6 D FGalaxy Survey BAOandthelocal Hconstant

Mon. Not. R. Astron. Soc. 416, 3017–3032 (2011) doi:10.1111/j.1365-2966.2011.19250.
The 6dF Galaxy Survey: baryon acoustic oscillations and the local Hubble
constant
Florian Beutler,1 Chris Blake,2 Matthew Colless,3 D. Heath Jones,4

Lister Staveley-Smith,1 Lachlan Campbell,5 Quentin Parker,3,6 Will Saunders3
and Fred Watson3
1 International Centre for Radio Astronomy Research (ICRAR), University of Western Australia, 35 Stirling Highway, Perth, WA 6009, Australia
2 Centre for Astrophysics & Supercomputing, Swinburne University of Technology, PO Box 218, Hawthorn, VIC 3122, Australia
3 Australian Astronomical Observatory, PO Box 296, Epping, NSW 1710, Australia
4 School of Physics, Monash University, Clayton, VIC 3800, Australia
5 Western Kentucky University, Bowling Green, KY 42101 USA
6 Department of Physics and Astronomy, Faculty of Sciences, Macquarie University, Sydney, NSW 2109 Australia
Accepted 2011 June 14. Received 2011 May 25; in original form 2011 March 29
ABSTRACT
We analyse the large-scale correlation function of the 6dF Galaxy Survey (6dFGS) and detect
a baryon acoustic oscillation (BAO) signal at 105 h−1 Mpc. The 6dFGS BAO detection allows
us to constrain the distance–redshift relation at zeff = 0.106. We achieve a distance measure of
DV (zeff ) = 457 ± 27 Mpc and a measurement of the distance ratio, rs (zd )/DV (zeff ) = 0.336 ±
0.015 (4.5 per cent precision), where rs (zd ) is the sound horizon at the drag epoch zd . The low-
effective redshift of 6dFGS makes it a competitive and independent alternative to Cepheids
and low-z supernovae in constraining the Hubble constant. We find a Hubble constant of H 0 =
67 ± 3.2 km s−1 Mpc−1 (4.8 per cent precision) that depends only on the Wilkinson Microwave
Anisotropy Probe-7 (WMAP-7) calibration of the sound horizon and on the galaxy clustering
in 6dFGS. Compared to earlier BAO studies at higher redshift, our analysis is less dependent
on other cosmological parameters. The sensitivity to H 0 can be used to break the degeneracy
between the dark energy equation of state parameter w and H 0 in the cosmic microwave
background data. We determine that w = −0.97 ± 0.13, using only WMAP-7 and BAO data
from both 6dFGS and Percival et al. (2010).
We also discuss predictions for the large-scale correlation function of two future wide-angle
surveys: the Wide field ASKAP L-band Legacy All-sky Blind surveY (WALLABY) blind H I
survey (with the Australian Square Kilometre Array Pathfinder, ASKAP) and the proposed
Transforming Astronomical Imaging surveys through Polychromatic Analysis of Nebulae
(TAIPAN) all-southern-sky optical galaxy survey with the UK Schmidt Telescope. We find
that both surveys are very likely to yield detections of the BAO peak, making WALLABY the
first radio galaxy survey to do so. We also predict that TAIPAN has the potential to constrain
the Hubble constant with 3 per cent precision.
Key words: surveys – cosmology: observations – dark energy – distance scale – large-scale
structure of Universe.
to radiation through Thomson scattering. The radiation pressure

1 I N T RO D U C T I O N
drives sound waves originating from overdensities in the matter
The current standard cosmological model, cold dark matter distribution (Peebles & Yu 1970; Sunyaev & Zeldovich 1970; Bond
(CDM), assumes that the initial fluctuations in the distribution & Efstathiou 1987). At the time of recombination (z∗ ≈ 1090),
of matter were seeded by quantum fluctuations pushed to cosmo- the photons decouple from the baryons and shortly after that (at
logical scales by inflation. Directly after inflation, the Universe is the baryon drag epoch zd ≈ 1020) the sound wave stalls. Through
radiation-dominated and the baryonic matter is ionized and coupled this process, each overdensity of the original density perturbation
field has evolved to become a centrally peaked perturbation sur-
rounded by a spherical shell (Bashinsky & Bertschinger 2001, 2002;
E-mail: [email protected] Eisenstein, Seo & White 2007a). The radius of these shells is called

C 2011 The Authors
Monthly
Downloaded from Notices of the Royal Astronomical Society
C 2011 RAS
https://academic.oup.com/mnras/article-abstract/416/4/3017/976636
by FAC.MED.RIB.PRETO-BIBL.CENTRAL-USP user
on 05 June 2018
3018 F. Beutler et al.
the sound horizon rs . Both overdense regions attract baryons and of photometric redshift samples (Blake et al. 2007; Padmanabhan
dark matter and will be preferred regions of galaxy formation. This et al. 2007) and a cluster sample based on the SDSS photometric
process can equivalently be described in Fourier space, where dur- data (Huetsi 2009). Several years earlier, a study by Miller, Nichol &
ing the photon–baryon coupling phase, the amplitude of the baryon Batuski (2001) found first hints of the BAO feature in a combination
perturbations cannot grow and instead undergo harmonic motion of smaller data sets.
leading to an oscillation pattern in the power spectrum. Low-redshift distance measurements can directly measure the
After the time of recombination, the mean free path of photons Hubble constant H 0 with a relatively weak dependence on other
increases and becomes larger than the Hubble distance. Hence from cosmological parameters such as the dark energy equation of state
now on the radiation remains almost undisturbed, eventually be- parameter w. The 6dF Galaxy Survey (6dFGS) is the biggest galaxy
coming the cosmic microwave background (CMB). survey in the local Universe, covering almost half the sky. If 6dFGS
The CMB is a powerful probe of cosmology due to the good could be used to constrain the redshift–distance relation through
theoretical understanding of the physical processes described above. BAOs, such a measurement could directly determine the Hubble
The size of the sound horizon depends (to first order) only on the constant, depending only on the calibration of the sound horizon
sound speed in the early Universe and the age of the Universe at through the matter and baryon density. The objective of the present
recombination, both set by the physical matter and baryon densities, paper is to measure the two-point correlation function on large
m h2 and b h2 (Eisenstein & Hu 1998). Hence, measuring the scales for the 6dFGS and extract the BAO signal.
sound horizon in the CMB gives extremely accurate constraints Many cosmological parameter studies add a prior on H 0 to help
on these quantities (Komatsu et al. 2011). Measurements of other break degeneracies. The 6dFGS derivation of H 0 can provide an al-
cosmological parameters often show degeneracies in the CMB data ternative source of that prior. The 6dFGS H 0 -measurement can also
alone (Efstathiou & Bond 1999), especially in models with extra be used as a consistency check of other low-redshift distance cali-
parameters beyond flat CDM. Combining low-redshift data with brators such as Cepheid variables and Type Ia supernovae (through
the CMB can break these degeneracies. the so-called distance ladder technique; see e.g. Freedman et al.
Within galaxy redshift surveys, we can use the correlation func- 2001; Riess et al. 2011). Compared to these more classical probes
tion, ξ , to quantify the clustering on different scales. The sound of the Hubble constant, the BAO analysis has an advantage of sim-
horizon marks a preferred separation of galaxies and hence predicts plicity, depending only on m h2 and b h2 from the CMB and the
a peak in the correlation function at the corresponding scale. The sound horizon measurement in the correlation function, with small
expected enhancement at s = rs is only ξ ≈ 10−3 b2 (1 + 2β/3 systematic uncertainties.
+ β 2 /5) in the galaxy correlation function, where b is the galaxy Another motivation for our study is that the SDSS data after
bias compared to the matter correlation function, and β accounts data release 3 (DR3) show more correlation on large scales than
for linear redshift-space distortions. Since the signal appears at very expected by CDM, and have no sign of a crossover to negative ξ
large scales, it is necessary to probe a large volume of the Universe up to 200 h−1 Mpc (the CDM prediction is 140 h−1 Mpc) (Kazin
to decrease sample variance, which dominates the error on these et al. 2010a). It could be that the LRG sample is a rather unusual
scales (Tegmark 1997; Eisenstein, Hu & Tegmark 1998; Goldberg realization, and the additional power just reflects sample variance.
& Strauss 1998). It is interesting to test the location of the crossover scale in another
Very interesting for cosmology is the idea of using the sound red galaxy sample at a different redshift.
horizon scale as a standard ruler (Eisenstein et al. 1998; Cooray This paper is organized as follows. In Section 2 we introduce
et al. 2001; Blake & Glazebrook 2003; Seo & Eisenstein 2003). the 6dFGS survey and the K band selected subsample used in this
A standard ruler is a feature whose absolute size is known. By analysis. In Section 3 we explain the technique we apply to derive
measuring its apparent size, one can determine its distance from the correlation function and summarize our error estimate, which
the observer. The baryon acoustic oscillation (BAO) signal can is based on lognormal realizations. In Section 4 we discuss the
be measured in the matter distribution at low redshift, with the need for wide-angle corrections and several linear and non-linear
CMB calibrating the absolute size, and hence the distance–redshift effects which influence our measurement. Based on this discussion,
relation can be mapped (see e.g. Bassett & Hlozek 2009, for a we introduce our correlation function model. In Section 5 we fit the
summary). data and derive the distance estimate DV (zeff ). In Section 6 we derive
The Sloan Digital Sky Survey (SDSS; York et al. 2000) and the the Hubble constant and constraints on dark energy. In Section 7 we
2dF Galaxy Redshift Survey (2dFGRS; Colless et al. 2001) were the discuss the significance of the BAO detection of 6dFGS. In Section 8
first redshift surveys which have directly detected the BAO signal. we give a short overview of future all-sky surveys and their power
Recently, the WiggleZ Dark Energy Survey has reported a BAO to measure the Hubble constant. We conclude and summarize our
measurement at redshift z = 0.6 (Blake et al. 2011). results in Section 9.
Eisenstein et al. (2005) were able to constrain the distance– Throughout the paper, we use r to denote real space separations
redshift relation to 5 per cent accuracy at an effective redshift and s to denote separations in redshift space. Our fiducial model
of zeff = 0.35 using an early data release of the SDSS luminous assumes a flat universe with fid m = 0.27, w
fid
= −1 and fid
k = 0.
red galaxy (LRG) sample containing ≈47 000 galaxies. Subsequent The Hubble constant is set to H 0 = 100h km s−1 Mpc−1 , with our
studies using the final SDSS-LRG sample and combining it with fiducial model using hfid = 0.7.
the SDSS-main and the 2dFGRS sample were able to improve on
this measurement and constrain the distance–redshift relation at zeff
2 T H E 6 D F G A L A X Y S U RV E Y
= 0.2 and 0.35 with 3 per cent accuracy (Percival et al. 2010). Other
studies of the same data found similar results using the correlation
2.1 Targets and selection function
function ξ (s) (Gaztanaga, Cabre & Hui 2009; Labini et al. 2009;
Martinez et al. 2009; Sanchez et al. 2009; Kazin et al. 2010a), the The galaxies used in this analysis were selected to K ≤ 12.9 from the
power spectrum P(k) (Cole et al. 2005; Huetsi 2006; Tegmark et al. Two Micron All Sky Survey Extended Source Catalogue (2MASS
2006; Reid et al. 2010), the projected correlation function w(rp ) XSC; Jarrett et al. 2000) and combined with redshift data from
C 2011 The Authors, MNRAS 416, 3017–3032
Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

