Dark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies
B. Hoyle, D. Gruen, G. M. Bernstein, M. M. Rau, J. De Vicente, W. G. Hartley, E. Gaztanaga, J. DeRose, M. A. Troxel, C. Davis, A. Alarcon, N. MacCrann, J. Prat, C. Sánchez, E. Sheldon, R. H. Wechsler, J. Asorey, M. R. Becker, C. Bonnett, A. Carnero Rosell, D. Carollo, M. Carrasco Kind, F. J. Castander, R. Cawthon, C. Chang, M. Childress, T. M. Davis, A. Drlica-Wagner, M. Gatti, K. Glazebrook, J. Gschwend, S. R. Hinton, J. K. Hoormann, A. G. Kim, A. King, K. Kuehn, G. Lewis, C. Lidman, H. Lin, E. Macaulay, M. A. G. Maia, P. Martini, D. Mudd, A. Möller, R. C. Nichol, R. L. C. Ogando, R. P. Rollins, A. Roodman, A. J. Ross, E. Rozo, E. S. Rykoff, S. Samuroff, I. Sevilla-Noarbe, R. Sharp, N. E. Sommer, B. E. Tucker, S. A. Uddin, T. N. Varga, P. Vielzeuf, F. Yuan, B. Zhang, T. M. C. Abbott, F. B. Abdalla, S. Allam, J. Annis, K. Bechtol, A. Benoit-Lévy, E. Bertin, D. Brooks, E. Buckley-Geer, D. L. Burke, M. T. Busha, D. Capozzi, J. Carretero, M. Crocce, C. B. D'Andrea, L. N. da Costa, D. L. DePoy, S. Desai, H. T. Diehl, P. Doel, T. F. Eifler, J. Estrada, A. E. Evrard, E. Fernandez, B. Flaugher, P. Fosalba, J. Frieman, J. García-Bellido, D. W. Gerdes, T. Giannantonio, D. A. Goldstein, R. A. Gruendl, G. Gutierrez, K. Honscheid, D. J. James, M. Jarvis, T. Jeltema, M. W. G. Johnson, M. D. Johnson, et al. (38 additional authors not shown)
DDES 2017-0260Fermilab PUB-17-293-AE
MNRAS , 000–000 (0000) Preprint 15 May 2018 Compiled using MNRAS L A TEX style file v3.0
Dark Energy Survey Year 1 Results:Redshift distributions of the weak lensing source galaxies
B. Hoyle , (cid:63) , D. Gruen , † , G. M. Bernstein , M. M. Rau , J. De Vicente , W. G. Hartley , , E. Gaztanaga ,J. DeRose , , M. A. Troxel , , C. Davis , A. Alarcon , N. MacCrann , , J. Prat , C. S´anchez , E. Sheldon ,R. H. Wechsler , , , J. Asorey , , M. R. Becker , , C. Bonnett , A. Carnero Rosell , , D. Carollo , , M. Car-rasco Kind , , F. J. Castander , R. Cawthon , C. Chang , M. Childress , T. M. Davis , , A. Drlica-Wagner ,M. Gatti , K. Glazebrook , J. Gschwend , , S. R. Hinton , J. K. Hoormann , A. G. Kim , A. King ,K. Kuehn , G. Lewis , , C. Lidman , , H. Lin , E. Macaulay , M. A. G. Maia , , P. Martini , , D. Mudd ,A. M¨”oller , , R. C. Nichol , R. L. C. Ogando , , R. P. Rollins , A. Roodman , , A. J. Ross , E. Rozo ,E. S. Rykoff , , S. Samuroff , I. Sevilla-Noarbe , R. Sharp , N. E. Sommer , , B. E. Tucker , , S. A. Uddin , ,T. N. Varga , , P. Vielzeuf , F. Yuan , , B. Zhang , , T. M. C. Abbott , F. B. Abdalla , , S. Allam ,J. Annis , K. Bechtol , A. Benoit-L´evy , , , E. Bertin , , D. Brooks , E. Buckley-Geer , D. L. Burke , ,M. T. Busha , D. Capozzi , J. Carretero , M. Crocce , C. B. D’Andrea , L. N. da Costa , , D. L. DePoy ,S. Desai , H. T. Diehl , P. Doel , T. F. Eifler , , J. Estrada , A. E. Evrard , , E. Fernandez , B. Flaugher ,P. Fosalba , J. Frieman , , J. Garc´ıa-Bellido , D. W. Gerdes , , T. Giannantonio , , , D. A. Goldstein , ,R. A. Gruendl , , G. Gutierrez , K. Honscheid , , D. J. James , M. Jarvis , T. Jeltema , M. W. G. Johnson ,M. D. Johnson , D. Kirk , E. Krause , S. Kuhlmann , N. Kuropatkin , O. Lahav , T. S. Li , M. Lima , ,M. March , J. L. Marshall , P. Melchior , F. Menanteau , , R. Miquel , , B. Nord , C. R. O’Neill , ,A. A. Plazas , A. K. Romer , M. Sako , E. Sanchez , B. Santiago , , V. Scarpine , R. Schindler , M. Schubnell ,M. Smith , R. C. Smith , M. Soares-Santos , F. Sobreira , , E. Suchyta , M. E. C. Swanson , G. Tarle ,D. Thomas , D. L. Tucker , V. Vikram , A. R. Walker , J. Weller , , , W. Wester , R. C. Wolf , B. Yanny ,J. Zuntz (DES Collaboration)
15 May 2018
ABSTRACT
We describe the derivation and validation of redshift distribution estimates and theiruncertainties for the populations of galaxies used as weak lensing sources in the DarkEnergy Survey (DES) Year 1 cosmological analyses. The Bayesian Photometric Red-shift (BPZ) code is used to assign galaxies to four redshift bins between z ≈ . ≈ .
3, and to produce initial estimates of the lensing-weighted redshift distributions n i PZ ( z ) ∝ d n i / d z for members of bin i . Accurate determination of cosmological pa-rameters depends critically on knowledge of n i but is insensitive to bin assignmentsor redshift errors for individual galaxies. The cosmological analyses allow for shifts n i ( z ) = n i PZ ( z − ∆ z i ) to correct the mean redshift of n i ( z ) for biases in n i PZ . The∆ z i are constrained by comparison of independently estimated 30-band photometricredshifts of galaxies in the COSMOS field to BPZ estimates made from the DES griz fluxes, for a sample matched in fluxes, pre-seeing size, and lensing weight to the DESweak-lensing sources. In companion papers, the ∆ z i of the three lowest redshift binsare further constrained by the angular clustering of the source galaxies around redgalaxies with secure photometric redshifts at 0 . < z < .
9. This paper details theBPZ and COSMOS procedures, and demonstrates that the cosmological inference isinsensitive to details of the n i ( z ) beyond the choice of ∆ z i . The clustering and COS-MOS validation methods produce consistent estimates of ∆ z i in the bins where bothcan be applied, with combined uncertainties of σ ∆ z i = 0 . , . , . , and 0 .
022 inthe four bins. Repeating the photo- z proceedure instead using the Directional Neigh-borhood Fitting (DNF) algorithm, or using the n i ( z ) estimated from the matchedsample in COSMOS, yields no discernible difference in cosmological inferences. Key words: catalogues: Astronomical Data bases, surveys: Astronomical Data bases,methods: data analysis: Astronomical instrumentation, methods, and techniques
Affiliations are listed at the end of the paper. (cid:63) corresponding author: [email protected] † corresponding author: [email protected]; Einstein fellowc (cid:13) a r X i v : . [ a s t r o - ph . C O ] M a y DES Collaboration
The Dark Energy Survey (DES) Year 1 (Y1) data placesstrong constraints on cosmological parameters (DES Col-laboration et al. 2017) by comparing theoretical models tomeasurements of (1) the auto-correlation of the positionsof luminous red galaxies at 0 . < z < . redMaGiC algorithm (Rozoet al. 2016); (2) the cross-correlations among weak lensingshear fields (Troxel et al. 2017) inferred from the measuredshapes of “source” galaxies divided into four redshift bins(Zuntz et al. 2017); and (3) the cross-correlations of sourcegalaxy shapes around the redMaGiC (“lens”) galaxy posi-tions (Prat et al. 2017). There are 650,000 galaxies in the redMaGiC catalog covering the 1321 deg DES Y1 analysisarea, and 26 million sources in the primary weak lensing cat-alog. For both the lens and the source populations, we relyon DES photometry in the griz bands to assign galaxies toa redshift bin i . Then we must determine the normalized dis-tribution n i ( z ) of galaxies in each bin. This paper describeshow the binning and n i ( z ) determination are done for thesource galaxies. These redshift distributions are fundamen-tal to the theoretical predictions of the observable lensingsignals. Uncertainties in the n i ( z ) must be propagated intothe cosmological inferences, and should be small enough thatinduced uncertainties are subdominant to other experimen-tal uncertainties. The bin assignments of the source galaxiescan induce selection biases on the shear measurement, sowe further discuss in this paper how this selection bias isestimated for our primary shear measurement pipeline. Theassignment of redshifts to the lens galaxies, and validationof the resultant lens n i ( z )’s, are described elsewhere (Rozoet al. 2016; Elvin-Poole et al. 2017; Cawthon et al. 2017).A multitude of techniques have been developed for es-timation of redshifts from broadband fluxes (e.g. Arnoutset al. 1999; Ben´ıtez 2000; Bender et al. 2001; Collister &Lahav 2004; Feldmann et al. 2006; Ilbert et al. 2006; Hilde-brandt et al. 2010; Carrasco Kind & Brunner 2013; S´anchezet al. 2014; Rau et al. 2015; Hoyle 2016; Sadeh, Abdalla &Lahav 2016; De Vicente, Sanchez & Sevilla-Noarbe 2016).These vary in their statistical methodologies and in theirrelative reliance on physically motivated assumptions vs em-pirical “training” data. The DES Y1 analyses begin with aphotometric redshift algorithm that produces both a pointestimate—used for bin assignment—and an estimate p PZ ( z )of the posterior probability of the redshift of a galaxy givenits fluxes—used for construction of the bins’ n i ( z ) . The key challenge to use of photo- z ’s in cosmological in-ference is the validation of the n i ( z ), i.e. the assignment ofmeaningful error distributions to them. The most straight-forward method, “direct” spectroscopic validation, is to ob-tain reliable spectroscopic redshifts for a representative sub-sample of the sources in each bin. Most previous efforts atconstraining redshift distributions for cosmic shear analy-ses used spectroscopic redshifts either as the primary vali-dation method, or to derive the redshift distribution itself(Benjamin et al. 2013; Jee et al. 2013; Schmidt & Thor-man 2013; Bonnett et al. 2016; Hildebrandt et al. 2017). While there is Y band data available, due to its lower depth,strong wavelength overlap with z , and incomplete coverage, wedid not use it for photo- z estimation. Direct spectroscopic validation cannot, however, currentlyreach the desired accuracy for deep and wide surveys like theY1 DES, because the completeness of existing spectroscopicsurveys is low at the faint end of the DES source-galaxy dis-tribution (Bonnett et al. 2016; Gruen & Brimioulle 2017),and strongly dependent on quantities not observed by DES(Hartley et al. in preparation). In detail the larger area ofthe DES Y1 analysis compared to other weak lensing sur-veys, including the DES SV analysis (Bonnett et al. 2016),reduces the statistical uncertainties such that the system-atic uncertainties from performing a direct calibration usingspectra become dominant.The validation for DES Y1 source galaxies thereforeuses high-precision redshift estimates from 30-band photom-etry of the COSMOS survey field (Laigle et al. 2016), whichare essentially complete over the color-magnitude space ofthe Y1 source catalog, in a more sophisticated version ofthe approach used in Bonnett et al. (2016). This direct ap-proach is then combined with constraints on n i ( z ) derivedfrom cross-correlation of the source galaxy positions withthe redMaGiC galaxy positions as an independent methodof photometric redshift validation (see, e.g. Newman 2008for an introduction to the method and Gatti et al. 2017;Cawthon et al. 2017; Davis et al. 2017a for the applicationto DES Y1). The cross-correlation redshift technique will bereferred to as “WZ,” and the validation based on the 30-band COSMOS photometric redshifts will be referred to as“COSMOS,” and the estimates returned from photo- z algo-rithms run on the DES griz photometry will be marked as“PZ.” Indeed we suggest reading this paper in conjunctionwith those of Gatti et al. 2017; Davis et al. 2017a, whichare dedicated to documenting the WZ procedure in greaterdetail. We also summarise the salient parts of these papersthroughout this manuscript and discuss the issue of the fail-ure of the redMaGiC sample to span the full redshift rangeof the Y1 lensing sources, which leaves gaps in our knowl-edge of n i derived from WZ.For the analysis in this work, the cosmological inferencewill assume that the redshift distribution in bin i is givenby n i ( z ) = n i PZ ( z − ∆ z i ) , (1)where n i PZ ( z ) is the distribution returned from the photo-metric redshift code using DES griz photometry, and ∆ z i isa free parameter to correct any errors resembling a shift ofthe photo- z result (see also, Jee et al. 2013; Bonnett et al.2016). The cosmological inference code is given a probabilitydistribution for ∆ z i , which is the normalized product of theprobabilities returned by the WZ and COSMOS analyses.It is apparent that Equation (1) essentially allows the meansource redshift returned by the PZ method to be alteredby the information provided by the COSMOS and WZ val-idation procedures, but the shape of n i ( z ) about its meanretains its PZ determination.This paper begins in § § §
