Galaxy Survey data analysis using SDSS-III as an example

Transcription

1 Galaxy Survey data analysis using SDSS-III as an example Will Percival (University of Portsmouth) showing work by the BOSS galaxy clustering working group"

2 Cosmology from Spectroscopic Galaxy Surveys" What are the constituents of matter?" What is the physics of inflation?" e.g. neutrino mass, primordial P(k)" Galaxy Redshift" Survey" Redshift-space distortions" What is the expansion rate of the Universe?" e.g. quintessence, Λ" How does structure form within this background?" e.g. modified gravity, GR" Understanding " acceleration" Is the Universe homogeneous " on large scales?" Copernican principle, Non-Gaussianity" Other non-cosmology science" e.g. galaxy formation & evolution"

3 What does galaxy clustering mean?" Clustering strength = number of galaxy pairs" " "beyond random" P (k) = k(k) 2 (d) = (r) (r + d) 2dFGRS"

4 Correlation function (preview)" Monopole ξ 0! 0 (s) = 2 (s) =5 Z 1 0 Z 1 0 dµ (s, µ) dµ (s, µ)l 2 (µ) Quadrupole ξ 2!

5 Power spectrum (preview)" Spherically averaged power spectrum P(k)"!

6 The BOSS survey"

7 SDSS-III BOSS summary" Duration: Fall Summer 2014, dark time" Telescope: 2.5m Sloan " Upgrade to SDSS-II spectrograph" 1000 smaller fibers" higher throughput" Spectra: " 3600 A < λ < 10,000 A New spectrograph" R = λ/ λ = " (S/N) at mag. limit" 22 per pix. (averaged over Å)" 10 per pix. (averaged over Å) " Area: 10,000 deg2 " Targets: " massive galaxies, z < 0.7, i < 19.9" quasars, z>2.2, g<22.0 selected from candidates " 75,000 ancillary science targets, many categories " Measurements from Galaxies: " da(z) to 1.2% at z = 0.35 and 1.2% z = 0.6 " H(z) to 2.2% at z = 0.35 and 2.0% at z = 0.6 " Measurements from Lyα Forest: " da(z) to 4.5% at z = 2.5 H(z) to 2.6% at z = 2.5 "

8 Data served to the public" Availability through DAS, CAS & CASJOBS" DAS (Data Archive Server)" provides raw data in fits files" expert users only" CAS (Catalog Archive Server)" SQL database with fast search " "capabilities" CASjobs" online workbench" includes history, user personalized database, group options, download options, interface options" Periodic (yearly) data releases" SDSS collaboration published 516 papers to end 2008" 1572 refereed papers with SDSS in abstract to end 2008"

9 Galaxy distribution" SDSS-II main " galaxies" SDSS-II LRGs" BOSS CMASS " galaxies" Image credit: Blanton"

10 147 pages on large-scale galaxy clustering " Anderson et al. (alphabetical) arxiv: BAO measurement in power-spectrum and correlation function. " Reid et al. arxiv: Anisotropic clustering, redshift-space distortion measurements. " Sanchez et al. arxiv: Fits to the full shape of the correlation function. " Ross et al. arxiv: Large-scale systematics. " Manera et al. arxiv: PTHalo mocks. " Tojeiro et al. arxiv: Enhanced redshiftspace distortion measurements." Samushia et al. arxiv: Testing Λ & GR" Plus more to come soon "

11 A collaborative effort " Anderson et al."

12 The BOSS DR9 galaxy sample"

13 The CMASS galaxy sample" BOSS CMASS" Standard colour selec.on criteria SDSS-II LRGs" Also include star- galaxy separa.on criteria Extends SDSS- II LRG cuts to cover both blue and red massive galaxies at higher

14 Galaxy distribution: not like an N-body simulation!" Needs care, but doesn t change numerical requirements

15 Catalog creation" need a random catalog to quantify the survey window " Anderson et al. 2012; Ross et al. 2012"

16 Redshift failures & close pairs" Spectra where we failed to get an accurate redshift are spatially correlated" Close pairs obviously correlated with density" Correct both by upweighting the nearest target with good classification " Ross et al."

17 Target density fluctuations" Target density correlates with stellar density and brightness" Corrected by weighting" See Ross et al. for more details" Ross et al."

18 Reconstruction of linear positions" Padmanabhan et al. 2012; arxiv: "

19 The (primary) BAO analysis"

20 Measuring a distance" Fit the observed acoustic feature using some way to parametrize over nuisance broad-band features (different approaches for P(k) and ξ(r))" Use a fiducial model to compare against observed features in spherically averaged statistics. Departures quantified by dilation scale α:" " "P(k/α) ξ(αr)" The dilation scale α depends on cosmology through:" " "D V /r s = α(d V /r s ) fid " " "D V = [cz(1 + z) 2 d A 2 H -1 ] 1/3 " Anderson et al."

