Issue 
A&A
Volume 513, April 2010



Article Number  A3  
Number of page(s)  5  
Section  Cosmology (including clusters of galaxies)  
DOI  https://doi.org/10.1051/00046361/200912763  
Published online  09 April 2010 
Absence of significant crosscorrelation between
WMAP and SDSS
(Research Note)
M. LópezCorredoira^{1,2}  F. Sylos Labini^{3,4}  J. BetancortRijo^{1,2}
1 
Instituto de Astrofísica de Canarias,
38200 La Laguna, Tenerife, Spain
2 
Departamento de Astrofísica, Universidad de La Laguna,
38205 La Laguna, Tenerife, Spain
3 
Centro Studi e Ricerche Enrico Fermi, via Panisperna 89 A, Compendio del
Viminale, 00184 Rome, Italy
4  Istituto dei Sistemi Complessi CNR, via dei Taurini 19, 00185 Rome,
Italy
Received 25 June 2009 / Accepted 17 January 2010
Abstract
Context. Several authors have claimed to detect a
significant crosscorrelation between microwave WMAP anisotropies and
the SDSS galaxy distribution. We repeat these analyses to determine the
different crosscorrelation uncertainties caused by resampling errors
and fieldtofield fluctuations. The first type of error concerns
overlapping sky regions, while the second type concerns nonoverlapping
sky regions.
Aims. To measure the resampling errors, we use bootstrap and
jackknife techniques. For the fieldtofield fluctuations, we use
three methods: 1) evaluation of the dispersion in the crosscorrelation
when correlating separated regions of WMAP with the original region of
SDSS; 2) use of mock Monte Carlo WMAP maps; 3) a new method (developed
in this article), which measures the error as a function of the
integral of the product of the selfcorrelations for each map.
Methods. The average crosscorrelation for b>30 deg
is significantly stronger than the resampling errors  both the
jackknife and bootstrap techniques provide similar results  but it is
of the order of the fieldtofield fluctuations. This is confirmed by
the crosscorrelation between anisotropies and galaxies in more than
the half of the sample being null within resampling errors.
Results. Resampling methods underestimate the errors.
Fieldtofield fluctuations dominate the detected signals. The ratio of
signal to resampling errors is larger than unity in a way that
strongly depends on the selected sky region. We therefore conclude that
there is no evidence yet of a significant detection of the integrated
SachsWolfe (ISW) effect. Hence, the value of
obtained by the authors who assumed they were observing the ISW effect
would appear to have originated from noise analysis.
Key words: cosmic microwave background  largescale structure of Universe
1 Introduction
Several authors (Cabré et al. 2006; Fosalba et al. 2003; Vielva et al. 2006; Ho et al. 2008; Granett et al. 2008; Raccanelli et al. 2008) have claimed that there is a significant crosscorrelation between cosmic microwave background radiation (CMBR) anisotropies and the density of galaxies, which is interpreted as the integrated SachsWolfe (ISW) effect. An anticorrelation caused by the SunyaevZel'dovich effect would also be expected on scales smaller than , but this is negligible when averaging large regions of the sky (HernándezMonteagudo & RubiñoMartín 2004). The conclusion of these authors is that the measured crosscorrelation should be interpreted as a detection of the ISW effect within a CDMcosmology and it serves to constrain the value of the cosmological parameters.
We reanalyze whether this correlation exists by considering galaxies observed by the Sloan Digital Sky Survey (SDSS), taking particular care in the calculation of the crosscorrelation errors. The root mean square (rms) of the crosscorrelation for distant, widely different areas of sky (here called ``fieldtofield'' errors) infer much larger errors than those calculated using resampling crosscorrelations techniques i.e., when these are determined in different strongly overlapping and thus not independent subsamples of a given sample (resampling errors). We conclude that measurements of the errors in the crosscorrelation function for overlapping subfields lead to an underestimate of the true scatter in the signal.
2 Data
We consider two types of data for the two fields that we crosscorrelate:
 1.
