more robust than the usual Fourier-
space intuition would suggest, because the beat frequency between the peaks and troughs of the acoustic oscillations
is a very small wavenumber that is well inside the linear regime. We argue that any possible shift of the acoustic scale
is related to infall on a scale of 150 Mpc, which is O(0.5%) fractionally at first order, even at z = 0. For the matter,
there is a first-order cancellation such that the mean shift is O(10
4). However, galaxy bias can circumvent this can-
cellation and produce a subpercent systematic bias.
Subject headingg s: cosmic microwave background — cosmological parameters — distance scale —
large-scale structure of universe
Online material: color figures

1. INTRODUCTION relations. In particular, simulations have shown that nonlinear

The imprint in the late-time clustering of matter from the baryon structure formation and, to a lesser extent, redshift distortions
acoustic oscillations in the early universe (Peebles & Yu 1970; erase the higher harmonics of the acoustic oscillations ( Meiksin
Sunyaev & Zel’dovich 1970; Doroshkevich et al. 1978) has et al. 1999; Seo & Eisenstein 2005; Springel et al. 2005; White
emerged as an enticing way to measure the distance scale and ex- 2005). This degrades the measurement of the acoustic scale. More
pansion history of the universe ( Eisenstein et al. 1998; Cooray worrisome is the possibility that these effects might actually bias
et al. 2001; Eisenstein 2002; Blake & Glazebrook 2003; Hu & the measurement.
Haiman 2003; Seo & Eisenstein 2003; Linder 2003; Matsubara In this paper, we present a physical explanation of these non-
2004; Amendola et al. 2005; Blake & Bridle 2005; Glazebrook linear distortions. We use a hybrid of analytic and numerical meth-
& Blake 2005; Dolney et al. 2006). The distance that acoustic ods to produce a quantitative model for the degradation. We also
waves can propagate in the first million years of the universe be- investigate whether the nonlinearities can measurably shift the
comes a characteristic scale, measurable not only in the cosmic acoustic scale. The results offer the opportunity to calibrate the ef-
microwave background (CMB) anisotropies (Miller et al. 1999; fects of nonlinearities with simulations far smaller than what would
de Bernardis et al. 2000; Hanany et al. 2000; Halverson et al. be required to see the effects directly in the acoustic signature.
2002; Netterfield et al. 2002; Bennett et al. 2003) but also in the We organize the paper as follows: In x 2, we present a pedagog-
late-time clustering of galaxies (Cole et al. 2005; Eisenstein et al. ical review of the acoustic phenomenon, explaining the effects in
2005; Hütsi 2006). This scale can be computed with simple lin- both the configuration and Fourier bases. In x 3, we discuss how
ear perturbation theory once one specifies the baryon-to-photon nonlinearities enter the process and why the large preferred scale
ratio and matter-radiation ratio, both of which can be measured of the acoustic oscillations provides an important simplification of
from the details of the CMB acoustic peaks (Bennett et al. 1997; the dynamics. In x 4, we present a quantitative framework for un-
Hu et al. 1997; Hu & Dodelson 2002; White & Cohn 2002). With derstanding the nonlinear effects based on Lagrangian displace-
this scale in hand, one can measure the angular diameter distance ments. We construct Zel’dovich-approximation estimates of the
and Hubble parameter as functions of redshift using large galaxy displacements in x 5 and then measure the required distributions
redshift surveys. This standard ruler offers a robust route to the numerically and compare them with the nonlinear two-point func-
study of dark energy. tions in x 6. The model is extended to biased tracers in x 7. In x 8,
However, the clustering of galaxies does not exactly recover we study how nonlinearities could create systematic shifts in the
the linear-theory clustering pattern. Nonlinear gravitational col- acoustic scale. We conclude in x 9.
lapse, galaxy clustering bias, and redshift distortions all modify Throughout the paper, we will refer to redshift space as the
galaxy clustering relative to that of the linear-regime matter cor- position of galaxies or matter inferred from redshifts, with no cor-
rection for peculiar velocities. The true position of the galaxies
or matter will be called real space. Sometimes, the term real space
1 is used in cosmology as the complement to the Fourier basis; we
Steward Observatory, University of Arizona, Tucson, AZ 85121.
2
Alfred P. Sloan Fellow. will instead refer to these as Fourier and configuration space (or
3
Departments of Physics and Astronomy, University of California, Berkeley, basis) to avoid confusion. We use comoving distance units in all
CA 94720. cases.
660
 BARYON ACOUSTIC OSCILLATIONS. I. 661

2. ACOUSTIC OSCILLATIONS IN CONFIGURATION nation slightly alters the damping from this exponential form (Hu
AND FOURIER SPACE & White 1997), but this detail is not of relevance to us here. Note
that since decoupling occurs slightly after recombination, both the
This section is pedagogical in nature. The physics of the acous-
sound horizon and the damping scale are slightly larger than the
tic oscillations have been well studied in Fourier space, but we
have found that the story in configuration space is rarely discussed corresponding scales in the CMB.
Once the photons release their hold on the baryons, the latter
even though it exhibits some of the most important features for our
can be treated as pressureless, allowing the perturbations to grow
purposes. Moreover, this pedagogy can warm up the reader to the
configuration-space view of the nonlinearities that we pursue in by gravitational instability (e.g., proportionally to the scale fac-
tor in a matter-dominated universe). The density and velocity
the rest of the paper.
perturbations from the tight-coupling era, including the CDM
The epoch of recombination marks the time when the universe
perturbations, must be matched onto the growing and decaying
has cooled enough that protons and electrons combine to form hy-
modes for the pressureless components. The final spectrum is the
drogen (Peebles 1968; Zel’dovich et al. 1968; Seager et al. 1999).
component that projects onto the growing mode. As is well known
Prior to this time, the mean free path of the photons against scat-
(Sunyaev & Zel’dovich 1970; Press & Vishniac 1980; for a text-
tering off the free electrons is much less than the Hubble distance.
book treatment, see x 8.2 of Padmanabhan 1993), at high k the
This means that gravitational forces attempting to compress the
growing mode is created primarily by the velocities, while at low
plasma must also increase the photon density. This produces an
k it is a mixture of density and velocity terms. For this reason,
increase in temperature and hence in radiation pressure. Any per-
turbation in the baryon-photon plasma thus behaves as an acoustic at high k the oscillations in the mass power spectrum are out of
phase with the peaks in the CMB anisotropy spectrum, which
wave.
arise predominantly from photon densities. In this limit, extrema
occur at ks = (2j þ 1)/2, where Ðj = 0, 1, 2, . . . and s is the
2.1. Fourier Space
sound horizon at decoupling: s = cs d. In detail, this differs
Before looking at this picture in configuration space, let us from the sound horizon at recombination (which controls the po-
remind ourselves of the physics in Fourier space. Consider a sition of the CMB peaks) but the two horizons are comparable.
(standing) plane-wave perturbation of comoving wavenumber k The amplitude of the oscillations depends both on the driving force
in Newtonian gauge (e.g., Kodama & Sasaki 1984). At early times, ( and
) and on R. Since the potentials decay in the radiation-
the density is high and the scattering is rapid compared with the dominated epoch, larger oscillations come from higher b h 2 and
travel time across a wavelength. We may therefore expand the mo- lower m h 2 (Eisenstein & Hu 1998; Meiksin et al. 1999).
mentum conservation (Euler) equation in powers of the Compton
mean free path over the wavelength k/
, ˙ where
˙ is the differ- 2.2. Configuration Space
ential Compton optical depth andÐ overdots denote derivatives With this in mind, it is instructive to switch to configuration
with respect to conformal time  = dt/a. To lowest order and set- space and consider what happens to a pointlike initial overden-
ting c = 1, we obtain the tight coupling approximation for the evo- sity.4 In an adiabatic model, the overdensity is present in all spe-
lution of a single Fourier mode of the baryon density perturbation: cies (Peebles & Yu 1970). In particular, because the region is
overdense in photons, it is overpressured relative to its surround-
d k2 d ings. This overpressure must equilibrate by driving a spherical
½(1 þ R)˙b þ b ¼
k 2 (1 þ R)