Monthly Notices of the Royal Astronomical Society
C 2011 RAS
on 05 June 2018
6dFGS: BAOs and the local Hubble constant 3019
the 6dFGS (Jones et al. 2009). The 6dFGS is a combined redshift
and peculiar velocity survey covering nearly the entire southern sky
with |b| < 10◦ . It was undertaken with the Six-Degree Field (6dF)
multifibre instrument on the UK Schmidt Telescope from 2001 to
2006. The median redshift of the survey is z = 0.052 and the 25:
50: 75 per cent percentile completeness values are 0.61: 0.79: 0.92.
Papers by Jones et al. (2004, 2006, 2009) describe 6dFGS in full
detail, including comparisons between 6dFGS, 2dFGRS and SDSS.
Galaxies were excluded from our sample if they resided in sky
regions with completeness lower than 60 per cent. After applying
these cuts, our sample contains 75 117 galaxies. The selection func-
tion was derived by scaling the survey completeness as a function
of magnitude to match the integrated on-sky completeness, using
mean galaxy counts. This method is the same adopted by Colless
et al. (2001) for 2dFGRS and is explained in Jones et al. (2006) in
detail. The redshift of each object was checked visually and care Figure 1. Redshift distribution of the data (black solid line) and the random
was taken to exclude foreground Galactic sources. The derived com- catalogue (black dashed line). The weighted distribution (using weights
pleteness function was used in the real galaxy catalogue to weight from equation 6) is shifted to higher redshift and has increased shot noise,
but a smaller error due to sample variance (blue solid and dashed lines).
each galaxy by its inverse completeness. The completeness func-
tion was also applied to the mock galaxy catalogues to mimic the
selection characteristics of the survey. Jones et al. (in preparation)
describe the derivation of the 6dFGS selection function, and inter- 3.1 Random catalogues
ested readers are referred to this paper for a more comprehensive To calculate the correlation function, we need a random sample of
treatment. galaxies which follows the same angular and redshift selection func-
tion as the 6dFGS sample. We base our random catalogue generation
on the 6dFGS luminosity function of Jones et al. (in preparation),
where we use random numbers to pick volume-weighted redshifts
2.2 Survey volume
and luminosity-function-weighted absolute magnitudes. We then
We calculated the effective volume of the survey using the estimate test whether the redshift–magnitude combination falls within the
of Tegmark (1997): 6dFGS K-band faint and bright apparent magnitude limits (8.75 ≤
2 K ≤ 12.9).
n(x)P0 Fig. 1 shows the redshift distribution of the 6dFGS K-selected
Veff = d3 x , (1)
1 + n(x)P0 sample (black solid line) compared to a random catalogue with the
same number of galaxies (black dashed line). The random catalogue
where n(x) is the mean galaxy density at position x, determined is a good description of the 6dFGS redshift distribution in both the
from the data, and P0 is the characteristic power spectrum am- weighted and unweighted case.
plitude of the BAO signal. The parameter P0 is crucial for the
weighting scheme introduced later. We find that the value of P0 =
40 000 h−3 Mpc3 (corresponding to the value of the galaxy power 3.2 The correlation function
spectrum at k ≈ 0.06 h Mpc−1 in 6dFGS) minimizes the error of the
We turn the measured redshift into comoving distance via
correlation function near the BAO peak. z
Using P0 = 40 000 h−3 Mpc3 yields an effective volume of c dz
DC (z) = (2)
0.08 h−3 Gpc3 , while using instead P0 = 10 000 h−3 Mpc3 (cor- H0 0 E(z )
responding to k ≈ 0.15h Mpc−1 ) gives an effective volume of
with
0.045 h−3 Gpc3 .
The volume of the 6dFGS is approximately as large as the volume E(z) = fid
m (1 + z) + k (1 + z)
3 fid 2
covered by the 2dFGRS, with a sample density similar to SDSS-

3(1+wfid ) 1/2
(3)
+ fid
(1 + z) ) ,
DR7 (Abazajian et al. 2009). Percival et al. (2010) reported success-
ful BAO detections in several samples obtained from a combination where the curvature fidk is set to zero, the dark energy density is
of SDSS DR7, SDSS-LRG and 2dFGRS with effective volumes in = 1 − m and the equation of state for dark energy
given by fid fid
the range 0.15–0.45 h−3 Gpc3 (using P0 = 10 000 h−3 Mpc3 ), while is wfid = −1. Because of the very low redshift of 6dFGS, our data
the original detection by Eisenstein et al. (2005) used a sample with are not very sensitive to k , w or any other higher dimensional
V eff = 0.38 h−3 Gpc3 (using P0 = 40 000 h−3 Mpc3 ). parameter which influences the expansion history of the Universe.
We will discuss this further in Section 5.3.
Now we measure the separation between all galaxy pairs in our
survey and count the number of such pairs in each separation bin. We
3 CLUSTERING MEASUREMENT
do this for the 6dFGS data catalogue, a random catalogue with the
We focus our analysis on the two-point correlation function. In the same selection function and a combination of data–random pairs.
following subsections we introduce the technique used to estimate We call the pair-separation distributions obtained from this analysis
the correlation function and outline the method of lognormal real- DD(s), RR(s) and DR(s), respectively. The binning is chosen to be
izations, which we employed to derive a covariance matrix for our from 10 h−1 Mpc up to 190 h−1 Mpc, in 10 h−1 Mpc steps. In the
measurement. analysis, we used 30 random catalogues with the same size as the

Downloaded
C 2011 RAS
from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636
on 05 June 2018
data catalogue. The redshift correlation function itself is given by
Landy & Szalay (1993):

DD(s) nr 2 DR(s) nr
ξ (s) = 1 +
data
−2 , (4)
RR(s) nd RR(s) nd
where the ratio nr /nd is given by
Nr
nr wi (x)
= Ni d , (5)
nd j w j (x)
and the sums go over all random (N r ) and data (N d ) galaxies. We

use the inverse density weighting of Feldman, Kaiser & Peacock
(1994):
Ci
wi (x) = , (6)
1 + n(x)P0
with P0 = 40 000 h3 Mpc−3 and Ci being the inverse completeness
Figure 2. The large-scale correlation function of 6dFGS. The best-fitting
weighting for 6dFGS (see Section 2.1 and Jones et al., in prepa-
model is shown by the black line with the best-fitting value of m h2 =
ration). This weighting is designed to minimize the error on the
0.135 ± 0.020. Models with different m h2 are shown by the green line
BAO measurement, and since our sample is strongly limited by (m h2 = 0.12) and the blue line (m h2 = 0.15). The red line is a linear
sample variance on large scales, this weighting results in a signif- CDM model with b h2 = 0 (and m h2 = 0.1), while all other models
icant improvement to the analysis. The effect of the weighting on use the WMAP-7 best-fitting value of b h2 = 0.022 27 (Komatsu et al.
the redshift distribution is illustrated in Fig. 1. 2011). The significance of the BAO detection in the black line relative to
Other authors have used the so-called J 3 -weighting which opti- the red one is 2.4σ (see Section 7). The error bars at the data points are the
mizes the error over all scales by weighting each scale differently diagonal elements of the covariance matrix derived using lognormal mock
(e.g. Efstathiou 1988; Loveday et al. 1995). In a magnitude lim- catalogues.
ited sample, there is a correlation between luminosity and redshift,
which establishes a correlation between bias and redshift (Zehavi
assuming Gaussian fluctuations. This density field is then Poisson
et al. 2005). A scale-dependent weighting would imply a different
sampled, taking into account the window function and the total
effective redshift for each scale, causing a scale-dependent bias.
number of galaxies. The assumption that the input power spectrum
Finally, we considered a luminosity-dependent weighting as sug-
has Gaussian fluctuations can only be used in a model for a density
gested by Percival, Verde & Peacock (2004). However, the same
field with overdensities 1. As soon as we start to deal with finite
authors found that explicitly accounting for the luminosity–redshift
rms fluctuations, the Gaussian model assigns a non-zero probability
relation has a negligible effect for 2dFGRS. We found that the effect
to regions of negative density. A lognormal random field LN(x), can
to the 6dFGS correlation function is 1σ for all bins. Hence the
avoid this unphysical behaviour. It is obtained from a Gaussian field
static weighting of equation (6) is sufficient for our data set.
G(x) by
We also include an integral constraint correction in the form of
ξ final (s) = ξ data (s) + ic, (7) LN(x) = exp[G(x)] (9)
where ic is defined as
which is positive-definite but approaches 1 + G(x) whenever the
s RR(s)ξ (s) perturbations are small (e.g. at large scales). Calculating the power
ic = . (8)
s RR(s)
spectrum of a Poisson sampled density field with such a distribution
will reproduce the input power spectrum convolved with the window
The function RR(s) is calculated from our mock catalogue, and ξ (s)
function. As an input power spectrum for the lognormal field, we
is a correlation function model. Since ic depends on the model of
use
the correlation function, we have to recalculate it at each step during
the fitting procedure. However, we note that ic has no significant Pnl (k) = APlin (k) exp[−(k/k∗ )2 ] (10)
impact to the final result.
Fig. 2 shows the correlation function of 6dFGS at large scales. where A = b2 (1 + 2β/3 + β 2 /5) accounts for the linear bias and
The BAO peak at 105 h−1 Mpc is clearly visible. The plot includes the linear redshift-space distortions. Plin (k) is a linear model power
model predictions of different cosmological parameter sets. We will spectrum in real space obtained from CAMB (Lewis, Challinor &
discuss these models in Section 5.2. Lasenby 2000) and Pnl (k) is the non-linear power spectrum in red-
shift space. Comparing the model above with the 6dFGS data gives
A = 4. The damping parameter k∗ is set to k∗ = 0.33h Mpc−1 , as
3.3 Lognormal error estimate
found in 6dFGS (see fitting results later). How well this input model
To obtain reliable error bars for the correlation function, we use matches the 6dFGS data can be seen in Fig. 9.
lognormal realizations (Coles & Jones 1991; Cole et al. 2005; We produce 200 such realizations and calculate the correlation
Kitaura, Jasche & Metcalf 2009). In what follows, we summa- function for each of them, deriving a covariance matrix
rize the main steps, but refer the interested reader to Appendix A in
which we give a detailed explanation of how we generate the lognor-
N
[ξn (si ) − ξ (si )][ξn (sj ) − ξ (sj )]
Cij = . (11)
mal mock catalogues. In Appendix B, we compare the lognormal
n=1
N −1
errors with jackknife estimates.
Lognormal realizations of a galaxy survey are usually obtained Here, ξ n (si ) is the correlation function estimate at separation si and
by deriving a density field from a model power spectrum, P(k), the sum goes over all N lognormal realizations. The mean value is