4. Thederivation of WZ constraints from angular clustering is thesubject of Gatti et al. (2017), Cawthon et al. (2017), andDavis et al. (2017a). In § MNRAS , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies on ∆ z i with those from COSMOS to yield the final con-straints. We describe the use of these redshift constraintsas priors for the DES Y1 cosmological inference, includingan examination of the impact of the assumption in Equa-tion (1) and other known shortcomings in our process, in § § n i ( z ) estimation and validation proce-dure not immediately required for Y1 lensing analyses willbe described in Hoyle et al. (in preparation) and Rau et al.(in preparation). Estimation and validation of the binning and n i ( z ) functionsfor the Y1 source galaxies require input photometry for thesegalaxies of course, but also Dark Energy Camera (DECam,Flaugher et al. 2015) data (S´anchez et al. 2014) and externaldata on the COSMOS field used for validation. Finally, ourvalidation uses simulations of the COSMOS catalog to esti-mate sample-variance uncertainties induced by the small skyarea of this field. Fluxes and photo- z ’s must be estimatedfor these simulated galaxies. The set of galaxies for which bin assignments and n i ( z ) es-timates are desired are the weak lensing (WL) sources de-fined in the Y1 shear catalogs documented in Zuntz et al.(2017). The primary shear catalog for DES Y1 is producedby the metacalibration algorithm (Huff & Mandelbaum2017; Sheldon & Huff 2017), and a secondary catalog using im3shape (Zuntz et al. 2013) is used as a cross-check. Forboth shear catalogs, we use a common photo- z catalog basedon our best measurements of fluxes (the “MOF” catalog de-scribed below) to estimate the n i ( z ) of each bin (see § n i ( z ) differ, however, because metacali-bration and im3shape implement distinct selection criteriaand bin assignments.The starting point for either shear catalog is the Y1 Gold catalog of sources reliably detected on the sum of the r, i , and z -band DES images (Drlica-Wagner et al. 2017).Detection and initial photometry are conducted by the SEx-tractor software (Bertin & Arnouts 1996). Photomet-ric zeropoints are assigned to each DES exposure usingnightly solutions for zeropoints and extinction coefficientsderived from standard-star exposures. Exposures from non-photometric nights are adjusted to match those taken inphotometric conditions.As detailed in Drlica-Wagner et al. (2017), the photo-metric calibration is brought to greater color uniformity andadjusted for Galactic extinction by stellar locus regression(SLR, Ivezi´c et al. 2004; MacDonald et al. 2004; High et al.2009): the i -band fluxes are adjusted according to the Galac-tic extinction implied by the Schlegel, Finkbeiner & Davis(1998) dust map with the O’Donnell (1994) extinction law.Then the zeropoints of other bands are adjusted to force thestellar color-color loci to a common template.Fluxes used as input to the photo- z programs for both shear catalogs are derived using ngmix (Sheldon 2014;Jarvis et al. 2016), which fits a model to the pixel val-ues of each galaxy in the Gold catalog. The ngmix codefits a highly constrained exponential+deVaucouleurs modelto each galaxy: the model is convolved with each expo-sure’s point-spread function (PSF) and compared to pix-els from all individual exposures covering the source. Thefitting is multi-epoch and multi-band: pixels of all expo-sures in all bands are fit simultaneously, assuming commongalaxy shape for all bands and a single free flux per band.The fitting is also multi-object: groups of overlapping galaxyimages are fit in iterative fashion, with each fit to a givengalaxy subtracting the current estimate of flux from neigh-bors. These “multi-object fitting” (MOF) fluxes are used asinput to photo- z estimators for im3shape and metacali-bration catalog member galaxies (although we use a dif-ferent flux measurement for bin assignment in the case of metacalibration , see below).The photo- z assigned to a galaxy depends on its mea-sured multi-band fluxes, which will vary if there is shear ap-plied to the galaxy. So the photo- z bin to which a galaxy isassigned might depend on how much it is sheared, leading toa potential selection bias. For im3shape , we have confirmed,using realistic image simulations, that these selection biasesare small (at or below the one per cent level), and have addeda term in the systematic uncertainty of the shear calibrationto account for them (cf. section 7.6.2 of Zuntz et al. 2017,called variation of morphology there). metacalibration ,on the contrary, can estimate and correct selection biaseson the WL shear inference by producing and re-measuringfour artificially sheared renditions of each target galaxy (by γ = ± .
01 and γ = ± .
01, where γ , are the two compo-nents of the shear). The selection bias correction in meta-calibration requires knowing whether each source wouldhave been selected and placed in the same bin if it had beensheared. It is thus necessary for us to run the photo- z es-timation software not only on the original fluxes, but alsoon fluxes measured for each of the four artificially shearedrenditions of each galaxy. The latter are not available fromthe MOF pipeline.For the metacalibration catalog, we therefore pro-duce an additional set of photo- z estimates based on a dif-ferent flux measurement made with the metacalibration pipeline. This measurement makes use of a simplified ver-sion of the ngmix procedure described above: the model fitto the galaxies is a PSF-convolved Gaussian, rather than asum of exponential and deVaucouleurs components. These“Metacal” fluxes do not subtract neighbors’ flux. In addi-tion to fluxes, metacalibration also measures pre-seeinggalaxy sizes and galaxy shapes (Zuntz et al. 2017).There are thus 6 distinct photo- z ’s for the WL sourcegalaxies: one produced using the MOF fluxes for galaxies ineither of the im3shape or metacalibration shape catalogs;one produced using Metacal fluxes of the as-observed sourcesin the metacalibration shape catalog; and four producedusing Metacal fluxes of the four artificially sheared rendi-tions of the sources in the metacalibration catalog. https://github.com/esheldon/ngmixMNRAS , 000–000 (0000) DES Collaboration griz
Our COSMOS validation procedure depends on having griz photometry and external redshift estimates for objects inthe COSMOS field. This field was observed by DES andby community programs using DECam. These observationswere combined, cataloged, and measured using the sameDES pipelines as the survey data; we use the
Y1A1 D04 cat-alog produced as part of the
Gold catalogs (Drlica-Wagneret al. 2017). MOF magnitudes and Metacal sizes are alsomeasured for all entries in this catalog. The COSMOS-fieldobservations used herein are ≈ n i ( z ).We must keep in mind, however, that the COSMOS data isbased on a single realization of SLR errors, and must there-fore allow for the consequent offset of COSMOS photometryfrom the Y1 mean ( § The COSMOS2015 catalog from Laigle et al. (2016) providesphotometry in 30 different UV/visible/IR bands, and proba-bility distribution functions (PDFs) p C30 ( z ) for the redshiftof each galaxy based on this photometry using the LeP-hare template-fitting code (Arnouts et al. 1999; Ilbert et al.2006). Typical p C30 ( z ) widths for DES source galaxies are ≈ . z ) , far better than the uncertainties in BPZ es-timates based on DES griz photometry. In § p C30 ( z )’s on our ∆ z i inferences.The validation procedure requires assignment of a p C30 ( z ) to each DES-detectable source in the COSMOS field.After limiting the catalogs to their region of overlap, we as-sociate COSMOS2015 objects with DES Gold objects with1 . (cid:48)(cid:48) p C30 ( z ) provided in COSMOS2015, without expla-nation. For these we synthesize a p C30 ( z ) by averaging thoseof ≈
10 nearest neighbors in the space of COSMOS2015
ZMINCHI2 and i -band magnitude, where ZMINCHI2 is the30-band photometric redshift point prediction correspond-ing to the the minimum χ fit between fluxes and templates.We remove from the sample galaxies whose fluxes orpre-seeing sizes could not be measured by the DES pipelines.We note that such objects would be flagged in the lensingsource catalog and removed. A total of 128,563 galaxies with Figure 1.
The effect of rescaling the COSMOS2015 photomet-ric redshift PDFs using the Probability Integral Transform (PIT)distribution. The PIT is the redshift cumulative distribution func-tion (CDF) values of the full sample of DES-detected sourcesevaluated at the spectroscopic redshift, for those sources withknown z spec . The original PDFs (blue) depart significantly fromthe expected uniform distribution (red dashed line). The p C30 ( z )rescaling procedure yields the orange histogram, much improved,as confirmed by the value of the Kullback-Leibler divergence be-tween the histogram and a uniform distribution. good DES Gold
MOF photometry remain in our final COS-MOS sample.We also use spectroscopic subsamples of this completesample of galaxies with COSMOS2015 results later to vali-date our calibration (cf. § Following a technique similar to Bordoloi, Lilly & Amara(2010), we rescale the estimated p C30 ( z )’s to make themmore accurately represent true distribution functions of red-shift.The method relies on using the subset of COSMOS2015galaxies with spectroscopic redshifts from the literature(Lilly et al. 2007, 2009a). While this subset is not represen-tative of the full photometric sample (Bonnett et al. 2016;Gruen & Brimioulle 2017), an excess of outliers in true, spec-troscopic redshift relative to p C30 ( z ) is still an indicationthat the rate of “catastrophic failures” in COSMOS2015photo- z determinations is higher than that estimated byLaigle et al. (2016). The procedure described here is nota panacea but will lessen such discrepancies.For each galaxy in COSMOS2015 having a spectro-scopic redshift and matching a DES detection, p C30 ( z ) isintegrated to a cumulative distribution function (CDF) 0 14 and 0 . → . 13, respec-tively, with even greater improvement for the full sample asnoted in Figure 1. The only parameter relevant for the meanredshift calibration performed in § p ( z ), µ . The sizes of these in each magnitude bin areall | µ | ≤ . We also draw upon simulated data sets generated specificallyfor the DES collaboration. Specifically, we make use of the Buzzard-v1.1 simulation, a mock DES Y1 survey createdfrom a set of dark-matter-only simulations. This simulationand the galaxy catalog construction are described in detailelsewhere (DeRose et al. 2017; Wechsler et al. 2017; Mac-Crann et al. 2017), so here we provide only a brief overview. Buzzard-v1.1 is constructed from a set of 3 N -body sim-ulations run using L-GADGET2 , a version of GADGET2 modified for memory efficiency, with box lengths rangingfrom 1–4 h − Gpc from which light-cones were constructedon the fly.Galaxies are added to the simulations using the AddingDensity Dependent GAlaxies to Light-cone Simulations al-gorithm [ ADDGALS , Wechsler et al. 2017]. Spectral en-ergy distributions (SEDs) are assigned to the galaxies froma training set of spectroscopic data from SDSS DR7 (Cooperet al. 2011) based on local environmental density. TheseSEDs are integrated in the DES pass bands to generate griz magnitudes. Galaxy sizes and ellipticities are drawn fromdistributions fit to SuprimeCam i (cid:48) -band data (Miyazakiet al. 2002). The galaxy positions, shapes and magnitudesare then lensed using the the multiple-plane ray-tracing code, Curved-sky grAvitational Lensing for CosmologicalLight conE simulatioNS [CALCLENS, Becker (2013)]. Thesimulation is cut to the DES Y1 footprint, and photomet-ric errors are applied to the lensed magnitudes by copyingthe noise map of the FLUX AUTO measurements in the realcatalog. More explicitly, the error on the observed flux isdetermined only by the limiting magnitude at the positionof the galaxy, the exposure time, and the noise-free apparentmagnitude of the galaxy itself. The source-galaxy samples in simulations are selected so asto roughly mimic the selections and the redshift distribu-tions of the metacalibration shear catalog described inZuntz et al. (2017). This is done by first applying flux andsize cuts to the simulated galaxies so as to mimic the thresh-olds used in the Y1 data by using the Y1 depth and PSFmaps. The weak lensing effective number density n eff in thesimulation is matched to a preliminary version of the shapecatalogs, and is about 7 per cent higher than for the final,unblinded metacalibration catalog. Truth values for red-shift, flux and shear are of course available as well as thesimulated measurements.COSMOS-like catalogs are also generated from the Buz-zard simulated galaxy catalogs by cutting out 367 non-overlapping COSMOS-shaped footprints from the simula-tion. In this section we describe the process of obtaining photo-metric redshifts for DES galaxies. We note that we only usethe g, r, i, z DES bands in this process. We have found thatthe Y band adds little to no predictive power. Posterior probabilities p PZ ( z ) were calculated for eachsource galaxy using BPZ ∗ , which is a variant of the Bayesianalgorithm described by Ben´ıtez (2000), and has been mod-ified to provide the photometric redshift point predictionsand PDFs required by the DES collaboration directly from fits -format input fluxes, without intermediate steps. The BPZ ∗ code is a distilled version of the distributed BPZ code, and in particular assumes the synthetic template filesfor each filter have already been generated. Henceforth wewill refer to these simply as “BPZ” results. The redshift posterior is calculated by marginalising overa set of interpolated model spectral templates, where thelikelihood of a galaxy’s photometry belonging to a giventemplate at a given redshift is computed via the χ betweenthe observed photometry and those of the filter passbandsintegrated over the model template. The model templatesare grouped into three classes, nominally to represent ellip-tical, spiral and star-burst galaxies. These classes, it is as-sumed, follow distinct redshift-evolving luminosity functionswhich can be used to create a magnitude-dependent prior on MNRAS , 000–000 (0000) DES Collaboration the redshift posterior of each object, a.k.a. the “luminosityprior”. The prior comprises two components, a spectral classprior which is dependent only on observed magnitude, andthe redshift prior of each class—which is itself also magni-tude dependent (see Ben´ıtez 2000 for more detail).Six base template spectra for BPZ are generated basedon original models by Coleman, Wu & Weedman (1980) andKinney et al. (1996). The stellar locus regression used for theDES Y1 data ensures uniformity of color across the foot-print, but there may be small differences in calibration withrespect to the empirical templates we wish to use. Moreover,these original templates are derived from galaxies at redshiftzero, while our source galaxies cover a wide range in redshift,with an appreciable tail as high as z ∼ . 