˙
½3(1 þ R) ð1Þ
d 3 d sound wave out into the baryon-photon plasma. The CDM per-
turbation is left at small radius. The wave travels out at the speed
( Peebles & Yu 1970; Doroshkevich et al. 1978; for a more re- of sound. At the time of decoupling, the wave stalls as the pressure-
cent derivation in the relevant gauge and notation, see Hu & supplying photons escape and the sound speed plummets. One
White 1996). Here R  3b /4 is the baryon-to-photon momen- ends up with a CDM overdensity at the center and a baryon over-
tum density ratio and the gravitational sources are , the pertur- density in a spherical shell 150 comoving megaparsecs in radius
bation to the spatial curvature, and
, the Newtonian potential. for the concordance cosmology. At z T103, both of these over-
At late times, the gravitational perturbations are dominated by the densities attract gas and CDM to them, seeding the usual gravi-
cold dark matter (CDM) and baryons (if b /m is large), while at tational instability (Peebles & Yu 1970). The fraction of gas to
early times they are dominated by the relativistic species. We re- CDM approaches the cosmic average as the perturbation grows by
cognize this equation as that of a driven oscillator with natural fre- a factor approaching 103, so that one ends up with an overdensity
quency cs k, where the speed of sound cs is c/[3(1 þ R)]1/2. During of all matter at the center and a spherical echo at 150 Mpc radius.
the tight coupling phase, the amplitude of the baryon perturbations Galaxies are more likely to form in these overdensities. The radius
cannot grow and instead undergoes harmonic motion with an am- of the sphere marks a preferred separation of galaxies, which we
plitude that decays as (1 þ R)
1/4 and a velocity that decays as quantify as a peak in the correlation function on this scale. This
(1 þ R)
3/4 (Hu & Sugiyama 1995; Hu & White 1996). For cur- evolution is shown graphically in Figure 1.
rently accepted values of b (Steigman 2006; Spergel et al. 2007) The universe is of course a superposition of these pointlike
R < 0.3 at z > 1000, so the decay of the period
R today is small. perturbations, but as the perturbation theory is exquisitely linear
If we define the optical depth
b ()   ˙ þ R), we
d
/(1 at high redshift, we can simply add the solutions. The width of
find that the baryons decouple from the photons when
b  1, a the acoustic peak is set by three factors: Silk damping due to pho-
time we refer to as ‘‘decoupling.’’ The oscillations in the baryons tons leaking out of the sound wave (Silk 1968), adiabatic broad-
are frozen-in at this epoch, which occurs after recombination for ening of the wave as the sound speed changes because of the
currently accepted values of b . Expanding to higher order in k/
, ˙ increasing inertia of the baryons relative to the photons, and the
one finds the oscillations are exponentially damped as a result of correlations of the initial perturbations. In practice, the acoustic
photon-baryon diffusion (Silk 1968) with a characteristic scale
that is approximately the geometric mean of the horizon and the 4
The development of the transfer function in configuration space was de-
mean free path at decoupling (on the order of 10 Mpc for a con- scribed in detail by Bashinsky & Bertschinger (2001, 2002), although that work
cordance cosmology). Including the finite duration of recombi- was based on a planar initial overdensity rather than a pointlike one.
 Fig. 1.—Snapshots of evolution of the radial mass profile vs. comoving radius of an initially pointlike overdensity located at the origin. All perturbations are fractional
for that species; moreover, the relativistic species have had their energy density perturbation divided by 4/3 to put them on the same scale. The black, blue, red, and green
lines are the dark matter, baryons, photons, and neutrinos, respectively. The redshift and time after the big bang are given in each panel. The units of the mass pro-
file are arbitrary but are correctly scaled between the panels for the synchronous gauge. Top left: Near the initial time, the photons and baryons travel outward as a pulse.
Top right: Approaching recombination, one can see the wake in the cold dark matter raised by the outward-going pulse of baryons and relativistic species. Middle left: At
recombination, the photons leak away from the baryonic perturbation. Middle right: With recombination complete, we are left with a CDM perturbation toward the center
and a baryonic perturbation in a shell. Bottom left: Gravitational instability now takes over, and new baryons and dark matter are attracted to the overdensities. Bottom right:
At late times, the baryonic fraction of the perturbation is near the cosmic value, because all of the new material was at the cosmic mean. These figures were made by suitable
transforms of the transfer functions created by CMBFAST (Seljak & Zaldarriaga 1996; Zaldarriaga & Seljak 2000).
 BARYON ACOUSTIC OSCILLATIONS. I. 663

peak is about 30 Mpc full width at half-maximum for the con- pling determines the positions of the peaks in the power spectrum,
cordance cosmology. but traveling waves. We imagine that each overdense crest of
There are some other interesting aspects of the physics of this the initial perturbation launches a planar sound wave that travels
epoch that are worth mentioning in the configuration-space pic- out for 150 Mpc (Bashinsky & Bertschinger 2001, 2002). If this
ture. First is that the outgoing wave does not actually stop at z  places the baryon perturbation on top of another crest in the initial
103 but instead slows around z  500. This is partially due to the wave (now marked by the CDM), then one will obtain construc-
fact that decoupling is not coincident with recombination but is tive interference in the density field and a peak in the power spec-
also because the coupling to the growing mode is actually dom- trum. If the sound wave lands in the trough of the initial wave, then
inated by the velocity field, rather than the density field, at z  a destructive interference is produced. Hence, one finds a harmonic
103 (Sunyaev & Zel’dovich 1970; Press & Vishniac 1980). In pattern of oscillations, with a damping set by the width of the post-
other words, the compressing velocity field in front of the wave recombination baryon and CDM perturbations from a thin initial
actually keys the instability at a later time. plane (i.e., the one-dimensional Green’s function).
Two other aspects of Figure 1 that may be surprising at first In our opinion, for understanding the late-time matter spectrum,
glance are that the outgoing pulse of neutrino overdensity does in particular the standard-ruler aspect, the traveling-wave view has
not actually remain as a delta function, as one might expect for a considerable pedagogical advantages. The fact that the baryon
population traveling radially outward at the speed of light, and signature in the correlation function has a single peak is a provoc-
that the CDM perturbation does not remain at the origin, as one ative indication that the physics is easily viewed as the causal
would expect for a cold species. Both of these effects are due to a propagation of a signal. Mathematically, the connection is that the
hidden assumption in the initial conditions: although the density correlation function and power spectrum are Fourier-transform
field is homogeneous everywhere but the origin, the velocity field pairs. The transform of a single peak is a harmonic sequence, and
cannot be for a growing mode. To keep the bulk of the universe the profile of the peak gives the damping envelope of the oscil-
homogeneous while growing a perturbation at the origin, matter lations. We shall draw upon this picture as we consider nonlinear
must be accelerating toward the center; this acceleration is sup- evolution in the following sections.
plied by the gravitational force from the central overdensity. How-
3. NONLINEAR DEGRADATION
ever, in the radiation-dominated epoch the outward-going pulse of
OF THE ACOUSTIC SIGNATURE
neutrinos and photons is carrying away most of the energy density
of the central perturbation. This outward-going pulse decreases At low redshift, nonlinear structure formation, redshift distor-
the acceleration, causing the inward flow of the homogeneous tions, and galaxy clustering bias all act to degrade the acoustic
bulk to deviate from the divergenceless flow and generating the signature in the galaxy power spectrum (Meiksin et al. 1999; Seo
behavior of the CDM and neutrinos mentioned above. Essen- & Eisenstein 2005; Springel et al. 2005; White 2005). These ef-
tially, the outgoing shells of neutrinos and photons raise a wake fects reduce the amount of information about the distance scale
in the homogeneous distribution of CDM away from the origin that can be extracted from a survey of a given volume (Seo &
of the perturbation. Eisenstein 2003; Blake et al. 2006). In principle, they might also
The smoothing of the CDM overdensity from a delta function shift the scale, resulting in biased information from uncorrected
at the origin is the famous small-scale damping of the CDM power estimators. One should of course distinguish such shifts from a
spectrum in the radiation-dominated epoch (Lifshitz 1946; Peebles simple increase in noise; a noisy statistical estimator is not the
& Yu 1970; Sunyaev & Zel’dovich 1970; Groth & Peebles 1975; same as a biased one.
Doroshkevich et al. 1978; Wilson & Silk 1981; Peebles 1981; How do the breakdowns of the ideal linear case enter? We be-
Blumenthal et al. 1984; Bond & Efstathiou 1984). The overden- gin in Fourier space. Here the acoustic signature is a damped har-
sity raised decreases as a function of radius because the radiation monic series of peaks and troughs. As nonlinearities progress,
is decreasing in energy density relative to the inertia of the CDM; mode-mode coupling begins to alter the power spectrum. Power
in the matter-dominated regime, the outward-going radiation has increases sharply at large wavenumbers, but the mode-mode cou-
no further effect. A universe with more radiation causes a larger pling also damps out the initial power spectrum on intermediate
effect that extends to larger radii; this corresponds to the shift in scales, decreasing the acoustic oscillations (Goroff et al. 1986;
the CDM power spectrum with the matter-to-radiation ratio. Jain & Bertschinger 1994; Meiksin et al. 1999; Jeong & Komatsu
Returning to the major conceptual point, that of the shell of 2006). Both of these effects reduce the contrast of the oscillations,
overdensity left at the sound horizon, we see immediately that making them harder to measure.
the sound horizon provides a standard ruler. The radius of the shell At z < 1, N-body simulations and galaxy catalogs derived
depends simply on the sound speed and the amount of propa- therefrom show an excess of power at the location of the higher
gation time (Peebles & Yu 1970; Sunyaev & Zel’dovich 1970; acoustic harmonics (k  0.2 h Mpc
1). Clearly, the nonlinear
Doroshkevich et al. 1978). The sound speed is set by the balance power spectrum deviates from the linear one at wavenumbers
of radiation pressure and inertia from the baryons; this is con- containing acoustic information. Does this mean that one cannot
trolled simply by the baryon-to-photon ratio, which is b h 2 . The recover ‘‘linear’’ information from these scales? As we demon-
propagation time depends on the expansion rate in the matter- strate next, the answer is not this pessimistic.
dominated and radiation-dominated regimes; this in turn depends In configuration space, the acoustic signature is a single peak
on the redshift of matter-radiation equality, which depends only in the correlation function at 150 Mpc separation. The width of the
on m h 2 for the standard assumption of the radiation density (i.e., peak is related to the decreasing envelope of the oscillations in the
the standard cosmic neutrino and photon backgrounds and noth- power spectrum. The action of the nonlinear effects is easy to see:
ing else). velocity flows and nonlinear collapse move matter in the universe
around by 10 Mpc relative to its initial comoving position. This
2.3. Connecting Fourier and Configuration Space
tends to move pairs out of the 150 Mpc peak, broadening it. This
Next it is useful to connect this picture to the Fourier-space broadening corresponds to the decay of the high harmonics in
view of the evolution of plane waves. Here we consider not stand- P(k). One can also see that because redshift distortions further in-
ing waves, for which the accumulated phase change until decou- crease the distance between a galaxy’s measured position and its
664 EISENSTEIN, SEO, & WHITE Vol. 664

