CMB-Normalized Predictions for Sunyaev-Zel'dovich effect fluctuations

Submitted to ApJL July 18,2002

Preprint typeset using L^T^ style emulateapj v. 14/09/00

CMB-NORMALIZED PREDICTIONS FOR SUNYAEV-ZEL'DOVICH EFFECT FLUCTUATIONS

(N
O
O

0^

>
cn
cn
\o
i>
o

(N

O ■

Oh
I

o

Gilbert P. Holder

School of Natural Sciences, Institute for Advanced Study, Princeton NJ, 08540; holder@ias.edu
Submitted to ApJL July 18,2002

ABSTRACT

We predict the level of small-scale anisotropy in the cosmic microwave background (CMB) due to the Sunyaev-
Zel'dovich (SZ) effect for the ensemble of cosmological models that are consistent with current measurements of
large-scale CMB anisotropy. We argue that the recently reported detections of the small-scale (arcminutes) CMB
anisotropy are only marginally consistent with being the SZ effect when cosmological models are calibrated to the
existing primary CMB data on large scales. The discrepancy is at more than 2-2.5(7, and is mainly due to a lower
CTg ^ 0.8 favored by the primary CMB and a higher > 1 favored by the SZ effect. A degeneracy between the
optical depth to Thomson scattering and the CMB -derived value of erg suggests that the discrepancy is reduced if
the universe was reionized very early, at redshift of ~ 25.

Subject headings: cosmic microwave background — cosmological parameters — cosmology: observations

1. INTRODUCTION

The current generation of cosmic microwave background
(CMB) experiments are producing a wealth of information on a
wide range of angular scales. With the recent measurements by
BIMA (Dawson et al. 2002) and CBI (Mason et al. 2002) com-
plementing earlier measurements (see Hu and Dodelson 2002,
for a recent compilation), CMB anisotropy has been measured
over a range of multipoles of I = 2-6000 (or angular scales of
2' -90°).

At low multipoles (I < 2000) anisotropy is primarily gen-
erated at z > 1000 except at very low multipoles (/ < 10)
where late-time decay of gravitational potential contributes sig-
nificantly. At higher multipoles (smaller angular scales) low-
redshift sources generate a significant amount of fluctuation
power At the observing frequencies of CBI and BIMA (^ 30
GHz), the largest sources of low-redshift anisotropy are radio
point sources and the thermal Sunyaev-Zel'dovich (SZ) effect.
The latter is of cosmological interest and may be large enough
to be detected, depending on cosmology (e.g.. Cole and Kaiser
1988).

The reported detections of power at small angular scales (I =
2000-6000) have argued that point-source contamination is not
a problem, suggesting that the detected power could be due to
the SZ effect (Bond et al. 2002; Dawson et al. 2002; Komatsu
and Seljak 2002). Since the number density and brightness of
the sources contributing to the SZ fluctuations (i.e., hot gas in
galaxy clusters at z < 1) depend on the background cosmology,
the level of the SZ fluctuations depends on cosmological param-
eters. The SZ angular power spectrum is sensitive to the matter-
fluctuation amplitude and the baryon density of the universe
but relatively insensitive to the matter density of the universe
(Komatsu and Kitayama 1999) or other cosmological parame-
ters (Komatsu and Seljak 2002). By fitting the CBI and BIMA
data to theoretical predictions, Komatsu and Seljak (2002) have
found a constraint on linear r.m.s. mass fluctuations within an
8 Mpc sphere, ag, as asiflbh/OmS)"-^'^ = 1.04±0.12 at the
95% confidence level.

On the other hand, observations of the primary CMB
anisotropy and the large-scale structure (LSS) of the universe
have already provided tight constraints on cosmological param-
eters (e.g., Wang et al. 2002). By using these constraints we

can predict how much SZ power ought to be seen at the CBI
and BIMA multipole bands. By doing so we can see if an SZ
interpretation of the small-scale fluctuations is consistent with
cosmological models favored by CMB or LSS.