21 Reconstruction on CMASS mocks" Anderson et al."

22 Reconstruction on CMASS" Anderson et al."

23 Reconstruction: error on α" Power spectrum Correla.on func.on Anderson et al."

24 BAO fitting, likelihood calculation" Anderson et al. 2012"

25 Analysis steps requiring significant computation" ~200 CPU hours" Data analysis" raw data reduction, spectral extraction" spectral fitting, redshift extraction" Catalog & mask creation" Reconstruction" calculation" Correlation function calculation" N gal 2 operation, some algorithms N gal log(n gal )" Power-spectrum calculation" N grid log(n grid ), N grid ~ 2048,4096" BAO & 2-pt statistics fitting and cosmological likelihood calculation"

26 Mock catalogues" Mocks needed to calculate covariances, and test analysis methods" Mock calculation would ideally be bases on N-body simulations" Too time-consuming so use approximate methods (e.g. based on 2 nd order fields)" For BOSS DR9, 600 mocks created by populating 2LPT field using the CMASS HOD (galaxy mass relation)" Manera et al. 2012"

27 Analysis steps requiring significant computation" Mock catalog creation" each takes ~200 CPU/hours" Mock catalog analysis" each takes ~200 CPU/hours (same as data)" Multiply by 600 gives significant computing request"

28 BOSS CMASS clustering measurements" Anderson et al."

29 Key BAO measurements" ξ(r) and P(k) based es.ma.ons are appropriate and unbiased, but they include the noise from small scales and shot noise differently We average the two results, and compute the error bar using the observed sca2er of the average value in the mocks. This shows no significant departure from a Gaussian distribu.on Anderson et al."

30 CMASS results" Anderson et al."

31 Extra information from additional measurements" Including the quadrupole allows us to measure H and d A separately (or include an additional measurement of F) " F = (1+z) d A (z)h(z)/c " F called the Alcock-Paczynski parameter " Measure the growth rate from the peculiar velocities through redshift-space distortions" fσ 8 (z=0.57)" amplitude of velocity field" See a wide variety of physics from the full shape of the clustering statistic (as a function of scale)" Ω m, Ω b, Ω ν, f NL, n s!

32 The Future "

33 The ESA Euclid Mission" Space-based Vis and NIR observations of galaxies" " VIS Imaging" NIR Photometry" Tomographic shear" machine" NIR Spectroscopy" NIR Imaging" Redshift machine" Dark Matter and Galaxy! PowerSpectra-meters! Explorer of gravity and expansion" Astronomical data base for " Legacy science"

34 The ESA Euclid Mission"

35 SDSS-II LRG clustering" SDSS LRGs at z~0.35" " Total effective volume" V eff = 0.26 Gpc 3 h -3 " " ~100,000 galaxies" Percival et al. 2009; arxiv: "

36 BOSS CMASS DR9 galaxy clustering" BOSS CMASS galaxies at z~0.57" " Total effective volume" V eff = 0.77 Gpc 3 h -3 " " ~300,000 galaxies" Anderson et al."

37 Redshift slice" 0.9 < z < 1.1" " Total effective volume (of Euclid)" V eff = 19.7 Gpc 3 h -3 " " ~52,000,000 galaxies" Predicted Euclid galaxy clustering"

38 Distance measurements for future surveys"

39 SDSS-II LRG BAO vs other data" ΛCDM models with curvature" flat wcdm models" Union supernovae" WMAP 5year" SDSS-II BAO Constraint on r s (z d )/D V (0.2) & r s (z d )/D V (0.35) " Percival et al. 2009; arxiv: "

40 Euclid BAO predictions" ΛCDM models with curvature" flat wcdm models" Union supernovae" WMAP 5year" SDSS-II BAO Constraint on r s (z d )/D V (0.2) & r s (z d )/D V (0.35) "

41 Conclusions" Science extraction from surveys requires significant computational effort" Recent BOSS analysis required ~10 6 CPU hours" Anderson et al. (alphabetical) arxiv: " Reid et al. arxiv: " Sanchez et al. arxiv: " Ross et al. arxiv: " Manera et al. arxiv: " Tojeiro et al. arxiv: " Samushia et al arxiv: " Spread between a number of groups and computers" Significant future resources needed as surveys increase in size (100x jump in next 10 years, with requirement going as N 2 )"