initial position, the acoustic oscillations are further reduced in red- nonlinear alterations at k = 0.1Y 0.2 h Mpc
1 does not mean that
shift space. the acoustic effects are hopelessly muddied by nonlinearities.
The key insight from the configuration-space picture is that the
processes that are erasing the acoustic signature are not at the fun- 4. MODELING THE DEGRADATION
damental scale of 150 Mpc but are at the cluster formation scale, a AS A LAGRANGIAN DISPLACEMENT
factor of 10 smaller. This is very different from the apparent behav- We have argued that the erasure of the acoustic signature can
ior in Fourier space that suggests the nonlinearities are encroach- be understood as being due to the motions of matter and galaxies
ing on the acoustic scale. The solution to the paradox is to note that relative to the initial preferred separation. It is therefore inter-
the acoustic information in Fourier space is actually carried by the esting to think about the displacement of matter or tracers from
comparison of adjacent peaks and troughs. The beat frequency their positions in the near-homogeneous initial state. This is the
between these nearby wavenumbers is much smaller, about k = Lagrangian displacement, well known from the Zel’dovich ap-
0.03 h Mpc
1, and is well below the nonlinear wavenumber. In proximation (Zel’dovich 1970).
other language, the power that is being built by mode-mode cou- We write the Lagrangian displacement as a vector u(r), where
pling must be smooth on the scale of k because the nonlin- r is the comoving coordinate system for the homogeneous cos-
earities are developing on much smaller scales.5 The existence of mology. One can then write the correlation function at a par-
this small beat frequency changes the behavior of nonlineari- ticular (large) separation as
ties relative to our usual intuition, which is based on broadband
Z
effects.
It is also interesting to note that the power that is being intro- (s) ¼ dr12 du12 d1 d2
duced by halo formation on small scales is particularly smooth at
the wavenumbers of interest (Schulz & White 2006). In the cor- p(1; 2 )1 2 p(u12 j r12 ; 1; 2 )(3) (r12 þ u12
s); ð2Þ
relation function, this nonlinearity adds very little to  at the acous-
tic scale; it is purely an excess correlation on halo scales. This where the subscripts 1 and 2 indicate the density values at two
reinforces the idea that such power can be removed by marginal- different points, r12 = r1
r2, and likewise for u12. The proba-
izing over smooth broadband spectra without biasing the recov- bility distribution p(1 , 2 ) is the distribution of the initial density
ery of the acoustic scale. The important effect of the small-scale field; on large scales, there is a small correlation (r12) between the
nonlinearity is the increase in noise. This is clear in the power densities. The probability distribution p(u12 | r12 , 1, 2 ) is the dis-
spectrum, where the noise, even in the Gaussian limit, is increased tribution for the Lagrangian displacements. If one were consider-
by the extra near-white noise at small wavenumbers. It is less ob- ing local clustering bias, then one could replace the product 1 2
vious in the correlation function, as (r) itself has not changed on by a stochastic bias g;1 g; 2 p(g;1 | 1 )p(g; 2 | 2 ).
the acoustic scale. However, the noise will be present in the co- In general, this integral is similar to the general redshift distor-
variance matrix of the correlation function, as it acts as extra shot tion problem ( Hamilton 1998; Scoccimarro 2004; Matsubara
noise that increases the variance in the correlation function. et al. 2004) and does not have an easy solution. In practice, given
As for the acoustic scale itself, in configuration space it is easy how large the desired scale is compared with the scale of clus-
to see that it will not be shifted much at all. The scale of 150 Mpc ter formation and likely galaxy physics, we consider a peak-
is far larger than any known nonlinear effect in cosmology. To background split. For this, we write the density field as a sum
shift the peak, we need to imagine effects that would systemat- of a large-amplitude small-scale density field  S and a small-
ically move pairs at 150 Mpc to smaller or larger separations. We amplitude large-scale density field  L . These fields are to be
will return to this point in detail in x 8 and argue that nonlinear thought of as statically independent. The small-scale fields at the
gravitational physics can only introduce shifts of 0.5% even at two points will be statistically independent. We do similarly for
first order and that linearly biased tracers imply a cancellation of the Lagrangian displacement field. This yields
the first-order term. Z
Hence, the picture of the nonlinear evolution of the acoustic (s) ¼ dr12 du1S du2S du12 L
d1S d2S d1L d2L
signature is easy to see in configuration space: the characteris-
tic separation of 150 Mpc is not shifted but rather is blurred by ½ p(1L ; 2L )p(1S )p(2S )(1S þ 1L )(2S þ 2L )
the action of cluster formation and bulk flows that move matter
; p(u1S j 1S )p(u2S j 2S )p(u12
L
j r12 ; 1L ; 2L )
around by 10 Mpc. This smears out the peak in the correlation
function, further damping out the higher harmonics in the power ;  (3) (r12 þ u12
L
þ u1S
u2S
s): ð3Þ
spectrum. The acoustic scale becomes harder to measure in a
survey because the wider correlation peak is harder to centroid or Only the 1L 2L cross term yields a nonzero result.
because the higher harmonics are at lower contrast in the power If we treat the large-scale Lagrangian displacement as a delta
spectrum. The existence of collapsed halos adds extra near-white L
function at u12 = 0 for the moment, then this integral reduces sim-
noise in the power spectrum and creates increased variance in ply to
measurements of the correlation function. Z Z
Again, there is nothing in the configuration-space version of (s) ¼ dr12 (r12 ) d1S du1S p(1S )p(u1S j 1S )
the story that cannot be stated in the Fourier-space version, but
Z
one must remember that the small beat frequency between the
peaks and troughs of the acoustic oscillations means that one’s d2S du2S p(2S )p(u2S j 2S ) (3) (r12 þ u1S
u2S
s): ð4Þ
usual intuition about broadband Fourier effects does not fully ap-
ply. In particular, the fact that the power spectrum does show In other words, this is simply the correlation function convolved
twice with the distribution p(u S ) of small-scale Lagrangian dis-
5
The power that is being removed by mode-mode coupling can have struc- placements. This function can depend on the galaxy bias model.
ture on the beat-frequency scale because a simple damping does bring out one In the case of the redshift-space correlation function, one can sim-
power of the initial power spectrum (e.g., Crocce & Scoccimarro 2006). ply treat p(u S ) as the difference between the initial position (which
No. 2, 2007 BARYON ACOUSTIC OSCILLATIONS. I. 665

is the same in real and redshift space) and the final position in red- In a simulation, this reduces to the distribution of u1 for galaxy
shift space; the function obviously becomes anisotropic in this ‘‘particles.’’ Again, we suggest simply computing u1
u2 for
case. When computing the small-scale displacement field uS, one pairs of galaxies separated by r12. Of course, the initial position
should high-pass filter to remove bulk flows on scales at or larger of a galaxy is ill defined in a simulation, but as we are interested
than |s|. This correction is important, because bulk flows are gen- in displacements similar in size to the width of the acoustic peak,
erated on quite large scales in CDM cosmologies. we expect that any reasonable measure of the initial position of the
Of course, the double convolution of the correlation function center of mass of the galaxy will agree to better than 1 Mpc, which
is a trivial multiplication in the power spectrum. Either way, one is negligible in the scatter of the displacements. It is important to
sees that the small-scale power has been strongly suppressed. One note that the distribution of the displacements of the galaxies may
should remember that this procedure only represents the memory be different from that of the mass displacement, particularly in
of the initial correlation pattern; it does not introduce the small- redshift space.
scale nonlinear clustering that builds up the small-scale correlation The picture of a peak and a spherical echo might lead one to
function or power spectrum. In real space, our method is an ap- hope that one could pick tracers that would preferentially accen-
proximation to the nonlinear propagator of Crocce & Scoccimarro tuate the peak or sphere, therefore boosting the height of the peak
(2006). in the correlation function. We show explicitly in the Appendix
We do not have a full solution to the case with an arbitrary that this is not possible for local bias, at least within the Gaussian
large-scale Lagrangian displacement field. However, we find a random field assumption; the large-scale correlation function of
significant simplification in equation (2) or (3) if we assume that galaxies always tracks that of the matter, as known from theorems
L
the distribution of u12 or u12 is independent of 1 and 2 . This is about local bias (Coles 1993; Scherrer & Weinberg 1998).
not true in detail: overdense regions will tend to move toward
one another, underdense regions away. The mean u12 depends on 5. ANALYTIC ESTIMATES
1L þ 2L at linear order, but as shown in x 8, the amount of this
shift is much less than the 30 Mpc width of the acoustic peak. It is In the Zel’dovich approximation, our interest in the Lagrangian
also much less than the standard deviation of the u12 distribution displacements translates into a study of correlations in the initial
computed in x 5 or measured in x 6. Hence, because the shifts are velocity field at two points. The displacement is
small and because the density dependence produces opposing Z
dk k ik = r1
shifts between the overdense and underdense regions, it is a good u12 ¼ 
3 k ik 2
(e
e ik = r2 ): ð8Þ
approximation to ignore the dependence when computing the (2)
broadening of the acoustic peak. This would of course not be a
good approximation for determining whether there is a mean shift This quantity has a mean of zero, but of course the variance is
in the scale; we will return to this topic in x 8. nonzero. The variance of the displacement along the direction
As both the large- and small-scale terms have boiled down to between the two points is
the distribution of displacements unconditioned by the density
field, we suggest that one can handle both of the effects simul- Z
2
dk k = r12
taneously and also do the proper filtering of the large-scale bulk
2
hu12;k i ¼ P(k) je ik = r12
1j 2
flows by computing the distribution of u12 for pairs of particles (2) 3 k 2 r12
Z 2
separated by a given r12, without regard for the dependence on 1 2 k dk
¼ r12 P(k) fk (kr12 ); ð9Þ
and 2 . Note that this avoids computing a peak-background split. 2 2
The resulting distribution p(u12 | r12) is our model for the kernel
that the correlation function will be convolved with, following where
equation (2). We will measure this kernel numerically in x 6 and