In this Letter we estimate the level of the SZ angular power
spectrum expected from cosmological models consistent with
CMB data at / < 2000, and compare it with the CBI and BIMA
data at 2000 < / < 10000. We use only CMB data and a prior
on the Hubble constant and do not include any constraints from
LSS. For the primary CMB data, we use the Markov Chain
Monte Carlo (MCMC) results of Lewis and Bridle (2002) as
an estimate of the appropriate weighting for each cosmological
model. For the SZ effect, we use a model for the SZ power
spectrum as a function of cosmology from Komatsu and Sel-
jak (2002). In §2 we oudine salient features of the MCMC re-
sults. In §3 we sketch out our calculation of the SZ power spec-
trum, including the effects of the non-Gaussian nature of the
SZ fluctuations. In §4 we compare MCMC realizations of the
SZ power spectrum weighted by the primary CMB data with
the CBI and BIMA data, and check for consistency between
them. In §5 we discuss implications for the reported detections
of power at small angular scales.

2. CMB-CALIBRATED COSMOLOGICAL MODELS

As a subset of cosmological models consistent with the cur-
rent CMB data, we use the MCMC results of Lewis and Bridle
(2002) (for details see http://cosmologist.info/cosmomc). Esti-
mating cosmological parameters using MCMC methods (Chris-
tensen et al. 2001) entails taking random steps in parameter
space, accepting a step if the new point is more likely, other-
wise accepting the step with some probability (less than one)
set by how much worse the new point is than the current point.
The resulting list of points should sample the likelihood func-
tion in the multidimensional parameter space.

An advantage of MCMC methods (besides often being faster
than grids) is that the resulting list of consistent models (the
"chain") provides a self-consistent sampling of the likelihood
distribution. A distribution of CMB-derived cosmological pa-
rameters using the MCMC chain therefore provides a maxi-
mally informative CMB prior. Covariances between parameters
are naturally incorporated. The chain provides more samples

2

HOLDER

where the density is higher; this feature is actually a disadvan-
tage when dealing with models lying in the tail of the distribu-
tion, as we will return to later.

The chain that we use (Lewis and Bridle 2002) assumes a
flat universe, no tensor component, and an equation of state
for the dark energy component of w = -1 (i.e., assumes a cos-
mological constant rather than quintessence). A prior on the
Hubble constant (h = 0.72 ± 0.08) consistent with the Hubble
Key Project results (Freedman et al. 2001) is assumed, but
no LSS priors. There are six free parameters in the chain:
^bh^, ^CDuh^, ^A, Zte, «s> and Ag, where fib is the baryon
density relative to the critical density, Ocdm the cold dark-
matter density, the cosmological-constant energy density,
Zre the redshift of reionization, the scalar spectral index,
and As the amplitude of density fluctuations. The constraint
that spatial curvature is zero defines the Hubble constant as

1 /2

h = [(Q\, + QcDM)h^/(i-^A)] ■ The total matter density
f^CDM + and (Tg are derived parameters for each model. There
are 2,596 MCMC samples in total.

For each of the cosmological models in the chain, we predict
the SZ power spectrum. The resulting distribution of powers
represents the prior probability of a given amount of SZ power,
given the current large-angle CMB data, i.e., CMB-normjiUzed
predictions for the SZ power spectrum.

3. ANGULAR POWER SPECTRUM OF THE SZ EFFECT

The SZ effect arises from Compton scattering of CMB pho-
tons with hot electrons in gas in halos (Sunyaev and Zel'dovich
1980). Many authors have estimated the SZ angular power
spectrum Q using analytic methods or hydrodynamic simula-
tions, with broad agreement between the calculations to within
a factor of two at / < lO'* (e.g., see Komatsu and Seljak 2002,
for a recent analytic prediction and comparison between ana-
lytic methods and numerical simulations; see references therein
for previous work). Komatsu and Seljak (2002) find that C; is
sensitive to the baryon density and erg with an approximate scal-
ing of C/ oc aliflbh)^, but is almost independent of any other
cosmological parameters.

We compute C; as a sum of one-halo Poisson contributions
over all halos which can contribute:

Q = 8ljdzfJdM'-!^\UM,z)\\ (1)

where V{z) is the comoving volume of the universe at z per
steradian, dn{M,z)/dM the comoving dark-matter halo mass
function, and the spectral function of the SZ effect (Sunyaev
and Zel'dovich 1980). We ignore the correlated contribution as
it is unimportant at / > 300 (Komatsu and Kitayama 1999). The
physics of gas in halos is encoded in yi{M,z), the 2D Fourier
transform of the Compton y-parameter as a function of halo
mass and redshift. The y-parameter is directly proportional to
the integrated gas pressure along the line of sight; this leads to
Q oc {n^hf.

