Effect of redshift bin mismatch on cross correlation between DESI Legacy Imaging Survey and Planck CMB lensing potential

We used the lensing potential maps from the third data release (PDR3)¹¹1https://pla.esac.esa.int/#cosmology of the Planck collaboration (Planck Collaboration et al., 2020b). PDR3 uses the SMICA DX12 CMB maps to reconstruct the lensing potential from the CMB temperature and polarization data covering $\sim 67\%$ of the sky. SMICA (Spectral Matching Independent Component Analysis; Delabrouille et al. 2003) is a foreground component separation method that had the best performance during the Planck foreground cleaning mock challenge Planck Collaboration et al. (2014). For our baseline analysis, we used the minimum-variance CMB lensing potential map (hereafter MV map) estimated from the inverse-variance weighting of all lensing estimators based on different correlations of CMB temperature and polarization maps. In addition, Planck PDR3 also provides lensing potential maps estimated only from the CMB temperature measurements (hereafter TT map) and with deprojection of the Sunyaev-Zeldovich sources (SZ-deproj map from now onwards). We used the TT and SZ-deproj maps in section 8.2.3 to study the dependence of our results on the choice of the CMB lensing potential map.

The lensing potential $\phi$ was converted to lensing convergence $\kappa$, where $\kappa$ is proportional to the two dimensional Laplacian of $\phi$. In spherical harmonic space, this relation can be expressed as (Hu, 2000)

The Planck PDR3 package provides spherical harmonics coefficients for the lensing convergence maps at HEALPix²²2https://healpix.jpl.nasa.gov/ (Górski et al., 2005) resolution parameter $\textit{N}_{\text{side}}=4096$, which we downgraded to $\textit{N}_{\text{side}}=1024$ for our analysis. The data package also provides an estimate of the noise power spectrum $N_{\ell}^{\kappa\kappa}$ for the lensing convergence maps along with binary maps masking the regions of sky not suitable for analysis.

The Dark Energy Spectroscopic Instrument Legacy Imaging Survey (DESI-LIS; Dey et al. 2019) is a combination of three legacy imaging surveys: the Dark Energy Camera Legacy Survey (DECaLS; Blum et al. 2016), the Beijing-Arizona Sky Survey (BASS; Zou et al. 2017) and the Mayall z-band Legacy Survey (MzLS) observed by the Mosaic3 camera (Dey et al., 2016), with the addition of the Dark Energy Survey (DES; The Dark Energy Survey Collaboration 2005). The legacy imaging surveys were motivated by the need to provide targets for the ongoing Dark Energy Spectroscopic Instrument (DESI; DESI Collaboration et al. 2016) survey, along-with supplementing the Sloan Digital Sky Survey (SDSS; eg. Abolfathi et al. 2018; Abazajian et al. 2009) photometry data. In our study, we used the galaxy catalogue and photometric redshifts prepared by H21 from the DESI-LIS Data Release 8³³3http://legacysurvey.org/dr8/, covering a total area of $\sim 17\,800$ deg² with sources observed in three optical bands ($g$, $r$, $z$) and three WISE (Wright et al., 2010) bands $W_{1}$, $W_{2}$ and $W_{3}$. Due to shallower effective depth of the $W_{2}$ and $W_{3}$ bands, H21 used only $W_{1}$ ($3.4\,\mu m$) band, resulting in the following selection criteria applied to the data

1. 1.

   PSF-type objects are excluded which eliminates most stars and quasars.

2. 2.

   Objects are detected in four bands, i.e. FLUX G|R|Z|W1 ¿ 0.

3. 3.

   MW TRANSMISSION G|R|Z|W1 are applied to the fluxes for Galactic extinction correction.

4. 4.

   Magnitude cuts are applied with $g<24$, $r<22$, and $W_{1}<19.5$, where all magnitudes are computed by $m=22.5-2.5\log_{10}\text{(flux)}$.

In addition to above selection criteria, H21 generated completeness map to account for the foreground contaminations at the map pixel level. These foreground contaminants included masks for bright stars, globular clusters and incompleteness in optical bands.

