ebook img

Galaxies in the Illustris simulation as seen by the Sloan Digital Sky Survey - I: Bulge+disc decompositions, methods, and biases PDF

8.4 MB·
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Galaxies in the Illustris simulation as seen by the Sloan Digital Sky Survey - I: Bulge+disc decompositions, methods, and biases

MNRAS000,1–30(2016) Preprint9January2017 CompiledusingMNRASLATEXstylefilev3.0 Galaxies in the Illustris simulation as seen by the Sloan Digital Sky Survey - I: Bulge+disc decompositions, methods, and biases. Connor Bottrell,1(cid:63) Paul Torrey,2 Luc Simard,3 and Sara L. Ellison1 † 1UniversityofVictoria,DepartmentofPhysicsandAstronomy,Victoria,BritishColumbia,V8P1A1,Canada 2DepartmentofPhysics,KavliInstituteforAstrophysicsandSpaceResearch,MassachusettsInstituteofTechnology,Cambridge,MA02139,USA 3NationalResearchCouncilofCanada,HerzbergInstituteofAstrophysics,5071WestSaanichRoad,Victoria,BritishColumbia,V9E2E7,Canada 7 1 0 AcceptedXXX.ReceivedYYY;inoriginalformZZZ 2 n a ABSTRACT J We present an image-based method for comparing the structural properties of galax- 5 ies produced in hydrodynamical simulations to real galaxies in the Sloan Digital Sky Survey. The key feature of our work is the introduction of extensive observational ] realism, such as object crowding, noise and viewing angle, to the synthetic images A of simulated galaxies, so that they can be fairly compared to real galaxy catalogs. G We apply our methodology to the dust-free synthetic image catalog of galaxies from . the Illustris simulation at z = 0, which are then fit with bulge+disc models to ob- h tain morphological parameters. In this first paper in a series, we detail our methods, p - quantify observational biases, and present publicly available bulge+disc decomposi- o tion catalogs. We find that our bulge+disc decompositions are largely robust to the r observational biases that affect decompositions of real galaxies. However, we identify t s a significant population of galaxies (roughly 30% of the full sample) in Illustris that a are prone to internal segmentation, leading to systematically reduced flux estimates [ by up to a factor of 6, smaller half-light radii by up to a factor of 2, and generally ∼ 1 erroneous bulge-to-total fractions of (B/T)=0. v 1 Key words: galaxies: structure – hydrodynamics – surveys – catalogues 5 4 1 0 . 1 INTRODUCTION ing projects such as the Large Synoptic Survey Telescope 1 (LSST) provides an increasingly precise and complete test- 0 A range of tools has been developed in the past several bed against which the models must be benchmarked. 7 decades to model the formation and evolution of galaxies 1 withtheoverarchinggoalofreproducingtheobservedprop- The most direct way of deriving observable properties v: erties of galaxies and their populations in nearby and dis- of galaxies from theoretical predictions is to numerically i tant epochs of the universe (see Somerville & Dav´e 2015 trackthecoevolutionofdarkandbaryonicmatterinhydro- X forarecentreview).Validationofthemodelsisdetermined dynamical cosmological simulations (e.g. Katz et al. 1992, r through comparisons with observations. (e.g. Abadi et al. 1996; Weinberg et al. 1997; Murali et al. 2002; Springel a 2003a; Brooks et al. 2011; Agertz et al. 2011; Guedes et al. & Hernquist 2003; Kereˇs et al. 2005; Ocvirk et al. 2008; 2011;Christensenetal.2014;Agertz&Kravtsov2015;Fur- Crain et al. 2009; Croft et al. 2009; Schaye et al. 2010; Op- long et al. 2015). The growth in observational data from penheimer et al. 2010; Vogelsberger et al. 2012). Tracking modern observational large galaxy redshift surveys such the dynamics of both the dark and baryonic matter self- as the Sloan Digital Sky Survey (SDSS) (Eisenstein et al. consistently to small spatial scales allows predictions to be 2011), the Two-degree-Field Galaxy Redshift Survey (2dF) made about the internal structure of galaxies including the (Folkes et al. 1999), the Deep Extragalactic Evolutionary distributionofgas(Kereˇsetal.2012;Torreyetal.2012)and Probe2survey(DEEP2)(Newmanetal.2013),theCosmic the structures formed from the stellar components (Abadi Assembly Near-infrared Deep Extragalactic Legacy Survey et al. 2003a,b; Governato et al. 2004; Agertz et al. 2011; (CANDELS)(Groginetal.2011)aswellasfromforthcom- Sales et al. 2012; Marinacci et al. 2012). Numerical simula- tionsareusefultoolsforinterpretingtheobservedproperties ofgalaxiesbecausetheyfacilitatecontrolledexplorationsof (cid:63) E-mail:[email protected] galaxyformationandevolution.Inparticular,thelatestgen- † HubbleFellow erationofcosmologicalhydrodynamicalsimulations(e.g.Il- (cid:13)c 2016TheAuthors 2 Bottrell et al. lustrisVogelsbergeretal.2014a,EAGLESchayeetal.2015, ric surface-brightness decomposition analysis tool (Simard FIRE Hopkins et al. 2014, APOSTLE Sawala et al. 2016, 1998).Specifically,theanalysisofSimardetal.(2011)using BAHAMAS McCarthy et al. 2016) are designed specifically the same quantitative morphology pipeline provides an ob- toallowstraight-forwardcomparisonsbetweenobservedand servational reference catalog against which our results can simulated galaxy populations across a range of redshifts. bebenchmarked.Weclarifythatouraimisnottooptimize Explicittrackingofthebirthmass,chemicalevolution, our quantitative morphology pipeline in its capacity to ac- mass loss, ages and motions of stellar populations within curatelymodelthesimulatedgalaxieswithrealism.Theex- galaxies in hydrodynamical simulations allows the assign- perimentdescribedhereisrathertoapplytheidentical,cal- ment of a full spectrum to each stellar population at every ibrated pipeline for deriving morphologies from real galaxy time-step. Synthetic images can be constructed from this images to realistic images of galaxies from a cosmological information that provide representations of the simulated simulation. Only in this way do we enable a comparison galaxies as observed with real photometric or spectroscopic that is procedurally consistent between observed galaxies instruments (Jonsson 2006; Li et al. 2008; Baes et al. 2011; and simulated galaxies with realism. We apply an