 Microwave temperature anisotropies () from the 5th year WMAP release (Hinshaw et al. 2009). We use the Vband (61 GHz) data because of its lower level of pixel noise. We checked that the results of this paper are approximately similar if we use the Wband (94 GHz) data. There is no need to subtract foreground Galactic contamination because this is not correlated with galaxy counts (corrected for extinction), and because this is small in offplane regions. In any case, the published foreground corrections might not be enough accurate (LópezCorredoira 2007). We assign the same weight to each WMAP pixel of equal size.
 2.
 Galaxy counts (G) are obtained from the survey SDSS, photometric catalog, data release DR7 (Abazajian et al. 2009). They cover an area 11 663 deg^{2} (28% of the sky) mostly in the northern Galactic hemisphere. We did not use the striped region data with to ensure low Galactic extinction and avoided negative latitudes because these are small isolated regions dominated by edge effects. We used only galaxies with r magnitudes in the range [18,21] (Galacticextinction corrected) (within these limits galaxy counts are complete) and ``clean photometry'' according to an SDSS algorithm (e.g., we removed sources close to saturated objects with contamination of their by other objects), and avoiding the borders by 0.3 deg. In this situation, the total used area is 7 349 deg^{2} (18% of the sky), containing 2.2, 6.4 and 17.1 million galaxies in the r magnitude ranges [18,19], [19,20], and [20,21], respectively.
3 Methods
By defining the galaxy count (G) density contrast to be
,
and denoting by
the
fluctuations in the CMBR with respect to the average temperature
T_{0}, the crosscorrelation function can be written as
The estimator of Eq. (1) computes the crosscorrelation to be the average over all pixels with separations , where is the step between successive values of . In what follows, we set deg.
There are two kinds of errors in the crosscorrelation, associated with two distinct ways of constructing subfields over which they are computed (Sylos Labini et al. 2009):
 1.
 Resampling errors: for point distributions, there is a component of the total error that is caused by the finiteness of the number of points and is closely related to that given by the resampling techniques BetancortRijo (1991); however, here, in the correlation of two continuous fields, the association is not at all clear. These may be estimated with a resampling technique, for instance jackknife or bootstrap. In the latter case, we calculate times the crosscorrelation by removing each time a different fraction of the N pixels. By using the bootstrap method, we also calculate a number of times the crosscorrelation that each time chooses the same number N of pixels from the original sample, but randomly selected (so that there are some pixels that are selected several times, while others are not selected at all). Both in bootstrap and jackknife, we then calculate the rms of these resamplings, which provides our error. We use , which implies that the relative error in the rms is % for Gaussian errors. We note that for both techniques the determination have been performed on overlapping subsamples, and they are thus not independent.
 2.
 Fieldtofield fluctuations: these are caused by intrinsic fluctuations in
both the largescale structure of galaxies and the microwave temperature field.
We propose three methods for estimating these fluctuations:
 (a)
 Different fields: We crosscorrelate the G field in the full area with a different field of the same power spectrum as the original WMAP data, although uncorrelated with G. One simple way of applying this method is assigning to the value of its own WMAP data but in other regions of the sky that are completely separated. For instance, we define with different values of (we consider different values: , i=1 to ), and calculate the rms for the realizations. In this case, we use a small enough number () of regions, so the relative error in the rms is 20% for Gaussian errors. The crosscorrelations at scales 60180 might produce some signal, but this would be small, given that the selfcorrelation of is almost zero for Copi et al. (2009). The possible largescale crosscorrelations of the different fields infer that this estimation of the rms value is a conservative upperlimit value.
 (b)
 Monte Carlo simulations of WMAP: We generate a number of Monte Carlo realizations of WMAP by using the software ``synfast'' to generate random mock maps of anisotropies corresponding to the theoretical power spectrum (Hinshaw et al. 2009) filtered for the Vband. We perform realizations, and then calculate the rms of their crosscorrelation with the fixed SDSS galaxy counts map. The relative error in the rms is %.