C 2011 RAS
on 05 June 2018
We find that the wide-angle corrections have only a very minor
effect on our sample. For our fiducial model, we found a correc-
tion of ξ = 4 × 10−4 in amplitude at 100 h−1 Mpc and ξ =
4.5 × 10−4 at 200 h−1 Mpc, (Fig. C2 in the appendix). This is much
smaller than the error bars on these scales. Despite the small size of
the effect, we nevertheless include all first-order correction terms
in our correlation function model. It is important to note that wide-
angle corrections affect the correlation function amplitude only and
do not cause any shift in the BAO scale. The effect of the wide-angle
correction on the unweighted sample is much greater and is already
noticeable on scales of 20 h−1 Mpc. Weighting to higher redshifts
mitigates the effect because it reduces the average opening angle
between galaxy pairs, by giving less weight to wide-angle pairs (on
average).
4.2 Non-linear effects

There are a number of non-linear effects which can potentially
Figure 3. Correlation matrix derived from a covariance matrix calculated influence a measurement of the BAO signal. These include scale-
from 200 lognormal realizations. dependent bias, the non-linear growth of structure on smaller scales,
and redshift-space distortions. We discuss each of these in the con-
defined as text of our 6dFGS sample.
As the Universe evolves, the acoustic signature in the corre-
1
N
ξ (si ) = ξn (si ). (12) lation function is broadened by non-linear gravitational structure
N n=1 formation. Equivalently, we can say that the higher harmonics in
the power spectrum, which represent smaller scales, are erased
The case i = j gives the error (ignoring correlations between bins,
(Eisenstein et al. 2007a).
σ i2 = Cii ). In the following, we will use this uncertainty in all
The early Universe physics, which we discussed briefly in the
diagrams, while the fitting procedures use the full covariance matrix.
introduction, is well understood and several authors have produced
The distribution of recovered correlation functions includes the
software packages (e.g. CMBFAST and CAMB) and published fitting
effects of sample variance and shot noise. Non-linearities are also
functions (e.g. Eisenstein & Hu 1998) to make predictions for the
approximately included since the distribution of overdensities is
correlation function and power spectrum using thermodynamical
skewed.
models of the early Universe. These models already include the
In Fig. 3 we show the lognormal correlation matrix rij calculated
basic linear physics most relevant for the BAO peak. In our anal-
from the covariance matrix. The correlation matrix is defined as
ysis, we use the CAMB software package (Lewis et al. 2000). The
Cij non-linear evolution of the power spectrum in CAMB is calculated
rij = , (13)
Cii Cjj using the HALOFIT code (Smith et al. 2003). This code is calibrated
by N-body simulations and can describe non-linear effects in the
where C is the covariance matrix (for a comparison to jackknife
shape of the matter power spectrum for pure CDM models to an
errors, see appendix B).
accuracy of around 5–10 per cent (Heitmann et al. 2010). How-
ever, it has previously been shown that this non-linear model is
4 M O D E L L I N G T H E BAO S I G N A L a poor description of the non-linear effects around the BAO peak
(Crocce & Scoccimarro 2008). We therefore decided to use the lin-
In this section we will discuss wide-angle effects and non-linearities. ear model output from CAMB and incorporate the non-linear effects
We also introduce a model for the large-scale correlation function, separately.
which we later use to fit our data. All non-linear effects influencing the correlation function can
be approximated by a convolution with a Gaussian damping factor
exp[−(rk∗ /2)2 ] (Eisenstein et al. 2007b,a), where k∗ is the damping
4.1 Wide-angle formalism
scale. We will use this factor in our correlation function model intro-
The model of linear redshift-space distortions introduced by Kaiser duced in the next section. The convolution with a Gaussian causes
(1987) is based on the plane-parallel approximation. Earlier sur- a shift of the peak position to larger scales, since the correlation
veys such as SDSS and 2dFGRS are at sufficiently high redshift function is not symmetric around the peak. However this shift is
that the maximum opening angle between a galaxy pair remains usually very small.
small enough to ensure the plane-parallel approximation is valid. All of the non-linear processes discussed so far are not at the
However, the 6dFGS has a maximum opening angle of 180◦ and a fundamental scale of 105 h−1 Mpc, but are instead at the cluster-
lower mean redshift of z ≈ 0.1 (for our weighted sample) and so formation scale of up to 10 h−1 Mpc. The scale of 105 h−1 Mpc is
it is necessary to test the validity of the plane-parallel approxima- far larger than any known non-linear effect in cosmology. This has
tion. The wide-angle description of redshift-space distortions has led some authors to the conclusion that the peak will not be shifted
been laid out in several papers (Szalay, Matsubara & Landy 1997; significantly, but rather only blurred out. For example, Eisenstein
Matsubara 2004; Szapudi 2004; Papai & Szapudi 2008; Raccanelli, et al. (2007a) have argued that any systematic shift of the acoustic
Samushia & Percival 2010), which we summarize in Appendix C. scale in real space must be small (0.5 per cent), even at z = 0.

Downloaded
C 2011 RAS
on 05 June 2018
However, several authors report possible shifts of up to 1 per cent where Plin (k) is again the linear power spectrum. CDM predicts a
(Guzik & Bernstein 2007; Smith, Scoccimarro & Sheth 2007, 2008; value of k∗ 0.17 h Mpc−1 . However, we will include k∗ as a free
Angulo et al. 2008). Crocce & Scoccimarro (2008) used renormal- fitting parameter.
ized perturbation theory and found per cent level shifts in the BAO The scale dependence of the 6dFGS bias, B(s), is derived from
peak. In addition to non-linear evolution, they found that mode the GiggleZ simulation (Poole et al., in preparation); a dark mat-
coupling generates additional oscillations in the power spectrum, ter simulation containing 21603 particles in a 1 h−1 Gpc box. We
which are out of phase with the BAO oscillations predicted by lin- rank-order the haloes of this simulation by V max and choose a con-
ear theory. This leads to shifts in the scale of oscillation nodes with tiguous set of 250 000 of them, selected to have the same clustering
respect to a smooth spectrum. In real space, this corresponds to a amplitude of 6dFGS as quantified by the separation scale r0 , where
peak shift towards smaller scales. Based on their results, Crocce & ξ (r0 ) = 1. In the case of 6dFGS, we found r0 = 9.3 h−1 Mpc.
Scoccimarro (2008) propose a model to be used for the correlation Using the redshift-space correlation function of these haloes and
function analysis at large scales. We will introduce this model in of a randomly subsampled set of ∼106 dark matter particles, we
the next section. obtain
B(s) = 1 + (s/0.474 h−1 Mpc)−1.332 , (20)

4.3 Large-scale correlation function
which describes a 1.7 per cent correction of the correlation function
To model the correlation function on large scales, we follow Crocce amplitude at separation scales of 10 h−1 Mpc. To derive this func-
& Scoccimarro (2008) and Sanchez, Baugh & Angulo (2008), and tion, the GiggleZ correlation function (snapshot z = 0) has been
adopt the following parametrization:1 fitted down to 6 h−1 Mpc, well below the smallest scales we are

ξ model (s) = B(s)b2 ξ (s) ∗ G(r) + ξ11 (r)ξ (s) . (14) interested in.
Here, we decouple the scale dependency of the bias B(s) and the
linear bias b. G(r) is a Gaussian damping term, accounting for
5 E X T R AC T I N G T H E BAO S I G N A L
non-linear suppression of the BAO signal. ξ (s) is the linear corre-
lation function (including wide-angle description of redshift-space In this section, we fit the model correlation function developed in
distortions; equation (C4) in the appendix). The second term in the previous section to our data. Such a fit can be used to derive the
equation (14) accounts for the mode coupling of different Fourier distance scale DV (zeff ) at the effective redshift of the survey.
modes. It contains ξ (s), which is the first derivative of the redshift-
space correlation function, and ξ 11 (r), which is defined as
∞ 5.1 Fitting preparation
1
ξ11 (r) = dk kPlin (k)j1 (rk), (15)
2π2 0 The effective redshift of our sample is determined by
with j1 (x) being the spherical Bessel function of order 1. Sanchez Nb Nb

wi wj
et al. (2008) used an additional parameter AMC which multiplies zeff = (zi + zj ), (21)
the mode-coupling term in equation (14). We found that our data i j
2Nb2
are not good enough to constrain this parameter, and hence adopted
AMC = 1 as in the original model by Crocce & Scoccimarro (2008). where N b is the number of galaxies in a particular separation bin
In practice, we generate linear model power spectra Plin (k) from and wi and wj are the weights for those galaxies from equation (6).
CAMB and convert them into a correlation function using a Hankel We choose zeff from bin 10 which has the limits 100 h−1 Mpc and
transform: 110 h−1 Mpc and which gave zeff = 0.106. Other bins show values
∞ very similar to this, with a standard deviation of ±0.001. The final
1
ξ (r) = dk k 2 Plin (k)j0 (rk), (16) result does not depend on a very precise determination of zeff , since
2π2 0
we are not constraining a distance to the mean redshift, but a distance
where j0 (x) = sin (x)/x is the spherical Bessel function of order 0.
ratio [see equation (24), later]. In fact, if the fiducial model is
The ‘∗’ symbol in equation (14) is only equivalent to a convo-
correct, the result is completely independent of zeff . Only if there is
lution in the case of a 3D correlation function, where we have the
a z-dependent deviation from the fiducial model do we need zeff to
Fourier theorem relating the 3D power spectrum to the correlation
quantify this deviation at a specific redshift.
function. In case of the spherically averaged quantities, this is not
Along the line of sight, the BAO signal directly constrains the
true. Hence, the ‘∗’ symbol in our equation stands for the multipli-
Hubble constant H(z) at redshift z. When measured in a redshift
cation of the power spectrum with G̃(k) before transforming it into
shell, it constrains the angular diameter distance DA (z) (Matsub-
a correlation function. G̃(k) is defined as
ara 2004). In order to separately measure DA (z) and H(z), we re-
G̃(k) = exp[−(k/k∗ )2 ], (17) quire a BAO detection in the 2D correlation function, where it
with the property will appear as a ring at around 105 h−1 Mpc. Extremely large vol-
umes are necessary for such a measurement. While there are studies
G̃(k) → 0 as k → ∞. (18) that report a successful (but very low signal-to-noise ratio) detec-
The damping scale k∗ can be calculated from linear theory (Crocce tion in the 2D correlation function using the SDSS-LRG data (e.g.
& Scoccimarro 2006; Matsubara 2008) by Gaztanaga et al. 2009; Chuang & Wang 2011, but see also Kazin
∞ −1/2 et al. 2010b), our sample does not allow this kind of analysis. Hence
1
k∗ = dk P lin (k) , (19) we restrict ourselves to the 1D correlation function, where we mea-
6π2 0
sure a combination of DA (z) and H(z). What we actually measure is a
superposition of two angular measurements (RA and Dec.) and one
1Note that r = s, the different letters just specify whether the function is line-of-sight measurement (redshift). To account for this mixture of
evaluated in redshift space or real space. measurements, it is common to report the BAO distance constraints