exten- Robitaille 2011; see Steinacker et al. 2013 for a recent re- sive suite of observational realism to the Illustris synthetic view).Thetoolsforconvertinghydrodynamicalsimulations images to ensure that the same biases in resolution, signal- into synthetic images bridge the gap between theory and to-noise, and crowding affect both model and real galaxies. observations such that consistent analysis of the data of Furthermore,weprovideadetailedcharacterizationofthese each camp are possible. However, the crucial requirement biases by conducting several experiments that quantify the of a fair comparison between the properties derived from random and systematic errors associated with these biases observed galaxies and simulations is that the same biases andidentifytheircorrelationswithmorphology.Wealsoin- affecttheinterpretationoftheirresults(Conroyetal.2010; vestigate the biases associated with the projected viewing Hayward et al. 2012, 2013b,a). Furthermore, the apparent angleonthemorphologicalparameters.Questioningthead- disparity between structural estimates, such as bulge- and equacy of the synthetic images themselves, we examine the disc-to-total ratios (e.g. Scannapieco et al. 2010), derived biasesassociatedwithpost-processingchoicesofhowstellar from photometric decomposition analyses of synthetic im- light is propagated from unresolved stellar populations and ages and estimates from the orbital properties obtained di- theireffectonstructuralestimates.Wereservecomparisons rectly from the simulations also demonstrates that consis- of the morphologies to a forthcoming paper (Bottrell et al. tency in methodology is fundamental to the interpretative 2017). power of such comparisons. Thispaperisorganizedasfollows.Section2providesa In an image-based comparison between the properties basic description of the simulation products and synthetic of galaxies from a hydrodynamical simulation and obser- images and a detailed description of our observational real- vations, Snyder et al. 2015 (S15) obtained non-parametric ismsuite.Section3describestheapplicationofour2Dpara- optical morphology estimates for Illustris’ galaxy synthetic metricquantitativemorphologiestomock-observedgalaxies image catalog (Torrey et al. 2015, T15) and demonstrated and the resulting catalogs. The biases on structural param- thediversityofmorphologiesproducedtherein.Twokeyel- eters from realism are investigated and discussed in Sec- ementsofS15werecrucialtotheircomparisonofsimulated tion 4. We adopt cosmological parameters that are consis- and observed galaxies: (1) the application of observational tentwithWilkinson Microwave Anisotropy Probe 9-yearre- realism by inserting the synthetic images directly into real sults in a ΛCDM cosmogony: Ωm = 0.2726; ΩΛ = 0.7274; fields; (2) consistent methods of deriving morphological es- Ωb = 0.0456; σ8 = 0.809; ns = 0.963; and H0 = 100h km timates. S15 showed that their non-parametric morpholo- s−1Mpc−1 where h=0.704 (Hinshaw et al. 2013). giesforsimulatedgalaxieswithrealism(specificallytheGini and M20 parameters, Lotz et al. 2004), roughly occupied the same space as real galaxies – an important success in 2 SIMULATED GALAXIES the structural comparison of galaxies from hydrodynami- cal simulations and the real universe. While several puzzles Inthissection,wedescribethe Illustrissimulation,produc- arehighlightedbyS15,themorphologicalsimilaritiesshown tion of synthetic stellar mock galaxy images, and applied between simulated and observed galaxies in their analysis observational realism used to generate the image that form motivates a complimentary and orthogonal exploration of the basis of the analysis in this paper. For details of the galaxy structures that may be obtained through paramet- simulationmodels,wereferthereadertoVogelsbergeretal. ric decompositions of simulated galaxies. Furthermore, the (2013, 2014a,b). realismsuiteofS15canbebuiltuponbyincludingthestatis- ticalbiasesfromcrowding,skybrightness,andpoint-spread 2.1 Illustris Simulation function(PSF)resolutiontoenableevenmorethoroughcon- sistency in comparisons with observations. The Illustris simulation is a large-volume cosmological hy- Inthisfirstpaperinourseries,wedescribeourmethod- drodynamicalsimulation(Vogelsbergeretal.2014b,a;Genel ology and data products that facilitate consistent, image- etal.2014).Thesimulationemploysabroadphysicalmodel based comparisons between structural properties derived that includes a sub-resolution interstellar medium (ISM), from real galaxies and galaxies evolved in hydrodynam- gas cooling (including primordial and metal-line cooling), ical simulations. In particular, we derive structural esti- star-formation, stellar evolution and enrichment, supernova mates from SDSS renderings of the synthetic galaxy im- feedback,black-holeseedingandmerging,andactivegalac- age catalog from the Illustris simulation (T15) that are tic nucleus (AGN) feedback. Parameters of the model are consistent with SDSS analyses using the gim2d paramet- tuned to reproduce the stellar mass function at z = 0 and MNRAS000,1–30(2016) Illustris galaxies as seen by the SDSS 3 global star-formation rate density across cosmic time (Tor- particlebasedonitsmass,metallicity,andage.Thesunrise rey et al. 2014; Genel et al. 2014). code(Jonsson2006;Jonssonetal.2010)isusedtomapstel- S15 showed that Illustris contains a variety of galaxy lar light from each particle and generate a synthetic image. morphologies at z = 0 including spirals, ellipticals, and The dust absorption, scattering, or emission functionalities irregulars whose non-parametric structures broadly agree of the sunrise code are not used in the radiative transfer withobservations.Inparticular,thepresenceofrotationally- (see Section 2.2.3). supported disc galaxies in coexistence with elliptical popu- Stellar particles represent discretized unresolved stel- lations is an important result in the current generation of lar populations in Illustris. T15 use adaptive smoothing of cosmologicalhydrodynamicalsimulations.Notonlyisitrel- each star particle using a gaussian kernel with full-width at evant to the capacity with which galaxy populations from