H21 used a direct approach to estimate photometric redshifts of galaxies from observed spectroscopy by assigning a redshift to a given location in multi-colour space. The spectroscopic surveys used to calibrate photometric redshifts included Galaxy And Mass Assembly (GAMA) survey Data Release 2 (Liske et al., 2015), SDSS-III’s Baryon Oscillation Spectroscopic Survey (BOSS) Data Release 12 (Alam et al., 2015), Extended Baryon Oscillation Spectroscopic Survey (eBOSS) Data Release 16 (Ahumada et al., 2020), VIMOS Public Extragalactic Redshift Survey (VIPERS) Data Release 2 (Scodeggio et al., 2018) and DEEP2 (Newman et al., 2013). Two photometric surveys Cosmic Evolution Survey (COSMOS) (Ilbert et al., 2009) and Dark Energy Survey (DES) Y1 redMaGiC (Cawthon et al., 2018) were also included with the spectroscopic surveys for their highly accurate photometric redshifts.

The calibration samples, except DES Y1 redMaGiC, were binned into three-dimensional grids of $g-r$, $r-z$ and $z-W_{1}$ with a pixel width of about $0.03$. The range of the colours used in the three-dimensional grid were $-0.5<g-r<2.5$, $-2<r-z<3$, and $-2<z-W_{1}<4$. Pixels containing more than five objects from the calibration samples were assigned the mean redshift of these objects. The DES Y1 redMaGiC samples were used to fill out empty pixels from the initial binning. When calibrating redshift of galaxies, objects falling in pixels that lack a redshift calibration were excluded. This ensured selection of objects that occupy the same colour space as the calibration sample. To marginalise digitization artefacts in the redshift distribution, the assigned photometric redshifts were added with a random top-hat dither of $\pm 0.005$. The resulting redshift calibration yielded $68\%$ percent of samples having photometric redshifts within $\pm 0.027$ of their spectroscopic redshifts.

The galaxies were divided into $4$ tomographic bins based on their photometric redshifts with intervals $(0.0,0.3,0.45,0.6,0.8]$. The galaxy number count maps and the photometric redshift data used in this study are publicly available at https://gitlab.com/qianjunhang/desi-legacy-survey-cross-correlations. A summary of the four tomographic bins including number of objects, mean density of objects per pixel and per steradian is given in Table 1. The galaxy over-density maps for every tomographic bin were build at the HEALPix resolution parameter $\textit{N}_{\text{side}}=1024$ using the relation

where $n(\hat{\textbf{n}})$ is the completeness corrected number of galaxies at angular position $\hat{\textbf{n}}$ and $\overline{n}$ is the mean number of objects. The galaxy over-density maps smoothed with a Gaussian beam of $60^{\prime}$ FWHM (for illustrative purpose only) are shown in Fig. 1.

For the $i^{\text{th}}$ redshift bin, the true redshift distribution $\frac{\mathrm{d}N^{i}}{\mathrm{d}z_{t}}$ was estimated using the observed photometric redshift distribution $\frac{\mathrm{d}N(z_{p})}{\mathrm{d}z_{p}}$ and the redshift error distribution $p^{i}(z_{s}-z_{p}|z_{p})$ using the relation

where $W^{i}(z_{p})$ is the window function defining the boundaries of the $i^{\text{th}}$ redshift bin

In Fig. 3, we show the galaxy redshift distributions estimated using Eq. (4). The left and the right panels show the redshift distributions estimated with modified Lorentzian fit of the error distribution and the sum of Gaussians model, respectively. The sum of Gaussians model produces broader tails in the redshift distributions than the modified Lorentzian function, which suggests stronger leakage of objects across redshift bins.

The true redshift distribution $\frac{\mathrm{d}N}{\mathrm{d}z_{t}}$ in the redshift range $0\leq z\leq 0.8$ can be estimated by the deconvolution method following the Fourier ratio approach (Saraf & Bielewicz, 2023)

where $\frac{\mathrm{d}N(z_{p})}{\mathrm{d}z_{p}}$ is the observed photometric redshift distribution and $p(z_{p}-z_{s}|z_{s})$ is the redshift error distribution built from the spectroscopic calibration sample (section 2.2). We used the deconvolution method to compute the scattering matrix in section 8.2.2.