C 2011 RAS
on 05 June 2018
as (Eisenstein et al. 2005; Padmanabhan & White 2008)
1/3
cz
DV (z) = (1 + z)2 DA2 (z) , (22)
H0 E(z)
where DA is the angular distance, which in the case of k = 0 is
given by DA (z) = DC (z)/(1 + z).
To derive model power spectra from CAMB, we have to specify
a complete cosmological model, which in the case of the simplest
CDM model (k = 0, w = −1) is specified by six parameters:
ωc , ωb , ns , τ , As and h. These parameters are the physical cold dark
matter and baryon density (ωc = c h2 , ωb = b h2 ), the scalar
spectral index (ns ), the optical depth at recombination (τ ), the scalar
amplitude of the CMB temperature fluctuation (As ) and the Hubble
constant in units of 100 km s−1 Mpc−1 (h).
Our fit uses the parameter values from WMAP-7 (Komatsu et al.
2011): b h2 = 0.022 27, τ = 0.085 and ns = 0.966 (maximum
likelihood values). The scalar amplitude As is set so that it results
in σ 8 = 0.8, which depends on m h2 . However, σ 8 is degenerated
with the bias parameter b which is a free parameter in our fit.
Figure 4. Likelihood contours of the distance DV (zeff ) against m h2 . The
Furthermore, h is set to 0.7 in the fiducial model, but can vary freely
corresponding values of α are given on the right-hand axis. The contours
in our fit through a scale-distortion parameter α, which enters the show 1σ and 2σ errors for both a full fit (blue solid contours) and a fit over
model as 20–190 h−1 Mpc (black dashed contours) excluding the first data point. The
black cross marks the best-fitting values corresponding to the dashed black
ξ final (s) = ξ model (αs). (23)
contours with (DV , m h2 ) = (462, 0.127), while the blue cross marks the
This parameter accounts for deviations from the fiducial cosmolog- best-fitting values for the blue contours. The black solid curve corresponds
ical model, which we use to derive distances from the measured to a constant m h2 DV (zeff ) (DV ∼ h−1 ), while the dashed line corresponds
redshift. It is defined as (Eisenstein et al. 2005; Padmanabhan & to a constant angular size of the sound horizon, as described in the text.
White 2008)
DV (zeff ) V (zeff ) = 440.5 Mpc, where we have followed
fiducial model gives Dfid
α = fid . (24) the distance definitions of Wright (2006) throughout. For each fit we
DV (zeff ) derive the parameter β = m (z)0.545 /b, which we need to calculate
The parameter α enables us to fit the correlation function derived the wide-angle corrections for the correlation function.
with the fiducial model, without the need to recalculate the correla- The maximum likelihood distribution of k∗ seems to prefer
tion function for every new cosmological parameter set. smaller values than predicted by CDM, although we are not able
At low redshift we can approximate H(z) ≈ H 0 , which results in to constrain this parameter very well. This is connected to the high
significance of the BAO peak in the 6dFGS data (see Section 7). A
H0fid
α≈ . (25) smaller value of k∗ damps the BAO peak and weakens the distance
H0 constraint. For comparison, we also performed a fit fixing k∗ to the
Compared to the correct equation (24) this approximation has an CDM prediction of k∗ 0.17 h Mpc−1 . We found that the error
error of about 3 per cent at redshift z = 0.1 for our fiducial model. on the distance DV (zeff ) increases from 5.9 to 8 per cent. However,
Since this is a significant systematic bias, we do not use this ap- since the data do not seem to support such a small value of k∗ , we
proximation at any point in our analysis. prefer to marginalize over this parameter.
The contours of DV (zeff )–m h2 are shown in Fig. 4, together
with two degeneracy predictions (Eisenstein et al. 2005). The solid
5.2 Extracting DV (zeff ) and rs (zd )/DV (zeff )
line is that of constant m h2 DV (zeff ), which gives the direction of
Using the model introduced above we performed fits to 18 data degeneracy for a pure CDM model, where only the shape of the
points between 10 and 190 h−1 Mpc. We excluded the data below correlation function contributes to the fit, without a BAO peak. The
10 h−1 Mpc, since our model for non-linearities is not good enough dashed line corresponds to a constant rs (zd )/DV (zeff ), which is the
to capture the effects on such scales. The upper limit is chosen to be degeneracy if only the position of the acoustic scale contributes to
well above the BAO scale, although the constraining contribution the fit. The dashed contours exclude the first data point, fitting from
of the bins above 130 h−1 Mpc is very small. Our final model has 20 to 190 h−1 Mpc only, with the best-fitting values α = 1.050 ±
four free parameters: m h2 , b, α and k∗ . 0.075 [corresponding to DV (zeff ) = 462 ± 33 Mpc], m h2 = 0.127
The best fit corresponds to a minimum χ 2 of 15.7 with 14 degrees ± 0.025 and b = 1.59 ± 0.15. The contours of this fit are tilted
of freedom (d.o.f.) (18 data points and four free parameters). The towards the dashed line, which means that the fit is now driven
best-fitting model is included in Fig. 2 (black line). The parameter by the BAO peak, while the general fit (solid contours) seems to
values are m h2 = 0.135 ± 0.020, b = 1.65 ± 0.10 and α = have some contribution from the shape of the correlation function.
1.039 ± 0.062, where the errors are derived for each parameter by Excluding the first data point increases the error on the distance
marginalizing over all other parameters. For k∗ , we can give a lower constraint only slightly from 5.9 to 7.1 per cent. The value of m h2
limit of k∗ = 0.19h Mpc−1 (with 95 per cent confidence level). tends to be smaller, but agrees within 1σ with the former value.
We can use equation (24) to turn the measurement of α into a Going back to the complete fit from 10 to 190 h−1 Mpc, we can
measurement of the distance to the effective redshift DV (zeff ) = include an external prior on m h2 from WMAP-7, which carries
V (zeff ) = 457 ± 27 Mpc, with a precision of 5.9 per cent. Our
αDfid an error of only 4 per cent (compared to the ≈15 per cent we

Downloaded
C 2011 RAS
on 05 June 2018
Focusing on the path from z = 0 to zeff = 0.106, our data set
can give interesting constraints on m . We derive the parameter
(Eisenstein et al. 2005)
√
m h2
A(zeff ) = 100DV (zeff ) , (27)
czeff
which has no dependence on the Hubble constant since DV ∝ h−1 .
We obtain A(zeff ) = 0.526 ± 0.028 √ with χ 2 /d.o.f. = 15.7/14. The
value of A would be identical to m if measured at redshift
z = 0. At redshift zeff = 0.106, we obtain a deviation from this
approximation of 6 per cent for our fiducial model, which is small
but systematic. We can express A, including the curvature term k
and the dark energy equation of state parameter w, as
⎧ √ 2/3
⎪
⎪ sinh k χ (z)
⎪
⎪ √ k > 0
⎪
⎪ k z
Figure 5. The distance measurement DV (z) relative to a low-redshift ap- ⎪
⎪
√ ⎪
⎪
m ⎨ χ (z) 2/3
proximation. The points show 6dFGS data and those of Percival et al. (2010).
A(z) = k = 0 (28)
obtain by fitting our data). Marginalizing over m h2 now gives E(z)1/3 ⎪
⎪ z
⎪
⎪
DV (zeff ) = 459 ± 18 Mpc, which reduces the error from 5.9 to 3.9 per ⎪
⎪ √ 2/3
⎪
⎪ sin |k |χ (z)
⎪
⎪ √ k < 0
cent. The uncertainty in m h2 from WMAP-7 contributes only about ⎩
|k |z
5 per cent of the error in DV [assuming no error in the WMAP-7
value of m h2 results in DV (zeff ) = 459 ± 17 Mpc]. with
In Fig. 5 we plot the ratio DV (z)/Dlow−z (z) as a function of redshift, z
V H0 dz
where Dlow−z (z) = cz/H 0 . At sufficiently low redshift, the approx- χ (z) = DC (z) = (29)
V
c E(z )
imation H(z) ≈ H 0 is valid and the measurement is independent 0
of any cosmological parameter except the Hubble constant. This and

figure also contains the results from Percival et al. (2010). E(z) = m (1 + z)3 + k (1 + z)2
Rather than including the WMAP-7 prior on m h2 to break the 1/2
degeneracy between m h2 and the distance constraint, we can fit the + (1 + z)3(1+w) ) . (30)
ratio rs (zd )/DV (zeff ), where rs (zd ) is the sound horizon at the baryon Using this equation, we now linearize our result for m in k and
drag epoch zd . In principle, this is rotating Fig. 4 so that the dashed w, and get
black line is parallel to the x-axis and hence breaks the degeneracy
if the fit is driven by the BAO peak; it will be less efficient if the fit m = 0.287 + 0.039(1 + w) + 0.039k ± 0.027. (31)
is driven by the shape of the correlation function. During the fit, we For comparison, Eisenstein et al. (2005) found
calculate rs (zd ) using the fitting formula of Eisenstein & Hu (1998).
The best fit results in rs (zd )/DV (zeff ) = 0.336 ± 0.015, which has m = 0.273 + 0.123(1 + w) + 0.137k ± 0.025 (32)
an error of 4.5 per cent, smaller than the 5.9 per cent found for DV based on the SDSS-LRG DR3 sample. This result shows the reduced
but larger than the error in DV when adding the WMAP-7 prior on sensitivity of the 6dFGS measurement to w and k .
m h2 . This is caused by the small disagreement in the DV –m h2
degeneracy and the line of constant sound horizon in Fig. 4. The χ 2
is 15.7, similar to the previous fit with the same number of degrees 6 C O S M O L O G I C A L I M P L I C AT I O N S
of freedom. In this section we compare our results to other studies and discuss
the implications for constraints on cosmological parameters. We
first note that we do not see any excess correlation on large scales
5.3 Extracting A(zeff ) and R(zeff )
as found in the SDSS-LRG sample. Our correlation function is in
We can also fit for the ratio of the distance between the effective agreement with a crossover to negative scales at 140 h−1 Mpc, as
redshift, zeff , and the redshift of decoupling (z∗ = 1091; Eisenstein predicted from CDM.
et al. 2005):
DV (zeff ) 6.1 Constraining the Hubble constant, H0
R(zeff ) = , (26)
(1 + z∗ )DA (z∗ ) We now use the 6dFGS data to derive an estimate of the Hubble
with (1 + z∗ )DA (z∗ ) = DC (z∗ )2 being the CMB angular distance. constant. We use the 6dFGS measurement of rs (zd )/DV (0.106) =
Besides the fact that the Hubble constant H 0 cancels out in the 0.336 ± 0.015 and fit directly for the Hubble constant and m . We
determination of R, this ratio is also more robust against effects combine our measurement with a prior on m h2 coming from the
caused by possible extra relativistic species (Eisenstein & White WMAP-7 Markov Chain results (Komatsu et al. 2011). Combining
2004). We calculate DA (z∗ ) for each m h2 during the fit and then the clustering measurement with m h2 from the CMB corresponds
marginalize over m h2 . The best fit results in R = 0.0324 ± 0.0015, to the calibration of the standard ruler.
with χ 2 = 15.7 and the same 14 d.o.f. We obtain values of H 0 = 67 ± 3.2 km s−1 Mpc−1 (which has an
uncertainty of only 4.8 per cent) and m = 0.296 ± 0.028. Table 1
and Fig. 6 summarize the results. The value of m agrees with the
2 In case of k = 0. value we derived earlier (Section 5.3).