halfmaximumequaltothe16th nearestneighbourdistance the simulations can be compared with the observable uni- to convert the discrete simulation particle distribution to a verse, but also in resolving the long-standing problem of smoothdistributionoflight.However,T15cautionthatnei- general angular momentum deficit, high central concentra- ther this, nor any other adaptive or fixed-length smoothing tions, and unrealistic rotation curves in disc-formation ex- prescriptionthattheyexploreisanymoreorlessvalidwhen periments (Navarro & Steinmetz 2000; Scannapieco et al. treating individual star particles as full unresolved stellar 2012;Torreyetal.2012).TheimplicationoftheS15analysis populations. While a more physical light assignment pro- is that realistic galaxy morphology can emerge from simu- cedure may exist, we limit our exploration of methods for lations that couple sophisticated numerical methods torea- distributing stellar light only to the biases on specific pa- sonably complete galaxy formation (specifically feedback) rametersinthemorphologicaldecompositions(Section4.1). modules. Currentlarge-volumehydrodynamicalsimulationssuch 2.2.2 Creation of Synthetic Images as Illustris and EAGLE have been shown to broadly re- producefundamentalrelationsandgalaxymorphologiesob- Photonpacketsfromthestellarparticlesarepropagatedinto served in the real universe (e.g. Snyder et al. 2015; Furlong pinhole cameras with 256×256 pixel resolution providing et al. 2015; Crain et al. 2015; Trayford et al. 2015, 2016; each pixel with the integrated SED of the photon packets Lange et al. in prep.). Their new levels of fidelity make incident upon it - effectively creating a mock integral field themwell-suitedforexperimentscomparingtheobservation- unit (IFU) data cube. The field of view from each camera ally accessible properties of simulated galaxies with those is 10rhm(cid:63) where rhm(cid:63) is the stellar half-mass radius mea- of real ones. A quantitative morphology analysis that em- suredfromthegravitationalpotentialminimumofagalaxy. ploysbulge+discsurfacebrightnessdecompositionsandwill Cameras are located on the arms of a tetrahedron whose yieldstructuralinformationaboutthephysicalcomponents centroid is located at the gravitational potential minimum. isideallyfashionedforthistask.Thekeytotheinterpretive The camera locations provide 4 camera angles with effec- power in relating the results with observations is that the tively random orientation with respect to the galaxy of in- theoretical/syntheticimagesincludethesamerealismfound terest. The mock IFU can then be convolved with a filter in observational datasets. We describe our methods for cre- transmissionfunctionandresampledontothedesiredangu- ating realistic synthetic images in the following sections. lar resolution. The broadband image data products of T15 canbe manipulatedto create syntheticgalaxyimages in 36 unique bands using the dedicated python module sunPy.1 2.2 Synthetic Galaxy Images WeemploythesyntheticgalaxyimagecatalogofT15which 2.2.3 Dust obscuration iscomprisedof6891galaxieswithN(cid:63) >104stellarparticles. The particle resolution cut places a lower limit on the total Thesyntheticimagesarecreatedwithoutusingthedustab- stellar mass of galaxies in our sample at logM(cid:63)/M(cid:12) (cid:38) 10. sorption, scattering, or emission functionalities of the sun- Systemsbelowthiscutareneglectedowingtotheirpoorin- risecode.Theinclusionofdusteffectsintheradiativetrans- ternal resolution. All galaxies in our sample are taken from fer stage is a vital element of a truly comprehensive proce- redshift z=0 and their surface brightness distributions are dure for observational realism. However, the mass and spa- redshifted to z = 0.05. The corresponding luminosity dis- tial resolutions afforded by the Illustris simulation are not tance and angular scale at this redshift, assuming the cos- expectedtoenableconvergedradiativetransferresultswith mological parameters stated at the end of Section 1, are dust(Jonssonetal.2010).Furthermore,giventhatthedust dL =221.3 Mpc and 0.973 kpc/arcsecond. is not traced explicitly incurrent large-volume state-of-the- artcosmologicalsimulationsonthescaleofIllustris(though, see recent work by McKinnon et al. 2016b), the properties 2.2.1 Stellar Light & Surface Brightness Smoothing derivedfromsyntheticimagesofgalaxiesthataregenerated usingradiativetransferwithdustmaydependsensitivelyon Themethodsforproducingsyntheticspectralenergydistri- how dust is implemented in post-processing. butions (SEDs) and idealized simulated galaxy images are In principle, the dust column density and correspond- described in detail in T15. In short, stellar particles inherit ing obscuration along the line of sight to a stellar parti- an initial mass M(cid:63) ≈ 1.3×106 M(cid:12), time of birth, and a clewithinacosmologicalhydrodynamicalsimulationcanbe metallicityfromthelocalISMwheretheparticleiscreated. computed by adopting a model for the distribution of dust Usingthestarburst99single-agestellarpopulationsynthe- sis(SPS)models(Leithereretal.1999;Va´zquez&Leitherer 2005;Leithereretal.2010)anSEDisassignedtoeachstellar 1 http://www.github/ptorrey/sunpy-master/ MNRAS000,1–30(2016) 4 Bottrell et al. thatfollowsthegasdensity,temperature,andspatialmetal- fortheCCDgainthatiscontainedinitsimageheader.The licitydistribution(e.g.Hopkinsetal.2005;Robertsonetal. photometric zero points, airmasses, extinction coefficients 2007; Wuyts et al. 2009a,b). However, a proper treatment (PhotoPrimary table), and CCD gain (Atlas image) for the of dust in generating synthetic images and spectra requires SDSSfieldsarecollectedtobeusedtoconvertthesynthetic the capacity to resolve the gas and dust distributions on Illustris galaxy image fluxes from nanomaggies to counts. smallspatialscales((cid:28)1kpc).IntheIllustrissimulation,the (2) Photometric Segmentation and Loca- complexstructureoftheinterstellarmedium(ISM)ismod- tion Assignment: We employ Source Extractor elledwithapressurized,effectiveequation-of-state(Springel (SExtractor) (Bertin & Arnouts 1996) to create a & Hernquist 2003) that is limited to ∼ kpc scale resolu- segmentation map