5. The colors ofgalaxies evolve significantly over this redshift range, even atfixed spectral type. Failure to account for this evolution caneasily introduce biases in the redshift posteriors that sub-sequently require large model bias corrections (see Bonnettet al. 2016, for instance). To address these two issues, wecompute evolution/calibration corrections to the templatefluxes.We match low-resolution spectroscopic redshifts fromthe PRIMUS DR1 dataset (Coil et al. 2011; Cool et al.2013) to high signal-to-noise DES photometry and obtainthe best fit of the six basic templates to each of the highestquality PRIMUS objects (quality = 4) at their spectroscopicredshift. The flux of each template in each filter is then cor-rected as a function of redshift by the median offset betweenthe DES photometry and the template prediction, in a slid-ing redshift window of width δz = 0 . 06. The calibration sam-ple numbers 72,176 galaxies and reaches the full depth of ourscience sample ( i DES < . 5) while maintaining a low rateof mis-assigned redshifts. Although the incompleteness inPRIMUS is broadly independent of galaxy color (Cool et al.2013) and each template is calibrated separately, we never-theless expect small residual inaccuracies in our calibrationto remain. Our COSMOS and WZ validation strategies serveto calibrate such errors in BPZ assignments.A complete galaxy sample is required for deriving theluminosity prior we use with BPZ. No spectroscopic samplesare complete to the limit of our source galaxy sample, andso we turn to the accurate photometric redshift sample inthe COSMOS field from Laigle et al. (2016), which is com-plete to the depth of our main survey area despite beingselected in the K -band. The prior takes the form of smoothexponential functions (see Ben´ıtez 2000), which we fit to theCOSMOS galaxy population by determining galaxy typesat their photometric redshift. Because BPZ uses smoothfunctions rather than the population directly, the luminos-ity prior used for obtaining posterior redshift probabilitiesdoes not replicate the high-frequency line-of-sight structurein the COSMOS field.BPZ is run on the MOF fluxes (see § 2) to determine p PZ ( z ) for metacalibration and im3shape , while for thefive metacalibration catalogs—the real one and the four The outlier fraction of 7 . 85% quoted in Cool et al. (2013) in-cludes all objects that lie more than δz > . 025 from their trueredshift. The difference in template photometry caused by sucha small change in redshift is well within the scatter of our com-puted DES - template offsets. Of greater concern is the fractionof objects with large redshift differences, which is < artificially sheared versions—BPZ is run on the metacal-ibration fluxes to determine bin assignments (cf. § i -band fluxes for both catalogues. For the Buzzard simulatedgalaxy catalogs, BPZ is run on the single mock flux mea-surement produced in the simulation.We also explored a further post-processing step as in § i -band magnitude. Wefind that this rescaling did not noticeably change the meanor widths of the PDFs on average, and that the statisticalproperties of the redshift distributions in each tomographicbin also remain unchanged. During BPZ processing of the Y1 data, three configurationand software errors were made.First, the metacalibration catalogs were processedusing MOF i -band magnitudes for evaluating the BPZ priorrather than Metacal fluxes. This is internally consistent forBPZ, but the use of flux measurements that do not exist forartificially sheared galaxies means that the metacalibra-tion shear estimates are not properly corrected for selectionbiases resulting from redshift bin assignment. We note thatsmall perturbations to the flux used for assigning the lu-minosity prior have very little impact on the resulting meanredshift and the colors used by BPZ in this run are correctlymeasured by metacalibration on unsheared and shearedgalaxy images. Rerunning BPZ with the correct, Metacal in-puts for i -band magnitude on a subset of galaxies indicatesthat the induced multiplicative shear bias is below 0 . 002 inall redshift bins, well below both the level of statistical errorsin DES Y1 and our uncertainty in shear bias calibration. Wetherefore decide to tolerate the resulting systematic uncer-tainty.Second, the SLR adjustments to photometric zeropointswere not applied to the observed Metacal fluxes in the Y1catalogs before input to BPZ. The principal result of this er-ror is that the observed magnitudes are no longer correctedfor Galactic extinction. This results in a shift in the average n i ( z ) of the source population of each bin, and a spatiallycoherent modulation of the bin occupations and redshift dis-tributions across the survey footprint. In § n i ( z ) can be accurately estimatedby mimicking the SLR errors on the COSMOS field. In Ap-pendix B we show that the spurious spatial variation of theredshift distributions causes negligible errors in our estima-tion of the shear two-point functions used for cosmologicalinference, and zero error in the galaxy-galaxy lensing esti-mates.Finally, when rewriting BPZ for a faster version, BPZ ∗ ,two bugs were introduced in the prior implementation, onecausing a bias for bright galaxies ( i band magnitude < . z that are subdominant to our calibration uncertainties(below 0.006 among all individual bins). In addition, theyare fully calibrated by both COSMOS (which uses the sameimplementation) and WZ, and hence do not affect our cos- MNRAS , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies < > ( z ) = -0.1(1+z) + 0.12(1+z) Figure 2. The average width of the posterior distributions ofBPZ photometric redshifts for data selected in bins of mean BPZredshift. The posterior width is defined as the 68% spread of thePDF p PZ ( z ) about its median. The error bars correspond to thestandard deviation of the individual source’s σ around the av-erage. mological analysis. We have since implemented all of theabove bug fixes, and applied the SLR adjustments correctly,and find negligible changes in the shape and mean of theBPZ PDFs, which are fully within the combined systematicuncertainties. z precision While n i ( z ) are the critical inputs to cosmological inference,it is sometimes of use to know the typical size of the redshiftuncertainty for individual galaxies. We define σ for each p PZ ( z ) as the half-width of the 68 percentile region aroundthe median. We select 200,000 galaxies from the metacali-bration catalog at random, and determine the average σ in bins of redshift according to the median of p PZ ( z ). Wefind that this mean σ ( z ) is well fit by a quadratic poly-nomial in mean BPZ redshift and present the best-fittingparameters in Figure 2.Further metrics of the performance of individual galax-ies’ photo- z ’s but with respect to truth redshifts are providedin § Directional Neighborhood Fitting (DNF) (De Vicente,Sanchez & Sevilla-Noarbe 2016) is a machine-learning algo-rithm for galaxy photometric redshift estimation. We haveapplied it to reconstruct the redshift distributions for the metacalibration catalogs. DNF takes as reference a train-ing sample whose spectroscopic redshifts are known. Basedon the training sample, DNF constructs the prediction hy-perplane that best fits the neighborhood of each targetgalaxy in multiband flux space. It then uses this hyperplaneto predict the redshift of the target galaxy. The key featureof DNF is the definition of a new neighborhood, the Direc-tional Neighborhood. Under this definition — and leavingapart degeneracies corresponding to different galaxy types— two galaxies are neighbors not only when they are closein the Euclidean multiband flux space, but also when theyhave similar relative flux in different bands, i.e. colors. In this way, the neighborhood does not extend in multibandflux hyperspheres but in elongated hypervolumes that bet-ter represent similar color, and presumably similar redshift.As described in § z predictions are usedto classify the galaxies in tomographic redshift bins.A random sample from the p PZ ( z ) of an object is ap-proximated in the DNF method by the redshift of the nearestneighbor within the training sample. It is used as the samplefor n i ( z ) reconstruction and interpreted in section § z ’s. Objects near the COSMOS data were removed from thetraining sample. Since the machine learning algorithm cancorrect for imperfections in the input photometry giving arepresentative training set, both training and photo- z pre-dictions are based on Metacal photometry without SLR- ad-justments, for all runs on DNF.The fiducial DES Y1 cosmological parameter estimationuses the BPZ photo- z ’s, and DES Collaboration et al. (2017)demonstrate that these estimates are robust to substitutionof DNF for BPZ.The n i ( z ) distributions of BPZ and DNF are not ex-pected to be identical, because the algorithms may makedifferent bin assignments for the same source. We thereforedo not offer a direct comparison. We do, however, repeatfor DNF all of the validation processes described herein forthe BPZ n i ( z ) estimates. The results for DNF are given inAppendix C. n i ( z ) estimation Both photo- z codes yield 6 different posterior distributions p PZ ( z j ) for each galaxy j in the Y1 shape catalogs, condi-tional on either the MOF, the unsheared Metacal, or thefour sheared Metacal flux measurements. In this section, wedescribe how these are used to define source redshift bins i and provide an initial estimate of the lensing-weighted n i ( z )of each of these bins. Table 1 gives an overview of thesesteps.Galaxies are assigned to bins based on the expectationvalue of their posterior, (cid:104) z j (cid:105) = (cid:82) z j p PZ ( z j ) d z j . We use fourbins between the limits [0 . , . , . , . , . (cid:104) z j (cid:105) < . (cid:104) z j (cid:105) > . z biases. We place three tomographicbins at (cid:104) z j (cid:105) < . n eff , a proxy for the statistical uncertainty of shearsignals in the metacalibration catalog, since z = 0 . . < (cid:104) z j (cid:105) < . , is thus validated only by the COSMOSmethod.For metacalibration sources, this bin assignment ismade based on the (cid:104) z j (cid:105) of the photo- z run on Metacal pho-tometry, instead of MOF photometry. The reason for thisis that flux measurements, and therefore photo- z bin as-signments, can depend on the shear a galaxy is subject to.This can cause selection biases in shear due to photo- z bin-ning, which can be corrected in metacalibration . The lat-ter requires that the bin assignment can be repeated using MNRAS , 000–000 (0000) DES Collaboration Table 1. Binning, n i ( z ) estimation, and mean z calibration for the variants of the shear and photo-z catalogsshear catalog step BPZ DNFbinned by: Metacal griz (cid:104) z j (cid:105) Metacal griz (cid:104) z j (cid:105) metacalibration n i ( z ) by stacking: MOF griz z PZ j Metacal griz z PZ j calibration by: COSMOS + WZ COSMOS + WZbinned by: MOF griz (cid:104) z j (cid:105) — im3shape n i ( z ) by stacking: MOF griz z PZ j —calibration by: COSMOS + WZ — a photo- z estimate made from measurements made on arti-ficially sheared images of the respective galaxy (cf. Huff &Mandelbaum 2017; Sheldon & Huff 2017; Zuntz et al. 2017),and only the Metacal measurement provides that.For im3shape sources, the bin assignment is made basedon the (cid:104) z j (cid:105) of the photo- z run on MOF photometry, whichhas higher S/N and lower susceptibility to blending effectsthan Metacal photometry. This provides more precise (andpossibly more accurate) photo- z estimates.We note that this means that for each combination ofshear and photo- z pipeline, bin assignments and effectiveweights of galaxies are different. The redshift distributionsand calibrations derived below can therefore not be directlycompared between the different variants.The stacked redshift distribution n i ( z ) of each of thetomographic bins is estimated by the lensing-weighted stackof random samples z PZ j from the p PZ ( z j ) of each of all galax-ies j in bin i . Given the millions of galaxies in each bin, thenoise due to using only one random sample from each galaxyis negligible. For both the metacalibration and im3shape catalogs, we use random samples from the p PZ ( z ) estimatedby BPZ run on MOF photometry to construct the n i ( z ),this being the lower-noise and more reliable flux estimate.In the case of DNF, we use the Metacal photometry run forboth the binning and initial n i ( z ) estimation.By the term lensing-weighted above, we mean the ef-fective weight w eff j a source j has in the lensing signals wemeasure in Troxel et al. (2017) and Prat et al. (2017). Inthe case of metacalibration , sources are not explicitlyweighted in these papers. Since the ellipticities of galaxies in metacalibration have different responses to shear (Huff &Mandelbaum 2017; Sheldon & Huff 2017), and since we mea-sure correlation functions of metacalibration ellipticitiesthat we then correct for the mean response of the ensemble,however, sources do have an effective weight that is propor-tional to their response. As can be derived by considering amixture of subsamples at different redshifts and with differ-ent mean response, the correct redshift distribution to use istherefore one weighted by w eff j ∝ ( R γ , ,j + R γ , ,j ) , wherethe R ’s are shear responses defined in Zuntz et al. (2017).In the case of im3shape , explicit weights w j are used in themeasurements, and sources have a response to shear (1+ m j )with the calibrated multiplicative shear bias m j (Zuntz et al.2017). The correct effective weights for im3shape are there-fore w eff j ∝ (1 + m j ) × w j .We note that for other uses of the shape catalogs, suchas with the optimal ∆Σ estimator (Sheldon et al. 2004), theeffective weights of sources could be different, which has tobe accounted for in the photo- z calibration. In Bonnett et al. (2016) we made use of COSMOS photo-metric redshifts as an independent estimate and validation ofthe redshift distribution of the weak lensing source galaxies.We made cuts in magnitude, FWHM and surface brightnessto the source catalogue from DECam images in the COS-MOS field that were depth-matched to the main survey area.These cuts approximated the selection function of the shapecatalogues used for the cosmic shear analysis. Similar tech-niques that find COSMOS samples of galaxies matched to alensing source catalog by a combination of magnitude, colorand morphological properties have been applied by numer-ous studies (Applegate et al. 2014; Hoekstra et al. 2015; Ok-abe & Smith 2016; Cibirka et al. 2017; Amon et al. 2017).In the present work, we modify the approach to reduce sta-tistical and systematic uncertainty on its estimate of meanredshift and carefully estimate the most significant sourcesof systematic error.We wish to validate the n i ( z ) derived for a target sampleA of galaxies using a sample B with known redshifts. Ideally,for every galaxy in A, we would find a galaxy in B thatlooks exactly like it when observed in the same conditions.The match would need to be made in all properties we useto select and weight the galaxy in the weak lensing samplethat also correlate with redshift.Then the mean redshift distribution of the matched Bgalaxies, weighted the same way as the A galaxies are for WLmeasures, will yield the desired n i ( z ). This goal is unattain-able without major observational, image processing and sim-ulation efforts, but we can approximate it with a methodrelated to the one of Lima et al. (2008) and estimate the re-maining uncertainties. We also need to quantify uncertain-ties resulting from the finite size of sample B, and from pos-sible errors in the “known” redshifts of B. Here our sampleA are the galaxies in either the im3shape or metacalibra-tion Y1 WL catalogs, spread over the footprint of DES Y1,and sample B is the COSMOS2015 catalog of Laigle et al.(2016). We begin by selecting a random subsample of 200,000 galax-ies from each WL source catalog, spread over the wholeY1 footprint, and assigning to each a match in the COS-MOS2015 catalog. The match is made by griz MOF fluxand pre-seeing size (not by position), and the matching algo- MNRAS , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies rithm proceeds as follows, for each galaxy in the WL sourcesample:(i) Gaussian noise is added to the DES griz MOF fluxesand sizes of the COSMOS galaxies until their noise level isequal to that of the target galaxy. COSMOS galaxies whoseflux noise is above that of the target galaxy are consideredineligible for matching. While this removes 13% of potentialDES-COSMOS pairs, this is unlikely to induce redshift bi-ases, because the noise level of the COSMOS griz catalog iswell below that of the Y1 survey in most regions of either,so it should be rare for the true COSMOS “twin” of a Y1source to have higher errors. The discarded pairs predom-inantly are cases of large COSMOS and small Y1 galaxies(since large size raises flux errors), and the size mismatchmeans these galaxies would never be good matches. Otherdiscarded pairs come from COSMOS galaxies lying in a shal-low region of the DECam COSMOS footprint, such as neara mask or a shallow part of the dither pattern, and this ge-ometric effect will not induce a redshift bias. Note that theMOF fluxes used here make use of the SLR zeropoints, forboth COSMOS and Y1 catalogs. The size metric is the oneproduced by metacalibration .(ii) The matched COSMOS2015 galaxy is selected as theone that minimizes the flux-and-size χ ,χ ≡ (cid:88) b ∈ griz (cid:18) f Y1 b − f COSMOS b σ b (cid:19) + (cid:18) s Y1 − s COSMOS σ s (cid:19) , (3)where f b and s are the fluxes in band b and the size, respec-tively, and σ b and σ s are the measurement errors in these forthe chosen source. We also find the galaxy that minimizesthe χ from flux differences only: χ ≡ (cid:88) b ∈ griz (cid:18) f Y1 b − f COSMOS b σ b (cid:19) . (4)If the least χ is smaller than ( χ − 4) of the galaxy withthe least flux-and-size χ , we use this former galaxy instead.Without this criterion, we could be using poor matches influx (which is more predictive of redshift than size) by re-quiring a good size match (that does not affect redshift dis-tributions much). It applies to about 15 per cent of cases.(iii) A redshift z true is assigned by drawing from the p C30 ( z ) of the matched COSMOS2015 galaxy, using therescaling of § § griz fluxes of the COS-MOS match, using the mean value of each galaxy’s posterior p PZ ( z ), as before. For the im3shape catalog, the MOF pho-tometry of the COSMOS galaxy is used, just as is done forthe Y1 main survey galaxies. The metacalibration treat-ment is more complex: we generate simulated Metacal fluxes f meta , COSMOS b for the COSMOS galaxy via f meta , COSMOS b = f MOF , COSMOS b f meta , Y1 b f MOF , Y1 b . (5)This has the effect of imposing on COSMOS magnitudes thesame difference between Metacal and MOF as is present inY1, thus imprinting onto COSMOS simulations any errors inthe Y1 catalog metacalibration magnitudes due to neglectof the SLR or other photometric errors. For the flux uncer-tainty of these matched fluxes, for both MOF and Metacal,we assign the flux errors of the respective Y1 galaxy. − nu m b e r o f C O S M O S ga l a x i e s Figure 3. Repetitions of COSMOS galaxies in the fiducialmatched metacalibration sample of 200,000 objects. The overallweight of galaxies with more than 20 matches, which are typicallybright, is below one per cent of the total weight. The high-usageoutliers are a few of the very bright COSMOS galaxies. (v) The effective weak lensing weight w of the originalsource galaxy is assigned to its COSMOS match (cf. § χ between matched galaxies. Thedistribution is skewed toward significantly lower χ valuesthan expected from a true χ distribution with 4 degreesof freedom. This indicates that the COSMOS-Y1 matchesare good: COSMOS galaxies are photometrically even moresimilar to the Y1 target galaxies than they would be to re-observed versions of themselves.A second check on the matching algorithm is to askwhether the individual COSMOS galaxies are being resam-pled at the expected rates. As expected, most sufficientlybright galaxies in COSMOS are used more than once, whilethe faintest galaxies are used more rarely or never. Figure3 shows the number of times each of the COSMOS galax-ies is matched to metacalibration (if it is bright enoughto be matched at all) in our fiducial matched catalog. Wesee that there is no unwanted tendency for a small fractionof the COSMOS galaxies to bear most of the resamplingweight. All COSMOS galaxies with more than 50 repeti-tions are brighter than i = 18 . z ≈ . griz -predicted distribution and the “truth” provided byCOSMOS2015 for all galaxies assigned to a given source bin:∆ z = (cid:80) i w i z true i (cid:80) i w i − (cid:80) j w j z PZ j (cid:80) j w j , (6)where the sums run over all matched COSMOS2015 galaxies i and all galaxies in the original source sample j .This construction properly averages ∆ z over the ob-serving conditions (including photometric zeropoint errors)and weights of the Y1 WL sources. These estimated ∆ z val-ues using BPZ are tabulated in Table 2 for both WL sourcecatalogs.The COSMOS validation also yields an estimate of MNRAS , 000–000 (0000) DES Collaboration Figure 4. The redshift distributions n i ( z ) derived from threedifferent methods are plotted for each of the 4 WL metacalibra-tion source bin populations i = 1 . . . . The top (bottom) figureshows the 1st and 3rd (2nd and 4th) tomographic redshift bins.The clustering methodology (WZ) can only constrain n i ( z ) for0 . < z < . 9, and the normalization of the distribution is arbi-trary for the bins extending beyond this range. The band aroundthe COSMOS n i ( z ) depicts the uncertainties as described in § n ( z ) . We demonstrate in § n i ( z ) by a weighted average of the rescaled p C30 ( z )’s of thematches (or, equivalently, of samples drawn from them). Fig-ure 4 plots these resampled-COSMOS estimates along withthe original n i PZ ( z ) from BPZ. Here it is apparent that insome bins, these two estimates differ by more than just asimple shift in redshift—the shapes of the n i ( z ) distribu-tions differ significantly. In § z i . All of these arepresented for the metacalibration sample binned by BPZredshift estimates. For im3shape galaxies with BPZ, we usethe same uncertainties. Results for DNF are in Appendix C.From the resampling procedure, we also determine com-mon metrics on the photo- z performance in § The first contribution to the uncertainty in the COSMOS∆ z i ’s is from sample variance from the small angular sizeof the COSMOS2015 catalog. Any attempt at analytic es-timation of this uncertainty would be complicated by thereweighting/sampling procedure that alters the native n ( z ) Figure 5. Redshift distributions of the full simulated lensingsample from the Buzzard catalog (grey) and two examples of sam-ples from COSMOS-sized footprints in the Buzzard catalogs thathave been resampled and weighted to match the full distribution(blue and orange). Figure 6. Correlation coefficients of error on ∆ z i due to sam-ple variance in COSMOS-resamplings between our four sourceredshift bins. Shown is the correlation matrix for the metacali-bration sample binned by BPZ. of the COSMOS line of sight, so we instead estimate the co-variance matrix of the ∆ z i by repeating our procedures ondifferent realizations of the COSMOS field in the Buzzardsimulated galaxy catalogs.The resampling procedure of § § n ( z ) ofthe matched COSMOS catalogs (cf. Figure 5), and conse-quently an independent sample variance realization of the∆ z i . There are significant correlations between the ∆ z i bins,especially bins 1 and 2, as shown in § 6. The diagonal ele-ments are listed as “COSMOS footprint sampling” in Ta-ble 2.Since we use the same subset of the Buzzard lensingsample for each of the COSMOS-like resamplings, this vari-ance estimate does not include the uncertainty due to thelimited subsample size of 200,000 galaxies. We estimate the MNRAS , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies Table 2. Values of and error contributions to photo- z shift parameters of BPZ n i ( z ) . Value Bin 1 Bin 2 Bin 3 Bin 4 z PZ range 0.20–0.43 0.43–0.63 0.63–0.90 0.90-1.30COSMOS footprint sampling ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . z i uncertainty ± . ± . ± . ± . metacalibration COSMOS final ∆ z i , tomographic uncertainty − . ± . − . ± . 