show that it reproduces the nonlinear degradation of the acoustic 2 1 sin x 2 cos x 2 sin x
signatures in simulated power spectra. fk (x) ¼ 2

þ : ð10Þ
x 3 x x2 x3
If we consider a local clustering bias (Coles 1993; Scherrer &
Weinberg 1998; Dekel & Lahav 1999), then we can write
Note that fjj ! 1/5
x 2 /84 as x ! 0, so the result behaves sim-
Z ilarly as k ! 0 to the integral for computing the density fluctua-
(s) ¼ dr12 du12 d1 d2 p(1 ; 2 )dg;1 dg;2 tions, rather than the k
2 behavior of the rms bulk flow integrand.
This is because very large wavelength modes move both particles
½ p(g;1 j 1 )p(g;2 j 2 )g;1 g;2 p(u12 j r12 ; 1 ; 2 ) together. The variance of the displacement in a direction orthog-
; (3) (r12 þ u12
s); ð5Þ onal to r12 is the same formula, but with


2 1 sin x cos x
where u refers to the displacements of the galaxies. In the peak- f? (x) ¼ 2
3 þ 2 ; ð11Þ
S x 3 x x
background split, the g;1 factor becomes g;1 þ b1L , where b is
the large-scale bias. In equation (4), the terms of the form
which behaves as 1/15
x 2 /420 as x ! 0.
Z For a concordance CDM cosmology with 8 = 0.85 (i.e., z  0),
d1S du1S p(1S )p(u1S j 1S ) ð6Þ we find rms displacements of 9 h
1 Mpc along the separation di-
rection and 8 h
1 Mpc transverse for separations of 100 h
1 Mpc.
These displacements are mostly generated on large scales, with
become
50% of the variance at k < 0.05 h Mpc
1 for radial displacements
Z and k < 0.08 h Mpc
1 for transverse. Rather little of the variance
b d1S dg;1 duS1 p(1S )p(g;1
S
j 1S )p(uS1 j 1S ): ð7Þ is generated at k < 0.02 h Mpc
1; wavelengths much longer than
100 h
1 Mpc move both points similarly. The displacements scale
666 EISENSTEIN, SEO, & WHITE Vol. 664

Fig. 2.—Distribution of pairwise Lagrangian displacements for particles initially separated by 100 h
1 Mpc. Left, real space; right, redshift space. Both are at redshift
z = 0.3. The plot is shown as the logarithm of probability vs. the square of the separation, so that a Gaussian distribution would be a straight line. The distribution in the
radial direction is slightly skew; we fold the distribution at zero and show the infalling and outflowing distribution as separate lines. In real space, the distribution is nearly
Gaussian; in redshift space, it is slightly cuspier. The displacement in the radial direction has slightly more variance than that in the tangential direction. For the redshift-
space plot, the displacement is always in the direction along the line of sight. The displacements in the direction across the line of sight are of course identical to those in real
space.

linearly with 8 (of the matter) and hence are smaller at high Figure 3 shows the rms displacement as a function of redshift
redshift. for 100 h
1 Mpc scale. The results are well modeled as increas-
In redshift space, the Lagrangian displacement in the Zel’dovich ing linearly with the growth function; the residuals are below 1%
approximation is simply increased by a factor of f = d( ln D)/ fractional. This is not surprising, because so much of the displace-
d( ln a)  0:56
m along the line of sight, because the same veloc- ment in the CDM cosmology is generated by flows that originate
ity that gives the real-space displacement also alters the posi- at k P 0.1 h Mpc
1, where the growth is close to linear theory.
tion in redshift space ( Peebles 1980; Kaiser 1987).
TABLE 1
6. NUMERICAL RESULTS Lagrangian Displacements Distribution for Matter
FOR THE MATTER DISTRIBUTION
We have measured the Lagrangian displacement of pairs Radial Transverse
of particles from the set of N-body simulations presented in z R  Skewness Kurtosis  Kurtosis
Seo & Eisenstein (2005). In brief, we use 30 simulations, each
256 3 particles in a 512 h
1 Mpc periodic cube. The cosmology Real Space
is m = 0.27, h = 0.72,  = 0.73, b = 0.046, n = 0.99, and
0.3.................. 50 7.59
0.092 0.102 6.12 0.053
8 = 0.9. We select pairs of particles whose initial separation falls
80 8.02
0.049 0.060 6.82 0.036
in a given bin, for example, narrowly centered on 100 h
1 Mpc, 100 8.15
0.034 0.049 7.10 0.029
and compute the difference of their final separation and initial 120 8.24
0.023 0.043 7.30 0.023
separation. We project this vector into components parallel and 150 8.33
0.013 0.041 7.51 0.018
perpendicular to the initial separation vector and refer to these as 0.4.................. 100 7.76
0.032 0.047 6.77 0.026
the radial and transverse displacements. 0.5.................. 100 7.40
0.031 0.045 6.46 0.022
In real space, Figure 2 shows the distribution of radial and 0.7.................. 100 6.74
0.028 0.040 5.89 0.016
transverse displacements. Both distributions have zero mean, as 1.0.................. 100 5.90
0.024 0.035 5.17 0.005
required by homogeneity, and only mild skewness and kurtosis. 1.5.................. 100 4.84
0.020 0.026 4.24
0.012
In other words, the distributions are close to Gaussian, with the 3.0.................. 100 3.07
0.011 0.010 2.70
0.053
radial displacements having slightly larger variance. The skew- Redshift Space
ness is such that large inward displacements are more likely than
large outward displacements. On small scales, many pairs of par- 0.3.................. 100 13.59
0.063 0.19 11.96 0.23
ticles fall into the same halo, so that u12 =
r12, and there is nec- 0.4.................. 100 13.30
0.061 0.18 11.72 0.23
essarily a positive tail that keeps the mean zero. In other words, 0.5.................. 100 12.98
0.059 0.18 11.45 0.22
the distribution becomes very skew. On 100 h
1 Mpc scales, 0.7.................. 100 12.26
0.056 0.17 10.82 0.21
1.0.................. 100 11.13
0.051 0.15 9.83 0.18
it appears that treating the distribution as Gaussian is a good
1.5.................. 100 9.41
0.044 0.11 8.30 0.14
approximation. 3.0.................. 100 6.10
0.031 0.05 5.35 0.05
The variance of the displacements is hence the relevant statis-
tic to quantify the distribution. We tabulate this in Table 1 at vari- Note.—Separations R and rms displacements are given in comoving h
1 Mpc.
ous redshifts and separations. The variance is a slow function of Skewness and kurtosis statistics are the usual dimensionless normalization. Neg-
scale: the radial standard deviations at 50, 100, and 150 h
1 Mpc ative skewness means that the heavier tail is inward. The mean displacements are all
consistent with round-off error, less than 10
4R. For redshift space, we list the dis-
are 7.59, 8.15, and 8.33 h
1 Mpc, respectively. We will therefore placements in the line-of-sight direction when this direction is radial and transverse
focus on 100 h
1 Mpc separation as a representative value, broadly to the initial separation vector. The displacements in the direction perpendicular to
applicable around the scales of interest. the line of sight are the same as in real space.
No. 2, 2007 BARYON ACOUSTIC OSCILLATIONS. I. 667

radial displacement rms is modulated roughly as the square of

the cosine of the angle relative to the line of sight. In the direction
along the line of sight, the boost in the standard deviation is very
close to the value f predicted by linear theory. We show this fit in
Figure 3 as well. The fit appears low by about 1% fractionally,
with about 1% residual around that correction.
In sum, the radial displacements across and along the line of
sight are well predicted by the following simple model:

? ¼ s0 D; k ¼ s0 D(1 þ f ); ð12Þ

where D is the growth function and f = d(ln D)/d(ln a)  0:56 .