of the SDSS r−band corrected image. tion.Consequently,gasdiscsinIllustrisgalaxieshavelarger We select a random location on the SDSS corrected image scale-heights and are more chemically and thermodynami- as the designated position of the centroid of the simulated cally homogeneous than in the ISM of real galaxies. It may galaxy images. While upholding our intention of recreating be possible to construct a sub-grid dust-obscuration model realistic crowding statistics of the simulated galaxy pop- that accounts for the unresolved properties of the ISM in ulation, we restrict selection of the centroid location to Illustris and facilitates accurate and convergent radiative pixels which have not been flagged as belonging to other transfer. However, applying such a model would require a objects identified in the segmentation map. Our choices of dedicatedexplorationoftheimpactofsub-griddustproper- deblending parameters in SExtractor are the same as tiesontheresultingsyntheticimagesinordertounderstand those used in S11. themodeluncertainties.Indeed,recentzoom-insimulations (3) SDSS PSF Image Convolution and Flux Con- of Milky Way halos and 25 h−1 Mpc volumes that employ version: We reconstruct the g− and r−band PSF images theIllustrismodelandexplicitlyimplementdusthavechal- specific to the selected location in the corrected image us- lenged the notion that the dust broadly traces the gas den- ing the read_PSF software utility from SDSS2. We remove sity and metallicity averaged over large scales (McKinnon thesoftbiasaddedbySDSS,normalizethePSFimages,and et al. 2016b,a). Evidently, dedicated investigations of the convolve them with the noiseless synthetic galaxy images sub-grid or explicit modelling of dust in large-volume hy- to provide realistic SDSS resolution. Using the photometric drodynamicalsimulationsarerequiredtoresolvetheeffects information obtained in (1) specific to the choice of SDSS and uncertainties associated with dust on measured prop- correctedimages,weconvertthefluxesoftheconvolvedsyn- erties from synthetic galaxy images. Such an exploration is theticimagesfromnanomaggies(defaultdataproductunits) beyond the scope of this paper. Instead, we acknowledge to DN counts. that the lack of dust in the synthetic images may bias our (4) Addition of Signal Shot Noise to Simulated comparison with observations, and focus in this paper on Galaxy: Photon shot noise is generated and added to the analysisoftheunattenuatedsyntheticstellarlightdistribu- convolved synthetic galaxy images following Poisson statis- tions. Nonetheless, our dustless synthetic images will be a tics. The contribution of Poisson noise to the total vari- valuable standard for comparisons with future image cata- ance in each pixel is expected to be small relative to the logs and realism suites that include treatments for dust. sky – which is the dominant noise term for photometry in the SDSS (Simard et al. 2011). In a preliminary analysis of galaxies at z = 0 we found that the inclusion of Poisson 2.3 Observational Realism noise did not affect parameter estimates significantly, but we elect to include it for completeness. Observational realism is applied to the synthetic images of (5) Placement of Simulated Galaxy Image into the simulated galaxy sample from Illustris to enable direct theSDSS:Thesimulatedgalaxyimageswithrealismarein- comparisonsbetweenrealandsimulatedgalaxypopulations. sertedintotheSDSScorrectedimageswithimagecentroids First, we generate synthetic images of simulated galaxies in (whicharealignedwiththegravitationalpotentialminimum SDSS bands g and r using sunPy (as described in Section ofthegalaxy)atthedesignatedlocationfrom(2).Thispro- 2.2)tobeconsistentwiththecanonicalsimultaneousg−and videsthebiascontributionsfromthesky,crowding,andany r−band morphological decomposition analysis of 1.12 mil- other field-specific properties. lion real galaxies in the SDSS DR7 Legacy Survey (Simard et al. 2011, S11). We then use the following procedure to Figure 1 shows a demonstration of the steps described create an unbiased Illustris simulated galaxy population in above.TherightmostpanelofFigure1canbeanalyzedus- SDSS in g and r bands: ing the same pipeline described in S11. The only prior is the position of the simulated galaxy in the corrected im- (1) Selection of SDSS Fields and Photometric age–whichisconsistentwithanalysisofrealgalaxies.One Quantities: We randomly select a unique galaxy objID subtlety is that applying steps (1) and (2) to the place- from SDSS DR7 Legacy photometric galaxy catalog. The ment of simulated galaxies achieves effective matching to SDSS atlas (prefix“fpAtlas”), PSF (prefix“psFIeld”), and the statistics of crowding, resolution, sky brightness within g−andr−bandcorrectedimages(prefix“fpC”)fortherun, the real photometric galaxy catalog. Using this placement rerun, camcol and field containing this galaxy are then procedure, fields containing higher concentrations of galax- obtained from the SDSS Data Archive Server (DAS). The ies have a naturally higher probability of being selected for corrected images are field images that have been reduced placement. While galaxies in isolation will still be repre- throughbias-subtraction,flat-fielding,andpurgingofbright sented, this method guarantees that any systematic biases stars. The PSF image contains all of the necessary meta- data to reconstruct the PSF in any band and location on the corrected image field. The Atlas image is only collected 2 http://classic.sdss.org/dr7/products/images/read psf.html MNRAS000,1–30(2016) Illustris galaxies as seen by the SDSS 5 subhaloID: 312287 camera: 1 PSF convolution r-band synthetic image + Poisson noise insertion into SDSS sky Figure 1. Addition of observational realism to synthetic images of simulated galaxies. Left panel: Synthetic SDSS r−band image of Illustris subhaloID 312287 in a nearly face-on projection. Middle panel: convolution of the r−band synthetic image with the PSF correspondingtoitsinsertionintotheSDSSandadditionofsignalshotnoise.Rightpanel:placementofPSF-convolvedsyntheticimage flux into an SDSS corrected image. The logarithmic scale is identical in each image. The flux in each pixel in the right panel has been liftedby5countssothattheskyisvisibleusingthisscale. in the recovered parameters from quantitative morpholog- – each of which have associated dedicated experiments for ical analysis are (a) consistent with the real galaxy popu- characterizationofthebiasesthatourrealismconsiderations lation and (b) statistically quantifiable through analysis of have on structural parameter estimates. the same simulated galaxy across multiple placements and viewing angles. 