 (c)
 Integral of the selfcorrelations: A calculation of the
fieldtofield variance in the crosscorrelation of two noncorrelated
fields can be given by (see Appendix A)
where stands for the average extended over all groups of four pixels (1,2,3,4) in a region in which the separation between pixels 1,3 and 2,4 is between and , is the separation of pixels 1,2, is the separation of pixels 3,4, and and are the selfcorrelations, respectively, for the fields and . We note that with this method we assume that and are uncorrelated (as in Monte Carlo simulations); therefore, refers to the limits of pure noncorrelated fields within the corresponding probabilities (68%). In addition, we note that we use the selfcorrelations that we measure in our fields (see Fig. 1), i.e., we have only one realization. Cosmic variance would introduce some extra uncertainty.
4 Results
Figure 1: Loglog of the selfcorrelations of the fields and . 

Open with DEXTER 
Figure 2: Crosscorrelation function WMAPSDSS (black line) for galaxies with and in the magnitude range 18 < r < 19 ( left panel), 19 < r < 20 ( center panel), and 20 < r < 21 ( right panel). The rms value calculated by resampling errors and fieldtofield fluctuations are also plotted. 

Open with DEXTER 
In Fig. 1, we plot the selfcorrelations. In Fig. 2, we show the determination of the crosscorrelation function for different ranges of magnitude, and the errors computed by using resampling errors and fieldtofield determinations. On the one hand, the errors computed by both the bootstrap and the jackknife method are of the same order, and on the other hand the three ``fieldtofield'' methods yield similar results, which however are much larger than the resampling errors. The ``Different fields'' method yields in general a slightly lower rms than the integral of the selfcorrelations, possibly because of small positive largescale correlations, which slightly reduce the dispersion, as mentioned in discussing ``different fields'' in Sect. 3. The ``Monte Carlo'' method might yield slightly higher values of rms than the integral of the selfcorrelations due to the larger amplitude of the lowmultipoles in the theoretical power spectrum.
The fieldtofield fluctuations obtained by using independent determinations of the crosscorrelation function are similar to the amplitude of the detected signal or even larger. Figure 3 illustrates this point by showing that there are no positive average crosscorrelations in a sky region of area more than half of the full angular coverage.
From all these analyses, we cannot exclude the value of being compatible with zero for any within fieldtofield fluctuations. Thus we conclude that there is no significant crosscorrelation detection. This situation is similar to that found for the SDSS 3D selfcorrelation by SylosLabini et al. (2009), who also demonstrated that the fieldtofield fluctuations are of the order of the signal in the previously announced discovery of baryon acoustic oscillations and largescale anticorrelations.
Figure 3: Crosscorrelation ( ) WMAPSDSS, 20<r<21: the average of the whole selected SDSSDR7 area, and the average for and ; error bars represent jackknife resampling errors. 

Open with DEXTER 
5 Comparison with previous works
Other authors who calculated the crosscorrelation between WMAP and SDSS galaxy counts measured a significant signal. For instance, Cabré et al. (2006) measured a value of K and a significant positive for all angles lower than 20 degrees for the subsample 20<r<21 over 5500 square degrees of SDSSDR4. Giannantonio et al. (2008) obtained a value of K for the subsample 18<r<21 of SDSSDR6, excluding the southern Galactic hemisphere and high Galactic extinction regions. In addition, they found a significant positive signal out to degrees. Their values are more or less compatible with our estimate of the crosscorrelation function, within the resampling error bars and taking into account that their subsamples are slightly different. However we do not measure significant crosscorrelations, whereas these authors do a result we cannot explain.