C 2011 RAS
on 05 June 2018
Table 1. All parameter constraints from 6dFGS obtained in this paper. affected by parameters like k and w because of its lower redshift.
The priors used to derive these parameters are listed in square brackets. In any case, our result of the Hubble constant agrees very well with
All parameters assume b h2 = 0.022 27, and in cases where a prior on earlier BAO analyses. Furthermore, our result agrees with the latest
m h2 is used, we adopt the WMAP-7 Markov Chain probability distribution CMB measurement of H 0 = 70.3 ± 2.5 km s−1 Mpc−1 (Komatsu
(Komatsu et al. 2011). A(zeff ) is the acoustic parameter defined by Eisenstein
et al. 2011).
et al. (2005) [see equation (27) in the text] and R(zeff ) is the distance ratio
of the 6dFGS BAO measurement to the last-scattering surface. The most
The SH0ES programme (Riess et al. 2011) determined the Hub-
sensible value for cosmological parameter constraints is rs (zd )/DV (zeff ), ble constant using the distance ladder method. They used about
since this measurement is uncorrelated with m h2 . The effective redshift 600 near-IR observations of Cepheids in eight galaxies to im-
of 6dFGS is zeff = 0.106, and the fitting range is from 10 to 190 h−1 Mpc. prove the calibration of 240 low-redshift (z < 0.1) Type Ia su-
pernova, and calibrated the Cepheid distances using the geometric
Summary of parameter constraints from 6dFGS distance to the maser galaxy NGC 4258. They found H 0 = 73.8 ±
2.4 km s−1 Mpc−1 , a value consistent with the initial results of the
m h 2 0.135 ± 0.020 (14.8 per cent) Hubble Key project (H 0 = 72 ± 8 km s−1 Mpc−1 ; Freedman et al.
DV (zeff ) 457 ± 27 Mpc (5.9 per cent)
2001), but 1.7σ higher than our value (and 1.8σ higher when we
DV (zeff ) 459 ± 18 Mpc (3.9 per cent) [m h2 prior]
combine our data set with WMAP-7). While this could point towards
rs (zd )/ DV (zeff ) 0.336 ± 0.015 (4.5 per cent)
R(zeff ) 0.0324 ± 0.0015 (4.6 per cent) unaccounted or underestimated systematic errors in either one of
A(zeff ) 0.526 ± 0.028 (5.3 per cent) the methods, the likelihood of such a deviation by chance is about
10 per cent and hence is not enough to represent a significant dis-
m 0.296 ± 0.028 (9.5 per cent) [m h2 prior]
crepancy. Possible systematic errors affecting the BAO measure-
H0 67 ± 3.2 (4.8%) [m h2 prior]
ments are the modelling of non-linearities, bias and redshift-space
distortions, although these systematics are not expected to be sig-
nificant at the large scales relevant to our analysis.
To summarize the finding of this section, we can state that our
measurement of the Hubble constant is competitive with the lat-
est result of the distance ladder method. The different techniques
employed to derive these results have very different potential sys-
tematic errors. Furthermore, we found that BAO studies provide the
most accurate measurement of H 0 that exists, when combined with
the CMB distance priors.
6.2 Constraining dark energy

One key problem driving current cosmology is the determination
of the dark energy equation of state parameter, w. When adding
additional parameters like w to CDM, we find large degeneracies
in the WMAP-7-only data. One example is shown in Fig. 7. WMAP-7
Figure 6. The blue contours show the WMAP-7 m h2 prior (Komatsu

et al. 2011). The black contour shows constraints from 6dFGS derived by
fitting to the measurement of rs (zd )/DV (zeff ). The solid red contours show
the combined constraints resulting in H 0 = 67 ± 3.2 km s−1 Mpc−1 and
m = 0.296 ± 0.028. Combining the clustering measurement with m h2
from the CMB corresponds to the calibration of the standard ruler.
To combine our measurement with the latest CMB data, we use

the WMAP-7 distance priors, namely the acoustic scale
πDA (z∗ )
A = (1 + z∗ ) , (33)
rs (z∗ )
the shift parameter
√
m h2
R = 100 (1 + z∗ )DA (z∗ ) (34)
c
and the redshift of decoupling z∗ (tables 9 and 10 in Komatsu et al.
2011). This combined analysis reduces the error further and yields
Figure 7. The blue contours shows the WMAP-7 degeneracy in H 0 and
H 0 = 68.7 ± 1.5 km s−1 Mpc−1 (2.2 per cent) and m = 0.29 ±
w (Komatsu et al. 2011), highlighting the need for a second data set to
0.022 (7.6 per cent). break the degeneracy. The black contours show constraints from BAO data
Percival et al. (2010) determine a value H 0 = 68.6 ± 2.2 km incorporating the rs (zd )/DV (zeff ) measurements of Percival et al. (2010) and
s−1 Mpc−1 using SDSS-DR7, SDSS-LRG and 2dFGRS, while Reid 6dFGS. The solid red contours show the combined constraints resulting in
et al. (2010) found 69.4 ± 1.6 km s−1 Mpc−1 using the SDSS-LRG w = −0.97 ± 0.13. Excluding the 6dFGS data point widens the constraints
sample and WMAP-5. In contrast to these results, 6dFGS is less to the dashed red line with w = −1.01 ± 0.17.

Downloaded
C 2011 RAS
on 05 June 2018
Table 2. wCDM constraints from different data sets. Com-
paring the two columns shows the influence of the 6dFGS
data point. The 6dFGS data point reduces the error on w
by 24 per cent compared to WMAP-7+LRG which contains
only the BAO data points of Percival et al. (2010). We as-
sume flat priors of 0.11 < m h2 < 0.16 and marginalize
over m h2 . The asterisks denote the free parameters in each
fit.
Parameter WMAP-7+LRG WMAP-7+LRG+6dFGS
H0 69.9 ± 3.8(*) 68.7 ± 2.8(*)

m 0.283 ± 0.033 0.293 ± 0.027
0.717 ± 0.033 0.707 ± 0.027
w −1.01 ± 0.17(*) −0.97 ± 0.13(*)
alone cannot constrain H 0 or w within sensible physical boundaries Figure 8. The number of lognormal realizations found with a certain

(e.g. w < −1/3). As we are sensitive to H 0 , we can break the χ 2 , where the χ 2 is obtained by comparing a fit using a CDM
degeneracy between w and H 0 inherent in the CMB-only data. correlation function model with a no-baryon model. The blue line indicates
Our assumption of a fiducial cosmology with w = −1 does not the 6dFGS result.
introduce a bias, since our data are not sensitive to this parameter
and any deviation from this assumption is modelled within the shift best pure CDM model with our previous fit, we estimate that the
parameter α. BAO signal is detected with a significance of 2.4σ (corresponding
We again use the WMAP-7 distance priors introduced in the to χ 2 = 5.6). As a more qualitative argument for the detection
last section. In addition to our value of rs (zd )/DV (0.106) = 0.336 of the BAO signal we would like to refer again to Fig. 4 where the
± 0.015, we use the results of Percival et al. (2010), who found direction of the degeneracy clearly indicates the sensitivity to the
rs (zd )/DV (0.2) = 0.1905 ± 0.0061 and rs (zd )/DV (0.35) = 0.1097 ± BAO peak.
0.0036. To account for the correlation between the two latter data We can also use the lognormal realizations to determine how
points, we employ the covariance matrix reported in their paper. likely it is to find a BAO detection in a survey like 6dFGS. To do
Our fit has three free parameters: m h2 , H 0 and w. this, we produced 200 lognormal mock catalogues and calculated
The best fit gives w = −0.97 ± 0.13, H 0 = 68.7 ± the correlation function for each of them. We can now fit our cor-
2.8 km s−1 Mpc−1 and m h2 = 0.1380 ± 0.0055, with a χ 2 /d.o.f. = relation function model to these realizations. Furthermore, we fit
1.3/3. Table 2 and Fig. 7 summarize the results. To illustrate the im- a no-baryon model to the correlation function and calculate χ 2 ,
portance of the 6dFGS result to the overall fit, we also show how the distribution of which is shown in Fig. 8. We find that 26 per
the results change if 6dFGS is omitted. The 6dFGS data improve cent of all realizations have at least a 2σ BAO detection, and that
the constraint on w by 24 per cent. 12 per cent have a detection >2.4σ . The lognormal realizations
Finally we show the best-fitting cosmological parameters for show a mean significance of the BAO detection of 1.7 ± 0.7σ ,
different cosmological models using WMAP-7 and BAO results in where the error describes the variance around the mean.
Table 3. Fig. 9 shows the 6dFGS data points together with all 200 lognor-
mal realizations (grey). The red data points indicate the mean for
each bin, and the black line is the input model derived as explained
7 S I G N I F I C A N C E O F T H E BAO D E T E C T I O N
in Section 3.3. This comparison shows that the 6dFGS data contain
To test the significance of our detection of the BAO signature, a BAO peak slightly larger than expected in CDM.
we follow Eisenstein et al. (2005) and perform a fit with a fixed The amplitude of the acoustic feature relative to the overall nor-
b = 0, which corresponds to a pure CDM model without a BAO malization of the galaxy correlation function is quite sensitive to
signature. The best fit has χ 2 = 21.4 with 14 d.o.f. and is shown the baryon fraction, f b = b /m (Matsubara 2004). A higher BAO
as the red line in Fig. 2. The parameter values of this fit depend on peak could hence point towards a larger baryon fraction in the local
the parameter priors, which we set to 0.7 < α < 1.3 and 0.1 < m universe. However, since the correlation function model seems to
h2 < 0.2. Values of α much further away from 1 are problematic agree very well with the data (with a reduced χ 2 of 1.12) and
since equation (24) is only valid for α close to 1. Comparing the is within the range spanned by our lognormal realizations, we
Table 3. Parameter constraints from WMAP7+BAO for (i) a flat CDM model, (ii) an open
CDM (oCDM), (iii) a flat model with w = constant (wCDM) and (iv) an open model with w
= constant (owCDM). We assume flat priors of 0.11 < m h2 < 0.16 and marginalize over m h2 .
The asterisks denote the free parameters in each fit.
Parameter CDM oCDM wCDM owCDM
H0 69.2 ± 1.1(*) 68.3 ± 1.7(*) 68.7 ± 2.8(*) 70.4 ± 4.3(*)

m 0.288 ± 0.011 0.290 ± 0.019 0.293 ± 0.027 0.274 ± 0.035
k (0) −0.0036 ± 0.0060(*) (0) −0.013 ± 0.010(*)
0.712 ± 0.011 0.714 ± 0.020 0.707 ± 0.027 0.726 ± 0.036
w (−1) (−1) −0.97 ± 0.13(*) −1.24 ± 0.39(*)