3.1 Deblending The methods that are used to delineate object boundaries (deblending) have been shown to affect morphological pa- 3 QUANTITATIVE MORPHOLOGIES rameters – particularly in crowded images (Simard et al. The observational realism described in the previous section 2011). The standard SDSS photo pipeline attempts to iso- puts the Illustris simulated galaxy population in an obser- latethefluxfromanobjecttoreconstructtheimageofwhat vationalcontextthatiswellsuitedforquantitativecompar- the object would have looked like if it were the only source isons with real galaxy populations. To facilitate our com- intheimage.Therefore,pixelsthatsharefluxfrommultiple parison, we use the same quantitative morphology analy- sources are attributed to the area of each source with an sis of S11 which performs 2D photometric surface bright- associated fractional flux contribution based on the recon- ness decomposition with parametric component models. struction modelling. However, deblending with the photo Bulge+disc and single component decompositions contrast pipeline has been shown to produce erroneous photomet- thedetailedstructuralpropertiesofsimulatedandobserved ric and structural estimates such as the production of red- galaxies.Furthermore,ourrealismsuiteplacesthemorpho- outliers andlarge scatter inthe colour-magnitude diagrams logicaldecompositionsofsimulatedgalaxiesonlevelground ofpairgalaxies–whichwerepreviouslyanderroneouslyas- with observations. While S15 employ the same quantitative cribedtoanewpopulationofextremelyredgalaxiesinpair analysisandmethodsforphotometricdeblendinginthesim- systems (Alonso et al. 2004; Perez et al. 2009; Darg et al. ulatedandobservedgalaxypopulations,anybiasesthatare 2010). Inaccurate photometric estimates in pairs is an indi- intrinsic tothesource-delineationormorphologicalanalysis cationthatthesameinaccuraciesarerelevantinallobjects that correlate with resolution, signal-to-noise, and crowd- with closely projected external sources. S11 showed that ing will manifest themselves differently in the observations photometric and structural estimates derived from SEx- and simulations if the statistical distributions for these re- tractor3 deblending combined with gim2d sky measure- alisms are not the same. Our more extensive treatment of observational biases may have important consequences in comparisons between models and observations. 3 SExtractordeblendingusesamulti-thresholdfluxtree.Start- ing with the lowest isophotal threshold and moving up, troughs In this section, we detail our quantitative morpholo- are identified that separate branches which meet the criterion gies analysis. We describe our methods for delineating pho- ofcontainingaspecifiedfractionofthetotalflux.Theminimaof tometric boundaries between sources and the sky and be- thesetroughsdelineatethefluxboundariesandeachpixelisgiven tweencloselyprojectedsources.Wethenexplainourchoice aflagcorrespondingtoauniqueobject–creatingasegmentation of parametric models in the surface brightness decomposi- image. No pixels are shared and therefore no object’s segmen- tions and the structural parameters that are afforded by tation map area extends into the area associated with another thesechoices.Finally,wedescribethedesignofourcatalogs object identified through this scheme. Although the fluxes mea- MNRAS000,1–30(2016) 6 Bottrell et al. ment and bulge+disc decompositions improved upon other schemesusingseveralsensitivetests:thesize-luminosityre- lationofdiscs,andthecolour-magnitudediagramsandfiber colours of pairs. S11 showed that the deblending used in the photo pipeline and associated magnitude and colour estimates was the source of the outliers and scatter in the colour-magnitude diagrams of pairs. S11 also demonstrated that deblending using SExtractor in tandem with para- metric bulge+disc decompositions reduced the scatter and eliminated the outlier populations in the colour magnitude diagrams of pairs – leaving a tight red sequence and clearly separatedbluecloudusingSDSSpaircatalogofPattonetal. (2011). We therefore employ the SExtractor source de- blending procedure used by S11. We do not presume that the S11 scheme is optimal or unique in defining object-sky boundaries and separating objects whose constituent pixels mayhavesharedcontributionsfromothersources.However, we note that although any biases from the S11 deblending andskyestimationschememayaffectourmorphologies,ap- plicationofthesameschemetoourmock-observedsimulated galaxies ensures that the biases are consistent (we explore the impact of alternative deblending on morphological pa- rameters in Appendix B). Before application of our five-step realism suite de- scribedinSection2.3,thesyntheticimagesgeneratedusing sunPyhavenoskyornoiseotherthanaresidualnoisecon- tributionfromtheMonteCarlophotonpropagationscheme in sunrise. However, since the synthetic images are con- structed from the Friends-of-Friends halo finder in Illustris, theremaybecontributionsfromotherstellarsourceswithin thefieldofviewthatgiverisetofluxesthatdonottruncate to zero at the synthetic image boundaries. When added to Figure 2.Fourcameraanglerealizationsofthesamesimulated the SDSS corrected image after the other steps in Section galaxy with observational realism. The galaxy (subhaloID 119) 2.3, the non-zero flux from other sources in the FoF group is inserted into the same location in SDSS for each camera an- mayresultinboxlikefluxboundariesbetweensyntheticim- gle–separatedbyrowintheFigure.Thesciencer−bandimage ageandSDSScorrectedimageintowhichitisplaced.Figure stamp,g−bandimagestamp,andsegmentationstampareshown 2showsanexamplewhereFoFcompanionstothegalaxyof