We use the method of Komatsu and Seljak (2002) for com-
puting Q. In brief, we compute a gas-pressure profile using the
universal gas-density and temperature profiles derived by Ko-
matsu and Seljak (2001) which make three assumptions about
gas in halos: (1) hydrostatic equilibrium between gas pressure
and the gravitational force induced by a universal dark-matter
density profile (Navarro et al. 1997), (2) the gas density tracing
the dark matter density at large radii, as observed in simulated
galaxy clusters (e.g., Frenk et al. 1999), and (3) a constant poly-
tropic equation of state for the gas. For the mass function we

use the mass function of Jenkins et al. (2001). This prescrip-
tion has no free parameters, and is in broad agreement with SZ
simulations to within a factor of two at / < lO'*, agreeing with
simulations at least as well as different simulations agree with
each other

Gas coohng, energy feedback, and star formation are not in-
cluded in our models; Komatsu and Seljak (2002) have argued
that these effects are not very important for Q, which is dom-
inated by gas outside the core of halos. Recently, White et al.
(2002) have used hydrodynamic simulations to show that these
effects affect Q by no more than a factor of two. These studies
suggest that our theoretical prediction for Q is accurate to bet-
ter than a factor of two, and thus MCMC-derived predictions
for C; in the next section can be trusted up to this accuracy.

The SZ fluctuations are highly non-Gaussian, as character-
ized by large skewness (Seljak et al. 2001; Zhang et al. 2002;
White et al. 2002); thus, sampling variance of C; for the SZ fluc-
tuations differs from that for Gaussian fluctuations, and mea-
surements of Ci in adjacent bins are highly correlated (Cooray
2001). We have to include the non-Gaussianity in the error
analysis to interpret observational SZ data correctly. We do this
by explicitly including those trispectrum configurations which
contribute to the power-spectrum covariance matrix. The error
of Ci in a bin of size A/ is given by

sky

2{Q + Cff ^ Tu
(2/+l)A/ 47r

(2)

where Tu is the angular trispectrum relevant for the non-
Gaussian sampUng variance (Cooray 2001),

dz—

dz

(3)

The expression for AC; is correct when both C/ and Tu are suf-
ficiently smooth in I as is the case here. Komatsu and Seljak
(2002) have found that the predicted AC; agrees with hydrody-
namic simulations to within a factor of two.

For the instrumental-noise power spectrum Cf we use l{l +
l)Cf /(27r) = 500, 1000, 1500, and 3000 /xK^ at / = 1703, 2183,
2630, and 3266 for CBI, and 720 and 2000 ijlK? at / = 5237 and
8748 for BIMA. For sky coverage of observations A-nf^ky we

use 1 deg^ for CBI, and 0.1 deg^ for BIMA. The bin sizes are
A/ = 565, 378, 612, 1000, 2870, and 4150. Alfliough we do not

use the exact window functions for the experiments, C; is very
flat at these multipoles, reducing the importance of the window
functions.

4. RESULTS

In figure 1 we show the histogram (weighted according to
the weights of Lewis and Bridle (2002) for each model) of ex-
pected SZ power for CBI and BIMA high-/ bins, where the
three highest-/ CBI bins and the two BIMA bins are averaged to
single bins for each experiment. The most likely values for the
expected SZ power are less than 100 jiK?. The weighted mean
values are 97 and 86 /iK^ for CBI and BIMA, respectively,
while the median values are 79 and 69 /xK^. We also find that
the analytic approximate scaling of Komatsu and Seljak (2002)
at / - 2000-6000, /(/ + l)C//(27r) ~ 330 aliVL^h / msf ,
gives similar results; the weighted mean is 100 /iK^, while the
median is 88 /xKl Bond et al. (2002) find 20-30% lower
normalization for the scaUng relation at the CBI band, shghtly
shifting the histogram leftward.