Cabré et al. (2006) and Giannantonio et al. (2008) performed Monte Carlo simulations using mock maps, and obtained similar values or ones only slightly larger than a jackknife. We do not know whether these disagreement are caused by mistakes in their calculations or whether their claim is that resampling errors represent the full errors. Other authors used only jackknife technique errors (e.g., Sawangwit et al. 2009). A similar problem may affect the results of Raccanelli et al. (2008), who measured the crosscorrelation between NVSS radio sources and WMAP anisotropies. Raccanelli et al. (2008) calculate the error in simulating 1000 mock NVSS maps by randomly distributing the unmasked pixels of the true NVSS maps. We are concerned that this process might destroy part of the selfcorrelation of each map, and that the errors might not represent the full fieldtofield fluctuations. There has been considerable discussion of these errors Giannantonio et al. (2008); Cabré et al. (2007). However, against their claims one can infer from the analyses of this paper that: i) jackknife or bootstrap methods do not calculate the whole error; ii) the level fieldtofield fluctuations is as large as the measured average signal. In addition, our conclusion is that the signal is largely dependent on the specific subregion chosen. We find that in the large area of , (3906 square degrees available with SDSSDR7, more than half of the sample) we do not measure any signal, so the average signal of the entire sample must be caused by a fluctuation.
One remarkable aspect of the analysis of WMAP/SDSSDR4 is that Cabré et al. (2006) obtain a 3.6 detection for 20<r<21, while Giannantonio et al. (2008) with a wider areal coverage (SDSSDR6) and broader range of magnitudes, 18<r<21, achieve only a 2.3 detection. This decrease in the significance is unexpected if the signal were real. We also note that some authors calculated the combined signal to noise ratio of different crosscorrelations in different samples, obtaining values over 4, by summing them quadratically Cabré et al. (2006). This is incorrect because they do not take into account the correlation between the samples, thus neglect an important part of the estimated error.
A higher significance in the crosscorrelation of WMAP/SDSS is claimed to be obtained Granett et al. (2008) when only superclusters/supervoids are correlated with WMAP instead of the entire SDSS survey: a value of 4.4. Apart from our questions raised above, we are also concerned about possible a posteriori fitted parameters used to obtain this correlation. For instance, Granett et al. (2008) separate regions of the sky centered on superclusters with radii of 4 degrees; we ask why 4 degrees? These authors illustrate that the significance is only 3.5 for radius 3 deg or 3.8 for 5 deg. The significance also changes with the number of superclusters/supervoids selected, being 4.4 with N=50 but only 2.8 with N=70. A signal to noise ratio of 23 is provided by other authors without any selection of superclusters, so they appear to have considered both a radius and number of superclusters/supervoids that achieves the maximum increase in the signal to noise ratio (from 23 up to to 4.4).
We note that Bielby et al. (2010) measured the correlation of WMAP anisotropies with emissionline galaxies selected photometrically from SDSS and inferred a nonsignificant correlation, with large fieldtofield errors comparable to those we obtain (we are cautious in interpreting their result, however, because the crosscorrelation in the different subfields are not independent and this affects the way in which they have been using to determine the rms). They claim that their result implies that possibly emission line galaxies are more strongly clustered and less correlated with microwave anisotropies, something that is not entirely clear to us. In our opinion, the results of Bielby et al. (2010) of nonsignificant crosscorrelation may be correct and there is unlikely to be a difference in the interaction of galaxies with the background CMBR that is caused entirely by them having emission lines. Sawangwit et al. (2010) failed to measure a significant crosscorrelation between the luminous red galaxies of SDSSDR5 and WMAP. On the other hand, they found some positive correlation of WMAP with 2SLAQ survey, and negative correlation of WMAP with the AAOmega survey. Analyses by HernándezMonteagudo (2008) demonstrated that the crosscorrelation of WMAP/SDSSDR4 should have at least within a signal/noise ratio of 0.71.7, much lower than the significance obtained by the authors cited above. These results might be interpreted as independent confirmations of our results here.
Apart from those analyzing data from SDSS, previous studies reach the general conclusion that the ISW effect was not detected significantly in: (1) crosscorrelations with Xray XRB, Boughn & Crittenden (2003) claiming an absence of the ISW using Xray data; (2) nearinfrared 2MASS, Francis & Peacock (2009) not finding any corresponding ISW signal; or (3) radio sources NVSS, HernándezMonteagudo (2009) casting doubt on the correlation between WMAP and NVSS radio sources, since the cumulative signal to noise ratio of the crosscorrelation with multipoles l<60 is lower than 1, and the ISW itself, since the signal to noise ratio should be around 5 theoretically.