021 +0 . ± . − . ± . z i +0 . ± . − . ± . 017 +0 . ± . 014 — Combined final ∆ z i − . ± . − . ± . 013 +0 . ± . − . ± . im3shape COSMOS final ∆ z i , tomographic uncertainty +0 . ± . − . ± . 021 +0 . ± . − . ± . z i +0 . ± . − . ± . − . ± . 014 — Combined final ∆ z i +0 . ± . − . ± . − . ± . − . ± . latter effect by resampling of the ∆ z i in this sample, andfind it to be subdominant ( σ i ∆ z < . 003 in all redshift bins,“limited sample size” in Table 2). The griz DECam photometry of the COSMOS field has un-certainties in its zeropoint due to errors in the SLR-basedcalibration. While the Y1 catalog averages over the SLR er-rors of many fields, the validation is sensitive to the singlerealization of SLR errors in the COSMOS field. We esti-mate the distribution of zeropoint errors by comparing theSLR zeropoints in the Y1 catalog to those derived from thesuperior “forward global calibration module” (FGCM) andreddening correction applied to three years’ worth of DESexposures by Burke et al. (2017). In this we only use regionswith Galactic extinction E ( B − V ) < . 1, since the COS-MOS field has relatively low extinction and strong reddeningmight cause larger differences between the FGCM and SLRcalibration. The root-mean-square zeropoint offsets betweenSLR and FGCM calibration are between 0.007 ( z ) and 0.017( g ). We estimate the impact on ∆ z i by drawing 200 mean-subtracted samples of photometric offsets from the observed(FGCM-SLR) distribution, applying each to the COSMOSfluxes, and repeating the derivation of ∆ z i as per § z i of each of the fourtomographic bins due to those, which are 0 . − . We have matched COSMOS galaxies to the shear cataloggalaxies by their griz fluxes and by their estimated pre-seeing size. This set of parameters is likely not completelypredictive of a galaxy’s selection and weight in our shearcatalog. Other morphological properties (such as the steep-ness of its profile) probably matter and do correlate withredshift (e.g. Soo et al. 2017). In addition, the matching insize is only done in 85 per cent of cases to begin with § z i for metacalibra-tion to be (+0 . , +0 . , +0 . , +0 . σ and out-lier fraction is galaxy size. Since we therefore expect the sizeto have the strongest influence on both lensing and redshift,and we are correcting for size, we estimate the potential in-fluence of any further variables as no more than half of thesize effect. We do not assume that these systematic errorsfound in simulations are exactly equal in the data - rather,we only assume that the two are of similar size, and thus usethe rms of offsets found in the simulation as the width of aGaussian systematic uncertainty on the data. We take halfof the quadratic mean of the shifts in the four redshift bins, ± . , as our estimate of the hidden-variable uncertaintyin each bin. These biases are likely to be correlated betweenbins. In § Even in the absence of the above uncertainties, the re-sampling algorithm described above might not quite repro-duce the true redshift distribution of the input sample. Thematching algorithm may not, for example, pick a COSMOSgalaxy which is an unbiased estimator of the target galaxy’sredshift, especially given the sparsity and inhomogeneousdistribution of the COSMOS sample in the four- to five-dimensional space of griz fluxes and size.We estimate the size of this effect on ∆ z i using the mean offset in binned mean true redshift of the 367 realiza-tions of resampled COSMOS-like catalogs in the Buzzardsimulations (see § MNRAS , 000–000 (0000) DES Collaboration do not attempt to correct the result of our resampling withthese values. Rather, we take them as indicators of possi-ble systematic uncertainties of the resampling algorithm.Following the argument in section § The final uncertainties on the ∆ z i are estimated by addingin quadrature the contributions listed above, yielding the“COSMOS total ∆ z i uncertainty” in Table 2. These val-ues are derived independently for each redshift bin, but it iscertain that the ∆ z i have correlated errors, e.g. from sam-ple variance as shown in Figure 6, and such correlationsshould certainly be included in the inference of cosmologi-cal parameters. The values of the off-diagonal elements ofthe combined COSMOS ∆ z covariance matrix, are, how-ever, difficult to estimate with any precision. In Appendix Awe demonstrate that by increasing the diagonal elements ofthe covariance matrix by a factor (1 . and nulling the off-diagonal elements, we can ensure that any inferences basedon the ∆ z i are conservatively estimated for any reasonablevalues of the off-diagonal elements. We therefore apply afactor of 1.6 to all of the single-bin uncertainties in derivingthe “COSMOS final ∆ z i ” constraints for metacalibration and im3shape given in Table 2. Although not a critical input to the cosmological tests ofDES Collaboration et al. (2017), we determine here somestandard metrics of photo- z performance. We define theresidual R as the difference between the mean of the p PZ ( z )using the MOF photometry and a random draw from theCOSMOS p C30 ( z ) matched during resampling. We use arandom draw from p C30 ( z ) rather than the peak, so thatuncertainty in these “truth” z ’s is included in the metrics.Because the width of p C30 ( z ) is much smaller than that of p PZ ( z ), this does not affect the results significantly.We define σ ( R ) as the 68% spread of R around itsmedian. In this section σ ( R ) measures the departure ofthe mean of p PZ ( z ) from the true z , whereas the σ inFigure 2 is a measure of width of p PZ ( z ) independent ofany truth redshifts. We also measure the outlier fraction,defined as the fraction of data for which | R | > × σ . Ifthe redshift distribution were Gaussian, the outlier fractionwould be 5%, and this metric is a measure of the tails of the R distribution.We calculate the uncertainties on these metrics fromsample variance, COSMOS photometric calibration uncer-tainty, and selection of the lensing sample by hidden vari-ables (cf. § metacalibra-tion sample and binning. To supplement the constraints on ∆ z i derived above usingthe COSMOS2015 photo- z ’s, we turn to the “correlationredshift” methodology (Newman 2008; M´enard et al. 2013;Schmidt et al. 2013) whereby one measures the angular cor-relations between the unknown sample (the WL sources)and a population of objects with relatively well-determinedredshifts. In our case the known population are the red-MaGiC galaxies, selected precisely so that their griz colorsyield high-accuracy photometric redshift estimates.An important complication of applying WZ to DES Y1is that we do not have a sufficient sample of galaxies withknown redshift available that spans the redshift range of theDES Y1 lensing source galaxies – the redMaGiC galaxiesdo not extend beyond 0 . < z < . 9. Constraints on themean redshift of a source population can still be derived inthis case, but only by assuming a shape for the n ( z ) dis-tribution, whose mean is then determined by the clusteringsignal in a limited redshift interval. A mismatch in shapebetween the assumed and true n ( z ) is a source of system-atic uncertainty in such a WZ analysis. One of the mainresults of Gatti et al. 2017, which describes the implementa-tion and full estimation of uncertainties of the WZ methodfor DES Y1 source galaxies, is that while this systematicuncertainty needs to be accounted for, it is not prohibitivelylarge. This statement is validated in Gatti et al. 2017 forthe degree of mismatch between the true n ( z ) and the n ( z )found in a number of photometric redshift methods appliedto simulated galaxy catalogs. The redshift distributions ofthe DES weak lensing sources as estimated by BPZ, as far aswe can judge this from the comparison with the COSMOSestimates of their true n ( z ), show a similar level of mismatchto the truth. The systematic uncertainty budget derived inGatti et al. (2017) is therefore applicable to the data. Wedo not, however, attempt to correct the systematic offsets inWZ estimates of ∆ z i introduced due to this effect – for this,we would require the galaxy populations and photometricmeasurements in the simulations to be perfectly realistic.The method is applied to DES Y1 data in Davis et al.(2017a). A similar analysis was performed on the DES SVdata set in Davis et al. (2017b). The resultant estimatesof ∆ z i are listed in Table 2 and plotted in Figure 7. Thefull n i ( z )’s estimated from the WZ method are plotted inFigure 4. Note that the WZ method obtains no useful con-straint for bin 4 because the redMaGiC sample is confinedto z < . z i that arefully consistent. Indeed even their n i ( z ) curves show qual-itative agreement. We therefore proceed to combine theirconstraints on ∆ z i to yield our most accurate and reliableestimates. The statistical errors of the COSMOS and WZmethods are uncorrelated (sample variance in the COSMOSfield vs. shot noise in the measurements of angular corre-lations in the wide field). The dominant systematic errorsof the two methods should also be uncorrelated, e.g. short- MNRAS , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies Table 3. Common performance metrics and uncertainties measured using BPZ point predictions and draws from the rescaled COS-MOS2015 PDFs. The quantity σ ( R ) is the 68% spread of the residual distribution R , about the median. The outlier fraction is definedas the fraction of galaxies with griz redshift estimates than 2 × σ ( R ) from the COSMOS2015 value.metric 0 . < z < . 43 0 . < z < . 63 0 . < z < . 90 0 . < z < . metacalibration binning, MOF p PZ ( z ) σ ( R ) 0 . ± . 01 0 . ± . 01 0 . ± . 01 0 . ± . . ± . . ± . . ± . . ± . metacalibration σ ( R ) 0 . ± . 01 0 . ± . 01 0 . ± . 01 0 . ± . . ± . . ± . . ± . . ± . comings in our resampling for COSMOS vs. uncertainties inthe bias evolution of source galaxies for WZ. We are there-fore confident that we can treat the COSMOS and WZ con-straints as independent, and we proceed to combine them bymultiplying their respective 1-dimensional Gaussian distri-butions for each ∆ z i , i.e. inverse-variance weighting. In bin4, the final constraints are simply the COSMOS constraintssince WZ offers no information.The resultant constraints, listed for both metacalibra-tion and im3shape catalogs in Table 2, are the principalresult of this work, and are adopted as input to the cosmo-logical inferences of Troxel et al. (2017) and DES Collabo-ration et al. (2017). The adopted 68%-confidence ranges foreach ∆ z i are denoted by the gray bands in the 1-d marginalplots of Figure 7.One relevant question is whether our calibration findsthat significant non-zero shifts are required to correct thephoto- z estimates of the mean redshift. For the fiducial metacalibration BPZ, this is not the case: the χ = (cid:80) i (∆ z i /σ ∆ z i ) is 3.5 with 4 bins. However, the combined∆ z is non-zero at 2 . σ for im3shape BPZ and the ∆ z is non-zero at 3 . σ for metacalibration DNF, indicatingthat there are significant alterations being made to some ofthe n i PZ ( z ) estimates.A further check of the accuracy of our n i ( z ) estimationis presented by Prat et al. (2017) using the ratios of lens-ing shear on the different source bins induced by a commonset of lens galaxies. Initially proposed as a cosmological test(Jain & Taylor 2003), the shear ratio is in fact much less sen-sitive to cosmological parameters than to potential errors ineither the calibration of the shear measurement or the de-termination of the n i ( z ) . We plot in Figure 7 the constraintson ∆ z i inferred by Prat et al. (2017) after marginalizationover the estimated errors in shear calibration and assuminga fixed ΛCDM cosmology with Ω m = 0 . 