C 2011 RAS
on 05 June 2018
Figure 11. Redshift distribution of 6dFGS, WALLABY and two different

Figure 9. The different lognormal realizations used to calculate the covari-
versions of the proposed TAIPAN survey. See text for details.
ance matrix (shown in grey). The red points indicate the mean values, while
the blue points show actual 6dFGS data (the data point at 5 h−1 Mpc is not
included in the fit). The red data points are shifted by 2 h−1 Mpc to the right
for clarity. We included this selection to show that a stronger BAO peak does
not bias the estimate of α in any direction. The Gaussian fit gives
χ 2 /d.o.f. = 5/11, with a mean of 1.007 ± 0.007. The distribution of
α shows a smaller spread with σ = 0.041 ± 0.008, about 2σ below
our error on α. This result shows that a survey like 6dFGS is able
to constrain α (and hence DV and H 0 ) to the precision we report in
this paper.
8 F U T U R E A L L S K Y S U RV E Y S
A major new wide-sky survey of the local Universe will be the Wide
field ASKAP L-band Legacy All-sky Blind surveY (WALLABY).3
This is a blind H I survey planned for the Australian Square Kilome-
tre Array Pathfinder telescope (ASKAP), currently under construc-
tion at the Murchison Radio-astronomy Observatory in Western
Australia.
The survey will cover at least 75 per cent of the sky with the
potential to cover 4π of sky if the Westerbork Radio Telescope
Figure 10. This plot shows the distribution of the parameter α derived from
the 200 lognormal realizations (black). The distribution is well fitted by a
delivers complementary northern coverage. Compared to 6dFGS,
Gaussian with a mean of μ = 0.998 ± 0.004 and a width of σ = 0.057 ± WALLABY will more than double the sky coverage including the
0.005. In blue we show the same distribution selecting only the lognormal Galactic plane. WALLABY will contain ∼500 000–600 000 galax-
realizations with a strong BAO peak (>2σ ). The Gaussian distribution in ies with a mean redshift of around 0.04, giving it around four times
this case gives a mean of 1.007 ± 0.007 and σ = 0.041 ± 0.008. greater galaxy density compared to 6dFGS. In the calculations that
follow, we assume for WALLABY a 4π survey without any exclu-
sion around the Galactic plane. The effective volume in this case
cannot claim any discrepancy with CDM. Therefore, the most turns out to be 0.12 h−3 Gpc3 .
likely explanation for the excess correlation in the BAO peak is The TAIPAN4 survey, proposed for the UK Schmidt Telescope
sample variance. at Siding Spring Observatory, will cover a comparable area of sky,
In Fig. 10 we show the distribution of the parameter α obtained and will extend 6dFGS in both depth and redshift (z 0.08).
from the 200 lognormal realizations. The distribution is well de- The redshift distribution of both surveys is shown in Fig. 11,
scribed by a Gaussian with χ 2 /d.o.f. = 14.2/20, where we employed alongside 6dFGS. Since the TAIPAN survey is still in the early
Poisson errors for each bin. This confirms that α has Gaussian dis- planning stage, we consider two realizations: TAIPAN1 (406 000
tributed errors in the approximation that the 6dFGS sample is well galaxies to a faint magnitude limit of r = 17) and the shallower
described by lognormal realizations of an underlying CDM power TAIPAN2 (221 000 galaxies to r = 16.5). We have adopted the
spectrum. This result increases our confidence that the application same survey window as was used for 6dFGS, meaning that it covers
of Gaussian errors for the cosmological parameter fits is correct. the whole southern sky excluding a 10◦ strip around the Galactic
The mean of the Gaussian distribution is at 0.998 ± 0.004, in agree- plane. The effective volumes of TAIPAN1 and TAIPAN2 are 0.23
ment with unity, which shows that we are able to recover the input and 0.13 h−3 Gpc3 , respectively.
model. The width of the distribution shows the mean expected error
in α in a CDM universe for a 6dFGS-like survey. We found σ =
0.057 ± 0.005, which is in agreement with our error in α of 5.9 per 3http://www.atnf.csiro.au/research/WALLABY
cent. Fig. 10 also contains the distribution of α, selecting only the 4TAIPAN: Transforming Astronomical Imaging surveys through Polychro-
lognormal realizations with a strong (>2σ ) BAO peak (blue data). matic Analysis of Nebulae.

Downloaded
C 2011 RAS
on 05 June 2018
of the BAO detection for each lognormal realization by performing
fits to the correlation function using CDM parameters and b
= 0, in exactly the same manner as the 6dFGS analysis described
earlier. We find a 3.5 ± 0.8σ significance for the BAO detection for
TAIPAN1, 2.1 ± 0.7σ for TAIPAN2 and 2.1 ± 0.7σ for WALLABY,
where the error again describes the variance around the mean.
We then fit a correlation function model to the mean values of
the lognormal realizations for each survey, using the covariance
matrix derived from these lognormal realizations. We evaluated the
correlation function of WALLABY, TAIPAN2 and TAIPAN1 at the
effective redshifts of 0.1, 0.12 and 0.14, respectively. With these in
hand, we are able to derive distance constraints to respective pre-
cisions of 7, 6 and 3 per cent. The predicted value for WALLABY
is not significantly better than that from 6dFGS. This is due to the
significance of the 6dFGS BAO peak in the data, allowing us to
Figure 12. Predictions for two versions of the proposed TAIPAN survey. place tight constraints on the distance. As an alternative figure of
Both predictions assume a 2π-steradian southern sky-coverage, excluding merit, we derive the constraints on the Hubble constant. All surveys
the Galactic plane (i.e. |b| > 10◦ ). TAIPAN1 contains 406 000 galaxies recover the input parameter of H 0 = 70 km s−1 Mpc−1 , with abso-
while TAIPAN2 contains 221 000 (see Fig. 11). The blue points are shifted lute uncertainties of 3.7, 3 and 2.2 km s−1 Mpc−1 for WALLABY,
by 2 h−1 Mpc to the right for clarity. The black line is the input model, which TAIPAN2 and TAIPAN1, respectively. Hence, TAIPAN1 is able to
is a CDM model with a bias of 1.6, β = 0.3 and k∗ = 0.17h Mpc−1 . For a constrain the Hubble constant to 3 per cent precision. These con-
large number of realizations, the difference between the input model and the straints might improve when combined with Planck constraints on
mean (the data points) is only the convolution with the window function.
b h2 and m h2 which will be available when these surveys come
along.
Since there is significant overlap between the survey volume of
6dFGS, TAIPAN and WALLABY, it might be interesting to test
whether the BAO analysis of the local Universe can make use of
a multiple tracer analysis, as suggested recently by Arnalte-Mur
et al. (2011). These authors claim that by employing two different
tracers of the matter density field – one with high bias to trace the
central overdensities and one with low bias to trace the small density
fluctuations – one can improve the detection and measurement of
the BAO signal. Arnalte-Mur et al. (2011) test this approach using
the SDSS-LRG sample (with a very large bias) and the SDSS-
main sample (with a low bias). Although the volume is limited by
the amount of sample overlap, they detect the BAO peak at 4.1σ .
Likewise, we expect that the contrasting high bias of 6dFGS and
TAIPAN, when used in conjunction with the low bias of WALLABY,
would furnish a combined sample that would be ideal for such an
Figure 13. Prediction for the WALLABY survey. We have assumed a
analysis.
4π-steradian survey with 602 000 galaxies, b = 0.7, β = 0.7 and k∗ = Neither TAIPAN nor WALLABY is designed as BAO surveys,
0.17h Mpc−1 . with their primary goals relating to galaxy formation and the local
Universe. However, we have found that TAIPAN1 would be able to
improve the measurement of the local Hubble constant by about 30
To predict the ability of these surveys to measure the large- per cent compared to 6dFGS going to only slightly higher redshift.
scale correlation function, we produced 100 lognormal realizations WALLABY could make some interesting contributions in the form
for TAIPAN1 and WALLABY, and 200 lognormal realizations for of a multiple tracer analysis.
TAIPAN2. Figs 12 and 13 show the results in each case. The data
points are the mean of the different realizations, and the error bars
are the diagonal of the covariance matrix. The black line represents 9 CONCLUSION
the input model which is a CDM prediction convolved with a We have calculated the large-scale correlation function of the 6dFGS
Gaussian damping term using k∗ = 0.17h Mpc−1 (see equation 17). and detected a BAO peak with a significance of 2.4σ . Although
We used a bias parameter of 1.6 for TAIPAN (similar to that found in 6dFGS was never designed as a BAO survey, the peak is detectable
6dFGS) and following our fiducial model we get β = 0.3, resulting because the survey contains a large number of very bright, highly
in A = b2 (1 + 2β/3 + β 2 /5) = 3.1. For WALLABY, we used a bias biased galaxies, within a sufficiently large effective volume of
of 0.7 (based on the results found in the HIPASS survey; Basilakos 0.08 h−3 Gpc3 . We draw the following conclusions from our work.
et al. 2007). This results in β = 0.7 and A = 0.76. To calculate the
correlation function, we used P0 = 40 000 h3 Mpc3 for TAIPAN and (i) The 6dFGS BAO detection confirms the finding by SDSS and
P0 = 5 000h3 Mpc3 for WALLABY. 2dFGRS of a peak in the correlation function at around 105 h−1 Mpc,
The error bar for TAIPAN1 is smaller by roughly a factor
√ of 1.7 consistent with CDM. This is important because 6dFGS is an
relative to 6dFGS, which is consistent with scaling by Veff and is independent sample, with a different target selection, redshift
comparable to the SDSS-LRG sample. We calculate the significance distribution and bias compared to previous studies. The 6dFGS