fromlefttoright.Thethirdrowofpanelsindicatethatthecen- interest(centred)areintheprojectedfield-of-view(twobot- tralgalaxyhasatleastonecompanionthatisnotwithinthefield tomrowsofpanels).Theundesirableeffectisexemplifiedin of view for the first two rows. Note the delineation of flux for the bottom row, where a companion is projected along the thecompanionattheboundaryofinsertedsyntheticimage.Only pixelsforwhichthecentralgalaxyhasadominantfluxcontribu- line of sight between the galaxy and the camera position. tion are selected as data used in the fitting. In the fourth row, Galaxy projections with unrealistic artifacts such as another (more massive) companion’s orientation with respect to shown in Figure 2 are easily separable by comparing the the central galaxy places it in the line of sight of the cameras totalfluxatanarbitrarycameraanglewithanyother.Iden- fieldofview.Theprojectioneffectresultsinalargediscontinuity tificationofasignificantpositivesystematicbiasinthetotal betweenthefluxattheboundaryofthesyntheticimageandthe flux of a synthetic image of a particular galaxy subhaloID restoftheSDSSskyandanerroneoussciencemaskthatdoesnot with respect to, for example, an estimate of the mode flux properly delineate between the galaxy’s flux and the visible flux for all camera angles is a effective flagging scheme for these fromthecompanion. situations.Theseprojectioneffectsarealsorare.SubhaloID 119 is also a part of the most massive and most crowded FoFinthesimulation.TheprojectioneffectsinFigure2are of panels in Figure 2 are not problematic for our quanti- onlycommonamonggalaxiesbelongingtothemostmassive tative morphologies pipeline due to our deblending scheme groupsandclusters.Situationssuchasseeninthethirdrow and valid comparisons between the morphological param- eters to those from other camera angles and real galaxies canbemade.Theyarenotuseful,however,inanalysesthat sureddirectlyfromthepixelsassociatedwithanobjectwouldbe comparepropertiesofthesyntheticimagessuchastotalflux systematically be underestimated in this scheme in the presence andphotometricaperturehalf-lightradiuswiththoserecov- ofcloseneighbours,S11showedthatthemissingfluxisrecovered ered from the fitting. The aim of our analysis is to decom- by fitting a surface brightness profile model – which integrates pose the surface brightness profiles of the primary galaxy thefluxofthemodel(whoseformisdeterminedonlybythepix- els flagged as belonging to the object of interest) out to large – which contains the subhalo’s gravitational potential min- radii. In practice, fitting the model recovers the missing flux of imum. Therefore, parametric estimates of flux and size will SExtractordeblendedobjectsincrowdedenvironments. besystematicallybiasedrelativetothepropertiesderiveddi- MNRAS000,1–30(2016) Illustris galaxies as seen by the SDSS 7 rectlyfromthesyntheticimagesforgalaxieswithprominent and stellar half-mass radius. The RIG sample was assem- projectedcompanions/satellites.However,thebiasonlyoc- bled by uniformly sampling 100 galaxies in the M(cid:63)−rhm(cid:63) curscommonlyamongstmassivegalaxiesbelongingtolarge plane while omitting systems in mergers, with strong tidal groups. features,orproblematicprojectioneffects(seeFigure2).Fig- ures C1-C5 show each RIG in the CAMERA 0 projection. TheRIGsampleisusedtoperformrepeatedanalysesof 3.2 Parametric Decompositions galaxies in SDSS fields to obtain parameter estimate distri- butionsforeachgalaxyandeachcameraangle–facilitating We perform simultaneous parametric 2D surface brightness a quantitative analysis of the biases from field crowding, profile fitting of the g− and r−band image data using the gim2d software package (Simard et al. 2002). gim2d em- camera projection, and stellar light sourcing. Our final cat- alogincludesasingledecompositionforallcameraanglesof ploysaMetropolis-HastingsMarkov-ChainMonteCarloal- everygalaxyintheIllustrissyntheticimagecatalogofT15. gorithm for deriving the best-fitting models based on the image data. Only science image pixels identified by SEx- The characterization of biases with the RIGs preceding the tractor to belong to the principle object in the mask are analysesofeverygalaxyinIllustrisenablesquantitativeas- sessmentofthetypicalbiasesassociatedwithspecificregions usedinthelikelihoodcalculations.Theinitialconditionsare of the size-mass space of Illustris galaxies. In the following determinedfromacoarsesamplingoftheverylargevolume two sub-sections, we examine biases associated with galaxy of structural parameter space with generous limits that are imagegenerationandcameraprojectionbyholdingthesky computed from the image moments. The procedure for set- placement fixed. ting the initial conditions and optimization of the models are identical to the S11 analysis. Two models are used to fit the images: a bulge+disc 3.3.2 SMOOTHING catalog (B+D) model and a pure Sersic (pS) model. The B+D de- compositions employ Sersic profiles with indices nb = 4 TheSMOOTHINGcatalogisconstructedtoexplorethechoices and nd = 1 (de Vaucouleurs 1953, 1959; Freeman 1970), for distributing stellar light in creating the synthetic im- while the pS model allows the Sersic index to vary over ages (Torrey et al. 2015). Synthetic images are generated 0.5≤npS ≤8.0. The fourteen free parameters of the B+D for the RIG sample using three alternative stellar light dis- decompositions are the total fluxes in each band fg, fr, tribution(SLD)schemestocomplimenttheexistingimages bulge-to-totalratios(B/T)g,(B/T)r,semi-majoraxisbulge constructed using the fiducial scheme (Section 2.2.1). All effectiveradiusre,bulgeeccentricitye,bulgepositionangle SLDrealizationsofgalaxiesfromtheRIGsampleareplaced (clockwisey-axis≡0)φb,discscalelengthrd,discinclination in the same uncrowded location in SDSS and fitting is per- i,discpositionangleφd,andcentroidpositions(dx)g,(dy)g, formedforasinglecameraangle:CAMERA 0.TheSMOOTHING (dx)r, and (dy)r. The parameters for position angles of the catalogthereforecontains400decompositionswithfourde- bulgeanddisc,discinclination,bulgeellipticity,bulgeeffec- compositionsforeachRIG(oneforeachSLDscheme).The tive radius, and disc