3. The shear-ratiotest is fully consistent with the COSMOS and WZ estimatesof ∆ z i , though we should keep in mind that this test is alsodependent on the validity of the shear calibration and someother assumptions in the analysis, and importantly is co-variant with the WZ method, because both methods rely oncorrelation functions as measured with respect to the samegalaxy samples.. The final rows of Table 2 provide the prior on errors in theredshift distributions used during inference of cosmologicalparameters for the DES Y1 data, under the assumption thaterrors in the n i ( z ) resulting from the photo- z analysis fol-low Equation (1). Determination of redshift distributions isand will continue to be one of the most difficult tasks forobtaining precision cosmology from broadband imaging sur-veys such as DES, so it is important to examine the potentialimpact of assumptions in our analysis choices. Further, wewish to identify areas where our methodology can be im-proved and thereby increase the precision and accuracy offuture cosmological analyses. First, we base our COSMOS validation on the COS-MOS2015 redshift catalog derived from fitting spectral tem-plates to 30-band fluxes. Our COSMOS validation rests onthe assumption that Laigle et al. (2016) have correctly esti-mated the redshift posteriors of their sources. Overall, red-shift biases in the COSMOS2015 redshifts are significant,unrecognized sources of error in our cosmological inferencesif they approach or exceed the δz ≈ . z i constraints. More precisely, this bias mustaccrue to the portion of the COSMOS2015 catalog that isbright enough to enter the DES Y1 shear catalogs.For the subset of their sources with spectroscopic red-shifts, Laigle et al. (2016) report that galaxies in the mag-nitude interval 22 < i < 23 have “catastrophic” disagree-ment between photo- z and spectroscopic z for only 1.7%(0.6%) for star-forming (quiescent) galaxies (their Table 4).This is the magnitude range holding the 50% completenessthreshold of the DES Y1 shear catalogs. Brighter bins havelower catastrophic-error rates, and only about 5 per cent ofweight in the metacalibration lensing catalog is providedby galaxies fainter than i = 23. It would thus be difficult forthese catastrophic errors to induce photo- z errors of 0.01 ormore.About 30 per cent of the galaxies used for the COS-MOS weak lensing validation have spectroscopic redshiftsfrom the latest 20,000 I < . . of the COSMOS field to z < . MNRAS , 000–000 (0000) DES Collaboration Figure 7. Constraints on the shifts ∆ z i applied to the metacalibration n PZ ( z ) distributions for the weak lensing source galaxies areplotted for three different validation techniques. Shifts derived from resampling the COSMOS 30-band redshifts are described in thispaper, and agree well with those derived (for bins 1–3 only) using angular correlations between the source population and redMagicgalaxies (WZ) by Davis et al. (2017a) (COSMOS constraints plotted here have been expanded as per Appendix A to include the effectsof poorly known correlation between bins). These are also consistent with the weak lensing shear ratio tests conducted by Prat et al.(2017). The final validation constraints on ∆ z i are taken as the combination of the COSMOS and WZ results for each redshift bin (whereavailable), and yield the 68% confidence intervals denoted by the black points and error bars in the 1-d marginal plots. The dashed linesat ∆ z i = 0 indicate no mean shift from the BPZ posteriors—the validation processes yield shifts that are non-zero at ≈ σ level. troscopic subset are very similar (less than 1-sigma of ourerror estimate) to the corresponding shifts estimated withphotometric redshifts in the full sample. The difference be-tween the 30-band (corrected) photometric mean redshiftsand the corresponding spectroscopic redshifts for this subsetis also within our error estimates. These tests indicate thatthe potential (unknown) biases in the 30-band photometricredshifts are smaller than other sources of uncertainty in themean redshifts used for our WL analysis.Of greater concern is the potential for bias in the por-tion of the DES detection regime for which spectroscopic validation of COSMOS2015 photo- z ’s is not possible. Nei-ther we nor Laigle et al. (2016) have direct validation of thissubsample, so we are relying on the success of their template-based method and broad spectral coverage in the spectro-scopic regime to extend into the non-spectroscopic regime.Our confidence is boosted, however, by the agreement in ∆ z i between the COSMOS validation and the independent WZvalidation in bins 1, 2, and 3.Finally we note that we have also attempted to vali-date the photo-z distributions using only the galaxies withspectroscopic redshifts in the COSOMOS field, and find con- MNRAS000 Constraints on the shifts ∆ z i applied to the metacalibration n PZ ( z ) distributions for the weak lensing source galaxies areplotted for three different validation techniques. Shifts derived from resampling the COSMOS 30-band redshifts are described in thispaper, and agree well with those derived (for bins 1–3 only) using angular correlations between the source population and redMagicgalaxies (WZ) by Davis et al. (2017a) (COSMOS constraints plotted here have been expanded as per Appendix A to include the effectsof poorly known correlation between bins). These are also consistent with the weak lensing shear ratio tests conducted by Prat et al.(2017). The final validation constraints on ∆ z i are taken as the combination of the COSMOS and WZ results for each redshift bin (whereavailable), and yield the 68% confidence intervals denoted by the black points and error bars in the 1-d marginal plots. The dashed linesat ∆ z i = 0 indicate no mean shift from the BPZ posteriors—the validation processes yield shifts that are non-zero at ≈ σ level. troscopic subset are very similar (less than 1-sigma of ourerror estimate) to the corresponding shifts estimated withphotometric redshifts in the full sample. The difference be-tween the 30-band (corrected) photometric mean redshiftsand the corresponding spectroscopic redshifts for this subsetis also within our error estimates. These tests indicate thatthe potential (unknown) biases in the 30-band photometricredshifts are smaller than other sources of uncertainty in themean redshifts used for our WL analysis.Of greater concern is the potential for bias in the por-tion of the DES detection regime for which spectroscopic validation of COSMOS2015 photo- z ’s is not possible. Nei-ther we nor Laigle et al. (2016) have direct validation of thissubsample, so we are relying on the success of their template-based method and broad spectral coverage in the spectro-scopic regime to extend into the non-spectroscopic regime.Our confidence is boosted, however, by the agreement in ∆ z i between the COSMOS validation and the independent WZvalidation in bins 1, 2, and 3.Finally we note that we have also attempted to vali-date the photo-z distributions using only the galaxies withspectroscopic redshifts in the COSOMOS field, and find con- MNRAS000 , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies sistent, albeit uncompetitive results. The number of galaxieswith spectra (20k) is an order of magnitude less than thosewith reliable photometric redshifts which increases statisti-cal uncertainties, cosmic variance uncertainties and uncer-tainties from data re-weighting. n i ( z ) shape Equation (1) assumes that the only errors in the n i PZ ( z ) dis-tributions take the form of a translation of the distributionin redshift. We do not expect that errors in the photo- z dis-tribution actually take this form; rather we assume that theshape of n i ( z ) has little impact on our cosmological inferenceas long as the mean of the distribution is conserved—andour methodology forces the mean of the n i ( z ) to match thatderived from the COSMOS2015 resampling. The validity ofthis assumption can be tested by assuming that any errorsin the shifted-BPZ n i ( z ) from Equation (1) are akin to thedifference between these distributions and n i COSMOS ( z ) de-rived from the resampled COSMOS catalogs during the val-idation process of § n i COSMOS ( z ) distributions. We then fit this data using amodel that assumes the shifted BPZ distributions. The best-fit cosmological parameters depart from those in the inputsimulation by less than ten per cent of the uncertainty ofDES Collaboration et al. (2017). We therefore confirm thatthe detailed shape of n i ( z ) is not important to the Y1 anal-ysis. A third assumption in our analysis is that the n i ( z ) arethe same for all portions of the DES Y1 catalog footprint(aside, of course, from the intrinsic density fluctuations thatwe wish to measure). This is not the case: the failure toapply SLR adjustment to our fluxes in BPZ ( § z assignments. Even without this error, wewould have significant depth fluctuations because of vari-ation in the number and quality of exposures on differentparts of the survey footprint.Appendix B provides an approximate quantification ofthe impact of n i ( z ) inhomogeneities on our measurementsof the 2-point correlation functions involving the shear cata-log. There we conclude that the few per cent fluctuations insurvey depth and color calibration that exist in our sourcecatalogs should not significantly influence our cosmologicalinferences, as long as we use the source-weighted mean n i ( z )over the survey footprint. Both the COSMOS and WZ vali-dation techniques produce source-weighted estimates of ∆ z i , as required. We have estimated redshift distributions and defined tomo-graphic bins of source galaxies in DES Y1 lensing analysesfrom photometric redshifts based on their griz photometry.While we use traditional photo- z methods in these steps, we independently determine posterior probability distributionsfor the mean redshift of each tomographic bin that are thenused as priors for subsequent lensing analyses.The method for determining these priors developed inthis paper is to match galaxies with COSMOS2015 30-bandphotometric redshift estimates to DES Y1 lensing sourcegalaxies, selecting and weighting the former to resemble thelatter in their griz flux and pre-seeing size measurements.The mean COSMOS2015 photo- z of the former sample is ourestimate of the mean redshift of the latter. We determine un-certainties in this estimate, which we find have comparable,dominant contributions from(i) sample variance in COSMOS, i.e. the scatter in mea-sured mean redshift calibration due to the limited footprintof the COSMOS field,(ii) the influence of morphological parameters such asgalaxy size on the lensing source sample selection, and(iii) systematic mismatches of the original and matchedsample in the algorithm we use.A significant reduction of the overall uncertainty of the meanredshift priors derived in this work would thus only be pos-sible with considerable additional observations and algorith-mic advances.Subdominant contributions, in descending order, aredue to(i) errors in photometric calibration of the griz data inthe COSMOS field and(ii) the finite subsample size from the DES Y1 shear cat-alogs that we use for resampling.The COSMOS2015 30-band photometric estimates ofthe mean redshifts, supplemented by consistent measuresby means of angular correlation against DES redMaGiC galaxies in all but the highest redshift bins (Gatti et al. 2017;Cawthon et al. 2017; Davis et al. 2017a), have allowed us todetermine the mean redshifts of 4 bins of WL source galaxiesto 68% CL accuracy ± . n i PZ ( z ) distributions. These redshift uncertaintiesare a highly subdominant contributor to the error budgetof the DES Y1 cosmological parameter determinations ofDES Collaboration et al. (2017) when marginalizing overthe full set of nuisance parameters. Likewise, the method-ology of marginalizing over the mean redshift uncertaintyonly, rather than over the full shape of the n i ( z ), biases ouranalyses at less than ten per cent of their uncertainty. Thusthe methods and approximations herein are sufficient for theY1 analyses.DES is currently analyzing survey data covering nearly4 times the area used in the Y1 analyses of this paper andDES Collaboration et al. (2017), and there are ongoing im-provements in depth, calibration, and methodology. Thuswe expect > × reduction in the statistical and system-atic uncertainties in future cosmological constraints, com-pared to the Y1 work. Uncertainties in n i ( z ) will becomethe dominant source of error in future analyses of DES andother imaging surveys, without substantial improvement inthe methodology presented here. We expect that the linear-shift approximation in Equation (1) will no longer sufficefor quantifying the validation of the n i ( z ). Significant