C 2011 RAS
on 05 June 2018
BAO measurement is the lowest redshift BAO measurement ever Bashinsky S., Bertschinger E., 2001, Phys. Rev. Lett., 87, 081301
made. Bashinsky S., Bertschinger E., 2002, Phys. Rev. D, 65, 123008
(ii) We do not see any excess correlation at large scales as seen in Basilakos S., Plionis M., Kovac K., Voglis N., 2007, MNRAS, 378, 301
the SDSS-LRG sample. Our correlation function is consistent with Bassett B. A., Hlozek R., 2009, preprint (arXiv:0910.5224)
Blake C., Glazebrook K., 2003, ApJ, 594, 665
a crossover to negative values at 140 h−1 Mpc, as expected from
Blake C., Collister A., Bridle S., Lahav O., 2007, MNRAS, 374, 1527
CDM models.
Blake C. et al., 2011, preprint (arXiv:1105.2862)
(iii) We derive the distance to the effective redshift as Bond J. R., Efstathiou G., 1987, MNRAS, 226, 655
DV (zeff ) = 457 ± 27 Mpc (5.9 per cent precision). Alternatively, Chuang C. H., Wang Y., 2011, preprint (arXiv:1102.2251)
we can derive rs (zd )/DV (zeff ) = 0.336 ± 0.015 (4.5 per cent preci- Cole S. (The 2dFGRS Collaboration) et al., 2005, MNRAS, 362, 505
sion). All parameter constraints are summarized in Table 1. Coles P., Jones B., 1991, MNRAS, 248, 1
(iv) Using a prior on m h2 from WMAP-7, we find m = Colless M. (The 2DFGRS Collaboration) et al., 2001, MNRAS, 328,
0.296 ± 0.028. Independent of WMAP-7, and taking into account 1039
curvature and the dark energy equation of state, we derive m Cooray A., Hu W., Huterer D., Joffre M., 2001, ApJ, 557, L7
= 0.287 + 0.039(1 + w) + 0.039k ± 0.027. This agrees very Crocce M., Scoccimarro R., 2006, Phys. Rev. D, 73, 063519
Crocce M., Scoccimarro R., 2008, Phys. Rev. D, 77, 023533
well with the first value, and shows the very small dependence on
Efstathiou G., 1988, in Lawrence A., ed., Proc. 3rd IRAS Conf., Comets to
cosmology for parameter derivations from 6dFGS given its low
Cosmology. Springer, New York, p. 312
redshift. Efstathiou G., Bond J. R., 1999, MNRAS, 304, 75
(v) We are able to measure the Hubble constant, H 0 = 67 ± Eisenstein D. J., Hu W., 1998, ApJ, 496, 605
3.2 km s−1 Mpc−1 , to 4.8 per cent precision, using only the standard Eisenstein D. J., White M. J., 2004, Phys. Rev. D, 70, 103523
ruler calibration by the CMB (in form of m h2 and b h2 ). Com- Eisenstein D. J., Hu W., Tegmark M., 1998, ApJ, 504, L57
pared to previous BAO measurements, 6dFGS is almost completely Eisenstein D. J. (SDSS Collaboration) et al., 2005, ApJ, 633, 560
independent of cosmological parameters (e.g. k and w), similar Eisenstein D. J., Seo H. j., White M. J., 2007a, ApJ, 664, 660
to Cepheid and low-z supernovae methods. However, in contrast to Eisenstein D. J., Seo H. j., Sirko E., Spergel D., 2007b, ApJ, 664, 675
these methods, the BAO derivation of the Hubble constant depends Feldman H. A., Kaiser N., Peacock J. A., 1994, ApJ, 426, 23
Freedman W. L. (HST Collaboration) et al., 2001, ApJ, 553, 47
on very basic early Universe physics and avoids possible systematic
Frigo M., Johnson S. G., 2005, Proc. IEEE 93, 216231
errors coming from the build-up of a distance ladder.
Gaztanaga E., Cabre A., Hui L., 2009, MNRAS, 399, 1663
(vi) By combining the 6dFGS BAO measurement with those of Goldberg D. M., Strauss M. A., 1998, ApJ, 495, 29
WMAP-7 and previous redshift samples (from SDSS-DR7, SDDS- Guzik J., Bernstein G., 2007, MNRAS, 375, 1329
LRG and 2dFGRS; Percival et al. 2010), we can further improve Heitmann K., White M., Wagner C., Habib S., Higdon D., 2010, ApJ, 715,
the constraints on the dark energy equation of state, w, by breaking 104
the H 0 –w degeneracy in the CMB data. Doing this, we find w = Huetsi G., 2006, A&A, 459, 375
−0.97 ± 0.13, which is an improvement of 24 per cent compared Huetsi G., 2009, preprint (arXiv:0910.0492)
to previous combinations of BAO and WMAP-7 data. Jarrett T. H., Chester T., Cutri R., Schneider S., Skrutskie M., Huchra J. P.,
(vii) We have made detailed predictions for two next-generation 2001, AJ, 119, 2498
Jones D. H. et al., 2004, MNRAS, 355, 747
low-redshift surveys, WALLABY and TAIPAN. Using our 6dFGS
Jones D. H., Peterson B. A., Colless M., Saunders W., 2006, MNRAS, 369,
result, we predict that both surveys will detect the BAO signal,
25
and that WALLABY may be the first radio galaxy survey to do so. Jones D. H. et al., 2009, preprint (arXiv:0903.5451)
Furthermore, we predict that TAIPAN has the potential to constrain Kaiser N., 1987, MNRAS, 227, 1
the Hubble constant to a precision of 3 per cent improving the Kazin E. A. et al., 2010a, ApJ, 710, 1444
6dFGS measurement by 30 per cent. Kazin E. A., Blanton M. R., Scoccimarro R., McBride C. K., Berlind A. A.,
2010b, ApJ, 719, 1032
Kitaura F. S., Jasche J., Metcalf R. B., 2009, preprint (arXiv:0911.1407)
AC K N OW L E D G M E N T S Komatsu E. et al., 2011, ApJS, 192, 18
Labini F. S., Vasilyev N. L., Baryshev Y. V., Lopez-Corredoira M., 2009,
The authors thank Alex Merson for providing the random mock gen- A&A, 505, 981
erator, and Lado Samushia for helpful advice with the wide-angle Landy S. D., Szalay A. S., 1993, ApJ, 412, 64
formalism. We thank Martin Meyer and Alan Duffy for fruitful Lewis A., Challinor A., Lasenby A., 2000, ApJ, 538, 473
discussions, and Greg Poole for providing the relation for the scale- Loveday J., Maddox S. J., Efstathiou G., Peterson B. A., 1995, ApJ, 442,
dependent bias. We also thank Tamara Davis, Eyal Kazin and John 457
Peacock for comments on earlier versions of this paper. FB is sup- Martinez M. J., Arnalte-Mur P., Saar E., de la Cruz P., Pons-Borderia, J. M.,
ported by the Australian Government through the International Post- Paredes S., Fernandez-Soto A., Tempel E., 2009, ApJ, 696, L93
graduate Research Scholarship (IPRS) and by scholarships from Matsubara T., 2004, ApJ, 615, 573
Matsubara T., 2008, Phys. Rev. D, 77, 063530
ICRAR and the AAO. Part of this work used the ivec@UWA su-
Miller C. J., Nichol R. C., Batuski D. J., 2001, ApJ, 555, 68
percomputer facility. The 6dF Galaxy Survey was funded in part Norberg P., Baugh C. M., Gaztanaga E., Croton D. J., preprint
by an Australian Research Council Discovery–Projects Grant (DP- (arXiv:0810.1885)
0208876), administered by the Australian National University. Padmanabhan N., White M. J., 2008, Phys. Rev. D, 77, 123540
Padmanabhan N. (SDSS Collaboration) et al., 2007, MNRAS, 378, 852
Papai P., Szapudi I., 2008, preprint (arXiv:0802.2940)
REFERENCES
Peebles P. J. E., Yu J. T., 1970, ApJ, 162, 815
Abazajian K. N. (SDSS Collaboration) et al., 2009, ApJS, 182, 543 Percival W. J., Verde L., Peacock J. A., 2004, MNRAS, 347, 645
Angulo R., Baugh C. M., Frenk C. S., Lacey C. G., 2008, MNRAS, 383, Percival W. J. et al., 2010, MNRAS, 401, 2148
755 Raccanelli A., Samushia L., Percival W. J., 2010, preprint (arXiv:1006.
Arnalte-Mur P. et al., 2011, MNRAS, submitted, preprint (arXiv:1101.1911) 1652)

Downloaded
C 2011 RAS
on 05 June 2018
Reid B. A. et al., 2010, MNRAS, 404, 60 zero. The lognormal density field is then given by
Riess A. G. et al., 2011, ApJ, 730, 119
Sanchez A. G., Baugh C. M., Angulo R., 2008, MNRAS, 390, 1470 μL (x) = exp δ(x) − σG2 /2 , (A3)
Sanchez A. G., Crocce M., Cabre A., Baugh C. M., Gaztanaga E., 2009, which is now a quantity defined on [0, ∞[ only, while δ(x) is defined
preprint (arXiv:0901.2570)
on ]−∞, ∞[.
Seo H. J., Eisenstein D. J., 2003, ApJ, 598, 720
Smith R. E. (The Virgo Consortium Collaboration) et al., 2003, MNRAS,
Since we want to calculate a mock catalogue for a particular sur-
341, 1311 vey, we have to incorporate the survey selection function.
If W (x)
Smith R. E., Scoccimarro R., Sheth R. K., 2007, Phys. Rev. D, 75, is the selection function with the normalization W (x) = 1,
063512 we calculate the mean number of galaxies in each grid cell
Smith R. E., Scoccimarro R., Sheth R. K., 2008, Phys. Rev. D, 77, as
043525
Sunyaev R. A., Zeldovich Y. B., 1970, Ap&SS, 7, 3 ng (x) = N W (x) μL (x), (A4)
Szalay A. S., Matsubara T., Landy S. D., preprint (arXiv:astro-ph/9712007) where N is the total number of galaxies in our sample. The galaxy
Szapudi I., 2004, ApJ, 614, 51
catalogue itself is than generated by Poisson sampling ng (x).
Tegmark M., 1997, Phys. Rev. Lett., 79, 3806
Tegmark M. (SDSS Collaboration) et al., 2006, Phys. Rev. D, 74, 123507
The galaxy position is not defined within the subcell, and we
Weinberg D. H., Cole S., 1992, MNRAS, 259, 652 place the galaxy in a random position within the box. This means
Wright E. L., 2006, Publ. Astron. Soc. Pacific, 118, 1711 that the correlation function calculated from such a distribution
York D. G. (SDSS Collaboration) et al., 2000, AJ, 120, 1579 is smooth at scales smaller than the subcell. It is therefore im-
Zehavi I. (SDSS Collaboration) et al., 2005, ApJ, 621, 22 portant to make sure that the grid cells are smaller than the size
of the bins in the correlation function calculation. In the 6dFGS
calculations presented in this paper, the grid cells have a size of
A P P E N D I X A : G E N E R AT I N G L O G N O R M A L 5 h−1 Mpc, while the correlation function bins are 10 h−1 Mpc in
M O C K C ATA L O G U E S size.
Here we explain in detail the different steps used to derive a lognor-
mal mock catalogue, as a useful guide for researchers in the field. A P P E N D I X B : C O M PA R I S O N O F L O G N O R M A L
We start with an input power spectrum (which is determined as A N D JAC K K N I F E E R RO R E S T I M AT E S
explained in Section 3.3) in units of h−3 Mpc3 . We set up a 3D grid
with the dimensions Lx × Ly × Lz = 1000 × 1000 × 1000 h−1 Mpc We have also estimated jackknife errors for the correlation function,
with 2003 subcells. We then distribute the quantity P
(k)/V over this by way of comparison. We divided the survey into 18 regions and
calculated the correlation function by excluding one region at a time.
grid, where V is the volume of the grid and k = kx2 + ky2 + kz2
We found that the size of the error bars around the BAO peak varies
with kx = nx 2π/Lx and nx being an integer value specifying the by around 20 per cent in some bins, when we increase the number
x-coordinates of the grid cells. of jackknife regions from 18 to 32. Furthermore, the covariance
Performing a complex-to-real Fourier transform (FT) of this grid matrix derived from jackknife resampling is very noisy and hard to
will produce a 3D correlation function. Since the power spectrum invert.
has the property P (−k) = P (k)∗ , the result will be real. We show the jackknife errors in Fig. B1. The jackknife error
The next step is to replace the correlation function ξ (r) at each shows more noise and is larger in most bins compared to the log-
point in the 3D grid by ln [1 + ξ (r)], where ln is the natural loga- normal error. The error shown in Fig. B1 is only the diagonal term of
rithm. This step prepares the input model for the inverse step, which the covariance matrix and does not include any correlation between
we later use to produce the lognormal density field. bins.
Using a real-to-complex FT we can revert to k-space where we The full error matrix is shown in Fig. B2, where we plot the
now have a modified power spectrum, Pln (k). At this point, we correlation matrix of the jackknife error estimate compared to
divide by the number of subcells N c . The precise normalization
depends on the definition of the discrete FT. We use the FFTW
library (Frigo & Johnson 2005), where the discrete FT is defined as