CMB-NORMALIZED SZ PREDICTIONS

3

400

T3
O
O

3 200

lll III lllllj 1 1 1 llllll 1 1 1 llllll 1

CBI (95%) .

lll III llllll 1 1 1 lllllj 1 1 1 lllllj 1

_ BIMA (95%) _

■ -

1

1 \

'A '

iil-^i'iiiil 1 , , mV ,

1 10 100 1000 1 10 100 1000
l(l+l)Ci/27T [AiK2]

Fig. 1. — Unnormalized probability distribution of the SZ angular power
spectrum in CBI (left panel) and BIMA (right panel) high-/ averaged windows
(solid lines), weighted by the primary CMB likelihood. The gray areas show
95% confidence regions of the CBI (Mason et al. 2002) and BIMA (Dawson
et al. 2002) data, assuming Gaussian fluctuations. Note that the true error is
larger because of non-Gaussian nature of the SZ fluctuations, and depends on
cosmological models. Also shown is the result using analytic approximation
of Komatsu and Seljak (2002) for C; (dashed hnes).

For CBI the 95% lower limit (assuming Gaussian fluctua-
tions) to the observed anisotropy is 199 /xK^, making an SZ
interpretation of the CBI measurement only marginally consis-
tent with the CMB-cahbrated theories. We show this exphcitly
in figure 2 where we plot the number of standard deviations
of the CBI and BIMA data relative to the predicted levels of
SZ fluctuations. We have computed the errors AC; using equa-
tion (2). We find that the CBI measurements at Z = 2183 and
2630 are inconsistent with most of the models that are consis-
tent with the primary CMB data at greater than 2.5 and 2a lev-
els, respectively. The rest of the CBI bins and the BIMA bins
are consistent with the predicted levels of the SZ effect.

The cutoff at the 2.5 - 3(7 level in figure 2 is due to instrumen-
tal noise. At the 2.5 -3fT level the current data are consistent
with no SZ signal. Thus, lower noise experiments could extend
the histograms to larger cr values. Note that the largest values
of a in the figure are higher than signal-to-noise ratios of CBI
quoted by Mason et al. (2002). This is because AC/ depends
on the SZ power spectrum; a lower SZ power spectrum leads
to lower sample variance, and therefore a smaller AC;. As a
result, at the largest a the instrumental noise determines AC;
entirely, making it smaller than the quoted CBI errors.

Which models can account for the data at / = 2000-3000?
We find that those models typically have erg > 1 and ilbfi^
higher than the BBN-aUowed value of n^h^ = 0.020 ± 0.002
(Buries et al. 2001). A simple scaling of C; oc aliClhh)^ ac-
counts for this; thus, we need erg > 1 for a given BBN value
of Clhh to account for the CBI data by the SZ effect. This con-
clusion agrees with Bond et al. (2002) and Komatsu and Seljak
(2002).

In principle, it would be straightforward to do a joint like-
lihood analysis of the primary CMB and the SZ effect using
CMB data on all angular scales. In practice, the strong non-
Gaussianity of the SZ signal, combined with the uncertain field

selection effects (e.g., no bright point sources) make such a pro-
cedure still premature. Furthermore, the required erg > 1 is in
the tail of the CMB-preferred distribution, so the resolution of
the MCMC chain is poor in this region. Importance sampling
becomes unreUable in regions where the density of points is
low.

|Cobs_CSZ|/(^C,)

Fig. 2. — Number of standard deviations of observed data points relative to
predicted SZ fluctuations (each data point corresponds to a panel), weighted
by the primary CMB likelihood. The top four panels are the CBI data points
(Mason et al. 2002), while the bottom two panels the BIMA data (Dawson
et al. 2002). The effects of non-Gaussianity have been included in the error
estimation.

5. DISCUSSION

We have presented estimates of the SZ angular power spec-
trum at I > 2000 for cosmological models that are consistent
with the current CMB anisotropy data at / < 2000. The mode
of distribution of the predicted SZ power at / ~ 3000 is 50 /zK^,
while we find that the weighted mean is 97 /zK^ (the median
is 78.5 /iK^), still significantly lower than the reported CBI de-
tection of anisotropy on these angular scales, 508 /iK^. Mov-
ing out from the mode at levels of equal likelihood, the ranges
of power enclosed by 68% and 95% of the distribution are
20-200 ijK^ and 5-450 /zK^ respectively. This is effec-
tively a calculation of the primary CMB prior for the SZ power
spectrum at the CBI band. When compared to the quoted 68%
(359-624 /zK^) and 95% (199-946 /xK^) regions of flie CBI
detection at face value, we find that the CBI detection is only
marginally consistent with the SZ effect. The BIMA detection
is fully consistent with the SZ effect.