im-provement will be needed in some combination of: spec- MNRAS , 000–000 (0000) DES Collaboration troscopic and/or multiband photometric validation data;photo- z methodology; redshift range, bias constraints, andstatistical errors of WZ measurements; and treatment of sur-vey inhomogeneity. The redshift characterization of broad-band imaging surveys is a critical and active area of research,and will remain so in the years to come. ACKNOWLEDGEMENTS Support for DG was provided by NASA through EinsteinPostdoctoral Fellowship grant number PF5-160138 awardedby the Chandra X-ray Center, which is operated by theSmithsonian Astrophysical Observatory for NASA undercontract NAS8-03060.Funding for the DES Projects has been provided bythe U.S. Department of Energy, the U.S. National Sci-ence Foundation, the Ministry of Science and Education ofSpain, the Science and Technology Facilities Council of theUnited Kingdom, the Higher Education Funding Council forEngland, the National Center for Supercomputing Applica-tions at the University of Illinois at Urbana-Champaign, theKavli Institute of Cosmological Physics at the Universityof Chicago, the Center for Cosmology and Astro-ParticlePhysics at the Ohio State University, the Mitchell Institutefor Fundamental Physics and Astronomy at Texas A&MUniversity, Financiadora de Estudos e Projetos, Funda¸c˜aoCarlos Chagas Filho de Amparo `a Pesquisa do Estadodo Rio de Janeiro, Conselho Nacional de DesenvolvimentoCient´ıfico e Tecnol´ogico and the Minist´erio da Ciˆencia, Tec-nologia e Inova¸c˜ao, the Deutsche Forschungsgemeinschaftand the Collaborating Institutions in the Dark Energy Sur-vey. The Collaborating Institutions are Argonne NationalLaboratory, the University of California at Santa Cruz,the University of Cambridge, Centro de InvestigacionesEnerg´eticas, Medioambientales y Tecnol´ogicas-Madrid, theUniversity of Chicago, University College London, the DES-Brazil Consortium, the University of Edinburgh, the Ei-dgen¨ossische Technische Hochschule (ETH) Z¨urich, FermiNational Accelerator Laboratory, the University of Illi-nois at Urbana-Champaign, the Institut de Ci`encies del’Espai (IEEC/CSIC), the Institut de F´ısica d’Altes Ener-gies, Lawrence Berkeley National Laboratory, the Ludwig-Maximilians Universit¨at M¨unchen and the associated Ex-cellence Cluster Universe, the University of Michigan, theNational Optical Astronomy Observatory, the University ofNottingham, The Ohio State University, the University ofPennsylvania, the University of Portsmouth, SLAC NationalAccelerator Laboratory, Stanford University, the Universityof Sussex, Texas A&M University, and the OzDES Member-ship Consortium.Based in part on observations at Cerro Tololo Inter-American Observatory, National Optical Astronomy Obser-vatory, which is operated by the Association of Universitiesfor Research in Astronomy (AURA) under a cooperativeagreement with the National Science Foundation.The DES data management system is supported bythe National Science Foundation under Grant NumbersAST-1138766 and AST-1536171. The DES participants fromSpanish institutions are partially supported by MINECOunder grants AYA2015-71825, ESP2015-88861, FPA2015- 68048, SEV-2012-0234, SEV-2016-0597, and MDM-2015-0509, some of which include ERDF funds from the EuropeanUnion. IFAE is partially funded by the CERCA programof the Generalitat de Catalunya. Research leading to theseresults has received funding from the European ResearchCouncil under the European Union’s Seventh FrameworkProgram (FP7/2007-2013) including ERC grant agreements240672, 291329, and 306478. We acknowledge support fromthe Australian Research Council Centre of Excellence forAll-sky Astrophysics (CAASTRO), through project numberCE110001020.This manuscript has been authored by Fermi ResearchAlliance, LLC under Contract No. DE-AC02-07CH11359with the U.S. Department of Energy, Office of Science, Of-fice of High Energy Physics. The United States Governmentretains and the publisher, by accepting the article for pub-lication, acknowledges that the United States Governmentretains a non-exclusive, paid-up, irrevocable, world-wide li-cense to publish or reproduce the published form of thismanuscript, or allow others to do so, for United States Gov-ernment purposes.Based in part on zCOSMOS observations carried outusing the Very Large Telescope at the ESO Paranal Obser-vatory under Programme ID: LP175.A-0839. REFERENCES Amon A. et al., 2017, ArXiv e-prints:1707.04105Angus J. E., 1994, SIAM Rev., 36, 652Applegate D. E. et al., 2014, MNRAS, 439, 48Arnouts S., Cristiani S., Moscardini L., Matarrese S., Lucchin F.,Fontana A., Giallongo E., 1999, MNRAS, 310, 540Becker M. R., 2013, MNRAS, 435, 115Bender R. et al., 2001, in Deep Fields, Cristiani S., Renzini A.,Williams R. E., eds., p. 96Ben´ıtez N., 2000, ApJ, 536, 571Benjamin J. et al., 2013, MNRAS, 431, 1547Bertin E., Arnouts S., 1996, A&AS, 117, 393Bonnett C. et al., 2016, Phys. Rev. D, 94, 042005Bordoloi R., Lilly S. J., Amara A., 2010, MNRAS, 406, 881Burke D. et al., 2017, ArXiv e-prints: 1706.01542Carrasco Kind M., Brunner R. J., 2013, MNRAS, 432, 1483Cawthon R., et al., 2017, in prep.Cibirka N. et al., 2017, MNRAS, 468, 1092Coil A. L. et al., 2011, ApJ, 741, 8Coleman G. D., Wu C.-C., Weedman D. W., 1980, ApJS, 43, 393Collister A. A., Lahav O., 2004, PASP, 116, 345Cool R. J. et al., 2013, ApJ, 767, 118Cooper M. C. et al., 2011, ApJS, 193, 14Davis C. et al., 2017a, ArXiv e-prints:1710.02517Davis C. et al., 2017b, ArXiv e-prints:1707.08256Dawid A. P., 1984, Journal of the Royal Statistical Society. SeriesA (General), 147, 278De Vicente J., Sanchez E., Sevilla-Noarbe I., 2016, Mon. Not.Roy. Astron. Soc., 459, 3078DeRose J., Wechsler R., Rykoff E., et al., 2017, in prep.DES Collaboration, et al., 2017, to be submitted to Phys. Rev. DDrlica-Wagner A. et al., 2017, ArXiv e-prints:1708.01531Elvin-Poole J., et al., 2017, to be submitted to Phys. Rev. DFeldmann R. et al., 2006, MNRAS, 372, 565Flaugher B. et al., 2015, AJ, 150, 150Gatti M. et al., 2017, ArXiv e-prints:1709.00992Gruen D., Brimioulle F., 2017, MNRAS, 468, 769Gruen D. et al., 2017, ArXiv e-printsMNRAS , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies Gschwend J. et al., 2017, ArXiv e-prints:1708.05643Guzik J., Bernstein G., 2005, Phys. Rev. D, 72, 043503High F. W., Stubbs C. W., Rest A., Stalder B., Challis P., 2009,AJ, 138, 110Hildebrandt H. et al., 2010, A&A, 523, A31Hildebrandt H. et al., 2017, MNRAS, 465, 1454Hoekstra H., Herbonnet R., Muzzin A., Babul A., Mahdavi A.,Viola M., Cacciato M., 2015, MNRAS, 449, 685Hoyle B., 2016, Astronomy and Computing, 16, 34Huff E., Mandelbaum R., 2017, ArXiv e-prints:1702.02600Ilbert O. et al., 2006, A&A, 457, 841Ivezi´c ˇZ. et al., 2004, Astronomische Nachrichten, 325, 583Jain B., Taylor A., 2003, Physical Review Letters, 91, 141302Jarvis M. et al., 2016, MNRAS, 460, 2245Jee M. J., Tyson J. A., Schneider M. D., Wittman D., SchmidtS., Hilbert S., 2013, ApJ, 765, 74Kinney A. L., Calzetti D., Bohlin R. C., McQuade K., Storchi-Bergmann T., Schmitt H. R., 1996, ApJ, 467, 38Laigle C., McCracken H. J., Ilbert O., Hsieh B. C., Davidzon I.,Capak P., others, 2016, ApJS, 224, 24Lilly S. J. et al., 2009a, VizieR Online Data Catalog, 217Lilly S. J. et al., 2009b, ApJS, 184, 218Lilly S. J. et al., 2007, ApJS, 172, 70Lima M., Cunha C. E., Oyaizu H., Frieman J., Lin H., SheldonE. S., 2008, MNRAS, 390, 118MacCrann N., et al., 2017, in prep.MacDonald E. C. et al., 2004, MNRAS, 352, 1255M´enard B., Scranton R., Schmidt S., Morrison C., Jeong D., Bu-davari T., Rahman M., 2013, ArXiv e-prints: 1303.4722Miyazaki S. et al., 2002, PASJ, 54, 833Newman J. A., 2008, ApJ, 684, 88O’Donnell J. E., 1994, ApJ, 422, 158Okabe N., Smith G. P., 2016, MNRAS, 461, 3794Prat J. et al., 2017, ArXiv e-printsRau M., Gschwend J., Hoyle B., et al., 2017, in prep.Rau M. M., Seitz S., Brimioulle F., Frank E., Friedrich O., GruenD., Hoyle B., 2015, MNRAS, 452, 3710Rozo E., et al., 2016, Mon. Not. R. Astron. Soc., 461, 1431Sadeh I., Abdalla F. B., Lahav O., 2016, PASP, 128, 104502S´anchez C. et al., 2014, MNRAS, 445, 1482Schlegel D. J., Finkbeiner D. P., Davis M., 1998, ApJ, 500, 525Schmidt S. J., M´enard B., Scranton R., Morrison C., McBrideC. K., 2013, MNRAS, 431, 3307Schmidt S. J., Thorman P., 2013, MNRAS, 431, 2766Scodeggio M. et al., 2016, ArXiv e-prints: 1611.07048Sheldon E. S., 2014, MNRAS, 444, L25Sheldon E. S., Huff E. M., 2017, ApJ, 841, 24Sheldon E. S. et al., 2004, AJ, 127, 2544Soo J. Y. H. et al., 2017, ArXiv e-prints: 1707.03169Troxel M. A. et al., 2017, ArXiv e-prints:1708.01538Wechsler R., DeRose J., Busha M., et al., 2017, in prep.Zuntz J., Kacprzak T., Voigt L., Hirsch M., Rowe B., Bridle S.,2013, MNRAS, 434, 1604Zuntz J., et al., 2017, submitted to Mon. Not. R. Astron. Soc. APPENDIX A: COMPENSATION FORUNKNOWN COVARIANCES INTOMOGRAPHIC ANALYSES There are four parameters ∆ x i in our model for errors in theredshift distribution, which will be used when constrainingsome parameter(s) π of the cosmological model. While wehave produced reliable bounds σ i of the variance of each ofthese, we have less knowledge of the off-diagonal elements r ij σ i σ j of the covariance matrix C of the ∆ x i —perhaps weknow only that | r ij | ≤ r . We wish to make estimates of π and the uncertainty σ π that we are sure do not underestimatethe true error, for any allowed values of the r ij , in analysesthat combine these redshift bins. We show here this can bedone by amplifying the diagonal elements of C by a factor f while setting the off-diagonal elements to zero (cf. also Zuntzet al. 2017, their appendix D).We consider a general case where a parameter π de-pends on a vector x of N elements via a linear relation π = w T x for some unit vector w . Without loss of gener-ality we can assume that the covariance matrix C of x has C ii = 1 and C ij = r ij for i (cid:54) = j . Since C is positive-definite, | r ij | < . Our task is to seek a value f such that we canguarantee that our estimate of the error on π exceeds itstrue uncertainty: w T ( f I ) w ≥ σ π = w T C w , (A1)for all unit vectors w and any r ij meeting our criteria. An-other way to view this is that we wish to construct a spher-ical error region in x that is at least as large as the ellipsoiddefined by C in every direction.Clearly the condition is satisfied if and only if we canguarantee that f ≥ λ max , (A2)where λ max is the largest of the (positive) eigenvalues λ i of C ( i = 1 , . . . , N ). The eigenvalues are the solutions of apolynomial equation0 = | C − λI | (A3)= (1 − λ ) N − (1 − λ ) N − (cid:88) i>j r ij (A4)+ [lower-order terms in (1 − λ )] (A5)= λ N − Nλ N − + (cid:34) N ( N − − (cid:88) i>j r ij (cid:35) λ N − (A6)+ [lower-order terms in λ ] . (A7)One can see that the roots of this polynomial must satisfy (cid:88) i λ i = N, (A8)Var( λ ) = 2 N (cid:88) i>j r ij ≤ ( N − r . (A9)It is also straightforward to show that the maximum eigen-value must be within a certain distance of the mean eigen-value: f = λ max ≤ (cid:80) i λ i N + (cid:112) ( N − λ ) ≤ N − r. (A10)If we only know that r < 1, then we must increase thediagonal elements of the covariance matrix by f = √ N. This applies to the case when all N values of x are fullycorrelated ( r = 1), and our parameter responds to the meanof x . In the case of our ∆ z i , we have N = 4 , and we estimatethat correlation coefficients between bins should be mod-est, | r ij | ≤ . f = √ . ≈ . π . MNRAS , 000–000 (0000) DES Collaboration APPENDIX B: EFFECT OF N I ( Z ) INHOMOGENEITIES The DES Y1 analyses assume that the WL source galax-ies in bin i have a redshift distribution n i ( z ) that is inde-pendent of sky position θ, apart from the intrinsic densityfluctuations in the Universe. Our survey is inhomogeneousin exposure time and seeing, however, and furthermore isnot properly corrected for Galactic extinction. This inducesangular fluctuations both in the overall source density n S and in the redshift distribution n S ( z ) of the galaxies in thebin (here dropping the bin index i for simplicity). For afixed lens redshift, a fluctuation in source redshift distribu-tion changes the mean inverse critical density. This producesa multiplicative deviation between the measured shear andthe true shear in some angular region, which we will adoptas a rough description of the effect on shear measurementseven though the lenses are distributed in redshift:ˆ γ ( θ ) = [1 + (cid:15) ( θ )] γ ( θ ) . (B1)We can similarly define a deviation of the mean source andlens densities as n L ( θ ) = ¯ n L [1 + δ L ( θ )] , (B2) n S ( θ ) = ¯ n S [1 + δ S ( θ )] , (B3) (cid:104) δ L (cid:105) = (cid:104) δ S (cid:105) = 0 . (B4)The averages above are over angular position θ within