The length s0 is 12.4 h
1 Mpc for 100 h
1 Mpc separations in
our cosmology if we normalize D such that D = (1 þ z)
1 at
high z (D = 0.758 at z = 0). The length will depend slightly on
cosmology, at least scaling linearly with the clustering ampli-
Fig. 3.—The rms Lagrangian displacement in the radial direction as a function
tude (our simulations have a linear 8 = 0.9 at z = 0, but the rel-
of redshift for initial separations of 100 h
1 Mpc. The upper points are for redshift evant normalization scale is much above 8 h
1 Mpc); s0 is also a
space when the initial separation is along the line of sight. The lower points are for slowly increasing function of scale.
real space. The lines are a model in which the real-space result varies as the growth If one approximates the transverse displacements as the same,
function D, while the redshift-space result varies as D(1 þ f ), where f = d( ln D)/ then the distribution of the displacement vector is very nearly an
d( ln a). The residuals to the fit are 1%, although our simulations may not be ac-
curate to this level because of their limited mass resolution. elliptical Gaussian, independent of the angle of the separation
vector to the line of sight. To the extent that one treats s0 as con-
stant, the effects on the correlation function become a simple con-
However, it is possible that the coarseness of the simulation res- volution, meaning that the modification of the power spectrum is
olution may be causing us to slightly underresolve the nonlinear simply to multiply by a Gaussian. To be clear, to the extent that the
behavior at z = 3, so the excellence of the growth function’s scal- displacement distributions are Gaussians of rms k and ? , along
ing might be somewhat worse in reality. This does not affect the and across the line of sight, we have
result for the acoustic oscillations, because the intrinsic width of
the acoustic peak is broader than the displacement distribution at

kk2 k2
that time. P(k) ¼ Plinear (k) exp
2
?2 : ð13Þ
The rms displacements are actually slightly smaller than pre- 2k 2?
dicted by the Zel’dovich approximation. This is the well-known
effect that pancake collapse in the Zel’dovich approximation This is the model for the portion of the linear power spectrum,
slightly overshoots; this overshoot prompted development of with acoustic peak, that survives. Of course, this is a poor model
models such as the adhesion model ( Kofman & Shandarin on small scales, as one has filtered out all of the small-scale clus-
1988; Gurbatov et al. 1989). The slight offset between the radial tering. One can restore the small-scale linear spectrum, without
and tangential rms matches the prediction of the Zel’dovich wiggles, by adding the no-wiggle approximation from Eisenstein
approximation. & Hu (1998) multiplied by unity minus the smearing function.
In redshift space, the displacements along the line of sight are Figure 4 shows the comparison of the measured power in the
enhanced. This is primarily due to the fact that the velocities and simulations (Seo & Eisenstein 2005) with the smeared model
displacements are well correlated on large scales ( Kaiser 1987); with the small-scale power restored. To make this comparison,
small-scale velocity dispersions (e.g., fingers of God) play a sub- we need to subtract off the smooth nonlinear power contribution
dominant role. Again, the displacement distributions are close to and divide out a linear bias. We do this by removing a polynomial
Gaussian, with small skewness. The kurtosis is larger; the redshift- with constant, k 2 , k 4 , and k 6 terms and dividing by a constant,
space distribution has larger non-Gaussian tails (Fig. 3). The means fitting over the range 0 < k < 0.4 h Mpc
1 (Seo & Eisenstein
are again zero. 2005). For redshift space, we adopt a smoothing function that is
The redshift-space variance is of course a function of the angle the spherical average of the elliptical Gaussian in Fourier space. In
of the displacement vector to the line of sight, as well as of the detail, we have in Figure 4 shown the power divided by the z = 49
angle of the displacement to the separation vector. As before, the input power spectrum. Purely linear evolution would be a line at
difference between the radial and transverse displacement dis- unity; degradations of the acoustic signature put oscillations in the
tributions is small, provided one holds the angle of the dis- plot.
placement to the line of sight fixed. For example, at z = 0.3 and Overall, the model is a very good fit. The amplitude of the os-
on 100 h
1 Mpc scales, the radial displacement for pairs along cillations in the power spectrum is well matched. The higher har-
the line of sight is 13.59 h
1 Mpc, while the line-of-sight displace- monics show more residuals, and it is not clear how much of this
ment for pairs transverse to the line of sight is 11.96 h
1 Mpc. is a failing of the model versus noise in the power spectra. Many
The displacements in the direction transverse to the line of sight of the excursions at high wavenumber cannot be matched even if
are of course unchanged by peculiar velocities. the acoustic oscillations are completely erased; that is, they must
As it is the radial distribution that most affects the smearing of be non-Gaussian noise in the simulated power spectra. How-
the acoustic peak, we will focus on the radial displacement along ever, the model assumptions to ignore the small deviations from
and across the line of sight. The transverse displacement vari- a Gaussian distribution of displacements, the transverse displace-
ances are only slightly smaller, and given their small impact on ments, and the scale dependence of the rms displacement could
the problem in any event, it is a good approximation to simplify result in minor changes to the residuals. It is important to note that
by rounding them up to the radial displacement variances. The even 3.75 h
3 Gpc 3 still leaves enough sample variance in the
 Fig. 4.—Power in the simulations compared with the model. Left, real space; right, redshift space. The top row is z = 0.3, the middle z = 1, and the bottom z = 3. In all
panels, the simulation power has been fitted over the range 0 < k < 0.4 h Mpc
1 to a model of b 2 Psmear þ a0 þ a2 k 2 þ a4 k 4 þ a6 k 6 , where Psmear is the smeared linear
power spectrum Plinear , with the small scales restored by the no-wiggle form. What is plotted is the residual Pres = (Pmeasured
a0
a2 k 2
a4 k 4
a6 k 6 )/b 2 divided by
the linear power spectrum. In detail, we divide by the initial power spectrum of the simulation, so that the variations in the initial amplitudes of particular modes cancel out;
this does not change the sense of the plot but tightens the residuals slightly. The smeared model (curves) has been similarly divided by the linear power spectrum. If the
linear spectrum were fully intact, it would be flat in this plot. Erasure of the acoustic peaks produces oscillations in this plot. One sees that the model provides an excellent
match to the observed degradation. [See the electronic edition of the Journal for a color version of this figure.]
 BARYON ACOUSTIC OSCILLATIONS. I. 669

Fig. 5.— Correlation function in the simulations compared with the model with the small-scale linear power restored. Left, z = 0.3; right, z = 1. In both panels, the
simulation data are the solid line, the linear correlation function is the short-dashed thin line, and the model correlation function is the long-dashed line. We have not
removed any broadband nuisance spectra in making this figure. [See the electronic edition of the Journal for a color version of this figure.]

power spectrum that one cannot fully test this model at the level This would correspond to a 0.4% bias in the recovered distance
of 1% residuals. scale, although it is not clear that this mean displacement is in fact
Figure 5 shows the match between the correlation function in the correct statistic to predict the exact shift. The mean for the mat-
the simulation and that of the model with the small-scale power ter displacements was zero because of homogeneity, but setting
restored. Here we do not remove any smooth power spectrum; the tracer density according to the late-time matter density breaks
all of these nuisance terms appear at small separations in the the homogeneity assumption. We explore this effect in detail in the
correlation function. Again, the agreement is good: the model next section.
seems to correctly match the smearing of the peak in the cor-
relation function. 8. NONLINEAR SHIFTS OF THE ACOUSTIC SCALE
8.1. Spherical Collapse
7. NUMERICAL RESULTS FOR BIASED TRACERS
Until this point, we have been mostly studying how the acous-
We next turn to biased tracers. We use a very simple model of tic signature is washed out by nonlinear structure. This addresses
bias, described in Seo & Eisenstein (2005). We find halos with a the precision with which the acoustic scale can be measured in a
friends-of-friends algorithm ( Davis et al. 1985), place a thresh- given survey. We now turn to the question of how much non-
old on halo multiplicity, and include all of the particles in the linear collapse can actually shift the scale of the peak, so as to cre-
halo as valid tracers. This is equivalent to a halo occupation model ate a systematic bias in the measurement. In detail, the question
in which the number of galaxies in a halo is proportional to the of a shift cannot be separated from the question of what statistic
halo mass, if above some threshold, and all galaxies trace the or analysis one plans to use to measure the acoustic scale. That
velocity dispersion of the dark matter. This is a relatively extreme is, measuring the maximum of the peak or some centroid or
model and is not intended to be realistic but simply to explore the
basic effect.
Table 2 lists the moments of the Lagrangian displacement dis- TABLE 2
tributions for biased tracers of different mass threshold and red- Lagrangian Displacements Distribution for Biased Tracers
shift. These mass thresholds at z = 0.3 give large-scale bias similar
to that of the Sloan Digital Sky Survey luminous red galaxy sam- Radial Transverse
ple ( Eisenstein et al. 2001; Zehavi et al. 2005). The primary z R M Mean  Skewness Kurtosis  Kurtosis
conclusion is that the bias does not change the variance of the
distribution much. The rms increases by 10% in real space and Real Space
somewhat more in redshift space. The latter is not surprising,
0.3........ 100 10
0.39 8.78
0.028 0.034 7.80 0.007
because the high halo masses have more small-scale thermal ve- 30
0.48 8.97
0.026 0.024 8.02
0.005
locities, particularly in the bias model we use here. We interpret 1.0........ 100 10
0.34 6.55
0.017 0.012 5.91
0.028
the small increase in real space as indicating that bulk flows af-
fect all galaxies more or less equally, with the subdominant trend Redshift Space
that the high-mass halos have more action in their small-scale
0.3........ 100 10
0.65 15.00
0.051 0.22 13.51 0.26
environments. It appears that bias has rather little effect on the
30
0.81 15.45
0.049 0.22 13.99 0.25
nonlinear degradation of the acoustic peaks. Figure 6 shows the 1.0........ 100 10
0.64 12.93
0.037 0.18 11.80 0.21
power spectrum from the z = 0.3, M = 10 biased tracer, along
with the model spectrum. As before, the agreement is excellent Note.—Masses of halos are listed as the number of particles M; each particle
within the errors. has a mass of 6 ; 1011 h
1 M . Separations R, as well as mean and rms displace-
However, we do find in the biased case that the mean pairwise ments, are given in comoving h
1 Mpc. Skewness and kurtosis statistics are the
usual dimensionless normalization. Negative skewness means that the heavier
displacements are no longer zero, as they were for the matter. The tail is inward. For redshift space, we list the displacements in the line-of-sight di-
mean displacements are small, about 0.4 h
1 Mpc in real space for rection when this direction is radial to the initial separation vector. The displace-
separations of 100 h
1 Mpc, somewhat larger in redshift space. ments in the direction perpendicular to the line of sight are the same as in real space.
670 EISENSTEIN, SEO, & WHITE Vol. 664