6 Conclusions
We concluded that there is no significant crosscorrelation between the CMBR temperature anisotropies of WMAP and the galaxy counts of the SDSS, and any claims to have detected the ISW effect on the basis of significant crosscorrelation are unjustified. Fieldtofield fluctuations dominate the detected signals. Any detection of signal is very dependent on the selected region of the sky. Other authors erroneously claimed to have detect significant correlations because they had used a particular sky region with a fluctuation that is not representative of the average sky or because they had underestimated the statistical errors by using nonindependent resamplings. If our conclusion is correct, the value of obtained by those authors based on the assumption of observing the ISW effect would have been one induced by noise. Its value would be coincident with the expected value for CDM by chance, and in the spirit of accepting a scientific result when it indeed produces numbers expected a priori. AcknowledgementsWe thank an anonymous referee for useful comments and suggestions. Thanks are given to J. A. RubiñoMartín and R. GénovaSantos from IAC (Tenerife) for helpful comments and help in the use of HEALPIX and SYNFAST software. Thanks are given to Claire Halliday (language editor of A&A) for proofreading of the text. Funding for the SDSS has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the US Department of Energy, the NASA, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web Site is: http.//www.sdss.org/. WMAP is the result of a partnership between Princeton University and NASA's Goddard Space Flight Center. MLC was supported by the Ramón y Cajal Programme of the Spanish Science Ministry.
Appendix A: Fieldtofield errors in the crosscorrelation as an integral of the selfcorrelations for two uncorrelated fields^{}
We consider two continuous random scalar fields, F_{A} and F_{B}, in a space with d dimensions and any topology. Without loss of generality, we shall assume that the mean values (over realization) of both fields is zero and they are not correlated: . On the other hand, the values of each field at two different points , are not independent random variables: , where the average is over realizations. In practice, in most interesting cases the fields are statistically homogeneous and ergodic, so that depends only on , and the correlation may be defined as spatial averages, which is the useful definition since in most cases only one realization is available. If the fields are also statistically isotropic, depends only on . For the following derivation, we shall assume homogeneity and isotropy; the full expression might easily be recovered if needed.
We first derive the ``fieldtofield error'' (i.e., the true error)
for the zero lag estimator:
where . We have replaced the ddimensional volume integral over the sample with a sum over N equal volume cell indexed by iand centered on . We would have to multiply the contribution of the field for each pixel by a weight equal to the volume of the pixel in the case of nonequal volume cells.
For the variance in Eq. (A.1), we have:
(A.2) 
since, by construction, the mean value of E over realizations, , is assumed to be zero. Developing the square of expression (A.1), and taking its average, we have:
(A.3) 
In principle, the factor of 2 should not be there in the case i=j, but this will be negligible in the limit of arbitrarily small cells.
Now, since the fields F_{A} and F_{B} are uncorrelated
=  
=  (A.4) 
where . Thus, we have
where represents the volume of the sample.
The correlation estimator for any nonzero lag is
where . Following Eq. (A.6) using the same procedure as for Eq. (A.1), one obtains
Equation (A.7), and its particular case, Eq. (A.5), infer the variance over realizations of the estimator of the correlation between two uncorrelated fields A, B when any new global realization of both fields is carried out. In the case when we fix the realization of one of the fields while changing the other, Eq. (A.7) is also valid but the selfcorrelation of the fixed realization must be calculated by averaging over pixels in this fixed realization, rather than over realizations. Now, since the estimated selfcorrelation may fluctuate above and below the universal (mean of all realizations) value, it is clear that the variance in the estimator of the crosscorrelation of A and B when one of them is kept fixed may be slightly above or below the one corresponding to the case when both fields fluctuate.