thefootprint. In the DES Y1 analyses, the lenses are red-MaGiC galaxies, which are selected to be volume-limitedand hence nominally have δ L = 0 . To ensure that this istrue, Elvin-Poole et al. (2017) look for any correlation be-tween n L ( θ ) and observing conditions. If any such correla-tions are found, the lenses are reweighted to homogenize themean density. We can assume therefore that δ L = 0 every-where, i.e. any fluctuations in lens density are much smallerthan those in the sources.Both the determination of the shear response calibra-tion (Zuntz et al. 2017) and the validation of the redshiftdistribution (in this paper) are produced with per-galaxyweighting, which means that the nominal shear response iscalibrated such that (cid:104) n S (1 + (cid:15) ) (cid:105)(cid:104) n S (cid:105) = 1 (B5) ⇒ (cid:104) (1 + δ S ) (cid:15) (cid:105) = 0 . (B6)We also assume that the source density and depth fluctua-tions are uncorrelated with the shear signal, (cid:104) δγ (cid:105) = (cid:104) (cid:15)γ (cid:105) = 0 , since γ is extragalactic in origin while δ and (cid:15) have terrestrialor Galactic causes.First we consider the galaxy-galaxy lensing observable(Prat et al. 2017). It is an average of tangential shear ofsource galaxies about the positions of lens galaxies. Since itis calculated by summing over lens-source pairs, the resul-tant measurement converges to (cid:104) ˆ γ t ( θ ) (cid:105) = (cid:104) n L n S (1 + (cid:15) ) γ t ( θ ) (cid:105) θ (cid:104) n L n S (cid:105) θ (B7)= (cid:104) (1 + δ s )(1 + (cid:15) ) (cid:105) θ γ t ( θ ) (B8)= γ t ( θ ) . (B9)Here θ is the separation between lens and source, and theaverages are taken over lens-source pairs with separation in some range about θ. The last two lines are simplifica-tions that arise from δ L = 0 and the vanishing conditionsin (B4) and (B6) above. The tangential-shear measurementis, therefore, unaffected by survey inhomogeneity, as long asthe nominal shear and redshift calibrations are weighted bynumber of source galaxies, not by area.The other DES Y1 cosmological observable using thesource population is the two-point correlation function ofshear ξ γ ( θ ). The shear γ is a two-component field, and thereare two non-trivial correlation functions ξ ± , or equivalentlythe spin field can be decomposed into E and B -mode com-ponents. Guzik & Bernstein (2005) analyze the influenceof multiplicative inhomogeneities on the full E/B field, anddemonstrate that such systematic errors shift power between E and B modes at a level comparable to the change in the E mode. Here we will consider a simplified scalar version of theGuzik & Bernstein (2005) formalism, which we can think ofas quantifying the E -mode errors due to inhomogeneity. Ifthese are small, we do not have to worry about effects on B modes either.The calculation of ξ γ in Troxel et al. (2017) accumulatesthe shear products of all pairs of source galaxies 1 and 2separated by angles in a range near θ, yielding an estimateˆ ξ γ ( θ ) = (cid:104) n n (1 + (cid:15) )(1 + (cid:15) ) γ γ (cid:105) θ (cid:104) n n (cid:105) θ (B10)= (cid:104) (1 + δ )(1 + δ )(1 + (cid:15) )(1 + (cid:15) ) γ γ (cid:105) θ (cid:104) (1 + δ )(1 + δ ) (cid:105) θ (B11) ≈ (cid:104) γ γ (cid:105) θ [1 + (cid:104) δ δ (cid:105) θ ] − × (B12) (cid:104) (cid:104) δ δ (cid:105) θ + (cid:104) (1 + δ ) (cid:15) (cid:105) θ + (cid:104) (1 + δ ) (cid:15) (cid:105) θ + (cid:104) δ (cid:15) (cid:105) θ + (cid:104) δ (cid:15) (cid:105) θ + (cid:104) (cid:15) (cid:15) (cid:105) θ (cid:105) ≈ ξ γ ( θ ) [1 + 2 ξ δ(cid:15) ( θ ) + ξ (cid:15) ( θ )] . (B13)We have kept terms only to second order in δ and (cid:15) . Wealso exploited the lack of correlation between true shear andthe systematic errors. We find, following Guzik & Bernstein(2005), that the systematics lead to a multiplicative errorin ξ γ ( θ ) given by the correlation function ξ (cid:15) ( θ ) of the mul-tiplicative systematic; there are additional terms from thecross-correlation ξ δ(cid:15) of density and depth inhomogeneities,which we expect to be of the same order. Since ξ (cid:15) ( θ ) ≤ (cid:104) (cid:15) (cid:105) ,the fractional error in ξ γ is no larger than the square ofthe typical fluctuation in source catalog density or inversecritical density.The RMS fluctuation in source mean redshift inducedby failure to apply the SLR adjustment ( § δz (cid:46) . 01. We estimate the effect of variations in survey depthby removing sources above an i band MOF magnitude m lim from the matched COSMOS sample. The derivative of (cid:104) z (cid:105) w.r.t. m lim is below 0.05 mag − in the relevant range of m lim for all four source bins. At the variation of depth presentin DES Y1 (0.25 mag RMS, Drlica-Wagner et al. 2017),this leads to a RMS fluctuation in source mean redshift of δz (cid:46) . δz (cid:46) . 02 RMS.We expect the source density and inverse critical density(i.e. δ and (cid:15) ) to scale no faster than linearly with the meanredshift of the sample, and the lowest redshift bin has z ≈ MNRAS000 02 RMS.We expect the source density and inverse critical density(i.e. δ and (cid:15) ) to scale no faster than linearly with the meanredshift of the sample, and the lowest redshift bin has z ≈ MNRAS000 , 000–000 (0000) ark Energy Survey Year 1 Results: Redshift distributions of the weak lensing source galaxies < σ > σ ( z ) = 0.029(1+z) + 0.02(1+z) N o r m a li z e d C o un t s DNF DESY1COSMOSWZ Figure C1. Top panel: as Figure 2 showing the average width ofthe posterior distributions of DNF photometric redshifts. Bottompanel: the n i ( z ) for galaxies with bin assignments and estimatedusing DNF photo- z ’s rather than BPZ. The error bars correspondto the standard deviation of the individual source’s σ aroundthe average. The bottom panel is otherwise equivalent to Figure 4. . 3, so (cid:104) (cid:15) (cid:105) (cid:46) ( δz/z ) ≈ . ξ γ measurements at roughly this level.The most accurately measured combination of cosmo-logical parameters in DES Y1 data is S = σ (Ω m / . . ,which is determined to a fractional accuracy of ≈ . 5% (DESCollaboration et al. 2017). Since ξ + ∝ S , roughly, its er-ror due to uncorrected Galactic extinction is estimated tobe ≈ × smaller than the uncertainty level in the DES Y1analyses. APPENDIX C: CALIBRATION OF DNF N I ( Z )The COSMOS validation procedure of § z ’s in the same way as for BPZ, as was theWZ validation. The resultant ∆ z i are shown in Table C1and the n i ( z ) and photo- z precision metrics are plotted inFigure C1. Note that we do not require agreement betweenthe values for DNF and those for BPZ, because they applyto different binnings of the source galaxies. AFFILIATIONS Universit¨ats-Sternwarte, Fakult¨at f¨ur Physik, Ludwig-Maximilians Uni-versit¨at M¨unchen, Scheinerstr. 1, 81679 M¨unchen, Germany Max Planck Institute for Extraterrestrial Physics, Giessenbachstrasse,85748 Garching, Germany Kavli Institute for Particle Astrophysics & Cosmology, P. O. Box 2450,Stanford University, Stanford, CA 94305, USA SLAC National Accelerator Laboratory, Menlo Park, CA 94025, USA Department of Physics and Astronomy, University of Pennsylvania,Philadelphia, PA 19104, USA Centro de Investigaciones Energ´eticas, Medioambientales y Tecnol´ogicas(CIEMAT), Madrid, Spain Department of Physics & Astronomy, University College London, GowerStreet, London, WC1E 6BT, UK Department of Physics, ETH Zurich, Wolfgang-Pauli-Strasse 16, CH-8093Zurich, Switzerland Institute of Space Sciences, IEEC-CSIC, Campus UAB, Carrer de CanMagrans, s/n, 08193 Barcelona, Spain Department of Physics, Stanford University, 382 Via Pueblo Mall, Stan-ford, CA 94305, USA Center for Cosmology and Astro-Particle Physics, The Ohio State Uni-versity, Columbus, OH 43210, USA Department of Physics, The Ohio State University, Columbus, OH 43210,USA Institut de F´ısica d’Altes Energies (IFAE), The Barcelona Institute ofScience and Technology, Campus UAB, 08193 Bellaterra (Barcelona) Spain Brookhaven National Laboratory, Bldg 510, Upton, NY 11973, USA ARC Centre of Excellence for All-sky Astrophysics (CAASTRO) School of Mathematics and Physics, University of Queensland, Brisbane,QLD 4072, Australia Laborat´orio Interinstitucional de e-Astronomia - LIneA, Rua Gal. Jos´eCristino 77, Rio de Janeiro, RJ - 20921-400, Brazil Observat´orio Nacional, Rua Gal. Jos´e Cristino 77, Rio de Janeiro, RJ -20921-400, Brazil INAF - Osservatorio Astrofisico di Torino, Pino Torinese, Italy Department of Astronomy, University of Illinois, 1002 W. Green Street,Urbana, IL 61801, USA National Center for Supercomputing Applications, 1205 West Clark St.,Urbana, IL 61801, USA Kavli Institute for Cosmological Physics, University of Chicago, Chicago,IL 60637, USA School of Physics and Astronomy, University of Southampton,Southampton, SO17 1BJ, UK Fermi National Accelerator Laboratory, P. O. Box 500, Batavia, IL60510, USA Centre for Astrophysics & Supercomputing, Swinburne University ofTechnology, Victoria 3122, Australia Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley,CA 94720, USA Australian Astronomical Observatory, North Ryde, NSW 2113, Australia Sydney Institute for Astronomy, School of Physics, A28, The Universityof Sydney, NSW 2006, Australia Department of Astronomy, The Ohio State University, Columbus, OH43210, USA The Research School of Astronomy and Astrophysics, Australian Na-tional University, ACT 2601, Australia Institute of Cosmology & Gravitation, University of Portsmouth,Portsmouth, PO1 3FX, UK Jodrell Bank Center for Astrophysics, School of Physics and Astronomy,University of Manchester, Oxford Road, Manchester, M13 9PL, UK Department of Physics, University of Arizona, Tucson, AZ 85721, USA Purple Mountain Observatory, Chinese Academy of Sciences, Nanjing,Jiangshu 210008, China Cerro Tololo Inter-American Observatory, National Optical AstronomyObservatory, Casilla 603, La Serena, Chile Department of Physics and Electronics, Rhodes University, PO Box 94,Grahamstown, 6140, South Africa LSST, 933 North Cherry Avenue, Tucson, AZ 85721, USA CNRS, UMR 7095, Institut d’Astrophysique de Paris, F-75014, Paris,France Sorbonne Universit´es, UPMC Univ Paris 06, UMR 7095, Institutd’Astrophysique de Paris, F-75014, Paris, France George P. and Cynthia Woods Mitchell Institute for FundamentalPhysics and Astronomy, and Department of Physics and Astronomy, TexasA&M University, College Station, TX 77843, USA Department of Physics, IIT Hyderabad, Kandi, Telangana 502285, India Department of Physics, California Institute of Technology, Pasadena,CA 91125, USA Jet Propulsion Laboratory, California Institute of Technology, 4800 OakGrove Dr., Pasadena, CA 91109, USA Department of Astronomy, University of Michigan, Ann Arbor, MI48109, USA Department of Physics, University of Michigan, Ann Arbor, MI 48109,USA MNRAS , 000–000 (0000) DES Collaboration Table C1. Values of and error contributions to photo- z shift parameters of DNF n i ( z )Value Bin 1 Bin 2 Bin 3 Bin 4 z PZ range 0.20–0.43 0.43–0.63 0.63–0.90 0.90-1.30COSMOS footprint sampling ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . ± . z i uncertainty ± . ± . ± . ± . metacalibration COSMOS final ∆ z i , tomographic uncertainty − . ± . − . ± . 021 +0 . ± . 021 +0 . ± . z i +0 . ± . − . ± . 014 +0 . ± . 019 — Combined final ∆ z i − . ± . − . ± . 012 +0 . ± . 014 +0 . ± . Instituto de Fisica Teorica UAM/CSIC, Universidad Autonoma deMadrid, 28049 Madrid, Spain Institute of Astronomy, University of Cambridge, Madingley Road, Cam-bridge CB3 0HA, UK Kavli Institute for Cosmology, University of Cambridge, MadingleyRoad, Cambridge CB3 0HA, UK Department of Astronomy, University of California, Berkeley, 501 Camp-bell Hall, Berkeley, CA 94720, USA Astronomy Department, University of Washington, Box 351580, Seattle,WA 98195, USA Santa Cruz Institute for Particle Physics, Santa Cruz, CA 95064, USA Argonne National Laboratory, 9700 South Cass Avenue, Lemont, IL60439, USA Departamento de F´ısica Matem´atica, Instituto de F´ısica, Universidadede S˜ao Paulo, CP 66318, S˜ao Paulo, SP, 05314-970, Brazil Department of Astrophysical Sciences, Princeton University, PeytonHall, Princeton, NJ 08544, USA Instituci´o Catalana de Recerca i Estudis Avan¸cats, E-08010 Barcelona,Spain Department of Physics and Astronomy, Pevensey Building, Universityof Sussex, Brighton, BN1 9QH, UK Instituto de F´ısica, UFRGS, Caixa Postal 15051, Porto Alegre, RS -91501-970, Brazil Instituto de F´ısica Gleb Wataghin, Universidade Estadual de Campinas,13083-859, Campinas, SP, Brazil Computer Science and Mathematics Division, Oak Ridge National Lab-oratory, Oak Ridge, TN 37831 Excellence Cluster Universe, Boltzmannstr. 2, 85748 Garching, Germany Institute for Astronomy, University of Edinburgh, Edinburgh EH9 3HJ,UK MNRAS000