N c −1
√
Yi = Xj exp[±2πij −1/Nc ]. (A1)
j =0
The modified power spectrum Pln (k) is not guaranteed to be neither

positive defined nor a real function, which contradicts the defini-
tion of a power spectrum. Weinberg & Cole (1992) suggested to
construct a well-defined power spectrum from Pln (k) by
Pln (k) = max [0, Re[Pln (k)]] . (A2)
We now generate a real and an imaginary Fourier amplitude δ(k)
for each point on the grid by randomly sampling from a Gaussian
distribution with rms Pln (k)/2. However, to ensure that the final
overdensity field is real, we have to manipulate the grid, so that all Figure B1. Correlation function error for different values of P0 . The weight-
subcells follow the condition δ(−k) = δ(k)∗ . ing with P0 = 40 000 h3 Mpc−3 reduces the error at the BAO scale by almost
Performing another FT results in an overdensity field δ(x) from a factor of 4 compared to the case without weighting. The red dashed line
which we calculate the variance σ 2G . The mean of δ(x) should be indicates the jackknife error.

C 2011 RAS
on 05 June 2018
The total correlation function model, including O(θ 2 ) correction
terms, is then given by Papai & Szapudi (2008):
ξ (φ, θ, s) = a00 + 2a02 cos(2φ) + a22 cos(2φ) + b22 sin2 (2φ)

+ [−4a02 cos(2φ) − 4a22 − 4b22 − 4a10 cot2 (φ)
+ 4a11 cot2 (φ) − 4a12 cot2 (φ) cos(2φ) + 4b11
− 8b12 cos2 (φ)]θ 2 + O(θ 4 ) (C1)
This equation reduces to the plane-parallel approximation if θ = 0.
The factors axy and bxy in this equation are given by

2β 2β 2 2
a00 = 1 + + ξ0 (r)
3 15

β 2β 2 2 3β 2 2
− + ξ2 (r) + ξ (r)
3 21 140 4

β 3β 2 2 β2 2
a02 = − + ξ2 (r) + ξ (r)
2 14 28 4
Figure B2. Correlation matrix of the jackknife errors (upper-left triangle)
β2 2 β2 2 19β 2 2
and lognormal errors (lower-right triangle). a22 = ξ0 (r) − ξ2 (r) + ξ (r)
15 21 140 4
β2 2 β2 2 4β 2 2
b22 = ξ0 (r) − ξ2 (r) − ξ (r)
15 21 35 4
the lognormal error. The jackknife correlation matrix looks much
more noisy and seems to have less correlation in neighbouring 4β 2 1 1 β2 1
a10 = 2β + ξ1 (r) − ξ (r)
bins. 5 r 5r 3
The number of jackknife regions cannot be chosen arbitrarily. 4β 2 0
Each jackknife region must be at least as big as the maximum a11 = 2
ξ0 (r) − 2ξ20 (r)
3r
scale under investigation. Since we want to test scales up to al-
most 200 h−1 Mpc, our jackknife regions must be very large. On β2 1
a21 = 3ξ3 (r) − 2ξ11 (r)
the other hand, we need at least as many jackknife regions as 5r
we have bins in our correlation function, otherwise the covari- 4β 2 0
b11 = ξ (r) + ξ20 (r)
ance matrix is singular. These requirements can contradict each 3r 2 0
other, especially if large scales are analysed. Furthermore, the small 2β 2 1
b12 = ξ1 (r) + ξ31 (r) , (C2)
number of jackknife regions is the main source of noise (for a
5r
more detailed study of jackknife errors, see e.g. Norberg et al.
2008). where β = m (z)0.545 /b, with b being the linear bias. The correlation
Given these limitations in the jackknife error approach, correla- function moments are given by
tion function studies on large scales usually employ simulations or ∞
1
lognormal realizations to derive the covariance matrix. We decided ξlm (r) = dk k m Plin (k)jl (rk), (C3)
to use the lognormal error in our analysis. We showed that the jack- 2π2 0
knife errors tend to be larger than the lognormal error at larger scales with jl (x) being the spherical Bessel function of order l.
and carry less correlation. These differences might be connected to The final spherically averaged correlation function is given by
the much higher noise level in the jackknife errors, which is clearly
visible in all our data. It could be, however, that our jackknife regions π π/2
are too small to deliver reliable errors on large scales. We use the ξ (s) = ξ (φ, θ, s)N (φ, θ, s) dθdφ, (C4)
0 0
minimum number of jackknife regions to make the covariance ma-
trix non-singular (the correlation function is measured in 18 bins). where the function N(φ, θ , s) is obtained from the data. N(φ, θ ,
The mean distance of the jackknife regions to each other is about s) counts the number of galaxy pairs at different φ, θ and s, and
200 h−1 Mpc at the mean redshift of the survey, but smaller at low includes the areal weighting sin (φ) which usually has to be included
redshift. in an integral over φ. It is normalized such that
π π/2
N (φ, θ, s) dθ dφ = 1. (C5)
0 0
APPENDIX C: WIDE-ANGLE FORMALISM
If the angle θ is of order 1 rad, higher order terms become dominant
The general redshift-space correlation function (ignoring the plane- and equation (C1) is no longer sufficient. Our weighted sample
parallel approximation) depends on φ, θ and s. Here, s is the sep- has only small values of θ, but growing with s (see Fig. C1). In
aration between the galaxy pair, θ is the half opening angle, and φ our case, the correction terms contribute only mildly at the BAO
is the angle of s to the line of sight (see fig. 1 in Raccanelli et al. scale (red line in Fig. C2). However, these corrections behave like
2010). For the following calculations, it must be considered that in a scale-dependent bias and hence can introduce systematic errors if
this parametrization, φ and θ are not independent. not modelled correctly.

Downloaded
C 2011 RAS
on 05 June 2018
Figure C1. The half-opening angle θ as a function of separation s of the Figure C2. The black line represents the plain correlation function without
6dFGS weighted catalogue. The plane-parallel approximation assumes θ = redshift-space distortions (RSDs), ξ (r), obtained by a Hankel transform of
0. The mean half-opening angle at the BAO scale is 10◦ . The colour bar our fiducial CDM power spectrum. The blue line includes the linear model
gives the number of pairs in each bin. for RSDs (linear Kaiser factor) using β = 0.27. The red line uses the same
value of β but includes all correction terms outlined in equation (C1) using
the N(φ, θ , s) distribution of the weighted 6dFGS sample employed in this
analysis.
This paper has been typeset from a TEX/LATEX file prepared by the author.

C 2011 RAS
on 05 June 2018

6 D FGalaxy Survey BAOandthelocal Hconstant

Uploaded by

Copyright:

Available Formats

6 D FGalaxy Survey BAOandthelocal Hconstant

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

6 D FGalaxy Survey BAOandthelocal Hconstant

Uploaded by

Copyright:

Available Formats

Mon. Not. R. Astron. Soc. 416, 3017–3032 (2011) doi:10.1111/j.1365-2966.2011.19250.

Florian Beutler,1 Chris Blake,2 Matthew Colless,3 D. Heath Jones,4

to radiation through Thomson scattering. The radiation pressure

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

covered by the 2dFGRS, with a sample density similar to SDSS- 

and the sums go over all random (N r ) and data (N d ) galaxies. We

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

4.2 Non-linear effects

B(s) = 1 + (s/0.474 h−1 Mpc)−1.332 , (20)

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

of any cosmological parameter except the Hubble constant. This and

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

6.2 Constraining dark energy

Figure 6. The blue contours show the WMAP-7 m h2 prior (Komatsu

To combine our measurement with the latest CMB data, we use

Parameter WMAP-7+LRG WMAP-7+LRG+6dFGS

H0 69.9 ± 3.8(*) 68.7 ± 2.8(*)

Parameter CDM oCDM wCDM owCDM

H0 69.2 ± 1.1(*) 68.3 ± 1.7(*) 68.7 ± 2.8(*) 70.4 ± 4.3(*)

C 2011 The Authors, MNRAS 416, 3017–3032

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

Figure 11. Redshift distribution of 6dFGS, WALLABY and two different

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

The modified power spectrum Pln (k) is not guaranteed to be neither

C 2011 The Authors, MNRAS 416, 3017–3032

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

ξ (φ, θ, s) = a00 + 2a02 cos(2φ) + a22 cos(2φ) + b22 sin2 (2φ)

C 2011 The Authors, MNRAS 416, 3017–3032

Downloaded from https://academic.oup.com/mnras/article-abstract/416/4/3017/976636

You might also like

covered by the 2dFGRS, with a sample density similar to SDSS-

H0 69.9 ± 3.8() 68.7 ± 2.8()

H0 69.2 ± 1.1() 68.3 ± 1.7() 68.7 ± 2.8() 70.4 ± 4.3()