scale length have the added constraint biases on photometric and structural parameters from the that they must be equivalent in both bands. Furthermore, SLD schemes may be evaluated with respect to the fiducial the centroid positions of the bulge and disc components of scheme and to each other. Four SLD schemes are explored themodelareconstrainedtobethesame.Similarly,theten in our analysis: free parameters for the pS model are fg, fr, Sersic index npS, semi-major axis disc effective radius re, disc eccentric- (1) Fiducial Smoothing (fn16): Light from each stel- ity e, disc position angle φd, and centroid positions (dx)g, lar particle within the FoF halo is projected from an SPH (dy)g,(dx)r,and(dy)r.Thepositionangleoftheprofile,el- kernel4 with a characteristic scale set by the distance to lipticity,andeffectiveradiusareconstrainedtobethesame the16thneareststellarparticle.Adaptivesmoothingallows in each band in the pS fits. forsmootherdistributionsoflightwhileavoidingunrealistic compactnessaroundlargelyisolatedstellarparticlesatlarge distances from the galactic centre. 3.3 Simulated Galaxy Population Samples and (2) Constant Smoothing (fc1kpc): Same as (1), but Catalogs with the characteristic scale length set to a constant 1 kpc for all stellar particles. Total stellar light is conserved with Several experiments were conducted with increasing com- respect to the fiducial scheme, but compact surface bright- pleteness to characterize the complexities and biases affect- ness features with projected spatial distributions less than ing estimates for the full Illustris galaxy population. Our the characteristic scale are distributed more broadly. samples and catalogs are described in this section. The de- (3) Resampled Adaptive Smoothing (rn16): Young, composition catalogs are summarized in Table 1 at the end bright stellar populations associated with the ∼ 106 solar ofthissectionandaremadepubliclyavailableintheonline masssimulationstellarparticlescanresultinartificiallydis- supplementary information with this paper. Descriptions of tinct,circularlightfeaturesinthesyntheticimages.Tomit- catalog parameters are given in Appendix A in Tables A1 igatethiseffect,weresamplethelightassociatedwiththese and A2. young, bright star particles into 100 particles (child parti- cles). The child particles contain 1/100th the mass of the parent particle, are spatially distributed within the parent 3.3.1 Representative Illustris Galaxy (RIG) Sample We began by selecting a small, but representative Illustris galaxy(RIG)sampleintheparameterspaceofstellarmass 4 AcubicB-splineprofile. MNRAS000,1–30(2016) 8 Bottrell et al. particlesoriginallightkernel,andareassignedGaussianage Table 1. Decomposition catalogs. Each catalog was fitted with and metallicity distributions with 1σ values of 10% of the both a pure Sersic and bulge+disc model. SLD denotes which originalageandmetallicity.Theresultingstellarpopulation stellarlightdistributionmethodsareusedtocreatethesynthetic flux is roughly conserved, but the sharp light profile edges images of each galaxy (Section 3.3.2). Cameras denotes which aresomewhatreduced.Thelightfromallchildparticlesand camera angle projections of each galaxy are used in a catalog. remaining particles is then propagated in the same way as Whether the insertion into the SDSS sky is fixed to a single lo- for (1). cation for all decompositions (Fixed) or follows the randomized (4) Resampled Constant Smoothing (rc1kpc): Same proceduredescribedinSection2.3(Random)isindicatedbyIn- resamplingasin(3)butbutaconstantcharacteristickernel sertion.Ngalisthenumberofgalaxiesusedinthedecompositions. ((cid:63))Decompositionsofallcameraanglesofgalaxiesareperformed scale of 1 kpc is employed. ineachinsertionlocation. 3.3.3 CAMERAS catalog Catalog SLD Cameras Insertion Ngal Ndecomp ASKA fn16 0-3 Random(cid:63) 100 42319 The CAMERAS catalog is constructed to evaluate the biases SMOOTHING all 0 Fixed 100 400 onparameterestimatesfromprojection.Eachgalaxyinthe CAMERAS fn16 0-3 Fixed 100 400 RIG sample is placed in a single, uncrowded location in DISTINCT fn16 0-3 Random 6891 27564 SDSS and fitting is performed for all four camera angles. Thisguaranteesconsistentandcontrolledenvironment,res- olution, and sky in each decomposition and across analyses be sampled to match the luminosity or stellar mass distri- of each RIG. The CAMERAS catalog contains 400 decomposi- butions of real galaxies around z=0.05. tions with four decompositions for each RIG (one for each cameraangle).Thevariationinbest-fittingphotometricand structuralestimatesineachRIGareobtainedfromtheCAM- 4 CHARACTERIZATION OF BIASES ERAS catalog. In this section we explore estimates for total flux, circu- lar half-light radii, and bulge-to-total light ratios in our de- 3.3.4 ASKA catalog composition catalogs. We explore the consequences of post- processingchoicesinhowstellarlightisdistributedspatially AnAllSKyAnalysiswasperformedontheRIGsampleto from particles that embody full unresolved stellar popula- examine insertion effects. The RIGs are placed all over the tions in the simulation. We then assess the sensitivity of SDSS sky with an average of ∼ 100 unique fields for each photometric estimates to the observational realism of pro- galaxy following the steps in Section 2.3. All four camera jection and crowding, respectively. In each analysis of a po- anglesofagalaxyaremodelledineachplacement–provid- tentialbias,wetakeprecautionstocontrolallotherpotential ing a distribution of ∼ 100 unique sets of best-fitting mor- biasessuchthatthevariationinparameterswillbesensitive phological parameters for each camera angle. The full ASKA exclusively to the bias under examination. While we focus catalog contains ∼ 40,000 decompositions of galaxies from ondemonstratingtheB+Ddecompositionresultsinthispa- theRIGsample.Thedistributionofbest-fittingparameters per,wehaveverifiedthatthebiasesreportedforgalaxysizes foreachcameraangleofagalaxysamplestherealstatistics and magnitudes are broadly consistent in the characteriza- for crowding, signal-to-noise, and resolution that exists for tionofbiasesforthepS decompositions(seeSection4.5and SDSS galaxies as