By doing a careful statistical analysis, taking into account
the non-Gaussianity of SZ fluctuations and sampling variance,
we find that an SZ interpretation of the CBI data is inconsis-
tent at more than 2— 2.5cr for those cosmological models which
are most consistent with the primary CMB data. The mod-
els with the highest SZ power share several characteristics: all
models with more than 400 /xK^ of power but a few exceptions
have 0.95 < erg and 0.05 < Jlh- Furthermore, such models pre-
fer a low Hubble constant {h < 0.7) and a high matter density
(0.3 < ricDM + fib < 0.8). A reason for a low h is that the pri-
mary CMB data tightly constrain il^h^ while we need a higher
flbh to make the SZ effect larger. We can accomplish this by

4

HOLDER

increasing while reducing h slightly within the HST-h prior.
A model with large red tilt (ng < 0.95) can not produce a large
amount of SZ power.

The possible presence of radio point sources that partially
"fill in" the SZ effect from clusters (Holder 2002) further en-
hances the possible discrepancy. A rehable determination of ag
significantly lower than 1 would be very difficult to reconcile
with an SZ interpretation of the measured high-/ power.

Uncertainty in estimates for the SZ power spectrum is not yet
fully understood, making a more detailed interpretation of the
excess power comphcated. A major issue is that different sim-
ulations have not yet converged, even for adiabatic simulations.
Although our current knowledge of missing physics such as gas
cooling, star formation, or energy feedback is still limited, these
effects appear to not be very important (White et al. 2002) ex-
cept on very small angular scales (< 1' or / > 10"^) where other
effects like non-sphericity or merging of halos may also play a
role. Nevertheless, the current differences among analytic mod-
els, simulations, and estimates of the effects of missing physics
are at the level of a factor of two in Q, while the discrepancy
between these predictions and the CBI data is about a factor of
five.

Given the strong dependence of Q on erg, we argue that the
discrepancy is due to the difference between a low ag < 0.8 fa-
vored by the primary CMB and a high trg > 1 favored by the
SZ effect (Bond et al. 2002; Komatsu and Seljak 2002). Lahav
et al. (2002) and Melchiorri and Silk (2002) have found simi-
larly low (jg from the primary CMB data with the same prior on
h. If the excess power is really due to the SZ effect, then this dis-
crepancy is suggesting that there are some missing components
in our analysis. Multi-band SZ observations covering several
frequencies will be required to verify the apparent discrepancy.

What is missing in our analysis? The chain that we have
used has a strong HST prior on h. Lower values of h reduce
the discrepancy, but h <0A would be required to explain the
entire difference. Additional components such as massive neu-
trinos would make the discrepancy worse by driving erg to even
lower values. Allowing tensor modes will have competing ef-
fects of reducing the overall normalization of scalar modes at
large scales but also allowing a blue tilt (higher n,), leaving the
effects on cluster scales largely unchanged. The effect of al-
lowing a general equation of state for the dark energy will be to
slightly enhance the SZ fluctuations for a fixed value of erg (Ko-
matsu and Seljak 2002), but to significantly reduce the CMB-
pref erred value of erg. With the SZ fluctuation power going as

i7g, the latter effect will dominate and w > -1 will generally
reduce the expected fluctuation power. The addition of isocur-
vature fluctuations may help to reconcile the discrepancy, while
it significantly expands the allowed range of many cosmologi-
cal parameters (Trotta et al. 2001).

A very early reionization of the universe (zre > 20) will in-
crease the CMB-pref erred value of erg, helping to reduce the
discrepancy. We find a broad peak in the distribution of mod-
els in the chain around cge"^ ~ 0.8, where r is the Thomson-
scattering optical depth of the universe; r > 0.22 or Zk 20
would comfortably allow ag > 1.0. The discrepancy thus dis-
appears if the universe was reionized early. The current CMB
data cannot break the degeneracy between erg and t, and allow
this area of parameter space, although r > 0.3 (zre ^ 30) ap-
pears to be ruled out. CMB-polarization experiments on large
angular scales (e.g., MAP or Planck) should be able to break
this degeneracy, and detect the signature of reionization (Zal-
darriaga et al. 1997; Eisenstein et al. 1999; KapUnghat et al.
2002).