References
 Abazajian, K. N., AdelmanMcCarthy, J. K., Agüeros, M. A., et al. 2009, ApJS, 182, 543 [NASA ADS] [CrossRef] [Google Scholar]
 BetancortRijo, J. E. 1991, A&A, 245, 347 [NASA ADS] [Google Scholar]
 Bielby, R., Shanks, T., Sawangwit, U. et al. 2010, MNRAS, 403, 1261 [NASA ADS] [CrossRef] [Google Scholar]
 Boughn, S. P., & Crittenden, R. G. 2003, in: The emergence of Cosmic Structure, AIP Conf. Proc., 666, 67 [Google Scholar]
 Cabré, A., Gaztañaga, E., Manera, M., Fosalba, P., & Castander, F. 2006, MNRAS, 372, L23 [NASA ADS] [Google Scholar]
 Cabré, A., Fosalba, P., Gaztañaga, E., & Manera, M. 2007, MNRAS, 381, 1347 [NASA ADS] [CrossRef] [Google Scholar]
 Copi, C. J., Huterer, D., & Starkman, G. D. 2009, MNRAS, 399, 295 [NASA ADS] [CrossRef] [Google Scholar]
 Francis, C. L., & Peacock, J. A. 2009, [arXiv:0909.2494] [Google Scholar]
 Fosalba, P., Gaztañaga, E., & Castander, F. J. 2003, ApJ, 597, L89 [NASA ADS] [CrossRef] [Google Scholar]
 Giannantonio, T., Scranton, R., Crittenden, R. G. et al. 2008, Phys. Rev. D, 77, 3520 [Google Scholar]
 Granett, B. R., Neyrinck, M. C., & Szapudi, I. 2008, ApJ, 683, L99 [NASA ADS] [CrossRef] [Google Scholar]
 HernándezMonteagudo, C. 2008, A&A, 490, 15 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
 HernándezMonteagudo, C. 2009, [arXiv:0909.4294] [Google Scholar]
 HernándezMonteagudo, C., & RubiñoMartín, J. A. 2004, MNRAS 347, 403 [NASA ADS] [CrossRef] [Google Scholar]
 Hinshaw, G., Weiland, J. L., Hill, R. S., et al. 2009, ApJS, 180, 225 [NASA ADS] [CrossRef] [Google Scholar]
 Ho, S., Hirata, C., Padmanabhan, N., Seljak, U., & Bahcall, N. 2008, Phys. Rev. D, 78, 3519 [NASA ADS] [CrossRef] [Google Scholar]
 LópezCorredoira, M. 2007, J. Astrophys. Astron. 28, 101 [CrossRef] [Google Scholar]
 Raccanelli, A., Bonaldi, A., Negrello, M. et al. 2008, MNRAS, 386, 2161 [NASA ADS] [CrossRef] [Google Scholar]
 Sawangwit, U., Shanks, T., Cannon, R. D., et al. 2010, MNRAS, 402, 2228 [NASA ADS] [CrossRef] [Google Scholar]
 Sylos Labini, F., Vasilyev, N. L., Baryshev, Yu. V., & LópezCorredoira, M. 2009, A&A, 505, 981 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
 Vielva, P., MartínezGonzález, E., & Tucci, M. 2006, MNRAS, 365, 891 [NASA ADS] [CrossRef] [Google Scholar]
Footnotes
 ...field^{}
 By J. BetancortRijo.
All Figures
Figure 1: Loglog of the selfcorrelations of the fields and . 

Open with DEXTER  
In the text 
Figure 2: Crosscorrelation function WMAPSDSS (black line) for galaxies with and in the magnitude range 18 < r < 19 ( left panel), 19 < r < 20 ( center panel), and 20 < r < 21 ( right panel). The rms value calculated by resampling errors and fieldtofield fluctuations are also plotted. 

Open with DEXTER  
In the text 
Figure 3: Crosscorrelation ( ) WMAPSDSS, 20<r<21: the average of the whole selected SDSSDR7 area, and the average for and ; error bars represent jackknife resampling errors. 

Open with DEXTER  
In the text 
Copyright ESO 2010
Current usage metrics show cumulative count of Article Views (fulltext article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 4896 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.