a result of our placement criteria and re- Appendix B). alism procedures. We use the ASKA catalog to quantify the random and systematic effects of crowding, signal-to-noise, and resolution. The distributions of best-fitting parameters 4.1 Distribution of Stellar Light foragalaxyineachfixedcameraanglefacilitatethestatisti- Weproducesyntheticimagesforeachofthe100galaxiesin calquantificationofthescatterandsystematicsfrombiases theRIGsampleforasingleprojectionusing4uniquestellar associated with placement. lightdistribution(SLD)schemestocharacterizethebiasesof ourfiducialandalternativeSLDschemes(seeSection3.3.2). Ourcontrolofplacementandprojectionfordecompositions 3.3.5 DISTINCT catalog in the SMOOTHING catalog ensures that the variations in pa- TheDISTINCTcatalogcontainsdecompositionsforallgalax- rameterestimatesforeachgalaxyisexclusivelyduetoSLD iesinsyntheticimagecatalogofT15.Eachcameraanglefor schemes. a given galaxy is assigned probabilistically to a location in The decompositions from the SMOOTHING catalog are theSDSSfollowingitem(2)ofSection2.3.Thecatalogcon- used to quantify the biases from the SLD schemes on in- tains 27,564 decompositions from the 4 camera angles for tegrated magnitude, circular aperture half-light radius, and each of 6891 galaxies from Illustris. Figure 3 shows exam- photometric(B/T).Integratedmagnitudeandhalf-lightra- ples of B+D and pS decompositions of galaxies taken from dius can be computed from the synthetic images as well theDISTINCTcatalog.Thecatalogisdesignedtoinvestigate as the best-fitting (B+D) models for each galaxy and SLD theglobalobservationalpropertiesofthefullIllustrisgalaxy scheme.Thesystematicbiasesonintegratedmagnitudeand population. The DISTINCT catalog also forms the basis for half-light radius are determined by comparing properties comparisons between simulated galaxies and populations of of the best-fitting models mb+d and rhlb+d,c with the cor- realgalaxiesinSDSS.GalaxiesintheDISTINCTcatalogcan responding properties of the synthetic images msynth and MNRAS000,1–30(2016) Illustris galaxies as seen by the SDSS 9 bflags:10 -22.3 10.0 0.23 -22.4 10.4 1.63 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 275172, camera: 2, field matched to SDSS objID: 587731869627449512 bflags:3 -22.6 8.6 0.03 -22.7 9.2 1.28 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 15, camera: 0, field matched to SDSS objID: 587739826592481415 bflags:4 -22.1 11.0 0.84 -21.9 7.1 2.54 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 281154, camera: 3, field matched to SDSS objID: 588848898827747605 bflags:3 -22.1 11.2 0.23 -22.1 11.4 1.55 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 161363, camera: 1, field matched to SDSS objID: 587725993575055610 bflags:8 -22.2 11.5 0.10 -22.2 11.7 1.22 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 256888, camera: 1, field matched to SDSS objID: 587729150373265515 bflags:8 -22.3 6.3 0.69 -22.4 7.6 4.98 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 225517, camera: 1, field matched to SDSS objID: 588017115048771604 bflags:8 -20.3 8.2 0.07 -20.5 9.3 1.44 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 440035, camera: 2, field matched to SDSS objID: 587731187281494209 bflags:2 -21.4 7.0 0.48 -21.5 7.5 1.95 g-band stamp g-b+d model g-b+d residual g-ps model g-ps residual science mask Illustris subhaloID: 290750, camera: 1, field matched to SDSS objID: 587733398111322320 Figure3.Mosaicofg-bandimagesandbest-fittingmodelsforsimulatedgalaxieswithrealism.LefttoRight:(1)sciencecut-out(stamp) ofsimulatedgalaxywithrealism;(2)B+Dmodelabsolutemagnitude,circularhalf-lightradius,andbulge-to-totalratioparameterslisted from left to right; (3) B+D residual image (model subtracted from science cut-out); (4) pS model with absolute magnitude, circular half-lightradius,andSersicindexparameterslistedfromlefttoright;(5)pS residualimage;(6)sourcedelineationmask(sciencemask). Thesciencemasksarecolour-codedbyflagsassociatedwithdistinctobjects.Redcorrespondstothepixelsthatareusedinthefitting. bflags isthenumberofsourcesthatdirectlyneighbourpixelsusedinthefitting.Ascaleinthetopleftofeachpaneldenotes10kpcat z=0.05. MNRAS000,1–30(2016) 10 Bottrell et al. 3 3 3 3 m(fn16-fn16)∆,rphotinput−−21012 m(fc1kpc-fn16)∆,rphot−−21012 m(rn16-fn16)∆,rphot−−01221 m(rc1kpc-fn16)∆,rphot−−21012 3 3 3 3 −12 13 14 15 16 17 18 −12 13 14 15 16 17 18 −12 13 14 15 16 17 18 −12 13 14 15 16 17 18 mr,phot fn16input mr,phot fn16 mr,phot fn16 mr,phot fn16 3 3 3 ) m(fc1kpc-fc1kpc,rphotinput−−21012 m(rn16-fc1kpc)∆,rphot−−21012 m(rc1kpc-fc1kpc)∆,rphot−−01221 ∆ 3 3 3 −12 13 14 15 16 17 18 −12 13 14 15 16 17 18 −12 13 14 15 16 17 18 mr,phot fc1kpcinput mr,phot fc1kpc mr,phot fc1kpc 3 3 (rn16-rn16)input 012 (rc1kpc-rn16) 012 1 1 m∆,rphot−−2 m∆,rphot−−2 3 3 −12 13 14 15 16 17 18 −12 13 14 15 16 17 18 mr,phot rn16input mr,phot rn16 3 ) cinput 2 p k c1 1 r - pc 0 k 1 c r 1 (− m,rphot−2 ∆ 3 −12 13 14 15 16 17 18 mr,phot rc1kpcinput Figure 4. Variation and systematic biases in integrated magnitude for alternative SLD schemes. The panels compare the r-band integratedmagnitudemeasurements,mr,phot,fortheRIGsampletakenfromtheSMOOTINGcatalog.Thefirstpanelofeachrowshowsthe systematicbiasonr-bandintegratedmagnitudecomputedfromthedifferencebetweenthemagnitudeofthebest-fittingB+Dmodeland the synthetic image for the specified SLD scheme, ∆mr,phot. All other panels compare the magnitude estimates from one SLD scheme withanother.Thegreydashedlineineachpanelindicates∆mr,phot=0.Thereddashedlineanddottedlinesshowthemedianand16th to84thpercentilerangeineachcomparison,respectively.GreenmarkershighlighttheRIGswithsystematicoffsetsinthefiducialscheme thatexceedthe84th percentilein∆mr,phot (i.e.,thoselyingabovetheupperdottedredlineinthetopleftpanel).Panelsshowingthe systematicbiasesinotherschemesshowthatthechoiceofschemedoesnotimproveagreementinthemagnitudeestimatesfortheRIGs identified as outliers in the fiducial scheme. The systematic outliers in ∆mr,phot for the fiducial scheme also consistently generate the majorityofthescatterincomparisonsofeachSLDschemewitheachother. MNRAS000,1–30(2016)

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.