We have presented an example of the ease and power of
MCMC methods in applying CMB constraints to calculations
that include non-trivial dependence on cosmological parame-
ters. It would be worthwhile to investigate the effects of LSS
priors, which should make the discrepancy worse by making ag
smaller (Lewis and Bridle 2002; Bond et al. 2002). With the
strong preference of the high-/ measurements for high values
of erg and flbh, it would require running a new chain that in-
cludes the CBI and BIMA data points, as importance sampling
is unrehable in the tails of the current distribution.

Measurements of CMB anisotropy will continue to improve.
MCMC methods provide a natural way to incorporate strong
constraints on cosmological parameters from CMB experi-
ments into other cosmological studies. Using CMB information
to understand the effects of cosmology will allow better under-
standings of systematic errors in the measurements and insight
into important astrophysical processes.

Many of the calculations and some of the text were gener-
ously provided by Eiichiro Komatsu, of Princeton University,
who could not be a co-author due to considerations of a pos-
sible conflict of interest but was an important contributor. We
would like to thank David N. Spergel and Uros Seljak for useful
discussions. GPH is supported by the W. M. Keck Foundation
at the IAS. We are grateful to Sarah Bridle and Antony Lewis
for generously providing their MCMC results publicly.

REFERENCES

Bond, J. R., et al. 2002, ApJ, submitted (astro-ph/0205386)
Buries, S., NoUett, K. M., and Turner, M. S. 2001, ApJ, 552, LI
Christensen, N., Meyer, R., Knox, L., and Luey, B. 2001, Class. Quant. Grav.,
18, 2677

Cole, S., and Kaiser, N. 1988, MNRAS, 233, 637
Cooray, A. 2001, Phys. Rev. D, 64, 063514.

Dawson, K. S., Holzapfel, W. L., Carlstrom, J. E., LaRoque, S. J., Miller, A.,
Nagai, D., and Joy, M. 2002, ApJ, submitted (astro-ph/0206012)

Eisenstein, D. J., Hu, W., and Tegmark, M. 1999, ApJ, 518, 2-23.

Freedman, W. L., et al. 2001, ApJ, 553, 47

Frenk, C. S., et al. 1999, ApJ, 525, 554

Holder, G. R 2002, ApJ, in press (astro-ph/0205467)

Hu, W., and Dodelson, S. 2002, ARA&A, in press (astro-ph/01 10414)

Jenkins, A., Frenk, C. S., White, S. D. M., Colberg, J. M., Cole, S., Evrard, A.
E., Couchman, H. M. R, and Yoshida, N. MNRAS, 321, 372

KapWghat, M., Chu, M., Haiman, Z., Holder, G. P., Knox, L., and Skordis, C.
ApJ, submitted (asB-o-ph/0207591)

Komatsu, E., and Kitayama, T. 1999, ApJ, 526, LI

Komatsu, E., and Seljak, U. 2001, MNRAS, 327, 1353

Komatsu, E., and Seljak, U. 2002, MNRAS, submitted (asn-o-ph/0205468)

Lahav, O., et al. 2002, MNRAS, 333, 961

Lewis, A., and Bridle, S. 2002, Phys. Rev. D, submitted (astro-ph/0205436)
Mason, B. S., et al. 2002, ApJ, submitted (astro-ph/0205384)
Melchiorri, A., and Silk, J. 2002, preprint (astro-ph/0203200)
Navarro, J. F, Frenk, C. S., and White, S. D. M. 1997, ApJ, 490, 493
Seljak, U., Burwell, J., and Pen, U. L. 2001, Phys. Rev. D, 63, 063001
Sunyaev, R. A., and Zel'dovich, Y. B. 1980, ARA&A, 18, 537
Trotta, R., Riazuelo, A., and Durrer, R. 2001, Phys. Rev. Lett., 87, 231301
Wang, X., Tegmark, M., and Zaldarriaga, M. 2002, Phys. Rev D, 65, 123001
White, M., Hemquist, L., and Springel, V. 2002, ApJ, submitted

(astt-o-ph/0205437)
Zaldarriaga, M., Spergel, D. N., and Seljak, U. 1997, ApJ, 488, 1
Zhang, R, Pen, U. L., and Wang, B. 2002, ApJ, submitted (astro-ph/0201375)