Fig. 6.—Sane as Fig. 4, but for a biased tracer (M = 10) at z = 0.3. Left, real space; right, redshift space. The biased power spectrum is noisy compared with the matter
power spectrum because of additional shot noise. However, one can see that the model remains a good match to the data. [See the electronic edition of the Journal for a
color version of this figure.]

performing a likelihood analysis with different templates and herently to the acoustic effect. With this smoothing, 0 is about
nuisance parameters could yield different answers when affected unity at z = 0.
by nonlinearities. The choice of a measurement scheme is beyond The value of 3J3 (150 Mpc) in the concordance cosmology
the scope of this paper. Here we offer a discussion of the mean (Spergel et al. 2007) at z = 0 is about 0.007, and so the mean
flows on 150 Mpc scale, which should explore the dynamics of (<R) is of this order. It is worth remembering that this over-
the nonlinearities. density is dominated by the correlations in the initial density
We again begin by considering the configuration-space picture. field carried by the CDM, not the acoustic effects. Translating the
A peak at one location creates a faint spherical echo and a pre- overdensity into a change in the radius of the sphere yields (<R)/3,
ferred separation of galaxies. It is clear that the nonlinear motion so the scale is changed only by O(0.003). In the line-of-sight di-
caused by clusters and flows will be largely independent between rection, redshift distortions double the apparent infall (Peebles
the two galaxies in a given pair. In other words, most of the non- 1980; Kaiser 1987) in EinsteinYde Sitter, and less so in low-
linear blurring of the peak (or erasure of the high harmonics in density, universes. It seems difficult to push this to be more than a
the power spectrum) does not shift the pairs systematically to a 1% effect.
different separation. But it is clear that peaks in the universe sep- The rms overdensity inside a sphere of 150 Mpc radius is about
arated by 150 Mpc will tend to fall toward each other, and that 0.07 in the standard cosmology at z = 0. However, this over-
voids will tend to move away from each other. Hence, there is a estimates the infall because the small overdensities that domi-
small anisotropy that could shift the acoustic scale. nate the total in the sphere can be anywhere inside the sphere,
How big is this effect? Let us first consider spherical collapse. not at the center, so that the infall is not radial and does not affect
If there is a matter overdensity at one point, there will tend to be the entire spherical acoustic echo.
a matter overdensity inside a cocentered sphere of radius R = Thus far, we have only discussed the infall around overden-
150 Mpc. The mean value of this overdensity is related to the sities. However, there is a equivalent outflow from underden-
J3 integral of the correlation function: sities. For the matter, these cancel to leading order. The next-order
Z terms do not cancel, and we therefore expect the changes in scale
30 R
30 to be O(10
4).
h(<R)i ¼ dr r 2 (r)  J3 (R): ð14Þ
R 3 02 0 02
8.2. Biasing the Acoustic Scale
Here 0 is the overdensity at the center, 02 is the variance in that We next study whether applying a local galaxy bias can create
overdensity, and J3 (R) is the integral of the correlation function, a first-order motion of the acoustic scale. We wish to compute
Z Z whether two galaxies separated initially by a given large distance
s
r 2 dr k 2 dk j1 (ks) will tend to fall toward each other. In the Zel’dovich approxi-
J3 (s) ¼ (r) ¼ P(k) ð15Þ mation, the pairwise displacement of two initial positions, pro-
0 s3 2 2 ks
jected along the initial separation vector, is (from eq. [8])
( Peebles 1980), where j1 is a spherical Bessel function. One can Z
u12 = r12 dk k = r12
consider the density field to be smoothed, provided that the u12;k ¼ ¼ 3
k 2 (e ik = r1
e ik = r2 ): ð16Þ
smoothing length is much smaller than R. Since 0 /0 will be of r12 (2) ik r12
order unity, the mean (<R) will scale as 
1 0 . If unsmoothed,
then 0 is very large and the mean (<R) is near zero; this is It is clear that hu12;k i = 0; one is performing an average weighted
simply a statement that behavior at a single point does not bias by the initial volume, not final density, and this cancellation is re-
an entire filled sphere very much. However, if one smooths the quired by homogeneity. If we consider hu12;k 1 i, then we find that
density field, 0 becomes smaller. The natural smoothing scale this is equal to
r12 J3 (r12 ), where J3 is defined in equation (15).
for this problem is something similar to the width of the acous- Hence, we do find that objects fall toward overdensities and away
tic shell, since that is the size of the central region that adds co- from underdensities. We note that although this is the same result
No. 2, 2007 BARYON ACOUSTIC OSCILLATIONS. I. 671