The analytical covariance matrix between Gaussian fields $A,B,C$ and $D$ estimated using Saraf et al. (2022) is given by

where $\{A,B,C,D\}\in\{\kappa,g\}$, $f_{\text{sky}}^{AB}$ is the fraction of sky common to fields $A$ and $B$. $f^{AC,BD}$ is the composite quantity that acts as a weight when taking into account different observed sky areas for different fields. The analytical covariance has the advantage that sampling fluctuations, known as cosmic variance are already included in the matrix. It can also be computationally fast to estimate, using arbitrary theoretical power spectra.

From the $300$ simulations of the galaxy density and CMB convergence fields (as described in section 6), one can compute an unbiased estimator of the covariance matrix as

where $\overline{C}_{L}$ is the average power spectrum from $300$ Monte Carlo simulations, $\hat{C}_{L}^{i}$ represents the $i^{th}$ simulated power spectrum estimate and $\text{N}_{\text{sim}}$ is the number of simulations.

The advantage of jackknife method is that it is independent of any assumptions about the cosmological model. As this method uses data for the estimation of the covariance matrix, it also naturally takes into account the different survey selections effects and any unforeseen systematic errors. However, the jackknife method assumes that the observed data is an accurate representation of the distribution of measurements and is inherently blind to cosmic variance.

To apply the jackknife method, we followed the delete-one approach and make $\text{N}_{\text{sub}}=90$ disjoint partitions of the DESI-LIS survey footprint. We then computed the angular power spectra by omitting one sub-sample at a time. If we denote by $C_{L,i}$ the angular power spectrum computed by removing the $i^{\text{th}}$ sub-sample, the covariance matrix can be estimated as (Norberg et al., 2009)

where $\overline{C}_{L}$ is the mean power spectrum over all sub-samples.

H21 estimated the covariance matrix from the errors computed directly from the observed data. The observed angular power spectra were binned into groups with multipole binwidth $\Delta\ell=10$

The error on the binned data points were computed by

The diagonal covariance matrix is then given by $\text{Cov}_{LL^{\prime}}=\text{diag}(\sigma_{\ell}^{2})$.

In Fig. 4 we show the power spectrum errors estimated as square root of the diagonal elements of the covariance matrix from different methods for the redshift slice $0.45\leq z<0.6$. We found that the internal methods, i.e. jackknife method and covariance matrix from H21 showed larger contributions to the diagonal elements of the covariance matrix. The analytical covariance and the sample covariance from mock realisations were underestimated suggesting that the nuances of the data are not captured by these methods. We found similar results for other redshift bins and chose to use the covariance matrix from H21 for the estimation of parameters.

In section 6, we described the mock catalogues prepared using the code FLASK with DESI-LIS observed photometric redshift errors. We validated our pipeline and estimated the impact of redshift bin mismatch of DESI-LIS galaxies in this section. We followed the procedure from Saraf & Bielewicz (2023) to estimate galaxy bias and cross correlation amplitude from simulated photometric datasets. In the left panel of Fig. 5, we compare the galaxy linear bias from the average power spectra of $300$ simulations before (blue circles) and after (red squares) accounting for redshift bin mismatch of objects. The average power spectra and relative differences between the averages and best-fit theoretical spectra are shown in Appendix A (Figs. 16 and 17, respectively). In the right panel of Fig. 5, we show the deviations between the estimated amplitude of cross correlation from its true value ($A=1$) in terms of standard deviation of the amplitude. Before leakage correction, the galaxy bias were estimated higher with $7-42\,\sigma$ deviations in three out of four tomographic bins. The second redshift bin gives a $\sim 30\sigma$ lower estimate of the galaxy bias. Such large deviations in terms of errors are a consequence of small uncertainties on the galaxy bias constrained from the galaxy auto-power spectrum. The amplitude of cross correlations was found smaller than unity by $\sim 2\sigma,\sim 0.4\sigma\text{ and }\sim 3\sigma$ in three redshift bins, and higher by $\sim 2\sigma$ in the second redshift bin. The estimated parameters closely follow their expected values after accounting for the leakage across redshift bins.

In this section, we present the measurements of angular power spectra and parameter estimates from cross correlation of DESI-LIS photometric galaxy catalogue with Planck CMB lensing potential maps.