as from spherical collapse, we did not assume spherical symmetry remind the reader that 2 J3 (r12 ) is about 0.5% for the standard
here. cosmology at z = 0 (and of course smaller at high redshift).
We now wish to consider the mean displacement weighted by The pair-weighted mean displacement of equation (18) is not
pairs, that is, the same as the mean displacement between pairs of tracers, as
studied in x 7. The difference is effectively that of weighting by
hu12;k 1 2 i g or by (1 þ g )/(1 þ m ). These clearly disagree on the relative
upair
k  : ð17Þ
h1 2 i weighting of overdense and underdense regions, and the tracer-
based counting may fail to include the voids properly. However,
We consider these densities to be those of the initial linear den- in the limit of a bias model in which galaxies form only above a
sity field. The denominator h1 2 i is simply (r12 ), which can be high density threshold, the two calculations converge to the same
zero, but we will work in the limit where  is small but nonzero; answer.
we find that the ratio upair
k approaches a finite number as  ! 0.
For a toy model inspired by Press-Schechter (1974) theory,
In the Zel’dovich approximation, u12;k is linear in the density we consider a model in which galaxy formation obeys a simple
field  and hence upair
k = 0 for a Gaussian initial density field, as the
density threshold t , typically 1.7. For a threshold t  t /,
numerator is a three-point function. The displacement field will we find h 2 g i = t . This yields upair 2
k /r12 =
J3 (r12 )(2 t /t ). The
have corrections at orders above first, but this will produce upair
k at
usual problems in accounting for underdense regions in Press-
O(J32 (r12 )) or worse. In other words, there is a first-order cancel- Schechter theory recommend that one only consider this model
lation in the pair-weighted mean displacement when one uses an for reasonably extreme peaks, that is, t k 2. For a fixed t and
unbiased tracer. smoothing scale (and hence a rapidly declining number density
We next consider weighting by the galaxy overdensity field of objects at higher redshift), the redshift dependences of t2 and
g . We assume a local bias model in which g is a function simply J3 cancel, matching the behavior in Table 2. Matching the bias
of the initial matter density . We write g;1 = g (1 ). Now we have models used at z = 0.3 in x 7 would suggest t  1.6, which
pair
makes u k /r12 
0.6%, somewhat larger than the measured
hu12;k g;1 g;2 i mean displacement but in qualitative agreement. However, it is
upair
k  : ð18Þ
hg;1 g;2 i not surprising that the actual halo bias is closer to linear than the
sharp-threshold model.
Importantly, the displacement u12;k still depends on the matter Estimating an exact level of the bias in the measured acoustic
density field, not the galaxy density field. In the limit that |(r12 )| scale is not straightforward from this local bias formalism, as one
is much smaller than the variance at a single point,  2 = h12 i, has to decide what smoothing scale to apply to the initial density
the mean Zel’dovich displacement given 1 and 2 is field, both to define the local bias model g ( ) and to compute the
matter variance  2 . One sees, however, that one is caught between
r12 J3 (r12 ) two limits in which the cancellation is restored: if one chooses a
hu12;k i1 ;2 ¼
(1 þ 2 ): ð19Þ
2 large smoothing scale, then  can be small, but the local bias be-
comes more linear, whereas with a small smoothing scale the bias
Keeping only the leading-order terms, we have will be nonlinear, but  is larger. Hence, it seems likely that the
combination h 2 g i
1 will be somewhat less than 1 at z = 0,
r12 J3 (r12 ) h(1 þ 2 )g;1 g;2 i
upair
k 
: ð20Þ yielding a shift of about 0.5%. The shift in the acoustic scale
2 hg;1 g;2 i should decrease at high redshift, somewhat slower than the square
of the growth function, for tracers of a given number density.
We next consider the form of the function g . We must have To summarize, for objects that trace the matter, the inflow be-
hg i = 0, where the averaging is over the distribution of the mat- tween positive-density pairs exactly cancels the outflow between
ter density. This simply defines the mean density of galaxies. negative-density pairs. Galaxy bias can shift the balance, yield-
We also have hg i = b 2 , where  2 = h 2 i and b is the familiar ing a net flow. Fortunately, the effect involved, namely, the flows
large-scale bias. Allowing g to be a stochastic distribution de- on 150 Mpc scale, is very easy to simulate. We expect that large-
pending on the matter density at that point (e.g., Dekel & Lahav volume gravitational simulations with relatively poor mass reso-
1999) does not alter the result; to leading order, only the mean lution should be able to give accurate estimates of the effect for
galaxy density as a function of  enters. various galaxy bias prescriptions.
To compute the expectation values, we must include the fact
that 1 and 2 are correlated. In the limit that |(r12 )|T 2 , we find 9. CONCLUSIONS
hg;1 g;2 i = b 2 (r12 ) and h1 g;1 g;2 i = [b(r12 )/ 2 ]hg;1 12 i. If we We have investigated the nonlinear degradation of the baryon
define rescaled density fields = / and g = g /b, then we acoustic signature through several different methods. We argue
reach the result that as the clustering signature is manifested as a peak in the cor-
relation function on large scales, the degradation enters primarily
upair
k 2J3 (r12 ) 2 through the motion of the two galaxies altering the separation be-
¼
h g( )i; ð21Þ tween them. This smearing is due to small-scale thermal motion,
r12 
for example, cluster formation, and coherent motions, for exam-
where the expectation value is computed with being distrib- ple, bulk flows into superclusters or out of voids between the two
uted as a Gaussian of unit variance. The function g is also sub- objects. This smearing broadens the peak in the correlation func-
ject to h g i = 0 and h g i = 1, the latter being due to the scaling tion and decreases the higher harmonics in the power spectrum.
by b. We model this smearing by measuring the distribution of
Hence, the fractional change in the pair-weighted separation differences of the Lagrangian displacements of pairs of particles
can be first order in the large-scale correlation function if the initially separated by a separation equal to the sound horizon.
local bias is suitably chosen. However, for a linear bias, g / These distributions are close to Gaussian in both real and redshift
(which must actually be g = ), the first-order effect cancels. We space, and we find that the redshift dependence of the rms width
672 EISENSTEIN, SEO, & WHITE Vol. 664

is well predicted by the linear-theory scaling predictions. We pro- tic scale at levels of 0.5% or better will have fantastic data on the
pose convolving the linear-theory correlation function with this small-scale clustering and environments of the tracer galaxies.
Gaussian. The linear-theory power spectrum can be multiplied By viewing the nonlinear acoustic signature as the spreading
by the Fourier transform of this Gaussian, and the small-scale of a preferred separation, one sees that the acoustic scale must be
power restored using the no-wiggle form from Eisenstein & Hu robust to the effects of scale-dependent bias and nonlinear clus-
(1998). tering. This is not obvious in the power spectrum, where the higher
This model has significant advantages. First, the displacements acoustic harmonics (k  0.2 h Mpc
1) appear at the same wave-
can be calculated with a relatively modest volume of cosmological number as the quasi-linear regime. The key point is that the
simulations as compared with those required to measure the de- peaks and troughs of the acoustic series define a beat frequency
graded power spectrum. However, one should use box sizes of at that is at very large wavelength. Alternatively stated, nonlineari-
least 512 h
1 Mpc to model the required wavelengths. Second, the ties on 10 h
1 Mpc scales can only introduce broadband (k 
Lagrangian displacements are easy to calculate in real or redshift 0.1 h Mpc
1) modifications of the power spectrum and therefore
space and can be applied to arbitrary tracers of the field, as the cannot affect the peaks and troughs differentially, save to damp
fuzzy nature of the initial position of a galaxy is still well confined out the entire linear spectrum.
relative to the width of the Lagrangian displacement distribution. The fact that small-scale nonlinear clustering produces smooth
Third, in the context of survey parameter estimation, one can now power spectra compared with the acoustic oscillations is also im-
avoid the assumption of simply truncating the linear-mode count- portant for making unbiased measurements of the acoustic scale.
ing at a given maximum wavenumber, as has been the standard The measured power spectra will generally be tilted relative to
practice (e.g., Eisenstein et al. [1998] and Seo & Eisenstein the linear spectrum, and it is well known that when measuring the
[2003], among many others). Instead, we now have an accurate position of a peak, one must avoid tilting the baseline on which the
model for the quasi-linear acoustic signature, in which the higher peak sits. This leads to the speculation that nonlinearities can bias
harmonics are gradually erased. We plan to include this model in the measurement of the acoustic scale. We see here that this need
survey forecasts in a future paper. not be the case. In configuration space, these effects collapse to
Although degradations of the acoustic signature are important small separations, well separated from the acoustic scale. Alterna-
to estimating the statistical performance of a given volume, an tively, in Fourier space one can marginalize over smooth nuisance
actual shift in the acoustic scale would be more important, as this functions. The linear power spectrum predicted as part of the
would lead to a bias in the inferred distance scale. The idea that sound-horizon calculation provides an easy template against
the acoustic signature is a single peak in the correlation function which to make an unbiased measurement.
allows one to argue that these effects must be small: to shift the In short, the acoustic scale is a robust standard ruler because it
peak, one must systematically move pairs initially at 150 Mpc is fundamentally a 150 Mpc clustering signature. This is far larger
scale outward or inward on average. We have investigated this than the characteristic scales for halo formation and galaxy bias,
with two different models, the mean pairwise displacement of tra- and the 10% width of the peak makes it very implausible to mimic
cers and the pair-weighted mean displacement in the Zel’dovich by astrophysical processes. We have shown here that the degrada-
approximation. In both cases, we find that linearly biased tracers tions of the signature have benign quasi-linear causes. These can
imply a cancellation of the first-order terms, which would make be computed to high accuracy with N-body gravitational simula-
the shift negligible at O(10
4). With biased galaxy formation, tions. The remaining challenge for the method is the enormous
we estimate shifts of order 0.5% for objects with b  2 at z = 0, survey volumes required to measure the scale to high precision.
dropping at higher redshift. However, neither of our models is
exact, nor have we considered the issue of how a particular choice
of statistic by which one measures the acoustic scale might be D. J. E. thanks Michael Joyce for useful conversations, and
affected by nonlinearities. We believe that large volumes of simu- the Miller Institute at the University of California for support
lations will be needed to refine the estimates and calibrate partic- during a key period in the development of this paper. D. J. E. and
ular measurement techniques. It is highly plausible that corrections H.-J. S. are supported by grant AST 04-07200 from the National
could be derived from simulations that would decrease the resid- Science Foundation. D. J. E. is further supported by an Alfred P.
ual shift by a factor of a few and reach 10
3 accuracy. After all, Sloan Research Fellowship. M. W. is supported by NASA and
any survey capable of reaching statistical precision on the acous- the NSF.

APPENDIX
AN INVESTIGATION OF LOCAL BIAS

In the configuration-space picture of the acoustic signature, the sound waves created by a peak in the density field leave a small
residual overdensity in a shell at 150 Mpc radius. We consider in this appendix whether there is a judicious choice of tracers that would
increase the contrast of the shell or peak, so as to increase the strength of the signature in the correlation function. Unfortunately, we
will conclude that there is not, reinforcing the concepts of linear bias.
The thickness of this shell is about 10 Mpc because of Silk damping and the damping of oscillatory modes. Hence, two points
separated by T10 Mpc create acoustic shells that essentially overlap. In what follows, we consider the density field smoothed on
10 Mpc scales.
The variance at low redshift in the overdensity in 10 Mpc patches is roughly unity. The corresponding echoes at 150 Mpc separation
are only 1% in amplitude, far smaller than the primary variations at those locations (which are again order unity). The smallness of the
echo suggests that a local bias model should apply. We therefore consider the response of the behavior in 10 Mpc patches when acted
upon by bulk shifts in their density, corresponding to long-wavelength perturbations.
We will consider two regions of size 10 Mpc separated by a distance of 150 Mpc. We write the linear matter overdensity (ignoring
the acoustic effect) in the two as m;1 and m;2 . The m will be Gaussian-distributed with variance ; we write the distribution as p(m ).
No. 2, 2007 BARYON ACOUSTIC OSCILLATIONS. I. 673

Because of the acoustic effect, the density at location 2 will actually be m; 2 þ m;1 , where is about 0.01. There is a reciprocal effect
on the density at location 1, which we ignore for clarity.
Now we imagine that galaxies are formed in a region of density m with a mean density of g (m ). Of course,Ðthe actual number will
scatter from the mean, but only the mean enters this calculation. The homogeneous density of galaxies is ¯ g = dm p(m )g (m ). We
define the galaxy overdensity as g (m ) = [g (m )
]/
¯ .
¯
With such a model, the bias at large scales (small wavenumber) should be the response of this density to a small long-wavelength
fluctuation in the matter density:
R R Z
dm p(m )g (m þ )
¯ dm p(m )g (m þ ) dg
b ¼ lim ¼ ¼ dm p(m ) : ðA1Þ
!0 ¯ dm

Meanwhile, the amplitude of the acoustic effect should be the product of the galaxy overdensities at the two locations:
Z Z
B 2 ¼ dm;1 p(m;1 )g (m;1 ) dm; 2 p(m; 2 )g (m; 2 þ m;1 ): ðA2Þ

Because is small, we can use the limit in equation (A1) to form

Z
2
B ¼ dm;1 p(m;1 )g (m;1 )b m;1 : ðA3Þ

One might at this point have thought that by picking g judiciously, one could produce an amplitude for the acoustic peak that would
increase its contrast relative to the square of the large-scale bias. In other words, one has a large peak in region 1 and a faint echo in
region 2. The galaxy density in region 2 is unavoidably controlled by the standard large-scale bias, as the acoustic signature is a sub-
dominant echo at that location. But one could hope that one could pick a tracer that would emphasize region 1 more. One could even
imagine cross-correlating two tracer sets, one that maximizes the signal-to-noise ratio for small long-wavelength perturbations (region 2)
and another that picks out high peaks (region 1).
Unfortunately, this is all for naught. Integrating equation (A3) by parts yields

B 2 ¼ b 2 2 ðA4Þ

for any g (m ) bias function. The trick is that the indefinite integral of m p(m ) is just
 2 p(m ) because the initial density distribution
p is a Gaussian.
This result shows that at the level of local bias, one cannot pick the tracer to maximize the contrast in the acoustic peak in the linear
regime. This is a simple corollary of the usual theorems about local bias (Coles 1993; Scherrer & Weinberg 1998). It may be possible
to find tracers that minimize the nonlinear spreading of the acoustic peak or that minimize the amount of small-scale near-white noise; such
tracers would improve the signal-to-noise ratio of the measurement. But there is no opportunity at the simple level of pair counting.

REFERENCES
Amendola, L., Quercellini, C., & Giallongo, E. 2005, MNRAS, 357, 429 Goroff, M. H., Grinstein, B., Rey, S.-J., & Wise, M. B. 1986, ApJ, 311, 6
Bashinsky, S., & Bertschinger, E. 2001, Phys. Rev. Lett., 87, No. 081301 Groth, E. J., & Peebles, P. J. E. 1975, A&A, 41, 143
———. 2002, Phys. Rev. D, 65, No. 123008 Gurbatov, S. N., Saichev, A. I., & Shandarin, S. F. 1989, MNRAS, 236, 385
Bennett, C. L., Turner, M. S., & White, M. 1997, Phys. Today, 50(11), 32 Halverson, N. W., et al. 2002, ApJ, 568, 38
Bennett, C. L., et al. 2003, ApJS, 148, 1 Hamilton, A. J. S. 1998, in The Evolving Universe: Selected Topics on Large-
Blake, C., & Bridle, S. 2005, MNRAS, 363, 1329 Scale Structure and on the Properties of Galaxies, ed. D. Hamilton ( Dordrecht:
Blake, C., & Glazebrook, K. 2003, ApJ, 594, 665 Kluwer), 185
Blake, C., Parkinson, D., Bassett, B., Glazebrook, K., Kunz, M., & Nichol, R. C. Hanany, S., et al. 2000, ApJ, 545, L5
2006, MNRAS, 365, 255 Hu, W., & Dodelson, S. 2002, ARA&A, 40, 171
Blumenthal, G. R., Faber, S. M., Primack, J. R., & Rees, M. J. 1984, Nature, Hu, W., & Haiman, Z. 2003, Phys. Rev. D, 68, No. 063004
311, 517 Hu, W., & Sugiyama, N. 1995, ApJ, 444, 489
Bond, J. R., & Efstathiou, G. 1984, ApJ, 285, L45 Hu, W., Sugiyama, N., & Silk, J. 1997, Nature, 386, 37
Cole, S., et al. 2005, MNRAS, 362, 505 Hu, W., & White, M. 1996, ApJ, 471, 30
Coles, P. 1993, MNRAS, 262, 1065 ———. 1997, ApJ, 479, 568
Cooray, A., Hu, W., Huterer, D., & Joffre, M. 2001, ApJ, 557, L7 Hütsi, G. 2006, A&A, 446, 43
Crocce, M., & Scoccimarro, R. 2006, Phys. Rev. D, 73, No. 063520 Jain, B., & Bertschinger, E. 1994, ApJ, 431, 495
Davis, M., Efstathiou, G., Frenk, C. S., & White, S. D. M. 1985, ApJ, 292, 371 Jeong, D., & Komatsu, E. 2006, ApJ, 651, 619
de Bernardis, P., et al. 2000, Nature, 404, 955 Kaiser, N. 1987, MNRAS, 227, 1
Dekel, A., & Lahav, O. 1999, ApJ, 520, 24 Kodama, H., & Sasaki, M. 1984, Prog. Theor. Phys. Suppl., No. 78, 1
Dolney, D., Jain, B., & Takada, M. 2006, MNRAS, 366, 884 Kofman, L. A., & Shandarin, S. F. 1988, Nature, 334, 129
Doroshkevich, A. G., Zel’dovich, Ya. B., & Sunyaev, R. A. 1978, AZh, 55, 913 Lifshitz, E. M. 1946, Zh. Eksp. Teor. Fiz., 16, 587 ( English transl. J. Phys.
( English transl. Soviet Astron., 22, 523) USSR, 10, 116)
Eisenstein, D. J. 2002, in ASP Conf. Ser. 280, Next Generation Wide-Field Linder, E. V. 2003, Phys. Rev. D, 68, No. 083504
Multi-Object Spectroscopy, ed. M. J. I. Brown & A. Dey (San Francisco: Matsubara, T. 2004, ApJ, 615, 573
ASP), 35 Matsubara, T., Szalay, A. S., & Pope, A. C. 2004, ApJ, 606, 1
Eisenstein, D. J., & Hu, W. 1998, ApJ, 496, 605 Meiksin, A., White, M., & Peacock, J. A. 1999, MNRAS, 304, 851
Eisenstein, D. J., Hu, W., & Tegmark, M. 1998, ApJ, 504, L57 Miller, A. D., et al. 1999, ApJ, 524, L1
Eisenstein, D. J., et al. 2001, AJ, 122, 2267 Netterfield, C. B., et al. 2002, ApJ, 571, 604
———. 2005, ApJ, 633, 560 Padmanabhan, T. 1993, Structure Formation in the Universe (Cambridge:
Glazebrook, K., & Blake, C. 2005, ApJ, 631, 1 Cambridge Univ. Press)
674 EISENSTEIN, SEO, & WHITE
Peebles, P. J. E. 1968, ApJ, 153, 1 Silk, J. 1968, ApJ, 151, 459
———. 1980, The Large-Scale Structure of the Universe ( Princeton: Princeton Spergel, D. N., et al. 2007, ApJS, 170, 377
Univ. Press) Springel, V., et al. 2005, Nature, 435, 629
———. 1981, ApJ, 248, 885 Steigman, G. 2006, Int. J. Mod. Phys. E, 15, 1
Peebles, P. J. E., & Yu, J.-T. 1970, ApJ, 162, 815 Sunyaev, R. A., & Zel’dovich, Ya. B. 1970, Ap&SS, 7, 3
Press, W. H., & Schechter, P. 1974, ApJ, 187, 425 White, M. 2005, Astropart. Phys., 24, 334
Press, W. H., & Vishniac, E. T. 1980, ApJ, 236, 323 White, M., & Cohn, J. D. 2002, Am. J. Phys., 70, 106
Scherrer, R. J., & Weinberg, D. H. 1998, ApJ, 504, 607 Wilson, M. L., & Silk, J. 1981, ApJ, 243, 14
Schulz, A. E., & White, M. 2006, Astropart. Phys., 25, 172 Zaldarriaga, M., & Seljak, U. 2000, ApJS, 129, 431
Scoccimarro, R. 2004, Phys. Rev. D, 70, No. 083007 Zehavi, I., et al. 2005, ApJ, 621, 22
Seager, S., Sasselov, D. D., & Scott, D. 1999, ApJ, 523, L1 Zel’dovich, Ya. B. 1970, A&A, 5, 84
Seljak, U., & Zaldarriaga, M. 1996, ApJ, 469, 437 Zel’dovich, Ya. B., Kurt, V. G., & Sunyaev, R. A. 1968, Zh. Eksp. Teor. Fiz.,
Seo, H.-J., & Eisenstein, D. J. 2003, ApJ, 598, 720 55, 278 ( English transl. Soviet Phys.YJETP, 28, 146 [1969])
———. 2005, ApJ, 633, 575